Term Structure of Interest Rates: Termstructureofinterestrates

Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

T E R M S T R U C T U R E O F I N T E R E S T R A T E S

Term Structure of Interest Rates


This is the rst of two articles on the term structure. In it, the authors
discuss some term structure fundamentals and the measurement of the
current term structure. They also illustrate the Vasicek and the
Cox-Ingersoll-Ross models of the term structure. A succeeding article will
discuss the Black-Derman-Toy and Black-Karasinsky models of the term
structure.
by Simon Benninga and Zvi Wiener
I
nterest rates and their dynamics provide probably the
most computationally difcult part of the modern -
nancial theory. The modern xed income market in-
cludes not only bonds but all kinds of derivative securities
sensitive to interest rates. Moreover interest rates are im-
portant in pricing all other market securities since they are
used in time discounting. Interest rates are also impor-
tant on corporate level since most investment decisions
are based on some expectations regarding alternative op-
portunities and the cost of capitalboth depend on the
interest rates.
Intuitively an interest rate is something very clear. Con-
sider a payment of $1 which will be made with certainty at
time t (throughout this article we will consider only pay-
ments made with certainty and consequently only risk-free
interest rates). If the market price of $1 paid in time t from
now is P
0
, then the interest rate for time t can be found
from the simple discount formula P
0
=
$1
(1+rt )
t
. The inter-
est rate r
t
in this formula is known as the pure discount
interest rate for time t.
At this point a short and very incomplete review of
bond terminology is in order. A standard bond has the
following characteristics:
D the bonds face value (sometimes called its notional
value)
C the bonds coupon payments. Often quoted as a per-
centage of its face value (hence the terminology cou-
pon ratethe percentage rate of the face value paid as
coupons)
N the bonds maturity; the date of the last payment (con-
sisting of the face value plus the coupon rate)
Once the term structure of interest rates is known, the
price of a bond with yearly coupons is given by:
P
0
=
N

t =1
C
(1 + r
t
)
t
+
D
(1 + r
N
)
N
,
For analytic computations it is sometimes more conve-
nient to assume that a bond makes a continuous stream
of payments between time 0 and time N. Denoting the
payment during the time from t till t + dt by C
t
dt , the
bond price is given by P
0
=

N
t =1
e
-tr t
C
t
dt . This expression
makes use of continuous compounding.
A zero coupon bond (sometimes referred to as a pure
discount bond) is a bond with no coupon payments. In
our initial calculations in this section, we have calculated
the term structure from such pure discount bonds. The
term structure of interest rates describes the curve r
t
as a
function of t. Although most of this article deals with the
theory of term structures, it helps to look at some actual
term structures. Consider the following graphs, based on
data collected by John McCullogh. The rst graph gives
12 term structures, one for each month of 1947:
The graph of 1947 term structures may suggest that the
interest rate increases with the length of the time period
over which the payment is promised, but the following
picture shows that this is not always true:
Vol. 7 No. 2 1998 Mathematica in Education and Research 1
T E R M S T R U C T U R E O F I N T E R E S T R A T E S
FORWARD INTEREST RATES
The forward interest rate is a rate which an investor can
promise herself today, given the term structure. Here is a
simple example: Suppose that the interest for a maturity
of 3 years is given by r
3
= 10% and the interest rate for
a maturity of 5 years is given by r
5
= 11%. Furthermore
suppose that lending and borrowing rates are equal. Now
consider the following package of lending and borrowing
made by an investor:
Lend $1000 for 3 years at 10%.
Borrow $1000 for 5 years at 11%.
The cash ow pattern of this lending/borrowing looks
like:
time 0: Borrow $1000 and lend $1000 - net zero cash ow.
time 3: Get repayment of $1000 lent at time 0 for 10% -
inow of $1000 * (1.10)
3
= $1331.
time 5: Repay $1000 borrowed at time 0 for 11% - outow
of $1000 * (1.11)5 = $1658.06.
The upshot is that the package looks exactly like a 2-
year loan at time 3 arranged at time 0. The annual interest
rate - the two-year forward interest rate at time 3 on this
loan is: 12.517%
In[1]:= Sqrt[1.115/1.13] 1
Out[1]= 0.125171
Thus, given a discrete term structure of interest rates,
r
1
, r
2
, . . . the n-period forward rate at time t is dened by:
K
(1 + r
t +n
)
(t +n)
(1 + r
t
)
t
O
1 / n
= r
f
t ,n
We most often use a continuous version of this formula:
(Exp[(t + n)r
t +n
] / Exp[t r
t
])
1 / n
= Exp[r
f
t ,n
]
Or using the linear approximation,
r
f
t ,n
=
1
n
[t (r
t +n
- r
t
) + n r
t +n
] =
t (r
t +n
- r
t
)
n
- r
t +n
.
