IEOR E4706: Financial Engineering: Discrete-Time Asset Pricing c 2005 by Martin Haugh

Fall 2005

Deterministic Cash Flows and the Term Structure of Interest Rates

Basic Theory of Interest
Cashow Notation: We use (c0 , c1 , . . . , ci , . . . , cn ) to denote the series of cashows where ci is received at time t = i . The length of a period, i.e. the interval of time between t = i and t = i + 1, will usually be understood from the context. Negative cashows refer to cash payments. The initial cashow, c0 , will often be negative while the remaining cashows are positive. This situation models the cashows of many securities such as stocks and bonds. In such circumstances, c0 denotes the cost of the security, while later cashows refer to dividends, coupons or sale receipts which are all positive.

Principal and Interest

Denition 1 If A is invested in an account for n periods with a simple interest rate of r per period, then after n periods the account will be worth A(1 + rn). Denition 2 If A is invested in an account for n periods with a compound interest rate of r per period, then after n periods the account will be worth A(1 + r)n . Interest rates are usually quoted on an annualized basis, even if the compounding period is less than a year. For example, the phrase 10% interest, compounded quarterly implies that an investment of A will be worth A(1 + .1/4)4 one year later. Similarly, the phrase 10% interest, compounded semi-annually implies that an investment of A will be worth A(1 + .1/2)2 after one year. In general, if there are n compounding periods per year and the interest rate is r%, then an investment of A will be worth V = A(1 + r/n)mn after m years. Denition 3 Continuous compounding refers to the situation where we let the length of the compounding period go to 0. That is, after m years we see that an investment of A will be worth
n

lim A(1 + r/n)mn = Aerm .

Denition 4 The eective interest rate is that rate which would produce the same result if compounding were done per year rather than per period. So if the length of a compounding period is one year, the eective interest rate is the same as the quoted or nominal rate. For example, A invested for 1 year at 10% interest, compounded quarterly, will be worth 1.1038A and so the eective interest rate is 10.38%.

Present and Future Values of Cash Flow Streams

The value of $1 today is clearly not the same as the value of $1 next year. However, the simple concepts of present value and future value allow us to fairly compare cash ows that occur at dierent dates. For example

Deterministic Cash Flows and the Term Structure of Interest Rates $1 invested today at t = 0 would be worth $(1 + r) next year at t = 1 assuming1 annual compounding. So for r > 0, we can conclude that $1 at t = 0 is worth more2 than $1 at t = 1.

We can then reverse the argument to say that $(1 + r) at t = 1 is worth $1 at t = 0. That is, the present value of $(1 + r) at t = 1 is $1. We say that we are discounting the cash ow at t = 1 back to t = 0. Likewise the future value at t = 1 of $1 at t = 0 is $(1 + r). More generally, the present value of the cashow, (c0 , c1 , . . . , cn ), is P V = c0 + c1 c2 cn + + ... + 1 + r (1 + r)2 (1 + r)n

where we have assumed compounding is done per period. Likewise, the future value (at t = n) of the cashow is F V = P V (1 + r)n = c0 (1 + r)n + c1 (1 + r)n1 + c2 (1 + r)n2 + . . . + cn . You must be careful to observe the compounding convention. For example, if cashows occur yearly, interest rates are quoted on an annual basis (as usual) but are compounded m times per year then P V = c0 + and F V = P V (1 + r/m)nm . Note that the individual cash ows, ci , may be positive or negative. c1 c2 cn + + ... + (1 + r/m)m (1 + r/m)2m (1 + r/m)nm

Internal Rate of Return (IRR)

Denition 5 Given a cash ow stream (c0 , c1 , . . . , cn ), the internal rate of return (IRR) is a number, r, that satises c1 cn c2 0 = c0 + + ... + . (1) + 2 1 + r (1 + r) (1 + r)n

In most contexts, c0 will be negative and we can interpret it as the price paid today for the cash ow stream, (c1 , . . . , cn ). Then r is just the number that makes this a fair deal. If we let z = 1/(1 + r) we can see that (1) is a polynomial in degree n. In general, (1) could then have zero, one or as many as n real roots so it is not always clear what the appropriate value of r is. (But see the following exercise.) Exercise 1 Show that when c0 < 0 and ck 0 for all k 1, there exists a unique positive root, z , to the equation 0 = c0 + c1 z + c2 z 2 + . . . + cn z n . Furthermore, if
n k=0 ck

> 0, then show the corresponding IRR, r = 1/z 1, is positive.

