You are on page 1of 9

Active Portfolio Management a la Grinold & Kahn

Active portfolio management is based on the fundamental belief that there are opportunities
available in the market and one can do better than just passively managing the portfolio.
Information – whether internally or externally generated – is at the heart of active portfolio
management. Portfolio is managed actively in order to beat the performance of a benchmark
portfolio. Information gives rise to forecast which is different from the consensus market
opinion. Based on this forecast, the portfolio manager engages in active trading of the portfolio.
Since the risk and return profile of the benchmark portfolio is accommodated in the active
portfolio, what matters to the portfolio manager is the active return and active risk as opposed
to total return and the total risk. Any setback in benchmark return gets reflected in the total
portfolio return also, but that risk is borne by the owner of the fund. In the absence of benchmark
timing when the portfolio is tilted towards high/low beta stocks away from the benchmark, the
portfolio risk is kept along the line of the benchmark and the active and residual risks and returns
are identical.

Return Risk

Active r =r − r
PA P B ψ = S.D. (r )
P PA r = active excess (net of r ) return
PA F

=θ +β .r P PA B = √ω2 + β2 . σ2
P PA B r = portfolio excess return
P

β = β –1
PA P = tracking error r = benchmark excess return
B

β = beta (on benchmark) of portfolio


P

Residual θ =r −β .r
P P P B ω = S.D. (θ )
P P β = beta of active portfolio
PA

α = E (θ )
P P θ = residual return on portfolio
P

ω = residual risk
P

σ2 = benchmark return variance


B

Ex-post α is the average of the realised residual returns and can be obtained by running the
following time-series regression of stock n and benchmark excess returns:

r (t) = α + β . r (t) + ε (t)


N N N B N

where, ε (t) is the random component of residual return with expected value being zero.
N

This breaks the asset excess return into α and β components – the uncorrelated and the correlated
parts to the benchmark portfolio.

α has the portfolio property, e.g., weighted average of individual stock alphas is α . Benchmark P

portfolio does not have any residual return and therefore, α = 0 and so is the alpha of risk-free B

asset.

Active portfolio manager is on the search for alphas; that is the avenue for outperforming the
benchmark. Manager is also responsible for the portfolio residual risk and not for the overall risk.
An institutional money manager is willing to accept the benchmark portfolio with 20% risk and
loath to take on a portfolio with 21% risk if it contains 20% benchmark risk and 6.4% residual
risk. A high level of residual risk means there is a large chance of earning low or negative
residual return, which will be harmful for the manager’s career. The observed residual risk
assumption, therefore, is quite low (see the table next page).
2

An information ratio, denoted by IR, is a ratio (usually annualised) of residual return to


residual risk. The ex-post information ratio is a measure of achievement of the portfolio manager
and is

IR = α / ω
P P

obtained by dividing realised residual return with the residual risk taken to obtain that return. A
realized IR can be anything from positive to negative. Often, it is the latter. Here is some
perspective.

US Active Equity Mutual Funds (300 funds during the period Jan. `91 to Dec. `93)

Percentile IR(before fees) IR(after fees) Annual Active Risk


90 1.33 1.08 9.87%
75 0.78 0.58 7.00%
50 0.32 0.12 4.76%
25 - 0.08 - 0.33 3.66%
10 - 0.47 - 0.72 2.90%

IR is independent of manager’s level of aggressiveness, i.e., if the manager can get a residual
return of 2% with 4% residual risk, a higher level of weights (3 times the previous weights) on
long-short positions can take the return to 6%; but residual risks also will rise by the same
multiple so as to make the IR unchanged. Don’t believe? Here is the proof.
Active portfolio is just the difference in holdings (the weights: w) between the benchmark and
the managed portfolios. Active holding w = w – w and V is the covariance matrix.
PA P B

Active variance ψ2 = w . V . w
P PA
T
PA And if the active holdings change from w to 3w , then
PA PA

the active risk changes from ψ to 3ψ = √3w . V . 3w .


P P PA
T
PA

IR is assumed to be independent of level of risk. However, this does break down in real life due
to constraints imposed on the manager at higher level of risk. IR, however, does depend upon the
time horizon. It is usually expressed in 1-year time horizon. Monthly returns, when annualised,
will go up by 12 times. However, the standard deviation will grow only by √12. This will change
the IR with the time horizon.

Ex-ante α is the forecast of residual return and ex-ante IR – expected level of annual residual
return per unit of annual residual risk – is the measure of opportunity.