Or alternatively
r
f
t ,n
= -
logP(t , t + n)
n
,
where P(t , t + n) is the time t price of $1 paid at time
t + n.
1. ESTIMATING THE TERM STRUCTURE FROM BOND
MARKET DATA
In principle, estimating the term stucture of interest rates
from bond market data ought to be a matter of a fairly
simple set of calculations. The simple idea is to take a set
of bonds {B
j
}, decompose each of them into series of pay-
ments, discount each series according to some unknown
term structure and then to equate the resulting expres-
sion to the observed prices. In practice there are many
complications caused by a variety of market complica-
tions: non-simultaneity of bond prices, bid-ask spreads,
liquidity premiums, bond covenants and embedded bond
options, etc.
In this section we give only a simple illustration of the
problematics of estimating the term structure. Let us as-
sume that the true term structure is given by interest rates
are r
1
= 5.5%, r
2
= 5.55%, r
3
= 5.6%, r
4
= 5.65%, r
5
= 5.7%,
for 1,2,3,4 and 5 years respectively. Suppose we observe
prices of the following bonds: (1 years, 3%), (2 years, 5%),
(3 years, 3%), (4 years, 7%), (5 years, 0%). The rst number
is time to maturity and the second number is the coupon
rate; we assume that each bonds coupon is paid once a
year. The face value of each bond is $1000, and we assume
yearly payments for simplicity. Thus, for example, the
bond
(4 years, 7%) pays 3 payments of $70 (one year from
today, two years from today, and three years from today);
in four years the bond will pay $1070.
The term structure allows us to price the bonds:
In[2]:= termStructure =
{0.05, 0.058, 0.059, 0.0595, 0.06};
The bonds can be described as:
In[3]:= bond1 = {1., 0.03};
bond2 = {2., 0.05};
bond3 = {3., 0.03};
bond4 = {4., 0.07};
bond5 = {5., 0.};
The pricing function is:
In[4]:= Clear[bondPrice]
bondPrice[bond ] := Sum[1000 bond[[2]]
E( termStructure[[i]] i),
{i, bond[[1]]}]
+1000
E( termStructure[[ bond[[1]] ]]
bond[[1]]);
The prices of these bonds are:
In[5]:= bprices = Map[bondPrice,
{bond1, bond2, bond3, bond4, bond5}]
Out[5]= {979.766, 982.56, 918.164, 1030.94, 740.818}
We form a formal expression for pricing bonds using
the unknown interest rates
In[6]:= Clear[BPriceExpr]
IRunknown = {r1, r2, r3, r4, r5};
BPriceExpr[bond ] :=
Sum[1000 bond[[2]]
E( IRunknown[[i]] (i)),
{i, bond[[1]]}]
+1000 E( IRunknown[[bond[[1]]]]
bond[[1]]);
Solving the appropriate pricing equations gives us back
the precise term structure of interest rates:
2 Mathematica in Education and Research Vol. 7 No. 2 1998
T E R M S T R U C T U R E O F I N T E R E S T R A T E S
In[7]:= impliedIR = FindRoot[
{bprices[[1]] == BPriceExpr[bond1],
bprices[[2]] == BPriceExpr[bond2],
bprices[[3]] == BPriceExpr[bond3],
bprices[[4]] == BPriceExpr[bond4],
bprices[[5]] == BPriceExpr[bond5]},
{r1, 0.045}, {r2, 0.045}, {r3, 0.045},
{r4, 0.045}, {r5, 0.045}]
Out[7]= {r1 0.05, r2 0.058, r3 0.059,
r4 0.0595, r5 0.06}
Here we use a at 4.5% term structure as an initial
guess. Note that Solve or even NSolve are very inef-
cient in this case, and that we have used the FindRoot
operator. In this case our computation of the implied term
structure gives us back the true underlying term structure.
So Whats the Problem?