Even when we know that there exists a unique positive solution to (1), we usually have to solve for it numerically.
1 The compounding convention should always be observed so that if we are using quarterly compounding, for example, then $1 invested at t = 0 would be worth $(1 + r/4)4 at t = 1. 2 This statement assumes that there will be no ination over the next year. If there is ination then we should use real interest rates before concluding that $1 today is worth more than $1 next year; see Luenberger, Section 2.6. We will not consider ination or related issues in this course.

Deterministic Cash Flows and the Term Structure of Interest Rates

Project Evaluation Criteria: NPV versus IRR

Many investment problems may be formulated as the problem of optimally allocating funds among a number of competing projects. The word project is signicant as it usually implies that the projects cannot be scaled up or down. For example, a project may represent developing an oil eld or introducing a new product to a market and the level of investment required for such projects is usually xed3 . These investment allocation problems are often called capital budgeting problems and they constitute an important4 topic in corporate nance. When faced with a capital budgeting problem, we need to be able to evaluate the competing projects so that the appropriate decision can be made. There are a number of dierent criteria available for evaluating projects but the two most common are Net Present Value (NPV) and Internal Rate of Return (IRR). The NPV criterion amounts to selecting the project with the highest present value5 while the IRR criterion amounts to selecting the project with the highest IRR. It is widely believed by theorists that the NPV criterion, when correctly applied, is the correct criterion. However, conicting recommendations can often be reconciled6 when common sense is applied. A particular diculty that arises when applying the NPV criterion is that of choosing the correct discount factor. This problem does not arise when cash ows are deterministic in which case the appropriate risk-free interest rates should be used for discounting. More generally, however, it is not always clear how to discount cash ows that are stochastic since the riskiness of the cashows also needs to be considered. This is an important problem in nancial economics and nancial engineering. In fact, we shall see later in the course that the problem of pricing derivative securities can also be viewed as one of choosing the appropriate stochastic discount factor. When markets are complete7 , it turns out that there is only one stochastic discount factor that can be used.

Further Examples and Applications

Example 1 (Sunk Costs (Exercise 2.6 in Luenberger)) A young couple has made a nonrefundable deposit of the rst months rent (equal to $1, 000) on a 6-month apartment lease. The next day they nd a dierent apartment that they like just as well, but its monthly rent is only $900. They plan to be in the apartment only 6 months. Should they switch to the new apartment? What if they plan to stay 1 year? Assume an interest rate of 12%. Solution: Dierent interpretations are possible here. For example, the deposit may be never be refundable and therefore represent a brokerage free that is paid up front. Alternatively, the deposit may only be nonrefundable if the couple choose not to take the appartment. We will adopt the latter interpretation but note that regardless of the interpretation, the important aspect of this example is that sunk costs are exactly that: sunk! (a) Compare the two alternatives. 1. Stay in the original apartment, for an NPV, C1 , given by
5

C1 =
i=0

1000 1000 + 4, 222. i 1.12 1.126

2. Take the new apartment where we assume a security deposit is again required. The NPV then is
5

C2 = 900
i=0

900 900 + 4700. 1.12i 1.126

3 This contrasts with portfolio optimization problems where it is commonly understood that an investor may invest as little or as much as she chooses in a particular asset such as a stock or bond. In this context of portfolio optimization, the term asset is preferred to the term project. 4 See, for example, Section 5.1 in Luenberger and Principle of Corporate Finance, by Brealey and Myers. 5 The criterion is called NPV since some of the cash ows are usually negative and these have to be included in the analysis. 6 See, for example, Examples 2.4 and 2.5 in Luenberger and the discussion that follows. 7 We will dene market completeness when we study martingale pricing later in the course.

Deterministic Cash Flows and the Term Structure of Interest Rates The couple should then take the $1000 apartment. (b) You can do this as an exercise. What result do you expect?

Example 2 (An Appraisal (Exercise 2.9 in Luenberger)) You are considering the purchase of a nice home. It is in every way perfect for you and in excellent condition, except for the roof. The roof has only 5 years of life remaining. A new roof would last 20 years, but would cost $20, 000. The house is expected to last forever. Assuming that costs will remain constant and that the interest rate is 5%, what value would you assign to the existing roof? Solution: We know a new roof costs 20, 000 and that it lasts 20 years. We can therefore infer the value per year, A, of a roof by solving 19 A . 20, 000 = 1.05i i=0 We nd that A = 1, 528.4. If V is the value of a roof that has 5 years of life remaining, we can obtain that V = 6, 948. Exercise 2 Find an alternative method for solving the problem in Example 2? Example 3 (Valuation of a Firm) A simple model that is sometimes used to determine the value of a corporation is the Gordon Growth Model. It assumes that there is a constant interest rate, r, and that dividends are paid annually and grow at a rate of g. The value of the rm may then be expressed as V0 = D1 (1 + g) D1 + + . . . = D1 1+r (1 + r)2

