IRR

! ! ! !

Term Structure Estimation

Sanjay K. Nawalkha Gloria M. Soto
!

!
Electronic copy available at: http://ssrn.com/abstract=1096182

Term Structure Estimation

Sanjay Nawalkha* Gloria M. Soto** http://www.fixedincomerisk.com/ Last updated: February 19, 2009

Abstract The term structure of interest rates gives the relationship between the yield on an investment and the term to maturity of the investment. Since the term structure is typically measured using default-free, continuously-compounded, annualized zero-coupon yields, it is not directly observable from the published coupon bond prices and yields. This paper focuses on how to estimate the default-free term structure of interest rates from bond data using three methods: the bootstrapping method, the McCulloch cubic-spline method, and the Nelson and Siegel method. Nelson and Siegel method is shown to be more robust than the other two methods. The results of this paper can be implemented using user-friendly Excel spreadsheets.
Practical Guide Series in Fixed Income The aim of the Practical Guide Series in Fixed Income is to provide a simple introduction to the basics of fixed income valuation, interest rate modelling, and credit risk modelling. These guides are based upon selected chapters in the Trilogy on the Fixed Income Valuation Course by Wiley Finance. The guides are accompanied with free excel spreadsheets that allow readers with basic excel skills to instantly play with the various interest rate models and credit risk models. A more detailed analysis of the topics herein and other subjects related to fixed income analysis, is provided in the Trilogy on the Fixed Income Valuation Course. A description of the Trilogy can be found at http://www.fixedincomerisk.com/ *Department of Finance and Operations Management, Isenberg School of Management, University of Massachusetts, Amherst, MA 01003. E-mail: nawalkha@som.umass.edu ** Departamento de Economia Aplicada, Universidad de Murcia, Murcia, Spain.

Electronic copy available at: http://ssrn.com/abstract=1096182

1.

Introduction
The term structure of interest rates, or the TSIR, can be defined as the relationship

between the yield on an investment and the term to maturity of the investment. The TSIR is typically measured using continuously-compounded, annualized zero-coupon yields. The TSIR is not directly observable due to the non-existence of zero-coupon bonds over a continuum of maturity dates. As a consequence, the TSIR is generally estimated by applying some term structure estimation method to a set of coupon bearing bonds with different maturities. As noted by Bliss [1997], the TSIR estimation requires making three important decisions. First, one must consider the assumptions related to taxes and liquidity premiums in the pricing function that relates bond prices to interest rates or discount factors. Second, one must choose a specific functional form to approximate the interest rates or the discount factors. And third, one must choose an empirical method for estimating the parameters of the chosen functional form. This practitioner guide focuses on the three popular TSIR estimation methods given as the boostrapping method, the McCulloch cubic spline method and the Nelson and Siegel method. The structure of the guide is as follows. First, we review the notation and some important concepts. Next, we describe each of the three tern structure estimation methods and illustrated them using numerical examples. Finally, we describe the excel spreadsheet file for applying the three term structure estimation methods, using a few snapshots of the spreadsheet file.

2.

The Building Blocks: Bond Prices, Spot Rates, and Forward Rates
The TSIR can be expressed in terms of spot rates, forward rates, or prices of discount

bonds. This section shows the mathematical relationship between these concepts. 2.1. The Discount Function Under continuous compounding, the price (or present value) of a zero-coupon bond with a face value of $100 and a term to maturity of t years can be written as:

P (t ) "

100 " 100 e ! y ( t )t " 100 d (t ) y ( t )t e

(1)

where y(t) is the continuously-compounded rate corresponding to the maturity term t. The function y(t) defines the continuously-compounded term structure based upon zero-coupon rates. The expression e-y(t)t is referred to as the discount function d(t). The typical shape of the discount function is shown in Figure 1. This function starts at 1, since the current value of a $1 payable today is $1, and it decreases with increasing maturity due to the time value of money.

0 Term

Figure 1

The discount function

If a series of default-free zero-coupon bonds exist for differing maturities, then it is possible to extract the term structure by simply inverting equation (1) to obtain y(t). However, due to the lack of liquidity and unavailability of zero-coupon bonds for all maturities, the term structure cannot be simply obtained by using zero-coupon bonds such as U.S. Treasury STRIPS.

2.2.

Bond Price and Accrued Interest

A coupon bond can be viewed as a portfolio of zero-coupon bonds. Consequently, the

3

price of a coupon-bearing bond that makes a total of N coupon payments, k times a year at times t1, t2,..., tN, and a face value F, be given as:1

P "$
j "1

C e
y (t j )t j

F e
y ( tN ) tN

(2)

where C is the periodic coupon paid k times a year. Equivalently, we have:

P " $ C % d (t j ) # F % d (t N )
j "1

(3)

These formulas give what is called the cash price of a bond. This is the price that purchaser pays when buying the bond. However, bond prices are not quoted as cash prices. The quoted prices are clean prices, which exclude the accrued interest. Accrued interest is the interest accumulated between the most recent interest payment and the present time. If t0 denotes the current time, tp denotes the date of the previous coupon payment, and tq denotes the date of the next coupon payment, then the formula for accrued interest is given as:

& t !t AI " C ( 0 p ( t !t * q p

' ) ) +

(4)

and the bonds quoted price is then expressed as:

When compounding is discrete, each ey(t)t is replaced by (1+APR(t)/k)tk, where APR(t) is the Annual Percentage

Rate for term t compounded k times a year.

Quoted Price " P ! AI " $

j "1

C e
y ( t j )t j

F e
y ( t N ) tN

& t !t !C( 0 p (t !t * q p

' ) ) +

(5)

Computation of accrued interest requires the day count basis used in the market. The day count basis defines how to measure the number of days: i) between the current date and the date of the previous coupon payment, or t0 tp, and ii) between the coupon payment dates before and after the current date, or tq tp. The most widely used day count basis are given as follows:
,

Actual/Actual: both t0 tp and tq tp are measured using the actual number of days between the dates.

Actual/360: t0 tp is measured using the actual number of days between t0 and tp, and tq tp equals 360/k, where k is the number of coupon payments made in one year.

30/360: t0 tp is measured as 30 % number of remaining and complete months + actual number of days remaining between the dates t0 and tp, and tq tp equals 360/k, where k is the number of coupon payments made in one year.

The Actual/Actual basis is used for Treasury bonds, the Actual/360 basis is used for U.S. Treasury bills and other money market instruments, and the 30/360 basis is used for U.S. corporate and municipal bonds.

Example 1 Consider a semi-annual coupon bond with a $1,000 face value and a 5% annualized coupon rate with coupon payments on June 1 and December 1. Suppose we wish to calculate the accrued interest earned from the date of the last coupon payment to the current date, November 3.