The above example may lead the unsuspecting reader to
imagine that the term structure of interest rates is simple to
calculate. We now showby introducing small changes in
the examplehow difcult the computational problem can
become. First imagine that our set of bonds is differentwe
still have 5 bonds, but they do not cover every maturity:
In[8]:= bond1 = {2., 0.03};
bond2 = {2., 0.05};
bond3 = {5., 0.03};
bond4 = {4., 0.07};
bond5 = {4., 0.};
Print["Bond prices = ",
bprices = Map[bondPrice,
{bond1, bond2, bond3, bond4, bond5}] ]
Out[8]= Bond Prices =
{945.726,982.56,867.073,1030.94,788.203}
Now suppose that we do not know the exact bond
prices, but know only the prices rounded to the nearest
dollar (this could easily happen because of noise in the
prices, or bid-ask spreads, for example):
In[9]:= roundBPrices = Round[bprices]
Out[9]= {946, 983, 867, 1031, 788}
Using our previous function to determine the term
structure:
In[10]:= impliedIR = FindRoot[
{roundBPrices[[1]] == BPriceExpr[bond1],
roundBPrices[[2]] == BPriceExpr[bond2],
roundBPrices[[3]] == BPriceExpr[bond3],
roundBPrices[[4]] == BPriceExpr[bond4],
roundBPrices[[5]] == BPriceExpr[bond5]},
{r1, 0.045}, {r2, 0.045}, {r3, 0.045},
{r4, 0.045}, {r5, 0.045}]
Out[10]= {r1 0.041343, r2 0.0579861,
r3 0.0607358, r4 0.0595643,
r5 0.0600487}
Plotting the actual term structure (solid line) against
this "computed" term structure (the dots) gives:
In[11]:= plot1 =
ListPlot[Table[impliedIR[[i, 2]], {i, 5}],
PlotStyle PointSize[0.03],
DisplayFunction Identity];
plot2 = ListPlot[termStructure,
PlotJoined True,
DisplayFunction Identity];
Show[plot1, plot2,
DisplayFunction $DisplayFunction];
2 3 4 5
0.045
0.055
0.06
Clearly this procedure mis-estimates the term structure!
Instead of solving the set of equations described above,
a better idea is to nd a smooth curve which gives an
approximate solution to the pricing equations. An elegant
solution of this problem which uses Mathematica can be
found in [Fisher and Zervos 1996]. The general idea is to
form a functional which measures how smooth the term
structure is and how well it approximates the given set
of bonds. Then one can form a continuous term structure
from a set of discrete observations as we show below with
a linear interpolation.
3. ARBITRAGE IN A FLAT TERM STRUCTURE
Why do we need a complicated model of interest rates?
Cant we use a at term structure as a reasonable approxi-
mation? There are at least two important reasons why this
assumption is problematic. First, as we showed above, the
data suggest that the term structure is not at. The second
reason why a at term structure is impossible is that a at
term structure is not arbitrage free. In this section we use
Mathematica to illustrate why this is so.
Assume that interest rates at time t are r
t
and that the
term structure is at: i.e., The rate r
t
applies at time t to
any loan or investment regardless of its length. We know
the current level of interest rates r
0
but in the future this
number can change. Thus r
t
is a random variable, but we
make no assumption about its distribution.
To show that this at term structure induces an arbi-
trage opportunity, we build a portfolio of zero-coupon
bonds with the following properties.
a. The current price of the portfolio is zero.
b. The derivative of the price of the portfolio with respect
to the interest rates is zero.
The minimal number of bonds that we need for such
portfolio is 3. Take a zero coupon bond with time to
maturity 1 year, 2 years and 3 years. We buy a units of the
Vol. 7 No. 2 1998 Mathematica in Education and Research 3
T E R M S T R U C T U R E O F I N T E R E S T R A T E S
rst one, sell b units of the second one and buy c units of
the third one. Then the price of the portfolio is:
a e
-r
- b e
-2r
+ c e
-3r
Its derivative with respect to the interest rates is:
a e
-r
- 2 b e
-2r
+ 3 c e
-3r
We can solve these two equations using Mathematica.
Below we illustrate the solution for the case where a = 1
and r (0) = 10%:
In[12]:= portfolio = Solve[ {
a Exp[ r] b Exp[ 2r] c Exp[ 3r] == 0,
Exp[ r]
2b Exp[ 2r] 3 c Exp[ 3r] == 0},
{b, c}] /.
{a 1, r 0.1}
Out[12]= {{b 2.21034, c 1.2214}}
This is a zero investment portfolio. When interest rates
change, the price of this portfolio changes as well. How-
ever there is no linear term in this change (since the re-
quirement was that this derivative vanishes). If the sec-
ond derivative is different from zero (either positive or
negative) we have found an arbitrage portfolio, since this
means that the price of the portfolio when interest rates
become r
t
is either positive or negative for all values of
r close to r
0
. In fact one can even prove that this is an
arbitrage in big and not only in a small neighborhood.
Draw the payoff of this portfolio at time t as a function of
interest rates r
t
:
In[13]:= Plot[Exp[ r]
b Exp[ 2r] + c Exp[ 3r] /. portfolio,
{r, 0.05, 0.15}];
0.06 0.08 0.12 0.14
0.0005
0.001
0.0015
0.002
0.0025
The nal payoff is never negative, and it is positive for
all values of r different from the current level of interest
rates. This is an obvious arbitrage.