k=1

(1 + g)k1 (1 + r)k

so that we obtain V0 = D1 /(r g) for g < r. Clearly V0 is very sensitive to changes in g and r when g r. For this reason, the Gordon model has been used to provide some intuition for the volatility of growth stock prices; e.g. internet, biotech stocks. Obviously this model is very simple and is easy to generalize. For example, we could assume that dividends only grow for a certain xed number of years, possibly beginning at some future date. More realistic models should of course assume that dividends are stochastic. Moreover, since dividend policy is set by directors, we might prefer to focus not on dividends, but instead on the underlying cash ows of the corporation.

Fixed Income Securities

Traditionally, the term xed income securities refers to securities whose cash ows are xed in advance and whose values therefore depend largely on the current level of interest rates. The classic example of a xed income security is a bond which pays a xed coupon every period until expiration when the nal coupon and original principal is paid. A stock, on the other hand, is the classic example of a non-xed income security as the dividend payments (if they exist) and stock value vary stochastically through time. It should be mentioned, however, that since interest rates vary stochastically through time, so too do bond prices. Furthermore, many securities (e.g. convertible bonds) have xed-income and non-xed-income characteristics so the distinction is often blurred. Question: Are xed-income securities risk-free securities?

Deterministic Cash Flows and the Term Structure of Interest Rates

Annuities, Perpetuities and Amortization

Denition 6 An annuity is a contract that periodically pays a pre-determined amount of cash over some interval of time. Traditionally annuity payments were made on an annual basis (hence the term annuity) and the interval of time was xed. However, there are many variations. Pensions, for example, sometimes periodically pay a pre-determined amount of cash until a random time, T , usually the time of death of the recipient or the recipients spouse. Denition 7 A perpetual annuity or a perpetuity pays a xed amount of cash periodically for ever. Perpetuities are rare but do exist in some countries. In the UK, for example, they are called consols. A perpetuity is easily priced. Suppose it pays a xed amount, A, per period beginning at the end of the current period, and that the interest rate per period is r. Then the price, P , of the perpetuity satises

P =
k=1

A A = . (1 + r)k r

(2)

The price of an annuity that pays A per period beginning at the end of the current period for a total of n periods, satises n 1 A A 1 . = P = k (1 + r) r (1 + r)n
k=1

(3)

As always, the formulae in (2) and (3) depend on the compounding convention and how interest rates are quoted. They may also be inverted to express A as a function of P . For example, we may also write (3) in the form r(1 + r)n P . (4) A = (1 + r)n 1 This form of the annuity pricing formula is useful for determining the periodic payments that correspond to a xed value, P . It is also useful for amortization, which is the process of substituting periodic cash payments for an obligation today.

Yield-to-Maturity
The yield-to-maturity (YTM) of a bond is the interest rate (always quoted on an annual basis) that makes the present value of all associated future payments equal to the current value of the bond. In particular, the YTM is exactly the IRR of the bond at the current price. The YTM, , therefore satises
n

P =
k=1

C/m 1 F C 1 + = k n [1 + (/m)] [1 + (/m)] [1 + (/m)]n

F [1 + (/m)]n

(5)

where F is the face value of the bond, C is the annual coupon payment, m is the number of coupon payments per year, n is the exact total number of coupon payments remaining, and it is assumed that compounding is done m times per year. A lot of information is present in equation (5). In particular, it may be seen that P is a decreasing function of . When = 0, the bond price is simply the sum of the future payments, while if = C/F , it can then be shown (do it!) that P = F . Dierent bonds can have dierent yields but they generally track one another quite closely, particularly when the bonds have similar maturities. It should also be emphasized that bond yields, and therefore bond prices, generally change stochastically through time. Hence bonds are risky securities, despite the fact that their payment streams are xed8 .
8 It

is worthwhile reading Section 3.4 of Luenberger to further develop your intuition regarding yield-to-maturity.