If we use the Actual/Actual day count basis to measure the number of days between dates, then, t0 tp = the actual number of days between June 1 and November 3 = 155, and tq tp = the actual number of days between June 1 and December 1 = 183. The accrued interest is given as:

AI " C

& 155 ' " 25 ( ) " $21.17 tq ! t p * 183 +

t0 ! t p

where C = 50/2 = $25, is the semi-annual coupon payment. If we used the Actual/360 convention, then, t0 tp = the actual number of days between June 1 and November 3 = 155, and tq tp = equals 360/k = 360/2 = 180. The accrued interest is given as:

AI " C

& 155 ' " 25 ( ) " $21.53 tq ! t p * 180 +

t0 ! t p

Finally, if we used the 30/360 convention, then, t0 tp = 30 %5 (five complete months times 30 days each month) plus 2 (number of actual days from November 1 to November 3) = 152, and tq tp = equals 360/k = 360/2 = 180. The accrued interest is given as:

AI " C

& 152 ' " 25 ( ) " $21.11 tq ! t p * 180 +

t0 ! t p

Now, assume that the bond in the above example is a U.S. Treasury bond quoted at the price of $95-08 on November 3. Since Treasury bonds are quoted with the accuracy of thirty-two seconds to a dollar, a quoted price of 95-08 corresponds to a price of $95.25 on a $100 face value, and hence a price of $952.50 for the $1,000 face value. price equal to: Using an Actual/Actual day count basis, the accrued interest on November 3 is $21.17, giving a cash

P " Quoted Price + AI = 952.5 # 21.17 " $973.67

It is not the cash price, but the quoted price that depends on the specific day count convention being applied. Any increase (decrease) in the accrued interest due to a specific day count convention used is exactly offset by a corresponding decrease (increase) in the quoted price, so that the cash price remains unchanged. Since the TSIR is computed using cash prices, it is also independent of the day count convention used. price and the accrued interest. Of course, it is necessary to know the day count convention in order to obtain the cash price using the quoted

2.3.

Yield to Maturity

The yield to maturity is given as that discount rate which makes the sum of the discounted values of all future cash flows (either of coupons or principal) from the bond equal to the cash price of the bond, that is:2

When compounding is discrete, each eyt is replaced by (1+y/k)tk. Since cash price is used in equation (6), sometimes the discount rate is also called the adjusted yield to maturity.

P "$
j "1

C e
y %t j

F e
y %tN

(6)

A comparison of equations (6) and (2) reveals that the yield to maturity is a complex weighted average of zero-coupon rates. The size and timing of the coupon payments influence the yield to maturity, and this effect is called the coupon effect. In general, the coupon effect will make two bonds with identical maturities but with different coupon rates or payment frequencies have different yield to maturities if the zero-coupon yield curve is non-flat. The coupon effect makes the term structure of yields on coupon bonds lower (higher) than the term structure of zero-coupon rates, when the latter is sloping upward (downward).

2.4.

Spot Rates, Forward Rates and Future Rates

Zero-coupon rates as defined above are spot rates because they are interest rates for immediate investments at different maturities. The forward rate between the future dates t1 and t2 is the annualized interest rate that can be contractually locked in today on an investment to be made at time t1 that matures at time t2. The forward rate is different from the future rate in that the forward rate is known with certainty today, while the future rate can be known only in future. Consider two investment strategies. The first strategy requires making a riskless investment of $1 at a future date t1, which is redeemed at future date t2 for an amount equal to:

1% e f ( t1 ,t2 )( t2 !t1 )

(7)

The variable f(t1, t2) which is known today is defined as the continuously-compounded annualized forward rate, between dates t1 and t2. Now consider a second investment strategy that requires shorting today (which is the same as borrowing and immediately selling) a $1 face value riskless zero-coupon bond that matures at time t1 and investing the proceeds from the short sale in a two year riskless 8

investment maturing at time t2. The proceeds of the short sale equal P(t1), the current price of $1 face value riskless zero-coupon bond that matures at time t1. This investment costs nothing today, requires covering the short position at time t1 by paying $1, and receiving the future value of the proceeds from the short sale, which at time t2 equals:

P (t1 )e y ( t2 )%t2 "

e y ( t2 )%t2 e y ( t1 )%t1

(8)

where y(t1) and y(t2) are zero-coupon rates for terms t1 and t2. Since both riskless investment strategies require $1 investment at time t1, and cost nothing today, the value of these investment strategies at time t2 must be identical. This implies that the compounded value in expression (7) must equal the compounded value in equation (8), or:

e f ( t1 ,t2 )( t2 !t1 ) "

e y ( t2 )%t2 e y ( t1 )%t1

(9)

Taking logarithms on both sides of the above equation and further simplification we obtain:

f ( t1 , t 2 ) "

y (t 2 )t 2 ! y (t1 )t1 t 2 ! t1

(10)

Rearranging the terms:

f ( t1 , t 2 ) " y ( t 2 ) #

y ( t 2 ) ! y ( t1 ) t1 t 2 ! t1

(11)

The above equation implies that if the term structure of zero-coupon rates is upward (downward) sloping, then forward rates will be higher (lower) than zero-coupon rates. For a flat term structure, zero-coupon rates and forward rates are identical and equal to a constant.
9

Example 2 Table 1 illustrates the calculation of forward rates. The second column of the table shows the continuously-compounded zero-coupon rates for terms ranging from 1 to 10 years. The third column gives the one-year forward rates implied by the zero-coupon rates. Table 1 Computing implied forward rates Maturity t 1 2 3 4 5 6 7 8 9 10 Spot rates y(t) 5.444 5.762 5.994 6.165 6.294 6.393 6.471 6.532 6.581 6.622 Forward rates f(t-1,t) 6.080 6.457 6.679 6.811 6.888 6.934 6.961 6.977 6.987

For example the forward rate f(2,3) is computed using equation (10) as follows:

f (2,3) "

0.05994 % 3 ! 0.05762 % 2 " 0.06457 " 6.457% 3!2

(12)

All forward rates derived in the above example apply over the discrete time interval of one year. In general, forward rates can be computed for any arbitrary interval length, and each length implies a different term structure of forward rates. To avoid this indeterminacy, the term structure of forward rates is usually defined using instantaneous forward rates. Instantaneous forward rates are obtained when the interval length becomes infinitesimally small. Mathematically, the instantaneous forward rate f(t), is the annualized rate of return

10

locked-in today, on money to be invested at a future time t, for an infinitesimally small interval dt - 0, and can be derived using equation (11), by substituting t2 = t + dt and t1 = t:

f (t ) " lim f (t ,t # dt ) " y (t ) #

dt -0

.y (t ) t .t

(13)

The instantaneous forward rates can be interpreted as the marginal cost of borrowing for an infinitesimal period of time beginning at time t. Using equation (13), the term structure of instantaneous forward rates (or simply forward rates, from here on) can be derived from the term structure of zero-coupon rates. Equation (13) can be expressed in an integral form as follows:3

y (t ) / t " 0 f (s )ds
0

(14)

or,

y (t ) "

0 f (s)ds
0

(15)

The above equation gives a relationship between zero-coupon rates and forward rates. It implies that the zero-coupon rate for term t is an average of the instantaneous forward rates beginning from term 0 to term t. Since averaging reduces volatility, this relationship suggests that forward rates should be in general more volatile than zero-coupon rates, especially at the longer end. An excellent visual exposition of the difference in the volatilities of

Equation (14) is the solution to the first order differential equation given in (13). This can be verified by taking the

partial derivative of both sides of equation (14), with respect to t.

