journal

the journal of financial transformation

Pricing
Services Real options Assets

Recipient of the APEX Awards for Publication Excellence 2002-2004

04/2005/#13

Our 200 mph laboratory.

The BMW WilliamsF1 Team chose HP to provide the supercomputer used to design the car and to conduct thousands of race simulations. And before the car even hits the track, HP servers and notebooks are used to analyze research data that enables the team to make precise suspension and engine adjustments. Its mission-critical computing for fast-moving enterprises, and then some. www.hp.com/plus_bmwwilliamsf1

bmw williamsf1 team

2003 Hewlett-Packard Development Company, L.P.

Pricing

Editor Shahin Shojai, Director of Strategic Research, Capco Advisory Editors Predrag Dizdarevic, Partner, Capco Bill Irving, President, Capco John Owen, Partner, Capco Editorial Board Franklin Allen, Nippon Life Professor of Finance, The Wharton School, University of Pennsylvania Joe Anastasio, CEO, Cross Border Exchange, and Partner, Capco Philippe dArvisenet, Group Chief Economist, BNP Paribas Jacques Attali, Chairman, PlaNet Finance Rudi Bogni, Former Chief Executive Officer, UBS Private Banking Bruno Bonati, Strategic Consultant, Bruno Bonati Consulting David Clark, NED on the board of financial institutions and a former senior advisor to the FSA Gry Daeninck, former CEO, Robeco Douglas W. Diamond, Merton H. Miller Distinguished Service Professor of Finance, Graduate School of Business, University of Chicago Elroy Dimson, Professor of Finance, London Business School Nicholas Economides, Professor of Economics, Leonard N. Stern School of Business, New York University Michael Enthoven, Chief Executive Officer, NIB Capital Bank N.V. Jos Luis Escriv, Group Chief Economist, Grupo BBVA George Feiger, Executive Vice President and Head of Wealth Management, Zions Bancorporation Gregorio de Felice, Group Chief Economist, Banca Intesa Wilfried Hauck, Chief Executive Officer, Allianz Dresdner Asset Management International GmbH Thomas Kloet, Senior Executive Vice-President & Chief Operating Officer, Fimat USA, Inc. Herwig Langohr, Professor of Finance and Banking, INSEAD Mitchel Lenson, Global Head of Operations & Technology, Deutsche Bank Group David Lester, Chief Information Officer, The London Stock Exchange Donald A. Marchand, Professor of Strategy and Information Management, IMD and Chairman and President of enterpriseIQ Colin Mayer, Peter Moores Professor of Management Studies, Sad Business School, Oxford University Robert J. McGrail, Chairman of the Board, Omgeo Jeremy Peat, Group Chief Economist, The Royal Bank of Scotland Jos Schmitt, Partner, Capco Kate Sullivan, Chief Operating Officer, e-Citi John Taysom, Founder & Joint CEO, The Reuters Greenhouse Fund Graham Vickery, Head of Information Economy Unit, OECD Norbert Walter, Group Chief Economist, Deutsche Bank Group David Weymouth, Chief Information Officer, Barclays Plc

TABLE OF CONTENTS
NOBEL LAUREATE VIEW
8 The international monetary system
A discussion with Prof. Robert A. Mundell, University Professor of Economics, Columbia University, and Winner of The Bank of Sweden Prize in Economic Sciences in Memory of Alfred Nobel 1999

SERVICES
12 Opinion: Pricing traditional vs. alternative asset management services
Francois-Serge Lhabitant, Head, Investment Research, Kedge Capital, and Professor of Finance, H.E.C. University of Lausanne, and Professor of Finance, EDHEC Business School

93

Venture investment contracts as baskets of real options

Didier Cossin, UBS Professor of Finance, IMD Benot Leleux, Stephan Schmidheiny Professor of Entrepreneurship and Finance, IMD Entela Saliasi, FAME and HEC, University of Lausanne

17

Opinion: Propensity-based pricing

Keith MacDonald, Partner, Capco Simon Caufield, Managing Director, Nomis Solutions

103

A real options approach to bankruptcy costs: Evidence from failed commercial banks during the 1990s
Joseph R. Mason, LeBow College of Business, Drexel University, Wharton Financial Institutions Center, and Federal Reserve Bank of Philadelphia

19

Opinion: Pricing outsourcing

Simon Pilkington, Vice President, State Street Corporation

111 22 Opinion: All banks are not alike: Getting out of the commodity trap
Reed K. Holden, Founder, Holden Advisors

Strategic investment under uncertainty: A survey of game theoretic real options models
Kuno J. M. Huisman, Consultant, Centre for Quantitative Methods CQM B.V, and Researcher, Department of Econometrics & Operations Research and CentER, Tilburg University Peter M. Kort, Professor, Department of Econometrics & Operations Research and CentER, Tilburg University, and Professor, Department of Economics, University of Antwerp Grzegorz Pawlina, Lecturer, Department of Accounting and Finance, Management School, Lancaster University Jacco J.J. Thijssen, Lecturer, Department of Economics, Trinity College Dublin

25

Opinion: Powerful pricing processes: How banks can escape the profit crisis
Georg Wuebker, Partner, Financial Sevices, Simon-Kucher & Partners

30

Opinion: Market impact: Transaction cost analysis and the financial markets
Anders Amundson, Managing Director, Elkins/McSherry

ASSETS
120 Opinion: Pricing with time-technology and timescapes
Bala R. Subramanian, Associate Professor, DeVry University

35

Measuring trade execution costs from public data

Hendrik Bessembinder, A. Blaine Huntsman Presidential Chair in Finance, David Eccles School of Business, University of Utah

124 43 Best execution regulation: From orders to markets

Jonathan Macey, Sam Harris Professor of Corporate Law, Corporate Finance, and Securities Law, Yale Law School Maureen OHara, Robert W. Purcell Professor of Management, Professor of Finance, Johnson Graduate School of Management, Cornell University

Opinion: Inflation-induced valuation errors in the stock market

Kevin J. Lansing, Senior Economist, Research Department, Federal Reserve Bank of San Francisco

127

Opinion: Is the investor sentiment approach the solution to the IPO underpricing phenomenon?
Andreas Oehler, Professor of Finance, Bamberg University Marco Rummer, Ph.D. Student, Teaching and Research Assistant, Department of Economics and Management, Bamberg University Peter N. Smith, Professor of Economics and Finance, University of York

57

Credit card pricing developments and their disclosure

Mark Furletti, Payment Cards Center, The Federal Reserve Bank of Philadelphia

REAL OPTIONS
68 Opinion: The interaction between real options and financial hedging in non-financial firms
Tom Aabo, Associate Professor, Aarhus School of Business Betty J. Simkins, Associate Professor, Department of Finance, Oklahoma State University

131

Opinion: The credit spread puzzle

John Hull, Professor of Finance, Director, Bonham Centre for Finance and Maple Financial Group Chair in Derivatives and Risk Management, Rotman School of Management, University of Toronto Mirela Predescu, PhD. Candidate in Finance, Rotman School of Management, University of Toronto Alan White, Peter L. Mitchelson/SIT Investment Associates Foundation Chair in Investment Strategy and Professor of Finance, Rotman School of Management, University of Toronto

73

Opinion: Mergers and acquisitions as a response to economic change

Bart M. Lambrecht, Professor of Finance, Lancaster University Management School

135

Opinion: Impact of seasonality on inflation derivatives pricing

Nabyl Belgrade, Interest Rates Derivatives Research, CDC IXIS CM Eric Benhamou, Head of Quantitative Interest Rates Derivatives Research, CDC IXIS CM

77

Opinion: Valuing real options: Frequently made errors

Pablo Fernndez, PricewaterhouseCoopers Professor of Corporate Finance, IESE Business School - University of Navarra

139

Pricing default-free fixed rate mortgages: A primer

Patric H. Hendershott, Professor, Aberdeen University Robert Van Order, Lecturer, University of Pennsylvania

83

Real options and flexibility

Thomas E. Copeland, Managing Director of Corporate Finance, Monitor Group, and Senior Lecturer, Sloan School of Management, MIT

151

Efficient pricing of default risk: Different approaches for a single goal

Damiano Brigo, Head of Credit Models, Banca IMI Massimo Morini, University of Milan Bicocca

The pricing (r)evolution

Ever since man started to use trading instead of hunting or farming to provide for daily food, mankind has tried to frame the abstract concept of pricing. At first this was done in an unconventional manner through barter but later, with the invention of money, more sophisticated methods were developed to express value in a way which allows all parties to a transaction to understand and appreciate the exchange of money for goods or services. Nowadays it seems that newer, faster, and more reliable methods to calculate prices are invented every day. The financial services industry is no exception to this trend, as is outlined in the current issue of the Journal. The pricing of wholesale and retail financial services changes with each new service or product brought to the market, and pricing creativity or should we say confusion is the menu du jour: offers like basic custody services are provided for free if you enroll in our securities lending and borrowing program, or credit cards with cash back and low interest rates if you sign up today, or the notorious practice of soft dollars in the investment industry are all testimonial to the evolution of FSI pricing models. Correct pricing of financial assets is a complex task, which grows more complex with the degree of sophistication of the asset. Prices for widely traded straightforward instruments are relatively easy to obtain and to interpret, but as soon as the instrument is unlisted, unconventional, or unknown to the public at large, we start to lose our sense of fair value. It is nevertheless a very precise exercise, during which minuscule errors can make or break a deal. The mutual fund industry is a very sensitive subscriber to this theory: a USD 0.01 error on a NAV-calculation could cost upwards of a million dollars in missed subscription, or overpaid redemption fees. These types of monetary risks explain the constant evolution of newer, faster, more reliable pricing models. Another reason is regulation: in the United States, the Financial Accounting Standards Boards plan to require companies to treat employee stock options as an expense directly affecting issuing companies bottom line, has led to the demise of the heretofore favored traditional Black Scholes options pricing model a true revolution. New and existing binomial models, such as the Cox, Ross, Rubinstein model, allow for greater flexibility, more accuracy, and less overstating of value, all to the benefit of issuing companies. I hope you will appreciate the 13th issue of the Journal, which is dedicated to the (r)evolution of pricing in all its facets. As such it may prove for some of you to be a priceless issue, others may discover that they have made assumptions in the past which have pricey consequences today. Nevertheless, this issue of the Journal is a prize in itself.

Rob Heyvaert, Founder, Chairman and CEO, Capco

Has pricing fully evolved?

Pricing of assets and services has come a long way since the first pricing models were developed. We now have complex models to price shares, bonds, derivatives, and any combination thereof. However, despite all these highly developed tools, the valuation of assets still remains far from being totally accurate. The reason, of course, is that there are far too many unknown variables in our models. What we hope to do in this issue of the Journal is provide you with a number of the most advanced pricing models and strategies, with the caveat that while they are the best possible tools available, they are still not totally accurate. Of course, one of the most complex assets to price are currencies, as they involve a deep understanding of national economics and international asset and capital flows. Who better to discuss this topic with us than Prof. Robert Mundell, the recipient of the Nobel Prize in Economics in 1999, and the undisputed world authority on monetary economics. Prof. Mundell has kindly shared with us some of his views about the world of economics and currency pricing. Unlike currency pricing, however, which has had many decades of expert time dedicated to it, the development of accurate models for pricing financial services is only in its infancy. It is quite remarkable that while we have spent billions of dollars developing tools to price complex and not so complex products, we have generally failed to address the pricing of financial services themselves. As in other industries, the correct delivery and pricing of services can be very effective in distinguishing your capabilities from those of your peers. The articles in the first section of this Journal provide some guidance on the tools available to financial services institutions. Section two introduces a pricing model that, though available for some time now, has yet to become mainstream. The use of options pricing models to price flexibilities in all types of projects allows us to quantify behavioral decisions made by the management of most institutions. It also enables us to value the potential benefits of undertaking a negative NPV project in order to have the capability to take on a very positive NPV project in the future. The final section looks at pricing more complex products. Consequently, it is very technical. We cover pricing models used by the most advanced thinkers in the world of finance. We appreciate that many of our readers do not deal with this aspect of the industry, but this section does provide a good overview of how we have advanced in pricing assets. We hope that you enjoy this rather specialized issue of the Journal and that you continue to support us by sending us your best ideas.

On behalf of the board of editors

THE NOBEL LAUREATE VIEW

The international monetary system

A discussion with Prof. Robert A. Mundell
Robert A. Mundell, University Professor of Economics, Columbia University, and Winner of The Bank of Sweden Prize in Economic Sciences in Memory of Alfred Nobel 1999

We are very honored that Prof. Robert A. Mundell, one of the worlds most respected macro-economists and undoubtedly the leading authority in the field of monetary systems, has kindly taken the time to answer a few questions about the impact of currencies on the world economy. Below, you will find the questions posed and the answers kindly provided by Prof. Mundell.

cially against the smaller countries. As a result, such instability of exchange rates creates instability in the financial sectors of every country. There are no advantages for services. Q: There has also been a lot of debate about how economic productivity of countries should be compared, whether nominal currency rates or PPP should be used. Which one would you recommend?

Currency and the economy

Q: In recent years we have observed significant fluctuations of major currencies against each other. Commentators are, of course, very good at using the same explanation to describe why a currency has appreciated against another as they do for its depreciation. What in your opinion are the most important factors that determine currency prices today? Prof. Mundell: Exchange rates are determined by supply and demand. The most important factor in affecting supply and demand is expectations about the future. Under hard fixes, such as the gold standard or credible currency board systems, or even traditional fixed exchange rate systems, such as the Bretton Woods arrangements, the expectations are for stability, and that lets the exchange rate be used as a guide for private-sector planning. Under flexible rates, the main factor in determining expectations is government policy, including central bank policy, especially with regard to changes in interest rates and fiscal policy. Q: In recent years we have experienced very high volatilities among the major currencies. How do you feel such high volatilities will impact the purchase of services like banking where differentiation is weak and substitution opportunities are strong? Prof. Mundell: I think the dollar will hover around where it is for a while, and then depreciate, but not cataclysmically. Q: The dollar has come under a great deal of pressure in recent years. Economic development, which should be used to help it appreciate, is used as a justification that it should fall, since economic growth in the U.S., unlike most countries, is tied to imports, and consequently large trade deficits. Where do you see the dollar go from here? Prof. Mundell: The relation between PPPs and exchange rates is not random. There is lots of evidence that the disparities are a function of, i.e., per capita income. I think more theoretical work needs to be done. Q: And, doesnt the fact that the factors used in the evaluation of inflation rates across countries are very different make the whole exercise of evaluating PPP very hard, if not redundant? Prof. Mundell: Both methods are needed for the two different meanings of productivity. It depends on what you want to use the productivity measure for.

World currency
Prof. Mundell: I believe that the arrangement of flexible exchange rates between national paper currencies with its attendant instability and volatility among each of the nearly 200 countries of the IMF is an absurd arrangement that is at the root of the global macroeconomic problems that we have today. Apart from its global inefficiency, it discriminates espeProf. Mundell: Yes. Q: You have suggested that if the world had a single currency a world currency, as you call it there could be significant benefits gained, such as common, or similar, inflation and interest rates.

8 - The

Journal of financial transformation

Q: How well do you think the euro has been able to achieve similar objectives across Europe? Prof. Mundell: Superbly. Q: Looking at the euro from another perspective, how well has it performed generally? Prof. Mundell: From the perspectives and goals of both Europe as a whole and each individual country, without exception, the euro has been an unqualified success. It needs, however, to be accompanied by improvements in the financial structure with euro bills and bonds playing a larger role. Q: I am aware that you are a big fan of the Tuscany region in Italy and I am sure I do not need to inform you just how much the prices of goods and services have risen since the arrival of the euro, despite what inflation figures show. Dont you feel that the same threat could face the establishment of a world currency? Prof. Mundell: There is the puzzle that people think prices have risen since the euro, but it does not show up in the statistics. That could mean that people notice the prices that have risen, but not those that have fallen. I have seen both the fall and the rise in prices in Tuscany. But, the most noticeable are the rises in prices, because they involve restaurants and tourist and export goods that would be expected to rise with the increase in trade brought about by the euro. The evidence is still anecdotal. But I believe that special effect could be modeled fairly easily. There is a general view that the mental conversion of 1936 lire to 1 euro has involved price surges because the conversion ratio is close to 2000. If people are used to paying 30,000 lire for a lunch, then why not 30 euros instead of 15 euros? It does seem that there has been an escalation of this type. Is it harmful? The producers are helped, the consumers are hurt, and the country as a whole benefits or loses depending

on whether the products in question are net exports or imports. I believe the increases in prices have been mainly in exportables, including in this hotel prices, restaurant prices, etc., bearing in mind that Italy is among the worlds top three tourist destinations. It means, therefore, that Italy has gained, not lost, by the increase in prices. Before the euro, the lira was undervalued on a PPP basis, now it is less undervalued or possibly at par. Could it happen in the world economy? The answer is, yes. I dont believe any country would be hurt by it. But they should be aware of it. Would a world currency joined by China and India involve rising prices? I think that given the undervaluation (by the PPP consideration), they would. But Indias prices are rising fairly briskly staying out. Notice, however, that I have not advocated a single currency for the world economy. A common currency yes, but a single currency no. The latter would not be possible short of a global empire or dictatorship. Q: Today, we already have three major currencies, and if China and most parts of Asia-Pacific peg their currencies to the yen, or even choose it to replace their own, we could even have three major currency blocks. Would better coordination of these currencies achieve the same objective as the world currency or are political pressures too powerful to maintain such relationships? Prof. Mundell: The possibility of an Asia-Pacific fixed exchange rate zone becomes more likely the more the Americans follow policies that have been called Japan-bashing and China-bashing. The tripling of the yen against the dollar from September 1985 to April 1995 ruined the Japanese banking system. China wants to avoid a similar problem. So unless the United States accepts the need for global monetary reform and perhaps a reformation of a Bretton Woods type system, China and Japan will be pushed into each others arms, however reluctantly.

Q: Can a world currency really help improve international trade? Would you say that has happened across Europe? Prof. Mundell: It would be a great benefit to people the world over and the only losers would be the speculators who thrive on instability. Q: Prof. Mundell, thank you very much for giving us this time. We hope that our readers have benefited from your insight.

10 - The

Journal of financial transformation

Services

Pricing traditional vs. alternative asset management services Propensity-based pricing Pricing outsourcing All banks are not alike: Getting out of the commodity trap Powerful pricing processes: How banks can escape the profit crisis Market impact: Transaction cost analysis and the financial markets Measuring trade execution costs from public data Best execution regulation: From orders to markets Credit card pricing developments and their disclosure

Pricing traditional vs. alternative asset management services

Francois-Serge Lhabitant, Head, Investment Research, Kedge Capital, and Professor of Finance, H.E.C. University of Lausanne, and Professor of Finance, EDHEC Business School
The past thirty years have witnessed an increased separation between the ownership and the control of financial wealth. The emergence of modern portfolio theory, the increased efficiency of markets, and the growing sophistication of financial instruments have convinced many, if not most, investors to delegate the management of their portfolios to professional asset managers and their collective investment vehicles. Investment advice is now becoming a commodity. Initially, actively managed funds took the lead and intermediated much of the consumers investments in financial securities. However, their dismal average performance simply provided more general evidence of just how difficult it is to beat the market. It also opened the way for passive strategies and indexed funds, which were then perceived as a cost-effective way of buying equity market exposure a strategy that made sense in an environment of rapidly rising market valuations. However, the end of the technology bubble and the subsequent bear market significantly froze the development of passive funds and provoked interest in alternative investments, such as hedge funds and private equity. Since then, the number of highly specialized, non-traditional asset management firms has been growing exponentially. Many of them are born from the ashes of the failures of mainstream fund managers. Whatever the investment vehicle and investment strategy selected, the delegation of portfolio management activities can be seen as a particular case of the principal-agent model initially introduced in the seminal work of Jensen and Meckling (1976). Since the costs to wealth owners of monitoring those who are charged with managing their financial holdings are rather large, agency theorys most basic suggestion is that principals (investors) should compensate the agents (portfolio managers) through incentive contracts in order to align their respective interests. The nature and intensity of these incentives should depend upon a series of parameters, such as the incremental profit generated by an additional unit of effort from the manager, the precision with which investment performance and risk can be measured and monitored, By contrast, alternative asset managers target an absolute performance, and charge both a management fee (typically 1% of assets under management) and an incentive fee (typically 20 percent of profits) based on their funds overall performance. Anecdotal evidence suggests that for most hedge funds, the management fee is roughly equal to operating costs1 and the primary compensation is the incentive fee. In most cases, a hurdle rate of return must be exceeded by some multiple and any prior losses must be repaid before the fund manager is eligible to receive any incentive income. Over How are incentive contracts implemented in practice? Surprisingly, the empirical evidence seems to suggest that traditional and alternative asset managers have taken diametrically opposed choices. Most traditional investment managers are monitored and evaluated against appropriate style benchmarks, but their compensation is not linked to their relative performance. Rather, they charge a management fee that is generally expressed as a fixed percentage of the assets of their fund. The level of this fee varies depending upon the complexity of the strategy and the asset class considered, but is typically between 1 and 3 percent per annum. Over recent years, asset-based fees have been subject to highly competitive pressures and declined. This is not surprising, as investors have the option of shifting their assets to another asset manager or investment vehicle as soon as they identify a better opportunity. the risk tolerance of the portfolio managers, and their responsiveness to incentives.

Traditional vs. alternative incentives

From a theoretical perspective, incentive contracts may combine three elements, namely, a profit sharing rule (i.e. fee structure) to align incentives in terms of returns; a relative performance component measured against a benchmark to monitor performance, make returns comparable, and audit for common uncertainty; and checks on risk-taking, such as maximum allowable tracking error, reporting requirements, and constraints on available investment choices.

12 - The

Journal of financial transformation

Liang (1999) calculated the average annual management fee for hedge funds to be 1.36%, with a median of 1%. This base fee proved to be much smaller than total management fees surveyed from retail mutual funds.

recent years, these fees have risen, particularly those of established managers who have been able to create a scarcity for their fund, which they then use to increase fees and introduce a lock-up clause. On the contrary, with start-up funds in
2

strategy that works may continue selling it past the asset capacity for which it was designed, just because they are rewarded essentially on the basis of the size of their assets under management. By contrast, performance fees seem to do a better job at aligning the interests of managers (desire for high fees) and investors (desire for high excess returns). When subject to a performance fee, a manager will sell his strategy only up to the asset capacity for which it was designed. Then, he will close his fund to additional investment, as he has stronger incentives for performance than for asset growth. Adding too much assets means being forced to put some money into second-best ideas, and these ideas do not often deliver the kind of returns desired, so asset growth is de-facto limited. At some point, managers may even have to implement net share repurchases. In this context, an increase in revenues should essentially come from improving the excess returns delivered to investors rather than by increasing the assets under management. This partially explain the relatively small size of hedge funds about 80% of the hedge funds reporting to commercial databases manage less than U.S.$100 million of equity capital. However, performance fees also have their drawbacks. The most important ones are linked to their asymmetric nature, the manager participates in the upside, but not in the downside. This corresponds to a potentially perpetual call option with a path-dependent payoff the payoff at any time depends on the high-water mark, which is related to the maximum asset value achieved. This option-like payoff structure may lead to possible adverse incentive effects, because the manager simultaneously owns the option and controls its underlying asset (the portfolio), as well as its volatility. Therefore, near the end of an evaluation period, some managers may decide to increase portfolio risk in order to increase the value of their option.3 On the contrary, outperforming managers may attempt to lock-in their positive performance and dampen portfolio volatility. Alternatively, some fund managers may also try to improve the return of their portfolios by window dressing them, for example by using stale prices

the course of raising capital, investors often obtain discounts on the fees in exchange for early money.

Asset-based fees vs. incentive fees

One may wonder which of the two models, asset-based or incentive fees, is preferable to reduce the agency costs of portfolio management delegation. Fees uniquely based on the size of the assets under management offer a small implicit incentive to managers. As the assets in the fund grow, due to capital inflows or the appreciation of the underlying holdings, the fee collected will grow in tandem. If on the contrary, assets decrease, then the fee collected will be reduced proportionately. Several empirical academic studies have confirmed the positive relationship that exists between a funds relative performance and subsequent inflow of new investments [Sirri and Tufano (1998)], as well as the fact that some investment funds voluntarily waive their stated fees in an attempt to boost net performance and, thereby, to attract additional assets (fee waiving). This suggests that, even though the link between performance and compensation is not direct, it nevertheless appears to be an important factor in determining fund managers behavior. However, we should also note that academic research has evidenced the convex nature of the relationship between fund flow and performance. That is, while superior relative performance generates an increase in the growth of assets under management and, in turn, managerial compensation, there tends to be no symmetric outflow of funds in response to poor relative performance, at least over the shortterm. The convex flow/performance relationship creates an incentive for fund managers to increase risk taking, especially after poor performance. Therefore, the effective incentive of an asset-based fee needs to be carefully assessed on a caseby-case basis. However, in the case of skill-based and capacity constrained strategies, asset-based fees may also create a fiduciary conflict because adding new assets can harm the interests of existing ones. Managers who have developed a

2 As an illustration, Steve Mandel at Lone Pine Capital can charge half of the performance fee (i.e. 10%) of any gain the fund makes from its low. This 10% performance fee continues until the fund has made up 150% of the drawdown from the previous high, then the standard 20% fee kicks in again.

3 Carpenter (2000) studies the optimal portfolio strategy of a manager compensated with a convex option-like payoff and proves this is optimal behavior.

13

rather than real market values (or vice-versa) for illiquid stocks or non-traded assets around the end of an evaluation period. Between the lack of agreed-upon standards, different views about illiquid marks, and moral hazard, valuation can be akin to numerical quicksand. It is interesting to note that although mutual funds and hedge funds seem to disagree on what is the best choice between asset-based and performance-based fees for their external investors, they both agree on their own internal compensation structures, which involve asset management firms and individual fund managers. The compensation of portfolio managers tends to be performance-based, with a fixed base salary topped by bonuses based, partially or entirely, on relative performance. This should be kept in mind, as a complete discussion on the incentives facing mutual funds must consider two layers of agency problems: the agency relationship between the fund company and the fund investors and the agency relationship between the fund company and fund management [Chevalier and Ellison (1999)].

business. It establishes that the law or the fund rules must prescribe the remuneration and the expenditure which a management company is empowered to charge to a unit trust and the method of calculation of such remuneration. Therefore, legal restrictions to the way companies managing mutual funds can be compensated for their services, if any, are to be found only at the national level. Several countries, such as Spain, France, or the U.K., have left a large degree of latitude when it comes to portfolio managers deciding on the mechanism and the value of their compensation. Strikingly, in practice, even though it is legally permissible, most mutual fund companies are almost never compensated through incentive contracts. Instead, they are paid a fixed percentage of assets under management, and the incentive intensity is set to zero. At the other extreme, hedge funds and other lightly regulated private investments companies are primarily charging incentive fees.

The soft dollar arrangements

Our discussion of fees would not be complete if we did not mention soft dollar brokerage, or simply soft dollars. Soft dollar brokerage is a popular arrangement between a fund and its broker. Basically, the fund manager agrees to place a designated dollar value of trading commission business with a broker over a given period of time. In exchange for this promise, the broker provides the manager with research credits equal to some part, say 50 percent, of the promised commissions. Rather than rebating these credits back to investors, the manager keeps them and uses them to buy services and any of the large number of broker-approved research products (hardware, software, subscriptions, databases, etc.) supplied by third-party research vendors. The broker then pays the managers research bill and simultaneously cancels the appropriate number of credits from the managers soft dollar account. From a functional perspective, soft dollars are simply one form of bundling research and execution together into a single commission payment. They are unique in allowing research and execution to be provided by entirely separate firms, thereby promoting vertical disintegration of the research and execution functions.

The regulatory view

An interesting viewpoint on the question of asset management fees is that of regulators, which varies from one country to another. In the U.S., for example, mutual funds are registered investment companies and they are highly regulated by the S.E.C. The latter allows performance incentive fees and enables a fund to charge higher fees when it beats a benchmark, so long as it is willing to charge less when it fails to beat it. As one could expect, many fund managers are perfectly happy to sell their funds to the public on the grounds that it can beat the market, but despite the offer, very few of them are willing to put their own money where their mouths are and take the other side of the bet. According to the Lipper database, less than 2 percent of the U.S. equity mutual funds apply a performance fee. In Europe, a European Council Directive sets the general legal framework within which undertakings for collective investment in transferable securities (UCITS) may carry on their

14 - The

Journal of financial transformation

Do soft dollars reduce or increase agency costs of delegated portfolio management? Both views are defendable. On the one hand, one may argue that soft dollars allow managers to misappropriate investors wealth by churning their portfolios to subsidize research for which they should pay directly. This, in turn, generates various inefficiencies, such as the choice of a broker for his willingness to provide research credits rather than on expected execution quality. At the end of the day, because brokerage commissions are included in the price basis of the underlying security, investors implicitly pay the underlying research costs. Soft dollars, therefore, subsidize the managers use of research inputs, and in some cases the existence or amount of the subsidy is unknown to investors. Thus, portfolio managers shift expenses that are normally shouldered by them onto fund shareholders. But on the other hand, one may also argue that soft dollars are aligning the interests of asset managers with those of their investors. Fund managers typically own a very small percentage of their portfolio, directly as co-investors or via an annual management fee. If managers were required to pay for all research and execution out of their own pockets, they would bear a disproportionate share of the costs of generating portfolio returns in relation to the private benefits based on their portfolio share. Seen in this light, the agency problem faced by portfolio investors is that in the absence of agreement, managers will do too little research, identify too few profitable trading opportunities, and execute too few portfolio trades. Thus, soft dollar arrangements allow investors to subsidize investment research and thereby encourage managers to do more of it, which ultimately benefits the portfolio performance. Last but not least, soft dollars may also be unique in aligning the incentives of brokers and managers. When a broker provides soft dollar research credits to a manager, it typically does so in advance of the commission payments it expects from the manager. But the manager has no legal obligation to trade and may in particular terminate the executing broker relationship with the balance of the soft dollar account unpaid. The broker will then lose a stream of commissions that would have included a premium above the cost of providing low-qual-

ity brokerage. The threat of termination dramatically increases the expected losses to brokers who provide low-quality services, and may therefore perform an effective quality assuring function.

What makes a good performance fee?

Coming back to the main topic of our discussion, at this stage, we may wonder what the necessary characteristics of a good performance fee should be. Ideally, a performance fee should be structured to achieve five main objectives. It should reward a proficient manager for excess return earned over the measurement period, it should control portfolio risk, it should contain fair but significant consequences for manager underperformance, the performance fee agreement should be explicit in its description of the fee structure to eliminate client misunderstandings and properly frame client expectations, and it should be designed so that there is little economic incentive for the manager to grow the assets under management beyond the level at which the performance fees max out. The performance fee structure encourages investment firms to run their strategies at optimal asset levels that permit the maximization of dollars of excess return.

Are hedge fund fees exaggerated?

Throughout the bull market of the 1990s most investors overlooked the fees charged by mutual fund companies because returns were so impressive. But times have changed. We are now in an era of difficult markets and the level of fees have come under close scrutiny. Many traditional investors who are just beginning to venture into alternative investments find their levels of fees overwhelming. If the industry standard seems to be 1 percent for the management fees and 20 percent for the performance fee, several funds among the largest and top-performing ones are far above that. For instance, Caxton Corporation which oversees more than U.S.$10 billon charges 3 percent and 30 percent, while Renaissances U.S.$6.7 billion Medallion fund charges a 44 percent incentive fee, more than twice the industry average. Interestingly, both funds are closed to new investors and have returned money to their existing investors in 2003 in order to be able to maintain

15

positive returns. Of course, only the best performing funds are able to dictate conditions like this. Nevertheless, the list of the top ten earners in the hedge fund industry is impressive. According to Institutional Investor, the top 10 managers earned the following sums in 2003 from a combination of their share of the fees generated by the funds they managed and the gains on their own capital in the funds: George Soros of Soros Fund Management, U.S.$750 million, David Tepper of Appaloosa Management, U.S.$510 million, James Simons of Renaissance Technologies, U.S.$500 million, Edward Lampert of ESL Investments, U.S.$420 million, Steven Cohen of SAC Capital Advisors, U.S.$350 million, Bruce Kovner of Caxton Associates, U.S.$350 million, Paul Tudor Jones of Tudor Investment, U.S.$300 million, Kenneth Griffin of Citadel Investment, U.S.$230 million, Daniel Och of OCH-Ziff Capital Management, U.S.$150 million, and Leon Cooperman of Omega Advisors, U.S.$145 million. Not surprisingly, traditional investors first reaction may be to dismiss the hedge fund industry due to excessive layers of fees. Performance fee structures with 25 and 35 percent carry can work out to be tremendous fees, and immediately prompt the question: Does the return justify the fee? The answer is twofold. Firstly, outsiders invest in a hedge fund because they believe the manager has an expertise that they can not replicate for themselves, or that replication is too costly. This is a fact to remember when looking at hedge fund fees you get what you pay for. Secondly, if investors achieve their objectives after expenses, the fees are justified, even if their level is an especially hard pill to swallow.4 But if a fund delivers poor performance, it is not worth a low fee; in fact, it is worth no fee at all. Thus, fees should be directly related to providing what the investor wants. Consequently, when evaluating or selecting an investment fund, the fee charged should not be the unique determinant. The investment philosophy and quality and tenure of management are also important considerations, amongst others.

ture for asset management? One argument often encountered is that poorly performing managers will be paid less and, therefore, benefit the plan sponsor. On the other hand, managers who perform well will also be paid more. But since the fund earns more, this extra fee will really not cost anything at all. Perhaps, proponents contend, the carrot of higher fees and the stick of lower ones will make the managers work harder. The objectives of performance fees are to reduce them for flat and negative performance and to reward managers for positive absolute performance. Structured properly, this makes a lot of sense for the investor and the manager if added value is properly identified. Then the client and manager are simply entering a profit-sharing plan, and profit sharing is effective in aligning incentives. The problem with performance fees starts when they are not structured properly, that is, if the client is giving a manager a fee based on something other than added value (the true alpha). This is not sustainable in the long-run. Nevertheless, many traditional managers are still reluctant to use performance fees. If the entire industry shifted to performance fees, one of the things that might happen is a reduction in fees in general. For instance, if two-thirds of the managers underperformed, they would draw one-third of their normal fees, while the one-third that outperformed would draw four-thirds of their normal fees. The industry-wide fees would then be cut by a third. Not surprisingly, at least two thirds of the asset management industry will keep fighting such a trend.

References
Carpenter J. N., 2000, Does option compensation increase managerial risk appetite? Journal of Finance, 50, 2311-2331 Chevalier J. and G. Ellison, 1997, Risk taking by mutual funds as a response to incentives, Journal of Political Economy, 105, 1167-1200 Goetzmann, M., J. Ingersoll, and S. Ross, 1998, High water marks, NBER Working Paper 6413, National Bureau of Economic Research, Cambridge, MA Jensen, M. C., and W. H. Meckling, 1976, Theory of the firm: Managerial behavior, agency costs, and ownership structure, Journal of Financial Economics, 3, 305-360 Liang B., 1999, On the performance of hedge funds, Financial Analysts Journal, 55, 72-85 Sirri E. R. and P. Tufano, 1998, Costly search and mutual fund flows, Journal of Finance, 53, 1589-1622

Why so much resistance?

So, in conclusion, are performance-based fees a desirable fea-

16 - The

Journal of financial transformation

4 As an illustration, Goetzmann et al. (2001) use an option approach to calculate the present value of the fees charged by a hedge fund manager and show that the present value of the incentive fees can be quite high (i.e. for a volatility of 15%, the fee can be as high as 13% of the assets under management).

Propensity-based pricing
Keith MacDonald, Partner, Capco Simon Caufield, Managing Director, Nomis Solutions

Financial institutions are now collecting vast amounts of customer data. Spurred on by falling storage costs and more efficient channels, such as the Internet which help automate the process of data gathering and verification, as well as new regulations, such as Know Your Customer and anti-money laundering, financial institutions are now in possession of a wealth of information about their clients. However, despite having access to all this information, very little is being done to utilize it effectively to increase revenues. Advances in analytical tools and massive increases in processing power are allowing highly sophisticated analytical routines to be run across the data warehouses, with benefits being derived in areas such as real time marketing (customized and targeted propositions) and behavioral risk management. Processing of this type can learn from itself to constantly improve results. One area being experimented within financial services is propensity-based pricing (PBP). This involves using customer data to maximize revenue based on how much a customer is willing to pay for a product or service, based on understanding of their price elasticity. As an approach PBP has existed for many years, industries such as telecommunications and business services have been using it to set prices in competitive bidding situations. In the last couple of years, the same principles have been piloted in certain aspects of financial services with dramatic results.

based price for an unsecured loan is x%, but they may have taken the loan at x% plus 10bp. The rate at which a customers demand changes with a change in the underlying price determines that customers price elasticity. The reality is that customers and segments of customers have different price elasticities. For some customers, even small changes in the price of a given product will translate into large changes in demand for that product. Conversely, other customers react minimally to changes in price and are price inelastic. Price elasticity is measured on a sliding scale, with different customers falling along different points on an elasticity continuum. Failure to incorporate customer price elasticity in a pricing decision leads institutions to price sub-optimally, and leads to over- or under-pricing. Traditionally, product marketers focused primarily on the first type of error, and are painfully aware of situations where customers that would have purchased a product at a price lower than the one offered do not accept the offer, and the sale is lost. However, the second type of error is just as economically inefficient, but usually more difficult to spot. A customer who does accept a product offer, but pays a lower price than he would otherwise have been willing to pay, represents lost potential margin. If institutions can differentiate customers by their price elasticity, across the product range, the potential to tailor prices and thus optimize revenue would become a reality. Within regulatory constraints, one price need not necessarily fit all. Propensity-based pricing manages the trade-off between margin and volume and seeks a pricing solution that optimizes revenue. For financial institutions the cycle is iterative and comprises four stages. Firstly, you should segment the market by creating customer groupings based on common buying behaviors and other attributes. Secondly, determine segment price sensitivities. If current pricing has created some price variation, this may be done analytically. If not, create product campaigns targeting different customers within the same segment with different price points. Plot segment-specific demand curves and price elasticities by accounting for both

Principles of propensity-based pricing

Discretionary pricing has been used in financial services since the services were invented. The power of the bank manager to say yes or no is historical. However, in recent years pricing has become more automated, especially with risk-based tools defining acceptable levels of exposure a client handler can introduce. Customer facing roles have in many ways been deskilled by taking some or all of the risk decision away. However, risk-based pricing may not capture the full revenue and margin potential. For example, suppose a customers risk-

17

successful and unsuccessful offers. Thirdly, calculate segmentspecific price points, through the combination of price elasticity results and product business logic, to identify the price point at which the volume/price trade-off is optimized. Finally, track, measure, and recalibrate by monitoring results and over time refining pricing models to reflect both deeper customer segment understanding and changing market conditions.

develop recommendations for immediate pricing changes to drive short-term benefits, and to construct business case and outline implementation plans for deployment aligned with the institutions business environment and needs. Outputs would include a business case, including outline implementation and systems integration plans, short-term profit improvement from recommendations for immediate pricing changes, segmentation and pricing models for future use and refinement, and enhanced understanding of how price optimization could operate across the institution.

The product heatmap

Propensity-based approaches are only effective in certain product/customer situations. The ideal situation would involve one or more of the following, a product which is individually underwritten, such as unsecured lending or general insurance, quotations made based on specific customer situation, such as foreign exchange, a competitive bid process, such as responding to a request for proposal for credit facilities, availability of data on offers that were not accepted by customers as well as win data, availability of product profitability information, and constrained optimization, such as balancing targets for revenue growth, margin, and risk. Looking across a portfolio of products offered via different channels to specific customer segments, some products will be ideal (hot), some not (cold), but many in between. For example, the profitability of many card propositions is based more on usage than price given the cost of supporting transactions. The assessment of the heatmap can define the potential benefit of applying PBP, and the priorities for piloting and rollout.

Conclusion
While some financial institutions are positioning themselves as offering all customers the same price for a product, there are significant benefits to be gained from understanding and applying price sensitivities. Propensity-based pricing is now a realistic tool for financial services institutions to use the power of the latest analytical tools to increase profitability.

Proof of concept
Assessing PBP requires careful piloting, particularly as it can involve changes in behavior as well as how prices are calculated. A proof of concept can be used to cover all aspects of a potential rollout, including data quality and availability, and raising awareness of the principles of the approach. Typically, a proof of concept takes two to three months, with three main objectives. These are to validate expected benefits from deploying the approach by developing initial segmentation and pricing models and applying them to past transactions, to

18 - The

Journal of financial transformation

Pricing outsourcing
Simon Pilkington Vice President, State Street Corporation

A lot of ink has been spilled in recent years on the subject of investment management outsourcing. The concept has gained widespread acceptance as investment managers realize that they can focus better upon their core competencies managing money, developing new products, distribution methods, client service, and so on by shifting their investment operations, fund accounting, and other administrative functions to specialist third-party providers. Managers who outsource these functions do so with the expectation of controlling risk levels, improving service quality, gaining access to proven and superior technologies, and, ultimately, saving money or at least gaining better control of what can often be a variable cost. An intelligent, strategic outsourcing solution should yield all of these benefits, but as is often the case in many aspects of investment management, the bottom line remains the top concern for many managers thinking about outsourcing. Even clients who say that saving money is not their main concern end up focusing upon direct costs or other indirect costrelated business drivers, such as service quality, increased instrument coverage, or improved technology. And this focus inevitably raises a tricky issue: how does one put a fair price on outsourcing services? In what is still a maturing field of the investment business, where costs have not yet coalesced around generally accepted industry standards, there are no easy answers to this question. But analysis of this question from the standpoint of both outsourcing clients and providers considering such factors as length of the deal, up-front contract premiums, benchmarking, and rate card structure is critical in establishing what both sides really want and where the middle ground might lie in any given outsourcing contract. Immediate cost savings are usually high on the agenda of potential outsourcing clients, and in some cases they also look for an upfront premium based on the value they place on their business. In order to close an outsourcing deal, both sides will need to spend a lot of time, and therefore money, in due diligence and implementation planning processes. So if immedi-

ate cost savings are to be achieved, this would require the provider to underwrite the costs incurred in the initial years of the agreement. At this stage in the development of the outsourcing industry, it is questionable whether large, well established providers need to pay upfront premiums, as all of the large providers now have existing clients and infrastructures. It is possible that some smaller players looking to enter the market may be prepared to pay a premium, but prospective clients need to balance this immediate financial benefit against the risk of using a provider which has not yet established its skills and abilities in the market. The advantage for providers and potential clients in this situation concerns the clients intellectual capital, as experienced staff will bring significant benefits to the provider in return for cost savings for the client. This is especially true in a lift-out arrangement, in which the provider takes over the clients existing operations, effecting a complete transfer of both technology and staff. But again, established providers might not necessarily find such arrangements appealing in the future. Obviously, the willingness to compromise on both sides is crucial in establishing a mutually beneficial outsourcing arrangement. One of the advantages of outsourcing from a clients perspective is that it helps them move from a largely fixed set of costs to a more variable cost structure, but while they want to benefit from economies of scale as their business grows, providers need protection against their business declining. One could argue that the ideal tariff for a manager is to pay for services measured by basis points on assets under management. This is how managers are typically compensated, giving them the certainty of costs being a fixed percentage of revenue as assets under management grow or shrink. But most outsourcing clients have indicated a preference to pay a tariff based on volume, and tend to be less interested in ad valorum charges. Providers also want a tariff which reflects the economics of providing the business, structured to reflect the drivers of volume and complexity. So while clients typically gain the variable cost advantage they

19

are after, they also need to accept that there is still a fixed cost aspect from the providers perspective. One of the questions providers usually ask early in the process of negotiating an outsourcing deal is how much the clients operations currently cost. This question usually puts clients on the defensive, probably out of suspicion that the provider is going to take their cost and pitch a fee just below it, whereas without the existing costs as a benchmark the provider might offer a better deal. But in reality, this information is critical to the potential provider, especially if a lift-out is being contemplated, for until the business converts to the providers technology, the existing costs will be incurred by the provider. By being cagey on this subject, a client may put its potential provider in an impossible situation, leaving it unable to adequately assess one of its key drivers in any deal. Successful outsourcing relationships are partnerships based on trust, and this trust needs to be established early in the negotiation process so that both parties can deal with each other in an open, honest way. Length of a proposed deal is an important factor, and often a complicating one. The longer the arrangement, the more compelling a provider can make its proposition, benefits can be gained over the long haul by using more than one service. Clients, however, will need a potential exit strategy covering operational, IT, and cost considerations in case things go wrong:
Operational aspects may be complicated in the case of a

opportunity to benchmark their services and prices. With benchmarking exercises, clients hope to protect against the possibility that the overall market value for services may fall, which would put them at a disadvantage relative to firms which outsource at a later date. In fairness to providers, the benchmarking exercise should also include upside potential for them, providing compensation for them if the cost of service provision becomes more expensive than originally envisaged. Especially across longer deals, how does one anticipate future events such as acquisitions by the client, or of the client by a third party? The contract will probably contain clauses which allow both parties to reassess the relationship in the case of significant change, but most clients have an acquisitive strategy, and are likely to want cost certainty in case they grow by acquisition. This is not always easy to model: it is one thing to create a rate card which is tiered for the gradual growth of existing business, quite another to accommodate the sort of explosive growth which may result from an acquisition. And pricing the transaction is only one element of crystal ball-gazing. Will the operations be located in the same city? Will the acquired entity have similar or different systems? The price of an outsourcing deal already varies if multiple sites have to be used instead of a single site, or if the existing providers site is not sufficient and a new site has to be arranged. Connectivity is an important issue when pricing a deal: a single pipe into the clients data warehouse or data store is more straightforward and less expensive than a multiple point-to-point system. Factoring in the possibility of acquisitions, and their effect on considerations like these, makes pricing a deal that much more difficult. Naturally, the type of deal being pursued will greatly affect the type of savings a client can hope to gain. Most deals are either lift-outs or straight conversions, although some are a mix of the two. Within our current context, lift-outs can be thought of as a means to an end: they will involve the transfer of qualified and experienced staff who know the client to the provider, but they will inevitably cost more up front, as the business will ini-

lift-out, with the possibility that the operation may still be running as a separate and distinct unit at some point in the future.
IT revolves around connectivity and the clients ability to

retain ownership of their data.

Cost considerations will depend on the contract period still

to run, and the extent to which the provider has been able to absorb any up-front costs incurred. And in longer arrangements, clients are likely to look for the

20 - The

Journal of financial transformation

tially be run on legacy technology which is not leverageable. Conversions require the provider to have experienced staff immediately capable of redeploying to meet the new clients needs, and are likely to provide a more immediate financial advantage. However, this apparent advantage must be weighed against the costs of redeploying or laying off staff after the conversion is completed, since most lift-outs will eventually involve a conversion to leverageable strategic technology. Qualified and experienced staff joining the provider in a lift-out are likely to be closely involved in defining business requirements and conversion activity, and as such constitute an asset in their own right. In the final analysis, no two outsourcing deals will be hugely similar. The intricacies of establishing a close relationship between client and provider mean that the pricing of such deals can only be determined on a case-by-case basis. Ultimately, clients usually want a simple rate card, something which fixes their variable costs at advantageous rates which can be structured to match the economics of the deal. A tiered structure may provide protection against growth, but if the client expects to benefit from unit cost savings over the life of the deal, the rates need to be discounted accordingly. And of course, past growth of a client is no guarantee of future growth. It is difficult to assume anything in this regard, so financial analysis needs to include sensitivity analysis to determine the impact of various growth rates for different drivers on the economics of a deal. Establishing a worthwhile outsourcing arrangement requires a great deal of trust and coordination between both client and provider. It is fair to say that this same level of trust, and the establishment of a comprehensive understanding of each others needs, is equally necessary before any deal is signed.

21

All banks are not alike: Getting out of the commodity trap
Reed K. Holden Founder, Holden Advisors
In the eyes of many business clients, all banks look alike. Banking services are viewed by increasingly sophisticated business clients as a series of well-defined commodities that are available from a plethora of banks and financial services companies. Because banking service vendors have not updated their offering to meet the evolving needs of their business customers or communicated their unique value to each client, business customers have focused primarily on price in their acquisition process. To avoid this commodity trap, bank managers must become adept at diagnosing their customers business problems, developing targeted solutions, and communicating this value to the client in terms relevant to their business. Simply stated, the client manager must become a trusted advisor to the customer, partnering with them to deliver greater and greater value to their business operations. Absent this relationship, customers will view low-cost as the only value delivered by their banking services provider and will be happy to let multiple vendors duke it out with low prices for smaller and smaller pieces of their business. Research and years of working with clients tell us that the most productive answer is moving up the customer value chain with solutions based on effective value propositions. In our experience, most customers would like their vendors to offer additional services and solution packages, and are quite willing to pay for this added value. To accomplish this, banks and suppliers of financial services must recognize that while some customers will pay for this added value, others will not, while others will attempt to play poker in the hope of negotiating value at a lower price. They also need to understand the needs of their specific customers and then develop services and solutions that address these needs, and pass the acid test of value by demonstrating value to customers in economic terms. Since each of these types of buyers require a different response, banks need client representatives who have the knowledge to discern each different behavior, the skill to organize the right offering to respond to each type, and the poker skills to negotiate with customers who want to play poker. Differentiation is great, only if you can leverage it with the right customers at the right price.
Price buyers focus on purchasing commodities at the low-

with those customers who want a consultative relationship with their financial services suppliers. Unfortunately, not all customers will pay for these points of differentiation. Client managers must be able to understand and recognize the four primary purchasing behaviors that customers demonstrate:

est price and are unlikely to pay for additional value. They do not prioritize value propositions and they will not pay for differentiation. These customers want to purchase commoditized products and services based on price.
Value buyers are willing to pay for more value if it is rele-

vant to their business operations. These buyers are open to discussions of targeted value propositions. They will participate in the research to identify relevant points of differentiation and will expect a subsequent offering and price.
Relationship buyers rely on trusted advisors to meet their

banking needs, and generally have one. To be successful here, banks need to start with small packages of valuable services and prove their worth over time, gradually earning greater respect.
Poker players are value buyers who have learned that

they can get high value at a low price. These buyers will say that value propositions are unimportant when they are! By incrementally eliminating valued services while lowering prices, vendors can force these customers to reveal their hand about what they truly value.

Who are the poker players?

The first step in gaining pricing power is for client managers to believe that the higher price they are asking reflects tangible incremental value delivered to the customer. This knowledge lays the groundwork for developing a trusted advisor position

Building an effective value proposition

The first step in delivering value to the customer is targeting important or inefficient aspects of their business operations

22 - The

Journal of financial transformation

where new or enhanced financial service solutions can offer tangible business improvement. These solutions have to provide differentiated value; that is, not only do they have to be more than replications of competitive offerings, they have to be valued by customers. Such differentiation is the first step to gaining pricing power. The next step is tangible and measurable value propositions that demonstrate an understanding of the customers real business problem. Understanding of the customers business challenges drives the customers willingness to open a true business dialogue, giving the vendor a chance to sell a valuable solution. Too many banks and financial institutions miss the mark with their value propositions, they are either rhetorical or fail to address the customers real business or economic problems. Tangible and measurable value propositions serve to justify higher prices for both customers and salespeople. Intangible and generic value propositions do not. Banks that seek to replicate their retail brand campaigns for their B2B customer acquisition and retention strategies are at a significant competitive disadvantage. These retail campaigns rely on image rather than substance. Image is rarely compelling to a large business customer. Even if the financial institution is large enough to justify these campaigns, they must be backed by substantive and important differentiation to be successful. Selling treasury services to a business requires a solid understanding of that customers business climate and the needs of their customers. To be effective, points of differentiation should include value statements that address real customer business problems and provide a bottom line result. Validating customer business issues and priorities requires a systematic program of interviewing the right customers the ones who want the value and appreciate the incremental benefits of value at a higher price and the right buyers within the customer organization. For example, low to mid-level managers or back-office operations managers in an organization may focus on price when evaluating a solution, while senior managers or customer-fac-

ing managers may be able to identify points of business value in the proposed solution. One supplier we worked with had five pages of value statements. And their competitors delivered the exact same messages. They were in the middle of negotiations with a purchasing agent who was unlikely to provide any information on non-price, value requirements. Fortunately, it was possible to identify an operational requirement for reliable and timely shipment of products. This was collaborated with the client to craft a sales message around the suppliers ability to superior performance on this dimension. The customer, who had previously focused exclusively on price, chose this vendor and placed the order at a 15% price premium. The relationship proved to be the suppliers largest and most profitable customer!

The acid test

In todays competitive business marketing environment, effective differentiation must be based on quantification. That is the true acid test of whether a financial institutions value proposition is compelling and believable. Value propositions that are me too and fail to connect directly with the business customers financial performance will be lost in the noise of a crowded marketplace. This understanding of the linkages between the features of a banks offering and the real economic value that a customer achieves sends a strong message to prospects about the banks ability to partner with its clients to improve their business operations. In comparison, brand campaigns are blunt instruments that have less impact for business accounts. The process for building an understanding of customer needs and subsequent points of differentiation should be systematic. Researchers start with internal interviews with the marketing, sales, and service staff within the financial institution to identify differentiators that are particularly effective with customers. Once these areas of differentiation are identified, the next step is validation with customers and

23

building these differentiators into offerings and supporting value messages. Successful customer interviews require reaching the right individuals within the customers organization and conducting interviews that surface the customers needs, and how various solutions can provide tangible benefits to meet these needs. The most accessible person within a customer organization may not be the most appropriate person. The challenge is identifying people with the business insight and customer visibility to see the potential of a solution and be able to quantify its value. Senior managers, customer-facing managers, and operations staff can offer valuable insight and confirmatory views of the overall problem. The interviewer must be skilled at both in-depth questioning and the ability to synthesize qualitative interview results into a compelling value proposition. One effective approach is assigning marketing the task of interviewing customers, crafting value propositions, and developing a series of screening questions for a salesperson to use to qualify a prospects business needs. A skilled interviewer must be able to connect the dots between the firms business problems, the key features needed in an offering, and the subsequent value to the customer. For example, a customer, an automobile dealership, may say that the processing time for loan applications is causing lost sales. The next question should determine how many loans are closed with the current capabilities, followed by how fast the applications need to be processed. The last question would obtain a forecast as to what their close rate would be if the loans were able to be processed in ten minutes. From those questions, the interviewer is able to build a required feature set for the solution as well as the economic benefit a customer will experience from the improved offering. By connecting with the real needs of their customers and making tangible improvements in their customers operations via reduced costs, improved sales, or improved profits financial

institutions will become true trusted advisors for their customers. In doing so, they will pull ahead of the competition, provide service enhancements that deliver compelling offerings, communicate with their customers in terms that are meaningful to their business thus building the basis of sustainable differentiation and establish the foundations for a good customer relationship, the kind customers really want. Building the internal capability to understand and continuously analyze value that customers receive from their financial services is key to sustainable competitive advantage. When done properly, systematic monitoring of customer value provides the insight to drive sales, improve margins, and enhance skills when negotiating with the toughest buyer, even the poker player. This approach is essential to gaining the trusted advisor position with relationship buyers. Whichever customer type, this process will help banks avoid the commodity trap and win in highly competitive environments.

24 - The

Journal of financial transformation

Powerful pricing processes: How banks can escape the profit crisis
Georg Wuebker Partner, Financial Services, Simon-Kucher & Partners
Many banks are in the midst of a profit crisis, brought about to a large extent through lack of pricing strategies and poor judgment about the reactions of competitors and customers. Even those who fare quite well could do much better. The following article reveals that price is the primary profit driver. Top managers and decision makers must focus more on the profit potential hidden on the price side. This does not mean simply increasing prices. Rather, success depends on two factors, innovative pricing strategies orientated towards the customers needs and the subsequent complete reorganization of pricing processes. A bank sells its golden credit card for U.S.$100. Sales volume is one million units. The variable costs per unit are U.S.$60, which results in a contribution margin of U.S.$40 per unit. The banks fixed costs are U.S.$30 million. In this situation, the bank earns a profit of U.S.$10 million [(100-60) dollars/unit x 1 million units U.S.$30 million]. How does each of the four profit drivers price, variable costs, volume, and fixed costs change profit when improved by 10 percent? A 10 percent increase in price ($100 to U.S.$110) leads (ceteris paribus) to a profit increase of 100 percent, from U.S.$10 million to U.S.$20 million [(110-60) dollars/unit x 1 million units U.S.$30 million]. The effect of the other profit drivers is much lower. A 10 percent improvement of variable costs, volume, and fixed costs leads (ceteris paribus) to a profit increase of 60, 40 and 30 percent respectively. Bearing this in mind, management should concentrate more on intelligent price increases than on other measures, such as cost cutting. As a result, managers and decision makers must see the price as the primary profit driver. This is especially true in situations with low unit margins. Before a bank decides whether to increase or decrease prices it should analyze the impact of price changes given different cost-income ratios. The bank needs to ascertain how much volume has to be gained (price reduction)/can be lost (price increase) in order to keep the cost-income ratio constant. Example 1: An average lending rate increase of 10 basis points based on the credit volume of a bank could lead to additional profits of U.S.$1 million (based on a volume of credit of U.S.$1 billion), U.S.$10 million (based on a volume of credit of U.S.$10 billion), or U.S.$100 million (based on a volume of credit of U.S.$100 billion). Example 2: A U.S.$10 average increase in per customer profit could lead to additional profits of U.S.$1 million (based on Managers should only change a price if the expected effect on volume is higher/lower than the given percentages. For example, at a cost-income ratio of 0.75, a price reduction of 10% only increases profit, if a volume increase of 67% or more is expected. A 10% increase in price only increases profit, if the volume decrease is lower than 29%. Increased margins stimulated by professional pricing immediately increase profit and do not require expensive upfront These examples demonstrate the power of pricing, also known as power pricing [Dolan and Simon (1996)]. 100,000 customers), U.S.$5 million (based on 500,000 customers), or U.S.$10 million (based on 1,000,000 customers).