Deterministic Cash Flows and the Term Structure of Interest Rates

Macauley Duration
Consider a nancial security that makes payments m times per year for a total of n periods. Then if is the YTM of the security, we dene its Macauley duration, D, to be D =
n k=1 (k/m)ck / [1

+ (/m)]

(6)

where P is the present value of the security, and ck is the payment made in the k th period. Note that D is a weighted average of the times at which the payments are made, where the weight at period k is the contribution of ck to the present value of the security. In particular, time and duration have the same units and D satises 0 D T where T is the maturity time of the instrument. Exercise 3 What is the duration of a zero-coupon bond? The duration of a security may be interpreted as the length of time one has to wait in order to receive the securitys payments9 . We will soon see how it gives a measure of interest rate sensitivity and how it can be used to immunize a bond against adverse interest rate movements. In the case of a bond that has a coupon rate of c per period and yield y per period10 , the summation in (6) may be simplied to obtain 1+y 1 + y + n(c y) D = . (7) my mc [(1 + y)n 1] + my

Convexity
We dene the convexity, C, of a nancial security with YTM to be C = For the nancial security described above we obtain11 C = 1 P [1 + (/m)]
2 k=1 n

1 d2 P . P d2

(8)

ck k(k + 1) . k 2 m [(1 + (/m)]

(9)

We will now see how duration and convexity may be used to immunize a bond against adverse changes in interest rates, or to be more precise, yield-to-maturity.

Immunization
Let P () be the price of a nancial security (or a portfolio of nancial securities) when the YTM is . Then a simple second-order Taylor expansion implies (check!) P ( + ) = where DM :=
9 See 10 That

P () +

dP ()2 d2 P () + () d 2 d2 P ()C ()2 2 (10)

P () DM P () + D 1 + /m

the discussion on page 59 of Luenberger regarding the qualitative properties of duration. is, y = /m in the notation of equation 6. 11 Analogously to (7), it is possible to simplify the summation in (9) when the security is a bond with xed coupon rate, c.

Deterministic Cash Flows and the Term Structure of Interest Rates is called the modied duration12 .

Consider now a portfolio of n securities where we use Pk , Dk and Ck to denote the present value, duration and convexity of the k th security, 1 k n. Let Pp , Dp and Cp denote the present value, duration and convexity of the overall portfolio. It is easy to check (do it) that Pp , Dp and Cp satisfy
n

Pp

=
k=1 n

Pk Pk Dk Pp Pk Ck . Pp

Dp

=
k=1 n

Cp

=
k=1

Suppose now that we own a portfolio (which may have negative value) of xed income securities, and the current value of this portfolio is Po . As changes, so will Po , but in many circumstances we would like Po to remain constant and not change as changes. We can eectively arrange for this by adding securities to our portfolio in such way that we are immunized against changes in . The Taylor expansion in (10) implies that we can do this by adding securities to our portfolio in such a way that the combined present value, duration and convexity of the new securities match the present value, duration and convexity of the original portfolio13 . To do this, we generally need to add at least three securities to our portfolio. If we just use duration to immunize, then we generally only need to add two securities to the portfolio. The following example illustrates how to do this. Example 4 (Immunizing a Cash Flow) Suppose the current YTM is 8% and we have an obligation14 to pay 1 million dollars in 7 years. We wish to immunize this obligation by purchasing a portfolio of bonds in such a way that the value, duration and convexity of the obligation and bond portfolio coincide. Because this involves three equations, we will need at least three bonds in our portfolio. There are two 6-year bonds, with coupon rates 7% and 10%, and a 9-year bond with coupon rate 2%, available. Let P1 , P2 and P3 be the dollar amount invested in each of the three bonds, respectively. We must then solve the following equations: Po Do Co = = = P1 + P2 + P3 P2 P3 P1 D1 + D2 + D3 Po Po Po P1 P2 P3 C1 + C2 + C3 Po Po Po (11) (12) (13)

where Po , Do and Co are the present value, duration and convexity, respectively, of the obligation. We therefore have 3 linear equations in 3 unknowns and we can solve to obtain P1 = 12.78 million, P2 = 11.63 million and P3 = 576 thousand dollars. Note that this means we need to sell short15 the second and third bonds. Let us now convince ourselves that we are indeed immunized against changes in . We can consider our overall portfolio to be comprised of two sub-portfolios: (i) the obligation and (ii) the portfolio consisting of positions in
are implicitly assuming that compounding is done m times per year. that the present value of the new portfolio will be unchanged at Po though the duration and convexity will both be zero. This is because the purchase (or sale) of the new securities will be nanced by a cash amount of Po which will oset the cost of the new securities. The cash amount of P0 has zero duration and convexity. 14 Note that since we are immunizing an obligation, the signs on the left-hand-sides of equations (11), (12) and (13) below are positive and not negative. 15 The act of short-selling a security is achieved by rst borrowing the security from somebody and then selling it in the market. Eventually the security is purchased and returned to the original lender. Note that a prot (loss) is made if the security price fell (rose) in value between the times it was sold and purchased in the market.
13 Note 12 We

Deterministic Cash Flows and the Term Structure of Interest Rates

the three bonds. Moreover, we have arranged it so that the value, duration and convexity of the two sub-portfolios coincide with one another. As a result, if changes then the values of the two sub-portfolios will change by equal amounts according to (10). However, the changes will be in opposite directions and therefore cancel each other because the rst sub-portfolio is an obligation. See the Excel spreadsheet, DurationConvexity.xls , for the solution details and an analysis of how well the portfolio is immunized when YTM changes to 7% or 9%. These immunization methods are used a great deal in practice and have proven successful at immunizing portfolios of xed income securities against adverse interest rate movements. This is despite the fact that interest rates are assumed to be at and only parallel shifts in interest rates are assumed to be possible. Such assumptions are not at all consistent with reality and they do not allow for a satisfactory theory of interest rates16 .