11

the zero-coupon yields and those of the instantaneous forward rates is given in the excel file guide01movie.xls explained at the end of this guide.

3.

Other Practical Topics

In this section we discuss two important topics in the estimation of the TSIR: the

shape of the term structure and the selection of market data. For other advanced issues, such as the use error-weighting schemes or penalty parameters for the estimation, we refer the reader to Nawalkha, Soto and Beliaeva [2005], chapter 3.

3.1.

The Shape of the Term Structure of Interest Rates

Estimation of the term structure involves obtaining zero-coupon rates, or forward rates, or discount functions from a set of coupon bond prices. Generally, this requires fitting a functional form that is flexible in capturing stylized facts regarding the shape of the term structure. The TSIR typically takes four different shapes given as the normal shape, the steep shape, the humped shape and the inverted shape. Figure 2 shows these four typical shapes.

inverted Zero-coupon rates (%)

humped

normal steep

Term

Figure 2

Basic shapes of the term structure

12

The normal shape is generally indicative of an economy that is expanding normally. That is, the term structure tends to be sloping upwards, reflecting the fact that longer-term investments are riskier. A higher risk implies a higher risk premium and hence, a higher interest rate. The steep shape of the term structure typically occurs at the trough of a business cycle, when after many interest rate reductions by the central bank, the economy seems poised for a recovery in the future. The inverted shape of the term structure typically occurs at the peak of a business cycle, when after many interest rate increases by the central bank, the economic boom or a bubble may be followed by a recession or a depression. Finally, the humped shape typically occurs when the market participants expect a short economic recovery followed by another recession so that there are different expectations at different terms. It could also occur when moving from a normal curve to an inverted curved or vice versa4. It is also worthy to highlight that whatever the shape, the TSIR tend to be horizontal at longest maturities. The reason for this is twofold. First, although investors can hold different expectations about the future of interest rates for the short, medium, and long terms, their long term their expectations are more diffused, which makes it difficult to establish differences between different long rates. Second, risk premiums tend to be more stable for longer terms.

3.2.

Selection of Market Data

Usually, not all the bonds that trade in the market at a given time are used for the estimation of the TSIR. The bond selected must cover a wide spectrum of maturities, should have an enough degree of liquidity and their prices shouldnt incorporate high distortions due to tax effects or other market frictions. Usually, these requirements are fulfilled by the establishment of filtering criteria for determining the bonds that qualify for inclusion in the sample. For example, liquidity is related to both the daily trading volume and the outstanding

The shape of the term structure is also explained by other variables not related to expectations such as liquidity

premium, market segmentation, etc. Alternative term structure hypothesis have assigned different roles to these variables. For a brief discussion about the main hypothesis, see Nawalkha, Soto and Beliaeva [2005], pp. 52-55.

13

amount of a bond. The definition of filtering criteria, however, is not a simple task. A loose criterion might improve the robustness in estimation but at the cost of introducing unrepresentative data that might distort the resulting term structure. On the contrary, a tight filtering criteria reduces the number of data points - and hence, detrimental to the estimation - but ensures higher quality of data. Moreover, the choice of a threshold value for any filtering criteria might be somewhat arbitrary. Consequently, the empirical analysis and good judgment of the researcher must determine the filtering criteria.

4.

Basic Methods for Term Structure Estimation

First attempts to estimate the term structure relied on fitting smooth functions to the

yields to maturity of bonds using regression analysis.

However, this approach was

unsatisfactory due to its limitation in identifying the zero-coupon yields, and in dealing with the coupon effect. The seminal work of J. Huston McCulloch in 1971 suggested a new method based on quadratic splines, which focused directly on estimating zero-coupon yields and discount factors. Much research has extended the work of McCulloch in the past three decades. Methods for TSIR estimation must find a way to approximate the spot rates, or the forward rates, or the discount function. Generally, this requires fitting a parsimonious functional form that is flexible in capturing stylized facts regarding the shape of the term structure. A good term structure estimation method should satisfy the following requirements: , , The method ensures a suitable fitting of the data. The estimated zero-coupon rates and the forward rates remain positive over the entire maturity spectrum. , The estimated discount functions, and the term structures of zero-coupon rates and forward rates are continuous and smooth. , The method allows asymptotic shapes for the term structures of zero-coupon rates and forward rates at the long end of the maturity spectrum.

14

The commonly used term structure estimation methods are given as the bootstrapping method, the polynomial/exponential spline methods of McCulloch [1971,1975] and Vasicek and Fong [1982], and the exponential functional form methods of Nelson and Siegel [1987] and Svensson [1994]. Extensions of the above methods are given as the heteroscedastic error correction based model of Chambers, Carleton, and Waldman [1984], and the error weighing models such as the B-spline method of Steely [1991] and the penalized spline methods of Fisher, Nychka and Zervos [1995] and Jarrow, Ruppert, and Yu [2004]. In this section we focus on the three most commonly used term structure estimation methods: the bootstrapping method, the McCulloch polynomial cubic-spline method, and the Nelson and Siegel exponential-form method. Some applications of the Nelson and Siegel method for term structure modeling are analyzed in Nawalkha, Beliaeva and Soto [2007].

4.1.