However, do not throw away the at term-structure
model immediately. Looking at the vertical axis of the
above graph, we note that the values there (i.e., the ar-
bitrage prots) are very small. This means that even very
small transaction costs will destroy this arbitrage. In fact
what happens is that for long-term investment interest
rates are almost at. Only the short-term part of the term-
structure can be signicantly different from a at term
structure.
4. MODERN TERM STRUCTURE THEORY
Modern term structure theory falls into two broad classes:
Equilibrium models of the term structure derive the
term structure in models with consumer maximization and
occasionally production functions. The most famous ex-
ample is [Cox-Ingersoll-Ross 1985]: Their model has log-
arithmic utility functions for consumers and linear produc-
tion functions. Another example is [Benninga-Protopapadakis
1986]: They show that if consumers maximize concave
utility functions and if production functions are weakly
concave, then the real term structure will be upward slop-
ing. An important property of economic term structure
models is that any equilibrium term structure must ulti-
mately be at (see [Benninga and Wiener 1996]).
Non-equilibrium models of the term structure. This
is the fashion in nance today. The idea is to write a
plausible mathematical description of the term structure
which is numerically tractable. The models covered in
this series of papers are of this nature.
Some general facts about bonds
Suppose we denote by P(r,t,s) the price at time t of a pure
discount bond maturing at time s, s > t . Then the yield to
maturity R(r, t , T ) is the internal rate of return at time t
on a bond maturing at time t + T . Since
P(r, t , t + T ) = exp[-R(r, t , T )T ],
it follows that
R(r, t , T ) = -
1
T
log P(r, t , t + T ).
The integral of the forward rates gives the yield to matu-
rity:
R(r, t , T ) =
1
T

t +T
t
F(r, t , s) s.
The justication for this notation is that, when interest is
continously compoundedso that we exponentiate to get
future valuesthe sum of the forward rates gives the fu-
ture value. You can see this by taking a discrete example:
If r
0,1
is the interest rate from time 0 to 1, r
1,2
is the forward
rate from time 1 to 2, and so on, then the future value of
one dollar invested today for one period and rolled over
each period has future value e
r 0,1+r 1,2++rn-1,n
. Taking the
logarithm and dividing by the time period gives the above
integral equation. It follows that the forward rate can also
be written as:
F(r, t , s) = -

s
logP(r, t , s).
The structure of many bond pricing models is similar. We
often start off by assuming that interest rates follow a
diffusion process:
dr = m(r, t )dt + s(r, t )dB.
where B is a Wiener process. Given this assumption, a
pure discount bond price is a function of the current in-
terest rate r, the current time t, and the maturity T of the
4 Mathematica in Education and Research Vol. 7 No. 2 1998
T E R M S T R U C T U R E O F I N T E R E S T R A T E S
bond. Denoting this function by P(r,t,T), we have, by Itos
lemma:
dP =
P
r
dr +
P
t
dt +
s
2
2

2
P
r
2
dt
Substituting in for dr, we can write this equation as:
dP = P
r
[ mdt + sdB] + P
t
dt +
s
2
2
P
rr
dt
= KmP
r
+ P
t
+
s
2
2
P
rr
Odt + sP
r
dB
where subscripts indicate the appropriate derivatives. Di-
viding by dt and taking the expectation of dP/dt gives:
E K
dP
dt
O = mP
r
+ P
t
+
s
2
2
P
rr
At this point the models make use of an equilibrium pric-
ing model to postulate that E[dP/dt] must be equal to
the bond price P times the current risk-free rate r grossed-
up by a risk premium. Denoting this risk premium by l,
we get
E K
dP
dt
O = r(1 + l)P = mP
r
+ P
t
+
s
2
2
P
rr
or
0 = mP
r
+ P
t
+
s
2
2
P
rr
- r(1 + l)P.
The usual procedure to pricing the risk premium is to use
a result from Merton (1971, 1973). In Mertons papers it
is shown that in a continuous-time CAPM framework, the
ratio of each assets risk premium to its standard deviation
is constant when the utility function of the representative
investor is logarithmic:
E[R
i
] - r
s
Ri
=
l
s
Ri
= k
where E[R
i
] is the expected return on asset i, and s
Ri
is
asset is standard deviation of returns. For a pure discount
bond the instantaneous return is given by
E[R
i
] =
P + dP
P
= 1 +
dP
P
By Itos lemma the standard deviation of the rate of return
on the bond is given by:
s
RP
=
rs(r, t )
P
P
r
It thus follows that
l = kr
RP
=
krs(r, t )
P
P
r
The basic equation for our model thus becomes:
0 = mP
r
+ P
t
+
s
2
2
P
rr
- r(1 + l)P
or
0 = mP
r
+ P
t
+
s
2
2
P
rr
- rP -krs(r, t )P
r
(*)
The trick in the bond models now becomes nding a so-
lution to this differential equation! Another way of saying
this is that in bond option pricing models might be that
we have to try to nd a believable, yet solvable, diffusion
process for the interest rate.