Further Examples and Applications

Example 5 (Callable Bond (Exercise 3.5 in Luenberger)) The Z corporation issues a 10%, 20-year bond at a time when yields are 10%. The bond has a call provision that allows the corporation to force a bond holder to redeem his or her bond at face value plus 5%. After 5 years the corporation nds that exercise of this call provision is advantageous. What can you deduce about the yield at that time? (Assume one coupon payment per year.) Solution: If exercise of the call provision is advantageous, then it must be the case (why?) that
15

k=1

1 .1 + > 1.05. (1 + )k (1 + )15

In particular, it implies < 9.366%. Example 6 (Need for a Better Theory! (Exercise 3.16 in Luenberger)) Suppose that an obligation occurring at a single time period is immunized (matching duration, but not convexity) against interest rate changes with bonds that have only nonnegative cashows. Let P () be the value of the resulting portfolio, including the obligation, when the interest rate is r + and r is the current interest rate. By construction P (0) = 0 and P (0) = 0. We will show that P (0) 0, i.e. P (0) is a local minimum. With a yearly compounding convention, the discount factor for time t is dt () = (1 + r + )t . Without loss of generality it is assumed that the magnitude of the obligation is 1 and it is due at time t. If ct are the cashows of the bonds that have been used to immunize the obligation, the following equations must be satised (why?): P (0) =
t

ct dt dt = 0 tct dt tdt = 0.
t

(14) (15)

P (0)(1 + r) = (a) Show that for all values of and there holds P (0)(1 + r)2 =
t

(t2 + t + )ct dt (t + t + )dt .

(16)

16 See Example 6 or the Term-Structure Models lecture notes that we will study at the end of the course for a numerical example describing why the YTM framework allows arbitrage opportunities to exist.

Deterministic Cash Flows and the Term Structure of Interest Rates

(b) Show that and can be selected so that the function t2 + t + has a minimum at t and has a value of 1 there. Use these values to conclude that P (0) 0. Solution (a) Use equations (14) and (15) to obtain (16). 2 (b) Choose (why?) = 2t and = 1 + t . Using (14) we therefore see that P (0) 0. Remark 1 The result of the previous example is signicant. It seems to say that we can make money from nothing (arbitrage) if we assume that interest rates can only move in parallel. While duration and convexity are used a lot in practice and often result in a well immunized portfolio, this result highlights the need for a more sophisticated theory of interest rates. See Example 9 for a more explicit example where a model that only permits parallel movements in interest rates aords an arbitrage opportunity.

The Term Structure of Interest Rates

Until now, we have discussed a bonds yield-to-maturity and how this concept, together with duration and convexity, may be used to immunize or hedge the cashow of a xed income portfolio. We have also mentioned that bond prices in the real world are stochastic (hence the need for immunization techniques), but we have not discussed the precise stochastic nature of bond price processes. Any theory of these processes, and the stochastic evolution of interest rates more generally, should possess at least two qualities: it should provide an adequate t to real observed data, and it should preclude arbitrage possibilities. Towards this end, it is necessary to develop a much more sophisticated theory of interest rates. This need is evidenced in part by the existence in the real world of bonds with very dierent YTMs, and by Examples 6 and 9 where we see that arbitrage opportunities can exits in the YTM framework or frameworks where only parallel movements in interest rates are possible. In order to develop a more sophisticated theory of interest rates, we rst need to study the term structure of interest rates.