Price is the primary profit driver

Many banks are suffering from a crisis into which they have put themselves. A common solution is to cut costs drastically. Personnel cutbacks are almost daily occurrences. American banks, for example, cut more than 100,000 jobs between 2000 and 2004. These measures to reduce costs are inevitable. However, after heavy cost cutting in the recent years, companies have nearly exhausted the potential to reduce costs any further. Is there a way to escape the crisis? What else can be done to increase profits? Management must concentrate more on revenue and price. In these two areas, the potential to increase profits today are significantly higher than through cutting costs. Furthermore, many price measures immediately transform into a profit increase, the so-called quick wins. Our experience shows that for many banks these quick wins can provide a profit potential of millions of dollars. The following calculations illustrate the profit potential of pricing:

25

investments or severance pays. Relative to cost cutting, price management offers three opportunities, it gains time, avoids additional upfront expenses, and increases profit more strongly. Surprisingly, most managers either do not focus on the price or they make the wrong decisions. The following case reveals the power of pricing. An American bank has earned revenues of U.S.$1 billion. The profit is around U.S.$50 million. A sophisticated analysis shows that price increases of on average around 5% are possible without losses in volume. Consequently, the profit would increase by 100% from U.S.$50 million to U.S.$100 million. The key question is: How can such a measure be successfully implemented?

When companies offer such a large number of products as many banks do, with retail price lists of banks very often covering more than 200 price components, or prices are negotiated for each transaction, pricing processes are crucial. Due to the high number of necessary price decisions, the effort involved in each decision has to be limited. Precisely defined processes are required to determine and implement prices and thereby foster acceptable yields. The issue of pricing processes is rather new for several reasons. Traditionally, price decisions have been mostly made based on the feelings and subjective judgment of the person in charge. At present, the concept of pricing processes has been implemented consistently only by a few companies, mostly in the life science and automotive industries. The second reason why the issue is so new is that pricing processes have been a difficult topic for academic researchers. On the one hand, these processes are very industry-specific and quite often also company-specific, requiring time and labor-intensive research to understand. On the other hand, pricing processes are kept top secret. An automotive supplier, for instance, is not interested in initiating a public discussion about its pricing processes. According to our findings, effective pricing processes usually increase the return on sales by about 2 percentage points. In light of the dismal situation which most companies are currently facing, this is revolutionary. Figure 1 examines the increase in profit using cases from various sectors of the financial services industry.

Pricing processes: The key to success

Simple price increases, such as an increase of all prices by 5%, would be very risky and do not work. Just increasing existing prices or ordering sales persons to negotiate higher prices will fail. Banks must apply innovative price strategies that focus on the customers need, value pricing, and are supported by a completely restructured pricing process. In our experience, many banks and insurance companies today do not follow a systematic pricing process. Such a process is comprised of a system of organizational rules, structures, and measures intended to determine and implement prices. Pricing processes are complex and cover the following aspects: 1. The five phases of the pricing process:
Strategic guidelines (objectives, positioning, competition)

The increase in the return on sales in the right column of Figure 1 should be read as follows: In the case of a retail bank, the margin improved by 1.6 percentage points. The cases presented in the figure prove that the starting points for process improvements are very specific. An increase in the return on sales cannot be achieved based on simple price increases. Instead, more intelligent measures have to be applied, a crucial insight for success. Moreover, top management must commit to the new pricing process. After all, the reorientation of the companys competencies regarding pricing and the resulting profit increase are at stake. The following cases from the banking industry demonstrate the impact of a new pricing process.

What do we want? Where do we want to get to?

Status-quo check (current situation and processes) How

do we do it today?
Price decision (structure, level, customization, bundling)

What is the optimal price/price structure?

Implementation (organization, responsibility, IT, incentives)

How can the price be enforced into the market?

Controlling/monitoring How did the prices develop?

2. Information, methods, models, rules, qualifications, competencies, and deadlines. 3. Subjective (e.g. estimations, experience) and objective (e.g. market, competition data) components.

26 - The

Journal of financial transformation

 In the retail unit of a large bank, profits could be increased

through an improved differentiation of customer segments. One group of regular customers reacted strongly to the price increases, other groups and new customers did not show any reaction to the same measures. Improved coordination between pricing, segmentation, product customization, and communications significantly increased marketing efficiency. The bundling of certain products and prices to packages was crucial, especially for cross-selling [Wuebker (2003)].
Another bank was able to increase its revenue by 10 per-

Industry

Revenue category U.S.$500 million U.S.$1-5 billion

Main starting point for improved process Pricing process Pricing audit Extensive extraction of brand value Increased pricing competence of relationship managers Price implementation on indirect channels Intelligent price segmentation Restructuring and optimization of branches Optimization of sales strategy

Increase in return on sales (%) 5

Private banking Retail banking

1.6

Funds

U.S.$600 million U.S.$100500 million

Insurance

cent after a sophisticated analysis of the competitors prices, price elasticities, segmentation, and customized products. A detailed analysis of several product categories demonstrated that no systematic pricing process existed in the company. Every business unit used a different approach to determine prices. Pricing guidelines, a framework for the price policy, were missing. The price elasticities of analyzed products were unknown. Prices were determined based only on costs or competitors prices. The project served as a pilot for the future pricing organization.
A pragmatic analysis of price sensitivities was identified as

Figure 1: Efficient pricing processes lead to major profit increase

successful pricing process is about dramatically increasing pricing intelligence.

Optimization of pricing processes: A case study

Strategic guidelines (Phase 1) During the first phase of the pricing process, the price strategy and desired price positioning must be defined. In general, banks do not have guidelines for pricing their products and services. These guidelines have to be consistent with the future strategy and the positioning of the bank. One bank established the following guideline in a workshop: We are a premium bank. For the high value we deliver we set the appropriate prices. Our price position is in the upper quartile. In the next step, the strategic pricing objectives have to be defined and prioritized. In our experience, complex and unintelligible target systems are relatively widespread. The following example illustrates this. The conflict between profit and volume targets is well known. Most bank managers strive for higher prices without losing volume or market share. In many banks, only explicit volume targets have been defined (number of credit cards, number of new contracts, transactions, assets under management, etc.). Consequently, profit is not the only target that must be considered. A combination of profit and volume targets is common in business practice. Managers have to find a balance between volume and profit. This tradeoff must be resolved and made explicit.

a profit driver for the private banking division of a bank. Based on this analysis and some updated price structures, quick wins were generated. The additional revenue was several tens of millions dollars.
The sales force of another bank was identified as being

overly generous with discounts. They made price the central issue in negotiations with their customers. This problem was solved using argumentation guidelines and a new incentive system. These cases support the assertion, that generating profits through pricing is not an issue of simply increasing or decreasing prices. The parameters for improvement are more complex: information, knowledge, competence, responsibilities, incentives, price structures such as multidimensional or nonlinear prices, price bundling, multi-person pricing, and customization are some of the most important ones. Ultimately, a

27

Status-quo check and price optimization (Phases 2 and 3) The profitability of individual customer relationships is unknown to many banks. In light of this, a structured and substantiated data analysis is required. In general, banks store a great amount of customer information and data. However, this information is frequently not properly structured. An in-depth data analysis clarifies which customer relationships are profitable for the bank under consideration. The case study reveals that the contribution margin of many customer relationships is below zero and that there is no clear relation between assets under management and contribution per customer. This analysis is a crucial prerequisite for future price decisions. The analysis of the existing pricing process of the bank showed that price decisions in general were focused on costs or competitors. The impact of price variations on the sales volume was unknown to product managers. The lack of knowledge about the relationship between price and volume and price elasticities repeatedly caused poor decisions regarding the optimal price. In the future, product managers will gather sufficient information for price decisions. This information involves crucial elements, such as value drivers and benefits of the targeted customer segment, the willingness to pay, and segment-specific price elasticities. Methods like expert judgment or conjoint measurement have demonstrated their capability in the gathering of such information [(Wuebker and Mahajan (1999)]. The result of these methods is a priceresponse function, which serves as a base for the calculation of optimal prices. Implementation and monitoring (Phases 4 and 5) A profit improvement potential of 2 percent (measured in terms of return on sales) through the professional optimization of prices and products is realistic for many banks. To turn this potential into reality, the sales force must understand, accept, and communicate to the customer the new prices and their structures. In customer negotiations, the price, and not value-to-customer, is frequently the focus, which results in overly generous discounts. To avoid this, three instruments

have proven successful in enforcing higher prices and value on the market, an incentive system for the sales people, indicators for better evaluation and segmentation of customers (regarding their price elasticities), and sales guidelines. Incentive systems are crucial for the implementation of a centrally planned price process. They are the only means of achieving the intended price positioning. The purpose of an incentive system is to link the banks objectives to those of the sales person. Defining a sales persons objectives through a combination of volume (i.e. a number of successfully signed contracts or volume of invested capital) and profit targets has proven to make sense. In the long run, only profits, not volume alone, secure the market presence of a bank. Another way to enforce higher prices on the market is by using an indicator system to estimate the customers price sensitivity. Relationship managers could be asked to classify a representative sample of their customers in terms of these indicators. By aggregating these judgments, an overall estimation of the customers price sensitivity can be obtained. The advantage of this method is its simple, fast, and pragmatic operation. A quantification of the relationship between prices and volumes is not possible off-hand. When negotiating, relationship managers should never argue relying only on price. The customers willingness to pay always reflects the perceived value of the products and services. Consequently, value, not price, should be the focus of the negotiation. In business practice, we often notice the opposite as the following case shows. A wealthy customer intends to invest a huge amount of capital, more than U.S.$1 million, at a bank of his choice. He is welcomed with the following words: You are such a wonderful customer. Therefore we will grant you our special price, which is 30% below our regular price. In saying so, the customer relationship manager started an unnecessary price discussion. As a result, the customer, surprised by the generosity, demanded further discounts, ending with a final price 50% below list. To avoid such a disaster, relationship managers should emphasize the value and services

28 - The

Journal of financial transformation

generated by the bank. This requires a profound understanding of the value drivers. This is the basis for developing sales and argumentation guidelines. Supported by a value-based argumentation, relationship managers are able to convince the customer that the demanded price is right. Enforcing price and value can be significantly bolstered this way.

profit), improved price customization (U.S.$2.8 million additional profit), and implementation of centralized controlling (U.S.$1.0 million additional profit). A systematic pricing process is comprised of five phases: strategy, a status-quo check, the price decision, implementation, and controlling/monitoring. Most companies do not follow a systematic and standardized price decision process. The starting point of professional pricing is substantial information. Price elasticities, the willingness to pay for different products, etc. have to be known to optimize prices and products. Reliable and valid methods, such as conjoint measurement, to collect such information are necessary. The art of pricing lies in using intelligent and innovative means of price customization to exploit the customers willingness to pay [Schmidt-Gallas (2004)]. Some of these means are nonlinear pricing, multi-person pricing [Dolan and Simon (1996)] and price bundling [Simon and Wuebker (1999)]. Companies like Dell or Microsoft have successfully adopted these methods. They are a benchmark for value pricing.

Implications: Reaping the harvest

Many banks are suffering from a profit crisis. Simple price reductions in general reduce contribution margins dramatically, resulting in a profit collapse. Consequently, the cause of many crises is price erosion, which is frequently started because the reactions of competitors and customers are not correctly estimated. Many bank managers do not sufficiently understand the effects of price on volume. Yet price is the primary profit driver. The banks executive managers must focus more strongly on profits and prices. Companies need innovative pricing strategies, which are in line with customers needs and value perceptions and accompanied by an overall reorientation of pricing processes. At successful banks like UBS, pricing is closely linked to their organization. Other banks are just starting this process. The value-to-customer of products and services is not completely understood by many banks. Consequently, they do not charge the appropriate price for the value delivered. Increasing profits through more effective pricing processes is a challenge for top management. Gains in the area of several tens of millions of dollars can be achieved by implementing professional pricing processes. The additional profit comes from numerous measures. At one company the following price and product measures were implemented: Optimization of prices for core products (U.S.$5.1 million additional profit), guidelines for systematic discount strategy (U.S.$4.4 million additional profit), structuring of special discounts (U.S.$3.2 million additional profit), optimization of key account pricing (U.S.$3.1 million additional

References
Baumgarten, J. and G. Wuebker, 2004, Strategies against price wars in the financial service industry, February 10, offshoretoday.com. Dolan, R. J. and H. Simon. Power pricing How managing price transforms the bottom line. The Free Press, New York (1996). Hardock, P. and G. Wuebker, 2003, Bundling in the banking sector a promising strategy to create value, March 12, offshoretoday.com. Lauszus, D., Added value private equity: Professional price management to increase profitability. SKP White Paper (2004). Schmidt-Gallas, D., Profitable growth for insurance companies: Manage your pricing or lose money, SKP White Paper (2004). Simon, H., and G. Wuebker. Bundling A powerful method to better exploit profit potential, in: Herrmann, A., R. Frderer and G. Wuebker. Optimal Bundling. Springer, New York (1999). Wuebker, G., 2002, Bundles effectiveness is often undermined, March 18, Marketing News, p. 12. Wuebker, G. and V. Mahajan. A conjoint analysis based procedure to measure reservation price and to optimally price product bundles. Herrmann, A., R. Frderer, R. and G. Wuebker. Optimal Bundling. Springer, New York (1999).

29

Market impact: Transaction cost analysis and the financial markets

Anders Amundson Managing Director, Elkins/McSherry
In 1986, a simple regulatory concept focusing on investment transactions (U.S. Department of Labor Technical Bulletin, 861) forced the financial markets to place an increased emphasis on transaction costs. Almost 20 years later, transaction cost analysis has not only become ubiquitous amongst institutional investors, but created a niche industry focused on providing low cost stock transactions. This paper aims to describe the motivations behind measuring transaction costs, explain how transaction costs are measured in the financial markets, and outline how an intensified scrutiny of transactions will impact the future of the institutional investment community. Initially, investment managers performed the same type of This movement increased the pressure on investment managers to justify their brokerage relationships. Many managers began performing transaction cost analysis internally. Eventually, a regulatory concept called best execution1 all but explicitly demanded investment managers to regularly perform some type of transaction cost analysis. trustees and pension executives were able to identify managers and brokers whose costs diminished overall investment performance.

Why measure transaction costs?

Ideally, transaction costs are measured in an effort to improve the investment process, however, the intensity of detail and motivation behind a transaction cost study can vary depending on the audience. Originally, the idea of transaction cost analysis gathered widespread appeal with pension fund managers, however these days, transaction cost analysis has become universal amongst asset management firms and is very quickly becoming common within institutional brokerage houses. While the basic philosophy of transaction cost measurement can be applied to trades involving any type of financial instrument, the actual practice of transaction cost analysis is most pervasive in the equity markets.

basic, macro-level analysis common among pension funds. However, with the proliferation of order management technology, transaction cost analysis evolved from a basic evaluation tool to a much more precise methodology that investment managers could actually utilize at the trade level. The widespread adoption and integration of order management systems allowed asset management firms to assign a quantitative value to both their internal investment processes and external trading relationships. In fact, from a business perspective, many investment managers might agree that enhancing performance though a diligent improvement of the transaction process can dramatically improve the firms competitive position within the industry simply because so many managers have a difficult time delivering excess performance.

Institutional investment hierarchy

Pension funds (government, corporate, and union pension funds, as well as foundations and endowments) first started monitoring transaction costs as part of a government mandate that required pension trustees to monitor the brokerage relationships of their investment managers. The philosophy behind the mandate was that brokerage commissions were a pension asset and pension executives had a fiduciary responsibility to understand how the brokerage relationships of their external investment managers affected the overall performance of the pension fund. Before these rules, transaction costs were not even known, let alone understood, by the vast majority of pension trustees. As transaction costs began to be evaluated by the decision makers within a pension organization, To put the impact of transaction costs on portfolio performance in perspective, consider that during the fourth quarter of Standard and Poors recently calculated2 that there have been roughly 1096 distinct investment groups managing large capitalization investment funds in the U.S. consistently for the past three years. Of these funds, only 31.1% beat the S&P 500 large cap stock index. The percentage of index beaters drops even further for mid-cap and small-cap investment managers. Only 20.9% of mid-cap stock funds beat the S&P MidCap 400 index over the past three years and only 23.2% of small-cap stock managers beat the S&P SmallCap 600 index during the same time period.

30 - The

Journal of financial transformation

Gohlke, G., 2000, What constitutes best execution? Securities and Exchange Commission, November 30 2 Pane, R., and S. Dash, 2005, SPIVA active funds scorecard 4th Quarter 2004, Standard and Poors, January 18

2004,3 every NYSE transaction cost the average investment manager roughly 0.26% per trade and every NASDAQ transaction cost the average investment manager about 0.35% per trade. This total transaction cost encompasses commissions, fees, and market impact. While these costs might seem insignificant, over time, depending on the frequency of trading activity, transaction costs can add up, seriously hampering overall investment performance. Estimates have placed the average annual portfolio turnover for U.S. equity funds at around 90%.4 If this is true, the average active equity manager can expect transaction costs to reduce overall portfolio performance by roughly 0.50% annually or 1.50% over a three year period. During the same three year time period cited above, the S&P 500 beat large-cap equity managers by 0.95%. This suggests that an effort to become better than average in terms of transaction costs could help an investment manager discover a large portion of the extra performance needed to beat both the market and their peers. A recent article published in the Wall Street Journal5 helps to confirm this point. The article notes that fund behemoth Fidelity Investments adds as much as 0.30% to annual performance due to transaction cost analysis. Rounding out the bottom of the institutional investment hierarchy are brokerage firms. Brokerage firms have experienced the greatest impact to their business models as a result of the best execution mandate. The increased emphasis placed on transaction costs coupled with decreased spreads due to decimalization6 by both the NYSE and NASDAQ, has markedly squeezed margins for brokers. Over the past five years, brokers have witnessed a 40% reduction in average commissions7 for trades done on the NYSE. As a result, many brokerage firms have adapted their operations to cater to a more transaction focused marketplace. In fact, it is not uncommon to find brokers who actively tout their ability to help investment managers achieve best execution. A cursory glance at industry publications, such as Traders Magazine or Institutional Investor, will uncover dozens of brokerage

advertisements proclaiming expertise in delivering high quality executions.

Transaction costs in the financial markets

Academically, a transaction cost is described as the cost of making an economic exchange. In the financial markets, the measurable costs of an economic exchange can be characterized by a combination of explicit and implicit costs. Explicit transaction costs are generally considered easy to estimate. These transaction costs consist of commissions paid to brokers for trading services and fees or taxes paid to exchanges and regulatory agencies. Investors usually understand the financial impact of explicit transaction costs before the actual transaction takes place. Implicit transaction costs, on the other hand, can be much trickier to estimate. Implicit transaction costs, commonly described by the investment community as market impact, are the additional costs of actually implementing an investment idea, essentially the simple act of buying or selling a security. Market impact costs are different from explicit costs to the extent that market impact is not a cost actually levied by a broker or governing body, but by the supply and demand forces of the market itself. Market impact is usually not known before the transaction occurs and can be difficult to predict.

Measuring market impact

The measurement of any process begins with a benchmark. If you are jogging from point A to point B, the benchmark can be the amount of time it takes to run that distance. If you are manufacturing cars, the benchmark might be a minimum number of defects per 1,000 cars produced. In the financial markets, market impact is most commonly measured by comparing trades against benchmarks composed of price, time, and trading volume. Of the various benchmarks, the most common measurement metric is called the VWAP (Volume weighted average price). The VWAP is simply an average of all market activity throughout a specific time period, usually a day, in which larger trades

3 Source: Elkins/McSherry Universe. 4 Bogle, J., 2000, Mutual funds at the millennium, Bogle Financial Markets Research Center, May 15 5 Kelly, K., and J. Hechinger, 2004, How fidelitys trading chief pinches pennies, Wall Street Journal, October 12

6 The NYSE switched to decimal pricing in January of 2001 and the NASDAQ switched to decimal pricing in April 2001. 7 Source: Elkins/McSherry Universe.

31

have a greater overall impact on the average price. Common variants of the VWAP include available VWAP, which calculates the VWAP from the time the trader starts the order to the close of trading that day, and interval VWAP, which calculates the VWAP from the time the trader starts the order to the last trade execution for the order. Once a VWAP is created, trade executions are compared to that benchmark to measure market impact. Market impact is simply the positive or negative difference between the trade execution and the benchmark. While the basics of transaction cost measurement certainly begins with the comparison benchmark, understanding the personality of the order is critical to the determination of the correct benchmark. Some types of trades are considered very common and easy to perform, whereas others are considered quite difficult. Appreciating how all of the key variables contribute to the complexity of an order is of chief importance when estimating market impact. More so, many groups will compare themselves to multiple benchmarks because some of the most popular benchmarks, such as VWAP, contain inherent flaws that a savvy trader can game to their own advantage. Some of the other common trading benchmarks include implementation shortfall, which measures the absolute cost of a trade, and strike price, which compares the trade to a predetermined target price, such as the closing price of the stock the day before the trade occurred. With implementation shortfall, the metric generated is the price-to-price difference between the start of the trade and the final execution of the order. Implementation shortfall, available VWAP, and interval VWAP are similar inasmuch that all three benchmarks attempt to quantify costs within a very specific window of time. Using time segmented benchmarks is a recent innovation that has allowed asset managers to perform separate evaluations of each specific participant in the trade process (portfolio manager/ trader/broker). Breaking apart the trade process in this manner When looking at each segment of a transaction, it is important to understand the key variables that also contribute to the execution strategy. Many of these variables are also measured within a transaction cost analysis. These variables include the price volatility of the stock being traded, the size of the order relative to the daily volume of the market, the time horizon for completing the order, the opportunity cost of an uncompleted order, the type of broker used, and a cost described as information leakage.8 Information leakage arises when an institutional broker proves to be less than discreet while handling an order. If word leaks that money manager XYZ is buying or selling a major position, traders can and will jump in front of that order in an effort to capture a portion of the inevitable price movement, adversely affecting the overall market impact for manager XYZ and ultimately reducing the overall performance of the portfolio. Often, because of the potential for information leakage, investment managers will concede a certain degree of price to preserve anonymity for their trade executions.9 helps to determine how much or how little each order participant contributed to the overall cost of the transaction. At that level of analysis, groups can begin to streamline the trade process in an attempt to gain transaction cost efficiency.
Figure 1: Elkins/McSherry trade cycle, NYSE, 2nd Quarter, 2004 Segment Full day Portfolio manager to last execution Portfolio manager to trader Trader to broker Broker to last execution Broker to close Benchmark VWAP Implementation shortfall Implementation shortfall Implementation shortfall Interval VWAP Available VWAP Average cost 0.07% 0.52% 0.07% 0.11% 0.03% 0.05%

The future of transactions

Proof that transaction cost analysis has had a systemic influence on the brokerage business is evidenced by one of the hottest trends in the brokerage industry, algorithmic trading.10 Algorithmic trading is essentially the use of a computer gen-

32 - The

Journal of financial transformation

8 Sofianos, G., and J. Bacidore, 2002, Trading and market structure, Goldman Sachs, October 24 9 Birger, J., 2004, American Centurys secret weapon, Money Magazine, September 10 Risk Waters, 2004, Algorithmic trading explosion, Risk Waters Group, December 13

erated trading strategy to help buy-side traders achieve a specific benchmark. Of the various algorithmic strategies designed by brokerage houses, the core strategy offered by all of the major algorithmic desks is a trading engine designed to mimic the VWAP or its variants. Other brokerage groups have designed alternative trading systems that cater to the specific transaction needs of investment managers. Some of these systems aim to discover pools of liquidity or cross large blocks of stock anonymously. Pretrade analysis software that helps traders analyze orders before the executions begin is also starting to find a place on the trading desks of many investment management firms. One brokerage house even has plans to introduce a derivative product designed to hedge the VWAP.11 Many of these products are still in the nascent stages of growth. Over the next several years, a wide range of execution based tools and services will be available to the investment community. Also, it is highly likely that the idea of best execution will become increasingly common beyond the equity markets. Demand for fixed income transaction cost analysis has been gaining steam over the past 12 months and a few investment managers already analyze currency and futures transactions. All things considered, the capital markets are becoming a more transaction focused marketplace.

11 Chapman, P., 2004, J.P. Morgans analytical difference, Traders Magazine, December

33

Services

Measuring trade execution costs from public data1

Hendrik Bessembinder
A. Blaine Huntsman Presidential Chair in Finance, David Eccles School of Business, University of Utah

Abstract
This study assesses the sensitivity of trading cost estimates derived from publicly-available trade and quote data to two methodological issues: the time adjustment made before comparing trades to quotes, and the procedure used to designate trades as buyer or seller-initiated. The results indicate that making no allowance for trade reporting lags is optimal when assessing whether trades are buyer or seller-initiated, for both Nasdaq and NYSE stocks. However, trade prices are best compared to earlier quotations when assessing trade execution costs, in order to capture the effect of systematic quotation revisions in the seconds before trades are reported. Despite the sensitivity of trading cost measures to these methodological issues, inference as to whether the Nasdaq dealer market or the NYSE auction market provides lower trade execution costs is not sensitive.

This paper is extracted from Bessembinder, H., 2003, Issues in assessing trade execution costs, Journal of Financial Markets, 6:3, 233-258, with permission from Elsevier.

35

Measuring trade execution costs from public data

How much do investors pay to have their trades executed in financial markets? Obtaining accurate measures of trade execution costs and assessing the reasons for their systematic variation is important to individual investors, portfolio managers, those evaluating brokerage firm or financial market performance, and corporate managers considering where to list their shares. There seems to be increased interest in obtaining accurate measures of trading costs in recent years. This interest may be attributable in part to a realization that careful control of trade execution costs may be as, or more, important in determining portfolio performance than the ability to identify mis-valued securities in the highly competitive financial markets. The United States Securities and Exchange Commission (SEC) now requires (Regulation 11Ac1-5) individual market centers to use their internal order data to construct and disseminate monthly reports of average market quality. But, as Bacidore et al. (2003) emphasize, measures of trade execution quality are quite sensitive to detailed methodological choices. It is unclear the extent to which execution quality reports prepared by individual market centers will be comparable with each other. Those who wish to compare market quality measures for broad sets of traders and markets, while using completely uniform definitions and methods, are likely to have to construct their own measures and to rely on the publicly available trade and quote data. The same will be true for those who wish to evaluate market quality measures for time intervals finer than one month or those who wish to study measures, such as net order imbalances, other than those required by rule 11Ac1-5. One shortcoming of the public trade and quote data is that whether a trade was initiated by a buyer or a seller must be imperfectly inferred from the data. A second issue is that, although trade prices can be readily compared to quotes in effect at the trade report time, the appropriate comparison might be to quotes in effect at an earlier time, such as the time of the trading decision or at the time the order arrived at the market, and these times are generally not known.

The central methodological issue considered here concerns the relative timing of trades and quotes contained in public databases when inferring trade initiation and measuring trading costs. Lee and Ready (1991) recommend comparing trade prices to quotes in effect five seconds before the trade report time to allow for delays in trade reporting. However, comparing trade prices to an earlier reference quote might be justified even if trades were reported without delay. Traders are generally concerned with the possibility of adverse price changes between the time of their trade decision and execution. The possibility of adverse pre-trade price movements underlies the recommendation of Perold (1988) to compare trade prices to a benchmark at the time of the trading decision, as well as the common use in the practitioner literature of prior day quotes or trade prices as the benchmark. Comparing trade prices to earlier benchmark quotes potentially captures the effect of systematic price movements ahead of trades. Suppose, for example, that trade report times lag actual trade execution times by five seconds, and that quotes tend to be systematically revised in the fifteen seconds before trades are completed. Then, comparing trade prices to quotes five seconds before the trade report time would be optimal for inferring trade direction. However, comparing trades to quotes in effect twenty seconds before the trade report time would provide a more accurate measure of trade execution cost, including the effect of pre-trade price impact. This study assesses the effect on measured trading costs of comparing trade prices to quotes in effect from zero to thirty seconds prior to the trade report time. Roll (1984) introduces a technique for inferring trade execution costs from the serial covariance of price changes. His method does not require that trades be signed or matched to quotation data. Schultz (2000) reports that the Roll technique provides good estimates of trade execution costs for his sample of Nasdaq stocks. This study assesses whether difficulties arising from the need to sign trades and to match trades with prevailing quotes can be avoided by simply using the Roll

36 - The

Journal of financial transformation

Measuring trade execution costs from public data

method in broader samples that include data from the New York Stock Exchange (NYSE) as well as Nasdaq. These measurement issues are considered in the context of a broad comparison of trade execution costs across the Nasdaq dealer market and the NYSEs specialist-based auction market. The study examines 300 stocks traded on the Nasdaq stock market and 300 matched NYSE stocks during the July to December 1998 period. The results of the study can be summarized as follows. Firstly, measures of rates at which trades are executed at prices better (trades receive price improvement) or worse (trades receive price disimprovement) than the quotations are quite sensitive to whether trade prices are compared to contemporaneous or previous quotes. Comparing trade prices to earlier quotes decreases the percentage of trades that appear to receive price improvement while sharply increasing the percentage of trades that appear to be disimproved. Comparison of these results to those obtained in recent studies [Bacidore et al. (2003), Peterson and Sirri (2003), and Ellis et al. (2000)] that use proprietary order data suggests that contemporaneous comparisons are optimal when assigning trades as buyer or seller-initiated. Secondly, measures of effective trading costs increase with longer time adjustments because quotations systematically rise (fall) in the seconds before customer buy (sell) trades are reported. If the quotation movement prior to the trade report time occurs after the trading decision but before trade execution, then the increase in measured trading costs with a longer time adjustment is real, not illusory. Thirdly, the trading cost estimator due to Roll (1984) provides estimates of effective trading costs on the Nasdaq market that are very similar to those obtained when comparing trade prices to quotations, a result that is consistent with the findings of Schultz (2000). However, the Roll estimates do not correspond as closely to the quote-based estimates for NYSE stocks or for large trades on either market, implying that the difficulties arising from the need to match trades with quotes and to assign trade direction

cannot be avoided by simply using the Roll procedure instead. Finally, despite the sensitivity of trading cost measures to the methodological issues considered here, inference as to whether the dealer market or the auction market provided lower trade execution costs during the sample period is not sensitive. This paper is organized as follows. The next section describes the sample firms and the trade and quote data that are used. That is followed by a description of the measures of trading costs and procedures for inferring trade direction. Subsequently, I will report on the main empirical results, and finally conclude.

The sample
This study focuses on trades in 300 Nasdaq and 300 NYSElisted common stocks during the period July 1 to December 31, 1998. The Nasdaq and NYSE samples rely on the Trade and Quote (TAQ) database, and are matched based on beginningof-sample market capitalization, and include subsets of large, medium, and small capitalization stocks. Sample selection procedures and error filters are described in Bessembinder (2003). The final sample includes 43.5 million trades and 25.4 million quotes. Sample Nasdaq stocks are traded more frequently, averaging 946 trades per stock/day in the full sample, compared to 215 trades for sample NYSE stocks. Quotes are also updated more frequently on Nasdaq, 1094 times per stock/day, compared to 496 on the NYSE. Returns on sample Nasdaq stocks are more volatile: the cross-sectional average standard deviation of daily returns computed from 4 p.m. quotation midpoints is 4.1% for sample Nasdaq stocks, compared to 3.2% for sample NYSE stocks.

Research methods employed

Measures of trading costs This study considers four measures of trade execution costs.2 The first is the quoted bid-ask half-spread, defined as half the

2 Brokerage commission charges are not studied, for two reasons. Firstly, comprehensive data on commission payments is not available. Secondly, individual traders are aware of the commissions they pay, since these are reported directly to them. Trade execution costs are not reported to traders, but must be inferred from trade prices.

37

Measuring trade execution costs from public data

difference between the inside ask quote and the inside bid quote. The average quoted half-spread for each sample stock is computed on a time-weighted basis. Reported are crosssectional averages of the firm means. The quoted half-spread does not accurately measure trading costs when trades are executed at prices away from the quotes. A measure of trading costs that incorporates actual execution prices is the effective bid-ask spread, measured for trade t in stock i as: Effective Half-spreadit = Dit(Pit Mit), where Dit is an indicator variable that equals one for customer buy orders and negative one for customer sell orders, Pit is the price at which the trade is executed, and Mit is the midpoint of the reference bid and ask quotes, viewed as a proxy for the underlying value of the stock. If traders possess private information about security values on average, market prices will tend to rise after customer buys and fall after sells. These price movements reflect what Glosten (1987) refers to as adverse selection costs. Some observers have argued that trading costs should be measured based on trades temporary or non-informational price impact. The realized bid-ask half-spread is such a measure. It is defined as: Realized Half-spreadit = Dit(Pit Mit+n), where Mit+n is midpoint of the quotations in effect n periods after the trade, used as a proxy for the post-trade value of the stock. The post-trade movement in the quote midpoint reflects, on average, the markets assessment of the private information that the trade conveys, referred to as the trades price impact: Price Impactit = Dit(Mit+n Mit) = Effective Half-spreadit Realized Half-spreadit. I use the midpoint of the quotes in effect 30 minutes after the time of the reference quote, or the 4 p.m. quotations during the last half hour of trading, as Mit+n. Effective and realized half-spreads are measured for each trade and averaged across trades for each stock. Reported are simple cross-sectional averages of the firm-by-firm means. Obtaining estimates of effective or realized spreads requires

that trades be assigned as being buyer or seller-initiated, and that trades be matched with prevailing quotes. A measure of average trade execution costs that does not impose these requirements is the Roll (1984) spread, which exploits the movement of trade prices between the markets effective bid and ask prices, and infers the effective width of the bid-ask spread from the magnitude of the negative serial correlation induced in transaction price changes. I use the modification of the Roll estimator suggested by Schultz (2000). Letting Pit denote trade t in stock i and COV denote covariance, the estimator is: Si = [-COV (Pit Pit-1, Pit+1 Pit)]/[1 7/(8(n-1))], where n is the number of trades in the sample. Due to limitations on the number of trades that can be handled by available computing systems, I implement the Roll measure for each firm on a weekly basis, using all trades completed during the week. The final estimate for each stock is obtained as the trade-weighted average of the weekly estimates. Reported are cross-sectional averages of the firm-by-firm estimates. Methods used to assess statistical significance are described in Bessembinder (2003). Algorithms for assigning trade direction The most widely used technique for categorizing trades as buyer or seller-initiated is that recommended by Lee and Ready (1991), henceforth LR. Their algorithm assigns trades completed at prices above (below) the prevailing quote midpoint as customer buys (sells). Trades executed at the quote midpoint are assigned by the tick test, by which trades at a higher (lower) price as compared to the most recent trade at a different price are classified as buys (sells). Several studies, including Finucane (2000), Odders-White (2000), and Ellis et al. (2000) have recently emerged that use specialized datasets containing order information to assess the accuracy of the LR algorithm. These papers indicate that, while the LR algorithm works fairly well overall, classifying about 85% of trades correctly, alternative algorithms may perform better. Ellis et al. (henceforth EMO) pro-

38 - The

Journal of financial transformation

Measuring trade execution costs from public data

pose assigning trades executed at the ask (bid) quote as customer buys (sells), while using the tick test for all other trades, and show that their proposed method outperforms the LR method in their Nasdaq sample. Bessembinder (2003) notes that the results in Finucane (2000) imply that the EMO method would have outperformed the tick test in NYSE-based data as well.

Comparing with earlier quotations decreases the percentage of trades that appear to be executed within the quotes. For the NYSE sample the percentage of trades apparently receiving price improvement decreases from 36.4% for contemporaneous quotes to 34.0% when five seconds are deducted from trade report times to 32.5% when trade times are adjusted by thirty seconds. For the Nasdaq sample the effect is more dramatic, as the rate of price improvement declines from 28.1% with no lag, to 17.2% at a five-second lag, and to 14.5% at a thirty-second lag.

Empirical results
Quoted bid-ask half-spreads Though the main focus of this paper is on methodological issues that arise when computing effective or realized spreads, it is useful to initially examine quoted half-spreads as a basis for comparison. When we compared the average quoted bid-ask half-spreads for a capitalization-matched sample of 300 NYSE and 300 Nasdaq Stocks, during the July to December 1998 period, we found that the full-sample mean quoted half-spread on Nasdaq is 15.7 cents (0.74% of share price), compared to 8.7 cents (0.49% of share price) on the NYSE. There is, however, variation across firm size groups. For large-capitalization stocks differences in average quoted halfspreads across markets are minimal.3 The mean quoted halfspread for large stocks is 7.0 cents (0.21% of share price) on the NYSE versus 9.7 cents (0.24%) of share price on Nasdaq. The mean quoted half-spread for medium-size (small-size) stocks is 8.4 cents (10.8 cents) on the NYSE, compared to 17.7 cents (19.6 cents) on Nasdaq. Price improvement and allowances for trade reporting lags Recognizing that trade reports can be delayed, researchers have generally compared trade prices to quotes reported slightly earlier than the trade report time. The adjustment is typically accomplished by reducing trade report times by a fixed number of seconds before comparing trades to contemporaneous quotes. This section assesses the effect of alternative allowances for trade lags. I first report the percentage of trades that are completed at prices inside and outside the matching quotes, when the matching quote is defined as that in effect from zero to thirty seconds before the trade report time.

In contrast, comparing trade prices to earlier quotes monotonically increases the percentage of trades apparently receiving price disimprovement. For the NYSE sample, the percentage of trades at prices greater than the ask or lower than the bid is 0.6% when comparing trade prices to quotes at the trade report time. The apparent rate of price disimprovement rises to 1.9% if trade times are reduced by five seconds, and to 5.7% at a thirty-second lag. A similar effect is observed for Nasdaq firms, as the percentage of trades apparently completed outside the quotes rises from 5.0% with no trade reporting lag to 6.0% at a five-second lag and to 14.7% with a thirty-second lag. Some indication of the optimal amount by which to adjust trade times for purposes of assessing trade direction can be obtained by assessing the adjustment that leads to the most accurate measures of the proportion of trades executed inside and outside the quotations. This cannot be ascertained using the TAQ data, since the true trade times are not known. However, the percentages reported can be compared to data reported for specialized samples where actual trade times are available. EMO report that 25.2% of the trades in their proprietary Nasdaq sample are completed at prices within the quotes, while 4.3% of trades are executed outside the quotes. Their results, based on actual trade times, correspond most closely to the Nasdaq results obtained here when trades are compared to quotes in effect at trade report times, without any

3 Finding that quoted spreads for large stocks are quite similar across Nasdaq and the NYSE is consistent with the results reported by Weston (2000), who examines only large-capitalization stocks.

39

Measuring trade execution costs from public data

adjustment. Bacidore et al. (2003) use proprietary NYSE data to report that 0.7% of NYSE trades resulting from system market orders are executed at prices outside the quotes in effect at the trade time, while 41.5% of trades are executed at prices within the quotes. Again, the closest correspondence between results obtained here using publicly available data and results obtained when using the proprietary database arises when no allowance for trade reporting lags is made. The results obtained here suggest that trades are best compared to contemporaneous quotations when measuring price improvement rates, and when assigning trades as buyer or seller initiated.

average realized spreads are not sensitive to the adjustment for trade-reporting lags. When measuring the trades average price impact when the time of the benchmark quote is varied, I find that the average price impact for the full sample of NYSE trades is 4.7 cents when quotes 30 minutes after the trade report time are compared to quotes at the trade report time. Measured price impact for NYSE trades increases to 5.1 cents when quotes 30 minutes later are compared to quotes in effect 30 seconds before the trade report time. For the Nasdaq sample, measured price impact increases more dramatically as earlier quotes are used as the reference point. The signed movement in the quote midpoint from the trade report time until 30 minutes later is 4.6 cents, while the change in the quote midpoint from 30 seconds before the trade report time to 30 minutes later is 6.1 cents. These results indicate that quote midpoints move systematically away from trades (rising on buy orders and dropping on sell orders) in the seconds before trades are reported. On the NYSE the average movement in quote midpoints is 0.44 cents during the 30 seconds prior to the trade report. On Nasdaq, quote midpoints move away from the trade by an average 1.51 cents in the 30 seconds before the trade report. Greater adjustments to trade report times result in this pre-trade price impact being included in measures of effective trading costs. Pre-trade price impacts could potentially reflect trade reporting lags, i.e., that the quote is updated during the lag between the actual trade time and the trade report time. However, EMO report that quotes are rarely updated during this interval in their Nasdaq sample, and Peterson and Sirri (2003) report that TAQ trade report times lag actual trade execution times by only two seconds, on average. The other, and more likely, explanation is that quotes are systematically and adversely updated in the seconds before trades are executed. Peterson and Sirri (2003) and Werner (2003) both analyze proprietary NYSE system order data and

Measures of trading costs with varying trade time adjustments

Most studies of trading costs have compared trade prices to earlier quotes both when signing trades and when computing effective and realized bid-ask spreads. It is of interest to know whether measured trading costs are sensitive to the time adjustment used. To assess this issue I next report average effective and realized bid-ask spreads for the present sample, when the time of the reference quote precedes the trade report time between 0 and 30 seconds. The results reported rely on the EMO algorithm to sign trades, without any adjustment to the trade time stamps. I find that the measured effective spreads increase monotonically with the adjustment to trade times. For the sample of NYSE stocks, the measured average effective half-spread increases from 4.9 cents per share with no lag to 5.4 cents per share when trade report times are adjusted by thirty seconds. For the sample of Nasdaq stocks the measured effective half-spread increases from 9.4 cents with no lag to 10.8 cents with a thirty-second lag. T-tests indicate that the increase in measured trading costs is statistically significant for time adjustments of 20 seconds or greater for NYSE stocks and for adjustments of 25 seconds or greater for Nasdaq stocks. In contrast to the effect on effective spreads, measures of

40 - The

Journal of financial transformation

Measuring trade execution costs from public data

provide evidence consistent with this interpretation. Each reports that prices move significantly in the direction of the trades before execution. These adverse average movements in quotes ahead of trades could occur because information about pending orders leaks to the market ahead of trades. It could also reflect larger orders being broken up by brokerage firms into smaller orders sent to market in succession, or it may reflect several traders reacting similarly to common information events, with quotes being revised after the earliest orders are executed. Comparisons of effective and realized spreads to Roll-implied spreads The Roll (1984) technique provides estimates of trade execution costs that do not require that trades be signed or that trades be matched to quotes, so errors from these sources are avoided.4 Here, I assess the extent to which trading cost estimates obtained using the Roll procedure conform with those obtained from the effective and realized bid-ask spread methods. Because the results are of some interest, I report the average Roll half-spread for trades of varying sizes, using the procedure developed by Schultz (2000). Small trades are defined here as those of 1000 shares or less, medium trades are those from 1001 to 9999 shares, and large trades are those of 10,000 or more shares. The results indicate that average Roll-implied half-spreads are uniformly and significantly greater for the Nasdaq stocks than for NYSE stocks. The full sample mean for NYSE stocks is 3.38 cents, compared to 9.88 cents for Nasdaq stocks. Similar differentials are seen for each market-capitalization sub-sample. Comparing spread measures I find that average Roll-implied half-spreads in the present sample are quite similar to average effective half-spreads for Nasdaq stocks, consistent with the findings of Schultz (2000). For the full Nasdaq sample the average effective half-spread is 9.4 cents, while the average Roll-implied half-spread is 9.9 cents. Results for Nasdaq market capitalization sub-samples are similar.

The close correspondence between Roll spreads and effective spreads on the Nasdaq market might be viewed as suggesting that trades need not be signed or matched with quotes at all, as the Roll estimate could simply be used as a general procedure. However, estimated Roll spreads for NYSE stocks do not match effective spread estimates as closely, particularly for small stocks. For small-capitalization NYSE issues the average estimated Roll half-spread is 3.7 cents, compared to an average effective half-spread of 6.1 cents. For the full sample of NYSE stocks the estimated Roll half-spread is 3.4 cents, which is closer to, but still below, the effective half-spread estimate of 4.9 cents. Estimated Roll half-spreads are reasonably uniform across firm size groups, but not across trade sizes. On the NYSE, Roll half-spreads are 9.2 cents for large trades, 4.1 cents for medium trades, and 3.4 cents for small trades, indicating greater price reversals after large trades. A similar pattern is evident for Nasdaq stocks, where average Roll half-spreads are 18.1 cents for large trades, 11.8 cents for medium trades, and 9.7 cents for small trades. The Schultz (2000) result that Roll implied spreads provide good alternative estimates of effective spreads for Nasdaq stocks does not generalize well to trades of varying sizes or to NYSE stocks. One possible explanation for the NYSE result is the presence of price continuity rules that increase the serial dependence of price changes, thereby reducing Roll spread estimates. Regardless of the explanation, the implication is that difficulties in assessing trade execution costs resulting from the need to sign trades and to match trades with prevailing quotes cannot be readily sidestepped in studies involving NYSE stocks by simply using the Roll procedure instead.

Conclusions
This study conducts sensitivity analyses to assess the practical importance of some methodological issues that arise when attempting to measure trade execution costs using the publicly-available quotation and trade price databases. The publicly available databases contain trade report times, but not

4 However, the Roll method relies on a set of restrictive assumptions. Among these are the assumptions that the spread width is constant over time, and that trades do not convey private information about value, i.e. that trades average price impact is zero.

41

Measuring trade execution costs from public data

order submission or trade execution times, and do not indicate whether trades are buyer or seller-initiated. Adjusting trade report times as an allowance for possible reporting lags decreases the percentage of trades that appear to be executed within the quotes while increasing the percentage of trades that appear to be executed outside the quotes. Comparing results here with those obtained using proprietary databases that include accurate trade times and order data, it appears that trade direction and rates of price improvement are best assessed when making no adjustment for trade report lags. Estimated trading costs increase if trade prices are compared to earlier rather than contemporaneous quotes, reflecting adverse quote movements prior to trade report times. If these adverse quote movements occur after order submission but before trade execution then they comprise a real cost to traders that will not be captured when trade prices are compared to quotes in effect at trade report times. On balance, the results obtained here support recommendations to use the EMO technique in preference to the LR method to sign trades, implement the EMO technique on the basis of contemporaneous rather than earlier quotations, and use quotation midpoints in effect somewhat prior to the trade report time as the benchmark quote when measuring effective bid-ask spreads. This last recommendation is similar in spirit to the use of the quote midpoint at the time of the trading decision as the reference point [Perold (1988)], in order to include costs stemming from pre trade price impact.

References
Bacidore, J., K. Ross, and G. Sofianos, 2003 Quantifying market order execution quality at the New York Stock Exchange, Journal of Financial Markets, 6, 281-307 Bessembinder, H. 2003, Issues in assessing trade execution costs, Journal of Financial Markets, 6, 233-258 Ellis, K., R. Michaely, and M. OHara, 2000, The accuracy of trade classification rules: Evidence from Nasdaq, Journal of Financial and Quantitative Analysis, 35, 529-552 Finucane, T., 2000, A direct test of methods for inferring trade direction from intraday data, Journal of Financial and Quantitative Analysis, 35, 553-576 Glosten, L. 1987, Components of the bid-ask spread and the statistical properties of transactions prices, Journal of Finance, 42, 1293-1307 Lee, C., and M. Ready, 1991, Inferring trade direction from intraday data, Journal of Finance, 46, 733-746 Odders-White, E., 2000, On the occurrence and consequences of inaccurate trade classification, Journal of Financial Markets, 3, 259-286 Perold, A. 1988, The implementation shortfall, Journal of Portfolio Management, 14, 4-9 Peterson, M. and E. Sirri, 2003, Evaluation of the biases in execution cost estimates using trade and quote data, Journal of Financial Markets, 6, 259-280 Roll, R., 1984, A simple measure of the effective bid-ask spread in an efficient market, Journal of Finance 39, 1127-1139 Schultz, P., 2000, Regulatory and legal pressures and the costs of Nasdaq trading, Review of Financial Studies, 13, 917-958 Werner, I., 2003, NYSE order flow, spreads, and information Execution costs, Journal of Financial Markets, 6, 309-335

42 - The

Journal of financial transformation

Services

Best execution regulation: From orders to markets

Jonathan Macey
Sam Harris Professor of Corporate Law, Corporate Finance, and Securities Law, Yale Law School

Maureen OHara
Robert W. Purcell Professor of Management, Professor of Finance, Johnson Graduate School of Management, Cornell University

Abstract
This article traces the evolution of the legal obligation of best execution, from its origin as an application of the common law of agency, to its current incarnation as a rule implemented by stock exchanges and regulators as a means to regulate market structure. We show that, over time, the legal duty of best execution has shifted from a duty owed by brokers to their counterparties to a duty owed by brokers to the markets in general. From an economic perspective, this shift reveals a fundamental inconsistency between the historical legal obligation of brokers to give their customers best execution and the modern goal of securities regulators to ensure competitive capital markets. This paper reveals an as yet unrecognized policy trade-off between the goal of giving individual traders best execution on every trade, particularly where best execution is defined as best price, and the goal of promoting vigorous competition among trading venues. The article then turns to the question of which institution is most likely to provide optimal rules of best execution. We show that the best place to locate the decision-making authority over trading venue is the issuing firm.

43

Best execution regulation: From orders to markets

Among the clearest rules in U.S. securities law is the duty that brokers have to seek the best execution that is reasonably available for its customers orders whenever a broker executes an order for a customer. The legal duty of best execution
1

order flow and other arrangements that permit trading venues to contract ex-ante for the right to take and supply liquidity are legally impermissible. Applying the legal duty of best execution on an ex-ante market basis necessarily involves transfers of wealth from certain cohorts of traders to others. For example, a broker-dealer firm may decide that allocating all trades to a particular venue constitutes best execution because it lowers the average cost of trading for all customers, but such a decision is likely to raise costs for some clients, even as it lowers costs for others. In particular, exante preferencing arrangements that direct trades to a particular venue may benefit larger, block traders at the expense of smaller retail traders who would receive price improvement if their trades were shopped at an exchange rather than executed automatically. We will subsequently reconcile these conflicting conceptions of the legal duty of best execution. In our view, the problem with the current orientation of the policy discussion is that it has focused on the narrow, yet unanswerable, question of which venue provides traders with best execution. This is because, as we have pointed out in previous work, the term best execution does not connote a single execution attribute, such as price, but rather attaches to a vector of execution components. These certainly include trade price, but they also involve the timing of trades the trading mechanism used, the commission charged, and even the trading strategy employed. Such multi-faceted concerns have long been a feature of institutional trade execution, but their emergence now even in retail trading reflects the reality that markets are a great deal more competitive and complex than in times past [Macey and OHara (1997)]. Clearly, for example, it makes no sense to employ the same, or even a similar, legal definition of the duty of best execution for large institutional traders as for small retail traders. Institutional traders concerns with respect to the issue of best execution are focused around the impact of their trading on the average price on which their trades will execute. By contrast, retail traders are concerned

is derived from the common law fiduciary duty of loyalty, which requires brokers to maintain undivided allegiance to the interests of their clients and bars brokers, who act as agents on behalf of their customers, from using their position in any manner that allows him/her to garner a personal profit or a personal advantage. Building on our earlier work [Macey and OHara (1997)], this article examines the legal duty of best execution from an economics perspective. In the first section of the article, we examine the legal origins of the duty of best execution. We observe that, over time, there has been a shift from a case-by-case approach to the duty to a generalized regulatory approach that looks at the overall standards applied by brokerage firms when executing trades to determine whether the duty of best execution has been violated. Put another way, gradually the legal duty of best execution has shifted from a duty owed by brokers to their counterparties to a duty owed by brokers to the markets in general. In the following section we examine the most recent SEC initiatives to define the concept of best execution. We show that there is a fundamental inconsistency between the original goals of the historical legal obligation of brokers to give their customers best execution and the modern goals of securities regulators, which is to ensure competitive capital markets. The basic problem involves the policy choice that must be made between applying the duty of best execution on a caseby-case regulation or on a broader basis. Both approaches have problems (costs). Applying the legal duty of best execution on a case-by-case basis, which necessarily requires that brokers determine which venue is best for each order individually after it is received, can reduce competition among trading venues as market practices such as payment for

http://www.sec.gov/answers/bestex.htm (January 4, 2005)

44 - The

Journal of financial transformation

Best execution regulation: From orders to markets

with the transaction costs of their trading, since, unlike institutional traders, retail trades will not influence the underlying price of the securities being traded. In the next section we examine the alternative institutions most likely to generate optimal rules regarding best execution. The most likely choices include, the trading venues themselves (in their capacities as self-regulatory organizations), the government (in its administrative capacity as the Securities Exchange Commission), individual shareholders and institutions (in their capacities as principals to securities transactions), and issuing firms (in their capacity as issuers of securities who contract with investors via their articles of incorporation). We develop the argument that, for a variety of institutional and incentive-based reasons, while no institution is ideal, by far the best place to locate the decision-making authority over trading venue is the issuing firm. A conclusion follows.

buy or sell in such market so that the resultant price to the customer is as favorable as possible under prevailing market conditions.3 Thus, it is clear that best execution does not necessarily imply best price. (B)rokers have not been held...to an absolute requirement of achieving the most favorable price on each order... (W)hat has been required is that the broker endeavor, using due diligence, to obtain the best execution possible given all the facts and circumstances.4 We have observed previously that the legal requirement appears to require only that the broker tried to obtain the best price, not that he actually obtained the best price in fact [Macey and OHara (1997)]. For the purposes of the analysis here, what is important is the fact that the legal inquiry is particularized, that is, the legal inquiry is carried out on a case-by case basis. This is what is meant by the phrase given all the facts and circumstances. Each trade, under the common law approach is to be evaluated on a case-by-case basis. Such a narrow focus is relevant because the fiduciary principles of care and loyalty, which provide the historical underpinnings for the legal duty of best execution, are particularized duties, based on the common law of agency. These rules, in turn, are contractual in nature; they flow from brokers to their clients on an individual basis. Consequently, it is no defense against a customers claim that he or she was not given best execution for a broker to respond that some other customer, such as the customer on the other side of the transaction, received best execution. It also is no defense to respond that the marketplace in general was made better or more competitive as a result of a particular instance of best execution being denied in a particular circumstance. The duty of best execution is static not dynamic, what matters is this trade, not trades in the future.

The development of the legal obligation of best execution

The duty of best execution is firmly grounded in common law principle of agency, from which the common law fiduciary duties of care and loyalty are derived. A broker-dealers duty to seek to obtain the best execution of customer orders derives from the common law (duty of) agent(t) loyalty, which obligates an agent to act exclusively in the principals best interest...(the agent also) is under a duty to exercise reasonable care to obtain the most advantageous terms for the customer.2 Consistent with these common law duties, NYSE and the NASD both have rules requiring member firms to execute all customers orders at the best available prices. These rules require only that the price the customer receives be the best possible under the circumstances, [i]n any transaction for or with a customer, a member...shall use reasonable diligence to ascertain the best inter-dealer market for the subject security and

2 In re Merrill Lynch, 911 F. Supp 754, 760 -769 (1995) (cited in In re E.F. Hutton & Co., Securities Exchange Act Release No. 25887, (1988 Transfer Binder) Fed. Sec. L. Rep. (CCH) 84303, 89326 at 89326 (July 6, 1988); Restatement (Second) of Agency 1 (1957)).Market 2000: An Examination of Current Equity Market Developments, Division of Market Regulation of the SEC, Study V, 1994 SEC LEXIS 135, at *5-6 (1994)(hereinafter Market 2000); 15 U.S.C. 78k-1(a)(1)(c)(iv) (1994). Restatement of Agency 387, 424 (1958).).

3 New York Stock Exchange Rule 123A.41, 2 NYSE Guide (CCH) P 2123A; NASD Rules of Fair Practice, NASD Manual (CCH), Art. III, Sec. 1, 2151.03. 4 Merrill Lynch at 770 (citing Second Report on Bank Securities Activities: Comparative Regulatory Framework Regarding Brokerage-Type Services 97-98, n. 233 (Feb. 3, 1977), reprinted in H.R. Rep. No. 145, 95th Cong., 1st Sess. 2333.).