There are several aspects of bond models which make
them different from the other stochastic models in nance
(e.g., the option pricing models):
1. The basic differential equation to be solved (*) de-
pends on an equilibrium pricing model. In option
pricing models, this is not so. The usual assumption
about this equilibrium pricing model is that there
is a single representative consumer with logarithmic
preferences. This aspect of bond models is almost
unavoidable, since a bond in these models is itself
a risky asset in the model. We can thus only model
the change in the bonds price by having recourse
to the model itself.
2. The stochastic process for the interest rate is itself
problematic, and this for two reasons:
2.a. In some of the models the stochastic process postu-
lated for the interest rate allows the rate to become
negative. If we are discussing a real interest rate (and
hence trying to price real bonds), this is not, in itself,
a problem, since we know that real interest rates can
be negative. However, if we are trying to price nom-
inal bonds, it is an improbable assumption. The CIR
model gets around this problem by postulating an
interest rate process which cannot become negative.
2.b. The equilibrium process which gives rise to the in-
terest rate is unclear. In general equilibrium, interest
rates arise out of marginal rates of productivity. In
order to fully specify an interest rate process, there-
fore, we must show what the equilibrium technolo-
gies look like on the margin. The only model which
fully solves this problem is, again, [CIR 1985]. They
show a production process which gives rise to their
diffusion process for interest rates.
In the following sections, we discuss two specic mod-
els and show their results. In our next article we will dis-
cuss the discrete term structure models of Black-Derman-
Toy and Black-Karazinksi.
5. VASICEKS MODEL
This is one of the most widely-used models for the pric-
ing of bonds. [Vasicek 1977] uses the Ornstein-Uhlenbeck
process for the spot interest rate r :
dr = a(g - r)dt + sdB
Here g is the long-term mean spot interest rate, a is the
pressure to revert to the mean a > 0, and s is the in-
stantaneous standard deviation. The Vasicek model solves
for present value factors v(t , s, r). By v(t , s, r) we denote
the present value at time t of $1 paid at time s > t when
the current spot interest rate is r.
Our basic equation now becomes:
0 = a(g - r)P
r
+ P
t
+
s
2
2
P
rr
- rP -krsP
r
Vol. 7 No. 2 1998 Mathematica in Education and Research 5
T E R M S T R U C T U R E O F I N T E R E S T R A T E S
This partial differential equation can be solved
P(r, t , T ) = ExpC
1
a
(1 - E
-a(T -t )
)(R() - r)-
(T - t )R() -
s
2
4a
3
(1 - Exp[-a(T - t )])
2
G
where R() = g +
sq
a
-
s
2
2a
2
is the yield of a zero coupon
bond of innite maturity R() = lim
T
-
logP(t ,T )
T
and q is
the Sharpe ratio a general market price of risk measured
as a ratio of excess return to the standard deviation.
Simulating the Vasicek Model
The Vasicek model is a one-factor model: All rates ulti-
mately depend on the shortest-term interest rate, which
we call the spot interest rate (or simply the spot rate) and
denote by r. To simulate this rate we discretize the basic
Vasicek equation by considering changes in the interest
rate over a short period Dt :
Dr = a(g - r)Dt + sZ
0
Dt
Note thatas opposed to stock price models, which are
multiplicativeterm structure models are additive. This
means that if r
t
is the spot rate at time t, then the spot rate
at time t + Dt is given by
r
t +D
= r
t
+ Dr = r
t
+ a(g - r)Dt + sZ
0
Dt
We start off by simulating the short-term interest rate
process:
In[14]:= Needs[StatisticsNormalDistribution]
Clear[
nor, deltaR, shortTermR, , , r]
nor[mu , sigma ] :=
Random[NormalDistribution[mu, ]];
deltaR[
alpha , gamma , r , sigma , deltaT ] :=
( r) deltaT+
nor[0, 1] Sqrt[deltaT]
shortTermR[
alpha , gamma , r , sigma , deltaT ] := r+
deltaR[, , r, , deltaT]
Set a = 3%, g = 4%, s = 0.12, d
t
= 0.0001. This means
that the time step is about 1 hour and that r
today
= 3%. We
now simulate 300 steps.
In[15]:= SeedRandom[2]
tt = NestList[shortTermR[0.03,
0.04, #, 0.12, 0.0001] & , 0.03, 300];
ListPlot[tt, PlotJoined True,
PlotLabel "Vasicek Simulation"];
50 100 150 200 250 300
0.025
0.03
0.035
0.04
Vasicek Simulation
Running this simulation for 3000 time steps, we observe
one of the most problematic points of the Vasicekmodel
the spot interest rates can become negative.