Bootstrapping

The bootstrapping method consists of iteratively extracting zero-coupon yields using a sequence of increasing maturity coupon bond prices. This method requires the existence of at least one bond that matures at each bootstrapping date. To illustrate this method, consider a set of K bonds that pay semi-annual coupons. The shortest maturity bond is a six-month bond, which by definition does not have any intermediate coupon payments between now and six months, since coupons are paid semiannually. Using equation (2), the price of this bond is given as:

P (0.5) "

C0.5 # F0.5 e y (0.5)0.5

(16)

where F0.5 is the face value of the bond payable at the maturity of 0.5 years, C0.5 is the semiannual coupon payment at the maturity, and y(0.5) is the annualized six-month zero-coupon yield (under continuously-compounding). The six-month zero-coupon yield can be calculated by taking logarithms of both sides of equation (16), and simplifying as follows:

15

y (0.5) "

1 1 F0.5 # C0.5 2 ln 4 0.5 3 5 P (0.5) 6

(17)

In order to compute the 1-year zero-coupon yield, we can use the price of a 1- year coupon bond as follows:

P (1) "

1 y (0.5)0.5

F1 # C1 e y (1)

(18)

where F1 is the face value of the bond payable at the bonds 1-year maturity, C1 is the semiannual coupon, which is paid at the end of 0.5 years and 1 year, and y(1) is the annualized 1year zero-coupon yield. By rearranging the terms in equation (18) and taking logarithms, we get the 1-year zero-coupon yield as follows:

1 3 F1 # C1 y (1) " ln 3 C1 3 P (1) ! y (0.5)0.5 e 5

2 4 4 4 6

(19)

Since we already know the six-month yield, y(0.5) from equation (17), this can be substituted in equation (19) to solve for the 1-year yield. Now, continuing in this manner, the sixmonth yield, y(0.5), and the 1-year yield, y(1), can be both used to obtain the 1.5-year yield, y(1.5), given the price of a 1.5-year maturity coupon bond. Following the same approach, the zero-coupon yields of all of the K maturities (corresponding to the maturities of the bonds in the sample) are computed iteratively using the zero-coupon yields of the previous maturities. The zero-coupon yields corresponding to the maturities that lie between these K dates can be computed by using linear or quadratic interpolation. Generally, about 15 to 30 bootstrapping maturities are sufficient in producing the whole term structure of zero-coupon

16

yields. Instead of solving the zero-coupon yields sequentially using an iterative approach as shown above, the following matrix approach can be used for obtaining a direct solution. Consider K bonds maturing at dates t1, t2, , tK, and let CFit be the total cash flow payments of the ith (for i= 1,2,3,,K) bond on the date t (for t = t1, t2, , tK). Then the prices of the K bonds are given by the following system of K simultaneous equations:

& P (t1 ) ' & CF1t1 ( ) ( CF ( P (t 2 ) ) " ( 2t1 ( " ) ( " ( ) ( ( ) P t * K + ( * CFKt1

0 CF2t2 " CFKt2

0 ' & d (t ) ' )( 1 ) 0 ) d (t 2 ) ! ( ) ) ( ) " # " )( ) d (t K ) + ! CFKtK ) * + !

(20)

Note that the upper triangle of the cash flow matrix on the right hand side of equation (20) has zero values. By multiplying both sides of equation (20) by the inverse of the cash flow matrix, the discount functions corresponding to maturities t1, t2,...,tK can be computed as follows:

& d (t1 ) ' & CF1t1 ( ) ( CF ( d (t 2 ) ) " ( 2t1 ( " ) ( " ( ) ( ( ) d t * K + ( * CFKt1

0 CF2t2

"
CFKt2

0 ' & P (t ) ' ) ( 1 ) 0 ) P (t 2 ) ! ( ) ) ( " ) # " ) ( ) P (t K ) + ! CFKtK ) * +

!1

(21)

The above solution requires that the number of bonds equal the number of cash flow maturity dates. The zero-coupon rates can be computed from the corresponding discount functions using equation (1).

Example 3 This example demonstrates the bootstrapping method using the ten bonds given in 17

Table 2. For expositional simplicity all bonds are assumed to make annual coupon payments. Table 2 Bond data for bootstrapping method Bond # Price Maturity (years) 1 2 3 4 5 6 7 8 9 10 Annual coupon rate (%) 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5

1 2 3 4 5 6 7 8 9 10

$96.60 $93.71 $91.56 $90.24 $89.74 $90.04 $91.09 $92.82 $95.19 $98.14

Bond 1s price is given as follows:

96.60 " (2 # 100) e ! y (1)%1

which gives the 1-year zero-coupon yield as:

1 100 # 2 2 y (1) " ln 3 " 0.05439 " 5.439% 5 96.60 4 6

Using the 1-year yield of 5.439% to discount the first coupon payment from the 2-year bond, the price of the 2-year bond is given as:

93.71 " 2.5 % e !0.05439%1 # (2.5 # 100) e ! y (2)%2

18

Solving for y(2), we get,

11 2.5 # 100 2 " 0.05762 " 5.762% y (2) " 3 !0.05439%1 4 2 5 93.71 ! 2.5 % e 6

(22)

Following this iterative procedure, the zero-coupon rates for all ten maturities can be computed. More directly, we can use the matrix solution given by equation (21), as follows:

0 0 0 0 0 0 0 0 0 ' & d (1) ' & 102 ( ) ( ) 0 0 0 0 0 0 0 ) ( d (2) ) ( 2.5 102.5 0 ( d (3) ) ( 3 3 103 0 0 0 0 0 0 0 ) ( ) ( ) 0 0 0 0 0 ) ( d (4) ) ( 3.5 3.5 3.5 103.5 0 ( d (5) ) ( 4 4 4 4 104 0 0 0 0 0 ) ( )"( ) 0 0 0 ) ( d (6) ) ( 4.5 4.5 4.5 4.5 4.5 104.5 0 ( d (7) ) ( 5 5 5 5 5 5 105 0 0 0 ) ( ) ( ) 0 ) ( d (8) ) ( 5.5 5.5 5.5 5.5 5.5 5.5 5.5 105.5 0 ( d (9) ) ( 6 6 6 6 6 6 6 6 106 0 ) ( ) ( ) ( d (10) ) ( 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5 106.5 ) * + * +

!1

& 96.60 ' ( ) ( 93.71 ) ( 91.56 ) ( ) ( 90.24 ) ( 89.74 ) ( ) ( 90.04 ) ( 91.09 ) ( ) ( 92.82 ) ( 95.19 ) ( ) ( 98.14 ) * +

Multiplying the two matrices gives the solution as:

19

& d (1) ' & 0.947 ' ( ) ( ) ( d (2) ) ( 0.891 ) ( d (3) ) ( 0.835 ) ( ) ( ) ( d (4) ) ( 0.781 ) ( d (5) ) ( 0.730 ) ( )"( ) ( d (6) ) ( 0.681 ) ( d (7) ) ( 0.636 ) ( ) ( ) ( d (8) ) ( 0.593 ) ( d (9) ) ( 0.553 ) ( ) ( ) ( d (10) ) ( 0.516 ) * + * +

The zero-coupon rates are obtained from the corresponding discount functions by the following relationship derived from equation(1):

y (t ) "

! ln d (t ) t

(23)

The zero-coupon rates are displayed in Figure 3. The points between the estimated zero-coupon yields are obtained by simple linear or quadratic interpolation, and thus, the whole term structure of zero-coupon rates is obtained.