45

Best execution regulation: From orders to markets

To illustrate why this dynamic issue is important, suppose that a trade allocated to a particular ECN has a very low probability of being executed at a price better than the best exchange posted quote or the national best bid best offer (NBBO) price,5 while a trade allocated to the New York Stock Exchange has a reasonable (and much higher) probability of executing at a price better than the NBBO. It would not be a valid defense against the claim that the legal duty of best execution has been violated for a broker to say that to require that each trades be deployed to the NYSE (or any other particular trading venue) would make that venue a monopoly, and thus would result in other (future) traders receiving inferior prices due to the monopoly pricing power of the venue going forward. The duty of loyalty encompasses more than just price. This legal duty also addresses issues of conflicts of interest, and confidentiality. It is improper, for example, for a broker to use its position to advance a personal interest, say by front-running a customers order, by using the information embedded in the order for the advantage of other customers or traders associated with the firm. For example, on December 16, 2004, Knight Securities, L.P. settled a lawsuit brought against it by the Securities and Exchange Commission and agreed to pay U.S.$79 million in disgorgement and penalties for defrauding its institutional customers by extracting excessive profits out of such customers orders and failing to meet the firms duty to provide best execution to the institutions that placed those orders.6 The SEC found that between January 1999 and November 2000, Knight which was, at the time, one of the largest market-makers on the NASDAQ earned over U.S.$41 million in illegal profits by failing to provide best execution to its institutional customers. Specifically, Knight defrauded its institutional customers by extracting excessive profits out of its customers orders while failing to meet the firms duty to provide best execution to the institutions that placed those

orders. During the relevant time period, Knight, upon receipt of an institutional customer order, would acquire a substantial position (in the same security) in its own proprietary account. Rather than fill the order promptly on terms most favorable to the customer, Knight would wait to see if its proprietary position increased in value during the trading day. When the prevailing market price for the stock moved significantly away from Knights acquisition cost, Knight then filled the customers order and pocketed the difference as its profit on the transaction. For example, on April 4, 2000, Knight received a customer market not-held order to purchase 250,000 shares of Applied Micro Circuits Corporation (AMCC). Over the next 18 minutes, Knight acquired 147,000 shares of AMCC at an average cost of U.S.$91 per share. Rather than promptly selling the stock to the institutional customer at Knights cost (plus a reasonable profit), Knight sold the AMCC stock to the customer over a period of time at an average profit of approximately U.S.$2 per share. On the entire AMCC order, Knight realized a profit of over U.S.$1.1 million (or an average of U.S.$3.94 per share). When the market moved unfavorably in relation to the position Knight had established to fill the institutional customers order, Knight executed its remaining position in the order to the customer at prices that still generated a profit for Knight. By engaging in these trading practices, Knight extracted enormous profits as high as U.S.$9 per share by executing transactions that involved effectively no risk to Knight. Consequently, Knight improperly realized tens of millions of dollars in excessive per share profits from its institutional customers. During the same period, Knight failed reasonably to supervise Knights former leading sales trader who was primarily responsible for Knights fraudulent trading. Knights senior management not only allowed the leading sales trader to be directly supervised by his brother, but also permitted the two to evenly split the profits generated by the leading sales trad-

46

5 SEC rules require that any Nasdaq dealer that wishes to hold itself out as a market maker in a particular security must continually report the prices at which it is willing to buy and sell the security. Such market makers must be willing to honor their reported bid and offered prices up to the quotation size, i.e. the number of shares that the market maker reports as being willing to buy at the bid side or sell at the offered side. 17 C.F.R. Section 240.11Ac1-1(c)(1)(2), and (10). Nasdaq is required by

the SEC to collect, process and make available... the highest bid and lowest offer communicated... by each member... acting in the capacity of an OTC market maker for each reported security. 17 C.F.R. Section 240.11Ac1-1(b)(1)(ii). This latter quotation has come to be known as the NBBO (National Best Bid and Offer). 6 http://www.sec.gov/news/press/2004-173.htm

Best execution regulation: From orders to markets

ers trading with institutional customers. The two brothers relationship, their positions, and responsibilities within the firm, and their profit-sharing arrangement created an inherent conflict of interest that contributed to a substantial breakdown in the supervision over the Leading Sales Trader. Between 2000 and 2001, Knight failed reasonably to supervise Knights institutional sales traders while they were systematically misusing Automated Confirmation Transaction Service (ACT) trade modifiers. Knight had no written procedures, no adequate systems in place, and no supervisory personnel to prevent Knights sales traders from consistently misusing the modifiers over a two-year period. The misuse of ACT modifiers led to the sales traders recording inaccurate execution times on the firms trading blotters, in violation of the books and records provisions of the federal securities laws. By misusing the ACT trade modifiers, Knight sales traders were able to improperly input trades into Knights trading system at prices that were different from the inside market at the time the trades were reported. The repeated misuse of the ACT modifiers also limited the ability of Knights institutional customers to detect the fact that Knight was extracting excessive profits at their expense. By misusing ACT trade modifiers, Knights sales traders avoided limit order protection protocols and filled more profitable institutional not held. Without admitting or denying the SECs findings, Knight, (now known as Knight Equity Markets, L.P.), agreed to pay more than U.S.$41 million in disgorgement of illegal profits, over U.S.$13 million in prejudgment interest and U.S.$12.5 million in civil penalties. The settlement also required Knight to pay an additional U.S.$12.5 million in fines to settle a parallel NASD proceeding. In the settlement Knight agreed to cease and desist from committing or causing any future violations of the broker-dealer anti-fraud and books and records provisions of the Exchange Act, and was censured for its failure to supervise

its institutional sales traders. Knight also agreed to hire an independent compliance consultant to conduct a comprehensive review of its policies and procedures with respect to best execution obligations, and other regulatory obligations as well as the firms supervisory and compliance structure. In announcing the settlement, Stephen M. Cutler, the SECs Director of Enforcement observed that customers have a right to expect they are getting fair treatment when they entrust their broker with orders to buy and sell securities. That expectation is betrayed when the broker handling the orders puts its own financial interests ahead of its customers interests. The Knight settlement illustrates the severe conflicts of interest that exist whenever a broker or his firm has a financial interest in a transaction to which the organization may be a party, such as when a customers order is executed in-house or when the brokerage firm also acts in a dealer capacity. In response to these scandals, the SEC enacted new rules requiring disclosure of order execution policies and order routing protocols. Now, dealer firms are required to disclose each month precisely how they execute trades. Brokers who route customer orders must disclose quarterly the identity of each market center that receives a meaningful percentage of their orders. While there is no requirement that customers be told where their individual orders were executed, brokers must divulge this information if asked.7 Similarly, confidentiality obligations require brokers to protect from disclosure the trading plans and strategies of their clients. Thus, there are conflicts between brokers obligations to the firms they work for when those firms are engaged in underwriting or trading activities for their own accounts. The Knight scandal, along with other front-running scandals, such as those that periodically plague the New York and American stock exchanges, should be regarded as old fashioned best execution scandals. The wrong-doing arises

7 Goodman, B., 2002, Are things Kosher at your broker? The Street.Com, June 7, http://www.thestreet.com/funds/beverlygoodman/10026056.html.

47

Best execution regulation: From orders to markets

because it is extremely difficult for any investor, even sophisticated institutional investors, to monitor the agents charged with executing their orders. Market discipline works poorly in this setting because it is impossible for the client to know if the price received is due to agency problems or to stochastic market fluctuations beyond the brokers control. Just as the customer finds it difficult to monitor and control this behavior ex-ante, the SEC has problems formulating rules to deal with the problem ex-post. Another critical point that has been ignored in the current discussions of the legal duty of best execution is the effect of the legal rules regarding best execution on market structure. From a market structure perspective, use of the traditional, case-by-case approach to best execution poses a collective action problem. Individual traders acting rationally and in pursuit of their own self-interests may direct order flow in such a way as to damage the quality of secondary markets generally by generating trading patterns that give a particular trading venue a monopoly over the provision of secondary market liquidity. This puts regulators in a very difficult position. To maintain high quality markets, securities markets regulators must insure that there is robust competition among rival trading venues, or else spreads will widen and service will deteriorate. In the long-run, this will lead to less liquidity and to higher capital costs for investors. The common law approach to best execution, however, does not allow for this consideration. Another problem with the traditional, common law approach is that it ignores the fact that, once the decision to buy or sell has been made, the gains from improvements on the market bid-asked spread are zero sum. Gains to the seller from receiving more than the best extant bid translate directly into losses for the buyer, who ends up paying more. What then is best execution for firms executing in-house agency crosses or for trading venues that make continuous two-sided markets? This

point is particularly strong under the traditional view of best execution, which would require that customers placing orders with brokers receive the best price available, not just a price that is generally good or better than available in other markets [Macey and OHara (1997)]. But even where trading venues are highly efficient, and offer customers rebates on their trades (payment for order flow), such markets do not, from a technical legal perspective, necessarily offer best execution merely because the execution is good, or even excellent. The legal requirement is that the execution be the best of all available alternatives. Adding yet another layer of complication is the conflict between the interests of retail customers and institutional investors. Even at the system-wide level this conflict stymies the search for the illusive grail of best execution. This controversy manifests itself in the current debate over the rights and responsibilities of customers and markets regarding whether customers and/or markets can have their orders tradethrough the prior, better orders of other customers. A tradethrough is the execution of an order in a market at a price that is inferior to a price displayed in another market.8 Tradethrough rules bar traders from trading at worse prices in faster electronic markets when there is a better quote in the slower, exchange market.9 From the traditional legal perspective of best execution, the very concept that an order can execute at a price inferior to the price displayed in another market poses a problem, particularly where the transaction is consummated without the customers knowledge or consent. But the problem also exists where the trade-through rule prevents a customer from executing an order at an inferior price on a faster electronic market in order to obtain better overall execution for a large block trade. For example, if the best bid anywhere for a stock is U.S.$100, a large block seller, under the current incarnation of the trade-through rule, must execute the trade at that price, even if the bid were for only 1000 shares, and the

8 Junius W. Peake , Regulation NMS: Pools or an Ocean of Liquidity The Handbook of World Stock, Derivative and Commodity Exchanges. http://www.exchange-handbook.co.uk/articles_story.cfm?id=47680 (accessed January 7, 2005) 9 New York Stock Exchange, The Exchange, V11, No. 6/7, p 1-2 (July 2004).

48 - The

Journal of financial transformation

Best execution regulation: From orders to markets

seller would prefer to sell the entire 100,000 share block into an inferior bid at 99.75. Of course, if the block trader were permitted to elect to trade-through the superior U.S.$100 bid, as the SEC currently is proposing,
10

block trader filling its entire order, raising execution costs. Moreover, the longer the delay, the more difficult it will be for the block trader to keep its identity and trading plans anonymous. Once the institutional identity and trading plans become known, rational traders will enter the market in advance of some of the block traders orders, and this will deprive the block trader of best execution. The SEC has recently proposed, but not enacted, rules that will allow traders to choose speed of execution over best price.15 The proposed rule would limit the scope of the tradethrough rule in two important respects. Firstly, it would enable those placing orders to opt-out of the trade-through rule if the trader is able to make an informed decision to elect speed of execution over best price.16 Secondly, in automated markets, such as ECNs, where an order can be instantly filled by a computer system, the trade-through rule would not apply within a certain de minimis range of price discrepancy between the order price and the best bid or offered price in the system. This range would be from one to five cents per share, based on the total share price to be tradethrough-able. The NYSE has vigorously opposed relaxing the trade-through rule, arguing that trades should have to go the market posting the best price, which traditionally has been the NYSE. However, there is one aspect of the rule that the NYSE does support. The SEC is also proposing a uniform trade-through rule for all NMS market centers that, subject to the exceptions just described, would require a market center to establish, maintain, and enforce policies and procedures reasonably designed to prevent trading venues from executing orders at a price that is inferior to a price displayed in another market. This would extend the trade-through rule beyond the NYSE and certain other exchanges to the NASD, ECNs, and other market centers.

and consummate the

transaction at U.S.$99.75, the retail trader who entered the U.S.$100 bid might not obtain best execution, should the market move higher before the trade can be executed.11

The current conflict: Trading through best execution?

The trade-through rule that is now in place was promulgated pursuant to the Intermarket Trading System (ITS) Plan, and
12

it dictates that no market participating in this plan13 can trade at a price inferior to a price displayed in another market. The trade-through rule dictates that a trading venue receiving an order must match a better price available elsewhere or route the order to the other market for execution. At the market level, trade-through protection originally was intended to help enforce the general obligation of best execution discussed in the preceding section. A best execution problem arises on the buy-side when a specialist makes a risk-free profit by buying at a lower-pre-existing price to fill the bid or selling at a higher pre-existing offer price to fill an order. The trade-through rule was designed to promote best execution of customer orders by preventing specialists and other market makers from executing orders at inferior prices when superior prices are available elsewhere. Under the trade-through rule, trades must be executed at the best price, which is defined as the current best price offer regardless of the size order. Controversy has arisen because
14

many block traders would prefer to execute their trades automatically at inferior prices immediately if they can execute a trade that covers their entire block. This is because the delay itself poses the risk that the market will move prior to the

10 See Part I of SEC Release 34-47013, Final Rule: Repeal of the Trade-Through Disclosure Rules for Options, (2004), at http://www.sec.gov/rules/final/3447013.htm. The proposed new rules are part of Reg NMS, a proposal to change the current operation of the national market system. For details of Regulation NMS see http://www.sec.gov/news/press/2004-22.htm. For an analysis of these proposed changes see OHara [2004]. 11 Of course the large block trader might elect not to execute the block sale at all if it must fill the U.S.$100 order before hitting the larger bid at U.S.$99.75. This is one of the complexities in implementing a strict rule of best execution. 12 Section 11A(a)(2) was adopted by the Securities Acts Amendments of 1975 (1975 Amendments). Pub. L. No. 94-29 (June 4, 1975). 13 Current signatories to the ITS Plan include the American Stock Exchange LLC,

Boston Stock Exchange, Inc., Chicago Board Options Exchange, Inc., Chicago Stock Exchange, Cincinnati Stock Exchange, NASD, New York Stock Exchange, Inc., Pacific Exchange, Inc., and Philadelphia Stock Exchange, Inc. 14 Information on the trade-through rule can be found in Part I of SEC Release 3447013, Final Rule: Repeal of the Trade-Through Disclosure Rules for Options, (2004), at http://www.sec.gov/rules/final/34-47013.htm. 15 These new rules are part of Reg NMS, a proposal to change the current operation of the national market system. For details of Regulation NMS see http://www.sec.gov/news/press/2004-22.htm. For an analysis of these proposed changes see OHara [2004] 16 SEC Final Rule, supra note 45.

49

Best execution regulation: From orders to markets

As Junius Peake has observed, the trade-through rule, and its proposed opt-out exception, is by far the most controversial piece of proposed Regulation NMS because it will have a defining effect on the structure of U.S. equity markets. The New York Stock Exchange, which has been the principal beneficiary of the trade-through rule, is fighting tooth and nail to maintain the status-quo. Its competitors, and would-be competitors, such as ECNs, believe the present iteration of the trade-through rule to be old fashioned because it frequently requires manual intervention before an order can be executed, and gives an unfair advantage to NYSE specialists, who have time to make up their minds whether or not to participate in a trade.17 In our view, however, the trade-through rule must be evaluated in the context of all of the other rules governing best execution, and here our point is purely positive. The SEC has engineered a seismic shift in the legal conception of the duty of best execution from its traditional common law grounding in fiduciary duties and highly individualized contractual arrangements between customers and brokers into a generalized duty to promote competitive markets, regardless of the effects on individual traders. This change can be seen most clearly in the SECs own articulation of the general rules regarding the execution of trades by brokers on behalf of customers.18 Under the SECs conception of the rules regarding order execution, there is, stunningly, no duty of best execution. Instead, brokers have a choice of markets in which to execute trades. Thus, for stocks listed on an exchange, such as the New York Stock Exchange, it appears that brokers are free to direct the order to that exchange, to another exchange (such as a regional exchange), to a third market maker willing to buy or sell a stock listed on an exchange at publicly quoted prices, to an ECN that automatically matches buy and sell orders at specified prices, or even to another division of the brokerage firm receiving the customers order, such that the order is

filled from the brokers firms own inventory. The SEC is also, of course, aware that regional exchanges and third market makers make payments to attract order flow, both for NYSE stocks and for Nasdaq and other over-thecounter stocks. The SEC does not have the power to overturn or even to modify the state common law rules of fiduciary duty from which the general law of agency and the duties of best execution in particular are derived. However, the SEC has re-interpreted the duty of best execution as a general duty to the markets, rather than as a particularlized contractual obligation between market participants. The SEC observes, for example, that many firms use automated systems to handle the orders they receive from their customers. In deciding how to execute orders, your broker has a duty to seek the best execution that is reasonably available for its customers orders. That means your broker must evaluate the orders it receives from all customers in the aggregate and periodically assess which competing markets, market makers, or ECNs offer the most favorable terms of execution. The opportunity for price improvement which is the opportunity, but not the guarantee, for an order to be executed at a better price than what is currently quoted publicly is an important factor a broker should consider in executing its customers orders. Other factors include the speed and the likelihood of execution. Of course, the additional time it takes some markets to execute orders may result in your getting a worse price than the current quote especially in a fast-moving market. So, your broker is required to consider whether there is a trade-off between providing its customers orders with the possibility but not the guarantee of better prices and the extra time it may take to do so. Here, the SEC reveals its position that a brokers is, in its view, permitted to evaluate the orders it receives from its customers

17 Peake, Regulation NMS: Pools or an Ocean of Liquidity (2004). 18 http://www.sec.gov/investor/pubs/tradexec.htm

50 - The

Journal of financial transformation

Best execution regulation: From orders to markets

in the aggregate in order to determine where trading should be directed.19 The SECs statement on best execution also reflects the view that even that aggregate determination need be made only periodically. By contrast, under the traditional common law approach to best execution discussed in the previous section, each order must be evaluated individually, not in the aggregate, and the determination of what constitutes best execution must be made at the time each and every trade is made, not periodically. From an economic perspective, there are two implications to this analysis. Firstly, while wealth transfers among various investors were clearly impermissible under the traditional common law duty of best execution, wealth transfers are permissible under the SECs current approach. Secondly, while the traditional common law approach to best execution ignored the implications of the duty of best execution on market structure, the SECs approach is focused primarily on issues of market structure. This is not surprising, given the fact that the SECs mandate is to maintain the integrity of the markets and to promote fair competition among market participants in the context of the 1975 National Market System legislation.20 It clearly is the case that there are a few remaining vestiges of the traditional common law approach to best execution. For example, customers may, if they specifically request to do so, direct their trades to a particular exchange, market maker, ECN, or other venue, although, brokers are free to charge for that service or to decline to execute the trade unless they are given discretion about where to execute the transaction. Some brokers offer certain customers the ability to direct orders in Nasdaq stocks to the market maker or ECN of their choice when they place their orders. Customers are also permitted to obtain information about where their trades have been routed by their brokers during the previous six months, and to obtain information about their brokers policies on payment for order flow, internalization, and other trade routing practices.

It is tempting to view customers ability to direct order flow as a complete solution to the best execution problem. In theory at least, the problem could be solved simply by requiring brokers to ask customers where they want their orders sent and how they want them traded. Yet, this approach ignores a fundamental problem that the customer hires the broker to solve, and that is how best to trade his stock? As we have argued above, it is agency problems that give rise to the need for best execution duties.

Our proposal: An inquiry into institutional competence

As the recent spate of scandals illustrate, as with other principal-agency relationships, there are significant conflicts of interest associated with the broker-client relationship. Whether order flow is internalized, or not, brokers can benefit personally at the expense of customers when they receive orders. The private contracting model does not offer a good solution to this problem for two major reasons. Firstly, it is extremely costly for most traders, particularly retail traders, to develop the information and expertise necessary to permit them to make an informed choice about trading venue. Secondly, and perhaps more importantly, individual traders face a collective action problem when deciding where to allocate their trades. This is because the decision made individually by each market participant about which trading venue is best may be the same at a particular point in time, giving that venue a monopoly. If individual traders and regulators are unable to make the socially efficient, optimal decisions about best execution, where should the decision-making authority over trading be allocated? The remaining options are the trading venues themselves and the issuing firms.

The trading venues

Clearly the trading venues are conflicted with respect to this issue. After all, it is the venues themselves that stand most to

19 The same language is used in the SECs Order Execution Obligations, Exchange Act Release No 34-37619A, 60 Fed. Reg. 48290, at 48323, where the SEC uses the same phrasing when opining that: [b]roker-dealers routing orders for automatic execution must periodically assess the quality of competing markets to assure that order flow is directed to markets providing the most beneficial terms for their customers orders. Broker-dealers deciding where to route or execute small cus-

tomer orders must carefully examine the extent to which the order flow would be afforded better terms if executed in a market or with a market maker offering price-improvement opportunities. In making this evaluation the broker-dealer must regularly and rigorously examine execution quality likely to be obtained from different markets or market makers trading a security. 20 Section 11A(a)(1), National Market System legislation, 1975.

51

Best execution regulation: From orders to markets

gain from a decision that trading on their venue is consistent with the legal obligations of best execution. We pause only to observe that, for years off-board trading restrictions and prohibitions on delisting gave effective control over the decision about where to trade to the organized exchanges, first by prohibiting exchange members from trading in listed securities off the floor of an exchange, and then by making delisting virtually impossible. With the relaxation of these rules in recent years, the decisional authority lies uneasily among brokers, bureaucrats, and the traders themselves. We also observe that, at least for the New York Stock Exchange, the tradethrough rules are perhaps the last vestige of venue-based trading restrictions, as the NYSEs trade-through rule, which prevents trade-throughs in New York listed stocks, forces the execution of the vast majority of trades in NYSE listed firms onto the floor of the big board.

Thus, issuing firms, at the time of an initial public offering have strong incentives to establish the rules of best execution that maximize value for all shareholders. Unlike the case-bycase decisions that individuals make when they make specific trades, firms must make decisions about best execution exante, or essentially before the shares begin trading. And these decisions must be made on an aggregate basis. But intriguingly, any inefficiency in the trading restrictions imposed by the issuer will be reflected in the share price that investors must pay. This means that issuing firms, and their owners and venture capital providers, not subsequent investors, bear the costs associated with any inefficiency in the best execution regime established by the firm. If issuing firms, rather than bureaucrats, individual traders, or competing trading venues are better suited to establishing what the rules of best execution should be, then we need to consider the twin questions of how this legal regime should be effectuated, and what limits, if any, should be placed on firms powers to constrain shareholders secondary market trading practices.

The listing firms

So, by process of elimination, we are left with the listing firms themselves. We recognize that principal-agent problems also plague the relationship between shareholders and the firms in which they exist. But the severity of the agency costs between firms and investors is not constant over time. For example, when firms are in play, that is, subject to hostile acquisition, managers have strong incentives to maximize firm value, so that their firms share prices will be prohibitively high. Most importantly, from our perspective, when a firms shares are initially sold to the public the firm has extremely strong incentives to maximize the liquidity characteristics of the shares it is selling. After all, the whole point of going public for the prior owners and investors, particularly the firms founding entrepreneur, private equity owners, merchant bankers, and venture capitalists is to gain liquidity. The quest for liquidity is what induces firms to incur the substantial costs and potential liability of registering their securities with the SEC and going public (as opposed to raising capital through private placement or via a private sale of control or other strategy).

Implementation
We propose that issuing firms decide for themselves, and draft or amend their corporate charters (articles of incorporation) to establish the regime of best execution that best serves the interests of their shareholding population. This, in our view, can be done through the mechanism of drafting share transfer restrictions that define what, if any, restrictions should be placed on where firms shares are traded. As the name implies, a share transfer restriction is a provision in the articles of incorporation of a company that restricts, in some way or another, the ability of shareholders to transfer their shares to other investors. For example, share transfer restriction can require that the firms general counsel, or its board, (or, if the firm is small enough, the shareholder) gives permission for shareholders to sell. Alternatively, firms might

52 - The

Journal of financial transformation

Best execution regulation: From orders to markets

impose no share transfer restrictions, or require that trading be restricted to a particular exchange or constellation of exchanges and ECNs. Changes to the best execution rules contained in a firms corporate charter could be amended from time-to-time as technology changes the nature of the trading environment. In general, stock certificates, and other more modern indicia of share ownership, are regarded as personal property, and are subject to the traditional, common law rule that there be no unreasonable restraint on alienation.21 Share transfer restrictions, however, are widely used by corporations and serve a number of valid purposes, including ensuring that a corporation will continue to satisfy certain regulatory requirements, such as Subchapter S of the Internal Revenue Code, which grants preferred (pass-through) tax treatment to corporations but requires them to have no more than 35 shareholders, or ensuring that the company retains exemptions from the registration requirements of the Securities Act of 1933 that prohibit the public offering of unregistered securities. In addition, share transfer restrictions are commonly used to maintain a familys control over a particular corporation, or to maintain the status-quo ownership structure among shareholders, or to permit shareholders in closely-held corporations to regulate the identity of new investors.22 While the general rule is that shares of stock are freely transferable, share transfer restrictions are valid in the vast majority of states, so long as the restrictions they impose are reasonable under the circumstance. The modern rule of share
23

Limits
As Amihud and Mendelson (1996) have correctly observed, the SEC has taken a dim view of issuers efforts to restrict the trading venue of their securities, once those securities have been issued and are already being traded. In particular, the SEC routinely grants so-called unlisted trading privileges to securities exchanges. Unlisted trading privileges, as the name implies, are simply rights that give a particular trading venue the privilege of trading a security in situations in which the issuer of the securities has not asked permission for its securities to be traded in that venue. Under the Unlisted Trading Privileges Act of 1994,25 regional stock exchanges were actually encouraged to extend unlisted trading privileges to stocks listed on other trading venues. That statute simply codified a long-standing SEC policy of acquiescing in requests by regional exchanges for unlisted trading privileges.26 Occasionally, issuers have tried to control trading in their securities after their securities have already been issued, but these efforts have been uniformly unsuccessful.27 Two points about these efforts by issuers to control trading venue are worth making. Firstly, it appears that the SECs reluctance to block the grant of unlisted trading privileges, at least in some cases, has furthered shareholder interests by blocking managers efforts to entrench themselves in office. For example, in re Providence Gas Company, managers of a public utility wanted to limit ownership in the stock of the company to customers of the company. The managers sought to effectuate this policy by limiting trading in the firms securities to the local overthe-counter market. This policy was upset when the New York

transfer restrictions, in other words, balances two conflicting corporate tenets: Free alienability of corporate ownership interests and private corporate structuring to meet the participants needs.
24

21 Rafe v. Hinden, 29 A.D.2d 481, 288 N.Y.S.2d 662, affd mem. 23 N.Y.2d 759, 244 N.E.2d 469, 296 N.Y.S.2d 955 (1968); Allen v. Biltmore Tissue Corp., 2 N.Y.2d 534, 540, 161 N.Y.S.2d 418, 421, 141 N.E.2d 812, 814; Penthouse Properties v. 1158 Fifth Ave., 256 App. Div. 685, 690-691, 11 N.Y.S.2d 417. 22 This latter restriction is valuable in circumstances where, for example, state law limits share ownership in a professional medical corporation to doctors, or ownership in a professional legal corporation to licensed attorneys. These restrictions similarly allow shareholders to choose their business associates, to restrict ownership to family members, and to ensure congenial and knowledgeable associates. Share transfer restrictions also can prevent business competitors from purchasing shares. There are also important tax planning reasons for the restrictions. 23 RMBCA 6.27; Cal. Corp. 418; Del GCL 202. 24 Lewis D. Solomon & Alan R. Palmiter, Corporations: Examples and Explanations, (2d edition 1994).

25 Pub. L. No. 103-389, 108 Stat. 408 (1994) (codified as amended at 15 U.S.C. Section 78l(f) (1994) provides that unlisted trading privileges for securities originally listed on another national exchange may be granted to any security that is listed and registered on a national securities exchange. 26 140 Cong. Rec. H6508 (daily ed. August 1, 1994) (statement of Rep. Markey) 27 In re Edison Electric Illuminating Co., Securities Exchange Act Release No. 986 (December 17, 1937) (rejecting effort by company to block the granting of unlisted trading privileges in its bonds by the New York Curb Exchange); In re Providence Gas Co., Exchange Act Release No. 1992 (January 19, 1939) (rejecting issuers attempt to terminate the granting of unlisted trading privileges in the companys common stock by the New York Curb Exchange); In re Chicago Rivet & Machine Co., Exchange Act Release No. 3395, Fed. Sec. L. Rep. (CCH) 75,369 (March 17, 1943) (same); Ludlow Corp. v. SEC, 604 F.2d 704 (D.C. Circuit 1979) (permitting, over the issuers objection, the Boston Stock Exchange to grant unlisted trading privileges to stock already trading on the New York Stock Exchange).

53

Best execution regulation: From orders to markets

Curb Exchange (now the American Stock Exchange) granted unlisted trading privileges to the companys stock. The company unsuccessfully tried to block the New York Curb Exchanges move to extend unlisted trading privileges to the companys stock. Here it seems clear that the interests of management in restricting trading were unlikely to have been shared by the companys shareholders, whose interests were in expanding the investor base to improve liquidity in the companys shares. Management was, on the other hand, more interested in limiting outsiders access to the companys shares in order to reduce the possibility of a hostile takeover attempt. Thus, in this context at least, it appears that the liberal use of unlisted trading privileges serves shareholder interests and reduces agency costs. But merely because unlisted trading privileges are liberally granted does not mean that issuers are helpless to effectuate restrictions on the venues on which their securities trade. Share transfer restrictions, imposed as part of the initial stock issuance, can require stock to be traded in certain settings, effectively allowing issuers to influence the best execution of their securities.

trade price they are giving to investors reflects a fee for payment to order flow; and (4) whether allowing one investor to trade-through a price sacrifices best execution for the other trader. The issues surrounding attaining best execution are indeed complex. As we have argued here, however, even the concept of best execution is becoming untenable. Increasingly, trading takes place in multiple trading venues, and large orders are split and completed in stages. Where there are multiple trade executions taking place over a number of days, not only does the concept of best execution become extremely imprecise, but the common law idea of using fiduciary principles, and viewing each trade on a case-by-case basis in isolation, is difficult to defend. Moreover, whatever regime of best execution is selected, that legal regime should be organized to reflect the fact that the rules governing best execution will profoundly affect not only the costs associated with individual trades, but also the ultimate market structure as a whole. Thus, deriving a single, operational legal definition of best execution is simply not possible in todays complex markets, where traders preferences are heterogeneous and where trading venues offer a wide variety of benefits for traders. For all of these reasons, we are of the view that the critical question to answer in formulating a best execution regime is where to allocate the decision over trading, rather than what particular rules should be applied to particular trades. As among the various potential places to locate the decision about what regime of best execution to discuss, all have flaws. However, among individual traders themselves, the government (SEC), the trading venues themselves, and the issuers, clearly the issuers are the institution with the strongest incentives to formulate efficient rules of best execution.28 Perhaps the answer to how best to regulate best execution lies not at the order- or the market-level, but at the issuer level.

Conclusion
In this Article we have considered, from a law-and-economics perspective, how best to achieve the elusive goal of best execution of trades in todays increasingly fractured and complex trading markets. In earlier work we have observed that [d]espite the seeming simplicity of this concept, few issues in todays securities markets are more contentious than the debate surrounding best execution. The quest for a workable legal rule is confounded by issues such as: (1) whether clearing a trade in one market at the best available current quote constitute best execution if trades frequently clear between the quotes in another market; (2) whether trade size should be considered when determining what constitutes best execution; (3) whether brokers and investment advisers are in compliance with their legal right of best execution obligation if the

54 - The

Journal of financial transformation

28 Of course we say this with the proviso that issuing firms have the strongest incentives to establish execution rules at the time of the initial IPO, and at other times the protections and discipline required for charter amendments (including the requirement of a shareholder vote) should be required before issuers can change the rules relating to securities trading ex post, that is, after shares are issued.

Best execution regulation: From orders to markets

References
Amihud , Y., and H. Mendelson, 1996, A new approach to the regulation of trading across securities markets, 71 NYU L. Rev. 1411, 1416 Application of Rule 2320 to the Provision of Automated Execution for Orders in OTC Bulletin Board Securities NASD Regulation, Inc. Interpretive Letter from Alden S. Atkins to Mr. Leonard Mayer (May 1, 2000). Bacidore, J., R. Katharine, and G. Sofianos, 1999, Quantifying best execution at the New York Stock Exchange: Market orders, New York Stock Exchange Working Paper #99-05 Bacidore, J., R. Battalio, and R. Jennings, 2001, Changes in order characteristics, displayed liquidity, and execution quality on the New York Stock Exchange around the switch to decimal pricing, New York Stock Exchange Working Paper #2001-02 Bessembinder, H., 1999, Trade execution costs on Nasdaq and the NYSE: A postreform comparison, New York Stock Exchange Working Paper #98-03 (Updated Feb. 1999) Federal News, 2000, Broker-Dealers: Court approves U.S.$20M settlement in Schwab suit over best execution, 32(30) Sec. Reg. & Law Rep. (BNA) 1018, July 31 Federal News, 2000, Broker-Dealers: Firms should monitor best execution themselves, NASDR Official Counsels 32(6) Sec. Reg. & Law Rep. (BNA) 187, February 14 Federal News, 2000, Broker-Dealers: SEC staff to specify disclosure by firms to investors on order handling, Official Says 32(28) Sec. Reg. & Law Rep. (BNA) 940, July 17 Federal News, 2000, Stock Markets: Levitt calls for better linkages, public display of limit order books, 32(13) Sec. Reg. & Law Rep. (BNA) 429, April 3 Final Rule, 2000, Disclosure of order execution and routing practices, Exchange Act Release No. 34-43590, November 17, U.S. Securities and Exchange Commission Gordon DuMont v. Charles Schwab & Co., Inc. No. CIV. A. 99-2840, CIV. A. 99-2841, Memorandum Opinion and Orders, 2000 WL 1023231 (E.D. La., Jul. 21, 2000) (approving settlement of consolidated nationwide class actions based on allegations that Schwab purportedly failed to provide best execution of customers stock transaction orders and [accepted] payments for order flow from regional and third markets without disclosing the fact to its customers). Haddock, D. D., and J. R. Macey, 1987, Regulation on demand: A private interest model, with an application to insider trading regulation, Journal of Law and Economics, 30, 311-52 Hamilton, W., 2000, E-Trading a bad deal for investors? Los Angeles Times, August 21 Leland, H. E. 1992, Insider trading: Should it be prohibited? Journal of Political Economy, 100, 859-887 Levitt, A., 1999, Best execution: Promise of integrity, guardian of competition, Remarks by Chairman Arthur Levitt, U.S. Securities & Exchange Commission at Securities Industry Association, Boca Raton, Florida, November 4 Macey, J. and M. OHara, 1999,Globalization, exchange governance, and the future of exchanges, Brookings-Wharton Papers on Financial Services

Macey, J. and M. OHara, 2001, The economics of listing fees and listing requirements, Journal of Financial Intermediation, 11, 297-319 Macey, J. R., 1984, From fairness to contract: The new directions of the rules against insider trading, Hofstra Law Review, 13, 9-64 Mendiola, A. and M. OHara, 2003, Taking stock in stock markets: The changing governance of exchanges, Working Paper, Cornell University Macey, J. R., and M. OHara, 1997, The law and economics of best execution, Journal of Financial Intermediation, 6, 188-223 Merrill Lynch, Pierce, Fenner & Smith Incorporated, PaineWebber Incorporated, and Dean Witter Reynolds, Inc., Petitioners, v. Jeffrey Phillip Kravitz, Gloria Binder, and Bruce Zakheim IRA FBO Bruce Zakheim, Respondents, No. 97-1768, Brief of the Bond Market Association as Amicus Curiae in Support of the Petitioner (U.S. Sup. Ct., Jun. 1, 1998). Vaugha, A., 1999, NASDR to crack down on best execution claims on sites, Financial Net News, February 9 OHara, M., 2004, Searching for a new center: U.S. securities markets in transition, Economic Review, Federal Reserve Bank of Atlanta, forthcoming Online Brokerage: Keeping Apace of Cyberspace, Commissioner Laura S. Unger of the U.S. Securities and Exchange Commission (Nov. 22, 1999). Proposed Rule: Disclosure of Order Routing and Execution Practices Exchange Act Release No. 34-43084 (Jul. 28, 2000), U.S. Securities and Exchange Commission Proposed Rule: Disclosure of Order Routing and Execution Practices [Comments on Proposed Rule] SEC.gov (last update Nov. 2, 2000), U.S. Securities and Exchange Commission. Progress toward the development of a national market system, Joint hearings before the Subcommittee on Oversight and Investigations and the Subcommittee on Consumer Protection and Finance, 96th Congress, Serial 96-89, U.S. Government Printing Office, Washington DC Regulators Abuzz Over Best Execution Virtual Wall Street (Oct. 2000). Reerink, J., 1999, Schwab offers velocity software that speeds the trade process, USA Today, Aug. 25 Roberts, R. Y., and E. L. Pittman, 1999, SEC attention turns to best execution, eSecurities 1, Leader Publications, December SEC Order execution obligations, Exchange Act Release No 34-37619A, 60 Fed. Reg. 48290 Schaffer, Frederick P. & Sullivan, William F., Best Execution and Related Issues for On-Line Firms in Schulte Roth & Zabel LLP The Internet and Securities Law Current Issues and Trends [Seminar materials], Nov. 15, 1999, Tab 5. Sirri, E. R., Innovating around stasis: The exchange markets response to SEC regulation of institutional form, American Enterprise Institute for Public Policy Research, Draft, February 9 Stoll, H., Affirmative Obligation of Market Makers: An Idea Whose Time Has Passed?, Financial Markets Research Center, Owen Graduate School of Management, Vanderbilt University, Working Paper 97-14

55

56 - The

Journal of financial transformation

Services

Credit card pricing developments and their disclosure

Mark Furletti
Payment Cards Center, The Federal Reserve Bank of Philadelphia1

Abstract
Public data, proprietary issuer data, and data collected by myself from a review of over 150 lender-borrower contracts from 15 of the largest issuers in the U.S. suggest that, over the past 10 years, credit card issuers have drastically changed the way that they price their products. This paper outlines the history and dynamics of credit card pricing over the past 10 years and examines how new pricing methods are addressed by current regulatory disclosure requirements.

The views expressed here are not necessarily those of the Federal Reserve Bank of Philadelphia or of the Federal Reserve System.

57

Credit card pricing developments and their disclosure

Intense competition for new customers and the adoption of new technologies in the credit card industry has decreased the price of credit for most consumers, as measured by one wellunderstood metric, the nominal annual percentage rate (APR). For card issuers, this has meant surrendering some of the net interest margin they enjoyed as a result of high APRs in the late 1980s and early 1990s and instituting pricing strategies that consider an individual borrowers risk and behavior profile. As nominal APRs have decreased, issuers have come to rely on new pricing techniques to maintain or increase portfolio profitability. These techniques include new APR strategies, fee structures, and methodologies to compute finance charges. This paper outlines the history and dynamics of credit card pricing over the past 10 years and examines how pricing methods are disclosed to consumers. The analysis concludes by discussing the challenges that newer, more complex pricing strategies pose to the current disclosure framework established by the Truth in Lending Act.

that most consumers do not have to work very hard to find a new card. Product innovations, such as transferring balances and eliminating annual fees, have also made it easier for customers to switch cards. Customer loyalty, once ensured by an annual fee and a revolving balance built through years or months of purchases, can now be easily captured by competitors with a no-fee, low-rate offer to transfer balances. As a result of these developments, issuers have struggled to maintain customer loyalty through rewards programs, affinity/cobrand relationships, and enhanced customer service. Despite these efforts, a cards nominal APR remains one of its most distinguishing characteristics. The envelopes, letters, and applications that issuers use to solicit new business focus potential customers attention on either very low introductory APRs, such as 0 percent, or low permanent APRs, such as 7.9 percent. Low rates, however, are relatively new phenomena in the card industry. Researchers studying the card industry in the 1980s and early 1990s found that credit cards had substantially higher rates and returns than most other bank credit products.4 Further research showed that credit card rates remained high when other interest rates fell, leading Calem and Mester to conclude that card rates in that environment were sticky.5 From 1992 to 2001, however, the average interest rate that issuers charged revolving customers fell 320 basis points, from 17.4 percent to 14.2 percent. Issuer markup, a metric that normalizes for funding costs by subtracting the six-month Treasury bill rate from the average APR, decreased 330 basis points during the same period. Margins also narrowed compared with those of other consumer loan products. The difference between the average interest rate charged on a 24month personal installment loan and a revolving credit card loan fell from 3.8 percent in 1992 to 1.6 percent in 2001. Taken together, it is clear that for low-risk consumers who have revolving balances, credit card costs, as measured by APR, have significantly declined over the past 10 years. At the same time, more consumers gained access to credit cards, including

Industry pricing dynamics

Over the past 10 years, a series of innovations and market developments have significantly changed the credit card industry. Advances in credit scoring, response modeling, and solicitation technologies (i.e. e-mail, direct mail, telemarketing) have allowed experienced issuers to more efficiently market their products and enabled new issuers to enter the card market and grow quickly.2 At the same time, it has become easier for consumers to find better credit card alternatives and move their card balances from one issuer to another. From 1991 to 2001, the number of mailed credit card solicitations increased fivefold to 5.01 billion. According to BAI Global, these solicitations in 2001 reached 79 percent of U.S. households, which, on average, received five offers each month.
3

Issuers aggressive mail marketing efforts have been augmented by telephone, event, and Internet campaigns, such

58

2 In the eight months since launching its Visa program at the end of 2001, Target Corporation issued 6 million credit cards and captured over U.S.$2 billion in outstandings. Similarly, other new entrants like Sears and Juniper Bank have been able to grow very rapidly and compete against much larger issuers. 3 Compared with 73 percent of U.S. households receiving four offers in 2000. BAI Global, All time record high credit card mail volume set in 2001, press release, April 2002

4 Ausubel, L. M, 1991, The failure of competition in the credit card market, American Economic Review, March, pp. 50-81 5 Calem, P. S., and Loretta J. Mester, 1995, Consumer behavior and the stickiness of credit-card interest rates, American Economic Review, December, 1327-36. This paper is primarily focused on pricing changes that occurred in the mid- to late1990s. For detailed information about credit card pricing in the 1970s and 1980s, please refer to Lewis Mandells The Credit Card Industry: A History (G. K. Hall & Co., 1990).

Credit card pricing developments and their disclosure

those with lower incomes. Between 1989 and 1998, the largest increases in bankcard ownership were observed among consumers with the lowest levels of income.6 With generally lower liquidity buffers and weaker credit histories, lower income consumers are typically assessed higher annual percentage rates. An overall decrease in the average APR, coupled with an increase in the number of lower income credit users, suggests that the average rate decrease for many cardholders was even more pronounced than the average APR indicates. Consumer awareness of annual percentage rate as a key cost measure, combined with the ability to easily find new card offers and switch issuers, inevitably affected price competition and rate stickiness. According to surveys conducted in 2000 by the Survey Research Center of the University of Michigan, 91 percent of consumers who have a credit card are aware of the APR they are charged on their outstanding balances, based on a broad definition of awareness.7 Federal Reserve Board economist Thomas Durkin concludes that it is clear that awareness of rates charged on outstanding balances has risen sharply since implementation of the Truth in Lending Act in 1968.

or limits for any charges. Instead, it requires that issuers inform potential customers about specific pricing terms at specific times. Regulation Z specifies that select terms be disclosed at specific points, such as upon solicitation or application, before first use of the card, and upon receiving a statement.8 The level of detail for disclosure at each point varies. When first promulgated, TILA rules required that issuers of credit cards disclose information about the computation of APRs and finance charges to customers before the first transaction [was] made on the account. To meet this requirement, issuers mailed consumers a single written statement that explained the costs of the card after his or her account was opened. Since 1968 both Congress and the Board of Governors (the Board) have mandated changes to TILA disclosure requirements. One of the most well-known features, being a pricing disclosure box, resulted from the amendment of TILA by the Fair Credit and Charge Card Disclosure Act of 1988. Informally referred to as the Schumer box after the congressman from New York who was instrumental in the legislations passage the box displays APR and fee information on card applications and solicitations in a table designed to be easy for consumers to read and use for comparison purposes. By requiring that issuers display this box on applications and solicitations, the act enabled consumers to compare offers and rates before opening an account. Similarly, in response to the potentially confusing number of different APRs that can now be associated with a single credit card account such as balance transfer, cash advance, and purchase APRs the Board further modified Regulation Z in 2000. As a result of this modification, issuers are required to disclose the APR for purchases in at least 18-point type on applications and solicitations. The modification also requires them to disclose balance-transfer fees that apply to an account.

The Truth in Lending Act and price disclosure

The Truth in Lending Act (TILA) was enacted as Title I of the Consumer Credit Protection Act in 1968. The act stated that economic stabilization would be enhanced and that competition would be strengthened by the informed use of credit resulting from an awareness of credit costs on the part of consumers. TILA charged the Federal Reserve with creating and enforcing the specific rules needed to implement the legislation. These rules are embodied in the Board of Governors Regulation Z (Truth in Lending). TILA, as it applies to credit card accounts, is primarily disclosure focused. The act is silent about the number, amount, variety, or frequency of fees and credit-related charges that issuers can impose. It does not suggest ceilings, price controls,

The Board, through modifications to Regulation Z, and Congress, through legislation, have updated Truth in Lending to take into account product evolution. Recent changes in how

6 Durkin, T. A, 2000, Credit cards: Use and consumer attitudes, 1970-2000, Federal Reserve Bulletin, September 623-34. The following are the changes in the percentage of respondents to the survey cited by Durkin by income quintile who indicated that they owned a bank-type credit card: lowest +65 percent; second lowest +61 percent; middle +16 percent; second highest +13 percent; highest +7 percent. 7 Awareness was measured using a narrow and a broad definition. Under the broad definition, only those reporting that they did not know the rate were considered unaware. Under the narrow definition, those reporting a rate less than 7.9 percent were also considered unaware. Using the narrow definition, awareness in 2000

was measured at 85 percent. Previous measures of awareness from the Survey of Consumer Finance did not distinguish between narrow and broad. These measures showed 27 percent in 1969, 63 percent in 1970, and 71 percent in 1977. [Durkin (2000)]. 8 Regulation Z also requires that specific information be disclosed in advertisements for credit (i.e. adverts on television or in magazines) and when certain credit terms are changed. This paper does not examine these regulatory disclosure requirements.

59

Credit card pricing developments and their disclosure

issuers price credit cards, however, have resulted in new levels of pricing complexity and created a structure of credit costs that can impact some customers very differently than others. That is, the cost that a consumer faces greatly depends on the way he or she uses the credit card. This paper will explore the evolution of credit card pricing and examine the disclosure requirements of Truth in Lending (Regulation Z) that relate to these pricing changes. Analysis will rely on public data, proprietary issuer data, and data collected by the author from a review of over 150 lender-borrower contracts from 15 of the largest issuers in the U.S. over a five-year period. Pricing and fee changes are organized into
9

released to the media in the early 1990s.10 At the time, issuers generally had one rate that they extended to all customers. When they lowered this rate, they did so for almost all of their accounts. A 1993 issue of CardTrack, a publication of CardWeb.com, reported that Citibank was offering a 15.4 percent rate to all new applicants.11 This was the same rate it was offering to virtually all of its current customers. CardTrack also reported that 90 percent of Citibank cardholders had been paying an APR of 19.7 percent a few years earlier. Other large issuers, such as Chase, Chemical, AT&T, and Bank One, made rate-cut announcements that were similarly applied to all current and new customers.12 Competitive pressures and increasing price awareness among consumers, however, eventually made these undifferentiated pricing strategies obsolete.

three categories nominal APR, fee structure, and computational technique changes and presented in order of most to least consistent with the current format of regulatory disclosure requirements. Examples of each type of change are provided, along with an analysis of each changes impact on issuers revenues.

Risk-based solicitation APRs

Issuers have generally used risk-based pricing techniques in two ways. The first is in setting the interest rate initially offered to a consumer. Using credit bureau attributes, issuers assess the default risk of a consumer and essentially charge him or her a premium for that risk. This premium is typically reflected in the APR stated in the card application or solicitation. Prior to the early 1990s, by charging every customer the same rate, issuers made much higher profits from customers with very low default risk.13 These excess profits could be used to cover defaults generated by customers whose risk, over time, had increased. As issuers began competing on APR, however, they were forced to eliminate this cross-subsidization and assess APRs based on an analysis of individual borrower risk. Ultimately, a cards nominal APR became a competitive focal

Nominal APR changes

Until the early 1990s, credit card pricing, as it related to nominal APRs, might best be characterized in two ways, high and simple. Card issuers generally had one or two card products, such as a classic card and/or a gold card, that each had a single annual percentage rate of around 18 percent. If an applicant for credit could pass the risk threshold set by the issuer, he or she would receive a card. If the applicants credit behavior was determined to be too risky, his or her application was denied. This resulted in a portfolio of customers who were priced as if they had very similar probabilities of default. Evidence of these risk-indifferent APR strategies can be observed in public rate decrease announcements that issuers

point and drove widespread adoption of risk-based pricing. Issuers who failed to adjust pricing appropriately by risk seg-

60

9 Lender-borrower contracts are the documents issuers send their customers that often include fee disclosures, account usage terms and conditions, borrower and lender responsibilities, etc. Issuers typically refer to them as Cardmember Agreements or Required Disclosures, and modify them with Change in Term Notices. These documents are usually made available to cardholders before the first transaction is made on the account. 10 Stango (2002) observes that comments about high card rates from President George H. Bush in 1991 and the Senates passage of a bill in that same year capping APRs influenced many large issuers to lower rates. (The Senates bill was never signed into law.)

11 Cardweb.com, CardTrack. National Credit Education Week, <http://www.cardweb.com/cardtrack/> April 1993. 12 Stango, V., 2002, Strategic responses to regulatory threat in the credit card market, Federal Reserve Bank of Chicago Working Paper 2002-02, February 13 The risk-based pricing techniques referred to in this section impact customers only to the extent to which they carry a balance on their credit card. For customers who always pay their balance in full, such pricing techniques are effectively inconsequential.

Credit card pricing developments and their disclosure

ments would expose themselves to serious adverse selection problems. Issuers today may have hundreds of different APR price points. With few exceptions, these points are highly correlated to some risk measure. When we look at the difference
14

fail to make a payment to us when due, you exceed your credit line, or you make a payment to us that is not honored by your bank. I also found that issuers had taken these policies a step further in recent years. Agreements were changed to allow issuers to incorporate into the penalty pricing decision information they obtained from credit bureaus about other loan behavior. Newer policies read as follows: We may increase the annual percentage rate on all balances to a default rate of up to 24.99 percent if you fail to make a payment to us or any other creditor when due, you exceed your credit line, or you make a payment to us that is not honored by your bank [emphasis added].20

between the effective finance charge yield for the highest risk revolving customers (FICO scores less than 600) and customers in other risk cohorts, we find that the discount that lower risk customers receive on their APR has increased significantly since the early days of risk-indifferent pricing. The lowest risk customers, who once paid the same price as highrisk customers, now enjoy rate discounts that can reach more than 800 basis points. At the other end of the risk spectrum,
15

these strategies have enabled issuers to grant more people (i.e. immigrants, lower income consumers, those without any credit experience) access to credit, albeit at higher prices. Former Federal Reserve Governor Lawrence Lindsey has referred to this phenomenon as the democratization of credit.16 Examining data from the Survey of Consumer Finance, Stavins also noted the same risk-based pricing trend. She observes that consumers with higher ratios of unpaid credit card debt to income, and thus [who were] worse credit risks for the issuers, were charged higher interest rates.17

Nominal APR changes and regulatory disclosure requirements

The risk-based pricing strategies described above are exclusively focused on the nominal APR component of credit card pricing. Disclosures required by Regulation Z inform customers about such APRs upon solicitation in two sections of the Schumer box, APR for Purchases and Other APRs, and on periodic statements. In addition, Regulation Z requires that the non-introductory purchase APR be displayed in 18-point type in the Schumer box on new card offers. Overall, Truth in Lending disclosure requirements ensure prominent display of each APR associated with an account. The nominal APR-focus of Truth in Lending statements and of issuers marketing materials has no doubt contributed to consumer awareness of APRs as key determinants of credit cost.

Risk-based penalty APRs

Risk-based pricing strategies can also be used to modify a customers APR after he or she has started using the account. Issuers have recently implemented penalty APR strategies that allow them to adjust upward the nominal APR of customers whose risk, perhaps because of recent late payments or increasing levels of debt, is no longer in line with their original APR.18 In my own study of lender-borrower contracts I found that penalty pricing strategies were first introduced in the late 1990s.19 An example of language that explained these policies in 1997 read as follows: Your APRs may increase if you

Fee structure changes

Another way that credit card pricing has developed is in the unbundling of costs in the form of fees. As previously mentioned, card pricing in the 1980s and early 1990s was relatively simple. Issuers typically charged a relatively high interest

14 One notable exception to the risk-based pricing strategy is the co-branded airline portfolio. Co-branded air cards that reward users with frequent-flyer miles typically attract low-risk business travelers despite having a high rate (e.g., 18.9 percent) and an annual fee. 15 One could argue that the customers at any given FICO score might be riskier today than in 1998 (i.e., a 650 FICO score in 2002 carries a higher risk of default than a 650 did in 1998) because of changes in issuers underwriting standards or a less favorable economic environment. There are three reasons to doubt the material impact of such factors. Firstly, in an attempt to control for the impact of economic cycles, the data are presented relative to the yield of highest risk customers (FICO scores < 600). Secondly, credit modeling experts believe that Fair Isaac frequently recalibrates its FICO model in order to ensure that its score-odds ratio is relatively stable. This mitigates the effects that different economic environments might have on the score. Finally, the underwriting standards of the

prime/super-prime issuers in the Argus study are thought to have been stable throughout the period with little or no sub-prime origination. 16 Black, S. E., and D. P. Morgan, 1998, Risk and the democratization of credit cards, Federal Reserve Bank of New York, Research Paper 9815 17 Stavins, J., 2000, Credit card borrowing, delinquency, and personal bankruptcy, New England Economic Review, July/August, 15-30 18 Penalty pricing tactics employed by Direct Merchants (Metris) led CardTrack to observe that credit card interest rates have passed the 30% barrier! As reported in May 2000, a 31.99 percent APR was imposed on Metris customers who were late three times during the year or who fell 60 days delinquent. 19 I would like to thank MarketIQ, a direct marketing competitive intelligence firm in Fair Haven, New Jersey, for contributing to the study. 20 Some issuers policies explained that a consumer could get his or her pre-penalty rate back after making 12 consecutive on-time payments.

61

Credit card pricing developments and their disclosure

rate and an annual fee of around U.S.$25 that covered most of the expenses associated with card usage. Few issuers charged over-limit fees or late fees, and when they did, these fees were relatively small. The increased competition for new accounts
21

based fee levels and created new fees. For example, in 1997, the risk-based fee that most issuers charged customers who had exceeded their credit line was less than U.S.$20. By 2002, most top issuers had adopted an over-limit fee structure that was tiered by balance size and nearly doubled the fee for overlimit customers. These issuers now assess a U.S.$35 fee to those customers who exceed their credit line and have a balance of over U.S.$1000. Similar increases were observed for late fees and returned-check (NSF) fees.24 Issuers were also observed introducing new risk-based fees during this period. For example, in the late 1990s, issuers began assessing a returned-check fee for credit card convenience checks. Such checks allow customers to access their credit cards line of credit using a paper check. If a customer writes a convenience check for an amount that exceeds his or her available credit line and the issuer chooses not to honor the check, most issuers now assess that customer a fee that ranges from U.S.$29 to U.S.$35. The impact of higher risk-based fees on issuers revenues has been substantial. In May 2002, Cardweb.com estimated that half of all consumers in the U.S. who had a credit card had been late at least once in the previous 12 months. Issuers annual late fee revenues more than quadrupled from 1996 to 2001 (U.S.$1.7 to U.S.$7.3 billion) while average late fees only doubled (U.S.$13 to U.S.$27). This indicates that, in addition to an increase in the amount of the average late fee, there has been a substantial increase in late fee incidence. Cardweb.com also noted that late-fee revenues currently represent the third largest revenue stream for issuers after interest and interchange revenue.25 Argus Information & Advisory Services data compiled from top prime issuers during the first quarter of 2002 showed that 5 percent of issuers active cardholders were assessed an over-credit-limit fee during the three-month period. Information on the size and growth of other risk-based fee types is not available. The examples above, however, strongly suggest that risk-based fees have become an impor-

that developed in the mid-1990s, however, changed all of this. Rates came down, as described in the previous section, and issuers eliminated the once universal annual fee. Today, these fees are almost nonexistent in prime portfolios not associated with a rewards program. The data that Argus Information & Advisory Services collected from top issuers show that just 14 percent of customers who are not enrolled in a rewards program, such as frequent-flyer-miles program, paid an annual fee in 1998 and just 2 percent did in 2002.22 With average interest rates on the decline and annual fees becoming unpopular among their customers, issuers developed more targeted fee structures to replace lost revenues. In lieu of charging all of their customers an annual fee that subsidized the costs associated with the behaviors of a few, they began to assess fees directly on those customers whose card usage behaviors drove costs higher. As issuers started unbundling costs and creating behavior-based fees, fees rebounded and have again become an important component of issuer revenues. Ultimately, two distinct families of fees have emerged, risk-related fees and convenience/service fees.

Risk-related fees
In addition to using different APRs to better price for risk, issuers have significantly increased the use of risk-related fees. These include late fees, over-limit fees, and bouncedcheck fees. The industrys modeling and analysis efforts have shown that customers who are late or over their credit limit or who write bad checks are more likely to default. Risk-related fees help compensate issuers for this increased risk.23 For lower risk customers, risk-related fees can deter sloppy payment behavior and poor credit-line management. Examination of lender-borrower contracts from 1997 through 2002 revealed that issuers significantly increased traditional risk-

62

21 Typical late fees ranged from U.S.$5 to U.S.$10. The average late fee charged in 1990, according to CardWeb, was U.S.$9. 22 The data from Argus also show that the average annual fee charged on a nonrewards card has fallen from U.S.$3.31 in 1998 to U.S.$0.50 in 2002. 23 Issuers may also be aware that customers who are consistently paying these fees are likely to have fewer and less attractive credit alternatives.