Here is a quick stab (more will follow) at α prediction. As we can deduce from the table above
(median realized alpha [IR x Active Risk] is only 1.52%), the accuracy of a successful forecaster
of alphas is fairly low. A rule of thumb method, therefore, may be used. We can sort all assets in
five categories (more sophistication may be introduced by first segregating them into economic
sectors) – strong buy, buy, hold, sell and strong sell. Assign them alphas of 2%, 1%, 0%, -1% and
–2% respectively. Find the average alpha of the benchmark portfolio. If it is not zero (remember
α will have to be 0), then modify the alphas by subtracting from each original alpha number you
B

assigned, the benchmark average times the stock’s beta. This will make the alphas benchmark-
neutral (No? Try it. Just remember β = 1). Thus it may be difficult to forecast alphas correctly, it
B

is not difficult to forecast alphas directly.


3

As can be seen in the table above, high IR is difficult to achieve. We can classify IR = 1 as
exceptional, IR = .75 as very good and IR = .5 as good. The choices available to the active
manager are easier to see if we look at the alpha versus residual risk trade-offs. The ex-ante IR
determines the manager’s residual frontier, which describes the opportunities available to the
active manager. Effectively, the information ratio defines a ‘budget constraint’ for the manager.

Residual Frontier
12%
10%
Residual Return

8%
6%
IR = 1
4%
IR= .75
2%
IR = .5
0%
0% 2% 4% 6% 8% 10% 12%
Residual Risk

The objective of active management is to maximize the value added (akin to utility function)

VA[P] = α - λ . ω2 P R P

from residual return where λ measures the aversion to residual risk – it has the effect of
R

transforming residual variance into a cost in terms of loss in alpha. Following is a graph of
4

Constant Preference Curves

14.00%
12.00%
Residual Return

10.00%
8.00% VA = 3
VA = 1.5
6.00%
VA = .75
4.00%
2.00%
0.00%
0.00% 2.00% 4.00% 6.00% 8.00% 10.00% 12.00%
Residual Risk

lines of equal value added (three different levels), plotted as functions of expected residual return
and residual risk and for a risk aversion parameter λ value of 0.10.
The information ratio describes the opportunities open to the active managers. The active
manager should explore those opportunities and choose the portfolio that maximizes value
added. The tangency point between the residual frontier and the constant preference curve gives
the optimal portfolio. It indicates the optimum alpha and residual risk given the degree of risk
aversion and information ratio.

Value added function, when written in terms of residual risk, appears as follows:

VA[ω] = ω . IR - λ . ω2P R P

The optimal level of residual risk ω* which maximizes VA is : ω* = IR / 2λ . Desired level of


R

residual risk will increase with opportunities and decrease with residual risk aversion.
At this optimal risk level, Value Added becomes: VA[ω*] = IR2 / 4λ = ω* . IR / 2. This says the
R

ability of the manager to add value increases as the square of the information ratio and decreases
as the manager becomes more risk-averse. This has the effect of separating the decision as to
manager selection from the aggressiveness in investment strategy. Every investor seeks the
manager with the highest IR and then decides how aggressively to take risk.
Optimal residual risk also indicates the implied optimum risk aversion level as: λ = IR / 2 . ω*.
R

What determines the IR? Fundamental Law of Active Management gives an approximation to
IR. The law is based on two attributes of a strategy, breadth and skill. The breadth (BR) of a
strategy is the number of independent investment decisions that are made each year and the skill,
represented by the Information Coefficient (IC), measures the quality of those investment
decisions. IC is the correlation of each forecast with the actual outcome and is assumed for
convenience to be the same for all forecasts.

IR = IC . √BR
5

To increase the information ratio from .5 to 1, we need to either double our skill or increase our
breadth by a factor of 4, or do some combination of both. This relationship can be written in
terms of residual risk also.
ω* = IR / 2λ = IC . √BR / 2λ
R R

We see that the desired level of risk-taking will increase directly with the skill level and the
square root of the breadth. The breadth allows for diversification among the active bets so that
the overall level of aggressiveness ω* can increase. The skill increases the possibility of success;
thus, we are willing to incur more risk, since the gains appear to be larger.