7.0 Zero-coupon rates (%) 6.39 6.47 6.62 6.53 6.58

6.5 6.00 6.0 5.44 5.5 5.76

6.16

6.29

5.0 0 2 4 6 8 10 Term (years)

20

Figure 3

Zero-coupon rate estimated using the bootstrapping method

The bootstrapping method has two main limitations. First, since this method does not perform optimization, it computes zero-coupon yields that exactly fit the bond prices. This leads to over-fitting since bond prices often contain idiosyncratic errors due to lack of liquidity, bid-ask spreads, special tax effects, etc., and hence, the term structure will not be necessarily smooth as shown in Figure 3. Second, the bootstrapping method requires ad-hoc adjustments when the number of bonds is not the same as the bootstrapping maturities, and when cash flows of different bonds do not fall on the same bootstrapping dates. For example, when two or more bonds mature on the same bootstrapping maturity, the estimated spot rates resulting from using each of these bonds are usually averaged. In the opposite case, when no bond exists at a required bootstrapping maturity, a common practice is to estimate a par yield curve (that is, the yield to maturities of bond priced at par) using simple regression models that make the yields to maturity on current bonds depend on a series of bond characteristics including the coupon rate and the time to maturity. Then, the yields on par bonds are estimated by assuming that the coupon rate of each bond equals its yield to maturity. The next two methods overcome these difficulties by imposing specific functional forms on the term structure.

Cubic-spline method

Consider the relationship between the observed price of a coupon bond maturing at time tm, and the discount function. Using equations (1) and (3), the price of this bond can be expressed as:

P (t m ) " $ CFj / d (t j ) # 7
j "1

(24)

where CFj is the total cash flow from the bond (i.e., coupon, face value, or both) on date tj (j = 1,2,,m). Since bond prices are observed with idiosyncratic errors, we need to estimate some functional form for the discount function that minimizes these errors. We face two problems in 21

doing this. First, the discount functions may be highly non-linear, such that we may need a high-dimensional function to make the approximation work. Second, the error terms in equation (24) may increase with the maturity of the bonds, since longer maturity bonds have higher bid-ask spreads, lower liquidity, etc. Due to this heteroscedasticity of errors, estimation of the discount function using approaches such as least squares minimization, generally fits well at long maturities, but provides a very poor fit at short maturities (see McCulloch [1971] and Chambers Carleton and Waldman [1984]). The cubic-spline method addresses the first issue by dividing the term structure in many segments using a series of points that are called knotpoints. Different functions of the same class (polynomial, exponential, etc.) are then used to fit the term structure over these segments. The family of functions is constrained to be continuous and smooth around each knot point to ensure the continuity and smoothness of the fitted curves, using spline methods. McCulloch pioneered the application of splines to term structure estimation by using quadratic polynomial splines in 1971 and cubic polynomial splines in 1975. The cubic spline method remains popular among practitioners and is explained below. Consider a set of K bonds with maturities of t1, t2,..., tK. years. The range of maturities is divided into s-2 intervals defined by s-1 knot points T1, T2,..., Ts-1, where T1=0 and Ts-1=tK. A cubic polynomial spline of the discount function d(t) is defined by the following equation:

d (t ) " 1 # $ 8 i g i ( t )
i "1

(25)

where g1(t), g2(t),..., gs(t) define a set of s basis piecewise cubic functions and 81,..., 8s are unknown parameters that must be estimated. Since the discount factor for time 0 is 1 by definition, we have:

g i (0) " 0

i " 1,2,$, s

(26)

The continuity and smoothness of the discount function within each interval is ensured 22

by the polynomial functional form of each gi(t). The continuity and smoothness at the knotpoints is ensured by the requirement that the polynomial functions defined over adjacent intervals (Ti1,Ti)

and (Ti,Ti+1) have a common value and common first and second derivatives at Ti. The

above constraints lead to the following definitions for the set of basis functions g1(t), g2(t),..., gs(t):

Case 1: i 9 s :0 ; ; ; (t ! Ti !1 )3 ; ; 6(Ti ! Ti !1 ) ; ; g i (t ) " = 2 2 3 ; (Ti ! Ti !1 ) # (Ti ! Ti !1 )(t ! Ti ) # (t ! Ti ) ! (t ! Ti ) ; 6 2 2 6(Ti #1 ! Ti ) ; ; ; 2Ti #1 ! Ti ! Ti !1 t ! Ti #1 ' ;(Ti #1 ! Ti !1 ) & # ( ) 6 2 + ; * ? Case 2: i " s g i (t ) " t t 9 Ti !1 Ti !1 < t 9 Ti

Ti < t 9 Ti #1

t > Ti #1 (27)

Substituting equation (25) into equation (24), we can rewrite the price of the bond maturing at date tm as follows:

m s & ' P (t m ) " $ CFj ( 1 # $8 i g i (t j ) ) # 7 j "1 * i "1 +

(28)

By rearranging the terms, we obtain:

23

P (t m ) ! $ CFj " $8 i $ CFj g i (t j ) # 7

j "1 i "1 j "1

(29)

The estimation of the discount function requires searching of the unknown parameters, 81, 82,,8s, that minimizes the sum of squared errors across all bonds. Since equation (29) is linear with respect to the parameters 81, 82,,8s, this can be achieved by an ordinary least squares (OLS) regression.5 The above approach uses s 2 number of maturity segments, s-1 number of knotpoints, and s number of cubic polynomial functions. An intuitive choice for the maturity segments may be short-term, intermediate-term, and long-term, which gives three maturity segments of 0 to 1 years, 1 to 5 years, and 5 to 10 years, four knot points given as, 0, 1, 5, and 10 years, and five cubic polynomial functions. McCulloch recommends choosing knotpoints such that there are approximately equal number of data points (number of bonds maturities) within each maturity segment. Using this approach, if the bonds are arranged in ascending order of maturity, i.e., t1< t2 < t3... < tK, then the knot points are given as follows:

:0 ; Ti " =t h # @ (t h #1 ! t h ) ;t ?K

i "1 2< i < s!2 i " s !1

(30)

where h is an integer defined as:

Other functions can be optimized. McCulloch, indeed, proposed to weight the errors by the inverse of the buy-

ask spread. This, however, precludes the use of OLS techniques.

24

1 A i ! 1B K 2 h " INT 3 4 5 s !2 6

(31)

and the parameter @ is given as:

@"

A i ! 1B K ! h
s!2

(32)

McCulloch also suggests that the number of basis functions may be set to the integer nearest to the square root of the number of observations, that is:

2 s " Round 1 5 K6

(33)

This choice of s has two desired properties. First, as the number of observations (bonds) increases, the number of basis functions increases. Second, as the number of observations increases, the number of observations within each interval increases, too.