24 NSF is a return check reason code that stands for not sufficient funds. 25 Interchange revenue is derived from a fee set by the card associations that issuers assess merchants each time a credit card purchase is made. Depending on the card association, the fee can range from 1.5 to 4.0 percent of the value of the transaction.

Credit card pricing developments and their disclosure

tant source of revenue for card issuers and have replaced a significant portion of the revenues lost from the elimination of annual fees and lowered APRs.

Finally, most issuers have adopted balance transfer fees. These fees, typically around 3 percent of the amount transferred with various minimums and maximums, are often assessed on balances transferred from a competitors card. These balances often qualify for a discounted promotional APR. The balance transfer fee helps offset costs associated with customer service representatives who initiate the balance transfers and may help reduce rate surfing (i.e., the act of continually moving balances among cards to take advantage of short-term promotional rates).27 Information on the revenue impact of convenience and service fees is limited. Argus Information and Advisory Services data indicate that the top prime issuers are earning about U.S.$8 per active account per year in cash advance fees and U.S.$6 per active account per year in other fees (excluding risk-related fees). The actual impact that fees have on revenue per account can be observed only among the few issuers who separately list non-securitization fee income in their annual reports. One such annual report revealed a doubling of fee revenue per account from U.S.$4 in 1998 to U.S.$8 in 2001.

Convenience and service fees

Issuers have also unbundled servicing costs, introducing fees for services and conveniences that were once paid for by all customers out of annual fee and interest revenues. Some of these new fees, like those levied on the credit card purchase of casino chips or on cash advances, compensate issuers for the fraud risk thought to be inherent in cash or cash-equivalent transactions. Other fees, like those imposed for stop payment requests, statement copies, or replacement cards, more directly compensate issuers for out-of-pocket expenses, such as customer service representative time or telecommunications expense. In addition to defraying operational costs, these new fees are generally priced to provide attractive profit margins. Starting in the late 1990s, a number of issuers began assessing a foreign currency conversion fee of 2 percent on purchases that cardholders make outside the U.S. This fee was added on top of a 1 percent fee already assessed by MasterCard and Visa. The 1 percent fee charged by the associations covers the transaction costs associated with the actual exchange. Some industry sources suggest that the 2 percent fee levied by issuers is related to the long-distance telecommunications charges associated with customer service calls that originate in foreign countries from traveling customers. More recently, several issuers have added a phone payment convenience fee. This fee, which ranges between U.S.$10 and U.S.$25, is assessed when customers choose to pay over the phone instead of through the mail.26 While this fee may seem like an expensive alternative when compared to the cost of postage, customers who typically rely on phone payments are often close to missing their payment due date and being assessed a U.S.$35 late fee.

Fee changes and regulatory disclosure requirements

Regulation Z requires that issuers, upon application or solicitation, inform customers about the annual fee, minimum finance charge, cash advance fee, balance transfer fee, late fee, and over-limit fee associated with an account. Before the first transaction is made on the account, issuers must disclose other charges, that is any charge other than a finance charge that may be imposed as part of the plan, or an explanation of how the charge is determined. The official staff commentary on Regulation Z, which represents the Board staffs interpretations of the regulation, further provides that only significant charges must be disclosed as other charges. The commentary offers late fees, annual fees, over-limit fees, and account closure fees as examples of significant charges.

26 When an issuer accepts a payment over the phone, it receives authorization from its customer to debit the customers checking account for the payment amount. 27 If these fees become universally adopted, they will make it more expensive for consumers to switch cards.

63

Credit card pricing developments and their disclosure

Regulation Z does not explicitly address disclosure of the foreign currency conversion fee. Unlike most fees that can be observed upon a detailed review of a card statement, foreign currency conversion fees are often rolled into the transaction amount or the conversion factor.28 Other fees that are not specifically mentioned in the regulation include phone payment fees, wire transfer fees, and stop payment fees on credit card convenience checks. Issuers generally disclose these
29

card at an APR of 2.9 percent. Since the customers payments are allocated to the 2.9 percent balance first, the issuer effectively protects or locks in the U.S.$2500 balance at 18.9 percent until the lower rate balance is repaid.

Compounded interest
By 1997, most issuers had switched from monthly to daily compounding of interest by changing the computational method for calculating average daily balances. Before the adoption of daily compounding, disclosures typically explained that on each day of the billing period we subtract payments, add new purchases and fees, and make adjustments to calculate the average daily balance. By the end of the 1990s, however, the language had changed as follows: To get the daily balance we take the beginning balance for every day, add any new transactions, fees, and any finance charge on the previous days balance, subtract any credits or payments, and make other adjustments [emphasis added]. By adding finance charges to the balance each day, issuers increased finance charge revenue without increasing stated annual percentage rates.30 This has the effect of increasing the effective finance charge yield of a portfolio by as much as 10 to 20 basis points. For instance, the annual effective portfolio yield on a loan with an APR of 18.99 percent compounded 12 times a year is 20.73 percent. If the same loan is compounded 365 times per year, its effective yield increases 18 basis points to 20.91 percent.

fees to consumers by including a menu or a description of these other fees in welcome kit mailings to new customers or in Cardmember Agreements. The organization, detail, and prominence of these menus or descriptions vary by issuer.

Computational technique changes

In addition to adopting risk-based pricing and expanding fees, issuers are employing new computational practices that increase effective yields without affecting the disclosed nominal APRs. My own lender-borrower contract research uncovered many such examples among the major issuers. Three of the more common practices are detailed below.

Payment allocation
Many issuers have added sections to their contracts to explain how they allocate payments to revolving balances. As mentioned previously, the number of APRs that can be applied to the balances on an account has increased dramatically over the past 10 years, such as purchase, promotional, cash, and balance transfer APRs. Issuers have created various average daily balance categories to which these different rates are applied. One issuers disclosure statement explained the way in which payments would be applied to different balance categories as follows: We will allocate your payment and any credits to pay off balances at low periodic rates before paying off balances at higher periodic rates. This computational methodology effectively protects revolving balances at higher rates. For example, issuers can offer a customer with a U.S.$2500 revolving balance at 18.9 percent an opportunity to transfer another U.S.$2500 balance onto the

Double-cycle interest
Another pricing innovation involves a change in the treatment of the grace period during which interest does not accrue. One of the unique advantages of credit card borrowing has been the interest-free period consumers who pay their bill in full receive from the time they make a purchase until the date their payment is due. This period can vary from 40 to 60 days. The lender-borrower contract study revealed that a number of issuers have effectively eliminated the grace period for consumers who, after making a full payment or not having had a

64

28 Stellin, S., 2002, Credit card swipe: Concealed charges, New York Times, July 12, reports that just three major issuers separate out their foreign currency exchange charges on customers statements. 29 On November 26, 2002, the Board proposed credit card-specific revisions to Regulation Zs official staff commentary. One of these revisions would add phone payment fees to the list of other charges that must be disclosed before the first transaction occurs on an account.

30 The annual percentage rate (APR) disclosed on a customers statement is calculated by dividing finance charges for the period by the average daily balance for the period. While increasing the frequency of compounding increases the finance charge, the numerator, adding finance charges to the average daily balance, the denominator, each day offsets the effect of compounding on the disclosed APR.

Credit card pricing developments and their disclosure

balance in the previous month, do not make a full payment in the next month. For example, consider a customer without a previous balance who has a 10 percent APR on purchases and who makes a purchase of U.S.$1000 on May 1 (the first day of the customers May cycle). The customer then receives a bill for U.S.$1000 on June 1. Instead of paying the entire balance, the customer sends the issuer a minimum payment of U.S.$20, which arrives on June 30. When the customers account cycles on the night of June 30, the issuer will assess finance charges for the month of June and reach back and add finance charges for the entire month of May. In this example, instead of billing approximately U.S.$8 in finance charges (based on the APR of 10 percent), the issuer will bill approximately U.S.$16. It should be noted that double-cycle interest is assessed only in the month in which a customer moves from a non-revolving to a revolving state. The interest computation returns to a single-cycle method for the remaining months in the revolving period. Data on the revenue impact of the changes described above are limited. Given that approximately three-fifths of all cardholders pay interest on their balance and that half of those who pay interest make only the minimum payment [CardWeb (March 7, 2002)], it seems likely that these changes have had at least a material effect on issuers revenues and have contributed to the shift in industry revenue profiles.

chases), two-cycle average daily balance (excluding new purchases), adjusted balance, or previous balance. Additional explanations or definitions are not required by the regulation upon solicitation or application. Issuers are not required to provide detailed explanations of balance-computation techniques until after the account is opened. This means that consumers wanting to find out what the term two-cycle average daily balance signifies before filling out an application for a card would have to conduct their own research. As a practical matter, this may be easier for consumers with access to the Internet than for others.31 Detailed descriptions of the practices referred to above are usually not featured in application or solicitation materials.32 Issuers disclose the details of these computational techniques in various ways in their lender-borrower contracts and factor their effects into TILA-required periodic statement disclosures. For example, consumers who revolve a balance might become aware of the impact of daily compounded interest if, upon receiving their statement, they carefully review the issuers calculation of the total finance charges disclosed on their statement. Similarly, customers who pay finance charges on their late fees, on their balance between the statement date and payment date, or from their transaction date to their posting date might notice, upon an exceptionally careful review, increased finance charge amounts on their statement.

Conclusion Computational technique changes and disclosure requirements

Double-cycle billing is explicitly addressed by Regulation Z in a section of the Schumer box entitled Method of computing the balance for purchases. Here issuers are required to indicate the balance-computation technique they use with one of the following descriptors: average daily balance (including new purchases), average daily balance (excluding new purchases), two-cycle average daily balance (including new purSubstantial changes in the dynamics of credit card pricing have occurred over the past decade. The relatively straightforward pricing model of a single APR, an annual fee, and modest penalty fees has been replaced by a model with a complex set of APRs, new and increased fee structures, and sophisticated finance charge computation techniques. This unbundled pricing structure has created a card product for which consumers pay substantially different prices based on individual behavior. During this time, however, the overall disclosure

31 A search on two-cycle average daily balance on Google.com returns 10 non-governmental, consumer-oriented web sites, such as practicalmoneyskills.com, creditcards.com, and bankrate.com, that explain how two-cycle interest can influence the cost of credit.

32 The balance computation technique is disclosed with a brief descriptor in the Schumer box, but it does not give consumers the details necessary to understand how the interest is actually computed.

65

Credit card pricing developments and their disclosure

framework mandated by the Truth in Lending Act has changed very little.33 The adoption of new pricing structures has increased credit costs for some consumers and decreased it for others. Lowrisk borrowers, who behave in such a way as to avoid new and increased fees, generally experience lower credit costs than they might have several years ago. Higher risk borrowers, who may not have previously qualified for unsecured credit, can now obtain credit cards by paying a risk premium. Other borrowers, because of their consumption of fee-based services or perceived level of risk, now face higher credit costs. To a large extent the new pricing structure results in more credit card users paying their own way. A pricing structure that better allocates issuers risk and servicing expenses has likely come at a cost, in the form of a complex and customized product whose pricing is difficult to summarize. In todays environment of highly individualized pricing, it is difficult to imagine a generic disclosure requirement that could meet the burden of clearly explaining the total costs of credit that any given consumer would face. In 1996, in a report to Congress on TILA, the Board of Governors foresaw this possibility when it reported the following: The ability to ensure accurate disclosure of the true costs gets more difficult as creditors increase the number of credit products, pricing alternatives, and optional services. The permutations of possible costs to be disclosed and the potential for error also increase.34 While some of the pricing innovations described in this paper might easily fit into the existing regulatory disclosure format,

it is clear that others pose significant disclosure challenges. Is there a simple way to communicate two-cycle billing such that consumers understand that their credit costs will be higher if they occasionally revolve balances? Can payment allocation methods be explained in a way that customers with low-rate promotional balances understand the cost implications of making higher-rate purchases? Is there a simple way to explain that when consumers miss a payment with one issuer, it can affect the price they pay for credit to another? Recent survey results indicate that consumers have mixed feelings about Truth in Lending statements. Durkin (2002) points out that, in consumer surveys conducted for the Board of Governors in 1994, 1997, and 2001, over three-quarters of respondents agreed that Truth in Lending statements are complicated.35 Forty percent of those surveyed did not find the statements helpful as they relate to bank-type credit cards, and 77 percent said that the statements did not affect their decision to use credit cards in any way. Although consumers may find these disclosures complex and not always helpful, Durkin concludes that consumers appear to like knowing that the behavior of creditors is being monitored. Based on this survey data, it is not clear that requiring more details in regulatory disclosures would be useful for consumers. An alternative is to promote understanding of credit costs through education. Educated consumers can change the terms on which issuers compete and force transparency in price structures. In either case, understanding new developments in credit card pricing is important for ensuring that information is available for consumers to make an informed decision about credit.

66

33 As previously mentioned, the last major modification occurred with the passage of the Fair Credit and Charge Card Disclosure Act (FCCDA) 14 years ago. The FCCDA amended TILA and introduced the conspicuously placed Schumer box. Since that time, the Board of Governors has modified TILAs underlying regulation and regulation staff commentary to respond to some of these pricing changes, such as requiring larger point type, penalty rate disclosures, and cash advance and balance transfer APR disclosures.

34 Board of Governors Report to Congress, 1996, Finance charges for consumer credit under the Truth in Lending Act, April 35 Durkin, T. A., 2002, Consumers and credit disclosures: Credit cards and credit insurance, Federal Reserve Bulletin, April, 201-13. This includes Truth in Lending statements required for credit cards, home equity loans, and installment loans.

Real options

The interaction between real options and financial hedging in non-financial firms Mergers and acquisitions as a response to economic change Valuing real options: Frequently made errors Real options and flexibility Venture investment contracts as baskets of real options A real options approach to bankruptcy costs: Evidence from failed commercial banks during the 1990s Strategic investment under uncertainty: A survey of game theoretic real options models

The interaction between real options and financial hedging in non-financial firms1
Tom Aabo, Associate Professor, Aarhus School of Business Betty J. Simkins, Associate Professor, Department of Finance, Oklahoma State University
Financial hedging is often utilized to hedge transaction exposure (a nominal exposure tied to contractual obligations) and other short-term exposures caused by fluctuations in exchange rates. This article examines the hedging of operating exposure (sometimes called economic exposure, competitive exposure, or strategic exposure) through the use of real option strategies.2 Operating exposure is a real exposure and is concerned with the impact of unexpected changes in exchange rates on the operations of the firm and consequently on the operating profits, the cash flows, and the value of the firm. While financial hedges are well-suited to hedge short-term and certain exposures, hedging quickly becomes more complicated when managing operating exposure. In the long-run business conditions may change so that the underlying transaction may not materialize, and what was supposed to be a financial hedge turns into a financial speculation without underlying business rationale. Furthermore, when taking a long-term perspective, firms have other options than financial hedging. For example, a European firm that produces in Europe and exports its goods to the U.S. may react to a depreciating U.S. dollar by establishing production in the U.S. This real option strategy makes it possible for the European firm to regain its competitiveness. Accordingly, managers are not restricted to financial hedging when they are maneuvering in the minefield of operating exposure. An alternative is to change the set-up of the firm by exercising real options. The focus of this article is on the interaction between financial hedging and real options in managerial decision-making in non-financial companies. Shift production locations or factors of production A firm can react to exchange rate changes by varying the capacity utilization of different plants. Investing in production flexibility measures that facilitate production reallocation between countries can be profitable when cost fluctuations between countries are not identical. This real option is more valuable when the correlation between marginal costs (including exchange rates) in the two countries is low, the standardization of products is high, switching costs are low, and exchange rate volatility is high. Shift input sources across borders or between substitute inputs This is a switching option and is common in industries where production inputs are flexible. If adjustment costs (i.e., the cost to switch between suppliers) and lags (i.e., time lag in switching suppliers) are low, firms can expand purchases of inputs from suppliers in countries with beneficial exchange rates and reduce purchases from other suppliers. Abandon a foreign market Another real option strategy is the ability to abandon a foreign market when profitability is low due to such factors as low prices on outputs, high costs of inputs, or lack of competitiveness in the market relative to other producers. Effectively, this is the option to terminate all production and operations and sell the asset for its current market value. The abandonment decision is permanent unless the firm expects to return in the future. In the latter case the firm will face re-entry costs. Exploit growth options by entering a new foreign market or offering new products in an existing foreign market Investment in a foreign subsidiary network generates the real option to introduce new products in foreign countries. Firms with a presence in a country find it easier to introduce a new product than a firm without such presence. By establishing operations in a new country a firm also acquires the option to expand its product line at a later point in time.

Real option strategies

At least four real option strategies are relevant to non-financial firms that export and/or have foreign subsidiaries. These strategies are:

68

An extended version of this article is forthcoming in the Review of Financial Economics (2005). 2 Real options are options on real, or non-financial, assets. For example, a company that has real options has the right, but not the obligation, to make value-adding

managerial decisions through the strategic use of the firms operations. The real option approach applies financial option theory to real investments, such as manufacturing plants, product or business extensions, R&D investments, and global operations.

Firms in sample
This study is based on survey results from a representative sample of Danish, non-financial firms listed on the Copenhagen Stock Exchange as of the end of 2001. Denmark is a small (5.4 million inhabitants), open (exports as a percent of GDP = 45 percent) economy with a GDP per capita in 2001 of U.S.$30,000. The 2001 GDP per capita of the U.S. was 35,000 U.S. dollars. Denmark has been a member of the E.U. since 1973, but has not adopted the euro, although the Danish Krone (DKK) is tightly linked to it. A total of 117 Danish, non-financial firms listed on the Copenhagen Stock Exchange were mailed surveys in October 2001. The survey consisted of a questionnaire containing 17 closedended questions and was mailed to a target employee at each of the 117 firms. The target employees were primarily finance managers, but CFOs and treasurers were also surveyed. By the end of December 2001, 52 firms had sent in filled-out questionnaires, resulting in a total survey response rate of 44 percent. Using the Global Industry Classification Standard (GICS), 5 firms are from Materials, 19 firms from Industrials, 7 firms from Consumer Discretionary, 5 firms from Consumer Staples, 9 firms from Health Care, 6 firms from Information Technology, and 1 firm from Utilities. Except for Heath Care firms that are more likely to respond than the average firm, no non-response bias is detected. In other words, the sample should be representative for the population of Danish nonfinancial firms listed on the Copenhagen Stock Exchange. The average firm in our sample had a consolidated total assets of DKK 7,421 million (median DKK 1,756 million) in 2001. In U.S. dollar terms, this translates into average consolidated total assets of less than one billion dollars (8.41 DKK per U.S. dollar as of the end of 2001). Our average firm had foreign sales-to-total sales ratio of 61 percent (median 75 percent)

and had subsidiaries in 10 foreign countries (median 6 foreign countries).

Real options exercised

One of the main reasons for not hedging operating exposures by financial means is the inadequacy of such hedging to address the real problem of operating exposures. As an alternative, a firm may react to changes in exchange rates by exercising real options using the strategies previously discussed. Our survey of the decision-making process in Danish nonfinancial firms confirm that such considerations are made in a number of firms. More than one-half of the firms in our sample state that they sometimes or frequently do not hedge an operating exposure by financial means because the operating exposure is managed by real means.3 Various real options can be exercised when managing operating exposures caused by fluctuation in exchange rates. Figure 1 shows the real option strategies likely to be exercised by our sample firms within a time frame of two to three years and partly or fully due to the development in exchange rates. [Note: For brevity, hereafter, we use the words due to exchange rate changes in place of the actual words used in the survey partly or fully due to the development of exchange rates.] More than half of the firms (54 percent) indicated that it is
No Change sourcing between suppliers in different countries Buy a foreign firm (sales and/or production) Establish production in a country where your firm did not have production before Shift production between production outlets in different countries Sell a firm/close down a production subsidiary in a foreign country Enter a foreign market where your firm did not have sales before Abandon a foreign market (stop selling to that market) 46% 67% 69% 69% 75% 77% 81% Yes 54% 33% 31% 31% 25% 23% 19%

Figure 1: Real options likely to be exercised4

3 The question asked was: What are the likely reasons for your company to not hedge an operating exposure? The respondents were asked to prioritize different reasons. The most important reason was that the particular operating exposure was not significant. The second most important reason was: Operating exposure is managed by real means. Fifty-two percent of the firms stated that this reason was the first, second or third most important reason for not financially hedging an operating exposure.

4 The question asked was: A firm may react to changes in exchange rates by taking real actions (exercising real options) such as to reallocate production facilities, to abandon a market, to buy a foreign firm, etc. Is it likely that your firm within a time frame of two to three years and partly or fully due to the development of exchange rates would take the following actions? Respondents were asked to respond yes or no to the options listed.

69

35

Mean value = 2.2 real option strategies used

30

A large company should have the necessary resources to exercise real options. Furthermore, a company that has a large exposure to exchange rates (such as a high foreign sales-tototal sales ratio) should be more motivated to exercise real options. And finally, a company with subsidiaries in many countries should have a better ability to react to changing exchange rates by exercising real options. As a result, all three variables that we study (size, foreign sales ratio, and geographical coverage) are hypothesized to be positively correlated with real options usage. For our empirical tests, it is important to point out that several of the variables are highly correlated. The highest correlation coefficient (0.66) is between size (log of total assets) and geographical coverage (square root of the number of foreign countries in which the firm has subsidiaries). Large firms tend to have subsidiaries in more countries than small companies. Another high correlation coefficient (0.58) is between the foreign sales ratio (foreign sales/total sales) and geographical coverage. Firms that earn a lot of their revenues abroad also tend to have subsidiaries in several foreign countries. Finally, there is only a modest correlation (0.29) between the size of the company and its relative sales abroad. In order to minimize the problem of multicollinearity because of high correlations between the three variables (size, foreign sales ratio, and geographical coverage) and to test an alternative economic reasoning, we also construct a new integrated variable called multinationality.5 This new variable is a combination of the three variables. The economic reasoning is that when viewed separately, it is not size, nor foreign sales, nor foreign subsidiaries that determines the willingness and ability of a firm to react to changes in exchange rates by exercising real options. Rather, it is a combination of the three variables. A firm has to be large (ability) to have many foreign subsidiaries (ability) and to be exposed (willingness) in order to react to changes in exchange rates by exercising real options. We conduct regression analysis using an ordered probit regression to examine the significance of the four variables

25

Percentage of firms

20

15

10

0
1 2 3 4 5 6 7 8

Number or real option strategies used Figure 2: Number of real option strategies likely to be exercised

likely that they will change sourcing between suppliers in different countries. The equivalent likelihood of sample firms using other and more radical changes, such as buying or selling a foreign firm, is much smaller. The real option that is most unlikely to be exercised is to abandon a market (19 percent). Figure 2 illustrates the percentage of firms likely to use different real option strategies. As shown, the average number of real option strategies likely to be utilized within a time frame of two to three years due to exchange rate changes is 2.2. At one extreme, one third of the firms in this study (35 percent) do not think that it is likely that they will exercise any real options. At the other extreme, three firms (6 percent) think that all real options are likely to be exercised. All firms (except one) that indicated that only one option was likely to be exercised responded that this real option strategy was to change sourcing between suppliers in different countries.

Factors behind real options usage

The likelihood that real options are exercised within a time horizon of two to three years and in response to changes in exchange rates varies markedly among the firms in this study. At least three factors are candidates for explaining this variation: Size, foreign sales ratio, and geographical coverage of subsidiaries. We discuss this next.

70 - The

Journal of financial transformation

5 Multinationality is the square root of (log total assets times foreign sales/total sales times the square root of the number of foreign countries in which the firm has subsidiaries).

(i.e., size, foreign sales ratio, geographical coverage, and the combination variable, multinationality) in explaining the extent of real options usage.6 The dependent variable is the extent of options usage and ranges from 0 to 7 according to the number of real options likely to be exercised as shown in Figure 2. We also include economic sector variables in the regression analysis. Overall, while we do not find statistically significant coefficients for the three variables when analyzed separately, the combination variable multinationality, which captures the interaction between size, foreign sales and geographical coverage, is statistically significant in explaining the extent of options usage. The analysis suggests that the use of real options in relation to changes in exchange rates depends on the combined existence of a certain size, a certain exposure, and a certain range of foreign subsidiaries. In other words, just being a large firm, a firm heavily exposed to changes in exchange rates, or a firm with an extensive net of foreign subsidiaries is not sufficient. We also find that firms in Materials and in Consumer Discretionary tend to be statistically significantly more inclined to exercise real options in response to changes in exchange rates than firms in other sectors. However, due to the limited number of sample firms in each economic sector, we do not want to overemphasize this finding. Nevertheless, the finding that firms in Materials are more inclined to exercise real options in reaction to fluctuating exchange rates is supported by empirical evidence on U.S. firms. Furthermore,
7

In pursuit of flexibility
The ability to react to changes in exchange rates by exercising real options depends on the set-up of the firm. A firm that keeps contact and business relations with various suppliers in different countries should experience, at most, only minor obstacles when changing suppliers. Likewise, a company that has production facilities in several countries and excess capacity in these facilities should also experience, at most, only minor obstacles when moving production from one country to another. More generally, a firm with international experience and presence will be more flexible in exploiting the risk reducing and opportunity enhancing possibilities that the fluctuations in exchange rates create.

None Number of firms Percentage of firms Average number of real options likely to be exercised 20 39% 0.5

Partly 28 55% 3.3

Almost fully 3 6% 2.0

Figure 3: Extent to which flexibility is sought8

The level of flexibility is not just a by-product or something granted from above. It is also a result of discretionary actions taken by each firm. Figure 3 shows the extent to which sample firms pursue a strategy of being flexible. As we can see, 55 percent of the firms partly pursue a strategy of being flexible, 6 percent of the firms pursue a full (or almost full) strategy of being flexible, and 39 percent of the firms do not seek flexibility. There is a close relationship between the extent to which flexibility is sought and the likelihood that the firm will react to changes in exchange rates by exercising real options. While the firms that partly pursue a strategy of being flexible on average consider more than three real options likely to be exercised due to exchange rate changes, the corresponding figure for the group of firms that do not seek flexibility is as low as 0.5. The small group of firms that pursue a strategy of

economic intuition is supportive. The CIGS Materials economic sector includes commodity-related manufacturing industries, such as Chemicals, Construction Materials, Containers & Packaging, Metals and Mining, and Paper & Forest Products. In these industries firms face a high degree of uncertainty, especially in the form of volatile commodity prices. To survive in such an environment, firms must be able to react to risk factors by all means including the exercise of real options.

6 For more information on the ordered probit regression method, see Greene, William H. (2000), Econometric Analysis (4th Edition), Prentice Hall: New Jersey, pp 875-879 or Pindyck, Robert S. and Daniel L. Rubinfeld (1997), Econometric Models and Economic Forecasts (4th Edition), McGraw-Hill Inc: New York. 7 Carter, D., C. Pantzalis, and B. J. Simkins, 2003, Asymmetric exposure to foreignexchange risk: financial and real option hedges implemented by U.S. multinational corporations, Proceedings from the 7th Annual International Conference on Real Options: Theory Meets Practice, Washington D.C. The authors suggest that companies operating in industries with greater commodity-type inputs and outputs have a greater ability to employ real option strategies.

8 The question asked is: Exchange rates and business conditions change. In such circumstances, it may be profitable to improve flexibility partly at the expense of economies of scale (i.e. by having two minor production outlets in two different countries instead of one production outlet in one country; keeping contact with several suppliers in different countries instead of few suppliers in few countries; and by keeping options open by continuing to sell in markets that are not profitable at the present exchange rates). To what extent does your firm pursue a strategy of being flexible (at the expense of economies of scale)? The respondents were asked to choose between none, partly, or almost fully. One sample firm chose not to answer this question.

71

almost full flexibility illustrates that the picture of a clear linear relationship is not without deviations. The strong relationship between the extent to which flexibility is sought and the likelihood that the firm will react to changes in exchange rates by exercising real options is confirmed by regression analysis.

Conclusion
This study of the exchange rate exposure management of Danish non-financial firms shows that decisions on whether or not to financially hedge such an exposure is dependant on the ability of the firms to react to changes in exchange rates by exercising real options. Furthermore, the study shows that the combined interaction of a firms size, foreign sales, and foreign subsidiaries, as well as the firms managerial emphasis on flexibility, are important to real options usage. The results are important in understanding the degree to which a firm may react to changes in exchange rates by exercising real options. Although the conclusions are based on Danish firms, we believe that the generality of the findings can be extended to a wide range of firms in countries with open economies.

72 - The

Journal of financial transformation

Mergers and acquisitions as a response to economic change

Bart M. Lambrecht Professor of Finance, Lancaster University Management School
What causes mergers and acquisitions (M&A)? This is an issue that still remains to some extent a puzzle. While we can, of course, come up with a story or motive for each individual merger or takeover case, finance theory has not quite managed yet to formulate a coherent theory that is generally applicable. As in the first merger movement, the second wave of mergers also began with an upturn in business activity in 1922 and ended with the economic slowdown in 1929. As for the motivational factors Markham (1955) emphasized major developments in transportation (more consumer mobility through motor vehicles and automobiles), communication (more product differentiation through national-brand advertising with the rise of home radios) and merchandising (mass distribution with low profit margins). These developments caused an increase in the scale of operations and hence encouraged mergers. The second wave consisted mainly of vertical mergers. The advantages were related to technological economies, such as shortening processes or reliability of input supply. Merger activity reached its then historically highest level during the three year period of 1967 to 1969. The period was also one of a booming economy. The number of mergers declined sharply as the economic activity slowed down from 1970 onwards leading ultimately to the recession linked to the 1973 oil crisis. Most mergers in the sixties were conglomerate mergers.2 of scale with respect to transportation costs and the development of a national economy.

Mergers as a response to economic change

One good starting point for the development of a reasonable theory is to consider first historical and empirical evidence. When one considers M&A activity over the past century then a first striking observation is that it is not constant over time. Instead M&A activity seems to occur in waves. There are periods of high and low takeover activity. This leads us to the question why there are regime shifts in takeover activity. The most plausible answer is that a change in takeover activity is caused by a change in the economic environment within which firms are operating.1 This brings us to the key proposition of this article, namely the idea that M&A activity occurs in response to economic change. This inevitably prompts the next question: What kind of economic change causes M&A activity? Historical and empirical evidence may again shed some light on the matter.

Merger waves
There were five main merger waves in the 20th century. These waves typically coincided with economic expansion, booming stock markets, and fundamental changes to the economic environment, often under the form of technological innovation leading to higher economies of scale and larger geographical markets. The merger movement in the U.S. at the turn of the century (1895-1904) consisted mainly of horizontal mergers and coincided with a period of rapid economic growth which ended with the onset of an economic recession in 1903. Weston et al. (1990) comment that this wave was driven by major changes in economic infrastructure and production technologies, such as the advent of electricity and the completion of the transcontinental railroad system. The latter led to economies A fourth peak of merger activity was reached during the years 1988 and 1989. This was the era of raiders and leveraged buyouts. The fifth merger wave started in 1993 and slowed down after the collapse of the Internet bubble in 2000. The mergers boom in the eighties and especially the nineties coincided with strong economic growth, rising stock markets, and advances in telecommunication and computer technology. These technological advances resulted in an economy where firms were operating on a global level. The advent of the Internet brought with it the so called new economy and gave the concept of returns to scale a new dimension. Globalization has surprisingly not led to a substantial number of cross-border acquisitions. Improvements in technology and communications stimulated takeover activity particularly in banking, insurance, media, and telecommunications.

An alternative explanation, which we do not pursue in this paper, might be that merger waves are driven by some kind of herding behavior. The origin of this herding behavior could be informational or behavioral in nature.

2 Ravenscraft and Scherer (1987) note that many mergers and acquisitions during the 1960s were disappointing and were subsequently broken up in the 1980s.

73

While most of the above discussed economic shocks seem to be affecting the economy as a whole, this is not necessarily always the case. Regulatory changes, for example, may affect one particular industry and trigger a series of mergers in that industry alone. For example, in 1999 the Financial Services Modernization Act in the U.S. repealed the 1933 Glass-Steagall Act that forced investment banks, commercial banks, and insurance companies to be separate entities. This regulatory change may have contributed to the recent wave of mergers in the banking industry. Empirical evidence confirms that merger activity is related to shocks to the economic environment in which firms are operating. Mitchell and Mulherin (1996) studied industry-level patterns in takeover and restructuring activity during the 19821989 period. Across 51 industries, they find significant differences in both the rate and time-series clustering of these activities. They find that the inter-industry patterns in the rate of takeovers and restructurings are directly related to economic shocks borne by the sample industries. Theoretical models attempting to explain takeover waves are very rare. One notable exception is Gorts (1969) economic disturbance theory. Gort (1969) argues that at certain times shareholders have differing opinions as to the true value of a share because of imperfections in the information available and how it is assessed. These differences in valuation lead to takeover transactions. According to Gort, valuation differences are greater at times of dramatic change, such as rapid movements in stock prices, changes in technology, and in the relative price of energy.3 Brealey and Myers (2002) acknowledge that takeovers may result from mistakes in valuations on the part of the stock market, but argue that mistakes are made in bear markets as well as bull markets, and wonder therefore why we do not see just as many firms hunting for bargain acquisitions when the stock market is low. They conclude that it is possible that suckers are born every minute, but it is difficult to believe that they can be harvested only in bull markets (p. 967).

The pro-cyclicality of merger activity

The above argument by Brealey and Myers points to another empirical feature of M&A activity. Apart from the fact that mergers seem to occur in waves, a second notable feature is that merger activity is pro-cyclical, in that we observe more mergers during economic booms than during economic recession. A recent paper by Maksimovic and Philips (2001) provides further evidence on the procyclicality of merger waves. Analyzing the market for corporate assets, they find that over the period 1974-1992, on average, 3.89% of plants change ownership each year. In expansion years, close to 7% of plants annually change ownership. They found mergers and acquisitions to be strongly procyclical, rising in periods of economic expansion and falling during recessions. There is still very little theoretical evidence as to why merger activity should be procyclical. One exception is Lambrecht (2004) who analyzes the case of mergers motivated by economies of scale. In his model, the surplus (synergies) that firms receive from merging is generated by economies of scale: by combining their production facilities firms can produce more than the combined production when they each operate individually. The extra units of output that are created by merging can be sold at a stochastic product price. The surplus from merging is therefore not only stochastic, but it is also procyclical: it rises (falls) in periods of high (low) product market demand. Upon merging, each firm incurs a fixed merger cost. Since this is a sunk cost it introduces an element of irreversibility into the merger, and the decision to merge is therefore similar to the exercise of a call option. When merging, both companies therefore have to trade off the stochastic benefit of merging against the cost of merging. Since both firms have the right, but not the obligation to merge, each firms payoff resembles an option and the decision to merge resembles the exercise of an option. The higher profits that firms forgo by not merging act as an incentive to exercise this option, while the (at least partially) irreversible nature of the merger acts as an incentive to delay. The optimal merger timing strikes a balance between

74 - The

Journal of financial transformation

3 A more recent paper by Shleifer and Vishny (2001) also assumes that the stock market may misvalue potential acquirers, potential targets and their combinations. Managers of the firms understand stock market inefficiencies, and take advantage of them, in part through merger decisions. The takeover surplus and merger waves are driven by the relative valuations of the merging firms.

the two. Since the gains from mergers motivated by economies of scale are positively correlated to product market demand, mergers happen in rising product markets. This creates merger activity at high output prices and merger inactivity at low output prices. Cyclical product markets will therefore generate a pattern of merger waves with mergers being procyclical. Lambrecht (2004) also examines whether mergers take place at the efficient time when both firm act in a non-cooperative way. Each party holds a call option on a fraction of the enlarged firm (i.e. the ownership share each party has in the new, combined firm) with the strike price being the sum of its fixed merger cost and the standalone value of its existing firm. The merger can only take place if both parties agree on the timing of the merger (i.e. at what product price level the merger should happen) and the terms of the merger (i.e. the postmerger ownership share of each party in the new firm). It follows that, unlike financial options, the exercise of merger options is also influenced by strategic considerations since the payoff to each firm ultimately depends on the post-merger ownership share it obtains in the new firm. The restructuring mechanism (i.e. how the merger gains are divided up) can therefore also influence the timing of the restructuring. The paper considers two different mechanisms, namely friendly mergers and hostile takeovers, and shows that ceteris paribus they should occur at different stages in a merger wave. Since mergers are modelled as an investment decision, the model also allows us to analyze the determinants of this decision. For example, the model predicts that a lower interest rate, higher synergy benefits, higher economic growth, and reduced economic uncertainty (volatility) speed up and stimulate merger activity. This is consistent with empirical evidence. Melicher et al. (1983) examining the period 1947-1977 find support for viewing changes in aggregate merger activity as a capital market phenomenon. In particular, increased merger activity is associated with higher stock prices and lower interest rates.

Counter-cyclical takeover activity

While the above discussion emphasized the fact that merger waves are pro-cyclical, this does not mean that no takeover activity takes place in economic downturns [as was previously illustrated by the empirical evidence of Maksimovic and Philips (2000)]. However, the type of takeovers in booms versus recessions may be quite different. While in booms takeovers may be expansive in nature, during recessions takeovers are more likely to be consolidating or contractive in nature. For example Lambrecht and Myers (2004) study a category of takeovers that happen during downturns, namely takeovers motivated by agency problems and disinvestment. They argue that when the market declines firms may have certain assets that become unproductive and that ought to be sold off with the proceeds going to shareholders. Management may, however, be reluctant to do this and may prefer to keep the assets within the firm (even if these assets do not provide shareholders with a sufficient return). Those inefficiencies create a role for takeovers. A raider or hostile acquirer could take over the firm and create value by selling off some of the firms assets, returning the proceeds to shareholders, and by putting the firm on a diet deal. While the takeovers studied by Lambrecht and Myers (2004) are primarily driven by agency problems, one can, of course, also observe takeovers in downward markets that are driven by other considerations such as cost cutting, taxes, misvaluation, or where takeovers are an alternative to (costly) bankruptcy. Much research still needs to be done in those areas.

Conclusion
Empirical evidence suggests that merger activity happens in response to economic shocks. Furthermore, mergers tend to happen in waves and are pro-cyclical: merger activity is higher during economic booms than during recessions. To date there is still no theory that fully explains merger activity.

75

References
Brealey, R. A., and S. C. Myers, 2002, Principles of Corporate Finance. McGraw-Hill, Boston, MA, seventh edition. Gort, M., 1969, An economic disturbance theory of mergers, Quarterly Journal of Economics, 83:4, 624-642. Lambrecht, B. M., 2004, The timing and terms of mergers motivated by economies of scale, Journal of Financial Economics, 72:1, 41-62. Lambrecht, B. M., and S. C. Myers, 2004, A theory of takeovers and disinvestment, University of Lancaster, Lancaster. Maksimovic, V., and G. Philips, 2001, The market for corporate assets: Who engages in mergers and asset sales and are there efficiency gains? Journal of Finance, 56:6, 2019-2065. Markham, J. W., 1955, Survey of evidence and findings on mergers, in Business Concentration and Price Policy. Princeton University Press, Princeton, New Jersey Melicher, R. W., J. Ledolter, and L. J. DAntonio, 1983, A time series analysis of aggregate merger activity, Review of Economics and Statistics, 65:3, 423-430. Mitchell, M. L., and H. Mulherin, 1996, The impact of industry shocks on takeover and restructuring activity, Journal of Financial Economics, 41, 193-229. Ravenscraft, D. J., and F. Scherer, 1987, Mergers, sell-offs, and economic efficiency, The Brookings Institutions, Washington, D.C. Shleifer, A., and R. W. Vishny, 2003, Stock market driven acquisitions, Journal of Financial Economics, 70:3, 295-311. Weston, J. F., K. S. Chung, and S. E. Hoag, 1990. Mergers, restructuring, and corporate control. Prentice Hall International, Englewood Cliffs, New Jersey.

76 - The

Journal of financial transformation

Valuing real options: Frequently made errors

Pablo Fernndez, PricewaterhouseCoopers Professor of Corporate Finance, IESE Business School University of Navarra
There are many approaches used to valuing real options, many of which have serious inherent problems. The objective of this paper is to highlight some of these errors. Perhaps the best way of doing so is through an example. I will, therefore, analyze Damodarans proposal to value the option to expand the business of Home Depot and discuss some of the errors and problems of this and other approaches. The formulas used to value financial options are based on riskless arbitrage (the possibility of forming a portfolio that provides exactly the same return as the financial option) and are very accurate. However, we will see that very rarely does it make sense to use these formulas directly to value real options because real options are hardly ever replicable. However, we can modify the formulas to take non-replicability into account. Some problems we encounter when valuing real options are the difficulties faced in communicating the valuation due to its higher technical complexity than the present value, in defining the necessary parameters for valuing real options, in defining and quantifying the volatility of the sources of uncertainty, in calibrating the options exclusiveness, and in valuing the options adequately. In any case, their valuation is much less accurate than the valuation of financial options.
Contractual options Oil concessions Mining concessions Franchises Growth or learning options Expand Research and development Acquisitions New businesses New customers Internet venture Greater efficiency in increasing entry barriers Figure 1: Real options Flexibility options Defer the investment Downsize the project Alternative uses Renegotiations of contracts Outsourcing Abandon Modification of products

has the option of expanding its production facilities or canceling distribution, depending on the markets future growth. Investments in research and development can also be analyzed using options theory [Grenadier, S. and A. Weiss (1997)]. Corporate policy strategists and professors have repeatedly reproached finance and financial analysts for their lack of tools for valuing investment projects strategic implications. Before using options theory, most new investments were made solely on the basis of qualitative corporate policy criteria. The numbers, if any, were crunched afterwards so that they could give the results that the strategist needed to back his decision. Options theory seems to enable projects strategic opportunities to be valued. By combining quantitative analysis of the options with qualitative and strategic analysis of the corporate policy, it is possible to make more correct and more rational decisions about the firms future. Not considering the options contained in a project may lead us to undervalue it and, in general, turn down projects that we should undertake.1 One classification of real options is provided in Figure 1.

Real options
It is not possible to correctly value a firm or a project that provides some type of future flexibility real options using the traditional techniques for discounting future flows (NPV or IRR). There are many types of real options: options to exploit mining or oil concessions, options to defer investments, options to expand businesses, options to abandon businesses, options to change the use of certain assets, etc.

People also talk about compound options, which are those A real option exists in an investment project when there are future possibilities for action and when the solution to a current uncertainty is known. Oil concessions are a typical example. The oil well will be operated or not depending on the future price of oil. Designing a new product is also a real option. A firm that provide new options when they are exercised. Rainbow options is the term used to describe those that have more than one source of uncertainty, for example, an oil concession in which the uncertainty arises from the price of oil, an uncertain quantity of barrels, and uncertain extraction costs.2

Similarly, if the projects we are considering contain options that may be exercised by third parties (the future flexibility plays against us), non-consideration of the options contained by the projects will lead us to invest in projects that we should turn down.

2 For a compilation of the different types of real options, see the books published by Trigeorgis (1996), and Amram & Kulatilaka (1999), both with the same title: Real Options.

77

Frequently made errors when valuing real options the example

Home Depot is considering the possibility of opening a store in France. The stores cost will be 24 million and the present value of the expected cash flows is 20 million. Consequently, the projects value will be 4 million and it would not be a good idea. However, Home Depot believes that by opening this store, it will have the option to open another larger store in the next 5 years. The cost of the hypothetical second store would be 40 million and the present value of the expected cash flows is 30 million, although there is a lot of uncertainty regarding this parameter. Home Depot estimates the volatility (s) of the present value of the expected cash flows of the second store at 28.3%. Damodaran (2000) proposes valuing the option to open the second store using Black and Scholes formula. According to him, the option of opening the second store is a call with the following parameters: Option of opening the second store = Call (S=30; K=40; r = 1.06; t = 5 years; = 28.3%) = 7.5 million Consequently, according to Damodaran, Home Depot should open the store in France because the projects present value plus the value of the option to expand is 4 + 7.5 = 3.5 million. Some of the errors and problems of this approach are:
Assuming that the option is replicable This is why Black

obvious that the option to open a second store is not replicable.3

The estimation of the options volatility is arbitrary and

has a decisive effect on the options value Damodarans hypotheses regarding volatility (28.3%), present value of the expected cash flows (30 million), the options life (5 years) and the options replicability ( = ln(r) 2/2 = 1.82%) are synthesized in the distribution of the expected cash flows in 5 years time shown in Figure 2.4 It is obvious that a volatility of 28.3% per year means assuming an enormous scatter of cash flows, which is tantamount to having no idea what those cash flows may be. One thing is that a greater uncertainty increases real options value and another altogether that real options may have a high value (i.e. must undertake projects) because we have not the slightest idea of what may happen in the future. Figure 2 also shows the shape of two distributions with annual volatilities of 15%.
As there is no riskless arbitrage, the value of the option

to expand basically depends on Home Depots expectations about future cash flows Damodaran assumes that this parameter does not influence the options value because he assumes that the option is replicable.
It is not appropriate to discount the expected value of

the cash flows at the risk-free rate (as is done implicitly when Black and Scholes formula is used) Although a real option will be exercised when a future uncertainty is settled (in this case, if the first store is a success), this
0,04
= 1,82%, = 15%

and Scholes formula is used in the valuation. It is fairly

does not mean that it is a risk-free project. The present value of the cash flows (30 million in the above example) is calculated using a rate that reflects the estimated risk today. Once the outcome of the first store is known, if it is
= 8%, = 15%

0,03
= 1,82%, = 28,3%

0,02

a failure, the second store will not be opened; if it is a success, the second store will be opened, but the project of opening the second store will still have risks: the uncer-

0,01

0,00
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80

tainty of costs and sales in five years time may be greater or less than that estimated today. Therefore, the cash flows must be discounted at a rate greater than the riskfree rate.

Value of expected flows in year 5 (million euros) Figure 2: Distribution of the expected cash flows in 5 years time according to Damodaran

78

3 To get around the non-replicability issue, Amram and Kulatilaka define real options as the subset of strategic options in which the decision to exercise the option is basically determined by financial instruments or assets traded on markets. The problem is that, according to this definition, only a few oil and mining concessions would be real options. See page 10 of Amram and Kulatilaka (2000).

4 Another way of expressing the scatter is that Damodaran assumes that the value of the expected cash flows in 5 years time will lie between 22 and 79 with a probability of 66%; and between 12 and 149 with a probability of 95%.

 Damodarans valuation assumes that we know exactly

If they are not replicable, by any of the above methods but

the cost of opening the second store and that it will be 40 million Obviously, there is uncertainty as to how much it will cost to open a store in five years time. The formula used assumes that the risk of the opening cost is equal to the risk of the cash flows generated by opening the store, which is not entirely correct. Normally, the cash flows generated by opening the store will have a greater risk than the opening cost and should be discounted at a higher rate.
Another error is to assume that the options value

taking into account the non-replicability. For example, it is not possible to apply Black and Scholes formula but rather a modified formula that takes into consideration the expected value of the uncertainties and uses a discount rate higher than the risk-free rate [Fernandez (2002, chapter 22)]. The factors that determine the value of a financial option are different from those that affect a real option. These differences in the parameters are shown in Figure 3. If the real options cannot be replicated, using financial option formulas is completely inappropriate for valuing real options, as all the formulas are based on the existence of a replicate portfolio. The logic of options theory is based on arbitrage: as it is possible to form a replicate portfolio that will have exactly the same return as the option we are trying to value, (in order to avoid arbitrage) the option must have the same value as the replicate portfolio. If it is not possible to form the replicate portfolio, this reasoning loses its entire basis. In the following paragraphs, we have included a number of

increases when interest rates increase For example, Leslie and Michaels (1997) say that an increase in interest rates increases the options value, in spite of its negative effect on the net present value, because it reduces the present value of the exercise price. This is wrong because the negative effect of increased interest rates on the present value of the expected cash flows (as on the value of shares) is always greater than the positive effect of the reduction of the present value of the exercise price.

Applying options theory in a firm

Real options can be valued using the following methods:
If they are replicable, use Black and Scholes formula, the

considerations on the practical application of options theory to the analysis of investment projects.
High interest rates mean high discount rates, which

formulas developed for valuing exotic options, by simulation, the binomial formula, or by solving the differential equations characterizing the options.5

reduces the present value of future flows. Obviously, this

Financial call option Share price Exercise price Risk-free interest rate Shares volatility Time to exercise Dividends Its value does not depend on the expected appreciation of the underlying asset Real call option Expected value of cash flows Cost of investment Discount rate with risk Volatility of expected cash flows Time to exercise Cost of holding the option Its value does depend on the expected appreciation of the underlying asset

should decrease the value of the option to undertake a project. However, high discount rates also reduce the present value of the options exercise price. This compensatory effect helps sustain the options value when interest rates increase, which may give certain types of projects particularly growth options an enormous value that should be taken into account when analyzing investments.
Kester (1984) suggests one feature of options that should

be considered, the extent to which the holder of an option has an exclusive right to exercise it. Unlike share options, there are two types of growth option, exclusive and

Figure 3: Parameters that influence the value of a financial option and of a real option

5 It is important to realize that Black and Scholes formula interpreted as net present value considers = 0.379% = ln(r) 2/2 and discounts the options expected value E[Max(S-K,0)] with the risk-free rate r.

79

shared. The former are the more valuable because they give their holder the exclusive right to exercise them. These options derive from patents, unique knowledge of the market held by the firm, or technology that its competitors cannot imitate. Shared growth options are less valuable. They represent collective opportunities held by the industry, such as, for example, the possibility of entering a market that is not protected by high entry barriers or of building a new factory to supply a particular geographical segment of the market. Cost reduction projects are normally shared options, because, as a general rule, they can also be undertaken by competitors.
Kester also suggests that when analyzing investment

The firm must separate the projects that require an

immediate decision on the entire project from those in which there is decision flexibility in the future. Finally, the firm must ask itself if it can realize all the options benefits or whether they will also be available for other competitors.
When examining investment opportunities from the

option valuation viewpoint, managers will find it easier to recognize that: (a) the conventional NPV may undervalue certain projects by eliminating the value of the options already existing in the project; (b) projects with a negative NPV can be accepted if the value of the option associated with future flexibility exceeds the NPV of the projects expected cash flows; and (c) the extent of the undervaluation and the degree to which managers can justifiably invest more than what the conventional rules regarding the NPV would indicate can be quantified using options theory.6
The options framework indicates that the value of the

projects, firms should classify the projects in accordance with the options they include. The classification using the traditional criteria of replacement, cost reduction, capacity increase, and new product introduction is not very useful. A more appropriate classification would be to distinguish between projects whose future benefits are mainly generated through cash flows (simple options) and those that include subsequent investment options (compound options). Simple growth options, such as routine cost reductions and maintenance and replacement projects, only create value through the cash flows generated by the underlying assets. Compound growth options such as research and development projects, a major expansion in an existing market, entry in a new market, and acquisitions (of new businesses or firms) lead to new investment opportunities and affect the value of the existing growth options. Given their complexity, their role in giving shape to the firms strategy, and their impact even on the organizations survival, compound options require a deeper analysis. The firm must view these projects as part of a larger group of projects or as a series of investment decisions that follow a time continuum. In light of the firms strategy, its executives must ask themselves whether a particular option will provide suitable investment opportunities in the appropriate markets, within a suitable time frame, that are matched to their firms needs.

managements future flexibility is greater in more uncertain environments. This value is greatest in periods with high interest rates and availability of the investment opportunities for extended periods. Consequently, contradicting generally held opinion, greater uncertainty, high interest rates, and more distant investment horizons (when part of the investment can be deferred) are not necessarily harmful for an investment opportunitys value. Although these variables reduce a projects static NPV, they can also increase the value of the projects options (value of management flexibility) to a level that may counteract the previous negative effect.
A real option will only be valuable if it provides a sustain-

able competitive advantage. This competitive advantage basically depends on the nature of the competitors (normally, if competition is fierce and the competitors are strong, the advantages sustainability will be less) and on the nature of the competitive advantage (if it is a scarce resource, for example, scarce buildable land, the advantages sustainability will be greater).

80 - The

Journal of financial transformation

6 For a good study on the application of real options to mining companies, see Moel and Tufano (2002).

References
Amram M., and N. Kulatilaka, 2000, Strategy and shareholder value creation: The real options frontier, Journal of Applied Corporate Finance, 13:2, 8-21 Amram, M. and N. Kulatilaka, 1999, Real Options, Harvard Business School Press Damodaran, A., 2000, The promise of real options, Journal of Applied Corporate Finance, 13:2 Damodaran, A., 1999, The promise and peril of real options, Working Paper, Stern School of Business Fernandez, P., 2002, Valuation and shareholder value creation, Academic Press. San Diego, CA Grenadier, S. and A. Weiss, 1997, Investment in technological innovations: An option pricing approach, Journal of Financial Economics, 44, 397-416 Kester, W. C., 1984, Todays options for tomorrows growth, Harvard Business Review, March-April, 153-160 Leslie, K. J. and M. P. Michaels, 1997, The real power of real options, The McKinsey Quarterly, Number 3, 5-22 Luehrman, T. A., 1995, Capital projects as real options: An introduction, Harvard Business School, 9-295-074 McDonald R., and D. Siegal, 1986, The value of waiting to invest, Quantitative Journal of Economics, 101, 707-727 Tufano, P., and A. Moel, 2002, When are real options exercised? An empirical study of mine closings. Review of Financial Studies 15:1, 35-64 Margrabe, W., 1978, The value of an option to exchange one asset for another, Journal of Finance, 33:1, pp. 177-198. Trigeorgis, L., 1996, Real Options, The MIT Press

81

Real options

Real options and flexibility

Thomas E. Copeland
Managing Director of Corporate Finance, Monitor Group, and Senior Lecturer, Sloan School of Management, MIT

Abstract
Never heard of real options? It is a new technique for making large investment decisions that will replace Net Present Value as the dominant tool. In a survey of 4,000 U.S. based CFOs, fully 27 percent said that they always or almost always used it for important capital expenditures. Real options captures the value of flexibility and gives you trigger points that tell you when to take action to develop an oil field, to stop a research and development effort, to get out of a business, to expand a business, or to turn on and off a peak load power plant. This article gives several examples of how real options have turned the tide of sentiment for or against strategic investments.

83

Real options and flexibility

A better decision-making tool is invented only a few times each century, and each advance goes farther to capture human intuition and sometimes to improve it. Every year trillions of dollars are either invested in major capital projects or withheld from investment. Net present value (NPV) is the tool that is currently used most often for these decisions. It has been recommended for 50 years by graduate schools of management throughout the world. But now it is being replaced by a new tool called real options analysis (ROA). This article explains why CFOs are changing from NPV to ROA, gives some ROA examples based on actual business applications, and discusses some of the new implications of ROA for the financing of companies.

evolved in the future. After all, that is what it means to manage something. All of this flexibility is ignored when one estimates the expected cash flows of a project. Depending on the winds of fortune the project may do better than expected. If so, management can change course and spend additional funds to expand the project to take advantage of greater demand, or to extend its life. These actions are conditional on favorable states of nature within which they will be exercised, because the value captured, given the state of nature, is greater than the cost of the additional investment (the exercise price of a real option). In less favorable states of nature they will not be exercised. What we have just described are two types of real option

Why real option analysis is replacing NPV

In a recent survey of 4,000 chief financial officers of U.S. companies, when asked whether they were familiar with real options, fully 27% of the respondents answered that for major capital investment decisions they always or almost always used real options [Graham and Harvey (2001)]. Industries where the use of ROA is common include aerospace, oil and gas, computers, pharmaceuticals, high tech, and power generation. They are switching away from NPV because it fails to capture flexibility of decision-making. All it does is estimate the expected future cash flows of the project (including its operating profit after taxes, depreciation tax shield, capital expenditures, and working capital requirements), discounts them back to the present at the weighted average cost of capital, and subtracts the initial capital outlay. If the NPV is negative, executives who review the project are supposed to reject the capital spending request. But in my years as a professor and consultant, I have seen many negative NPV projects accepted by senior management for so-called strategic reasons. In other words, the wisdom of experienced managers outweighed the quantitative result of the NPV technique. One senior executive explained his objection to NPV this way. In his role as manager, he had the flexibility to manage the risk of the project by steering it, or changing its course, as it

expansion and extension of the life of a project. A real option is the right, but not the obligation, to take an action in the future contingent upon information that resolves (completely or partially) an uncertainty. Most projects have call options like expansion or extension to capture the value of an opportunity by paying an additional amount of money, called the exercise price. They also have put options to shrink a project or to abandon it altogether and receive an amount of cash, also called the exercise price. Finally, many projects can be deferred and that too is a valuable option. Every project has at least five real options, five types of flexibility, built in. The NPV methodology ignores all of them. Therefore, after over 30 years as a teacher of corporate finance at UCLA, NYU, MIT, and Harvard, I have reached the point of view that the NPV method provides an estimate of the value of a project without flexibility, and consequently it systematically undervalues every project, with the only question being by how much. For high positive NPV projects the value of flexibility may be low simply because the project should move forward as quickly as possible, no optionality is necessary. For highly negative NPV projects, no amount of flexibility can save them. But for projects that are close decisions, their flexibility can swing the answer from negative to positive. I have seen real options change the answer several hundred percent.