In[16]:= SeedRandom[2]
tt = NestList[
shortTermR[0.03, 0.04, #, 0.12, 0.0001] & ,
0.03, 3000];
ListPlot[tt, PlotJoined True,
PlotLabel "Vasicek Simulation"];
500 1000 1500 2000 2500 3000
-0.08
-0.06
-0.04
-0.02
0.02
0.04
Vasicek Simulation
Thus, starting from a current interest rate of 3%, with
s = 0.12, and the long-run spot rate of 4%, we get many
negative interest rates. Furthermore, the Vasicek process
can lead to negative expected interest rates. To see this,
look at the conditional expectation and variance of the
Ornstein-Uhlenbeck process:
E
t
r (T ) = g + (r (t ) - g)e
-a(T -t )
Var
t
r (T ) =
s
2
2a
(1 - e
-a(T -t )
)
It is not difcult to see that for r (t ) < 0, E
t
r (T ) can
be negative. This is an undesirable property. Note that
negative interest rates are not necessarily inconceivable.
If we are modeling a real term structure modelreal inter-
est rates being the after-ination return on a bondthen
negative interest rates may be perfectly normal; indeed,
real interest rates are often negative. It is more unlikely
that nominal interest ratesthe quoted interest rates on a
bondare negative.
The Shapes of the Term Structure
in the Vasicek Model
Recall that in the Vasicek model the spot rate denes the
whole term structure. Suppose that the spot interest rate
today is r. Then the price of a pure-discount bond is given
by the expression P(r,t,T) dened above when t = 0:
P(r, 0, T ) = ExpC
1
a
(1 - E
-aT )
)(R() - r)-
-TR() -
s
2
4a
3
(1 - Exp[-aT ])
2
G.
This means that the interest rate r
t
is given by Log[P(r,
0, T)/T]. Thus the Vasicek model enables us to plot the
whole shape of the term structure. Here is an example:
6 Mathematica in Education and Research Vol. 7 No. 2 1998
T E R M S T R U C T U R E O F I N T E R E S T R A T E S
In[17]:= Clear[P, pureDiscountRate, , , , ]
= 0.06;
= 0.02;
= 0.667;
= 1;
RInf = + /
2/(2 2);
P[r , t , T ] :=
Exp[1/ (1 Exp[ (T t)])(RInf r)
(T t)RInf (2/r 3)
(1 Exp[ (T t)])2] // N;
pureDiscountRate[r , T ] :=
If[T == 0, r, Log[P[r, 0, T]]/T]
Plot[pureDiscountRate[0.03, T],
{T, 0, 10}];
2 4 6 8 10
0.04
0.05
0.06
0.07
If we start with r above the long-run rate, the term-
structure will be downward sloping:
In[18]:= Plot[pureDiscountRate[0.1, T], {T, 0, 10}];
2 4 6 8 10
0.08
0.085
0.09
0.095
The Vasicek model can also produce humped term
structures:
In[19]:= = 0.12;
= 0.08;
Plot[pureDiscountRate[0.1, T], {T, 0, 10}];
2 4 6 8 10
0.085
0.09
0.095
0.105
0.11
0.115
Pricing Bonds with Vasicek Model
Given the term structure of interest rates (fully described
in this model by the spot rate) the pricing formula for
bonds is straighforward. Bond prices are determined by
summing the present value of the coupons and terminal
value, discounting at the discount factors P(r, 0, T ). Thus,
the value of a bond paying coupon c at time 1, 2, . . . , M
and having face value of 1 is given by the function:
In[20]:= bondPrice[c , r , M ] :=
Sum[P[r, 0, t] c, {t, 1, M}] + P[r, 0, M]
It is often convenient to consider continuously paid out
coupons, in which case the above formula becomes:
In[21]:= bondPrice[c , r , M ] :=
Integrate[P[r, 0, t] c, {t, 0, M}]
+ P[r, 0, M]
A European option on a bond can also be priced analyt-
ically with the Vasicek term structure model. The deriva-
tion of the option pricing formula and some applications
can be found in [Jamshidian 1989].
6. THE COX-INGERSOLL-ROSS TERM STRUCTURE
MODEL
CIR consider an interest rate process of the type:
dr = a(g - r)dt + s
0
rdB
where a > 0, g = long-run mean interest rate, r = current
interest rate. Consider a pure discount, default-free bond
which promises to pay one unit at time T . We denote
the price of this bond at time 0 < t < T by P(r, t , T ). It
follows from Itos formula that
dP =
P
r
dr +
P
t
dt +
rs
2
2

2
P
r
2
dt
Substituting for dr gives
dP =
P
r
Ia(g - r) dt +s
0
r dB M +
P
t
dt +
rs
2
2

2
P
r
2
dt
Dividing by dt and taking expectations gives
E C
dP
dt
G = a(g - r)
P
r
+
P
t
+
rs
2
2

2
P
r
2
Now the right-hand side, representing the expected
rate of return on the bond over a small instant of time,
is proportional to the risk-free interest rate and to a risk-
adjusted interest elasticity of the bond. Denoting by k the
covariance of the changes in the interest rate with the
market portfolio, we nd that:
E C
dP
dt
G = rP(1 + kP
r
P).