The fundamental law is designed to give us insight into active management. It isn’t an
operational tool. A manager needs to know the trade-offs between increasing the breadth of the
strategy BR – by either covering more stocks or shortening the time horizons of the forecasts –
and improving skill IC. Operationally, it will prove difficult in particular to estimate BR
accurately, because of the requirement that the forecasts be independent. This means that forecast
2 should not be based on a source of information that is correlated with the source of forecast 1.
For example, suppose that our first forecast is based on as assumption that growth stocks will do
poorly, and our second is based on an assumption that high-yield stocks will do well. These
pieces of information are not independent; growth stocks tend to have very low yields, and not
many high-yield stocks would be called growth stocks. We have just picked out two ways to
measure the same phenomenon. An example of independent forecasts is a quarterly adjustment
of the portfolio’s beta from 1 to either 1.05 or .95 as a market timing decision based on new
information each quarter. The following figure shows the trade-offs between breadth and skill for

Skill-Breadth Trade-off

800.00

(BR) 600.00


dth
400.00  
Brea

 
200.00

0.00

0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16

Skill (IC)

two levels of information ratio. At low level of IC (where most of them belong to), breadth has to
increase dramatically in order to raise the IR level.

We can see the power of the law by making an assessment of three strategies. In each strategy,
we want an information ratio of .50. Start with a market timer who has independent information
about market returns each quarter. The market timer needs an information coefficient of 0.25,
since 0.50 = 0.25√4. As an alternative, consider a stock selector who follows 100 companies and
revises the assessments each quarter. The stock selector makes 400 bets per year; he needs an
6

information coefficient of 0.025, since 0.50 = 0.025√400. As a third example, consider a


specialist who follows two companies and revises her bets on each 200 times per year. The
specialist will make 400 bets per year and require a skill level of 0.025. The stock selector
achieves breadth by looking at a large number of companies intermittently, and the specialist
achieves it by examining a small group of companies constantly. We can see from these
examples that strategies with similar information ratios can differ radically in the requirements
they place on the investor. The basic insight that we gain from the fundamental law is that it is
important to play often (high BR) and to play well (high IC).

The fundamental law is additive in the squared information ratios. Suppose the two portfolios (1
& 2) that a manager used to handle is combined into a single portfolio under him. The
information ratio for the manager will be

IR2 = BR1 . IC12 + BR2 . IC22

This assumes optimal implementation of the alphas across the combined set of stocks.

Information is the vital input into any active management strategy. Information
separates active management from passive management. Information, properly
applied, allows active mangers to outperform their information-less benchmarks.

Active managers use information to predict the future exceptional return on a group of
stocks. The emphasis is on predicting alpha, or residual return – beta-adjusted return
relative to a benchmark. We want to know what stocks will do better than average and
what stocks will do worse, on a risk-adjusted basis.

Information is made up of signal plus noise. The signal is linked with future stock
returns. The noise masks the signal and makes the task of information analysis both
difficult and exciting. Information analysis begins by transforming information into
something concrete: investment portfolios. Then it analyses the performance of those
portfolios to determine the value of the information. Information analysis can work
with something as simple and publicly available as an analyst’s buy and sell
recommendations; or it can be derived data, such as a change in consensus earnings
forecasts.

By way of example, we will use book-to-price data to generate return predictors


according to various standard schemes. Underlying the book-to-price example will be
the hypothesis that book-to-price ratios contain information concerning future stock
returns, and, in particular, that high book-to-price stocks will outperform low book-to-
price stocks. Is this hypothesis true? How much information is contained in book-to-
price ratios?

There are a great many ways to generate portfolios from information, and the
procedure selected could depend on the type of information. Here are six possibilities.
7

Procedure1. With buy and sell recommendations, we could equal- (or value-) weight
the buy group and the sell group [Rank according to B/P ratio – top half in buy list
and the bottom half in sell list].
Procedure 2. With scores, we can build a portfolio for each score by equal- (or value-)
weighting within each score category [Rank as above, divide into quintiles or deciles
and assign a score for each part]
Procedure 3. With straight alphas, we can split the stocks into two groups, one group
with higher than average alphas and one group with lower than average alphas. Then
we can weight the stocks in each group by how far their alpha exceeds or lies below
the average [One way is to assume that alphas and B/P ratios are linearly related and
weight the stocks with their deviations from mean B/P ratio].
Procedure 4. With straight alphas, we can rank the assets according to alpha, then
group the assets into quintiles or deciles and equal- (or value-) weight within each
group [Combination of 2 and 3].
Procedure 5. With any numerical score, we can build a factor portfolio that bets on the
information and does not make a market bet. The factor portfolio consists of a long
portfolio and a short portfolio. The long and short portfolios have equal value and
equal beta, but the long portfolio will have a unit bet on the information, relative to
the short portfolio. Given these constraints, the long portfolio will track the short
portfolio as closely as possible [For B/P ratio, we can build long and short portfolios
with equal value and beta, with the long portfolio exhibiting a B/P ratio one standard
deviation above that of the short portfolio, and designed so that the long portfolio will
track the short portfolio as closely as possible].
Procedure 6. With any numerical score, we can build a factor portfolio, consisting of a
long and a short portfolio, designed so that the long and short portfolios are matched
on a set of pre-specified control variables. For example, we could make sure the long
and short portfolios match on industry, sector, or small-capitalization stock exposures.
This is a more elaborate form of procedure 5, where we controlled only for beta (as a
measure of exposure to market risk) [Using B/P data, this is an extension of 5].