84 - The

Journal of financial transformation

Real options and flexibility

One of the first real option cases that I worked on serves to illustrate another shortcoming of NPV, namely that it forces one to compare false mutually exclusive alternatives. An Australian company was about to bid on a government lease that would give the winning bidder the exclusive right to develop and extract coal from a proven reserve. Like most leases of this type, development did not have to start immediately but if it did not start within five years, the lease would revert to the government. At the time of bidding the price per ton of coal was only U.S.$1 higher than the extraction cost, so no one was sure that the lease property should be developed immediately. A one dollar drop in the price could wipe out all profit while a one dollar increase would double it. The company analyzed five mutually exclusive alternatives using the NPV method, namely develop immediately, after one year, after two, and so on. It estimated that the highest value alternative was worth about U.S.$59 million. However, rumor had it that this bid would be too low. Upon rethinking the problem, the management decided to use real option analysis because they were paying not only for the developed lease but also for a five year deferral option that would give them the right to first see how the price of coal would change and then decide whether to develop the lease. Given the volatility of coal prices the project with the right to defer turned out to be a little over U.S.$100 million. Figure 1 contrasts the difference between NPV and real options for the coal lease example. NPV forecasts the expected price of coal as it grows over time and uses it for the starting point of each of the five mutually exclusive alternatives.

Notice that since the spread between the price of coal and the cost of extraction is positive now and since it continues to grow, all 5 NPV estimates are positive. ROA is a more sophisticated approach because it models not only the expected price but also the variability of that price. As the price goes up or down, the value does too. When the valuation is done we start at the back of the tree, i.e. in year 5, and determine what states of nature would have a coal price high enough to develop the lease property. These are the states where we would no longer defer (D), but would exercise our option to invest (I). In all other states we would defer any investment. Next, we would move to year 4 and examine the deferral decision in each state deciding whether we are better off deferring or investing. In this way we work back in the tree to its root where we have an estimate of the value of the project with the right to defer as well as decision rules made up of trigger prices that would cause us to invest in a given year. NPV logic forces us to assume that we develop for sure in a given year then compares the resulting mutually exclusive alternatives. ROA provides a single value, not 5, and that value represents the value of the project with the flexibility to defer the investment until the time when the price of coal goes high enough to trigger the exercise of our option by spending the money to develop the leased property. NPV fails because it forces false mutually exclusive choices.

Examples of real options with different types of flexibility

An example of a real call option nearly 2500 years ago

Mutually exclusive choices NPV if developed immediately NPV if developed in year 1 NPV if developed in year 2 NPV if developed in year 3 NPV if developed in year 4

ROA tree I I I D D D

Historians have found examples of real options in writings as old as the 4th century B.C. by the Greek philosopher Aristotle.
D D D I D D D I I D D D

He did not have the advantage of modern mathematics such as Itos lemma, but he did know a real option when he saw one. His story is about Thales the Milesian who read the tea leaves and interpreted them as predicting a bountiful olive harvest that year. Being a man of action, he took his meager savings and offered it to the owners of olive presses, which were used to extract the olive oil from the olives, in return for the right to

MAX NPV < ROA

D D

Key D = Defer I = Invest Figure 1: The ROA value will always be higher than NPV

85

Real options and flexibility

rent the presses at the normal rate during harvest. When they agreed, he had just purchased historys earliest recorded call option. It gave him the right, but not the obligation to purchase time on the presses at a fixed price. Later on, when the harvest was truly plentiful, the olive farmers rushed to the presses. Thales charged them a price that was higher than normal due to the unprecedented demand, paid the normal rental rate to the owners of the presses, and kept a large profit. His option had finished in-the-money and he became a wealthy man. We can see, in this simple example, all of the general factors that affect the value of a real option. First one must identify the underlying risky asset and its uncertainty. In this case it was the rental price of the olive presses and the annual variability of the rental rate. Although this uncertainty is driven by the unavailability of the olive harvest we must be careful to measure the standard deviation of the rental price of the presses and not the variability in the quantity of olives. The value of the option increases when the price of the underlying asset goes up, but it also goes up when the variance of the underlying goes up. The exercise price was the normal rental rate. Higher exercise prices decrease the value of a call option (the right to buy) and increase the value of a put option (the right to sell). The life of the option was the time between Thales purchase of the option and the date of the olive harvest, and the value of the option increases with longer life. Finally, the value of the option increases when the risk-free rate does. Modern compound real options Most real options are options on options, called compound options. They are much more complex than the original BlackScholes (1973) and Merton (1973) models that started the modern science of option pricing. To contrast them, BlackScholes prices a European call that can be exercised only at maturity, on an underlying risky asset that pays no dividends or cash flows of any kind, and that is driven by a single source of uncertainty. An example of a compound option is the phased construction of a large manufacturing facility. There

are usually several types of uncertainty, such as price, quantity sold, manufacturing cost per unit, and technology. The project pays and requires cash flows. There are multiple phases, such as design, engineering, pre-construction, and final construction. It has taken three decades since the original work of Black, Scholes, and Merton to develop lattice methods for solving these problems but there is literature to show the way [Copeland and Antikarov (2003)]. Examples of compound real options include any phased investment where each phase is an option and is contingent on other options. Research and development programs have multiple phases, so does oil and gas exploration and development, any new product development, phased construction, and merger and acquisition programs. A very simple example of a compound option is given in a Harvard Business Review article that I co-authored with Professor Peter Tufano entitled Real ways of managing real options. It is a simplification of the Xylenes Basement case that I wrote [Copeland (2002)]. The setting for the case is a commodity chemical company that is contemplating the construction of a PTA plant that will cost U.S.$1.26 billion in three stages: a design phase costing U.S.$60 million and payable immediately and requiring one year to complete, an engineering phase that will cost U.S.$400 million and take 2 years, and final construction that will cost U.S.$800 million. A traditional NPV analysis values the project at U.S.$260 million but assumes that all three phases will be completed lockstep one after the other without turning back. However, the company recognizes that the project is really a compound option with three phases and that it can decide to defer or abandon the project after each phase. The decision about whether to go into the design phase depends on the payouts of the engineering phase, which in turn depends on the payouts of the construction phase. Therefore the initial decision is an option on an option on an option, a three level compound option. Managements decision is driven primarily by two uncertainties, the price of the output commodity chemical PTA, and the cost of the input commodity chemical, p-xylene, and the cor-

86 - The

Journal of financial transformation

Real options and flexibility

relation between them. When flexibility to defer or abandon was taken into account, the value of the project increased to U.S.$71 million and management decided to start into the initial phase. We recently helped a high-tech company address the trade-off between a large, very efficient, but very expensive plant (several billion dollars to build) and a less efficient plant but one that provided more flexibility. The following, stylized example illustrates the essence of the idea, but is not exactly the same. Suppose a company in a high tech industry anticipates rapidly growing demand for its product 300,000 units in the first year, 600,000 in the second, and 900,000 units in the third year (then no demand from year 4 on). Demand is very uncertain and could be 30 percent higher or 23 percent lower. The revenue per unit is expected to be U.S.$2000. Two types of production facility are being considered. The first is a world class, efficient 900 thousand unit plant that can be constructed in time for the first year demand, lasts 3 years, and has variable cost of U.S.$1200 per unit an U.S.$800 profit margin. The second plant is smaller and less efficient. It produces
0 Quantity Price/unit Revenue Variable cost/unit Variable cost Depreciation EBIT Tax @ 40% Net income Capital expenditures Free cash flows Discount factor (9%) PV NPV (900) (900) 1.0000 (900) 111.85 1 300 2.00 600 1.20 (360) (300) (60) (24) (36) 264 0.9174 242 2 600 2.00 1,200 1.20 (720) (300) 180 72 108 408 0.8417 343 3 900 2.00 1,800 1.20 (1,080) (300) 420 168 252 552 0.7722 426

300 thousand units per year at a variable cost of U.S.$1450 each for a profit of U.S.$550. It too lasts 3 years and the company plans to build one of these smaller plants each year for 3 years. Either type of plant can be sold for its book value, the tax rate is 40%, the risk-free rate is 5%, and the weighted average cost of capital is 8% for the small plant and 9% for the big plant. Using these facts and assuming straight line depreciation, the company calculated the NPV of the large plant to be U.S.$112 million as shown in Figure 2. The NPV of the small plants is only U.S.$101 million (Figure 3). We cannot stop here, however, because there are two rather significant mistakes in our analysis. The first is that we have failed to account for a capacity cap that reduces expected output because in some states of nature we do not have enough capacity to supply all of the units that are demanded. The event tree shown in Figure 4 shows the demand in various states of nature. At the end of the first year, for example, if we had built the 900 thousand unit plant, there would be no
0 Quantity Price/unit Revenue Variable cost/unit Variable cost Depreciation EBIT Tax @ 40% Net income Capital expenditures Salvage Free cash flows Discount factor (8%) PV NPV (300) (300) 1.0000 (300) 101.24 1 300 2.00 600 1.45 (435) (100) 65 26 39 (300) (161) 0.9259 (149) 2 600 2.00 1,200 1.45 (870) (200) 130 52 78 (300) (22) 0.8573 (19) 3 900 2.00 1,800 1.45 (1,305) (300) 195 78 117 300 717 0.7938 569

Figure 2: Large plant NPV analysis Note: Quantities in thousands. Per unit costs in U.S.$ thousands. Aggregate cash flows in U.S.$ millions.

Figure 3: NPV analysis of three small plants Note: Quantities in thousands. Per unit costs in U.S.$ thousands. Aggregate cash flows in U.S.$ millions.

87

Real options and flexibility

problem if demand is high (390 thousand units), but if we build on the small plant, our capacity is only 300 thousand units. With the small plant we would stock out by 90 thousand units if demand is high. Calculations to the right of the binomial tree in Figure 4 indicate that the expected supply that can be sold is lower due to the capacity cap. Actual expected output becomes 261 units in year 1, 522 units in year two, and 725 units in year 3.
Market demand 1,014 600 355 1,977 1,170 692 410

The presence of a capacity cap is equivalent to selling a call option on the quantity produced, and the NPV methodology ignores it completely. We can attempt to adjust the expected output downward as shown in Figure 5, but the NPVs of both types of facility fall in unison so that their relative ranking does not change. Given the correction, the NPV of the large plant is U.S.$38 million and it is U.S.$21 million for the three small plants. Our second mistake was that we did not consider that we have the (compound) option of whether or not to invest in a second, and later on a third, or even a fourth small plant. If we take advantage of the modularity of the strategy that gains flexibility by investing in a sequence of smaller plants, we can build supply that evolves to respond to innovations in demand. Figure 6 shows that, by using a forward-backward algorithm we find the optimal solution to be that we build one small plant

390 231

Small plant output Year 1 Prob. .43 .57 Output 300 231 .18 .50 .32 Expected: 261 Year 2 Prob. Output 600 600 355 522 .08 .32 .41 .14 Year 3 Prob. Output 900 900 692 410 725

Flexible value tree Value: 1,056.0 Plants: 4 Demand: 1,977 Value: 1,046.1 Plants: 4 Demand: 1,170 Value: 378.0 Plants: 2 Demand: 1,170 Value: 378.0 Plants: 2 Demand: 692 Value: 478.0 Plants: 4 Demand: 1,170 Value: 478.0 Plants: 2 Demand: 692 Value: 139.0 Plants: 1 Demand: 692 Value: 139.0 Plants: 1 Demand: 410

Figure 4: Demand tree (thousands of units per year) Value: 679.2 Plants: 2 Demand: 1,014 Value: 467.3 Plants: 1 Demand: 390 Value: 638.0 Plants: 2 Demand: 600 ROA Value: 108.4 Plants: 0 Value: 294.2 Plants: 1 Demand: 600 Value: 386.1 Plants: 1 Demand: 231 Value: 274.4 Plants: 1 Demand: 355

0 Quantity Price/unit Revenue Variable cost/unit Variable cost Depreciation EBIT Tax @ 40% Net income Capital expenditures Salvage Free cash flows Discount factor (8%) PV NPV (300) (300) 1.0000 (300) 21.33

1 261 2.00 522 1.45 (378) (100) 44 17 26 (300) (174) 0.9259 (161)

2 522 2.00 1,044 1.45 (757) (200) 87 35 52 (300) (48) 0.8573 (41)

3 725 2.00 1,450 1.45 (1,051) (300) 99 39 59 300 659 0.7938 523

Figure 5: NPV analysis of three small plants with constrained capacity Note: Quantities in thousands. Per unit costs in U.S.$ thousands. Aggregate cash flows in U.S.$ millions

Figure 6: Value tree for flexible investment

88 - The

Journal of financial transformation

Real options and flexibility

to start with in the higher outcome states of nature and one plant in the lower state. From there we build another plant in the up state but no plants otherwise. In the third year we build 2 plants in the upper state for a total of 4, and no plants in the two middle states for a total of 2, and no plants in the lowest state, for a total of one. The modularity of the smaller plants provides us with the flexibility to increase capacity in order to meet high demand, or of saving capital by refusing to invest when demand does not grow. In the final analysis, the NPV of the small plant alternative is U.S.$108 million, compared with U.S.$38 million for the single large, efficient plant.

start up. If the price falls to U.S.$39.99 per barrel, the company loses U.S.$0.01 on every barrel it produces, but if it shuts down it loses the shutdown cost and the restart cost as well. Furthermore, there is a 50-50 chance that the next day the price will return to the U.S. U.S.$40 mark. Therefore, the well should operate at a loss until the expected loss equals the present value of the shut down and restart costs. Real options analysis provides the answer to how low the price can fall before it becomes optimal to shut down.

Multiple sources of uncertainty

In addition to the many types of real options (expansion, extension, shrinkage, abandonment, deferral, compound, and switching) there are often multiple sources of uncertainty to deal with. The uncertainties may also be complicated, for example they may be cyclical, mean-reverting, continuous, or discontinuous. The approach that I recommend is to start with each driver of uncertainty and estimate the way that it moves over time, for example, a commodity chemical may be cyclical (i.e. meanreverting). Then estimate the way that it covaries with other drivers of uncertainty over time, for example high prices are often associated with maximum production so that price and quantity are positively related. Then use these estimates and the spreadsheet for the NPV of the project to implement a Monte Carlo analysis that provides an estimate of the distribution of returns based on simulations of the changes in value. The resulting standard deviation of return is the basis for the up and down movements in the binomial lattice.

Financial flexibility and operating flexibility are substitutes

In the aforementioned Graham and Harvey survey, chief financial officers stated that flexibility was the single most important consideration when deciding how much to borrow. The previous example shows that flexibility on the assets side of the balance sheet is a substitute for flexibility on the liability side. If it turns out that the large plant is more desirable because its efficiency outweighs its lack of flexibility it will still require less debt and more equity financing as a hedge against downside risk. In contrast, the more flexible modular construction will allow more debt and less equity because in unfavorable states of nature the company need not have committed itself to excess capital.

Switching options
A switching option is the right to shut down and then restart an economic activity. For example, an automotive assembly plant, a mine, a computer assembly plant, or a peak load power generating plant can be shut down and restarted later on. The fixed cost of shutting down and restarting are exercise prices of the option. Not only does ROA provide the extra value due to the flexibility of switching, but it also gives the trigger points for taking action. To illustrate what we mean, suppose that a company operates an oil well that costs U.S.$40 per barrel to extract oil, and that the world price per barrel has been falling from U.S.$50 down to near U.S.$40. It costs U.S.$1 million to shut down and another U.S.$1 million to

Doing the mathematics

The very attempt to capture the subtlety of flexibility, if worthwhile, requires an investment in learning how to do the calculations well. This article is written for top management and for my friends in academia as an introduction to the importance of the idea. It is not a primer in mathematical methods. However, the interested reader should know that many of the more complicated problems have been solved using lattice approaches that are primarily algebraic and therefore, it is not necessary to use advanced stochastic calculus methods for solutions.

89

Real options and flexibility

At the risk of getting too detailed, one very simple example can be used to show the mathematics involved and most of the assumptions behind it. Suppose we are going to bid on an oil lease for a magic well that produces a barrel of oil immediately and then a barrel per year forever. The current profit per barrel is U.S.$200 but it has a 50% chance of rising to U.S.$300 per barrel or falling to U.S.$100 per barrel by the end of the first year. We make the simplifying assumptions that there are no costs of production, that the discount rate for the project is 10%, the risk-free rate is 5%, and that the price will remain at its new level forever. First, we estimate the net present value of the project. The current cash flow is U.S.$200, and the expected cash flow is .5($300) + .5($100 ) = $200 each year forever, therefore the present value of the project is: PV = $200 + ($200/0.1) = $2200 And if the investment to develop the lease is U.S.$1600, the net present value of the project is U.S.$600. This is illustrated in the left-hand side of Figure 7. Now suppose that we can defer development for one year. We can make an informed decision at that time that gives us the maximum of either the NPV of the project or zero, as illustrated in the right-hand side of Figure 7. Note that we would invest if the price goes up and would not invest if it goes down. Also, we cannot discount the payouts from the option at 10% because the option has different risk than the project.

To solve this problem we form a portfolio made up of two assets whose prices we already know, and do so in such a way that this portfolio gives exactly the same payouts as the deferral option. Called the replicating portfolio, it consists of m percent of the underlying risky project and B dollars of the risk-free bond. Since its payouts at the end of the period are identical in every state of nature with the deferral option, the current price of this replicating portfolio must be the same as the price of the option. Next, we calculate the value of the replicating portfolio by starting with its payout in the up state (equation 1) and in the down state (equation 2), then solve for the two unknowns, m and B: muV + (1 + rf)B = Max[uV I,0] = Cu = 1,700 mdV + (1 + rf)B = Max[dV I,0] = Cd = 1,700 m = (Cu Cd)/V(u d) = 1700/2200 = 0.773 (1) (2) (3)

B = (Cu muV)/(1 + rf) = [1700(0.773(3300))]/1.05 = -810 (4) Thus, if we put our money into a 77.3% share of the project and borrow U.S.$810 we will get exactly the same payouts as the option. In the up state we get: muV + (1+r)B= .773(3300) (1.05)810 = 1,700 and in the down state we get: mdV + (1+r)B = .773(1100) (1.05)810 = 0 The payouts on the replicating portfolio are the same as on the option, therefore they must have the same present value,
.5 V = 2200 I = 1600 NPV = 600 .5 300 uV = 300 + .1 ROA = 891 100 dV = 100 + .1

[uV-I,0]=Cu MAX [3300-1600,0]

MAX

namely ROA = mV + B = .773(2200) 810 = 891 (5)

[dV-I,0]=Cd MAX [1100-1600,0]=0

MAX

Underlying risky asset NPV = V I = $600

With a deferral option ROA = mV B = $891

Recall that the net present value of the project without any deferral option was U.S.$600. With the option to defer, the value has increased by U.S.$291, which is the value of the right to defer.

Figure 7: Payouts on an underlying asset and the asset with a deferral option

90 - The

Journal of financial transformation

Real options and flexibility

Conclusion
Real option analysis (ROA) is a much more intuitive concept than the traditional net present value (NPV) method for evaluating corporate capital investments and consequently it is replacing NPV. It takes time, of course, for people to learn about a new decision-making tool, more time to experiment with it, and still more time to change their habits. This article has been written to serve notice that there is a better way called real options.

References
Black, F., and M. Scholes, 1973, The pricing of options and corporate liabilities, Journal of Political Economy, May-June 1973, 637-654 Copeland, T., 2002, Xylenes basement, Harvard Business School, Case #9-202097, Autumn Copeland, T., and V. Antikarov, 2003, Real options: A practitioners guide, ThomsonTexere, New York Copeland, T. and P. Tufano, 2004, A real-world way to manage real options, Harvard Business Review, March Graham, J., and C. Harvey, 2001, The theory and practice of corporate finance: Evidence from the field, Journal of Financial Economics, 60, 187-243 Merton, R., 1973, The theory of rational option pricing, Bell Journal of Economics and Management Science, Spring, 125-144

91

92 - The

Journal of financial transformation

Real options

Venture investment contracts as baskets of real options1

Didier Cossin
UBS Professor of Finance, IMD

Benot Leleux
Stephan Schmidheiny Professor of Entrepreneurship and Finance, IMD

Entela Saliasi
FAME and HEC, University of Lausanne

Abstract
Valuing early-stage high-technology growth-oriented companies is a challenge to current valuation methodologies. This inability to come up with robust point estimates of value should not and does not lead to a breakdown of market liquidity. Instead, efforts are redirected towards the design of investment contracts which materially skew the distribution of payoffs in favor of the venture investors. In effect, limitations in valuation are addressed by designing the investment contracts as baskets of real options instead of linear payoff functions. In this article, we investigate four common features (covenants) of venture capital investment contracts from a real option perspective, using both analytical solutions and numerical analysis to draw inferences for a better understanding of the value of the contract features. The impact of this conceptual approach for pricing, valuation negotiation, and contract design are considered. It is shown, for example, how contingent pre-contracting for follow-up rounds is theoretically a better proposition than the simple rights of first refusal commonly found in many contracts. We also provide for results such as timing of investments, length of rounds, choices of liquidation levels, and conversion levels that take into account full interaction of the different features considered. We document some complex facts, such as the concavity of the VC contract value depending on the amount invested at the different stages, the actual share impact of the most common antidilution feature, some endogenous motivation for early VC exits from otherwise performing companies, and stress overall the importance of a full option analysis for efficient contract negotiation and understanding.

This paper is an abridged version of Understanding the economic value of legal covenants in investment contracts: A real-options approach to venture equity contracts by the same authors. We are grateful to Jerome Detemple, Michel Habib, Frederic Martel, Spiros Martzoukos, James Schallheim, Rene Stulz, Ernst-Ludwig Von Thaden, and Ton Vorst for their helpful comments and suggestions. We also would like to thank the seminar participants at Tilburg Institute of Advanced Studies Business Schools Law and Economics 2000 and University of Southern

California 2002 and conference participants at IAFE 2001, FMA 2001, and EFMA 2001. Financing was partly provided by the Swiss National Science Foundation (FNRS) and IMD. Financial support by the National Centre of Competence in Research Financial Valuation and Risk Management is gratefully acknowledged. The National Centre of Competence in Research are managed by the Swiss National Science Foundation on behalf of the Federal Authorities.

93

Venture investment contracts as baskets of real options

Valuing early-stage high-technology growth-oriented companies is a challenge to current valuation methodologies. This inability to come up with robust point estimates of value could potentially lead to a breakdown of market liquidity. However, this is not what is witnessed. In fact, billions of dollars of early stage venture capital has been poured into promising startups in Europe and the United States over the last few years. So, how do venture capitalists cope with the valuation uncertainties? The pioneering work of Sahlman (1990) pointed the way to the solution, or at least the coping mechanism used. He suggests that instead of expending useless amounts of time and effort coming up with a better estimate of an inherently uncertain future, efforts should be redirected towards the design of investment contracts which materially skew the distribution of payoffs in favor of the venture investors and active involvement in the development process of the invested company. In effect, limitations in valuation abilities are addressed by designing the investment contracts as baskets of real options instead of linear payoff functions and by directly intervening in the underlying processes. The key items outlined by Sahlman (1990) in the relationship between venture capitalists and entrepreneurial ventures include, the staging of the commitment of capital, the use of convertible securities instead of straight equity investments and the presence of liquidation preference for the VC, and anti-dilution provisions to secure the investors equity position in the company. While the recent empirical literature has explored further the presence of these features in VC contracts and analyzed theoretical arguments for their use [Kaplan and Stromberg (2000)], no paper to date has systematically valued them, both in isolation and as a whole. We exploit the real options literature to obtain a better understanding of the economic value of some classical features of VC contracts. Some authors have mentioned the optionality of VC contract features without developing systematically their analyses.

Obtaining a better understanding of the economic value of the legal features of investment contracts is a pre-condition to understanding their optimality as well. While our model includes many realistic features of current VC markets, the analysis remains very much a partial equilibrium approach. A richer model would incorporate agency issues, adverse selection and/or asymmetry of information that justify the use of these contract features in the first place. The model could be extended to combine the continuous time analysis we provide with a complex game theoretical approach. While this has been attempted in the credit risk literature [Anderson and Sundaresan (1996)], this is beyond the scope of this paper, since the combination of multiple complex optional features in itself presents a big enough challenge. The features specifically addressed here are the liquidation preference, the convertibility of the securities, the ex-ante staging of the investments, and the antidilution provisions. All these features have been shown to be common in venture capital contracts (with staging remaining in general informal rather than contractual: we analyze the strong difference this leads to) [see Sahlman (1990), Kaplan and Stromberg (2000/2002)]. In the original paper, we present closed-form solutions under restrictive assumptions, thus allowing for analytical experimentation. We also provide numerical analyses (using finite difference methods) of more general cases. From these analyses we obtain interesting insights into the value contribution of the various covenants. It is shown in particular how contingent pre-contracting for follow-up rounds [or ex-ante staging as described in Kaplan and Stromberg (2000)] is theoretically a better proposition than the simple rights of first refusal commonly found in many contracts. We also provide results on timing of investments, the impact of the length of rounds considered, the liquidation level to be chosen, the critical conversion level at which the VC will convert, and these with all features integrated and interacting. This abridged version of the paper is organized as follows. We

94 - The

Journal of financial transformation

Venture investment contracts as baskets of real options

first locate the paper at the intersection of different strands of literature on venture capital and real options. We then introduce the general framework and discuss the models underlying assumptions. The following sections discuss the key contractual features, such as liquidation preference and staging. We show that the liquidation preference itself is a package of liquidation rights and automatic conversion. It is important to note that the analytical and numerical solutions do take account of the complete interdependence of the features. Whereas liquidation preference sets up the framework for the VC end payoff and the conditions (event) under which it is exercised, staging aims to not only maximize the final payoff by optimally distributing the investment amount over discrete stages but also to minimize the downside risk, given the early exit option. We also tackle the most common antidilution feature, carefully analyzing the impact of the issuing price on the VC ownership stake. It highlights in particular the concave shape of the VC ownership stake.

The second class of research includes empirical studies that analyze the specificities of venture backed projects, such as Cochrane (2001), Gompers (1995), Kaplan and Stromberg (2000, 2002), Lerner (1994), Gompers and Lerner (1996), Seppa and Laamanen (2000), Manigart et al. (2002), etc. The closest papers to our concern study specifically the contractual features investigated here and their prevalence in VC contracts. The third class of research is closest in methodology to what we develop here, although its goal is different. The real options literature has contributed to a better understanding of the value of the investments, notably high-technology investments [Berk et al. (1998), Schwartz and Moon (2000), Schwartz and Gorostiza (2000)]. American options, in terms of valuation and optimal exercise, present some of the most challenging problems in finance. Consequently, many papers applying option models to real assets resort to the Black and Scholes approximation methodology [Benaroch and Kauffman (1999), Hull (1997)], or approximate the option to a European one [Trigeorgis and Panayi (1998)]. Others refer to Magrabes formula when modeling the investment opportunity as an exchange option with uncertain cash flows [Kumar (1996, 1999)]. Perlitz et al. (1999) applied Kemnas (1993) adjustment of the Geske approximation in an R&D project valuation. Jagle (1999) uses a binomial model as its numerical pricing technique.2 Schwartz and Cortazar (1998) use the Barraquand and Martineau (1995) methodology to price the underdeveloped oil field as a two dimension American Option. Schwartz and Moon (2000) used the Least Square Monte Carlo approach to price the Internet companies with uncertainty in expected return and growth rate. Lastly Schwartz and Gorostiza (2000) implement the successive over-relaxation and alternating direction implicit methodology in pricing the information technology acquisition and development with stochastic stream of cash flows.3 Beyond classical investment valuation, the real-options literature has also provided interesting strategic results in innovation management, with for example Grenadier and Weiss (1997) and more recently Bernardo and Chowdry (2002) sharing some technical similitude to ours.

Literature review
This paper lies at the crossroad of the venture capital and the real options literatures, breaking new ground by bringing them together in a way not attempted before for private equity investment contracts. The current literature on venture capital can be subdivided into three main classes, depending on whether it addresses the theoretical optimality of contracts, empirically analyses the existing contracts used in practice, or values the investments underlying the contracts (notably with real-options techniques). A rich, mostly game-theoretic literature addresses the agency and moral hazard problems arising in a multi-stage financial contracting environment with information asymmetries [Dessi (2001), Casamatta (2001), Aghion et al. (2000), Bergmann and Hege (1998), Bergmann and Hege (2000), Admati and Pfleiderer (1994), Noldeke and Schmidt (1998), Bascha and Walz (2001), and Bascha (2001)]. While this literature would have much to bring to this paper in terms of optimality (and vice versa), we have simplified the issue and taken actual contracts as given.

2 A more extended summary of these papers is provided in Schwartz and Gorostiza (2000), p. 2. 3 Recently Fourier transformation of the characteristic function has been used to calculate the respective probabilities, achieving a closed form solution, for high dimension state-variable European option.

95

Venture investment contracts as baskets of real options

We are not aware of prior literature addressing specifically the valuation of the contract features themselves, even though the idea has been mentioned before [Trigeorgis (1997)]. Our approach relies on an original application of the real options literature applied to investment contracts. Where most of the literature in that field tries to resolve the uncertainty of the project value through complex valuation techniques (and thus tries to reach a point estimate of project values), we attach a strong uncertainty to the project value itself (assumed to follow a stochastic process) and try to determine the impact of this uncertainty on the VC contract and its different features. While equity investment contracts have not been analyzed systematically in this way, a rich literature applies similar techniques to debt contracts, for example in the context of credit risk. Indeed, understanding the value of legal contracts using real options methodologies is now classical (and highly successful in practice) for credit risk. Since the seminal work of Merton (1974), which valued a simple zero-coupon straight debt contract, research has concentrated on more complex features, such as convertibility [Ingersoll (1977)], safety covenants [Black and Cox (1976)], and agency issues [Anderson and Sundaresan (1996)].4 The lack of consideration granted to private equity contracts is often justified by the lack of market completeness, which is not a satisfactory explanation since debt contracts face the same constraints (the assets of a firm are typically not traded). We show that such an analysis, when brought to equity contracts, provides for rich and meaningful results and should drive further research on contract optimality and design. The objective of this paper is to provide the first systematic analysis of the economic value of some key legal covenants used in venture capital contracts. It develops an intuitive framework to evaluate these main covenants both separately and together. The framework is flexible enough to accommodate any possible interaction between covenants, adding an entirely new dimension to the corresponding real-option literature. Mathematically speaking, we show that the pricing of a VC contract is similar to that of a complex package of financial

options. Each feature separately is either an option or a bundle of options. Bringing all the features together in one contract creates interactions between these options, interactions which affect their values.5 Because all the options considered tend to be of the American type (where early exercise is possible), we also address the issue of timing, a problem wellknown by practitioners and to which we add an analytical understanding.

General framework
This paper investigates specifically four common features (covenants) of venture capital investment contracts, liquidation preference, staging, convertibility, and anti-dilution. The preferred stock investment format provides the venture capitalist with preferences in liquidation, a feature reinforced by the accruing dividends. The latter are not meant to be paid unless the exit mechanism is not rich enough to provide adequate returns to the investors. In such circumstances, for example liquidation, accruing dividends (often 10-20% per annum) guarantee a disproportionate share of the assets to the venture capital investors. The net effect of the preferred equity format with accruing dividend is to skew the payoff distribution in case of liquidation in favor of the venture capitalist. Venture capitalists rarely invest in a single stage, instead, they stagger (stage) the investments over time in synch with distinct milestones in the development of the investee. Enough capital is provided at each stage to reach the next one, at which point the venture capitalist reserves the right to release the next round of capital or to abandon the project funding. Staging of the investment is said to be a mutually beneficial arrangement. It gives the venture capitalist the option to reinvest or abandon the project. It provides the investee with gradually cheaper funding, as the sources of uncertainty are progressively removed. To capture the upside potential of the investment, venture capitalists reserve the right to convert the preferred equity or bond into common stock at a predefined conversion rate. Most conversions tend to be automatic conversions. Black and

96 - The

Journal of financial transformation

4 See Cossin and Pirotte (2000) for an extensive review of that literature. 5 These options interactions have been analyzed in a different context by Trigeorgis (1997) and mentioned as a specific limitation of current optimal contract theory in Kaplan and Stromberg (2000).

Venture investment contracts as baskets of real options

Gilson (1998) and Kaplan and Stromberg (2002) argue that the effect of the conversion feature is to create a strong incentive for the entrepreneur to perform, as the venture capitalists give up their superior proposition explicitly modeled here in case of high performance by the firm. The conversion feature of the preferred equity or bond investment is often contingent on the pricing of future rounds of funding, providing a level of protection for the venture investors. The intensity of these anti-dilution provisions varies greatly from top-off provisions, which guarantee the maintenance of the equity percentage of the initial investor in future rounds, full ratchets, which reset the conversion price of all pre-existing series of funding to the lowest price of follow-up rounds, maintaining the value of the investors stake if not its actual equity percentage in the investee company, or weighted average ratchets, which provide for a readjustment of the conversion price of prior series to a weighted average of succeeding rounds of fund raising.

(through monitoring and active involvement in the business). Consequently, instead of using replication arguments that rely on complete markets (and lead to risk-neutral pricing), we make use of the more realistic dynamic programming argument developed in Dixit and Pindyck (1996) that uses the discount rate . Empirical results, such as Gompers (1995), and research studies, such as Berk et al. (1998), have shown that the systematic risk, as well as the volatility levels, are highest early in a projects life and decrease as the project approaches completion. The cost of capital should thus decrease through the life of the project, due to the higher leverage of the project early in its life. In order to study the impact of these factors in the contracting value process, we consider stepwise parameters such as volatility, drift rates, and cost of capital.

Firm value as a diffusion process

The dynamics of the firm value is modeled as a diffusion process, with a drift parameter affected by the VCs effort in the firm. The role of the VC value-add (his/her supportive nonfinancial contribution) affects the probability of success. The drift of the firm value process is defined as the sum of a drift of a similar project plus the VC value-add given its involvement in the project. The larger the VC contribution the higher the projects drift. The VC non financial contribution is a crucial component in valuing the venture project contract, since it captures the VC dual role as financier as well as project coach (non-financial contribution), which not only affect the drift of the firm value, but also entitles the venture capitalist to an additional reward (higher ownership stake on the final venture projects payoff). Although the VC value-add is beneficial for the firm value (i.e. increases the probability of success of the venture project), the VC does incur some additional costs by participating. This could lead to a premature VC exit from the venture backed project, if no additional reward, in terms of a higher ownership stake, is provided. The existence of the VC value-add has major implications, which make the private equity investment differ-

Discount rate
As described in Sahlman (1990): In theory, the required rate of return on an entrepreneurial investment reflects the riskfree interest rate in the economy, the systematic risk of the particular asset and the market risk premium, the liquidity of the asset, and compensation for the value added by the supplier of capital. This last adjustment is required to compensate the VC for monitoring and playing an active role in management. A crucial but realistic assumption concerns the stochastic changes in the firm value, which cannot be spanned or replicated from the existing assets in the economy. The underlying state variable in our model is a non traded asset (as often in the real options literature, and frequently in the credit risk literature as well). Therefore the riskless portfolio cannot be constructed. We are thus making use of the discount rate per stage which accounts for the decision makers subjective valuation of risk [Dixit and Pyndick (1996)]. Furthermore, the risk free rate cannot cover the VC cost for providing value-add

97

Venture investment contracts as baskets of real options

ent from an outside equity financing. Indeed, as these actions are costly for the VC, they deserve additional reward. The VC thus uses a higher discount rate, as presented by Sahlman (1990). Because of the efforts put in to increase the chances of success of the firm, the preference for liquidity effects, as well as the opportunity cost of focusing on a particular company rather than another one (the number of firms a venture capitalist can oversee is limited), we assume that the discount rate is larger than the drift of the process when the VC participates. We further propose that the VC ownership stake is a convex increasing function of the VC additional costs. The later the VC decides to invest, the lower his/her stake. In other words the VC ownership stake is a decreasing convex function of the investment time. The VC ownership stake increases at a decreasing rate for increasing investment levels. Lastly, the VC claim on project value is lower if the project value at that moment is higher.

money liquidation preference features have been known to be implemented in order to skew values even more towards the venture capitalist (or a specific member of a syndicate). The net effect of the preferred security format with accruing dividend is to skew the payoff distribution in case of liquidation. The most common security used by venture capitalists are convertible preferred stocks and debentures, as shown by Kaplan and Stromberg (2000) and Sahlman (1990). Schmidt (2001) summarizes the papers dealing with optimal contract design for an inside investor. The conversion feature can be understood as a reallocation of control rights in case of success of the project from the venture capitalist to the entrepreneur. It is an incentive to perform for the entrepreneur, above and beyond the direct financial incentives offered by the venture capitalist. The conversion feature can thus be understood as a barrier that transforms the considered features, such as liquidation preference, in barrier options of the up-and-out type. In other words, if a certain milestone is achieved, i.e. a certain level of value for the project, the previous options, in this case the liquidation preference, are canceled, thus reducing the differences in rights between the venture capitalist and the entrepreneur. The convertibility is thus as much a redistribution of rights towards the entrepreneur as it is an upside potential for the VC. From the VCs perspective, the object is to maximize the value of the shareholding in case the project is successful and recoup the maximum possible value in case the project is a failure. The value of the VC contract with both convertibility and liqui-

Liquidation preference and convertibility

Liquidation preferences appear in venture contracts in different formats. Most often, the liquidation preference is constructed as a participating feature in the convertible preferred securities design, whereby the first tranche of capital obtained in an exit is entirely committed to the investor group and any residual is then distributed pro-rata to the equity owners. The participating feature usually disappears once the exit valuations are sufficiently high to guarantee the outside investors a solid return on their initial investments. The participating preferred debentures or preferred stock investment format provides the venture capitalist with preference in liquidation, a feature reinforced by the accruing dividends. The accruing cash flows, dividends or coupons, are not meant to be paid unless the exit mechanism is not rich enough to provide adequate returns to the investors. In such circumstances, for example liquidation, accruing dividends, often 1020% per annum, guarantee a disproportionate share of the assets to the venture capital investors. Very high, out of the

dation preference integrated corresponds to the linear combination of two option payoffs, one corresponding to the liquidation right and the other to an automatic conversion feature. This expression seems similar to the classical analysis of a convertible as a bond + a call but differs. Indeed, the second option payoff differs from a pure upside potential. Because the VC gives up on the liquidation preference by converting, the liquidation preference level affects the conversion decision. The automatic conversion feature is less valuable than a sim-

98 - The

Journal of financial transformation

Venture investment contracts as baskets of real options

ple call, which has a pure upside potential. Liquidation levels affect conversion policies. There is thus a strong interaction between features. The analyses show indeed that liquidation preferences and convertibility interact strongly in venture capital contracts, in particular the liquidation level affects the level at which conversion occurs. A contract with a high liquidation value will be converted later, at a given conversion price, than a contract at low liquidation value. Typically, a larger shareholder, who will hold a higher liquidation value, will thus wait longer to liquidate. The two features considered illustrate the strong optionality integrated in VC contracts. We derived the option values in both infinite and finite horizons and show that the value of the automatic conversion is affected by all parameters value of the project, investment, volatility, drift, and time horizon while the liquidation right is mostly sensitive to changes in liquidation levels, and investment if liquidation is done at par, as is common.

similar to a compound call option. The next financing round can take place as soon as the project value reaches the optimal project value endogenously determined from the model. The venture capitalist has first rights to further investments in the company. We consider the possibility that these rights are formal (ex-ante staging), i.e. that the VC commits to further financing of a certain amount at a certain valuation if some milestone is achieved. We compare this to informal rights (linked notably to competitive issues, in which the VC has a first right to participate but where the pricing is set competitively at the time of the financing) and to the situation in which no rights (formal or informal) are acquired by the VC, so that the VCs investment corresponds to a short-term (full commitment) or a one shot (no staging) investment. The analyses show that the option to reinvest brings higher value to the VC contract. Formal staging dominates the informal staging, even though the latter is frequently seen in current VC contracts. It also dominates the absence of staging or a full commitment to future investments [Trigeorgis (1997)]. One shot investing is not the optimal decision. Determining the optimal investment level requires the full analysis to be drawn as the timing of investments through stages has a concave shape that varies strongly depending on the parameters considered. While the number of stages matter, the marginal impact seems to be decreasing with the number of stages. Overall, liquidation right values are less affected by timing and staging issues than the upside potential (and automatic conversion) is. They nonetheless can be affected in opposite directions by the timing of the investment.

Staged financing
Sequential investment is strongly related to the existence of great uncertainty concerning high technology, R&D, and generally VC-financed projects, a phenomenon specifically analyzed in Gompers (1995). Staged capital infusions allow venture capitalists to periodically gather information and monitor the progress of the project, and notably the evolution of the risk of the project, or project volatility, as it matures. Staging can be organized formally or informally. Kaplan and Stromberg (2000) for example analyze ex-ante staging in which VCs commit at initiation to future investments with contingencies on milestones being met. They also mention that most VC financings are at least implicitly staged, in the sense that even when all the funding in the initial round is released immediately, it is understood that future financing rounds will be needed to support the firm until the IPO. We consider also both the case of endogenous staging and the case of deterministic times for sequential investments. In the case of endogenous staging, the staging is considered to be

Antidilution
The antidilution feature appears in different forms in venture agreements. Its purpose is to protect the early rounds investors from dilution when additional stock splits, stock dividends, and any new financing at a lower price per share occur, driving the value of the initial investors down. The antidilution feature analyzed is similar to a long put option, which guarantees to the VC an additional value in case the share price drops below the old conversion rate. The additional value that the VC

99

Venture investment contracts as baskets of real options

receives consists of free shares to compensate for the dilution effect of the negative change in project value. Our analyses show the impact on the contract value of the antidilution feature and show how it transforms the underlying distribution. In particular, the classical weighted average antidilution clause leads to a situation where the VC share is not necessarily increasing with the issue price. The protected stake becomes a V-function of the issue price. This contractual feature can be balanced economically in a contractual negotiation versus other features. The analysis can be pushed further, for example to provide a comparison of the value impact of different antidilution formulae, such as the top off, which guarantees equity percentage, or the full ratchets that maintains value.

able to look at the basket of options as a basket, not a collection of individually priced options. A first attempt at this can be found in the original paper. This is but a first glimpse at the complex universe of venture capital investment contracts form a real options standpoint. The contribution from the real options approach should not be underestimated though, and it holds the promise of significant advances in the understanding of the true value-added of contractual features which have often been part of boilerplate contracts for decades with little understanding of their effect on the investors wealth nor behavior.

References
Admati A. and P. Pfleiderer, 1994, Robust financial contracting and the role of the venture capitalist, Journal of Finance, 49:2, 371-402 Aghion P., P. Bolton, and J. Tirole, 2000, Exit options in corporate finance: Liquidity versus incentives, Working Paper, IDEI and University of Toulouse Anderson R., and S. Sundaresan, 1996, Design and valuation of debt contracts, Review of Financial Studies 9, 37-68 Barraquand J., and D. Martineau, 1995, Numerical evaluation of high dimensional multivariate American securities, Journal of Financial and Quantitative Analysis, 30:3, 383-405 Bascha A., 2001, Venture capitalists reputation and the decision to invest in different types of equity securities, Working Paper, University of Tubingen Bascha A., and U. Walz, 2001, Convertible securities and optimal exit decisions in venture capital finance, Journal of Corporate Finance, 7, 285-306 Benaroch M., and R. J. Kauffman, 1999, A case for using real option pricing analysis to evaluate information technology project investments, Information System Research, 10:1, 70-86 Bergmann D., and U. Hege, 1998, Venture capital financing, moral hazard, and learning, Journal of Banking and Finance, 22, 703-735 Bergmann D., and U. Hege, 2000, The financing of innovation: Learning and stopping, Working Paper, Yale University and ESSEC Business School and CEPR Bernardo A. E., and B. Chowdhry, 2002 , Resources, real options, and corporate strategy, Journal of Financial Economics, 63:2, 211-234 Berk J. B., R. C. Green, and V. Naik, 1998, Valuation and return dynamics of new ventures, NBER Working Paper 6745 Black F., and J. Cox, 1976, Valuing corporate securities some effects of bond indenture provisions, Journal of Finance, 31:2, 351-367 Black B. S., and R. J. Gilson, 1998, Venture capital and the structure of capital markets: Banks versus stock markets, Journal of Financial Economics, 47:3, 243-277 Casamatta C., 2001, Financing and advising: Optimal financial contracts with venture capitalist, mimeo, University of Toulouse Cochrane J. H., 2001, Risk and return of venture capital, Working Paper, University of Chicago Cossin D., and H. Pirotte, 2000, Advanced credit risk analysis, John Wiley & Sons Dessi R., 2001, Start-up finance, monitoring and collusion, Working Paper, IDEI University of Toulouse Dixit A. K., and R. S. Pindyck, 1996, Investment under uncertainty, Princeton University Press Gompers P. A., and J. Lerner, 1996, The use of covenants: An empirical analysis of

Conclusion
We use advanced real option methodologies and frameworks to investigate the value of four classic features of venture capital contracts: staging, liquidation preference, convertibility, and antidilution. We show the effects of these covenants not only on the value of the contracts but also on the distribution of the final payoffs. We also show the complex interaction of the features and how they affect each other. The results of this work can shed some light on contract negotiations, but also provide valuable help in designing optimal contracts. The high non-linearities of the features analyzed should convince the reader that linear analysis of the IRR or NPV-type do not provide valid approaches to valuation. Beyond valuation, better design, as demonstrated in the staging feature of contracts, was shown to improve current contracts and contract negotiations. As outlined by Kaplan and Stromberg (2000), venture capital contracts are best seen as flexible contracting mechanisms for the contingent reallocation of control, ownership, and cash flow rights. But for this complex package of real options to attain its goal of optimally balancing incentives, risk protections, and the sharing of the upside potential, we need to be

100 - The

Journal of financial transformation

Venture investment contracts as baskets of real options

venture partnership agreement, Journal of Law and Economics, 39, 566-599 Gompers P. A., 1995, Optimal investment, monitoring, and the staging of venture capital, Journal of Finance, 50, 1461-1489 Grenadier S., and A. M. Weiss, 1997, Investment in technological innovations: An option pricing approach, Journal of Financial Economics, 44, 397-416 Ingersoll J. E., 1977, A contingent-claims valuation of convertible securities, Journal of Financial Economics, 4:3, 289-231 Jagle J. J., 1999, Shareholder value, real options, and innovations in technologyintensive companies, R&D Management, 29:3, 271-287 Kaplan, S. N., and P. Strmberg, 2000, Financial contracting theory meets the real world, an empirical analysis of venture capital contracts, NBER Working Paper No. W7660 Kaplan S. N., and P. Strmberg, 2002, Characteristics, contracts, and actions: Evidence from venture capitalist analyses, University of Chicago, Working Paper Kumar R. L., 1996, A note on project risk and option values of investments in information technologies, Journal of Management Information Systems, 13:1, 187-193 Kumar R. L., 1999, Understanding DSS value: An options perspective, The International Journal of Management Science, 27, 295-304 Lerner J., 1994, The syndication of venture capital investment, Financial Management, 23, 16-27 Manigart, S.; H. Sapienza, M. Wright, an K. Robbie, 2002, Determinants of required returns in venture capital context, Journal of Business Venturing, 17:4, 291 312 Margrabe W., 1978, The value of an option to exchange one asset for another, Journal of Finance, 33:1, 176-186 Merton R. C., 1974, On the pricing of corporate debt -the risk structure of interest rates, Journal of Finance, 29:2, 449-470 Noldeke G., and K. M. Schmidt, 1998, Sequential investment and options to own, RAND Journal of Economics, 29:4, 633-653 Sahlman W. A., 1990, The structure and governance of venture capital organizations, Journal of Financial Economics, 27, 473-521 Schmidt, K. M., 2001, Convertible securities and venture capital finance, Working Paper, University of Munich, CESifo and CEPR Schwartz E. S., and C. Z. Gorostiza, 2000, Valuation of information technology investment as real options, Working Paper UCLA Schwartz, E. S., and G. Cortazar, 1998, Monte-Carlo evaluation model of an underdeveloped oil field, Finance Working Paper CMS Schwartz, E. S., and M. Moon, 2000, Rational pricing of internet companies, Working Paper UCLA Seppa, T., and T. Laamanen, 2000, Valuation of venture capital investments: Empirical evidence, Working Paper, Helsinki University of Technology Trigeorgis L., 1996, Real options: Managerial flexibility and strategy in resource allocation, MIT Press.

101

Titel artikel

102 - The

Journal of financial transformation

Real options

A real options approach to bankruptcy costs: Evidence from failed commercial banks during the 1990s1
Joseph R. Mason
LeBow College of Business, Drexel University, Wharton Financial Institutions Center, and Federal Reserve Bank of Philadelphia

Abstract
Literature to date has identified three main aspects of liquidation costs, firm size, asset specificity, and industry concentration. This paper unifies the theory behind these three aspects of bankruptcy costs by treating them as components of a broader option valuation problem faced by the liquidating trustee. In the options valuation framework, at time t the trustee may choose to liquidate at current asset values and incur a known loss, or hold until the next period t+1 at a positive opportunity cost. The trustee may not sell in the current period if expected asset price growth is sufficiently large. Expectations of asset price growth are based on previous asset price growth and asset price volatility, which are related to firm size, asset specificity, and industry concentration. Testing the hypothesized asset price relationships on FDIC failed bank liquidation data with OLS, three-stage least squares, and duration specifications yields the appropriate results. Furthermore, it appears that liquidation time alone can be used as an effective second order proxy for asset value growth where market value estimates are unavailable.

Forthcoming, Journal of Business

103

A real options approach to bankruptcy costs: Evidence from failed commercial banks during the 1990s

Costly bankruptcies hurt creditors and may contribute to sluggish economic growth. Moreover, the substitutability of debt and equity in finance theory is predicated upon low bankruptcy costs. Hence, the financial literature has long sought a better understanding of bankruptcy. Though it is generally accepted that direct bankruptcy costs, such as legal and administrative fees, are determined primarily by the amount of time spent in liquidation, the determinants of time itself are not well understood. Literature to date has identified three main aspects of liquidation time (and costs), firm size, asset specificity, and industry concentration [Alderson and Betker (1995, 1996), Warner (1977), and Weiss (1990)]. This paper unifies the theory behind these three aspects of bankruptcy costs by treating them as components of a broader option valuation problem faced by the liquidating trustee. Intuitively, the paper models the trustees option valuation problem in the following manner. Assume the trustee bears a fiduciary duty to maximize creditor recoveries. Having taken possession of the firms assets at a loss to creditors, the trustees task becomes one of loss minimization. At any time t the trustee may choose to either liquidate at current asset values and incur a known loss, or hold until the next period t+1 at a positive opportunity cost. The trustee will not sell in the current period if expected asset price growth is sufficiently large. Expectations of asset price growth are partially based on asset price volatility, which is itself related to firm size, asset specificity, and industry concentration. Since the option to liquidate is not typically in the money, the trustee will rationally liquidate when marginal gains from waiting approach zero, that is, when the value of the option stabilizes. Hence, following Dixit and Pindyck (1994), the trustees divestment opportunity is equivalent to a perpetual put option. Therefore the decision to divest is equivalent to deciding when to exercise that option.2 The trustee therefore chooses the optimal time to exercise, such that the expected option value is maximized. Note, however, that in a real options context the value may not be analogous to a price,

but rather a target variable like optimal time or value growth. Researchers in the real options literature typically obtain the solution to a deterministic growth specification (one with no uncertainty) and use that to estimate feasible models. In the present context, the deterministic specification suggests that if asset values are expected to fall divestment will occur immediately (liquidation time will decrease) and if asset values are expected to rise divestment will be delayed (liquidation time will increase). Of course, those implications mask significant complexity. What determines expectations in an uncertain world? The stochastic growth representation parameterizes expectations as a function of asset value growth (above the discount rate) and volatility. The paper shows mathematically that the solution to the stochastic growth representation is denominated in target asset value growth across the (unknown) liquidation period rather than simple liquidation time. Furthermore, that optimal divestment value (V*) rises in response to greater volatility and declines in response to higher discount rate-price growth spreads. While the solution concept is intuitively appealing (in that investors target growth and holding period together rather than holding period alone) it is inherently unmeasurable in most applications, because market values for all but a small minority of financial contracts are typically unknown until the assets are actually sold. This paper contributes, therefore, not only a theoretical representation of the real options process, but also empirical tests of the stochastic growth model that have been hitherto unavailable. Those results are measurable because in some instances, investors have occasionally recorded market value estimates at the time of taking possession of the assets in order to guide their divestment decisions. The paper obtains and tests both deterministic and stochastic growth specifications and finds empirical results estimated using time and recovery value, for the most part, agree. The stochastic representation, however, yields logical extensions of the liquidation process that do not arise in the deterministic concept: ways that will require extensions of real options

104 - The

Journal of financial transformation

2 An American option can be though of as a variant of the perpetual option that is forced to exercise at a limit date. The perpetual option, however, has no such limit date so the exercise needs to be derived from a fundamental limit on the option value. It may be useful to bear in mind the results for a Black-Scholes model of the value of an option on an equity index that pays dividends through the following discussion.

A real options approach to bankruptcy costs: Evidence from failed commercial banks during the 1990s

theory in order to fully account for their effects. Specifically, the results that follow show evidence that divestment option values depend crucially upon the direction of the growth trend and may therefore contain a jump component depending on the expected direction of asset markets.

Figure 1 represents liquidation outcomes (recoveries) for banks that failed between 1980 and the present. The recoveries represented in the figure are measured at the end of 2000. While it is important to keep in mind that the figure includes results from liquidations that are not yet finished, almost all liquidations that began before 1996 are substantially complete. On the basis of Figure 1 it is immediately apparent that recoveries associated with failures during periods of banking industry difficulty, such as the early and late 1980s, were lower than those that occurred in other periods. Recoveries of banks that failed in 1981-1982 averaged 36.5%, those that failed in 1989-1990 averaged 44.1%, whereas recoveries for banks failing in other years over the period 1980-1995 averaged 69.8%, with a maximum aggregate recovery rate from 1995 bank failures of 86.6%. Such diminished recoveries during periods of economic or industry distress have been noted

Data
This paper measures liquidation outcomes using Federal Deposit Insurance Corporation (FDIC) Failed Bank Cost Analysis (FBCA) data on liquidations of 1,581 banks that failed between 1986 and 1996. The FDIC was the majority stakeholder in each case. Therefore in principle the FDIC attempted to extract at least enough value to cover the shortfall incurred from paying depositors (creditors) in full at the time of failure. For both individual and aggregate records, the report provides data on the (book value) amount of total assets and liabilities at time of failure, deposit insurance payouts, and the amount recovered up to the date of the report.

Annual percent of total recovery in each of years after failure Failure Number of Disbursements year new failures (U.S.$ thousands) 0 1 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 11 10 42 48 80 120 145 203 221 207 169 127 122 41 13 6 5 1 3 7 1581 152,355 888,999 2,275,150 3,807,082 7,696,215 2,920,687 4,790,969 5,037,871 12,163,006 11,445,829 10,816,602 21,412,647 14,084,663 1,797,297 1,224,797 609,045 169,397 25,546 285,763 1,234,278 U.S.$102,838,198 31.64 25.03 14.14 23.68 50.86 26.24 26.04 22.81 32.11 33.37 19.15 29.21 26.74 71.35 22.16 55.03 0.00 0.00 12.29 100.00 27.95 59.19 18.11 25.68 8.41 27.22 34.42 40.61 31.22 35.52 56.58 22.49 34.88 8.19 34.18 16.95 88.31 97.26 87.71

2 19.22 4.30 11.08 9.44 10.02 18.74 15.55 12.28 2.25 18.00 7.53 36.97 28.54 12.10 19.07 24.42 4.04 2.74

3 4.87 4.23 11.70 9.55 7.06 12.76 8.31 7.58 5.12 0.64 9.44 6.48 1.09 2.28 22.02 3.59 7.65

4 4.93 3.32 6.52 10.79 9.38 3.84 4.46 6.29 5.12 0.64 1.37 1.39 2.71 2.28 1.60 0.00

5 5.53 0.54 9.74 7.30 3.93 1.57 0.77 4.83 5.12 0.64 0.53 0.61 2.71 3.12 0.96

6 2.98 2.21 2.64 3.42 6.10 0.02 7.66 2.58 0.45 1.04 3.19 -1.74 2.29 0.68

7 1.77 0.69 28.18 1.30 1.48 3.41 1.94 0.94 0.40 6.43 1.23 3.18 1.05

8 0.53 0.02 0.24 2.14 2.55 4.73 0.75 0.87 0.90 2.72 0.72 1.41

9 0.60 0.47 -2.15 5.92 0.15 0.22 0.00 0.65 16.86 0.46 0.25

10 0.00 0.00 -0.20 0.78 0.05 1.26 0.10 0.56 0.43 0.55

Total recovery, 2000 (%) 75.32 36.75 36.36 59.47 71.55 58.26 63.24 59.79 43.07 45.84 74.19 70.68 73.64 63.73 84.28 86.58 75.41 77.00 15.46 0.90

Figure 1: Annual number, size, and progress of bank liquidations, 1980-2000 Sources: FDIC Annual Reports and authors calculations. Note: In most cases prior to 1996, total recovery reflects the percent of total claims that were recovered during the entire period of liquidation. Total recoveries after 1996 are more substantially incomplete. Negative recoveries may arise when banks return some assets previously purchased from the FDIC under put-back options. Dark shading indicates period over which liquidation proceeds at ten percent per year. Lighter shading indicates progress of five percent per year.