Thus the basic differential equation for the bond price in
the CIR model is given by:
rP(1 + kP
r
P) = a(g - r)
P
r
+
P
t
+
rs
2
2

2
P
r
2
It can be shown that a solution to this equation is given
by
P(r, t , T ) = A(t , T )e
-B(t ,T )r
Vol. 7 No. 2 1998 Mathematica in Education and Research 7
T E R M S T R U C T U R E O F I N T E R E S T R A T E S
where
A(t , T ) = C
2he
(a+h)(T -t )/ 2
(a + h)(e
h(T -t )
- 1) + 2h
G
2ag/ s
2
B(t , T ) =
2(e
h(T -t )
- 1)
(a + h)(e
h(T -t )
- 1) + 2h
and
h =
0
a
2
+ 2s
2
It follows that in this model the bond prices are log-
normally distributed with parameters
dP
P
= m(r, t ) dt +s(r, t ) dB
where
m(r, t ) = r (1 - kB(t , T )), s(r, t ) = -B(r, T )s
0
r
As the time to maturity lengthens, the yield to maturity
in the CIR model approaches
R(r, t , ) =
2ag
a + k + y
where k is the market price of risk and
y =
1
(a + k)
2
+ 2s
2
The CIR models term structures that are usually up-
ward or downward sloping, although it, can, within nar-
row limits, produce humped term structures. Here are
some examples:
In[22]:= Clear[A, B, P, pureDiscountRate,
, , , ];
A[alpha , gamma , sigma , t , T ] :=
Module[{ = Sqrt[2 + 2]},
((2 (Exp[( + ) (T t)/2]))
/(( + )
(Exp[ (T t)] 1) + 2 ))
(2 /2)
];
B[alpha , gamma , sigma , t , T ] :=
Module[{ = Sqrt[2 + 2]},
((2 (Exp[ (T t)] 1))
/(( + )
(Exp[ (T t)] 1) + 2 ))
(2 /2)
];
P[r , t , T , alpha , gamma , sigma ] :=
A[, , , t, T]
Exp[ B[, , , t, T] r];
pureDiscountRate[
r , T , alpha , gamma , sigma ] :=
If[T == 0, r,
Log[P[r, 0, T, , , ]]/T];
Plot[pureDiscountRate[
0.12, T, 1, 0.08, 0.05], {T, 0, 10},
PlotLabel "CIR Term Structure",
AxesLabel {"maturity", ""}];
2 4 6 8 10
maturity
0.025
0.03
0.035
0.04
0.045
0.05
CIR Term Structure
Simulating the CIR spot interest rate process
We can repeat the simulation we performed above for Va-
sicek, simulating the CIR short-term interest rate process:
In[23]:= ( the model is CIR :
dr = ( r) dt + Sqrt[r] dz
is the long run interest rate;
r is the current rate )
Clear[deltaR, shortTermR, , , ];
deltaR[
alpha , gamma , r , sigma , deltaT ] :=
( r) deltaT+
nor[0, 1] Sqrt[deltaT] Sqrt[r];
shortTermR[
alpha , gamma , r , sigma , deltaT ] :=
r + deltaR[, , r, , deltaT];
SeedRandom[3];
lst = NestList[shortTermR[1, 0.04,
#, 0.02, 0.08333] & , 0.04, 300];
ListPlot[lst, PlotJoined True];
50 100 150 200 250 300
0.036
0.038
0.042
0.044
0.046
0.048
Compare the graph for the Vasicek and the CIR simula-
tion: Both were produced with the same series of random
normal deviates. Whereas the Vasicek can produce nega-
tive interest rates, the CIR cannot. Heres a repeat of our
simulationwhich produced negative interest rates with
Vasicekfor the CIR model:
In[24]:= SeedRandom[2];
lst = NestList[shortTermR[0.03,
0.04, #, 0.12, 0.004] & , 0.03, 300];
ListPlot[lst, PlotJoined True,
PlotLabel CIRSimulation];
8 Mathematica in Education and Research Vol. 7 No. 2 1998
T E R M S T R U C T U R E O F I N T E R E S T R A T E S
50 100 150 200 250 300
0.015
0.025
0.03
0.035
0.04
CIR Simulation
7. OTHER TERM STRUCTURE MODELS
In addition to theVasicek and the Cox-Ingersoll-Ross mod-
els of the term structure, there are a variety of other mod-
els:
1. Merton, 1973, dr = bdt + sdB
2. Vasicek, 1977, dr = a(g - r)dt + sdB
3. Dothan, 1978, dr = brdt + srdB
4. Cox, Ingersoll, Ross (CIR), 1985, dr = a(g - r)dt +
s
0
rdB
5. Ho, Lee, 1986, dr = q(t )dt + sdB
6. Hull, White (extended Vasicek), 1990, dr = (q(t ) -
br)dt + sdB
7. Hull, White (extended CIR), 1990,
dr = (q(t ) - br)dt + s
0
r dB
8. Black, Karasinski, 1991, d logr = (q(t )-b logr)dt +sdB
In addition to these spot interest rate models there are
forward interest rates models like [Heath, Jarrow, Mor-
ton 1992] approach and many discrete schemes. Our next
article will explore two of these discrete models.