Example 4 Consider the fifteen bonds shown in Table 3. This set includes five more bonds with maturities ranging from 11 to 15 years in addition to the ten bonds given in Example 3.

25

Table 3

Bond data for cubic-spline method Bond # Price Maturity (years) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Annual coupon rate (%) 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 8 8.5 9

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

$96.60 $93.71 $91.56 $90.24 $89.74 $90.04 $91.09 $92.82 $95.19 $98.14 $101.60 $105.54 $109.90 $114.64 $119.73

According to the McCulloch criterion, the number of basis function is given as:

1 15 2 " 4 s " Round 5 6

which implies that the number of knot points is s 1= 3, and the number of intervals for the maturity range that extends from 0 to 15 years is s 2 = 2. According to equation (30), the first knot point, T1 = 0, and the last knot point T3=15 years. The second knot point, T2, is obtained as follows:

T2 " t h # @ (t h #1 ! t h )

(34)

where using equations (31) and (32), h and @ are given as: 26

1 A i ! 1B K 2 1 A 2 ! 1B % 15 2 h " INT 3 4 " INT 3 4 " INT [7.5] " 7 , and 5 s !2 6 5 4!2 6

@"

A i ! 1B K ! h " A 2 ! 1B % 15 ! 7 " 0.5

s !2 4!2

Also, using Table 3, t7 = 7 and t8 = 8. Substituting the values of h, @, t7, and t8, given above in equation (34), we get,

T2 " 7 # 0.5 % (8 ! 7) " 7.5

Hence, the three knot points are T1 = 0, T2 = 7.5, and T3=15. The three knot points divide the maturity spectrum into two segments, 0 to 7.5 year and 7.5 years to 15 years. The number of basis functions is given as s = 4. Using equation (27), these functions are given as follows6:

For calculation, note that the basis functions are continuous at the knotpoints and T1=0.

27

:t2 t 3 ; 2 ! 6T 2 ; ; g1 ( t ) " = ; T2 t ! T2 ' ;T2 & # ( ) ; *3 2 + ?

T1 < t 9 T2

t > T2

: t3 ; 6T 2 ; ; g 2 (t ) " = ; T 2 T ( t T ) (t T ) 2 ( t T ) 3 ; 2 # 2 ! 2 # ! 2 ! ! 2 2 2 6(T3 ! T2 ) ; ?6

T1 < t 9 T2

T2 < t < T3

:0 ; ; g 3 (t ) " = 3 ; (t ! T2 ) ; ? 6(T3 ! T2 )

t 9 T2 T2 < t < T3

g 4 (t ) " t

for all t

The shapes of these basis functions are shown in Figure 4. The vertical line divides the maturity range into two segments. As can be seen these basis functions ensure continuity and smoothness at the knot points. The discount function is given as a linear weighted average of the basis functions as follows:

d (t ) " 1 # 81g1 (t ) # 8 2 g 2 (t ) # 8 3 g 4 (t ) # 8 4 g 4 (t )

28

60 50 40 30 20 10 0 0 1 2 3 4 5 6 7 g2(t) 8 9 10 11 12 13 14 15 g4(t)

g1(t)

g3(t)

Figure 4

Basis functions of the discount function

The parameters 81, 82, 83, and 84 are estimated using an OLS linear regression model given as:

m 1m 2 1m 2 P (t m ) ! $ CFj " 81 3 $ CFj g1 (t j ) 4 # 8 2 3 $ CFj g 2 (t j ) 4 j "1 5 j "1 6 5 j "1 6 m m 1 2 1 2 +8 3 3 $ CFj g 3 (t j ) 4 # 8 4 3 $ CFj g 4 (t j ) 4 # 7 5 j "1 6 5 j "1 6

Using the bond data in Table 3, the estimated values of the parameters are:

81 " !0.00035 8 3 " 0.00095

8 2 " 0.00347 8 4 " !0.05501

29

The discount function is obtained using the above four parameter values and the four basis functions. The zero-coupon rates are obtained from the discount function using equation (23), and are displayed in Figure 5. The points in the curve correspond to the maturities of the bonds used in the estimation and can be compared with those of the bootstrapping method in Example 3.

7.0 Zero-coupon rates (%)

6.47 6.55

6.72 6.76 6.65 6.68 6.70 6.60 6.63

6.5
6.1 3 5.98

6.38 6.26

6.0
5.67

5.83

5.5

5.0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Term (years)

Figure 5

Estimated zero-coupon rates using cubic-spline method

A potential criticism of the cubic-spline method is the sensitivity of the discount function to the location of the knot points. Different knot points result in variations in the discount function, which can be sometimes significant. Also, too many knot points may lead to overfitting of the discount function. So, one must be careful in the selection of both the number and the placing of the knot points. Another shortcoming of cubic-splines is that they give unreasonably curved shapes for the term structure at the long end of the maturity spectrum, a region where the term structure must have very little curvature. Additionally, the OLS regression used for the estimation of the parameters in equation (29), gives the same weights to the price errors of the bonds with heterogeneous characteristics, such as liquidity, bid-ask spreads, maturity, etc. Other functions can be used

30

for optimization to overcome this limitation but at the cost of precluding the use of OLS techniques7. Finally, the choice of polynomials as basis functions is also controversial. It is argued that the shape of the discount function estimated using cubic splines is usually reasonable up to the maturity of the longest bond in the dataset but tend to be positive or negative infinity when extrapolated to longer terms. This implies that it is possible to generate unbounded positive or negative interest rates. Moreover, although the use of polynomial splines moderates the wavy shape of simple polynomials around the curve to be fitted, this shape might not disappear completely and hence, the fitted discount function might wave around the real discount function introducing a significant variability in both spot and forward rates. Despite these shortcomings, the use of polynomial splines to estimate the TSIR is widespread in the financial industry. In fact, the superiority of exponential splines over polynomial splines has not been clearly established.

Nelson and Siegel Model

An alternative approach that overcomes many of the shortcomings of spline techniques is the methodology of Nelson and Siegel. The Nelson and Siegel [1987] model uses a single exponential functional form over the entire maturity range. Nelson and Siegel suggest a parsimonious parameterization of the instantaneous forward rate curve given as follows:

In fact, there are many alternative error-weighing schemes which might lead to more robust estimates of the term

structure. For example, Bliss [1997] suggests weighting each bond price error by the inverse of the bonds duration as a way to improve the fitting of long interest rates, which might be poor. This is due to the fact that in absence of a weighting scheme for pricing errors, the quality of the fit of the term structure decreases with maturity. To understand this, consider the relationship between prices, yields and maturities. A same change in price implies a much greater change in yield in short-term bonds compared to long-term bonds. Therefore, following a price error minimization criterion in the estimation will make interest rates corresponding to long-term bonds to be over-fitted at the expense of shorter-term interest rates.