105

A real options approach to bankruptcy costs: Evidence from failed commercial banks during the 1990s

by other authors writing on bankruptcy costs and are sometimes thought to be associated with asset fire sales and/or rapid liquidation [Pulvino (1998) and Shleifer and Vishny (1992)]. Figure 1, however, suggests that the fire sale/rapid liquidation hypothesis may not adequately explain these liquidation outcomes. Figure 1 presents aggregate FBCA data on both recoveries and liquidation speed. The main body of the figure shows the percent of the total recovered amount that is collected in each year of the liquidation. Shaded areas in the figure indicate the length of time that liquidation progresses at over five (light) and ten (dark) percent per year. If liquidation slows below that rate and then re-accelerates, the intervening periods are also shaded. Although bank failures after 1997 are also included, their liquidations may not have progressed enough to be meaningful.

asset appreciation across the liquidation period, yielding a V* estimate that can be used to empirically test the stochastic real options specification. The FBCA database does not, however, contain details on bank asset and liability compositions. Furthermore, once a bank fails its charter is retired and each bank is assigned a unique liquidation case number. I therefore hand-matched each banks liquidation case number to its pre-failure charter number in order to link financial details of each at the last observed call prior to failure to the liquidation reports in the FBCA. The resulting cross-sectional data set relates ex-ante bank asset and liability compositions to liquidation experiences, measured by both liquidation speeds and asset value growth The crisis of the 1980s that caused the majority of bank fail-

In contrast to the fire sale/rapid liquidation hypothesis, the rates and shading in Figure 1 suggest that the periods of bank failures associated with years of industry distress (and low recovery rates) are associated with slower liquidation speed (shading extending further to the right) than those occurring in other years. This coexistence of reduced recoveries and slower liquidations may be the result of a rational application of the options valuation framework described above. Firstly, lower asset values during periods of distress add value to the timing option, so the trustee may rationally delay liquidation. Secondly, if periods of distress are accompanied by high asset price volatility and therefore high expected price growth in recovery, the option may be quite valuable and the trustee may rationally wait to liquidate.

ures in the sample is known to be the result of price volatility among several specific asset classes. The FDICs own accounts of the crises suggest that the favorable tax treatment for real estate development projects in the Economic Recovery Tax Act of 1981 provided substantial incentive for construction lending [Federal Deposit Insurance Corporation (1997, 1998)]. The Tax Reform Act of 1986, however, removed this incentive, reducing the potential profit margins of a substantial number of projects. Additionally, geographic areas that were particularly overbuilt (especially New England and the West) faced additional pressures when the national recession of 1990-91 reduced real estate demand, producing significant declines in rents, prices, and values of all types of real estate properties. Given the FDICs anecdotal evidence, I assume that banks real

For individual banks, the FBCA database provides the date of failure and the date of resolution (if complete). Beginning in 1991, the FBCA report also contains estimated market values of expected final recoveries. These market values were estimated each year by taking a sample portfolio of that years liquidations and reconciling those values with a statistical model of historical market value estimates maintained at the FDIC.3 I use changes in these estimated market values to measure

estate and C&I loan compositions were the primary source of their exposures to volatility and growth expectations that affect the real options problem we want to measure. I use the Freddie Mac Conventional Mortgage Home Price Index to derive real estate growth spreads and volatilities and commercial & industrial loan interest and fee income as a percent of total loans (derived from Call Report data) to derive C&I loan growth spreads and volatilities.

106 - The

Journal of financial transformation

3 I thank Richard Brown of the FDIC for providing details on the market value estimation procedure.

A real options approach to bankruptcy costs: Evidence from failed commercial banks during the 1990s

Analysis and results

Figure 2 presents results of an OLS asset value growth model for the 276 completed bank liquidations that began after 1990. The asset volatility and discount rate price growth spread variables alone (Column 1) explain 9.2 percent of the variation in asset value growth in the present sample. In Column 1, the C&I Loan Price Volatility, the Discount Rate-C&I Loan Price Growth Spread, and the Real Estate Price Volatility variables all obtain the correct signs and are statistically significant. The coefficient on the Discount Rate-Real Estate Price Growth Spread is statistically insignificant. The specification in Column 2 adds the control variables. The addition raises the adjusted r-squared of the specification (to 10.3 percent), and the signs on all of the real options valuation variables obtain the correct signs. However, only the coefficient on Real Estate Price Volatility remains statistically significant at conventional levels. Bank size is positively associated with asset value growth, V*, and banks located in the Southeast realize lower asset value growth than those in the Southwest (the omitted group). Figure 3 again relies on the restricted sample of 272 completed bank liquidations beginning after 1990. However, here I estimate a process where both asset appreciation and liquidation time are believed to be determined jointly by asset volatilities and discount rate price growth spreads, the control variables, and each other in a three-stage least squares (3SLS) specification. I use different liquidation strategies employed by the FDIC as exogenous variables. The three-stage least squares system in Figure 3 explains more than 48 percent of the combined variation in asset value growth and liquidation time and the two processes have a correlation coefficient of 0.26. The OLS results for asset value growth reported in Figure 2 are robust enough to jointly estimate liquidation time in via 3SLS in Figure 3. All asset volatility and discount rate price growth spread variables obtain the correct signs. Again, bank size and location in the Southeast are statistically significant control variables.
Southeast West Other real estate owned/total loans Total deposits/total liabilities Central Midwest Northeast Independent variable: asset appreciation Constant C&I loan price volatility Discount rate C&I loan price growth spread Real estate price volatility Discount rate real estate price growth spread Log of total assets Past due and non-accrual loans/total loans Model type: independent OLS n R2 Adjusted R2

(1) 276 0.105 0.092 Coefficient (std. err.) -0.290 (0.146) 125.703b (73.767) -0.011c (0.007) 79.723c (51.134) 8.12E-06 (0.004)

(2) 272 0.146 0.103 Coefficient (std. err.) -0.685a (0.292) 76.972 (80.658) -0.007 (0.007) 85.694c (52.802) -7.04E-05 (0.004) 0.015b (0.008) 0.040 (0.097) -0.087 (0.137) 0.180 (0.154) 0.031 (0.057) 0.055 (0.049) -0.005 (0.026) -0.053c (0.038) -0.027 (0.030)

(a) Statistical significance at 1% (b) Statistical significance at 5% (c) Statistical significance at 10% Figure 2: OLS estimates of asset appreciation, 1991-1996 failed banks with complete resolutions

The results of the liquidation time specification in Figure 3, Column 2, seem even stronger than those for asset value growth, which may be affected by errors in the FDIC market value estimation process. Both the Discount Rate Real Estate Price Growth and Discount Rate C&I Loan Price Growth Spread variable coefficients are negative and statistically significant. Coefficients on the C&I Loan and Real Estate Price Volatility variables are positive and statistically significant.

107

A real options approach to bankruptcy costs: Evidence from failed commercial banks during the 1990s

Model type: 3SLS n R2 (System-wide for 3SLS) Correlation between dependent variables: Independent variable:

272 0.487 0.259 (1) Asset appreciation Coefficient (std. err.) -0.869 (0.813) 71.615 (83.564) -0.006 (0.007) 82.317c (54.567) -4.00E-05 (0.004) 0.015b (0.008) 0.040 (0.097) -0.084 (0.137) 0.178 (0.154) 0.027 (0.110)

Loglinear survival model: Logistic Dependent variable (all models): Log likelihood Log likelihood (coef=0) LR test of Sig. (2k) Significance level Number of total obs. Number of obs. still active (1) 82.2 -20.2 204.8 0.00 272 0

Constant C&I loan price volatility Discount rate C&I loan price growth spread Real estate price volatility Discount rate real estate price growth spread Log of total assets Past due and non-accrual loans/total loans Other real estate owned/total loans Total deposits/total liabilities Length of liquidation (endogenous) Asset appreciation (endogenous) Central Midwest Northeast Southeast West

(2) (log) Liquidation time Coefficient (std. err.) 4.135a (0.333) 474.145a (93.727) -0.027a (0.008) 693.164a (61.216) -0.008c (0.005) 0.018b (0.009) 0.318a (0.110) 0.512a (0.156) 0.022 (0.175)

(log) Liquidation time (2) -437.0 -654.8 435.5 0.00 1200 64 Coefficient (std. err.) 5.736a (0.274) 0.013a (0.001) -0.012a (0.001) -335.088a (32.298) 0.042a (0.004)

(3) -335.9 -654.8 637.6 0.00 1200 64 Coefficient (std. err.) 2.341a (0.426) 297.175c (186.658) -0.012 (0.017) 693.440a (123.463) -0.008 (0.010) -297.174c (186.658) 0.011 (0.017) -770.667a (135.029) 0.013 (0.012) 0.118a (0.009) 0.034 (0.086) 0.134 (0.119) 0.430a (0.146) 2.987a (0.358) -0.065 (0.050) -0.363 (0.027) -0.066 (0.043) -0.051 (0.048) -0.064 (0.029)

Constant C&I loan price volatility Discount rate C&I loan price growth spread Real estate price volatility

Coefficient (std. err.) 4.266a (0.367) 502.935a (90.533) -0.029a (0.008) 719.644a

0.032 (0.058) 0.057 (0.050) -0.006 (0.026) -0.054c

(0.039) -0.027 (0.030) (a) Statistical significance at 1% (b) Statistical significance at 5% (c) Statistical significance at 10%

0.407 (0.456) 0.057 (0.065) 0.003 (0.056) 0.073a (0.030) -0.022 (0.044) 0.047c (0.034)

(55.149) Discount rate -0.007c real estate price growth spread (0.005) C&I loan price volatility * pre-trough indicator Discount rate C&I loan price growth spread * pre-trough indicator Real estate price volatility * pre-trough indicator Discount rate Real estate price growth spread * pre-trough indicator Log of total assets 0.005 (0.008) Past due and non-accrual loans/ 0.229a total loans (0.091) Other real estate owned/total loans 0.370a Total deposits/total liabilities Pre-trough indicator (0.144) 0.058 (0.225)

0.108a (0.009) 0.036 (0.089) 0.086 (0.122) 0.310b (0.177)

Figure 3: 3SLS estimates of resolution time and asset appreciation, 1991-1996 failed banks with complete resolutions

0.075 -0.087 (0.052) (0.055) Midwest 0.095 -0.356 (0.061) (0.029) Northeast 0.080 -0.105 (0.028) (0.044) Southeast 0.043 -0.078 (0.037) (0.051) West 0.047 -0.133 (0.027) (0.028) (a) Statistical significance at 1% (b) Statistical significance at 5% (c) Statistical significance at 10%

Central

Figure 4: Duration estimates of liquidation time

108 - The

Journal of financial transformation

A real options approach to bankruptcy costs: Evidence from failed commercial banks during the 1990s

The endogenous liquidation time and asset appreciation variables are insignificant in Columns 1 and 2, respectively, suggesting that the real options and control variables adequately capture the endogenous effects. Given the positive correlation coefficient between the two processes and the performance of the liquidation model, liquidation time may indeed serve as a useful second order proxy for asset value growth. The first column of Figure 4 contains duration estimates of liquidation time based on the previously analyzed sample of 272 completed liquidations that began after 1990. Although the sample size in Column 1 remains small all the volatility and discount rate price growth spread variables obtain the correct signs and are statistically significant. Column 2, reports results of a duration model using all 1,200 observations in the data set from 1986 to 1996.4 Here, the C&I Loan Price Volatility and Discount Rate C&I Loan Price Growth Spread variables again obtain the correct signs and are statistically significant. However, the Real Estate Price Volatility and Discount Rate Real Estate Price Growth Spread variables obtain the wrong sign and are statistically significant. Column 3, adds the Pre-Trough Indicator (a peakto-trough indicator variable), which allows the coefficients on the volatility and discount rate price growth spread variables to vary across the business cycle. The coefficients on the original Real Estate Price and C&I Loan Price Volatility and Discount Rate Real Estate Price Growth and Discount Rate C&I Loan Price Growth Spread variables in Column 3 again obtain their appropriate signs in the presence of the Pre-Trough Indicator and interaction variables. This result occurs because the original variables now reflect real option valuation only during the business cycle growth period. The signs on the Real Estate Price and C&I Loan Price Volatility and Discount Rate Real Estate Price Growth and Discount Rate C&I Loan Price Growth Spread interaction variables (reflecting the contractionary period) are exactly opposite those of their expansionary counterparts.

This result suggests that the options valuation problem may also depend upon the direction of the economy. Only the volatility coefficients are significant in this specification, and the signs on volatility are negative in falling markets (suggesting quick disposal of volatile assets) and positive in expanding markets (suggesting delayed disposal of volatile assets). Hence it appears that volatility is a double-edged sword, benefiting the trustee in expansionary periods but posing a risk of prolonged exposure to depressed asset values in contractionary periods.

Conclusion
Overall, I find evidence of a rational application of the trustees real options problem. Banks exposed to volatile assets or assets with low discount rate price growth spreads generally take longer to liquidate and yield a greater value appreciation (higher V*) than others. Moreover, OLS estimates of the stochastic model are correlated with the results of the threestage least squares and duration models, suggesting that liquidation time may provide a useful proxy for applying the options valuation approach in the real world, where market value asset appreciation may be difficult to measure. Work is already proceeding to test the robustness of the real options specification to corporate bond defaults and recoveries [Cangemi et al. (2005)] and in the historical context of nineteenth century and Depression-era bank liquidations [Mason and Redenius (2005a) and Calomiris and Mason (2005), respectively]. From an economic policy perspective, liquidation differences across the business cycle can be a source of a substantial procyclicality, where fast liquidations when the economy is contracting may help push prices downward and delay sales, heightening price volatility, economic uncertainty, and business cycle depth and persistence. Related work by Calomiris and Mason (2003a), Anari, Kolari, and Mason (2005), and Mason and Redenius (2005b) is already exploring these implications in detail, although much work remains to be done.

4 The vastly different sample sizes result because the previous models rely crucially upon estimated recovery data, which the FDIC only archived after 1990, to accurately measure asset appreciation. The earlier OLS was also estimated using 1991 estimated recovery data for banks failing before that year. The reported results were qualitatively robust to the data adjustment. FBCA data is not available before 1986.

109

A real options approach to bankruptcy costs: Evidence from failed commercial banks during the 1990s

References
Alderson, M. J., and B. L. Betker, 1995, Liquidation costs and capital structure, Journal of Financial Economics, 39, 45-69 Alderson, M. J., and B. L. Betker, 1996, Liquidation costs and accounting data, Financial Management, 25, 25-36 Anari, A., J. Kolari, and J. R. Mason, 2005, Bank asset liquidation and the propagation of the U.S. Great Depression, Journal of Money, Credit, and Banking (forthcoming). This is a revised version of Wharton Financial Institutions Center (Philadelphia, PA) Working Paper No. 02-35, August 2002 Calomiris, C. W. and J. R. Mason, 2005, Rational divestiture in real options-based liquidation cycles: Evidence from failed bank assets in the Great Depression, Drexel University Working Paper Calomiris, C. W., and J. R. Mason, 2003a, The consequences of bank distress during the Great Depression, American Economic Review, 93, 937-47 Calomiris, C. W., and J. R. Mason, 2003b, Fundamentals, panics and bank distress during the Depression, American Economic Review, 93, 1615-1647 Cangemi, R., J. R. Mason, and M. Pagano, 2005, A real options approach to bankruptcy costs: Evidence from corporate bond defaults and recoveries, Mimeo Dixit, A. K., and R. S. Pindyck, 1994, Investment Under Uncertainty, Princeton, NJ: Princeton University Press Federal Deposit Insurance Corporation, 1997, History of the Eighties Lessons for the future, Washington, D.C. Federal Deposit Insurance Corporation, 1998, Managing the crisis: The FDIC and RTC experience 1980-1994, Washington, D.C. Federal Deposit Insurance Corporation, Annual report, Various years, Washington, D.C. Federal Deposit Insurance Corporation, Failed bank cost analysis, Various years, Washington, D.C. Federal Deposit Insurance Corporation, Reports of condition and income, Various years, Washington, D.C. Mason, J. R. and S. Redenius, 2005a, A real options approach to bankruptcy costs: Evidence from National Bank-era liquidations, Mimeo Mason, J. R. and S. Redenius, 2005b, Bank asset liquidation and the propagation of nineteenth century business cycles, Mimeo Pulvino, T. C., 1998, Do asset fire sales exist: An empirical investigation of commercial aircraft transactions, Journal of Finance, 53, 939-978 Warner, J. B., 1977, Bankruptcy costs: Some evidence, Journal of Finance, 32, 337347 Weiss, L. A., 1990, Bankruptcy resolution: Direct costs and violation of priority of claims, Journal of Financial Economics, 27, 285-314

110 - The

Journal of financial transformation

Real options

Kuno J. M. Huisman
Consultant, Centre for Quantitative Methods CQM B.V, and Researcher, Department of Econometrics & Operations Research and CentER, Tilburg University

Peter M. Kort
Professor, Department of Econometrics & Operations Research and CentER, Tilburg University, and Professor, Department of Economics, University of Antwerp

Grzegorz Pawlina
Lecturer, Department of Accounting and Finance, Management School, Lancaster University

Strategic investment under uncertainty: A survey of game theoretic real options models

Jacco J.J. Thijssen

Lecturer, Department of Economics, Trinity College Dublin

Abstract
The non-exclusivity of real options for individual firms and the growing importance of strategic interactions have given rise to a research field that is situated on the intersection of real options theory and game theory. This paper provides an overview of the state of the art of game theoretic real options models.

111

Strategic investment under uncertainty: A survey of game theoretic real options models

The main difference between financial options and real options is that in most cases real options are not exclusive. Exercising a given option by one party results in the termination of corresponding options held by other parties. For example, an option to open an outlet in an attractive location is alive only until a competitive firm opens its own store there. Currently, however, real options theory mainly considers single decision-maker problems of firms operating in monopolistic or perfectly competitive markets. But capital budgeting decisions can be strongly influenced by existing as well as potential competitors. The growth of Asian economies like China and India, as well as the extension of the European Union are prime examples of developments leading to increased interdependencies among firms around the globe. As a result, former domestic market leaders now have to deal with competition. The conclusion is that there is a strong need to consider situations where several firms share the option to invest in the same project. This new topic requires a merger between game theory and real options. Until a few years ago only a small number of contributions dealt with the effects of strategic interactions on the option value of waiting associated with investments under uncertainty. One of the main reasons is that the application of game theory to continuous-time models is not well developed and often quite tricky. However, due to the importance of studying the topic of investment under uncertainty in an oligopolistic setting, more publications have appeared recently [Grenadier (2000), Boyer et al. (2004), and Huisman et al. (2004)]. This paper provides an overview of the state of the art, where we mainly concentrate on identical firms in a duopoly context. We begin by discussing a standard model. The model is a new market model and based on Dixit and Pindyck (1994). Since the firms are identical it seems natural to consider symmetric strategies. However, it can be expected that coordination problems arise in situations where investment is optimal

only if the other firm refrains from undertaking the project. While discussing the standard model we show that imposing mixed strategies can deal with this coordination problem in an economically meaningful way. This approach, being inspired by the deterministic analysis in Fudenberg and Tirole (1985), was developed in Huisman (2001) (see also Huisman and Kort (2003)) and formalized in Thijssen et al. (2002). A similar attempt can be found in Boyer et al. (2001). We show that joint investment can occur even if it is optimal for only one firm to invest. Furthermore, we discuss why it may be impossible to rule out such a joint investment even when preplay communication is allowed. In other words, we argue that the outcome with both firms coordinating and investing sequentially with probability one, as in Smets (1991) and Dixit and Pindyck (1994), may be unlikely to achieve. One of the main results of the strategic real options literature is the rent equalization principle. According to this principle, the payoffs of the first mover (leader) and of the second mover (follower) are equal. This results from the fact that the leader has to invest no later than when the stochastic demand reaches the preemption point, i.e. the level at which the leader and the follower value functions intersect. Waiting longer would ultimately result in a preemptive investment by the competitor, attracted by the opportunity of realizing the leaders payoff. A direct economic interpretation of the rent equalization principle is that competition erodes the value of the option to wait. Subsequently, we show that if the initial level of demand is higher than the demand level at the preemption point, but lower than the demand level at which the second firm invests, the only symmetric Nash equilibrium is the one in which the firms play mixed strategies. As a consequence, the firms may end up investing simultaneously when it is not optimal to do so, which could even lead to performing projects with negative net present values (NPVs). This is a result of the coordination problem associated with the selection of the leader and the follower roles.

112 - The

Journal of financial transformation

Strategic investment under uncertainty: A survey of game theoretic real options models

Standard model
The first paper dealing with a multiple decision-maker model in a real option context is Smets (1991). It considers an international duopoly where both firms can increase their revenue stream by investing. Like in Fudenberg and Tirole (1985) two equilibria arise: a preemption equilibrium, where one of the firms invests early, and a simultaneous one, where both firms delay their investment considerably. A simplified version was discussed in Dixit and Pindyck (1994) in the sense that the firms are not active before the investment is undertaken. The resulting new market model only admits the preemption equilibrium. In this section our symmetric mixed strategy approach is applied to the new market model [Dixit and Pindyck (1994), for a more thorough analysis see Thijssen et al. (2002)]. This model considers an investment project with sunk costs I > 0. After the investment is made the firm can produce one unit of output at any point in time. Since the number of firms is two, market supply is Q {0, 1, 2}. It is assumed that the firms are risk neutral, value maximizing, discount with constant rate r, and variable costs of production are absent. The market demand curve is subject to shocks that follow a geometric Brownian motion. In particular, it is assumed that the unit output price is given by P(t) = Y(t)D(Q), in which dY(t) = Y(t) dt + Y(t) d(t), Y(0) = y, (2) (3) (1)

Given the stochastic process (Y(t))t0, we can define the payoff functions for the firms. If there is a firm that invests first while the other firm does not, this firm is called the leader. When it invests at time t its discounted profit stream is given by L(Y(t)). The other firm is called the follower. When the leader invests at time t the optimal investment strategy of the follower leads to a discounted profit stream F(Y(t)). If both firms invest simultaneously at time t, the discounted profit stream for both firms is given by M(Y(t)). In the remainder of the paper, the time dependency of Y will be omitted. In most cases, finding the optimal investment rule of a firm entails finding the value-maximizing threshold level of Y at which the firm should exercise its real option. In a strategic case, it often happens (as in the game considered in this section) that no pure strategy symmetric equilibria exist. In such a case the equilibrium strategy entails exercising the option at a given threshold with a probability strictly smaller than 1. Let us first consider the optimal investment threshold of the follower, which we denote by YF. If the leader invests at Y < YF, the followers value is maximized when the follower invests at YF. The followers profit flow will be YD(2). Following familiar steps [cf. Dixit and Pindyck (1994)], we can find YF. It satisfies YF = [/(-1)] x [(r )I/D(2) where is given by = 1/2 2 + [(1/2 /2)2 + 2r/2]1/2 > 1 By rewriting (4) as YFD(2)/r = I, (6) (5) (4)

where y > 0, 0 < < r, > 0, and the d (t)s are independently and identically distributed according to a normal distribution with mean zero and variance dt. Furthermore, D(Q) is a decreasing function, comprising the strategic part of the inverse demand curve. Equations (1) and (2) imply that the output price P(t) fluctuates randomly with drift and standard deviation and that it always takes positive values. where /(1), we can observe that the optimal investment rule is a modified NPV formula with a mark-up , which is larger than 1. The mark-up reflects the impact of irreversibility and uncertainty (both not taken into account in the traditional NPV rule) and is increasing in uncertainty (it holds that / < 0).

113

Strategic investment under uncertainty: A survey of game theoretic real options models

Since firms are identical, there seems to be no reason why one of them should be given the leader role beforehand. The fact that firms are rational and identical also implies that it is hard to establish coordination on a non-symmetric equilibrium. Therefore, we concentrate on equilibria that are supported by symmetric strategies. We use the subgame perfect equilibrium concept for timing games as formalized in Thijssen et al. (2002). This approach extends the perfect equilibrium concept of Fudenberg and Tirole (1985) to stochastic games. We present a less formal discussion of the firms strategies. To describe the equilibrium, first define the preemption point YP = minY {Y\L(Y) = F(Y)}, (7)

Since we restrict ourselves to symmetric strategies, the only possibility left is to apply mixed strategies. Denote the probability that Firm i invests at Y by i. Consequently, i can be interpreted as the probability that Firm i chooses row 1 in the matrix game depicted in Figure 1.

Invest Invest Not invest ( M(Y), M(Y) ) ( M(Y), M(Y) )

Not invest ( M(Y), M(Y) ) repeat game

Figure 1: Payoffs and actions in the bimatrix game. Firm 1 is the row player and firm 2 the column player

The game is played at Y if no firm has invested so far. For Y > YP this can happen either by mistake or that Y is the initial value. In all the other cases, at least one of the firms would have invested before the process presented in (2) has reached Y. Playing the game costs no time and if Firm i chooses row 2 and Firm j column 2 the game is repeated. If necessary the game will be repeated infinitely many times. Since i and i are the probabilities that Firm i and Firm j invest at a given level of Y, they are the control variables that need to be optimally determined. To do so, define Vi as the value of Firm i, which is given by: Vi = Max [i (1 j)]L(Y) + (1 i)jF(Y) + ijM(Y) +
i

This point is called preemption point because to the right of this point the leader value, L(Y), exceeds the follower value, F(Y), and this results in strategic behavior of the firms trying to preempt each other by investing, as will become apparent from the description below. The equilibrium under consideration is therefore called a preemption equilibrium. There are three potential scenarios that we can consider. The first can be defined by Y YF. This outcome exhibits immediate joint investment. Here the unit output price is large enough for both firms to enter the market. The second scenario is where YP Y < YF. Immediate joint investment gives a payoff M(Y). This is not a Nash equilibrium since if one of the firms deviates by waiting with investment until the process Y hits the trigger YF, it obtains the follower value F(Y). This follower value exceeds M(Y) as long as YP Y < YF. M(Y) can be negative for Y values belonging to the interval (YP, YF), which is equivalent to a negative NPV. In case both firms refrain from investment and wait until Y hits YF, they get the follower payoff F(Y). Again this is not a Nash equilibrium, because if one of the firms deviates by investing, this firm receives a payoff L(Y) that exceeds F(Y) on this interval.

(1 i)( 1 j)Vi

(8)

Since Firm i invests with probability i and Firm j with probability j, the probability that Firm i obtains the leader role, and thus receives L(Y), is i [1j]. Similarly, with probability [1 i] j Firm i is the follower, ij is the joint investment probability, and with probability [1i] [1j] nothing happens and the game is repeated. After determining down the first order conditions for Firm i and Firm j, and imposing symmetric strategies, i.e. i = j = , it is obtained that = [L(Y) F(Y)]/[L(Y) M(Y)] (9)

114 - The

Journal of financial transformation

Strategic investment under uncertainty: A survey of game theoretic real options models

We know that M(Y) < F(Y) L(Y) on the relevant Y-interval [YP, YF), so that we are sure that the probability lies between zero and one. From (9) it is obtained that, given the difference L(Y) M(Y), the firm is more eager to invest when the difference between the payoffs associated with investing first and second is large. After substitution of = i = j into (8), the value of Firm i can be expressed as Vi = [(1 )L(Y) + (1 )F(Y) + 2M(Y)]/(2 2) (10)

which leaves less room for the equal probabilities of being the first or second investor. In the third scenario it holds that Y < YP, in which the follower value exceeds the leader value. Hence, investing first is not optimal so that both firms refrain from investing and wait until Y = YP. Then the second scenario is entered, and upon observing that L(YP) = F(YP), it can be obtained from (9) that = 0. From (12) we get that the probability for a firm to become leader is one half, and with the same probability this firm will be the second investor. Furthermore, from (11) it can be concluded that the probability of simultaneous investment at YP is zero. All this implies that one of the firms will invest at YP and the other one, being the follower, will wait until Y equals YF. Since the values of leader and follower are equal at YP, the firms have equal preferences of becoming the first or the second investor in this case, so that we have rent equalization.

Of course, both firms do not want to invest at the same time, because it leaves them with the lowest possible payoff M(Y). From (10) it can be obtained that the probability of occurrence of such a mistake is /(2 ) (11)

The first mover advantage results in equilibrium strategies in which increases with . We also see that, whenever is greater than zero, which is the case for Y (YP , YF), the probability that the firms invest simultaneously is strictly positive. This is not in accordance with many contributions in the literature. For instance, Smets (1991) and Dixit and Pindyck (1994) state that if both players move simultaneously, each of them becomes leader with probability one half and follower with probability one half. Our analysis is based on the assumption that the firms do not Similarly, it can be obtained that the probability of a firm being the first investor equals (1 )/(2 ) (12) communicate in an attempt to coordinate their actions. This results in a positive probability of making a mistake, i.e. investing jointly while the level of demand is not sufficiently high. Such an outcome is ruled out by some authors, e.g. Smets (1991) and Dixit and Pindyck (1994), who assume that coordiDue to symmetry this is also the probability of the firm ending up as the follower. Since the probability of simultaneous investment increases with , it follows that the probability of being the first investor decreases with , which is at first sight a strange result. But it is not that unexpected, because if one firm increases its probability to invest, the other firm does the same. This results in a higher probability of investing jointly, We argue that such a coordination as in Smets (1991) and Dixit and Pindyck (1994) seem unfeasible without introducing a nation is possible via tossing a coin. Consequently, the game analyzed in these papers require introducing a third player, nature, who assigns the roles to the firms in the situation, where both of them want to invest immediately. which both firms take a positive probability of making a mistake in order to get the leader payoff. Substitution of equation (9) in (10) eventually shows that a firm sets its intensity such that its expected value equals the follower value. Due to the risk-neutrality the firm is indifferent between obtaining the follower payoff for sure ( equal to zero) and obtaining the follower payoff as expected value ( as defined in (9)).

115

Strategic investment under uncertainty: A survey of game theoretic real options models

third player (nature), even when firms are allowed to communicate. Any collusive agreement among firms in region (YP, YF) would be hard to sustain because of the following arguments. Firstly, none of the firms would accept the follower role, which is associated with a lower payoff than that of the opponent. Consequently, the only remaining possibility is an agreement on the firms roles with a monetary transfer from the leader to the follower. However, even if we ignore the fact that such an act is illegal, the leader cannot credibly commit to meet his obligations once his investment is made. The follower, who anticipates the leaders default on its promised payment, enters the preemption game, which results in the mixed strategy equilibrium described above. The outcome of Smets (1991) and Dixit and Pindyck (1994) is unlikely to occur even if successful coordination is allowed for (e.g. if some mechanism exists that enables credible commitment of the leader). Allowing for the possibility of pre-play agreement on the roles of the leader and of the follower will neutralize the incentive to preempt (since preemption is not associated with the maximization of the firms joint value). So, any binding agreement will not result in an equilibrium la Smets (1991) and Dixit and Pindyck (1994). Instead, the leader will invest at some Y, say YL, which is greater than YP but smaller than YF, such that YL maximizes the leader value.

nates rent equalization present in the basic strategic real option model. Among other things, a surprising result is found that the value of the high cost firm can increase in its own investment cost. In Huisman and Kort (2004) firms take into account the occurrence of future technologies when deciding about investment. A scenario is identified where the possibility of the arrival of a new technology results in a game with a second mover advantage. In such a case, it is optimal for a firm to be the follower and to wait for the new technology rather than to be the first mover locked into the inferior older technology. Finally, Thijssen (2004, Part I) extends the existing real option literature by studying a framework where over time information arrives in the form of signals. This information reduces uncertainty. In analyzing a new market model it is found that the mode of the game depends on the first mover advantage relative to the value of information free riding of the second mover, who observes the true state of the market after the leaders entry. Consequently, a firm has to trade off the benefit of entering the market earlier with the risk of incurring the investment cost in the bad state of the market. Besides our own extensions, the framework being presented in the previous section is used for many other applications. Grenadier (1996) applies it to the real estate market, Weeds (2002) and Miltersen and Schwartz (2003) study R&D investments, Pennings (2004) and Pawlina and Kort (2002) analyze the product quality choice, Mason and Weeds (2003) and Lambrecht (2004) study merger policy and entry, Boyer et al. (2001) look at incremental indivisible capacity investments, Lambrecht (2001) takes into account debt financing, Nielsen (2002) and Mason and Weeds (2001) analyze the effects of positive externalities, Grenadier (1999), Lambrecht and Perraudin (2003), and Dcamps and Mariotti (2004) consider incomplete information, Pawlina and Kort (2003) explicitly model demand uncertainty, Cottrell and Sick (2001, 2002) study the interaction of first mover advantages and real options, Sparla (2004) and Murto (2004) consider the decision to close down, while Williams (1993), Baldursson (1998), Grenadier (2002), Aguerrevere (2003), and Murto et al. (2004) study oligopolies with more than two firms.

Extensions
As an illustration of the applicability of the strategic real option framework, we proceed by reviewing some of our own work. Huisman (2001) shows that if firms already compete in the product market, they may avoid entering the preemption game and invest jointly when demand is sufficiently high. This results from the fact that foregoing a part of the future cash flow due to postponing the investment beyond the leaders optimal threshold can be more than compensated by a reduction in the present value of the investment cost (which will be incurred later). In Pawlina and Kort (2001) it is shown that introducing asymmetry in the investment cost function elimi-

116 - The

Journal of financial transformation

Strategic investment under uncertainty: A survey of game theoretic real options models

Application of our method to the standard model showed that mixed strategy equilibria can be handled in a very tractable fashion. Nevertheless, in the literature the prevailing method is to rule out simultaneous exercise beforehand [besides our own work, an exception is Boyer et al. (2001)]. This is either done by (i) assumption or by (ii) avoiding cases where suboptimal simultaneous investment can occur. Examples of (i) are, for instance, Grenadier (1996) who assumes that if each tries to build first, one will randomly (i.e. through the toss of a coin) win the race, or Dutta et al. (1995) where it is assumed that If both i and j attempt to enter at any period t, then only one of them succeeds in doing so [for a similar argument, see Nielsen (2002)]. Examples of (ii) are Weeds (2002) who in a new market model assumes that the initial value lies below the preemption point, so that sequential investment is the only equilibrium outcome, or Pennings (2004), Mason and Weeds (2003) and Pawlina and Kort (2002), where the leader and follower roles are exogenously assigned. Overall, with this contribution we attempted to show that the strategic real option framework is a suitable tool to extend the industrial organization literature in a dynamic stochastic direction. By reviewing some existing research in this field, this paper proves that the interplay of game theory and real option valuation is a fascinating area that can generate economic results being significantly different from what is known from the existing industrial organization literature.

References
Aguerrevere, F. L., 2003, Equilibrium investment strategies and output price behavior: A real-options approach, The Review of Financial Studies, 16, 12391272 Baldursson, F. M., 1998, Irreversible investment under uncertainty in oligopoly, Journal of Economic Dynamics & Control, 22, 627644 Boyer, M., . Gravel, and P. Lasserre, 2004, Real options and strategic competition: A survey, Working Paper, CIRANO, Montral, Qubec, Canada Boyer, M., P. Lasserre, T. Mariotti, and M. Moreaux, 2001, Real options, preemption, and the dynamics of industry investments, Working Paper, Department des sciences economiques, Universit du Qubec Montral, Montral, Qubec, Canada. Cottrell, T. and G. Sick, 2001, First-mover (dis)advantage and real options, Journal of Applied Corporate Finance, 14, 4151 Cottrell, T. and G. Sick, 2002, Real options and follower strategies: The loss of real option value to first-mover advantage, The Engineering Economist, 47, 232263 Dcamps, J-P. and T. Mariotti, 2004, Investment timing and learning externalities, Journal of Economic Theory, 118, 80102 Dixit, A. K. and R. S. Pindyck, 1994, Investment under uncertainty, Princeton University Press, Princeton, New Jersey, United States of America Dutta, P. K., S. Lach, and A. Rustichini, 1995, Better late than early: Vertical differentiation in the adoption of a new technology, Journal of Economics & Management Strategy, 4, 563589 Fudenberg, D. and J. Tirole, 1985, Preemption and rent equalization in the adoption of new technology, The Review of Economic Studies, 52, 383401 Grenadier, S. R., 1996, The strategic exercise of options: Development cascades and overbuilding in real estate markets, The Journal of Finance, 51, 16531679 Grenadier, S. R., 1999, Information revelation through option exercise, The Review of Financial Studies, 12, 95129 Grenadier, S. R., 2000, Game Choices: The Intersection of Real Options and Game Theory, Risk Books, London, United Kingdom Grenadier, S. R., 2002, Option exercise games: An application to the equilibrium investment strategies of firms, The Review of Financial Studies, 15, 691721 Huisman, K. J. M., 2001, Technology Investment: A Game Theoretic Real Options Approach, Kluwer Academic Publishers, Dordrecht, The Netherlands Huisman, K. J. M. and P. M. Kort, 2003, Strategic investment in technological innovations, European Journal of Operational Research, 144, 209223 Huisman, K. J. M. and P. M. Kort, 2004, Strategic technology adoption taking into account future technological improvements: A real options approach, European Journal of Operational Research, 159, 705728 Huisman, K. J. M., P. M. Kort, G. Pawlina, and J. J. J. Thijssen, 2004, Strategic investment under uncertainty: Merging real options with game theory, Zeitschrift fr Betriebswirtschaft, 67, 97123 Lambrecht, B. M., 2001, The impact of debt financing on entry and exit in a duopoly, The Review of Financial Studies, 14, 765804 Lambrecht, B. M., 2004, The timing and terms of mergers motivated by economies of scale, Journal of Financial Economics, 72, 4162 Lambrecht, B. M. and W. Perraudin, 2003, Real options and preemption under incomplete information, Journal of Economic Dynamics & Control, 27, 619643 Mason, R. and H. Weeds, 2001, Irreversible investment with strategic interactions, CEPR Discussion Paper No. 3013, London, United Kingdom Mason, R. and H. Weeds, 2003, The failing firm defence: Merger policy and entry, University of Southampton, Southampton, United Kingdom Miltersen, K. R. and E. S. Schwartz, 2003, R&D investments with competitive interactions, EFA 2003 Annual Conference Paper No. 430 Murto, P., 2004, Exit in duopoly under uncertainty, The Rand Journal of Economics, 35, 111127 Murto, P., E. Nskkl, and J. Keppo, 2004, Timing of investments in oligopoly under uncertainty: A framework for numerical analysis, European Journal of Operational Research, 157, 486500

117

Strategic investment under uncertainty: A survey of game theoretic real options models

Nielsen, M. J., 2002, Competition and irreversible investments, International Journal of Industrial Organization, 20, 731743 Pawlina, G. and P. M. Kort, 2001, Real options in an asymmetric duopoly: Who benefits from your competitive disadvantage? CentER Discussion Paper 2001-95, Tilburg University, CentER, Tilburg, The Netherlands Pawlina, G. and P. M. Kort, 2002, The strategic value of flexible quality choice: A real options analysis, Working paper, Tilburg University, Tilburg, The Netherlands. Pawlina, G. and P. M. Kort, 2003, Strategic capital budgeting: Asset replacement under market uncertainty, OR Spektrum, 25, 443480 Pennings, E., 2004, Optimal pricing and quality choice when investment in quality is irreversible, The Journal of Industrial Economics, 52, 569589 Smets, F., 1991, Exporting versus FDI: The effect of uncertainty, irreversibilities and strategic interactions, Working Paper, Yale University, New Haven, Connecticut, United States of America Sparla, T., 2004, Closure Options in a Duopoly with Strong Strategic Externalities, Zeitschrift fr Betriebswirtschaft, 67, 125-155 Thijssen, J. J. J., 2004, Investment under Uncertainty, Coalition Spillovers and Market Evolution in a Game Theoretic Perspective, Kluwer Academic Publishers, Dordrecht, The Netherlands Thijssen, J. J. J., K. J. M. Huisman, and P. M. Kort, 2002, Symmetric equilibrium strategies in game theoretic real option models, CentER Discussion Paper 2002-81, Tilburg University, CentER, Tilburg, The Netherlands Weeds, H., 2002, Strategic delay in a real options model of R-D competition, The Review of Economic Studies, 69, 729747 Williams, J. T., 1993, Equilibrium and options on real assets, The Review of Financial Studies, 6, 825850

118 - The

Journal of financial transformation

Assets

Pricing with time-technology and timescapes Inflation-induced valuation errors in the stock market Is the investor sentiment approach the solution to the IPO underpricing phenomenon? The credit spread puzzle Impact of seasonality on inflation derivatives pricing Pricing default-free fixed rate mortgages: A primer Efficient pricing of default risk: Different approaches for a single goal

Pricing with time-technology and timescapes

Bala R. Subramanian Associate Professor, DeVry University
Wisdom, knowledge, and information are required to survive in the 21st century. Is there a way to provide these without having to spend years in schools, colleges, and universities? It would seem humanity is on the verge of making such a transformation. This article discusses some of the issues and their potential in a very novel way. The economics of pricing is a complex subject and includes
1

Review, the 1974 Nobel Laureate in Economics, Frederick Hayek (1945) admired the efficiency with which the market processes the pricing information and sparingly allocates the scarce resource in a multiplicity of its use throughout the economy. So, pricing has become central to our economic processes and hence perhaps even for the wellbeing of human kind.

many fields of human knowledge and endeavors. The principle of scarcity has been the basis of economic models. Scarcity2 is a function of time. Hence, speed [Mansfeld (2004)] is valued, advocated, and sought to bring about financial transformations of economic systems. Time and its measurement have thus become central to modern economies and their studies. Economists have computed their time series of economic data based on the geographi3

In an Internet culture, a functioning knowledge society has come to rely on the pricing mechanism to provide the appropriate signals necessary to adjust allocations of not only all resources, but also to impart value to actions in many contexts. Through differential pricing [Clemons et al. (2002)] it is believed, disintermediation could be avoided in the presence of customer heterogeneity. This paper examines the implications of this conventional wisdom and proposes an alternative to the complete dependence on the geographical time scale in order to solve societal conflicts. This proposed framework might foster true egalitarianism and pragmatism [James (2000)] which might provide true and complete freedom from any and all societal strife which has been the declared goal of modern governments [Davies (1996)]. The flow of time in addition to being geographical (i.e., based on the location of our planet in the solar system and the longitude and latitude on the globe, or what is denoted as spatial) is a function of societal relationships. When it comes to pricing or establishing a value of exchange the latter measure of time and its duration has a greater significance than understood by the economists and scientists of the industrial era who were preoccupied with our physical universe and the human relationship with that universe. When we eliminate an important step or an intermediary layer of societal-time4 that has a vital role on the outcome, we risk losing valuable information necessary for its measurement and inclusion in our valuations. Before we include the spatial time in our computations we need to account for the time and the durations inherent in societal relationships and processes. If we do not account for these variations and the resulting costs associated with that phenomenon of time then the outcomes of our

cal time (solar clocks and solar calendar); both micro and macro studies of economic variables are computed using this geographical time parameter. In international trade world wide pricing has acquired a unique complexity due to its influence on the profits, taxes, and regulations pertaining to its implications on domestic market structures [Wild et al. (2003)]. In the pricing of the financial products such as options, futures, and bonds time durations help measure the risks [McDonald (2004)]. The durations included in these computations also are traditionally a function of the solar clock times as well. Price differentials, both in the products and factors markets, cause economic agents to either speculate or conduct time and/or market arbitrages. Here again, the solar clock time plays the central role. An outcome of either a profitable or unprofitable speculation is essentially a solar clock dependent phenomenon. Determinants of economic profits are functions of costs and costs are measured using the geographic time units. When it comes to the costing of information we include the value of time spent to acquire it, since time is money. But, here again we are consistent in the use of solar clocks and solar calendar time to help value this scarce economic resource. According to Paul Gregory, writing in American Economic

120

A money price is a price expressed in monetary units, a relative price is a price expressed in terms of other commodities, while a price system is the entire set of relative prices. 2 Scarcity exists when the amount of the good or resource offered is less than what users would want if it were given away free.

3 A time series is a measurement of one or more variables over a designated period of time, such as months, years, or quarters. 4 Societal-time is a new phrase coined by this author to denote the unique combination of the relationships of individuals to one another and to their societal institutions, processes, and methods that govern the outcome of such relationships.

decision-making processes become unpredictable. In the parable of the householder (King James Version of Bible, Matthew ch. 20) the Kingdom of Heaven prices the services of a laborer equally at the first hour and the eleventh hour and attributes this inequity to the goodness of that householder. In the 21st century how can we account for our business cycles and pricing inequities beyond statistical and data input errors? The most effective way to include and measure the goodness of the heavenly householder and its present day equivalent societal values is to measure the phenomena of societal-time, using timescapes [Subramanian (2003b)] and by the development of time-technologies akin to our present day space-technologies that could further our knowledge about our valuations and perhaps improve the dismal status of our economic sciences. The flow of the geographical-time is a summary of many a societal-times akin to the white light from the suns rays we see which is a summary of many other colors of differing frequencies in the visible spectrum of lights. If our economic data series fail to differentiate the many shades and color values of time and the durations associated with those units of measures, then the interpretation of those series become questionable at best and meaningless and incorrect for useful economic analysis and its application at its very least. Time-technology is the science of decoding the mystery of geographical time and its relationship to human existence. The geographical time produces seasons and weather conditions that affect the lives of many kingdoms (plant, animal, and many other organisms). The bio-clocks in these kingdoms govern the flow of events necessary for their survival and their interdependence. In like manner the economic-clock that governs the value creation processes of the universe balance our ecosystems. Our present day monetary, pricing, trading, and exchanging mechanisms seems to have been created without regard to this truism. Could this anomaly be the reason that produces our social and family strife? If so, what could be done to improve this? How can we create and use the many economic-clocks that could provide the correct rhythms for the

value creation cycles of individuals and communities that create and sustain the wealth of nations? Many authors [Hayami (1997)] and economists have correctly identified the interdisciplinary nature of this wealth building task but have failed to recognize that one of the main reasons for this complexity is the misunderstanding and perhaps even misuse of our geographical time as a substitute for the societal time. In my view the measurement and streamlining of the societal processes using individual and appropriate clocks could lead to the successful integration of global economies much more easily. Even if we could successfully integrate regional economies using common markets and could modify our human behaviors to a single and uniform set of standards, which may take many years and many a political battles, liberating the human capital and obtaining the liquidity [Shojai et al. (2001)] would be impossible using our geographic-clocks and that measure of durations alone because of the variations inherent in the creativity, health, individuality, and thought processes of humans. Data mining experts in finance [Kovalerchuk and Vityaev (2004)] have concluded that matching tasks with methods is a very complex endeavor leaving one to conclude, it is more an art than a science. The noise described with those time series could perhaps be due to the failure to associate data with a societal-time framework prior to their conversion to the geographical-time used in these computations. With the introduction of the Internet and the benefits of ecommerce [Litan and Rivlin (2001)] the geographical time unit has entered a new phase of obscurity for economic data. According to Prof. Samuelson (2004), there is no valid reason to think that if we were only a little more knowledgeable and a little more energetic, we could converge on highly accurate macro forecasts. Mass behaviors answer to no simple discoverable set of rules. The best way to discover rules of behavior is by mapping events and their durations on a virtual timescape independent of any geographical limitations. The U.S. asset management industry as well as the global investors

121

whose equity market capitalization amounted to U.S.$36 trillion in 1999 [Shojai and Preece (2001)], lack a uniform measure of time to value their interests. The future no longer is merely represented by the calendar year millennia but resides in the thought processes of the future generations [Feiger and Shojai (2003)]. Any data collection and its use for meaningful analysis should begin by associating that data with the individual entities that cause an event to occur in their respective time horizons and yet be amenable to integration. The creation of trust and a sustainable point of differentiation for financial relationships have to begin with the use of timescapes that show changes over many generations and many societal cultures. Brands conceived in the transactions era may have to be evaluated in the relationship era [Boone and Kurtz (2003)] using cultural and social contexts of greater depth to show product performance. The predicted disintermediation [Shojai and Wang (2003)] of financial intermediaries might not happen even after the dust settles over the internet landscape as those authors argue due to the fact that improving the efficiency of the world we live in is likely to remain invisible in solar timescape. I tend to agree with Myron Scholes (2004) that complex securities are hard to value especially when using solar time alone. Using societal timescapes that measure the efficiency of markets in meaningful ways to an individual investor or a group of investors of uniform demography might bring about the much needed transparency in that industry for its survival. If we look at the evolution of currency [Turk (2004)] and try to provide all of the properties of money to money substitutes using new technologies as suggested by that author then, to guarantee the purchasing power to money substitutes we need to develop timescapes and time-technologies [Subramanian (2003)]. If we want to extract the business value of IT by increasing its usage [Marchand (2004)], we should be able to link every user to that technology in meaningful ways using each users societal time and societal roles by providing the timescapes to software agents that can help that individual enhance his/her

productivity by being in multiple locations concurrently to safeguard their interests in all relevant virtual meeting places. Collaborative networks, be they financial trading (CTN) [Shojai (2001)] or of any other kind, should be capable of detecting nuances to data not only according to the Enterprise Integration (EAI) principles of Extract, Transform and Load (ETL) [Dilkes (2004)] or Extract, Load and Transform (ELT) but also relate the many contexts affected by its societal relationships contained in timescapes [Subramanian (2004)]. The complexity of wealth management in the 21st century [Feiger and Shojai (2001)] may not be addressed fully without a sophisticated link to the social contexts of the individual economies and the global citizenry. Trust [Zak (2003)] and risk [Bryan (2002)] are all related to social contexts and societal processes that may not be summarized in solar timescape measurements alone. If the laboratory experiments belie the predicted outcome of Nash equilibrium we could attribute this deviation to the lack of data mapping of societal processes and their significance for understanding human behavior. Fear of the unknown is the natural outcome of giving greater importance to the landscapes and the real estates than to the timescapes and the human societal values. The perceived law of motion [Kedar-Levy (2003)] (PLM) and the actual law of motion (ALM) could be based on societal time and long periodic events caused by human sociology rather than market portfolio valuations or simple dynamic asset pricing models. Economists [Walter (2001)] will have to look beyond a New Economy to study the causes and effects of events. Integration for banks [Derrer (2002)] can not be complete by bringing together a wide range of different processes, based on different technologies, and synthesize them into a coherent whole unless that includes the human values and the human societal processes on which financial assets are created and used. The virtual matching utilities (VMU) that allow for the seamless real time matching of trade data throughout the trade lifecycle [Walsh (2001)] need not limit its concept of real-time to the present day geographical time. By not doing so, the cul-

122 - The

Journal of financial transformation

tural aspects of a societal relationship that caused that transaction are not utilized and are lost in the analytical studies. If we extend this understanding to hedge fund investing [Lhabitant (2003)] we might question the effectiveness of a diversification strategy that is limited to styles and market inefficiencies alone. Diversification by demographic characteristics and the societal timescapes of the investing and the marketing members of the global citizenry is likely to produce the hedge funds of the right radar type.

Conclusion
Societal time and timescape is a dynamic and comprehensive medium to value exchanges. Societal-time pricing mechanism could be trusted to incorporate all relevant variables in realtime and the economic analysis of that data-time series would be able to better signify the effects of macro and micro policy. Information technology has a role to play in bringing about the integration of an individual with all of the communities and all of the roles taken-up in those communities and their impact on the rest of those community members and their roles. Since all these relationships are changing, evolving, and occurring simultaneously data, information, knowledge, and wisdom have to be extracted continually to fit a societal need and its scope. The technology, architecture, and the logic to transform the meaning of these relationships are likely to preoccupy the minds of many next generation solution providers for several decades to come. I hope they could be trusted to look into these societal-time phenomena as well to impart a degree of utility to their integration mechanisms [Thillairajah (2004)].

References
Boone and Kurtz, Contemporary Business, 11th ed., Thompson-South-Western, 2005 p.13 Bryan, A., 2002, The many faces of risk, Journal of Financial Transformation, 5, 66-67 Clemons, et al., Impacts of e-Commerce and enhanced information endowments on financial services: Transparency, differential pricing and disintermediation, Journal of Financial Transformation, April 2002 p.9-18 Davies, G., From opportunity to entitlement : the transformation and decline of Great Society liberalism, 1996

Derrer, A., 2002, Integration, Integration, Integration, Journal of Financial Transformation, 6, 34-35 Dilkes A., 2004, Data management in financial services 2004 and beyond, Journal of Financial Transformation, 11, 58-61 Feiger, G., and S. Shojai, 2001, Wealth management in the 21st century: The imperative of an open product architecture, Journal of Financial Transformation, 3, 74-76 Feiger, G., and S. Shojai, 2003, 7, Market credibility and other dietary fads, Journal of Financial Transformation, 7, 63-70 Gregory, P. R., Essentials of Economics, 6th edition Hayami, Y., 1997, Development economics: From the poverty to the wealth of nations, Oxford; New York Oxford University Press Hayek, F. A., (1945) The Use of Knowledge in Society, American Economic Review, 35:4: 519-530 James, W. 2000, Pragmatism and the meaning of truth, Harvard University Press Kedar-Levy, H., 2003, Technology shocks and financial bubbles, Journal of Financial Transformation, 7, 53-62 Kovalerchuk, B., and E. Vityae, 2004, Data Mining in Finance: From extremes to realism, Journal of Financial Transformation, 11, 81-89 Lhabitant, F-S, 2001, Hedge fund investing: A quantitative look inside the black box, Journal of Financial Transformation, 1, 82-90 Litan, R. E, and. A. M. Rivlin, 2001, Projecting the Economic Impact of the Internet, Journal of Financial Transformation, 2, 35-41 Mansfield, W., 2004, A Single Market for Hedge Funds, Journal of Financial Transformations, Vol. 10, 80-81 Marchand, D. A., 2004, Extracting the business value of IT: It is usage, not just deployment that counts! Journal of Financial Transformation, 11, 125-131 McDonald, Robert L., Derivative securities, Pearson-Addison Wesley, 2003 p.218 Scholes, M. S., 2004, The future of hedge funds, The Nobel Laureate view, Journal of Financial Transformation, 10, 8-11 Shojai, S., 2001, Financial Trading Networks, Journal of Financial Transformation, 1, 30-37 Shojai, S., and R. Preece, 2001, The future of the US asset management industry, Journal of Financial Transformation, 1, 72-79 Shojai S., and S. Wang, 2003, Transformation: the next wave, Journal of Financial Transformation, December, 9, 11-16 Shojai, S., P. Gray, C. Keeling, and S. Wang, 2001, Liberating human capital: The search for the new wave of liquidity, Journal of Financial Transformation, 3, 117-126 Subramanian, Bala R., 2003a, Time-technology centers- Engineering Education for designing and using timescapes. Mid-Atlantic Conference of American Society of Engineering Educators (ASEE), Saturday April 12 Subramanian, B. R., 2003b, A case for U.S. Currency (Dollar) re-valuation, Unpublished paper Subramanian, B. R., 2004, Teaching economics in the 21st century, Paper presented on October 23rd at West Coast Conference, California State University, Fullerton, CA Samuelson, P., 2004, The World of Economic Data, The Nobel Laureate view, Journal of Financial Transformation, 11, 6-7 Thillairajah, V., 2004, Thoughts from The Integration Consortium, DM Review, December Turk, J., 2004, The evolution of currency, Journal of Transformation, 12, 108-109 Walsh, P., 2001, STP/T+1: The European Challenge, The Journal of Financial Transformation, 3, 53-61 Walter, N., 2001, Effects of September 11th on the World Economy, Journal of Financial Transformation, 3, 13-14 Wild, J. et al., International Business, Prentice-Hall, 2003 p.418-420 Zak, P. J., 2003, Trust, Journal of Financial Transformation, 7, 17-24

123

Inflation-induced valuation errors in the stock market1

Kevin J. Lansing, Senior Economist, Research Department, Federal Reserve Bank of San Francisco
The long-run rate of return on stocks is ultimately determined by the stream of corporate earnings distributions (cash flows) that accrue to shareholders. In assigning prices to stocks, efficient valuation theory says that rational investors should discount real/nominal cash flows using real/nominal interest rates. A recent front-page article in the Wall Street Journal, however, documented an increasing tendency among economists to move away from theories of efficient stock market valuation in favor of behavioral models that emphasize the role of irrational investors [Hilsenrath (2004)]. Twenty-six years ago, Modigliani and Cohn (1979) put forth the hypothesis that investors may irrationally discount real cash flows using nominal interest rates a behavioral trait that would lead to inflation-induced valuation errors. This article examines some recent research that finds support for the Modigliani-Cohn hypothesis. In particular, studies show that the Standard & Poors (S&P) 500 stock index tends to be undervalued during periods of high expected inflation (such as the late 1970s and early 1980s) and overvalued during periods of low expected inflation (such as the late 1990s and early 2000s). This result implies that the long bull market that began in 1982 can be partially attributed to the stock markets shift from a state of undervaluation to one of overvaluation. Going forward, the return on stocks could be influenced by a shift in the opposite direction. This valuation technique is often referred to as the Fed model, but it is important to note that the Federal Reserve neither uses nor endorses it. Many authors, including Ritter and Warr (2002) and Asness (2003), have pointed out that the practice of comparing a real number like the E/P ratio to a nominal yield does not make sense. It would be more correct to compare the E/P ratio to a real bond yield, but that comparison still ignores the different risk characteristics of stocks versus bonds and the reality that, over the past four decades, cash distributions to shareholders in the form of dividends have averaged only about 50% of earnings.