REFERENCES
nixxixca, s. and a. iioroiaiaoaxis (1986). General EquilibriumProp-
erties of the Term Structure of Interest Rates, Journal of Financial Eco-
nomics 16, 389410.
nixxixca, s. and z. wiixii (1996). An Investigation of Cheapest to
Deliver on Treasury Bond Futures Contracts. Mimeo.
niacx, i. and i. xaiasixsxi (1991), Bond and Option Pricing when
Short Rates are Lognormal. Financial Analysts Journal (JulyAugust),
5259.
cox, ;.c., ;.i. ixciisoii, and s. a. ioss (1985). A Theory of the Term
Structure of Interest Rates. Econometrica 53, 385402.
ooruax, i. uii (1978). On the Term Structure of Interest Rates. Journal
of Financial Economics 6, 5969.
iisuii x. and o. ziivos (1996), Yield Curve, in Computational Eco-
nomics and Finance, Modeling and Analysis with Mathematica, ed. H.
R. Varian, TELOS/Springer-Verlag, pp. 269302.
uiaru o., i. ;aiiow and a. xoirox (1992), Bond Pricing and the
Term Structure of Interest Rates: A New Methodology for Contingent
Claims Valuation, Econometrica 60, 77105.
uo, r. s. \. and s.n iii (1986). Term Structure and Pricing Interest
Rate Contingent Claims, Journal of Finance 41, 101129.
uuii, ;. and a. wuiri (1990). Pricing Interest Rate Derivative Securities.
Review of Financial Studies 3, 57392.
;axsuioiax, iaisuio (1989). An Exact Bond Option Formula. Journal
of Finance 44 (March), pp. 205209.
xccuiiocu, ;. u. (1990). "U.S. Term Structure Data, 19461987." Hand-
book of Monetary Economics, Volume I, pp. 672715, Amsterdam: North
Holland.
xiirox, ioniir c. (1973). An Intertemporal Capital Asset Pricing Model.
Econometrica 41 (September), pp. 867886.
vasicix, oioiicu a. (1977). An Equilibrium Characterization of the
Term Structure. Journal of Financial Economics 5 (November), pp. 177
88.
ABOUT THE AUTHORS
The authors acknowledge grants from Wolfram Research, the Krueger
and Eshkol Centers at the Hebrew University, and the Israeli Academy
of Science. Wieners research has beneted from a grant from the Israel
Foundations Trustees and the Alon Fellowship.
Simon Benninga is professor of nance at Tel-Aviv University (Israel) and
the Wharton School of the University of Pennsylvania. He is the author
of Financial Modeling (MIT Press, 1997) and of Corporate Finance: A
Valuation Approach (with Oded Sarig, McGraw-Hill, 1997); he is also
the editor of the European Finance Review.
Simon Benninga
Faculty of Management
Tel-Aviv University, Tel-Aviv, Israel
benninga@post.tau.ac.il
http://nance.wharton.upenn.edu/benninga
Zvi Wiener is assistant professor of nance at the business school of the
Hebrew University of Jerusalem. His nance research concentrates on
the pricing of derivative securities, value-at-risk, computational nance
and stochastic dominance. He wrote this article while visiting at the Olin
School of Business at Washington University in St. Louis.
Zvi Wiener
Finance Department, Business School
Hebrew University, Jerusalem, Israel
mswiener@mscc.huji.ac.il
http://pluto.mscc.huji.ac.il/mswiener/zvi.html
ELECTRONIC SUBSCRIPTIONS
Included in the distribution for each electronic subscription is the le
termStruct.nb containing Mathematica code for the material de-
scribed in this article.
Vol. 7 No. 2 1998 Mathematica in Education and Research 9

You might also like