31

f (t ) " 81 # 8 2e !t / C1 # 8 3e ! t / C2

(35)

Finding the above model to be over-parameterized, Nelson and Siegel consider a special case of this model given as:8

f (t ) " 8 1 # 8 2 e ! t / C # 8 3

e !t / C

(36)

The zero-coupon rates consistent the forward rates given by the above equation can be solved using equation (15), as follows:

y ( t ) " 8 1 # A8 2 # 8 3 B

A1 ! e B ! 8 e t
!t / C 3

!t / C

(37)

The Nelson and Siegel model is based upon four parameters. These parameters can be interpreted as follows: , ,

81+82 is the instantaneous short rate, i.e., 81+82 = y(0) = f(0). 81 is the consol rate. It gives the asymptotic value of the term structure of both the zerocoupon rates and the instantaneous forward rates, i.e., 81 = y(D) = f(D).

The spread between the consol rate and the instantaneous short rate is 82, which can be interpreted as the slope of the term structure of zero-coupon rates as well as the term structure of forward rates.

The new function resembles a constant plus a Laguerre function, which consists of polynomials multiplied by

exponentially decaying terms. This indicates a method for generalization to higher-order models.

32

83 affects the curvature of the term structure over the intermediate terms. When 83 > 0,
the term structure attains a maximum value leading to a concave shape, and when 83 < 0, the term structure attains minimum value leading to a convex shape.

C > 0, is the speed of convergence of the term structure towards the consol rate. A lower C value accelerates the convergence of the term structure towards the consol rate, while a
higher C value moves the hump in the term structure closer to longer maturities. Figure 6 illustrates how the parameters 81, 82, and 83, affect the shape of the term

structure of zero-coupon rates (given a constant C = 1). A change in 81 can be interpreted as the height or parallel change, a change in 82 can be interpreted as the slope change (though this parameter also affects the curvature change slightly), and a change in 83 can be interpreted as the curvature change in the term structure of zero-coupon rates.

1.2 Zero-coupon rate changes 1.0 0.8 0.6 0.4 0.2 0.0 Term

E8F

E8G

E8H

Figure 6

Influence of the alpha parameters of Nelson and Siegel on the term structure of

zero-coupon rates

Figure 7 demonstrates that Nelson and Siegel method is consistent with a variety of term structure shapes, including monotonic and humped, and allows asymptotic behavior of

33

forward and spot rates at the long end. For illustrative purposes, the consol and instantaneous rates have been set at the same level.

Zero-coupon rates

E8HI

Term

14.0 13.0 12.0 Zero-coupon rates 11.0 10.0 9.0 8.0 7.0 6.0 5.0 Term

ECI

Figure 7

Influence of the curvature and hump positioning parameters of Nelson and

Siegel on the spot curve

The discount function associated with the term structures in (36) and (37) is given as:

d (t ) " e

!81t ! C A8 2 #8 3 B 1!e ! t / C #8 3te ! t / C

(38)

34

Substituting this functional form into the pricing formula for a coupon-bearing bond, we have:

P (t m ) " $ CFj e
j "1

!81t j ! C A8 2 #8 3 B 1! e

! t j /C

B#8 t e
3 j

! t j /C

(39)

where tm is the bonds maturity and CFj is the cash flow of the bond at time tj. The parameters in equation (39) can be estimated by minimizing the sum of squared errors, that is:

81 ,8 2 ,8 3 ,C

Min

$7
i "1

2 i

(40)

subject to the following constraints:

81 J 0 81 # 8 2 J 0 C J0

(41)

where 7i is the difference between the ith bonds market price and its theoretical price given by equation (39). The first constraint in equation (41) requires that the consol rate remain positive; the second constraint requires that the instantaneous short rate remain positive; finally, the third constraint ensures the convergence of the term structure to the consol rate. Since the bond pricing equation (39) is a non-linear function, the four parameters are estimated using a non-linear optimization technique. As non-linear optimization techniques are usually sensitive to the starting values of the parameters, these values must be carefully chosen. Despite this computational difficulty, the Nelson and Siegel model, and its extended version given by Svensson [1994], have a prominent position among term structure estimation 35

methods. The smoothness of the estimated curves for both spot rates and forward rates, the asymptotic behavior of the term structure over the long end, and their robustness to outliers and errors in market data are the main advantages these methods compared to spline methods. In fact, as reported in BIS [2005], most Central Banks use these methods for term structure estimation. Also, in recent years, these models are attracting the interest of researchers in the area of interest modelling and portfolio risk management. Matzner-Lber and Villa [2004] and Diebold and Li [2006], for example, reinterpret them as modern threefactor models of level, slope and curvature factors in the most pure tradition of Litterman and Scheinkman [1991] and Bliss [1997] and obtain empirical evidence in favor of them.

Example 3.5 Reconsider the bond data in Table 3. The initial values of the parameters may be guessed using some logical approximations. For example, the starting value for the parameter

81, which indicates the asymptotic value of the term structure of zero-coupon rates, may be set
as the yield to maturity of the longest bond in the sample. In Table 3, the longest bond is Bond 15, whose continuously-compounded yield-to-maturity can be calculated using the solver function in excel and is given as 6.629%.9 The starting value for the parameter 82, which is the difference between the instantaneous short rate and the consol rate, may be set as the difference between the yields to maturity of the shortest maturity bond and the longest maturity bond in the sample. In Table 3, this corresponds to the difference between the yields to maturity of Bond 1 and Bond 15, given as 5.439-6.629 = 1.19%. The starting values of the other two parameters, 83 and C, which are associated with the curvature of the term structure and the speed of convergence towards the consol rate,

Generally, the yield to maturity is reported using discrete compounding. This yield may be used as the starting

value too, since it is quite close in value to its continuously-compounded counterpart.

36

respectively, are difficult to guess. Therefore, it might be necessary to consider a grid of different starting values. The feasible range for these parameters for realistic term structure data are 10 < 83 < 10, and C ranging from the shortest maturity and the longest maturity in the sample of bonds. Hence, using Table 3, 1 < C < 15. Using the data in Table 3, we obtain the following values of the parameters using nonlinear optimization:

81 " 0.07000 8 3 " 0.00129

8 2 " !0.01999 C " 2.02881

The corresponding term structure of zero-coupon rates is shown in Figure 8.

7.0 Zero-coupon rates (%)

6.47

6.75 6.68 6.71 6.73 6.62 6.66 6.53 6.58

6.5
6.1 7 5.99

6.29

6.39

6.0
5.44

5.76

5.5

5.00

5.0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Term (years)

Figure 8

Estimated zero-coupon rates using Nelson and Siegel method

5.