Long-run yield comparison

Wall Street practitioners often argue that comparing the E/P ratio to a nominal bond yield is justified by the observed comovement of the two series in the data. I tested this hypothesis by looking at the relationship between the E/P ratio of the S&P 500 index (based on 12-month trailing reported earnings) and both the nominal and real yields on a long-term U.S. Treasury bond (with a maturity near 20 years) over the period December 1945 to June 2004. The real yield is obtained by subtracting expected inflation from the nominal yield. Expected inflation is constructed as an exponentially weighted moving average of past (12-month) CPI inflation, where the weighting scheme is set to approximate the time-series behavior of the one-year-ahead inflation forecast from the Survey of Professional Forecasters. I found that the E/P ratio is more strongly correlated with movements in the nominal yield than with the real yield, particularly since the mid-1960s. This result is a puzzle from the perspective of efficient valuation theory. Observed movements in the nominal yield can be largely attributed to changes in expected inflation which, in turn, have been driven by changes in actual inflation. If investors were rationally discounting future nominal cash flows using nominal interest rates, they would understand that inflation-induced changes in the nominal bond yield are accompanied by inflation-induced changes in the magnitude of future nominal cash flows. Indeed, Asness (2003) shows that low-frequency movements

Stock as a disguised bond

Famed investor Warren Buffett has described a stock as a type of disguised bond. A stock represents a claim to a stream of earnings distributions whereas a bond represents a claim to a stream of coupon payments. Given that stocks and bonds can be viewed as competing assets in a portfolio, investors may wish to compare the valuations of these two asset classes in some quantitative way. Wall Street practitioners typically compare the earnings yield on stocks (denoted here by the E/P ratio, the inverse of the price-earnings ratio) with the nominal yield on a long-term U.S. Treasury bond. Stocks are supposedly undervalued relative to bonds when the E/P ratio exceeds the nominal bond yield and overvalued when the E/P ratio is below the nominal bond yield.

124 - The

Journal of financial transformation

This article is adapted from the Economic Letter series published by the Federal Reserve Bank of San Francisco. The views stated here are those of the author and not necessarily those of the Federal Reserve Bank of San Francisco or of the Board of Governors of the Federal Reserve System.

in the U.S. inflation rate are almost entirely passed through to changes in the growth rate of nominal earnings for the S&P 500 index. Thus, to a first approximation, a real valuation number like the E/P ratio (or its inverse, the P/E ratio) should not move at all in response to changes in the nominal bond yield. Similar logic applies to the residential housing market; the ratio of house prices to rents (a real valuation number) should not be affected by inflation-induced changes in mortgage interest rates. Nevertheless, the data show that the ratio of house prices to rents in the U.S. economy has been trending up since the mid-1980s as inflation and mortgage interest rates have been trending down. The observation that real valuation ratios are correlated with movements in nominal interest rates lends credence to the Modigliani-Cohn hypothesis. Investors and home buyers appear to be adjusting their discount rates to match the prevailing nominal interest rate. However, for some unexplained reason, they do not simultaneously adjust their forecasts of future nominal cash flows, i.e., earnings distributions or imputed rents. The failure to take into account the influence of inflation on future nominal cash flows is an expectational error that is equivalent to discounting real cash flows using a nominal interest rate.

out that this simple behavioral model can account for 70% of the variance in the observed E/P ratio from December 1945 to June 2004. In contrast, an otherwise identical model that employs real explanatory variables can account for only 26% of the variance in the observed E/P ratio. When the fitted E/P ratios from both the nominal and real models are inverted for comparison with the observed P/E ratio of the S&P 500 index we find that the nominal model does a much better job of matching the level and volatility of the observed P/E ratio. The real model, in contrast, predicts a relatively stable P/E ratio one that remains close to its long-run average over much of the sample period, particularly during the so-called new economy years of the late-1990s and beyond. At the end of the sample period in June 2004, the observed P/E ratio is 20.2. The predicted P/E ratio from the nominal model is 24.1, while that from the real model is 14.8.

Inflation-induced valuation errors

The predicted P/E ratio from real model can be interpreted as an estimate of the rational (or fundamentals-based) P/E ratio because the model assumes that investors discount real cash flows using real interest rates. In this case, the difference between the observed P/E ratio and the real-model prediction can be viewed as a measure of overvaluation. When I plot valuation errors against expected inflation, I find that overvaluation (measured as a percent of fundamental value) is negatively correlated with the level of expected inflation; overvaluation tends to be high when expected inflation is low, and vice versa. According to the analysis, overvaluation was highest during the late 1990s and early 2000s a period when expected and actual inflation were quite low. The late 1990s witnessed the emergence of the biggest bubble in history. The bubble burst in March 2000, setting off a chain of events that eventually dragged the U.S. economy into a recession in 2001 [Lansing (2003)]. The recession-induced collapse in corporate earnings caused the market P/E ratio to spike above 45, thereby exceeding the previous high that had prevailed near the bubble peak. In contrast, the high inflation era of the late 1970s and early 1980s was characterized by substantial undervaluation, as the market P/E ratio languished below its long-

Changing risk perceptions

Asness (2000) shows that movements in the E/P ratio also appear to be driven by changes in investors risk perceptions. According to Ibbotson Associates, monthly stock return volatility has declined relative to monthly bond return volatility, regardless of whether returns are measured in nominal or real terms. If investors risk perceptions are based on their own generations volatility experience, then stocks will appear to have become less risky relative to bonds over time. Following Asness, a statistical model of investor behavior can be constructed by regressing the E/P ratio on a constant term against three nominal explanatory variables, the yield on a long-term government bond, the volatility of monthly stock returns over the preceding 20 years, and the volatility of monthly bond returns over the preceding 20 years. It turns

125

run average for more than a decade. These findings reinforce those of Ritter and Warr (2002) and Campbell and Vuolteenaho (2004) who find strong support for the Modigliani-Cohn hypothesis using more sophisticated empirical methods.

References
Asness, C. S., 2003. Fight the fed model, Journal of Portfolio Management, Fall, 30:1, 11-24 Asness, C. S., 2000. Stocks versus bonds: Explaining the equity risk premium, Financial Analysts Journal, 56:2, 96-113 Campbell, J. Y., and T. Vuolteenaho, 2004, Inflation illusion and stock prices, American Economic Review, Papers and Proceedings, 94, 19-23 Hilsenrath, J. E., 2004, Stock characters: As two economists debate markets, the tide shifts, Wall Street Journal, October 18, p. A1 Lansing, K. J., 2003. Growth in the post-bubble economy, FRBSF Economic Letter, 2003-17 (June 24), http://www.frbsf.org/publications/economics/letter/2003/el200317.html Lansing, K. J., 2004, Lock-in of extrapolative expectations in an asset pricing model. FRBSF, Working Paper 2004-06, http://www.frbsf.org/publications/economics/papers/2004/wp04-06bk.pdf Modigliani, F., and R. A. Cohn, 1979, Inflation, rational valuation and the market, Financial Analysts Journal, 35, 24-44 Ritter, J. R. and R. S. Warr, 2002, The decline of inflation and the bull market of 1982-1999, Journal of Financial and Quantitative Analysis, 37:1, 29-61 Ibbotson Associates, Stocks, bonds, bills, and inflation 2004 Yearbook. Chicago

Conclusion
In recent years, contributors to the rapidly growing field of behavioral finance have been refining a new class of asset pricing models. These models are motivated by a variety of empirical and laboratory evidence which shows that peoples decisions and forecasts are often less than fully rational. Simple behavioral models can account for many observed features of real-world stock market data including: excess volatility of stock prices, time-varying volatility of returns, long-horizon predictability of returns, bubbles driven by optimism about the future, and market crashes that restore attention to fundamentals [Lansing (2004)]. Twenty-six years ago, Modigliani and Cohn (1979) put forth a behavioral finance model that predicted mispricing of stocks in the presence of changing inflation. The co-movement of the stock market E/P ratio with the nominal bond yield observed since the mid-1960s (when U.S. inflation started rising) is consistent with the Modigliani-Cohn hypothesis. A regression model that includes a constant term and three nominal variables can account for 70% of the variance in the observed E/P ratio over the past four decades. However, as noted by Asness (2003), the success of this model in describing investor behavior should not be confused with the models ability to forecast what investors should really care about, namely, long-run real returns. Investors of the early 1980s probably did not anticipate the 20-year declining trend of inflation and nominal interest rates that helped produce above-average real returns as stocks moved from a state of undervaluation to one of overvaluation in the manner described by Modigliani and Cohn (1979). Todays investors may suffer the opposite fate if a secular trend of rising inflation and nominal interest rates causes the stock market to move back towards a state of undervaluation.

126 - The

Journal of financial transformation

Is the investor sentiment approach the solution to the IPO underpricing phenomenon?1
Andreas Oehler, Professor of Finance, Bamberg University Marco Rummer, Ph.D. Student, Teaching and Research Assistant, Department of Economics and Management, Bamberg University Peter N. Smith, Professor of Economics and Finance, University of York

During the process of going public the company owners offer shares to outsiders for the first time and naturally try to raise as much money as possible in order to be able to finance future endeavors and growth. To achieve this, an appropriate offer price has to be found which should not be set too low, as this will be like a generous present to the new investors, or too high, as this dispels interested outsiders. With the purpose of finding the perfect price and preparing the company to go public, issuers spend on average a tremendously high 7% of the gross proceeds in underwriter fees [Chen and Ritter (2000)]. Therefore, it is rather surprising that most of the firms show a significant increase in share price between the offering and the first trading day. These astonishing and time varying initial returns have been labeled the underpricing phenomenon and have been confirmed for all major stock markets around the world [Loughran, Ritter and Rydqvist (1994)]. A vast number of theoretical models and even more empirical examinations have been proposed since the pioneering work of Logue (1973) and Ibbotson (1975). From our point of view there are two major possible scenarios which could cause the empirically observed increase in stock prices between the offering date and the first trading day. Firstly, it could be that the offer price is set too low due to ex-ante uncertainty about the true market value of the newly issued stock. Secondly, the offer price might be at a fair value but demand for the IPO and therefore the sentiment of investors is so overwhelmingly high that this generates the observed initial returns. The first strand of explanation focuses mainly on the assumption that underpricing is a deliberate act by either the underwriter or the issuer due to ex-ante uncertainty about the true market value of the firm planning to go public for the first

time. This argument is normally based on the assumption that not all groups involved in the IPO pricing process possess the same level of information, given the missing track record of the firm. Rock (1986) suggests, in his pioneering work, that underwriters have to underprice IPOs in order to compensate uninformed investors for receiving most of the least profitable issues, as the informed investors only place orders on the advantageous IPOs due to their informational advantage. Welch (1989) believes that high-quality issuers can afford to sell their shares at a discount to investors in order to signal the companys high quality. Given the degree of initial returns with a maximum of 444% and a mean of 52% in our sample examined below, these theories are hardly reconcilable with empirical findings. This is especially the case during the dot-com boom where most investors appeared to not really care about risk and uncertainty. Coming to a similar conclusion, Ritter and Welch (2002) argue that these theories are unlikely to explain the persistent pattern of high initial returns during the first trading day and discuss additionally that over-enthusiasm among retail investors may explain the pattern of high initial returns. The second strand of research therefore focuses on behavioral finance in order explain the pattern of time variance and persistence of initial returns. From our point of view the notion of investor sentiment seems to be most promising. This approach argues that high and fluctuating initial returns are caused by demand from different groups of investors and are not induced by a required discount due to asymmetrically distributed information and ex-ante uncertainty. Cornelli et al. (2004) corroborate our argument, that high pre-IPO prices, which indicate overly optimistic investors, are a good predictor of high initial returns during the first trading day. Ljungqvist et al. (2004) show for the first time in a rational

The paper benefited from discussions with Wolfgang Bhler, James Gill, Leslie Godfrey, John Hutton, Aydin Ozkan, Klaus Rder, Ray da Silva Rosa, Dirk Schiefer, Oliver Schwindler and Ben Werner. We also thank seminar participants at the Bambeg University and Freiburg University and participants at the workshop

Economics Meets Psychology 2004 held by the Deutsche Bundesbank, the Portuguese Finance Network Conference 2004, SAM/IFSAM-Conference 2004, Money Macro and Finance Conference 2004 and European Investment Review Conference 2004. The usual disclaimer applies.

127

model that investors are willing to pay a price in excess of their rational belief if sentiment is high towards newly issued stocks.

demand for the stocks to be issued will then lead to a reduction in initial returns. On the other side, a decrease in the bookbuilding range due to more potential demand will lead to an increase in initial returns.3 Our other measures capturing the impact of investor sentiment are related to the behavior of prices prior to the offering. The variable gm measures the impact of grey market4 prices, which are simply calculated as the midpoint of the bid-ask spread during the last pre-IPO trading day. We expect a positive sign on gm in explaining initial returns as a high demand should cause high pre-IPO trading prices, which additionally leads to high underpricing. Finally, a widely used measure of investor sentiment is the performance of a suitable stock market index prior to the offering. As we are analyzing IPOs going public on the German Neuer Markt we use the Nemax-All-Share-Index as this benchmark.5 This index covers all stocks listed on this stock market segment. We expect a positive sign for the variable nemax, as an increase in the stock market should indicate a positive attitude in the market and therefore should lead to high initial returns.

Empirical analysis
Given the argument above one could conclude that investor sentiment indeed seems to have a strong influence on initial returns.2 In order to provide an additional empirical explanation we combine the analysis of Oehler et al. (2004), Lffler et al. (2004) and Cornelli et al. (2004).

Hypothesis
To help assess which strand of research provides the least explanation of the behavior of initial returns we use the variable bbwidth which describes the width of the bookbuilding range, measured as the difference between the upper and lower bounds of the initial price range. With this variable we are able to measure the effect of ex-ante uncertainty, due to asymmetrically distributed information, and the consequence of investor sentiment, due to demand of potential investors, within one variable. The result simply depends on the sign of the impact of the bookbuilding range on initial returns measured from multivariate cross-section econometric analysis. If deliberate underpricing due to ex-ante uncertainty about the firm value is critical we expect a positive sign for the impact of the width of the bookbuilding range. For example, the issuer or underwriter is unsure about the potential demand for the stocks to be issued and the achievable market price. Therefore, they increase the width of the bookbuilding range. This is then expected to have a positive impact on initial returns, as traditional theories assume that increase in exante uncertainty about the firm value will lead to higher underpricing due to a necessary discount for potential investors. Contrary, if investor sentiment is critical we expect the impact of the width of the bookbuilding range to be negative. An increase in the bookbuilding range due to little expected

Database
We analyze 288 IPOs on the German Neuer Markt during the period 1998-2000. This stock market segment had been the centre of attention during the dot-com bubble and had become, besides NASDAQ, the market of choice for young firms undergoing IPOs during the dot-com boom. IPOs are identified by using the web-pages and the fact books of the Frankfurt Stock Exchange. All firms were double-checked using the web-pages of the Brsenzeitung and OnVista AG, as well as information from IPO prospectuses, companies homepages, and investor relations departments. Firms that had traded nationally or internationally prior to the offering at the Frankfurt Stock Exchange have been excluded from the study, as a market price already existed for these firms. The

128

2 For a more detailed empirical analysis on the importance of investor sentiment see for example Cornelli et al. (2004) or Oehler et al. (2004). 3 These arguments show that initial returns and underpricing, which have commonly been used interchangeably in prior research, describe in our opinion different effects, even if measured identically as the difference between the offer and first day trading price of an IPO. Therefore, the notion underpricing signifies that the described increase in share prices between primary and secondary market are due to a discount on the offer price. Contrary, the notion initial return refers to an increase due to the impact of investors demand on the first day trading price.

4 For a detailed description of the pre-IPO trading in Germany see Lffler, Panther and Theissen (2004) 5 Admittedly, this might induce some bias into our analyses as many firms listed in this index have gone public during the examined time period. Nevertheless, using a different benchmark like the DAX-Index, which covers the 30 biggest German blue chips, would not proxy the attitude towards IPOs well enough. Therefore, the bias is acceptable, as our goal is to study the impact of sentiment towards newly issued stocks.

information about the bookbuilding range is taken from the IPO prospectus, the web pages of the Frankfurt Stock Exchange, the Brsenzeitung and OnVista AG. The pre-IPO prices are obtained from Schnigge AG. The secondary market prices of the examined stocks are taken from the KKMDB database at the University of Karlsruhe6 and the daily closing prices of the Nemax-All-Share-Index are from Datastream.
Dependent variable constant bbwidth gm nemax R2 (adjusted) F-statistic T-stats in brackets IR -0.652b -0.129a 0.010a 1.18b 0.448 78.721a (a) Significant at the 1% level (b) Significant at the 5% level (-2.358) (-5.210) (7.700) (4.409)

Empirical analysis
Initial returns have an average of 52.35%, a minimum of -30.00% for Pixelpark AG and an astonishing 444.44% for Biodata Information Technology AG. It is very interesting to note that the Nemax-All-Share-Index performed very well before Biodatas IPO but very poorly prior to Pixelpark AG going public. The age of the examined firms, measured as the difference between the foundation and the initial offering, varies quite substantially and ranges from 62 years to one year. The latter group of IPOs (i.e. one year) therefore only marginally fulfill the minimum listing requirement imposed by the law during this time period. Our brief empirical analysis is based on the following regression, which is estimated using OLS (standard errors are adjusted for heteroskedasticity of the error term using Whites (1980) methodology): IRi = i + 1,ibbwidth + 2,igm + 3,inemax + i where the dependent variable IR is measured as the percentage difference between the first day trading price and the offering price. The explanatory variables bbwidth, gm, and nemax are described in detail above. The results can be found below in Figure 1. It can be seen that our model nearly explains 45% of the variations in initial returns, which is quite a substantial amount for a cross-sectional analysis of stock returns. The adjusted R2 is well above the 31.2% reported by Ljungqvist (1997) for

Figure 1

Germany. The variable bbwidth, whose sign is supposed to differ depending on the importance of investor sentiment and exante uncertainty, shows a negative relationship between the width of the bookbuilding range and the initial return during the IPO. From this we have to conclude that the investor sentiment effect is more important.7 This finding is supported by the positive signs on the highly significant variables gm and nemax. These estimates support the notion that investor sentiment is fundamental for explaining the IPO underpricing phenomenon. The magnitude of the impact of the variable nemax is quite substantial. According to our results, an increase of one percentage point in the Nemax-All-Share-Index prior to an IPO will lead to an increase in the initial return of 1.181 percentage points. This is explained, given our argument, by the positive sentiment in the stock market, which naturally has an effect on the demand for upcoming stock issues too. In our analysis, all variables are at least significant at the 5% level, meaning that the population value of our variables is in deed different from zero with a likelihood of 95%.

Conclusion
The extensively documented existence of positive initial returns during the first trading day of an IPO has been labeled the underpricing phenomenon. Most researchers assume that the issuer or underwriter fail to choose a higher offer price due to asymmetrically distributed information. In

6 We thank Hermann Gppl for providing the data. 7 For a more detailed analysis of the relative importance of investor sentiment in the IPO-Pricing process see Oehler, Rummer and Smith (2004).

129

a short literature review we have shown that the unusually high degree of first day returns can in fact hardly be due to deliberate actions by the underwriter or issuer. This argument is backed up by our brief empirical analysis during which we provide strong evidence that initial returns are determined by investor sentiment and not by ex-ante uncertainty.

References
Baron, D. P., 1982, A model of the demand for investment bank advising and distribution services for new issues, Journal of Finance, 37, 955-976 Chen, H. C., and J. R. Ritter, 2000, The seven percent solution, Journal of Finance, 55:3, 1105-1131 Cornelli, F., D. Goldreich, and A. Ljungqvist, 2004, Investor sentiment and pre-IPO markets, Working Paper, London Business School. Ibbotson, R. G., 1975, Price performance of common stock new issues, Journal of Financial Economics 2, 235-272 Ljungqvist, A. P., 1997, Pricing initial public offerings: Further evidence from Germany, European Economic Review, 41, 1309-1320 Ljungqvist, A. P., V. Nanda, and R. Singh, 2003, Hot markets, investor sentiment, and IPO pricing, forthcoming Journal of Business Lffler, G., P. F. Panther, and E. Theissen, 2004, Who Knows What When? The Information Content of Pre-IPO Market Prices, forthcoming Journal of Financial Intermediation. Logue, D. E., 1973, On the pricing of unseasoned equity issues: 1965-1969, Journal of Financial and Quantitative Analysis, 8, 91-103. Loughran, T., J. R. Ritter, and K. Rydqvist, 1994, Initial public offerings: International insights, Pacific-Basin Finance Journal, 2, 165-199. Oehler, A., M. Rummer, and P. N. Smith, 2004, IPO pricing and the relative importance of investor sentiment, Working Paper, University of Bamberg. Ritter, J. R., I. Welch, 2002, A review of IPO activity, pricing, and allocations, Journal of Finance, 57, 1795-1828 Rock, K., 1986, Why new issues are underpriced, Journal of Financial Economics 15, 187-212 Ruud, J. S., 1993, Underwriter price support and the IPO underpricing puzzle, Journal of Financial Economics 34, 135151 Welch, I., 1989, Seasoned offerings, initiation costs, and the underpricing of initial public offerings, Journal of Finance, 44, 421-450 White, H., 1980, A heteroskedasticity-consistent covariance matrix estimator and a direct test of heteroskedasticity, Econometrics, 48, 817-838

130 - The

Journal of financial transformation

The credit spread puzzle1

John Hull, Professor of Finance, Director, Bonham Centre for Finance and Maple Financial Group Chair in Derivatives and Risk Management, Rotman School of Management, University of Toronto Mirela Predescu, PhD. Candidate in Finance, Rotman School of Management, University of Toronto Alan White, Peter L. Mitchelson/SIT Investment Associates Foundation Chair in Investment Strategy and Professor of Finance, Rotman School of Management, University of Toronto
Estimation of default probabilities is becoming increasingly important in risk management. A new regulatory framework for banks, Basel II, is expected to be implemented in 2007. This gives banks much more freedom than its predecessor, Basel I, to use internal models to estimate default probabilities. The estimation of default probabilities is also important in the valuation of credit derivatives, the market for which is growing fast. Most credit derivatives provide payoffs contingent on whether particular companies default on their obligations. The estimation of default probabilities is therefore central to the evaluation of credit derivatives. One approach to estimating default probabilities involves looking at historical data and assuming that the future will in some sense be similar to the past. Another involves implying default probabilities from corporate bond yields. It turns out that the two approaches give quite different estimates. Our objective in this article is to quantify just how different the estimates are and then to provide possible reasons for the difference. The probability of default is greater than the credit spread because some recovery is made in the event of a default. We suppose a recovery rate of 40% (which is fairly close to the average of the recovery rates observed in practice). This means that for A-rated bonds the probability of default implied by bond prices is 0.769% / (1 - 0.4) or about 1.28% per annum. To look at this another way, when the average probaTo estimate the default probability implied by bond prices we first calculated the average spread over the risk-free rate for a seven year corporate bond. We used the Merrill Lynch bond indices between December 1996 and July 2004. Consider, for example, A-rated bonds. The average yield on bonds with a life of approximately seven years was 6.274% per annum. The average risk-free rate was 5.505% per annum. The average credit spread was therefore 0.769% per annum. The credit spread is an indicator of the probability of default. We receive an extra 0.769% per year in additional bond yield because we expect to lose about 0.769% of the principal per year due to defaults. are sometimes referred to as real-world default probabilities or physical default probabilities.

Default probability estimates

Figure 1 shows estimates of the average probability of default per year during a seven-year period calculated from historical data and bond prices. The historical data are cumulative default rates published by Moodys for the period between 1970 and 2003. This data give the probability of a company default during a seven-year period given that it has a particular credit rating at the beginning of the period. For example, a company that starts with an Aaa credit rating has an average default probability per year of 0.04% (or about 0.28% over the whole seven year period). A company that starts with a credit rating of Aa has an average default probability of 0.06% per year (or about 0.42% over the whole sevenyear period). Default probabilities implied from historical data

Rating

Default probabilities from historical data (% per year) 0.04 0.06 0.13 0.47 2.40 7.49 16.90

Default probabilities implied from bond prices (% per year) 0.67 0.78 1.28 2.38 5.07 9.02 21.30

Ratio

Difference

Aaa Aa A Baa Ba B Caa and lower

16.8 13.0 9.8 5.1 2.1 1.2 1.3

0.63 0.72 1.15 1.91 2.67 1.53 4.40

Figure 1: Average default probabilities per year over a seven-year period

The ideas in this paper will be published in a paper by the authors entitled Credit Spreads, Default Probabilities, and Risk Premiums in Journal of Credit Risk. We are grateful to Moodys Investors Service for supporting this research.

131

bility of default is 1.28% per year the loss of principal per year is 1.28% x (1 0.4) or 0.769%, which is the credit spread earned. Default probabilities backed out from bond prices in this way are known as risk-neutral default probabilities. Figure 1 shows that the ratio of the default probability estimates derived from bond prices to those derived from historical data decreases as the credit quality declines. However, the difference between the default probabilities increases as credit quality declines. The size of the difference between the two default probability estimates is sometimes referred to as the credit spread puzzle.

This leads many market participants to regard swap rates as better proxies for risk-free rates than Treasury rates.2 The credit default swap (CDS) market provides a way of estimating the benchmark risk-free rate used by participants in credit markets. If a five-year par yield corporate bond provides a yield of 6% and five-year protection can be bought against the issuer for 150 basis points a year, an investor can obtain a (approximate) risk-free return of 4.5% by buying the bond and buying credit protection. This suggests that the risk-free rate being used by market participants is 4.5%. Using this type of analysis across many corporations we estimate that the benchmark risk-free rate being used by market participants is the swap rate less 10 basis points.3 This is similar to estimates that have been made by Moodys KMV.4 We therefore set the risk-free rate equal to the seven-year swap rate minus 10 basis points in producing the results in Figure 1. It is worth noting that if we instead chose the risk-free rate to be to be the seven-year Treasury rate the default probabilities implied from bond prices would be even higher, making the differences in Figure 1 even more marked. For example the ratio of the risk-neutral to real world default probability for A-rated companies in Figure 1 would rise from 9.8 to over 15.

The benchmark risk-free rate

The default probabilities implied from a bond price depend on the bonds yield spread over the risk-free rate. The estimates in Figure 1 therefore depend critically on the assumption made about the risk-free rate. A natural choice for the risk-free rate is the Treasury rate. Treasury rates are yields on bonds that have no default risk. Furthermore the bond yield spreads that are quoted in the market are usually spreads relative to a Treasury bond that has a similar maturity. However, Treasury rates tend to be lower than other rates that have a very low credit risk for a number of reasons:

Risk premiums
Treasury bills and Treasury bonds must be purchased by

From the results in Figure 1 we can calculate the risk premiums earned by holders of corporate bonds. The expected excess return of corporate bonds over Treasuries has a number of components. One component is the difference between the Treasury yield and our estimate of the benchmark riskfree yield used by market participants. During the 92 months covered by our Merrill Lynch data this averaged 43 basis points. Another component is a spread to compensate for defaults. Given that we are assuming a recovery rate of 40%, this is the historic default probability in Figure 1 multiplied by 0.6. The final component is the extra risk premium earned by the holders of corporate bonds. Note that if the risk premium were zero, there would be no difference between historic default probabilities and those derived from bond prices.

financial institutions to fulfill a variety of regulatory requirements. This increases demand for these Treasury instruments driving the price up and the yield down.
The amount of capital a bank is required to hold to sup-

port an investment in Treasury bills and bonds is substantially smaller than that required to support a similar investment in other very low-risk instruments.
In the United States, Treasury instruments are given a

favorable tax treatment compared with most other fixedincome investments because they are not taxed at the state level.

132

2 A similar point is made forcefully by Duffee, G. R., 1996 Idiosyncratic variation of treasury bill yields, Journal of Finance, 51, 527-551. He argues that Since the early 1980s [Treasury] bill yields have become increasingly irrelevant as a benchmark. This is not news to market participantsbut nonetheless [is] likely a surprise to many academic economists.

3 See J. Hull, M. Predescu, and A. White, 2004 The relationship between credit default swap spreads, bond yields, and credit rating announcements, Journal of Banking and Finance, Nov 4 For example, Stephen Kealhofers estimate of the risk-free rate in his presentation at the Moodys/New York University conference on Recent Advances in Credit Risk Research in May 2004 was very close to our estimate.

bonds earn the risk premiums over the risk-free rate shown in
Rating Bond yield spread over treasuries
(% per annum)

Spread of risk-free rate used by market over treasuries

(% per annum)

Spread to compensate for historic default rate

(% per annum)

Extra risk premium

Figure 2? There appear to be four main reasons:

Corporate bonds are relatively illiquid. It is less easy to sell

(% per annum)

Aaa Aa A Baa Ba B Caa and lower

0.83 0.90 1.20 1.86 3.47 5.85 13.21

0.43 0.43 0.43 0.43 0.43 0.43 0.43

0.02 0.04 0.08 0.28 1.44 4.49 10.14

0.38 0.43 0.69 1.15 1.60 0.93 2.64

corporate bonds than many other types of securities. As a result investors in corporate bonds demand what is termed a liquidity risk premium. This is part of the risk premium shown in the final column of Figure 2. Most researchers estimate the liquidity risk premium to be between 10 and 25 basis points.
Figures 1 and 2 assume that traders use historical data to

Figure 2: Excess expected returns earned by bond traders

determine the probability of default in the future. In practice traders may be assigning positive subjective probabiliFigure 2 shows the allocation of the yield spread into its various components. For example, A-rated bonds earned 120 basis points more than Treasuries on average. Forty-three basis points of this is the difference between Treasuries and the markets risk-free benchmark. A further 8 basis points is necessary to cover defaults. The remaining 69 basis points is a risk premium earned by bondholders. We can see from Figure 2 that as the quality of the bond declines from Aaa to Ba the risk premium increases. It then declines as we move from Ba to B and increases sharply as we move from Ba to Caa. The extra risk premium reported in the last column of Figure 2 can be characterized as the expected return on a portfolio that is long a corporate bond and short a default-free bond. Our results are consistent with those produced by Edward Altman some time ago. He showed that, even after taking account of the impact of defaults, an investor could expect significantly higher returns from investing in corporate bonds than from investing in Treasury bonds. As the credit rating of the corporate bonds declined, the extent of the higher returns increased. (Altman found that B bonds ran counter to the overall pattern, just as we do in Figure 2.) ties to depression scenarios that are much worse than any seen since 1970. If we use average default statistics for the whole 1920 to 2003 period we find that the historic default probabilities in Figure 1 do increase somewhat. For Aaa the historic default probability increases from 0.04% to 0.06%; for Aa it increases from 0.06% to 0.22%; for A it increases from 0.13% to 0.29%; for Baa it increases from 0.46% to 0.73%; and so on.
Bonds do not default independently of each other. There

are periods of time when default rates are very low and other periods when they are very high. (Evidence for this can be obtained by looking at the default rates in different years. Between 1970 and 2003 the default rate per year ranged from a low 0.09% in 1979 to a high of 3.81% in 2001.) This phenomenon is sometimes referred to as credit contagion. It is a form of systematic risk (i.e., it is a form of risk that cannot be diversified away) and bond traders should require a return in excess of the risk-free rate for bearing this risk.
Bond returns are highly negatively skewed with limited

upside. As a result it is much more difficult to diversify risks in a bond portfolio than in an equity portfolio.6 A very large number of different bonds must be held. In practice, many bond portfolios are far from fully diversified. As a result bond traders may require an extra return for bearing unsystematic risk as well as for bearing the systematic risk mentioned above.

Reasons for the difference

Why are the probabilities backed out of bond prices in Figure 1 so much higher than those estimated from historical data? An equivalent question is: Why do investors in corporate

5 Altman, E. I., 1989 Measuring Corporate Bond Mortality and Performance, Journal of Finance, 44, 902-22. 6 See J. D. Amato and E. M. Remolona, The pricing of unexpected credit losses, Working paper, Bank for International Settlements, Basel Switzerland.

133

Conclusions
In comparing Figures 1 and 2 we see that there are huge differences between default probability estimates estimated from corporate bond prices and those estimated from historical data. These differences translate into relatively modest (but non-negligible) premiums demanded by bond traders for the risks they are bearing. We have identified four types of risk. The first is liquidity risk; the second is the risk that default losses may be much worse than anything seen in recent history; the third is the systematic (non-diversifiable) risk;s and the fourth is the risk which is diversifiable but in practice quite difficult to handle.

134 - The

Journal of financial transformation

Impact of seasonality on inflation derivatives pricing1

Nabyl Belgrade, Interest Rates Derivatives Research, CDC IXIS CM Eric Benhamou, Head of Quantitative Interest Rates Derivatives Research, CDC IXIS CM
Motivations and static seasonality modeling
Unlike interest rates, inflation forwards are not really smooth and exhibit repeatable seasonal patterns. These patterns take their roots in various recurrent economic phases, like consumer spending increase during Christmas, winter, and or summer sales, cyclic variations in energy and food consumption, and many other periodic effects. Consequently, the assumption of a linear growth for month-on-month inflation changes is clearly misleading. This month-on-month change may be far away from its monthly average for a month with strong seasonality effect. This advocates energetically for the incorporation of seasonality in any inflation pricing model. In fact, macro-economists have known for quite some time now that inflation markets should and do exhibit strong seasonality behavior [Bryan and Cecchetti (1995)]. But when inflation derivatives were still in their infancy with only a few traded products, seasonality was not really an issue. This is not the case any more. The market has experienced tremendous growth both in terms of hedging instruments (inflation linkers issued by states, sovereigns, or corporations) and OTC derivatives (inflation swap and other vanilla inflation derivatives). In Europe, an estimated 3 billion to 7 billions were being traded monthly during 2004. This compares with only 500 million a year before. And more is to come. In France, for instance, the total amount of regulated saving account Livret A, tied up to inflation rates, should reach 400 billions by the end of year. Seasonality modeling is now not a choice but a necessity. Incorporating seasonality within a stochastic model may at first sight look complex. One may think to include seasonality in the diffusion equation itself, via its drift, or the volatility term or both. The model could be taken as the limit of seasonal discrete time model like SARIMA and periodic GARCH. This would come at the price of both confusing the model and making it harder to calibrate. In this paper, we suggest a simpler approach, similar in spirit to the one used for end of year effect in interest rate derivatives. Instead of modeling seaThe incorporation of seasonality in our model is then straightforward. When computing the spot value of the CPI forward value CPI(0,T), maturing at time T with corresponding month m, we first look at the two adjacent liquid market points with corresponding times Td and Tu that bound lower and upper time T. We then compute the interpolated forward CPI value CPI(0,T) using the CPI market value CPI(0,Tdown) and CPI(0,Tup). This interpolation may be assuming linear, stepwise constant or cubic spline interpolation on either inflation spot zero coupon rates, forward CPI, or forward zero coupon. Insights of the paper are not on this interpolation method but rather on what follows. To the interpolated forward CPI value CPIint(0,T), we apply a seasonal bump, computed as the difference between the seasonal bump of month M and its interpolated seasonal bump using the same interpolation assumption as the one for the forward CPI. For the sake of simplicity, we will assume linear interpolation on forward CPIs. In this case, the seasonal bump for the time T and denoted by SB(T) is given by: SB(T) = B(m) (m md) sonality dynamically we use a static pattern to reshape the forward curve of CPIs. Hence, it is only the forward curve that is modified while the inflation dynamics stays unchanged. In addition, we take yearly seasonality on a monthly basis. Yearly pattern is not very restrictive as most of the seasonality is on a yearly basis. We, therefore, use a vector of yearly seasonal up and down bumps {B(i)}i=1..12 indexed by their corresponding months i with the convention that January equals 1.

B(mu) B(md) mu md

+ B(md)

(1.1)

where md and mu are the months corresponding to the times Td and Tu. We can notice that that the (linear) interpolation reshapes our vector of yearly up and down bumps to guarantee market points to have zero seasonality adjustment. This is because market points already incorporate seasonality in their price. In addition the seasonality interpolation is done consistently with the one of CPI to create similar interpola-

We would like to thank G. Couzineau and A. Bayle for useful remarks and comments. All ideas and opinion expressed herein are the ones of the authors and do not necessarily reflect those of CDC Ixis CM.

135

tion effect. Equation (1.1) assumes additive bumps, meaning that the corrected CPI is obtained as the interpolated CPI plus the seasonality bumps: CPI(0,T) = CPIint(0,T) + SB(T) (1.2)

data into a variety of components, more or less observable and easy to discriminate. We assume that our time series of CPI data is summarized by a general trend, that may be stochastic or not and that represents the long-term evolution of the CPI, a seasonal effect corresponding to yearly fluctuations and a noise represented by a random variable. Note also that a multiplicative correction is as an additive one on the logarithm of the CPI data. We will, therefore, look at additive correction as the processing for multiplicative is then

If we were to assume a multiplicative correction, we would correct the raw interpolated forward CPI as follows: CPI(0,T) = CPIint(0,T) * SB(T) (1.3)

straightforward. At this stage, we could decide to either assume a parametric form for the data and do an OLS to estimate the seasonal component, or determine sequentially the trend of the time series and on the de-trended data, the seasonal component. The first method is a parametric estimation of the seasonal

with the seasonal bump computed as the monthly bump renormalized by the interpolated bump as given in equation (1.4). This equation is the simple translation of the interpolation equation (1.1) of the seasonal bump but for multiplicative bump: SB(T) = B(m) / (m md)

B(mu) B(md) mu md

+ B(md)

(1.4)

component that spreads the noise error on both the trend and the seasonal component. But this advantage is counterweighted by the parametric assumption used to do everything in one go. The second method is a non-parametric method, straight application of the X11 method to our CPI data. Key advantage is to make no assumption on the trend and seasonal component. We let the filter recover the trend iteratively. However, because of this two-stage estimation, noise affects only the seasonal component and not the

Once the spot value for forward CPI has been estimated, it is easy to adapt the stochastic model described in Belgrade et al. (2004a, b). The forward CPI inflation index CPI(t,T) observed at time t that fixes at time T keeps its diffusion equation given by dCPI(t,T) CPI(t,T) = (t,T)dt + Inf(t,T)dW(t) (1.5)

trend.

Parametric estimation of seasonality

but the boundary condition is modified to include the seasonality adjustment. This is given by equation (1.2) in an additive correction and by equation (1.3) in a multiplicative one. In this framework, it is then easy to price any inflation linked derivatives. We understand that the seasonality estimation comes to the finding of a vector of 12 up and down (additive or multiplicative) bumps from historical CPI data. Following standard econometric theory, we decompose our time series of CPI Xt = Tt + St + t, t = 1,..T, with (2.1) The method first used by Buys-Ballot to see seasonality in astronomy assumes that our time series (Xt)t=1,..T, (here the logarithm of CPI), has an additive decomposition schema trend + seasonality + noise. The two first components have also parametric forms. For an annual seasonality, the model is written as follows:

136 - The

Journal of financial transformation

Tt = 0 +

j=1

jTtj =

jtj :
j=0

Tiao (1976)]. The non-parametric estimation of CPI seasonality is done according to the following steps: (2.2)
Determine the trend Tt by taking the moving average

the trend modeled by a polynomial of degree p,

12 12 i=1

St =

B(mi)Sti = B(mi)1{t mod i=0} :

i=1

over a period of thirteen data: Tt = M(Xt) with the filtering moving average given by:
6

the annual seasonality seen as 12 average bumps, (2.3)

(t)t=1,..T is a white noise sequence.

(3.1)

M= The parameters (aj)j=1,..p and (B(mi))i=1,..12 and are estimated by the OLSs Best Linear Unbiased Estimators. The seasonal average bumps (B(mi))i=1,..12 constitute successively a regular cycle and they compensate themselves

i= 6

i Li

(3.2)

where i = 1/6 for i = -5,..5 and the first and the last terms are equal to 1/12.
Estimate the seasonal component on the residuals with

12 i=1

B(mi) = 0.

St = (1 M)M(Xt Tt) with

2

(3.3)

This approach presents four advantages. Firstly, being parametric, it easily provides estimations error for forecast. Secondly, it can detect not only yearly seasonality but also monthly pattern. Thirdly, it spreads the error estimation not only on the seasonality but also on the trend estimation. Finally, it uses the whole set of data computing in one go on both the trend and the seasonality pattern.

M =

i= 2

i Li

(3.4)

with i = 1/9 for i { 2,2}, i = 2/9 for i { 1,1} and i = 1/3 else.
Look that applying the moving average M =

These advantages are offset by the strong assumption on the parametric form.

wastes the m1 data and the m2 last ones.

m2 i=m1

i Li

Please note that we could have used a different filtering periWhen we undertake an estimation of the seasonality and look at the trend component of the European and U.S. CPI, we find that there is less irregularity in the U.S. seasonality pattern than in the European one. This is consistent with empirical studies that confirm stronger seasonality of U.S. CPI data compared to European ones. This estimation is done with a polynomial estimation of order 5. When we compare the seasonality component extracted from the European and the U.S. CPI data we find that the crude U.S. seasonal component is more regular than the European one, with the amplitude being less variable. The difference in regularity between the European and U.S. inflation is even more striking than in the parametric method. When conducting a comparison between trend components extracted from the European and U.S. CPI data, we find that the two CPIs od and come up with the X8 or X12 or any Xn method which is also traditionally used in econometrics. But with regards to the noise of the estimation, Xn methods should conduct to similar quantitative results.

Non-parametric estimation of seasonality

Let us denote by Xt the CPI data and by L the standard lag operator. Standard econometric literature supports the use of a 12 data period in our filtering process, leading to a straight application of the X11 method as described in [Cleveland and

137

trend components are very close to the parametric results. To obtain the twelve monthly bumps (B(mi))i=1,..12 for the seasonality adjustment, we have to regularize the seasonality component to an annual cycle. For this, we can simply apply to seasonality component the parametric model (2.1) without trend. For both the European and U.S. CPI, the seasonality estimation is here less important in magnitude than in the parametric estimation. This is associated with the fact that the approximation replicates at best the yearly effect but smoothes the pattern.

Conclusion
Because of the tightening of quotes on inflation linked derivatives, models should incorporate seasonality for more accurate pricing. Using simple harmonic analysis tools, we can extract seasonality component of the inflation data and superimpose this effect to the bootstrapped inflation curve. Pricing shows that it can have a significant impact on seasonality sensitive trades.

References
Belgrade, N., 2004a, Market inflation Seasonality Management, CERMSEM Working paper 2004.51, April Belgrade, N., E. Benhamou, and E. Koehler., 2004b, A Market Model for Inflation, CERMSEM Working paper 2004.50 and SSRN Working paper Belgrade, N., and E. Benhamou, 2004, Reconciling year on year and zero coupon inflation swap: a market model approach, CDC Ixis Capital Market Research Note Bryan, M.F. and S. G. Cecchetti, 1995, The Seasonality of Consumer Prices, NBER Working Paper No. W5173, 1995. Buys-Ballot, C. 1847, Les changements priodiques de tempratures, Utrecht Shishkin, J., A. Young, and J. Musgrave, 1965, The X11 Variant of the Census Method X11 Seasonal Adjustment Program, technical paper 15, Bureau of Census Cleveland and Tiao, 1976, Decomposition of Seasonal Time Series: a model for the Census X11 program

Pricing impact on various inflation linked derivatives

To measure the influence of seasonality, we price with and without various products: Zero coupon and year-to-year inflation swap. Impact on more exotic structures such as options on real interest rates (and interest rates structured coupons floored with inflation-linked strikes) is similar but harder to analyze because of the increased complexity of these products. However, the same conclusions hold. First of all, accounting for seasonality imposes a periodic cycle on the nave price (price without seasonality). Impact varies according to the month of the fixing. For instance, the impact of seasonality on a 10 year IFRF inflation zero coupon swap fluctuates between -1.5bps to 2 bps. For a 10 year annual IFRF inflation year-to-year swap this difference goes from -2bps to 2bps. Secondly, we should not ignore seasonality on options. Obviously, this changes along the lines of the options characteristics. For example, on 1% IFRF inflation floor, seasonality adjusts the price by -6 to 9 bps.

138 - The

Journal of financial transformation

This is a revised version of the article published in International Payments, March 2004

Assets

Pricing default-free fixed rate mortgages: A primer1

Patric H. Hendershott
Professor, Aberdeen University

Robert Van Order

Lecturer, University of Pennsylvania

Abstract
It is by now generally recognized that a wide range of contracts can be viewed as contingent claims, which can be modeled like financial options. For instance, mortgages usually have prepayment and default options. Prepayment is a call option (i.e., an option to buy back or call the mortgage at par); whereas default can be viewed as a put option (an option to sell or put the house to the lender at a price equal to the value of the mortgage). The purpose of this paper is to provide a primer on the application of option pricing techniques to mortgages, focusing on the prepayment (call) option. The great insight of the option pricing/contingent claims models comes from determining equilibrium prices by imposing only the conditions of zero arbitrage profits and rational (wealthmaximizing) exercise of options. Such models give exact, rather than simply qualitative, predictions about prices, using relatively few parameters. In particular, they derive the central proposition, that the value of a security is the risk-adjusted expected present value of its cash flows, and they show how expected values and the risk-adjusted discount rate are determined. Our approach is heuristic; we do not provide proofs. Rather we try to provide intuition for the major results of this line of research.

This is adapted from Hendershott and Van Order (1987).

139

Pricing default-free fixed rate mortgages: A primer

It is by now generally recognized that a wide range of contracts can be viewed as contingent claims, which can be modeled like financial options. This approach is also applicable to mortgages [Hendershott and Van Order (1987), and Kau and Keenan (1995)]. For instance, mortgages usually have prepayment and default options. Prepayment is a call option (i.e., an option to buy back or call the mortgage at par); whereas default can be viewed as a put option (an option to sell or put the house to the lender at a price equal to the value of the mortgage). The application of formal stock and bond optionpricing methodology [Black and Scholes (1973), Brennan and Schwartz (1977), Cox, Ingersoll, and Ross (CI&R) (1985), and Merton (1973)] has been the centerpiece of much mortgage pricing research. The great insight of the option pricing/contingent claims models comes from determining equilibrium prices by imposing only the conditions of zero arbitrage profits and rational (wealth-maximizing) exercise of options. Such models give exact, rather than simply qualitative, predictions about prices, using relatively few parameters. In particular, they derive the central proposition, that the value of a security is the riskadjusted expected present value of its cash flows, and they show how expected values and the risk-adjusted discount rates are determined. When transaction and other costs are incorporated, pricing is more complex but still doable. Our focus is on frictionless models, that is, models without transaction costs and where borrowers exercise options rationally. The basic approach to pricing is the same for mortgages as it is to, say, corporate or Treasury bonds. Our intention is to provide some intuition and a unifying framework for analyzing long-term fixed-rate, pre-payable mortgages (FRMs), which comprise well over half of all U.S. mortgages. Finance theory has developed a large arsenal for pricing options, and the advent of trading mortgages in secondary markets has presented opportunities to apply these tools to an important part of the financial system. In this primer we focus on prepayment risk and ignore credit risk. For purposes
2

of analyzing mortgage pricing, this is not a bad abstraction, first because most (prime) mortgages3 that are traded have credit risk insured or guaranteed by a third party or they are structured in such a way that the bulk of what is traded has an AA or AAA rating, and second because default risk is generally much smaller than prepayment risk. On a typical, nationally diversified portfolio of conventional single family mortgages default rates (the fraction of loans that go through foreclosure in a year) typically range from 10 to 30 basis points per year. Prepayment speeds, on the other hand, run between 5% and 40% per year.4 Pricing depends on the contract structure of the mortgage and the way borrowers exercise their options. We assume that borrowers exercise their option in a way that maximizes their wealth. That means that at any moment in time when borrowers are contemplating whether to prepay or simply make their scheduled payment they will ask which of these alternatives leads to greater wealth. The wealth maximization strategy is also the strategy that minimizes the value of the mortgage. Our approach is heuristic; we do not provide proofs. We begin with a simple warm up model that assumes, rather than derives, that price is expected present value. This model captures most of the essentials of callable mortgages. We then turn to a more rigorous continuous time model and then to some extensions and modifications.

A ward up model
Until rather recently (less than 25 years ago) the dominant approach to pricing mortgages might be characterized as certainty equivalence. Models of this type forecast expected cash flows from the mortgage, including scheduled monthly payments and prepayments, and then treat this forecast as if it were certain. The value of the mortgage is the present value of the projected mortgage cash flows, discounted at a rate that adjusts for the risk that the scenario will not be right. For example, one might use the Public Securities Association (PSA) prepayment model, which assumes a prepayment rate that begins at 0.2 percent per year and rises

140

2 Hendershott and Van Order (1987) survey option approaches to default modeling. Kau et al. (1995) analyze the complications involved in the jointness of the options because exercising one of the options means giving up the right to exercise the other. 3 Prime refers to standard high quality mortgages (often defined by credit score of the borrower). A recent development has been the securitization of subprime

mortgages; this is still a rather small (probably 10%) part of the market. 4 In terms of value, default costs for conventional FRMs with 20% down payments of the type bought by Fannie Mae and Freddie Mac are on the order of 5 basis points per year; whereas the premium for the right to prepay is more like 50 basis points. For riskier, such as subprime or low down payment mortgages the two risks are comparable in value.

Pricing default-free fixed rate mortgages: A primer

gradually over 30 months to 6 percent or some multiple of it and remains at that rate. This is not a very good approach. It leaves the choice of risk adjustment unclear, and for long-term securities relatively small errors in discount rate can lead to large errors in price. More importantly, prepayments change with interest rates, so that, assuming anything like historic interest rate volatility, no benchmark is likely to be accurate for very long. This might not be a major problem if the gains from over-predicting prepayments balanced the losses from under-predicting them, but because prepayment is an option at the discretion of the borrower, the gains and losses do not balance. When rates go up, fixed rate mortgages fall in value, but when rates go down, investors reap smaller gains because borrowers tend to refinance their mortgages at par. This means that the value of investments in mortgages or mortgage-related securities depends critically and asymmetrically on the course of interest rates. And no one is very good at forecasting interest rates.

the current market interest rate. For example, if we are holding a bond bearing a 10 percent coupon when the market interest rate is 8 percent, we would simply discount the future cash flows at 8 percent, and the bond would be worth more than par. At a current interest rate of 10 percent, the bonds value would be exactly par. If we plotted the value of the bond at different interest rates, the result would be a convex5 shaped curve, expressing the usual downward-sloping relationship between interest rates and bond prices. This is a property of all noncallable fixed-income securities, and it has an important implication: a drop in interest rates increases bond value by more than the same rise in interest rates lowers bond value, which leads us to the result that the expected value of the security is greater than its current value from the price curve a 10% coupon bond is worth more than par when the market coupon rate is 10%.6 This means that the model must be reworked to account explicitly for uncertainty even without investor risk aversion. For example, assume that the basic risk-free rate for the time

The option-based approach looks at interest rates and prepayment in a probabilistic sense and explicitly acknowledges the inability to predict interest rates. It specifies the probability distribution of interest rates and evaluates the effects of random interest rate changes on mortgage values. This method has two advantages, it requires predicting only the probability of rates changing, rather than the rates themselves, and it explicitly addresses the borrowers prepayment option and illustrates how the sheer volatility (and hence the unpredictability) of interest rates affects mortgage pricing. Applying the method to both callable and non-callable mortgages reveals some important differences in the shapes of their price/interest-rate relationships and in their reactions to volatility.

period relevant for repricing, such as a day, changes randomly, and will either go up or down by basis points, and draws are independent over time. The probability that it will rise is p and the probability that it will fall is 1-p. Although the p and can change over time for example, when monetary policy changes we can almost certainly estimate these changes more accurately than we can predict future interest rates, and indeed there are markets where these changes can be inferred from prices of options, such as on Treasury securities.7 Consider the simple case depicted in Figure 1. Initially, the rate is 10 percent. If equals one percentage point and p equals 0.5, the rate will either go up or down by 100 basis points, and each possibility is equally likely. Figure 1 shows, for these values of and p, how rates can change over the next three periods.8

Interest rate variation

If we expected interest rates to remain close to current levels, the certainty equivalence approach would be fine; we could price any default free bond by calculating the present value of future cash flows (perhaps adjusting for taxes), discounting at

Pricing non-callable bonds

How do we price a non-callable bond with this model? Consider a bond that pays a U.S.$10 coupon at the end of the

5 The value of an n year pure discount bond is 1/(1+r)n, where r is the n year discount rate, which is convex in r (the second derivative with respect to r is positive). 6 This is nothing more than Jensens Inequality.

7 See Barter and Rendleman ((1979) for an early application of binomial rate models in a more rigorous, arbitrage free contest. 8 This process is not entirely realistic because it implies that interest rates can be negative.

141

Pricing default-free fixed rate mortgages: A primer

13 12 96.63

97.35

sibilities. This is: [.5($97.35) + .5($99.10) + $10]/1.12 = $96.63 The first term in the numerator is the value of the bond if the rate rises to 13 percent times the probability of the rise. The

11

11

97.59

99.10

second term is the value if the rate falls to 11 percent times the probability of rates falling. The third term is the U.S.$10 coupon, which is received in either case. The sum is discounted at the

10

10

100.07

100.01

102.56

100.92

prevailing 12 percent interest rate. Because this rate exceeds the 10 percent coupon rate, the bonds value is below par. We can use this same procedure to evaluate the bond for the

8 7 End of period: 1 2 3 End of period: 1

103.57 102.80 2 3

other two rate possibilities (8 and 10 percent) at the end of year two. Now we have three possible bond prices at the end of the second year, shown in the next-to-last column of Figure 2. Working backward, we can use these values to obtain two prices at the end of the first year (U.S.$$97.59 and U.S.$102.56), one for each possible rate (11 and 9 percent). Finally, we can solve backward to find the initial price of the bond, which is U.S.$100.07; i.e., it is worth more than the face value. It can be shown that for the bond to be worth par (100) at origination the coupon would have to be 9.98.10 This corresponds to the convexity of the bond curve; interest rate decreases have a bigger effect, in absolute value, on value than do interest rate increases, which increases the value of the bond. This is perhaps a surprising result, but it is only part of the story. Recall that we assumed that investors were risk neutral. In fact, they seem to be averse to the risk of capital losses on bonds. In terms of our model, they may behave as if they raise p (the probability of rates going up) in their calculations to compensate for the risk associated with higher volatility. If p is a positive function of volatility, then the net effect of volatility on bond price could be negative the coupon rate needed to get the bond to sell at par would exceed 10 percent.11

Figure 1: Distribution of interest rates over time

Figure 2: Pricing non-callable bonds

first three years and the coupon plus U.S.$100 in principal (its par value) at the end of the fourth year. At the end of the third year, a one-period bond paying U.S.$110 remains. At this point there are four possible one-period rates: 7, 9, 11, and 13 percent. Once the rate in the last period is known there is no further uncertainty, so the last payment can be priced easily. If rates have risen to 13 percent, the bond will be worth U.S.$110/1.13 or U.S.$97.35. If rates have fallen to 7 percent, the bond will be worth U.S.$110/1.07 or U.S.$102.80. The last column in Figure 2 shows all possible bond prices with one year to maturity. With two years to maturity, the problem is more complicated because of uncertainty. Suppose the one-year interest rate has risen to 12 percent (the top branch of the tree in Figure 1) at the end of two years. The rate will either rise to 13 percent in the next year, in which case the bond will be worth U.S.$97.35, or it will fall to 11 percent, in which case the bond will be worth U.S.$99.10. There is equal chance that either will happen. In both cases, we get the U.S.$10 coupon payment. Suppose traders in the market are risk neutral, meaning indifferent between a fair bet on rates and a sure thing.9 In that case, the value of the bond at the end of the second year, given a 12 percent rate, is the expected present value of the two pos-

Pricing callable fixed-rate mortgages

For callable mortgages, the shapes of the relationships among prices, interest rates, and volatility look quite different from those for non-callable mortgages. The reason is that the option to prepay (the call option) induces negative convexity

142

9 More generally, the discount rate would reflect the markets attitude toward risk. In the next section, we use an arbitrage free assumption to handle both the risk adjustment and the expected present value approach to pricing. Here we simply assume away risk aversion and postulate that value is expected present value.

10 The backward solution techniques are entirely analogous to Bellman equations in intertemporal programming [Sargent (1987)]. 11 Indeed, it probably will be negative. For instance, from Figure 4 it can be seen that the effect of high volatility on price in a risk neutral world is rather small.

Pricing default-free fixed rate mortgages: A primer

97.35 96.63

Suppose rates have risen from 10 percent at origination to 13 percent (highest rate in the last column of Figure 1). The borrower can make the final payment, which in present value terms costs him U.S.$97.35, or prepay it, which means buying it

97.40

99.10

back for U.S.$100. Clearly, being a wealth-maximizer, he will not want to pay U.S.$100 for something worth only U.S.$97.35, so he will not prepay. But suppose rates have fallen to 7 percent.

98.82

99.59

100

100

The present value of carrying the loan another period is U.S.$102.80, but he can buy back the loan for U.S.$100. Clearly the borrower increases wealth by paying off the loan and refinancing, with a one year loan at the new market rate. Indeed,

100 100 End of period: 1 2 3 Implies prepayment

he should prepay at any time the value of the remaining payments is greater than par. At the end of period three, he should prepay when rates have fallen to either 7 percent or 9 percent. Figure 3, which traces out callable mortgage prices just as

Figure 3: Pricing callable non-amortizing mortgages

for callable debt. Over some range of interest rates the shape of the curve is reversed, so that rate increases lower price by more than rate decreases raise them, and thus volatility lowers value. Negative convexity is the central difference between callable mortgages (and all callable debt) and non-callable bonds. To illustrate this, we compare the bond analyzed above with a non-amortizing mortgage (i.e. a callable bond). The instruments are the same in every respect except that the mortgage borrower has the option to prepay the loan at par (or unpaid balance), always U.S.$100 in this case of no amortization, and take out a new mortgage at the new market rate without penalty or other cost. Assume that the borrower is not planning to sell the house before the mortgage is paid off (three years in the example), or if the house is sold the new owner can costlessly assume the mortgage. Under these conditions the borrowers prepayment decisions are based solely on financial concerns, which require maximizing borrower wealth by minimizing the value of the mortgage. When should the borrower prepay the mortgage? As with the non-callable bond, we can begin by looking at the borrowers choice with just one period left.