The Term Structure, Discount Factors and Present Values

If a bank lends you money for one year and lends money to someone else for ten years, it is very likely that the rate of interest charged for the one-year loan will dier from that charged for the ten-year loan. Term-structure theory has as its basis the idea that loans of dierent maturities should incur dierent rates of interest. This basis is consistent with what we observe in practice and allows for a much richer and more realistic theory than that provided by the yield-to-maturity (YTM) framework. We will often assume that there are m compounding periods per year, but it should be clear what changes need to be made for continuous-time models and dierent compounding conventions. For example, we will also consider models where we compound on a per period basis. Time may be measured in periods or years, but it should be clear from the context what convention we are using. Spot Rates: Spot rates are the basic interest rates that dene the term structure. Usually dened on an annual basis, the spot rate, st , is the rate of interest charged for lending money from today (t = 0) until time t. In particular, this implies that if you lend A dollars for t years17 today, you will receive A(1 + st /m)mt dollars when the t years have elapsed. The term structure of interest rates may be dened to constitute the sequence of spot rates, {stk : k = 1, . . . , n}, if we have a discrete-time model with n periods. Alternatively, in a continuous-time model the set {st : t [0, T ]} may be dened to constitute the term-structure. The spot rate curve is dened to be a graph of the spot rates plotted against time. In practice, it is almost always upwards sloping in which case sti < stj whenever i < j. Discount Factors: As before, there are discount factors corresponding to interest rates, one for each time, t.
17 We

assume that t is a multiple of 1/m both here and in the denition of the discount factor, dt , above.

Deterministic Cash Flows and the Term Structure of Interest Rates The discount factor, dt , for a deterministic cash-ow occurring t years from now is given by dt := 1 . (1 + st /m)mt

10

Using these discount factors we can compute the present value, P , of any deterministic cash ow stream, (x0 , x1 , . . . , xn ). It is given by P = x0 + d1 x1 + d2 x2 + . . . + dn xn . where it is understood that xi is received at time ti and di := dti . Example 7 In practice it is quite easy to determine the spot rate by observing the price of U.S. government bonds. Government bonds should be used as they do not bear default risk and so the contracted payments are sure to take place. For example the price, P , of a 2-year zero-coupon government bond with face value $100, satises P = 100/(1 + s2 )2 where we have assumed an annual compounding convention. Forward Rates: A forward rate, ft1 ,t2 , is a rate of interest18 that is agreed upon today for lending money from dates t1 to t2 where t1 and t2 are future dates. It is easy using arbitrage arguments to compute forward rates given the set of spot interest rates. For example, if we express time in periods, have m periods per year, compound per period and quote rates on an annual basis, then we have (1 + sj /m)j = (1 + si /m)i (1 + fi,j /m)ji where i < j. Exercise 4 Prove that (17) must hold by using an arbitrage argument. (It is very important that you be able to construct these arguments. After a while you should be able to write equations such as (17) that correspond to dierent compounding conventions without explicitly needing to go through the arbitrage argument.) (17)

Forward Discount Factors: We can also discount a cash ow that occurs at time j back to time i < j. The correct discount factor is 1 di,j := (1 + fi,j /m)ji where again time is measured in periods and there are m periods per year. In particular, the present value at date i of a cashow, xj , that occurs at date j > i, is given by di,j xj . It is also easy to see that these discount factors satisfy di,k = di,j dj,k for i < j < k and they are consistent with earlier denitions. Short Forward Rates: The term structure of interest rates may equivalently be dened to be the set of forward rates. There is no inconsistency in this denition as the forward rates dene the spot rates and the spot rates dene the forward rates. We also remark that in an n-period model, there are n spot rates and n(n + 1)/2 f forward rates. The set19 of short forward rates, {rk : k = 1, . . . , n}, is a particular subset of the forward rates f that also denes the term structure. The short forward rates are dened by rk := fk,k+1 and may easily be shown to satisfy f f f (1 + sk )k = (1 + r0 )(1 + r1 ) . . . (1 + rk1 ) if time is measured in years and we assume m = 1.

is usually quoted on an annual basis unless it is otherwise implied. generally reserve the term rt to denote the short rate prevailing at time t which, in a stochastic model, will not be known until t.
19 We

18 It

Deterministic Cash Flows and the Term Structure of Interest Rates Example 8 (Constructing a Zero-Coupon Bond)

11

Two bonds, A and B both mature in ten years time. Bond A has a 7% coupon and currently sells for $97, while bond B has a 9% coupon and currently sells for $103. The face value of both bonds is $100. Compute the price of a ten-year zero-coupon bond that has a face value of $100. Solution: Consider a portfolio that buys negative (i.e. is short) seven bonds of type B and buys nine of type A. The coupon payments in this portfolio cancel and the terminal value at t = 10 is $200. The initial cost is 7 103 + 9 97 = 152. The cost of a zero with face value equal to $100 is therefore $76. (The 10-year spot rate, s10 , is then equal to 2.78%. Remark 2 Note that once we have the zero-coupon bond price we can easily determine the corresponding spot rate. (This was the point of Example 7.) We now demonstrate that even simple and apparently reasonable term-structure models can contain arbitrage opportunities. Example 9 (Arbitrage in a 1-Period Model) Suppose at t = 0 the 1-year, 2-year and 3-year spot rates are given by 10%, 11% and 12%, respectively. One year from now the 1-year, 2-year and 3-year spot rates rates will either have increased to [11%, 12%, 13%] or decreased to [9%, 10%, 11%]. Note that this example assumes that only parallel movements in the spot rates can occur. If we assume continuous compounding, then we can see that the forward rate, f1,2 , at t = 0 is given by f1,2 = 2(.11) 1(.1) = 12%. 1