Software description and download

In this section we briefly describe the Excel files that accompany this guide. They

allow readers with basic Excel skills to instantly play with term structures and the methods described in this guide. 37

There are two files associated to this guide, named TSIRmovie.xls and TSIR.xls. They are fully operative and can be downloaded from the website: www.fixedincomerisk.com. See the Guides tab.

5.1.

TSIRmovie.xls

This file gives a visual movie illustration of the evolution of the U.S. term structure of interest rates between December 1946 and February 1991 based upon McCulloch and Kwon [1993] term structure data.

The above picture gives a snapshot of the excel file. As can be seen, there are three boxes with options and controls and a graph showing different term structures. The first box in grey area in the top-left corner allows the user to choose one or more term structures among the zero-coupon yield curve (labeled as TSIR), the par-bond yield curve (PAR) and the instantaneous forward rate curve (FWD). The second box in grey area to the top-right corner gives the option of including in the graph an additional curve for comparative purposes. This curve is represented by a dashed line in the graph and will not change during the movie. Choosing the option None imply that no comparative curve will be shown. The option A 38

specific month shows the curve of the month entered by the user. The option A specific year shows the average curve of the year chosen. Finally, the yellow area in the bottom-left is the box of controls for the movie. The user must first enter the starting month for the movie. Then, the user can move forward or backward in time to see how the term structure changes dynamically in the graph using the spinner control or view the entire movie from the starting month to the last month in the dataset by pressing the button View Movie. In this latter case, when the box Slow motion is checked, the movie goes at a slower speed.

5.2.

TSIR.xls

This file allows estimating the term structure of interest rates from a set of bond data and then using the estimated curves to price bonds. The file has two sheets, one for each purpose, titled as TSIR Estimation and Bond Pricing, respectively. The picture below gives a snapshot of the sheet TSIR Estimation. The sheet includes a graph showing the estimated term structures and four areas of data labelled as Input bond data, Input terms for the graph, Cubic spline and Exponential form. In the top of the sheet, the button Clear all when pressed deletes all the data in the sheet.

39

In the first two areas, whose titles are in a yellow font, the user must enter the input data for term structure estimation. The blue arrows in the different columns of these two areas allow sorting the data either in ascending or descending order. The area Input bond data must include the list of bonds and their characteristics. The maximum number of bonds is 100. For each bond, the user must enter an alphanumeric code or name, the cash (or dirty) price in dollars, the face value, the coupon rate in annual terms, the number of coupons the bond pays a year and the bonds maturity in years. If bond characteristics are incomplete or wrong, the user will receive an error message during the estimation. Also, bonds lacking a code are discarded from the estimation if they are placed at the end of the list. The column Input terms for the graph must be filled with the terms that will constitute the estimated term structure of interest rates. Once the required data has been entered, the user can press the buttons Estimate 40

under Cubic spline or/and Exponential form to obtain the estimated interest rates for the terms specified by the user using McCullochs cubic splines or Nelson and Siegels method, respectively. The estimation could take some time, especially in the case of Nelson and Siegels method due to the non-linear optimization. The original data in the sheet allows the user to obtain the term structures shown in this guide for both methods. The snapshot of the second sheet labelled Bond Pricing is shown bellow. The sheet consists on a graph showing the bond prices obtained from the estimated term structures, an input area titled Input bond data and an output area titled Calculate bond price.

The bond data to be filled in the first area is the same described above with the only 41

exception of bond prices, not needed in this case. The estimated prices of the bonds with the specified characteristic are shown in the area Calculate bond price once the button Spline or Exponential is pressed. The bond prices under the button Spline use the term structure obtained using McCullochs cubic splines and those under Exponential use the term structure estimated using Nelson and Siegels method. The button Clear all at the top of the sheet delete all the data in the sheet.

42

References

BIS, 2005, Zero-coupon yield curves: Technical documentation, Monetary and Economic Department, BIS Papers 25, October 2005, Bank for International Settlements. Bliss, R.R., 1997, Movements in the Term Structure of Interest Rates, Economic Review, FRB of Atlanta, fourth quarter, 16-33. Chambers, D.R., W.T. Carleton and D.W. Waldman, 1984, A New Approach to Estimation of the Term Structure of Interest Rates, Journal of Financial and Quantitative Analysis 19(3), 233-252. Diebold, F.X., Ji, L. and Li, C., 2006, A Three-Factor Yield Curve Model: Non-Affine Structure, Systematic Risk Sources, and Generalized Duration, in L.R. Klein (ed.), Long-Run Growth and Short-Run Stabilization: Essays in Memory of Albert Ando. Cheltenham, U.K.:Edward Elgar, 240,274. Fisher, M., D. Nychka and D. Zervos, 1995, Fitting the Term Structure of Interest Rates with Smoothing Splines, Working Paper 95-1, Finance and Economic Discussion Series, Federal Reserve Board. Jarrow, R., D. Ruppert and Y. Yu, 2004, Estimating the Term Structure of Corporate Debt with a Semiparametric Penalized Spline Model, Journal of the American Statistical Association 99, 57-66. Litterman, R., Scheinkman J., 1991. Common factors affecting bond returns, Journal of Fixed Income, June, 54-61. Matzner-Lber, E. and Villa, C., 2004, Functional Principal Component Analysis of the Yield Curve, International Conference AFFI 2004, France. McCulloch, J. H. and H.C. Kwon, 1993, U.S. Term Structure Data 1947-1991, Ohio State University, Working Paper, 93-96. McCulloch, J.H., 1971, Measuring the Term Structure of Interest Rates, Journal of Business 44, 1931. McCulloch, J.H., 1975, The Tax Adjusted Yield Curve, Journal of Finance 30, 811830. Nawalkha, Sanjay K., Gloria M. Soto, and Natalia A. Beliaeva, 2005, Interest Rate Risk Modeling: The Fixed Income Valuation Course, Wiley Finance, John Wiley and Sons, NJ. Nawalkha, Sanjay K., Natalia A. Beliaeva, and Gloria M. Soto, 2007, Dynamic Term Structure Modeling: The Fixed Income Valuation Course, Wiley Finance, John Wiley and Sons, NJ. Nelson, C.R. and A.F. Siegel, 1987, Parsimonious Modeling of Yield Curves, Journal of Business 60(4), 473-489. 43

Steeley, J. M., 1991, Estimating the Gilt-Edged Term Structure: Basis Splines and Confidence Intervals, Journal of Banking, Finance and Accounting 18(4), 513-529. Svensson, L.E.O., 1994, Estimating and Interpreting Forward Interest Rates: Sweden 19921994, Institute for International Economic Studies. Vasicek, O.A. and H.G. Fong, 1982, Term Structure Modeling Using Exponential Splines, Journal of Finance 37(2), 339-348.

44