Figure 2 traced out non-callable bond prices, shows the effect of these prepayments at par by recording U.S.$100 in the lower two boxes in the last column. This means that if interest rates have fallen since origination, the mortgage holder should expect prepayment immediately, making his investment worth only U.S.$100 at the end of year 3. Using the same procedure as in the bond case, we now go back one year. Suppose the interest rate is 10 percent at the end of year two. If the mortgage is not prepaid it will be worth either U.S.$99.10 or U.S.$100 (and called) next year. The expected present value of these two possibilities plus the coupon is U.S.$99.59. The possibility of call the next period causes the mortgage to have a lower value than the non-callable bond at the same interest rate. Solving backward, as in the bond case, gives a mortgage value at origination of U.S.$98.82, which is U.S.$1.25 less than the value of the non-callable bond. This U.S.$1.25 represents the markets implied value of the call option. The difference between 100 and 98.82, 1.18, is often referred to as the points charged upfront on the mortgage. We could also calculate the coupon rate given by the discount rate that makes the present value of the promised payments equal to U.S.$100. That is, we discount at (1+coup) and search

143

Pricing default-free fixed rate mortgages: A primer

over coupon values until the initial value is U.S.$100. This coupon is 10.28 percent versus 9.98 percent for the noncallable bond. The 30 basis point spread between the two yields is the amount by which the call option increases the mortgage rate on a mortgage originated at par.12 Alternately, 10.28 is the rate on a loan with no points. Note that the decision-making by the borrower is not complicated even if the pricing calculations are. The borrower need only know the market value of his loan and does not need to make complicated present value calculations. In particular, the borrower need only know if the coupon on a mortgage with zero points (and a term equal to the term remaining on his loan) has fallen below the coupon on his loan, which means that the market value has fallen below the outstanding unpaid balance.
13

Volatility lowers callable mortgage value, especially when

the option is close to par; it increases value for the noncallable mortgage. For high rates the callable mortgage looks like a non-callable mortgage.

Suboptimal prepayment
The model depicted above focuses on purely financial motives for prepayment, but borrowers also prepay for other reasons. For example, borrowers typically prepay when they move because most mortgages are due-on-sale (cannot be assumed by the new owner without the lenders permission). In principle, we can handle this by extending the model to include the probability of moving. Suppose there is a 10 percent chance each period that the borrower will move. We can price the mortgage by including this possibility in our expected present value calculations. If mortgages are not assumable, the possibility of moving always makes them more valuable because it forces some prepayments when rates are high and borrowers would like to keep their mortgages. In contrast, if the mortgage can be assumed without cost, moving and refinancing will again be entirely separate decisions. If the mortgage is valuable to the current borrower (has a market value below par) the mortgage will be assumed by the new owner; if not, it will be prepaid in any case, so the analysis of the previous section still applies.
Initial rate 8 9 10 11 12 50 106.7 103.3 100.1 96.9 93.9 Volatility 100 150 106.7 106.8 103.4 103.4 100.1 100.1 97.0 97.0 94.0 94.0

Determinants of the value of the callable mortgage

We can redo both the non-callable and callable mortgage valuations for different coupon rates and volatilities (). Figure 4 recalculates for the non-callable mortgage and Figure 5 for the same mortgage, but with the call option. Basic results are:
The callable mortgage is always worth less than the non-

callable; that is, the call option is always worth something. But for high interest rates relative to the 10 percent coupon, borrowers are unlikely to prepay (the call option is out of the money) and callable mortgage values are close to non-callable values.
There is a basic asymmetry that callable value does not go
200 106.9 103.5 100.3 97.1 94.2

above 100 when rates fall but can drop sharply if rates rise. This, again, is negative convexity; the shape contrasts sharply with the noncallable curve, which becomes steeper as rates fall. Negative convexity is also called price compression. Both expressions refer to the same property: as rates fall there is an upper limit on mortgage price increases. Negative convexity reflects the market anticipating the possibility of exercising the prepayment option before it actually occurs.

Figure 4: Prices of non-callable mortgages for various initial rates and volatility

Initial rate 8 9 10 11 12 50 100 100 99.4 96.9 93.9 100 100 100 98.8 96.6 93.9

Volatility 150 100 100 98.3 96.2 93.7

200 100 99.5 97.7 95.9 93.4

Figure 5: Prices of callable mortgages for various initial rates and volatilities

144

12 This is not quite correct. When the original three-year mortgage is called, we are effectively replacing it with a mortgage with the remaining term of the original loan but assuming that the call premium on this mortgage is the same as that of a three-year mortgage. In fact, it must be less given the shorter life. Alternatively, we could just assume that the investor invests the called mortgage in a noncallable bond, paying the certain 10 percent.

13 Of course, markets are not all that complete and there may not actually be quotes on a 23.5 year mortgage. For relatively new mortgages with close to 30 years remaining, quotes on new 30 year mortgages will be pretty close.

Pricing default-free fixed rate mortgages: A primer

In reality, most mortgages14 are not assumable, and people might not prepay as ruthlessly as in the simple model because of transaction costs and/or general inertia.15 The model can be modified to conform more closely to real behavior by modifying the option approach. For example, we might assume (or better still have empirical results that imply) that only half the borrowers prepay when it is optimal and that 10 percent of the borrowers prepay when the mortgage value is below par (interest rates have risen) because they are moving and must give up the low rate loan. This is a modified, option-based approach, and it is a simple version of the more behavioral approach used by most Wall Street firms that trade mortgages. Models like this incorporate major extensions to take account of factors like household mobility, unemployment, divorce, etc. Such models are more realistic but are subject to the very real problem that their parameters cannot be expected to remain constant. For instance, as costs of refinancing have decreased over the past decade the probability of prepayment conditional on being in the money has increased.

The basic equilibrium condition for any fixed income security is that the expected yield on a security should equal the riskfree rate plus an appropriate risk premium, which also implies that price equals the expected present value of cash flows discounted at a risk-adjusted interest rate. These two notions come from the condition that pricing be arbitrage free.16 Contingent claims models derive the conditions [Cox, Ingersol and Ross (1985)], but because this notion of equilibrium is so straightforward, we could start by making it an initial assumption, as was done in the warmup model. The equilibrium condition for a default-free, fully-assumable FRM turns out to be a second-order partial differential equation in R and t. This sort of equation will apply to any claim that is contingent on those state variables. Hence, many functions satisfy the equation. To determine the one function that applies to the mortgage being priced, we incorporate conditions specifying the details of the contract, such as the coupon rate, the term of the mortgage, and the value of R at which the mortgage will be called (usually determined via an optimal call strategy). The model determines mortgage price, not yield. Yield is usually measured as the internal rate of return computed for a given assumption about prepayment (i.e. prepayment in 12 years or the PSA model discussed above). Because prepayment can vary greatly, depending on the mortgage coupon, and the expected volatility and drift of interest rates, conventional yield calculations can be very misleading. The models can be used to find the coupon rate that is consistent with some price, say par, but that is not the same as determining the expected yield. Here we discuss a simple frictionless continuous-time model of mortgage pricing, which draws heavily on Dunn and McConnell (1981), and Brennan and Schwartz (1985). We begin by specifying the state variables and the arbitrage condition and then derive the pricing equation. Finally, we add extensions.

Continuous time pricing models

The warm up model captures all of the important elements of pricing callable debt, but it needs to be extended. While the assumption of trading at discrete intervals with a simple binomial process is sometimes a good approximation (as the number of steps increases the binomial distribution approaches a normal distribution), it is not general enough. We shall develop models that take place in continuous time with broader probability distributions. Moreover, because investors appear to be risk averse, we need to allow explicitly for risk aversion. The formal option pricing methodology requires some technical apparatus, but that is not our major concern. Rather, we continue to focus on intuitive interpretations to underscore the economic logic underlying the analysis. In general a pricing model seeks to find and evaluate a function M(R,t), where R is a vector of interest rates and t is time from origination, that explains observed prices of mortgages.

14 FHA insured mortgages are generally assumable. 15 See for instance, Deng et al (2000) for a model with sluggish prepayment behavior and heterogeneous borrowers. 16 Discrete time models can also be solved from arbitrage free conditions.

145

Pricing default-free fixed rate mortgages: A primer

Equilibrium conditions
Because a mortgage can be outstanding for up to 30 years, all interest rates up to 30 years are potential state variables, which is to say M(R,t) could depend on a large number of variables. The problem can only be managed if a small number of basic interest rates determine the other rates. Like most authors, Cox, Ingersoll, and Ross (1985) being the first, we begin by assuming that all interest rates are driven by a single exogenous rate, the instantaneous short rate, r, which determines the entire yield curve. Changes in this rate are taken to follow an Ito process [Maliaris and Brock (1981), Merton (1995)], the evolution of which is governed by the following stochastic differential equation: dr = u(r, t)dt + (r, t)dz (1), where u(r, t)dt is the expected drift in r over a small interval of time of length dt, (r, t)dz is a disturbance made up of dz, which is normally distributed with zero mean and unit variance, and (r, t), which is the volatility of dr. Equation (1) is a continuous-time version of a standard difference equation. Cox, Ingersol and Ross (1985) and others have shown how (1) can be used to determine the entire Treasury yield curve. We assume that markets are complete enough that traders can construct portfolios of traded assets that over a short period of time mimic the cash flows of mortgages. Then mortgage prices come from the assumption of no arbitrage profits. In particular, under rather general circumstances a portfolio of default free bonds, such as Treasuries, can be constructed such that the combination of the mortgage and a short position in the Treasury portfolio (the hedge portfolio) absorbs zero wealth and has zero short-term (with continuous time instantaneous) risk. Brennan and Schwartz (1977, 1985), among others, show how this portfolio is derived. Absence of arbitrage profits implies a zero instantaneous return. From this zero return, the basic equilibrium condition is deduced: the instantaneous expected yield on the mortgage must equal the risk-free short rate plus a risk factor. In the case of one state variable, umM (r, t) = rM(r, t) + (r, t)(r,

t)Mr (2), or um = [r + (r, t)(r, t)Mr]/M(r, t) (2), where um is the expected instantaneous rate of return on the mortgage, r the instantaneous risk-free rate, the market price of risk, (r, t) the volatility of short rate changes, and Mr the partial derivative of M with respect to r. Equation (2) says that in equilibrium the expected return equals the current rate plus a risk-adjustment term, where the risk-adjustment term becomes the product of three terms: the price of the risk that r changes, , the amount of risk, , and the interest-rate sensitivity (alternatively the duration) of mortgage value, Mr/M(r, t). If more state variables exist, more risk-adjustment terms, each with its own risk price, come into being. The arbitrage model does not derive the s; their derivation is a general equilibrium problem that requires knowledge of such market forces as traders risk aversion. The model does imply that the s are objective prices, which are the same for all traders and can be inferred from market prices. Thus, the s may be viewed as competitive market prices for risk.

Pricing equations
The next step is the derivation of um, the expected rate of return on the mortgage. This is a technical step that requires some knowledge of Stochastic Calculus. Use is made of Itos lemma, which is the stochastic analogue of the chain rule of ordinary calculus. The result, which we simply assert, has a fairly straightforward interpretation. The expected instantaneous return consists of the coupon rate and expected percentage capital gains. The coupon rate is simply the coupon payment, C, divided by M. Expected percentage capital gains come from two sources. The first can be called certainty equivalent gains, which occur if variables change as expected. These are given by Mt/M (amortization and capital gain from selling at a discount) and u(r, t)Mt/M (expected change in r times the interest sensitivity (duration) of value with respect to interest rates). The second source flows from the stochastic nature of r and is the vehicle that introduces the negative convexity discussed above. Because M is, in general, not linear in r random

146 - The

Journal of financial transformation

Pricing default-free fixed rate mortgages: A primer

increases in r will not have the same effect on M, in absolute value, as random decreases. Thus, as above, the certainty equivalent approach of assuming that r changes exactly by u(r,t) will not fully reflect expected capital gains. Accounting for this extra capital gain requires using Itos lemma. Here we simply assert that expected capital gains from the dispersion of r are given by 1/22Mrr/M, that is, they depend on the volatility of r and the shape of M, disappearing if M is linear or r is nonstochastic.17 Adding the returns from coupons and capital gains, we have umM (r, t) = C + Mt + u(r, t)Mr + 1/22(r, t)Mrr (3). Equating (1) and (3), C + Mt + [u(r, t) (r, t)(r, t)]Mr + 1/22(r, t)Mrr= rM (4) This second-order partial differential equation is the basic equilibrium condition for the one state variable model. An infinite number of functions of r and t satisfy this condition (an infinite combination of coupon and capital gains streams provide a normal or equilibrium return). To obtain an expression for a specific debt instrument, we must specify the terms of the instrument. Mathematically we need three boundary conditions, one for t and two for r (equation 4 is second order in r). The t boundary is the terminal condition that comes from the amortization schedule of the mortgage. For a fully amortizing mortgage, M(r, T) = 0 (5), where T is the time at which the last payment is made.18 The other two conditions relate to how M is valued when r takes on extreme values. The first of these conditions incorporates the economic intuition that M becomes worthless as r approaches infinity. M(, t) = 0 (6). The final condition specifies the interest rate at which the mortgage is called, rc. The assumption is that the call option will be exercised in a way that maximizes borrower wealth. But before turning to that, we consider, as we did above, the pricing of a benchmark security, a non-callable mortgage, M*, which is equivalent to a portfolio of Treasury securities with constant payout for T years. This is easy to price because the value of M* is just the present value of the known cash flows

discounted at the appropriate rates read off the yield curve determined by (1). Hence M*(r, t) has the usual downward sloping concave shape of a fixed-income security [Brennan and Schwartz (1977) and Cox, Ingersol and Ross (1985)]. The curve for the callable mortgage, M(r,t), lies below M*(r, t) by an amount equal to the value of the call option. Because the mortgage can be called when M equals PAR, we know that points in the region above the PAR line cannot be points on M; the borrower would maximize wealth by prepaying if that happened. Rational borrowers (ignoring transaction costs) will choose the call strategy that minimizes the value of M, which maximizes wealth. Of all the M functions satisfying (4)(6), the rational borrower chooses the function that has the smallest value subject to touching the PAR line (the option will never be exercised at less than par). The curve that does this (assuming an interior solution) must be tangent to the PAR line, and the level of r at which it touches is the optimal call rate for a given t. Hence, the final boundary condition is Mr(rc, t) = 0 at M = PAR (7), which gives the minimum M(r,t).19 Note that if (7) represents the minimum value of M(r,t) subject to touching PAR, then M(r,t) must be concave from below (Mrr < 0) in the neighborhood of the exercise rate, rc, reflecting the price of the security anticipating the call option even when r is not especially close to rc. This concavity (negative convexity) comes from the second order condition for optimal exercise. From (3) negative convexity implies that volatility produces a lower return on mortgages due to capital losses on average for mortgages close to being called. Hence, volatility makes callable mortgages less valuable. As was the case in the binomial example above, the borrower does not have to go through the optimization problem described here and does not need to be acquainted with Itos Lemma for the model to work. A strategy of prepaying whenever the coupon on a new par (no points) mortgage with the same remaining term is less than the coupon on the loan will also minimize mortgage value. Because this new coupon rate is what is to be determined, outside pricers do have to solve

17 Note that if Mr > 0, then the premium is positive, and vice versa. The Mrr < 0 corresponds to negative convexity, which lowers value. A mortgage can take on both positive and negative values of Mrr as r varies (the options goes in or out of the money).

18 In the non-amortizing case in the warm up model the condition is that M(T) = PAR. 19 Not surprisingly this implies that the duration is zero near the exercise point; then the mortgage will be priced so that it earns the risk-free rate.

147

Pricing default-free fixed rate mortgages: A primer

the partial differential equation, so as to make M(r,t) a function of exogenous variables (like the yield curve), but the borrower need only know current market rates. The solution has an expected-present-value interpretation, but here it is derived rather than assumed, as was the case in the warm up model. In particular, M is the expected present value of future cash flows, discounted at r, with the expected value of dr given by u rather than just u [Cox, Ingersol and Ross (1985), Lemma 4].20 Risk aversion is allowed, and it is, in effect, a market price that is shared by all fixed income securities. Finally, optimal exercise of the option is derived, coming from wealth maximization at every decision point, and it leads to an intuitively appealing tangency condition.

condition (4), as discussed in the previous section. The expected return on the mortgage now equals the risk-free rate plus adjustments for the risks of the short rate or the long rate changing. This leads to a generalization of (4). If we let c be the long-term (consol) rate, 12 and 22 be the variances of changes in r and c, be their correlation coefficient, u1 and u2 be their means and 1 and 2 be their risk prices, then equilibrium is given by 1/2Mrr12 + Mrc12 + 1/2Mcc22 + Mr (u1 11) + Mc (u2 22) + Mt + C = rM (8), which is similar in spirit to (7) except now there is a more complicated convexity term and two risk prices instead of one. What is especially interesting in this case is that the values of

In general there is not a closed form solution for M(r.t). Solutions come from techniques that involve carving the rt space into a grid, converting (partial) differential equations into difference equations. It is necessary, as in the warm up model, to solve the model backwards from the last period back to the first, because at any decision point except at the end of the term it is necessary to know future options in order to make decisions [Kau and Keenan (1995)].

2 and u2 can be inferred. Because (8) applies to the value of a consol, by substituting the value, 1/c, of a consol paying U.S.$1 into (8) and evaluating the derivatives [Mc = -1/c2, Mcc = 2/c3, Mr = 0] we can solve for 22, and, by substituting the result back into (8) produce an expression that contains neither 2 nor U2 [Brennan and Schwartz (1985)]. This insight comes directly from the arbitrage approach and is directly analogous to the result in Black and Scholes (1973) that we do not need to know the mean-reverting value of a stock or its risk price to price an option on the stock. In the one-state model, we could not eliminate u or because the instantaneous security is not a traded asset (it matures too quickly). As a result it does not have a price that we can plug back into (8).21 Another extension is to allow sub-optimal prepayments because households do not always exercise their prepayment option as ruthlessly as the model we have elaborated implies and/or there are transaction costs, observable and unobservable, that make exercise more complicated than is depicted. Many researchers have developed ad hoc prepayment functions, fancier versions of the one in the warmup model, that allow prepayments for reasons other than hitting the boundary condition. These models are typically option-based in the sense that they use the same sorts of variables that the rigorous, frictionless models used, but they allow for a wider

Extensions
A logical extension of the model is to increase the number of interest rate variables. Taken literally, the one-rate model above implies a constant rate toward which the short rate reverts for all time. Given the obvious importance of changing inflation on interest rates, this seems like a difficult model to take seriously (although this may not be of much empirical significance). The nominal rate could be defined as the sum of the real rate and the expected inflation rate, and these components could be viewed as being governed by separate processes, requiring two state variables [Richard, (1978)]. Alternately, and equivalently, in their two-state variable model Brennan and Schwartz (1985) assume different mean-reverting processes for the short rate and the rate on a long-term consol bond. Adding a state variable changes the equilibrium

148

20 Indeed, from (4) there is a choice in use of the risk adjustment term. You can, as above, bias the mean change in rates by deducting from it and discount at the risk free rate, or you can project at the true mean and discount at the risk-adjusted rate r + Mr/M. The former is often easier because r can be inferred from the Treasury yield curve.

21 Note though that we do not need to know u or separately. Only u matters and that can be inferred from market prices. We do need to know .

Pricing default-free fixed rate mortgages: A primer

range of response elasticities. In the context of continuous time models, Dunn and McConnell (1981) and Brennan and Schwartz (1985) added random prepayments, which they modeled as Poisson processes, to rational prepayments, given by the boundary condition (7). If is the probability of a random prepayment, then the expected cash flow (C in (4)) is increased by (M-PAR). Boundary conditions are as before. While much empirical work on default and prepayment has been proprietary, several studies have been published. Foster and Van Order (1985) estimate models of default and prepayment using aggregate data, which allows them to use ordinary least regression techniques to obtain estimates, and Deng et al. (2000) use a proportional hazard model to estimate both prepayment and default models that allow for unobserved heterogeneity among borrowers. The basic structure of both models is option-based, using the extent to which the options are in the money as key explanatory variables. An advantage of the more behavioral models, like Deng et al., which do not explicitly model optimal exercise, is that they are easier to use for pricing because they do not require backward solving techniques to calculate model prices. Backward-solving models are often costly to solve, especially if there are many state variables.22 With models that are not explicitly optimizing simulation or Monte-Carlo techniques can be used to estimate expected present value. This allows much greater flexibility in modeling and appears to be the main tool used by traders.

In all cases, however, the basic structure of the model and the ultimate solutions are the same:
The value of the mortgage is the expected present value

of the cash flows, discounted at a risk-adjusted rate that is revealed by market prices.
The cash flows come from optimal exercise, which comes

from wealth maximization and in general involves tangency conditions.

There is in general no closed form solution. Solution

requires converting into discrete values and backwardsolving.

Practical modelers and traders use more complicated but

less elegant behavioral models. The models are optionbased but not explicitly value-maximizing, and Monte Carlo techniques are typically applied to obtain solutions.

References
Barter, R., and R. Rendleman, 1979, Fee based pricing of fixed rate loan commitments, Financial Management, Spring, 13-20 Black, F., and M. Scholes, 1973, The pricing of options and corporate liabilities, Journal of Political Economy, 81, 637-59 Brennan, M., and E. Schwartz, 1977, Savings bonds, retractable bonds and callable Bonds, Journal of Financial Economics, 5, 67-88 Brennan, M., and E. Schwartz, 1985, Determinants of GNMA mortgage prices, Journal of AREUEA, 13:1, 209-228 Cox, J., J. Ingersoll, and S. Ross, 1985, A theory of the term structure of interest rates, Econometrica, 53, 302-407 Deng, Y., J. Quigley, and R. Van Order, 2000, Mortgage terminations, heterogeneity and the exercise of mortgage options, Econometrica, 68:2, 275-307 Dunn, K. and J. McConnell, 1981, Valuation of GNMA mortgage-backed securities, Journal of Finance, 36, 613-623 Foster, C. and R. Van Order, 1985, FHA terminations: A prelude to rational mortgage pricing, Journal of AREUEA, 13:1, 273-291 Hendershott, P., and R. Van Order, 1987, Pricing mortgages: An interpretation of models and results, Journal of Financial Services Research, 1:1, 77-111 Kau, J., and D. Keenan, 1995, An overview of option-theoretic pricing of mortgages, The Journal of Housing, 6:2, 217-244 Malliaris, A., and W. Brock, 1981, Stochastic methods in economics and finance, North Holland Merton, R., 1973, The theory of rational option pricing, Bell Journal of Economics, 4, 141-183 Merton, R., 1995, Continuous time finance, Blackwell, Cambridge MA Richard, S., 1978, An arbitrage model of the term structure of interest rates, Journal of Financial Economics, 6:1, 33-57 Sargent, T., 1987, Dynamic macroeconomic theory, Harvard University Press, Cambridge MA

Conclusion
This paper describes the option-based approach to pricing fixed rate, prepayable mortgages using techniques used to price callable bonds. It is relatively straightforward to extend the model to analyze default in the same option framework [Hendershott and Van Order (1987) and Kau and Keenan (1995)] with default as a put option, selling the house back to the lender at a price equal to the value of the mortgage. More complicated is analysis of the two options together [Kau and Keenan (1995)]. In that case exercising one option means foregoing the right to exercise the other, which greatly complicates the analysis.

22 For instance, if there are lagged values of variables, such as in analyzing adjustable rate mortgages [Kau and Keenan (1995)] or delayed reaction, they need to be modeled as separate state variables in optimizing models.

149

150 - The

Journal of financial transformation

Assets

Efficient pricing of default risk: Different approaches for a single goal

Damiano Brigo
Head of Credit Models, Banca IMI

Massimo Morini
University of Milan Bicocca

Abstract
With the rapid development of the credit derivatives market, efficient pricing of default has become an extremely important issue for the credit risk management of banks and other investors. We consider here some of the opportunities and problems that the development of this market poses to quantitative research in academia and industry. We describe different modeling choices pointing out the practical pros and cons of the different frameworks. For all different frameworks, we present innovative solutions allowing both computational efficiency and high consistency with the increasingly liquid credit reference market, the market of credit default swaps.

151

Efficient pricing of default risk: Different approaches for a single goal

The last few years have been described by many as a period of downturn for the financial markets, or at least of increased worry and lack of confidence. Crises have struck investors in different markets, starting from the collapse of the LTCM hedge fund in 1998, to the burst of the Internet equity bubble in 2000, and to the debt crisis of both sovereigns (Russia 1998, Argentina 2001) and major industrial worldwide players (Enron 2001, WorldCom 2002). These events have pushed many investors out of different sectors of the financial market.

outstanding increased from U.S.$170 billion in 1997 to almost U.S.$1,400 billion in 2001. At the same time, the range of products is growing, in particular in portfolio credit derivatives and options on existing credit derivatives. Besides representing an important contribution to the stability of the financial market as a whole, the increasing liquidity of this market is opening up unmissable opportunities for credit risk management of banks and other investors. In the following section, we will explain the structure of the

However, a closer look at the situation reveals that the context is twofold. While a proportion of investors were actually pushed out by the recent crises, those who remained started to develop instruments to strengthen and defend themselves for the future. This process includes the development of a substantially new market, the credit derivatives market, which has been growing dramatically even when most other financial markets have been stagnating. Credit derivatives are mostly over-thecounter derivative securities whose final payout depends on the default of a reference entity. Default has several meanings, including the risk that one counterparty will not honor some of its obligations. While the types of structures that are being developed are quite complex and varied the common purpose of these products remains the same, to allow market participants to single out, transfer, and trade separately credit and default risk, namely the part of the risk in a contract which is related to the credit reliability of an obligor. In parallel, the development of this market has been fueled by the increased attention of regulatory agencies to exposures in OTC derivatives by many of the worlds major financial institutions. Regulations, such as Basel I and II, made the advantages of an efficient credit risk market even more apparent, in particular to major banks. The market for credit derivatives was created in parallel in Europe and in the U.S. during the 1990s, and it has recently experienced the highest growth rate of all derivatives markets. The vitality of this market is revealed both in terms of quantitative expansion and qualitative development and financial innovation. According to a Risk magazine survey in 2002, the notional value of credit derivatives

most common credit derivatives. Then we consider the relevance of the development of a liquid market for credit risk for reliable relative value pricing through quantitative modeling. We describe the general features of the major modeling frameworks, and present innovative solutions for exploiting the opportunities and reducing the problems of the different approaches.

Credit default swaps (CDS)

Since the market is experiencing a fast evolution, it is not yet possible to give a precise or definitive taxonomy of credit derivatives. However, an important feature of these instruments, pointed out for instance by Bielecki and Rutkowski (2001), is the precise extent of credit risk that a product allows to be transferred, namely the intrinsic definition of credit risk. We have securities that allow the entire risk associated with a transaction vulnerable to credit risk to be transferred, such as total rate of return swaps and equity return swaps. We also have products that transfer the risk related to changes in the value of an agreement due to movements of the credit quality (financial reliability) of one or more obligors, such as credit spreads swaps and options. Finally, there are products precisely focused on separating and transferring only the risk directly involved with a default event. This last category appears to be most attractive to market participants. Credit default swaps (CDS) and similar products, for example, represented over two thirds of the entire credit derivatives business in 2002. A credit default swap is a contract between two parties, called

152 - The

Journal of financial transformation

Efficient pricing of default risk: Different approaches for a single goal

the protection buyer and the protection seller, designed to transfer the financial loss that the protection buyer would suffer if a particular default event happened to a third party, called the reference entity. Usually, the reference entity is a debtor of the protection buyer. The protection buyer agrees to pay a periodic amount R (less frequently an upfront fee) to the seller in exchange for a single protection payment, made by the seller only in case the pre-specified feared default event happens. The CDS are quoted on the market though a fair R, often called CDS spread, making the current price of the contract equal to zero, like in interest rate swaps. In spite of all possible variations, the basic structure of the product is always the simple one described. This structure allows for a protection similar to an insurance contract, but a CDS is an autonomous security traded on a financial market like any other derivative. The relevance of this feature in a situation of increasing liquidity will be further pointed out in the next section, with particular attention to its consequences on quantitative valuation and risk management.

ance contracts and could not be traded separately from the reference obligation. This system did not favor efficiency, consistency, and competition in evaluating credit risk, and strongly limited the possibilities for credit risk management. Instead, the creation of an increasingly liquid market allowing investors to trade credit risk separately as any other tradable asset was a real major financial innovation, since it led to a strong increase in attention, competition, and precision in evaluating credit risk. From a quants perspective, it allowed for the extension of the advanced techniques for pricing and management that had been developed for different markets, such as equity, interestrate, and FX derivatives, to credit risk. The increasing liquidity of the CDS market, in particular, has played an important role in shaping credit modeling, making it more similar to those typical in other markets. The CDS, aimed at transferring only the risk directly involved with a default event, are the most interesting category also from a quants point of view. In fact, their evaluation poses in a clean way the problem of modeling and pricing the financial consequences of the central specific event, the default. Consequently, CDS quotations embed fundamental information on the market assessment of the risk neutral default probabilities for the reference obligor. Since, as is well known, pricing of contingent claims is correctly performed by expectation under risk neutral probabilities, this is actually what we are interested in for valuation of default-dependent payoffs. Obviously this information is not transparent and requires setting a precise modeling framework for being assessed quantitatively (although we will see that default probabilities in CDS appear quite robust to the choice of the model). The procedure for determining the parameters of a model in such a way that it gives answers consistent with known reference market prices is called calibration. When the model has been calibrated to the basic products, it can be used for pricing consistently more advanced default-risky payoffs.

CDS and other credit derivatives from a quants perspective

Traditional models for pricing and hedging contingent claims written on equity or on the term structure of interest rates do not include a specific consideration for credit risk. The increased attention of market participants and regulators to credit risk, and the growing amount of complex structured products for which default risk could not be overlooked, called for increased attention to credit risk also from quants in academia and industry. In particular, because of the contemporary development of the credit derivatives market, giving the possibility to transfer default risk through an OTC contract, the focus moved to the development of specific models for the valuation of these new contracts. In fact the agreements used for transferring credit risk needed to be correctly priced for making the transfer effective. Previous financial agreements which were used to protect investors from credit risk had features often similar to insur-

153

Efficient pricing of default risk: Different approaches for a single goal

However, before CDS reached a critical liquidity, the market price of credit risk expressed by them was not a reliable representation of the market expectations or risk premia. Therefore modeling concentrated on explaining default risk based on information external to its specific market. With the increase in liquidity and diffusion of the CDS, the precise information on the price of credit risk they provide, determined by a reliable level of demand and supply, becomes a fundamental benchmark for fair valuations. This moved the attention of quants to building models that are able to register efficiently, through calibration, the information provided by the reference CDS market. Such models are then reliable enough to be used for required applications. This last development set a context for relative value pricing similar to that of traditional derivatives markets, but many technical modeling problems posed by valuation of credit risk are peculiar. Most of them had been paid little attention to in the past, since possible applications were less necessary and relevant. The possible structural causes of credit risk, or on the other hand the sudden consequences of a default event, are both examples of specific features making this risk inherently different from that involved in modeling the underlying asset of more traditional derivatives. Another distinctive feature, not treated in this work, is the fundamental incongruence between the nature of default interdependency and the concept of interdependency comfortably used for interest rate and equity markets. Therefore, although many techniques and models previously used have turned out to be useful for this new application, the extension of the quantitative apparatus to this new market has called for a massive work on model innovation and development of new techniques. The alternative models for credit risk pricing do not differ simply due to technical details. The main modeling frameworks stem from different views on fundamental issues, such as the choice of the variables to model or the correct description of a default event. We briefly review some of these frameworks in the next three sections.

Structural models
The first important contribution to quantitative credit modeling dates back to the seminal paper of Black and Scholes (1973), which introduced the principles of modern option pricing. In this paper the valuation of company liabilities was indicated as a possible application of the pricing technique introduced (later known as the Black and Scholes formula), and the underlying idea was a structural explanation of default. Although the distinction is at times rather subtle, financial research labels as structural those models containing a stylized description of the economic causes for a market value or event; they are opposed to those models, at times called reduced form models, based on a quite general mathematical framework to be specified consistently with historical or crosssection data. In the Black and Scholes (1973) and then Merton (1974) default model, a company defaults if, at maturity T of its debts, the value of the company is lower than the reimbursement to be made. Based on this interpretation and by assuming a geometric Brownian Motion (GBM, also called lognormal diffusion) dynamics for the value of the company, all the mathematics required for computing default probabilities is that typical of plain vanilla equity option pricing. Black and Cox (1976) also remain in this framework for explaining default based on balance sheet notions, but increase realism by introducing the so-called first passage time models. Here the value of a company is again a GBM, and default still happens if the value of this process is lower than the level of its debts, but now this event is not limited to an unrealistic single maturity time. Here the debt level is a barrier Dt, possibly time varying, and the value of the company can fall below Dt at any time, causing default. For first time passage models the mathematics required was already in use for equity barrier options. The model was tractable (easy to make computations with) in case the underlying was assumed to have constant parameters, in this case the explicit probability distribution for the first time that a GBM hit a barrier is known. One may be unconvinced by the hypothesis on the firm value underlying these approaches, but the ultimate reason that led

154 - The

Journal of financial transformation

Efficient pricing of default risk: Different approaches for a single goal

to the development of alternative approaches was a practical one. Structural models appeared empirically unable to describe real market expectation of defaults. The gradual description of default implied by these structural approaches was not consistent with market expectation of default as a sudden event, not fully predictable by observing book values of a company. In particular structural models tend to imply unrealistically low default probability in the short-term.

credit spreads, but not for their volatility. For this one has to allow for continuous stochastic variation of intensity. In this case we move to Cox Processes. In this setting, conditional on the information on the path followed by intensity, default happens at first jump time of an inhomogeneous Poisson process with this time-varying intensity. Thus survival probability is P( > t) = E[P( > t | { (s): 0 s t})] = E[e(s)ds] This approach can exploit many results and well-known mod-

Later we will describe some possibilities to overcome these practical problems of structural models without leaving the structural representation of default. In the next section, instead, we see a radically different framework.

els typical of stochastic modeling of the short interest rate (instantaneous spot rate), already in use for interest rate derivatives. Notice in fact that the expressions for survival probabilities are analogous to the functions expressing bond prices in terms of the dynamics of the short rate. Advanced intensity models [Duffie and Singleton (1999)] are more flexible than structural models. However, depending on the dynamics chosen for the intensity, here too there can be differences in terms of tractability and calibration power. We will see later a model designed with specific attention to both aspects.

Intensity models
The direct way to take into account the sudden nature of the default event in the real word is modeling directly default as an unpredictable, exogenous event. This leads to reduced form models, also called intensity models, where default happens at the time of the first jump of a stochastic (jump) Poisson process with intensity . This means that the time of default is exponentially distributed with a probability of survival for t years given by P( > t) = et, with expected time to default 1/. In such models there is no attempt made to model the economic causes for default. Default is simply a random variable with a distribution parameter to be determined consistently with market evidence. Also in this framework many extensions have been put forward to increase realism and flexibility. A model with constant default intensity is neither realistic nor flexible enough to embed market information about default probabilities on different interims. As a solution, the intensity can be supposed to vary deterministically in time, also continuously (given some technical conditions). In this case, the default happens at the first jump of an inhomogeneous Poisson process with intensity (t), leading to a survival probability P( > t) = e(s)ds. The shape of this survival probability hints at the fundamental fact that the intensity can be interpreted as an instantaneous credit spread. This model can account for the term structure of

Market models
Intensity models, although developed for increasing consistency with the market, are based on modeling theoretical, non-observable quantities and lead to option formulas quite different from those traders are used to. Traders on the interest rates market are used to price options with the Black formula, developed originally for commodity futures. This formula is based on modeling the underlying as a lognormal process under the pricing measure. Without this structure, even the typical concept of implied volatility cannot be used consistently. To recover the features of classic option formulas in credit, the first step is modeling directly the underlying rates in credit derivatives payoffs. So default is not to be explicitly modeled, neither as a predictable nor as a structural event. In interest rate modeling, recent developments in probability made it possible to give a rigorous probabilistic foundation to the aforementioned market standard of pricing most liquid options via heuristically derived Black-like formulas. This led

155

Efficient pricing of default risk: Different approaches for a single goal

to the so-called market models, such as the Libor and the Swap Market Models. Generalizing the definition, in a market model we model directly market observables underlying common options, and they are assumed to follow the so-called standard market model, namely a lognormal diffusion under the reference pricing measure. This allows us to price options with Black-like formulas. Hull and White (2003) outline that a market model approach, which makes it possible to detect the implied volatilities in option quotations, can be useful also in a credit setting, in order to make option quotations more standard, and therefore transparent and understandable. This would help enhance liquidity in some growing markets for credit options, such as options on CDS. But there are some relevant issues to be considered, ranging from detecting the correct specification of the state variables helping to price real world derivatives, to technical issues in the definition of the reference probability measure.

enough flexibility to be calibrated to a term structure of CDS rates, and thus to the embedded default probability. Is it possible to enrich the calibration power of structural models while keeping tractability, namely consistent with the mathematics of barrier options? Thanks to recent advances in this last field, the answer is now affirmative. Lo et al (2003) show that, assuming a specific shape for the barrier, one can have explicit formulas also when the underlying is modeled as a GBM with time-varying parameters. For credit risk, this means that we can have a tractable structural model even if allowing for time-varying volatility of the underlying. Such a model and its calibration to CDS are described in Brigo and Tarenghi (2004). Assuming that the volatility of the firm value is piecewise constant (a step function), the model can be easily calibrated to CDS quotes using both volatility and the debt barrier. In Brigo and Tarenghi (2004) diagnostic tests are performed to verify that this is not only a formal consistency due to the many parameters, but corresponds to an increased capability to pick out relevant implied market structures. Implied default probabilities are computed and compared with those implied by a standard intensity model (the framework designed for this exact purpose). Surprisingly enough, the probabilities found are nearly the same. This evidence confirms two hypotheses. Firstly, that CDS information on default probabilities is robust. If the level of fitting power is kept equivalent, it is not highly dependant on modeling assumptions. Secondly, that the tractable structural model of Brigo and Tarenghi (2004) also matches, besides the CDS quotes, the whole structure of default probabilities as determined by intensity models. A recent development that is currently under investigation concerns the possibility to calibrate the CDS quotes via an uncertain debt barrier, using the barrier scenarios and probabilities as calibrating parameters and keeping the value of the firm volatility as an exogenous input from the equity market. Among other issues, the work of Brigo and Tarenghi (2004) suggests that practical problems of standard structural mod-

A tractable structural model consistent with market implied default probabilities

Typically credit models were specified according to credit spreads or balance sheets information. However credit spreads appear to be largely determined by causes different from default probabilities, for instance liquidity. On the other hand, some recent major defaults (Enron, Parmalat) suggest that default often happens in conjunction with particularly unreliable balance-sheets information, so that book values can be even more misleading for assessing default probabilities. Therefore, it seems that nowadays, even when using a structural model, it may be appropriate to refer to the credit derivatives market to extract default probabilities, and in particular to CDS. Indeed, the CDS market is becoming an increasingly reliable and liquid source of information for market default probabilities, and is rapidly updated when new information on the real financial situation of a debtor is disclosed, sometimes even anticipating it. However, there are technical problems in implementing this approach. Structural models have not

156 - The

Journal of financial transformation

Efficient pricing of default risk: Different approaches for a single goal

els in pricing are not an unavoidable consequence of the hypothesis made, but depend a lot on the data used for expressing these assumptions in quantitative terms, and on the fitting capability of the structural model to these data. Brigo and Tarenghi (2004) also show applications of the model. Firstly, they analyze a concrete default case, the Parmalat case, showing how the model can efficiently embed increasing proximity to default as information on the worsened credit quality of a company is disclosed in time. Secondly, they consider the pricing of a claim embedding counterparty risk, and show how in this setting default correlation can be accounted for in a rather natural way, exploiting the proximity between the underlying being modeled and the equity market.

an intensity) has the form dyt = k( yt)dt + v(ytdWt)1/2, where the vector = (k, , v, y0) has positive components. All parameters have a clear and intuitive role in determining the dynamics of the process. The condition 2k>v2 ensures the process remains positive, and Brigo and Alfonsi (2005) present a scheme to enforce this property also when the model is discretized for simulation. This process features a non-central chi-square distribution and is highly tractable, featuring closed-form formulas for pricing. CIR has often been used in interest rate derivatives pricing, and also for the pricing of credit derivatives. With the development of the interest rate derivatives, it soon appeared strongly limited by the low number of parameters. In fact these models cannot fit exactly the initial term structure of interest rates. As we noticed also in the last section, an increased calibration power is now desirable in the credit market too. When a reference market for a source of risk becomes reliable enough, using models fully consistent with that market price becomes a necessary requisite for giving a fair price to more advanced products. Furthermore, increased liquidity reduces the risk of perfectly fitting unreliable quotations. Using a technique developed on the interest rate market, one can extend the CIR dynamics above to fit a full term structure of CDS. This is done in Brigo and Alfonsi (2005), where references are given, and in the resulting CIR++ model they set the intensity to t = yt + (t,), where y has the dynamics seen before and is a deterministic function. If is determined consistently with the default probabilities extracted from the CDS market (say via a deterministic-intensity model), the model is exactly calibrated to CDS. And we are left with parameters for calibrating different products, although this feature will probably show its usefulness when options in the credit market become more liquid. More importantly, the model extended this way inherits the tractability of CIR.

A positive intensity model with increased calibration power

We mentioned earlier that not all possible specifications of the intensity dynamics are equivalent for applications. We can recall at least three different orders of problems. Firstly, the intensity has to be strictly positive (thus forbidding Gaussian models) for default probabilities to be meaningful. Secondly, not all candidate intensity processes have the same calibration power. If the range of products considered increases, even in a reasonable range, basic common processes can become inadequate. This was already revealed in interest rate modeling. Thirdly, computational times and complexity are extremely important in pricing. Models allowing for closed form formulas (not requiring simulation) are usually preferred, in particular when a model has to be calibrated both to credit and interest rate products. In Brigo and Alfonsi (2005) a model is proposed that aims to give a more complete answer to these requirements than in preceding models. To ensure positive intensity, the family of stochastic processes considered are the square-root diffusions, introduced by Feller (1951) for application in the field of genetics. In finance this process is usually called the CIR process, after the application to short interest rate modeling of Cox, Ingersoll and Ross (1985). The process y (for instance

157

Efficient pricing of default risk: Different approaches for a single goal

When building the general model, Brigo and Alfonsi (2005) assume extended CIR dynamics in parallel to the default intensity and the short interest rate, with the instantaneous correlation between the two Brownian drivers set to . This model has the potential to be fully consistent with interest rates and CDS default probabilities, and parameters are left to include information on both credit and interest rate option markets. In addition, when is set to zero, the calibration of the full model is automatic due to the tractability of the extended CIR coupled with the interesting feature of separability. The latter means that the credit derivatives desk of a bank can ask for interest rate parameters from the interest rates desk and add them in a model to be calibrated to CDS, keeping full consistency in valuation procedures. When 0 this is no longer true, but Brigo and Alfonsi (2005) show empirically that the implications of on CDS prices is bounded by a small fraction of the market bid-ask spread, and is thus negligible. So one can keep the efficient zero-correlation calibration and then set to the desired value in pricing. Approximated formulas are introduced for different payoffs, including CDS options, namely options to enter a CDS at a future time. And with deterministic interest rates and stochastic intensity an exact analytical pricing formula is derived. These formulas have a particular use for detecting the capability of this model to interpret and replicate the smile phenomenon. But for this we first have to see how the smile can be detected on the CDS option market.

ferent variations of the payoff in the market are analyzed in detail, and in particular one CDS payoff structure is singled out, realistic enough for application to the market but simple enough to allow for the construction of a market model. This CDS structure is shown to lead to a CDS option equivalent to a simplified callable defaultable floating rate note. Since this is another relevant payoff in credit options, this latter CDS structure is taken as the central one for building the market model (although the derivation is outlined also for alternative specifications). Setting the price of the CDS to zero and deriving the spread R one finds the correct underlying of the option. Denote by Ta+1, ,Tb the payment dates of the CDS, with time intervals i. Denote by Q the risk neutral probability and by D(t,Ti) the stochastic discount factor. The correct definition of CDS rate turns out: Ra,b (t) = [bi=a+1 E[D(t,Ti)1{Ti-1<Ti} | Ft]]/ [bi=a+1 iQ( > t | Ft) P (t,Ti)], with P (t,Ti) = [E[D(t,Ti)1{ >Ti} |Ft]]/[ Q( > t | Ft)] Here is default time, while E[| Ft)] represents expectation conditional on all information (up to t) except the default time. 1A represents the indicator function for set A. The protection payment is set to 1. The corresponding CDS option, to enter a CDS with fixed rate K at time Ta, has discounted payoff 1{ >Ta} D(t,Ta) [bi=a+1 iP(Ta,Ti)](Ra,b (ta) K)+ and taking risk neutral expectation one finds the price. The technical tools are omitted here and can be found for example in Bielecki and Rutkowski (2001). The technical tool to develop market models is the change of numeraire theory. To put it in a nutshell, this allows us to compute a price, expressed by a risk neutral expectation, as the expectation of a related quantity under a different probability measure (equivalent to the risk neutral one). The purpose of market models is to detect a suitable probabil-

A market model for real world credit payoffs with non-vanishing numeraire
We already described how a market model approach might be useful for the development of the credit options market, and which issues should receive particular attention. Accordingly, rather than starting from an abstract definition, Brigo (2004) focuses on a specific payoff, the options on CDS, since they are written on the most liquid single-name credit derivative, and their market is likely to expand in the future. In order to detect the most convenient state variable to be modeled, Brigo (2004) starts from the real world CDS payoff. The dif-

158 - The

Journal of financial transformation

Efficient pricing of default risk: Different approaches for a single goal

ity measure under which both the underlying is a martingale and the price expression reduces to a Black formula. In our case, this implies that all but (Ra,b (ta) K)+ should get out of the expectation. In Brigo (2004) a probability measure is considered, individuated by its so-called associated numeraire, such that Ra,b(t) is a martingale. In addition, if Ra,b(t) is assumed to follow a lognormal process under this measure, through the change of numeraire the price of the option reduces, before default, to the simple formula bi=a+1 iP(t,Ti)(Ra,b(t)(d1) K(d2)), where () is the cumulate Normal probability and d1 and d2 are as in the Black formula, with reference to the underlying Ra,b(t) at valuation time t. It is interesting to notice that, in this derivation of Brigo (2004), the chosen numeraire cannot drop to zero, in the spirit of Jamshidian (2002), thus ensuring equivalence of the pricing measures.

results, and the application of a model approximation to the derivation of a pricing formula for Constant Maturity CDS, a payoff receiving increasing attention in the market, are given in Brigo (2004).

Understanding the smile and the parameters from a comparison of intensity and market models
Another advantage of having obtained a rigorous market model is that we can detect the implications of the use of a different model (for instance the CIR++ intensity model) on options implied volatility. This means first of all understanding the effects of the model parameters (i.e. in CIR++) on implied volatility. Secondly, it means understanding the pattern of the smile effect intrinsic in the model dynamics. As is well known, the smile effect can be interpreted as a deviation from the lognormality assumption for underlying market observables which, when market quotations are made through a lognormal model, is revealed by a non-flat shape of the graph of implied volatilities plotted against strike prices. From the perspective of a trader, a clear understanding of these implications is of fundamental importance, often influencing the choice of a model at least as much as the technical advantages of the model itself. For the CIR++ stochastic intensity and interest rate model, this kind of analysis is done in Brigo and Cousot (2003). Among other issues, they obtain the following format, as presented in Figure 2, where K is the option strike. Smile patterns implied by explicit intensity models such as CIR++ might be considered unsatisfactory when this phenomenon is clearly established on the CDS option market. In this case one can obtain more flexible patterns modeling directly the market observable R under the relevant measure, similarly to what we have described above, but replacing lognormality with a tractable dynamics allowing for smile (displaced diffusion, CEV, mixture dynamics, SABR).

Rk,b(0) Option 1 Option 2 Figure 1 61 43.4

K 60 43

Price 32.5 24.5

Volatility 62.16% 63.71%

As long as the market has not yet reached a critical level of liquidity, this model would be hard to calibrate to reliable implied volatilities for pricing different products. However, it plays a very important role. It makes it possible to consistently translate the prices of different options into implied volatilities, helping to understand different implications of market options quotations. Let us look at the quotations in Figure 1. It is hard to compare them based on the Price column, provided by the market. If we move to the Volatility column, provided by the model just presented, the information provides a much more effective understanding and comparison. The specification of a general model, including the dynamics of R under a range of probability measures, is currently under investigation. Initial

Parameter K y0 K = 0 K = 1 K = 1 Figure 2

Implied volatility /flat // flat /flat

159

Efficient pricing of default risk: Different approaches for a single goal

Conclusion
In this work we have outlined some general features of quantitative modeling for relative value pricing in the field of credit derivatives. In doing so, we had to address different approaches to modeling, pointing out pros and cons of different frameworks. Then, although avoiding most technical details, we presented three different models in different frameworks, all designed to overcome possible limitations and inadequacies typical of earlier solutions. A final remark on some of the results summarized here is in order. It may seem that some of the solutions presented here are actually ahead of the market, in that they would require a further development in liquidity of certain markets for being fully exploited. However, it has sometimes happened that financial markets have developed only when sufficiently sound technical tools for dealing with such developments had been provided by research. This research typically required a lot of analysis and attempts before reaching a level which, associated to different external factors, allowed the market to take some steps forward in terms of efficiency and stability.

References
Bielecki T., and M. Rutkowski, 2001, Credit risk: Modeling, valuation and hedging, Springer Verlag Black F., and M. Scholes, 1973, The pricing of options and corporate liabilities, Journal of Political Economy, 81, 637-654 Black, F., and J. C. Cox, 1976, Valuing corporate securities: Some effects of bond indenture provisions, Journal of Finance, 31, 351-367 Brigo, D., 2004, Constant maturity credit default swap pricing with market models, available at ssrn.com Brigo, D., 2005a, Market models for CDS options and callable floaters, Risk Magazine, January 2005. Extended version available at damianobrigo.it Brigo, D., and A. Alfonsi, 2005, Credit default swaps calibration and derivatives pricing with the SSRD stochastic intensity model, Available at damianobrigo.it. Finance and Stochastics, Vol. X (1). Brigo, D., and L. Cousot, L., 2003, A comparison between the SSRD Model and a market model for CDS options pricing, Bachelier 2004 Conference Brigo, D., and M. Tarenghi, 2004, Credit default swap calibration and equity swap valuation under counterparty risk with a tractable structural model, Paper presented at the 2004 FEA conference at MIT Duffie D., K. Singleton, 1999, Modeling term structures of defaultable bonds, Review of Financial Studies, 12, 687-720 Hull, J., and A. White, 2003, The valuation of credit default swap options, Rothman school of management working paper Lo C. F., H. C. Lee, and C. H. Hui, 2003, A simple approach for pricing barrier options with time-dependent parameters, Quantitative Finance, 3 Merton, R., 1974, On the pricing of corporate debt: The risk structure of interest rates, Journal of Finance, 29, 449-470

160 - The

Journal of financial transformation

161

Guidelines for manuscript submissions

Guidelines for authors
In order to aid our readership, we have established some guidelines to ensure that published papers meet the highest standards of thought leadership and practicality. The articles should, therefore, meet the following criteria: 1. Does this article make a significant contribution to this field of research? 2. Can the ideas presented in the article be applied to current business models? If not, is there a road map on how to get there. 3. Can your assertions be supported by empirical data? 4. Is my article purely abstract? If so, does it picture a world that can exist in the future? 5. Can your propositions be backed by a source of authority, preferably yours? 6. Would senior executives find this paper interesting?

Manuscript guidelines
All manuscript submissions must be in English. Manuscripts should not be longer than 5000 words each. The maximum number of A4 pages allowed is 10, including all footnotes, references, charts and tables. All manuscripts should be submitted by e-mail directly to the editor@capco.com in the PC version of Microsoft Word. They should all use Times New Roman font, and font size 10. Where tables or graphs are used in the manuscript, the respective data should also be provided within a Microsoft excel spreadsheet format. The first page must provide the full name(s), title(s), organizational affiliation of the author(s), and contact details of the author(s). Contact details should include address, phone number, fax number, and e-mail address. Footnotes should be double-spaced and be kept to a minimum. They should be numbered consecutively throughout the text with superscript Arabic numerals. For monographs Jensen, M., Corporate Control and the Politics of Finance. Journal of Applied Corporate Finance (1991), pp. 13-33. For books Copeland, T., T. Koller, and J. Murrin. Valuation: Measuring and Managing the Value of Companies. John Wiley & Sons, New York, New York (1994). For contributions to collective works Ritter, J. R., 1997, Initial Public Offerings, in Logue, D. and J. Seward, eds., Warren Gorham & Lamont Handbook of Modern Finance, South-Western College Publishing, Ohio. For periodicals Griffiths, W. and G. Judge, 1992, Testing and estimating location vectors when the error covariance matrix is unknown, Journal of Econometrics 54, 121-138. For unpublished material Gillan, S. and L. Starks. Relationship Investing and Shareholder Activism by Institutional Investors. Working Paper, University of Texas (1995).

Subjects of interest All articles must be relevant and interesting to senior executives of the leading financial services organizations. They should assist in strategy formulations. The topics that are of interest to our readership include: Impact of e-finance on financial markets & institutions Marketing & branding Organizational behavior & structure Competitive landscape Operational & strategic issues Capital acquisition & allocation Structural readjustment Innovation & new sources of liquidity Leadership Financial regulations Financial technology

Manuscript submissions should be sent to Shahin Shojai, Ph.D. The Editor Editor@capco.com Capco Clements House 14-18 Gresham Street London EC2V 7JE Tel: +44-20-7367 13 21 Fax: +44-20-7367 1001

162 - The

Journal of financial transformation

Request for papers Deadline June 30th, 2005

The world of finance has undergone tremendous change in recent years. Physical barriers have come down and organizations are finding it harder to maintain competitive advantage within todays truly global market place. This paradigm shift has forced managers to identify new ways to manage their operations and finances. The managers of tomorrow will, therefore, need completely different skill sets to succeed. It is in response to this growing need that Capco is pleased to publish the Journal of financial transformation. A journal dedicated to the advancement of leading thinking in the field of applied finance. The Journal, which provides a unique linkage between scholarly research and business experience, aims to be the main source of thought leadership in this discipline for senior executives, management consultants, academics, researchers, and students. This objective can only be achieved through relentless pursuit of scholarly integrity and advancement. It is for this reason that we have invited some of the worlds most renowned experts from academia and business to join our editorial board. It is their responsibility to ensure that we succeed in establishing a truly independent forum for leading thinking in this new discipline. You can also contribute to the advancement of this field by submitting your thought leadership to the Journal. We hope that you will join us on our journey of discovery and help shape the future of finance.

Shahin Shojai Editor@capco.com

For more info, see page 162

2005 The Capital Markets Company. VU: Shahin Shojai, Prins Boudewijnlaan 43, B-2650 Antwerp All rights reserved. All product names, company names and registered trademarks in this document remain the property of their respective owners.

163

Design, production, and coordination: Cypres Daniel Brandt and Pieter Vereertbrugghen 2005 The Capital Markets Company, N.V. All rights reserved. This journal may not be duplicated in any way without the express written consent of the publisher except in the form of brief excerpts or quotations for review purposes. Making copies of this journal or any portion there of for any purpose other than your own is a violation of copyright law.

1 2

Four Key Questions for Wealth Managers

Rising costs, heavier regulation, tougher clients in a tougher market how can a Private Banking business cope? Can we respond quickly enough to all these changes?

3 4

Should we be redefining our private client value proposition?

SEI knows about the competitive benefits of outsourcing in private banking institutions. To get a copy of SEIs special report Outsourcing and the European Wealth Management Market*, call Francis Jackson on + 44 (0)207 297 6308, or Email fjackson@seic.com

How do we decide what is best done in house, and what needs to be outsourced?

Hear what the industry is saying.

* Research amongst CEOs, CFOs and Senior Executives of wealth management institutions in 10 European Countries SEIInvestments (Europe) Ltd, 4th Floor, The Economist Building, 25 St Jamess Street, London SW1A 1HA is regulated and authorised by the Financial Services Authority

SEI. New Ways. New Answers.

Antwerp Bangalore Boston Frankfurt London New York Paris San Francisco

T +32 3 740 10 00 T +91 80 527 0353 T +1 617 262 1135 T +49 69 9760 9000 T +44 20 7367 1000 T +1 212 284 8600 T +33 1 47 55 30 90 T +1 415 445 0968 T +65 6395 6998

www.capco.com

Singapore