This forward rate, however, is higher than either of the possible 1-year spot rates prevailing at t = 1 and so there is an arbitrage opportunity. Exercise 5 Construct a trading strategy that would take advantage of the arbitrage opportunity identied in Example 9. At the end of the course (after we have studied martingale pricing) we will see how to construct term structure models that are arbitrage free.

Term Structure Explanations

There are three well known hypotheses that are commonly used for explaining the observed term structure of interest rates: the expectations hypothesis, the liquidity hypothesis and the market segmentation hypothesis. Expectations Hypothesis: The expectations hypothesis states that the forward rates, fi,j , are simply the spot rates, sji , that are expected to prevail at time i. While this has some intuitive appeal, if the hypothesis was true then the fact that the spot rate curve is almost always upwards sloping would mean (why?) that the market is almost always expecting spot interest rates to rise. This is not the case. Liquidity Preference Hypothesis: This hypothesis states that investors generally prefer shorter maturity bonds to longer maturity bonds. This is because longer-maturity bonds are generally more sensitive to changes in the general level of interest rates and are therefore riskier. In order to persuade risk-averse individuals to hold these bonds, they need to be sold at a discount, which is equivalent to having higher interest rates at longer maturities. Market Segmentation Hypothesis: This states that interest rates at date t1 have nothing to do with interest rates at date t2 for t1 = t2 . The rationale for this is that short-term securities might be of interest to one group of investors, while longer term securities might be of interest to an altogether dierent group. Since these investors have nothing in common, the markets for short- and long-term securities should be independent

Deterministic Cash Flows and the Term Structure of Interest Rates

12

of one another and therefore the interest rates that are set by the market forces of supply and demand, should also be independent. This explanation is not very satisfactory and explains very little about the term structures that are observed in practice. It is commonly perceived that the term structure is reasonably well explained by a combination of the expectations and liquidity preference hypotheses.

Duration and Immunization

We saw earlier how to immunize portfolios against changes in the YTM when we assumed that a bonds YTM captured all the relevant economic information. Now we have progressed to a term structure of interest rates, and we know how to price a bond given the current term structure without making any reference to the bonds YTM. In this context, however, we could still dene a bonds duration and convexity, and use them to immunize against, for example, parallel movements in the term structure. The immunization procedure is very similar20 to the one described earlier in the YTM context.

Further Examples and Applications

Example 10 (Floating Rate Bonds) A oating rate bond is the same as a regular bond except that the coupon payments at the end of a period are determined by the prevailing short rate at the beginning of the period. Since the term structure varies stochastically through time, this implies that the payment stream of a oating rate bond is also stochastic. Nonetheless, it is easy to price such a bond at a reset point by using a simple dynamic programming type argument. In fact, at any such reset point, it can be seen that the value of the oating rate bond is equal to the face value, or par. The argument works by noting that at the nal reset point, the present value must be (why?) par. The same argument can then be iterated back to date t = 0. Remark 3 The above comment regarding the price of oating rate bonds can be useful for pricing interest rate swaps. We will see this later when we study Forwards, Swaps and Futures. Example 11 (Mortgage Mathematics (Exercise 4.14 in Luenberger)) This example is an introduction to the mathematics of mortgage-backed-securities (MBSs), a class of derivative instruments that play an enormous role in the U.S. mortgage industry. They are credited with enabling a more ecient allocation of risk and in the process, enabling consumers to obtain much cheaper mortgages. We assume the initial value of the mortgage is M = M (0), that equal periodic payments of size B are made, and that the interest rate is r per period. We also assume that there are a total of n periods in the mortgage and we use M (k) to denote the mortgage principal after the k th period. Note that after the n payments of size B, the mortgage principal and interest have all been paid. This means that each payment, B, pays both interest and some of the principal. With this in mind, we can see that M (k) satises M (k) = (1 + r)M (k 1) B This equation can be solved (how?) to obtain M (k) = (1 + r)k M Since M (n) = 0 we have (see also equation (4)) B=
20 Details

for k = 0, 1, 2, . . .

(1 + r)k 1 B. r

r(1 + r)n M . (1 + r)n 1

can be found in Section 4.8 of Luenberger.