The general idea should be clear. We are trying to establish some sort of relative
performance. In each case, we will produce two or more portfolios. In the first, third,
fifth and sixth procedures, we will have a long and a short portfolio. The long bets on
the information; the short bet against it. In procedure 2, we have a portfolio for each
score, and in procedure 4, we have a portfolio for each quintile.

The simplest form of performance analysis is just to calculate the cumulative returns
on the portfolios and the benchmark, and plot them. Some summary statistics, like the
means and standard deviations of the returns, can augment this analysis. We can also
do regression analysis on the portfolio returns, regressing the excess portfolio returns
against the excess benchmark returns. This regression will estimate the portfolio’s
alpha and beta, and we can evaluate, via the t statistic, whether the alpha is
significantly different from zero.

The other two measures are Information Ratio and Information Coefficient. IR is the
best single statistic for capturing the potential for value added from active
management. IC is the correlation between information data and realized alphas. If the
data item is all noise and no signal, the information coefficient is 0. If there is a
8

perverse relationship between the data item and the subsequent alpha, the information
coefficient can be negative.

More detailed analysis of information is possible. For instance, when we build long
and short portfolios to bet for and against the information, we can also observe
whether our information better predicts upside or downside alphas. We can also
investigate whether the data contain information pertaining to up markets or down
markets. We can construct portfolios controlling for variables like industry, size etc.
and see our performance or construct portfolios with different controls and analyse the
performance in each case.

Active management is forecasting. The consensus forecasts of expected returns,


efficiently implemented, lead to the market or benchmark portfolio. Active managers
earn their title by investing in portfolios that differ from their benchmark. As long as
they claim to be efficiently investing based on their information, they are at least
implicitly forecasting expected returns. Forecasting is beyond the scope of this course.
There exist a variety of methods – time series analysis, ARCH, GARCH, Kalman
Filters, Chaos Theory, Neural Nets, Genetic Algorithms ….. (At least now you know
the names! Pursue in econometrics course.).

Information when found to have value, may be used for forecasting expected returns.
In the case of one asset and one forecast, we refine the raw forecast by
1) Standardizing the raw forecast by subtracting the expected forecast and dividing by
the standard deviation of the forecasts. Standardized version of the raw forecast is
called a score or z score.
2) Scaling the score to account for the skill level of the forecaster (the IC) and the
volatility of the return we are attempting to forecast.
This leads to the forecasting rule of thumb:

Refined forecast = volatility. IC. score

This refinement process – converting raw forecasts into refined forecasts – controls
for three factors: expectations, skill and volatility. The score calculation controls for
expectations by the subtraction of the unconditional expected raw forecast. IC is the
control for skill level. Finally, the volatility term provides the dimensions of return.
Note that given a skill level and two stocks with the same score, the higher-volatility
stock receives the higher alpha. Perhaps a utility stock and an Internet stock are in the
buy list and both are expected to rise. The Internet stock should rise more.

Forecasts for returns have negligible effect on forecasts of volatility and correlation.
This result arises because risk measures uncertainty in the return. A skilful forecaster
can reduce the amount of uncertainty in the return and a perfect forecaster reduces the
uncertainty to zero (the returns can still vary from month to month, but only exactly
according to forecast). For any forecaster, however, the size of the remaining
uncertainty in the return stays the same, independent of any particular forecast.
9

The correlation between forecast and realized alphas is a critical component in


determining the information ratio, according to the fundamental law of active
management and is a critical input for refining alphas and combining signals.

You might also like