Deterministic Cash Flows and the Term Structure of Interest Rates We can therefore substitute for B and write M (k) = M (1 + r)n (1 + r)k . (1 + r)n 1

13

Since we know M (k 1) we can compute the interest, I(k) = rM (k 1), on M (k 1) that would be due in the next period. This also means that we can interpret the k th payment as paying B rM (k 1) of the remaining principal. (a) Find the present value, V , (at rate r) of the principal payment stream in terms of B, r, n, M . (b) Find V in terms of r, n, M only. (c) What is the present value, W , of the interest payment stream? (d) What is the value of V as n ? (e) Which stream do you think has the larger duration: principal or interest? Solution (a) The present value, V , of the principal payment stream is given by V = Does this agree with your intuition when n = 1? (b) Substitute for B to obtain V = rnM . (1 + r) [(1 + r)n 1] n(B rM ) . 1+r

(c) The present value, W , of the interest payment stream satises

n

W =
k=1

(1 + r)n+1 1 r nr I(k) = M . (1 + r)k (1 + r) [(1 + r)n 1]

Does this agree with your intuition when n = 1? (d) Check that V M as n . (e) The principal payment stream has longer duration. Why? What does this say about the relative riskiness of the principal and interest payment streams?

Dynamic programming (DP) is a very useful tool in nancial engineering. It may be used whenever a decision-maker needs to make a sequence of decisions through time. In such circumstances, the decision maker usually needs to nd a strategy that will describe her decision in each possible state of the world. DP21 is usually applied when there is uncertainty in the system but it also applies in deterministic settings and that is the setting of Example 12 below. Example 12 (Operating a Factory) You own a lease on a factory that expires worthless exactly three years from now. Each year the factory earns prots that are equal to the value of the factory that prevailed at the beginning of the year. At the end of each year you can either withdraw these prots or reinvest them in the factory, thereby doubling its value. The current value of the factory is $5 million and the interest rate is 20% per year. What strategy should you adopt? Solution The lattice below describes the various possibilities that can occur depending on the choices you make. An
21 It is worth remarking that DP problems very often need to be solved numerically. This is clearly true for discrete-time, discrete-state examples, but it is also often true for continuous-state models. In such circumstances DP is often applied after discretizing the state space.

Deterministic Cash Flows and the Term Structure of Interest Rates

14

upwards sloping arc corresponds to reinvesting the prots, whereas a horizontal arc corresponds to withdrawing the prots at the end of the year. The value at a node represents the value of the factory at that node. 40

20 20 20 0 0 10 10 10 10 10 0 0 0 5 5 5 5 5 5 5 t=0 t=1 t=2 t=3

We now compute the optimal strategy by working backwards from t = 3 to compute the value of the lease. (It is convenient22 to assume that the decision to reinvest prots at the end of a period will be made at the beginning of that period.) At t = 3 the lease expires worthless and so in the lattice below we place a zero at each terminal node. At t = 2 it is clearly optimal to choose to take the prots at the end of the period. Since these prots are received at the end of the node, their value must be discounted backwards by 1/(1 + r) to t = 2. For example, we obtain 20/1.2 = 16.67 at the uppermost node at t = 2. At t = 1, for example, the value of the lease at the lowermost node is given by max(8.33, 5 + 4.17)/1.2 = 7.64. Continuing in this manner we nd that the initial value of the lease is $12.73 million and that the optimal strategy is to reinvest the prots after the rst period and to withdraw the prots after each of the remaining two periods.

16.67 0 20 0 0 15.28 8.33 0 10 10 0 0 0 12.73 7.64 4.17 0 5 5 5 t=0 t=1 t=2 t=3

Remark 4 Note that in Example 12 the optimal strategy is the same for all values of r > 0. Exercise 6 Suppose you have an option to sell the lease at t = 1 for $11 million. What is the optimal strategy now? Is it still true that the optimal strategy is independent of r for r > 0?
22 Since

this is a deterministic problem there is no loss in making this assumption.

Deterministic Cash Flows and the Term Structure of Interest Rates

15

Even in the simple deterministic setting of these notes, it is possible to construct many more examples ranging from capital-budgeting applications to problems in asset-liability management. Standard mathematical tools for solving these problems include linear, integer and dynamic programming.

Challenge Question
A fund manager tells you that his Krypton Fund has outperformed all other funds and benchmarks on both an absolute and risk-adjusted basis over the past 5 years. For a considerable fee, you can invest in this fund but the fund manager points out that even after taking this fee into account, the Krypton Fund is still easily the best performing fund on the market. You decide not to invest. Are you a complete fool(!) or might there be a good reason for your decision?