See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/24071808

Markov interest rate models

Article in Applied Mathematical Finance February 1999
DOI: 10.1080/13504869950079275 Source: RePEc

CITATIONS

READS

237

2 authors, including:
Patrick S Hagan
AVM LP
78 PUBLICATIONS 1,315 CITATIONS
SEE PROFILE

Available from: Patrick S Hagan

Retrieved on: 29 August 2016

Markov Interest Rate Models

Patrick S. Hagan

Diana E. Woodward

October 27, 1997

Original draft: October 15, 1996

Abstract
We present a general procedure for creating Markovian interest rate models. The models
created by this procedure automatically t within the HJM framework and t the initial term
structure exactly. Therefore they are arbitrage free. Because the models created by this procedure have only one state variable per factor, two- and even three-factor models can be computed
eciently, without resorting to Monte Carlo techniques. This computational eciency makes
calibration of the new models to market prices straightforward.
Extended Hull-White, extended CIR, Black-Karasinski, Jamshidian's Brownian path independent models, and Flesaker and Hughston's rational log normal models are one state variable
models which t naturally within our theoretical framework. The \separable" n-factor models of Cheyette and Li, Ritchken, and Sankarasubramanian | which require n(n + 3)=2 state
variables | are degenerate members of the new class of models with n(n + 3)=2 factors.
We use our procedure to create a new class of one-factor models, the \  models." These
models can match the implied volatility smiles of swaptions and caplets, and thus enable us
to eliminate smile error from our books. The   models also exactly solvable in that their
transition densities can be written explicitly. For these models we present accurate | but not
exact | formulas for caplet and swaption prices, and indicate how these closed form expressions
can be used to eciently calibrate the models to market prices.

 The views presented in this report do not necessarily re ect the views of NumeriX, The Bank of Tokyo-Mitsubishi,
or any
of their aliates.
y Current mailing address: NumeriX, 546 Fifth Avenue, 17th Floor, New York, NY 10036.
z Current mailing address: The Bank of Tokyo-Mitsubishi, Ltd., 1251 Avenue of the Americas, New York, NY
10020.

Introduction.

In [1, 2, 3] Heath, Jarrow and Morton created a broad framework for developing arbitrage-free
term structure models. In general, HJM models are non-Markovian and require extensive Monte
Carlo simulation to calibrate model parameters to market prices, to value contingent claims, and to
determine hedges. This computational burden can be greatly reduced by using special cases of the
HJM models which are Markovian. Notable among these special cases are the \separable" n factor
models of Cheyette [6] and Li, Ritchken, and Sankarasubramanian [4, 5]. Although the \separable"
n factor models are Markovian, they require n(n + 3)=2 state variables, which still imposes a sti
computational burden. Moreover, to date there has been no systematic procedure for nding HJM
models which are Markovian.
Here we use the method of the undetermined numeraire to develop a general procedure for
creating Markovian term structure models. The resulting models automatically t within the HJM
framework and match the initial term structure. Thus they are arbitrage free. Extended HullWhite [8, 9], extended CIR [10], Black-Karasinski [12], Jamshidian's Brownian path independent
models [14], and Flesaker and Hughston's rational log normal models [15] all t naturally within our
theoretical framework.
Unlike the \separable" models, the new models require only n state variables for an n factor
model. This greatly reduces their computational complexity, allowing two and even three factor
models to be used without requiring Monte Carlo simulations. Although we can re-create the
separable models within our framework as degenerate models with n(n + 3)=2 factors and state
variables, we can also create models which are direct analogues of the separable models and require
only n factors and state variables. These analogues are asymptotically equivalent to their separable
counterparts in the limit of small volatilities, and, at normal market place volatilities, the two sets
of models should give nearly identical results.
We use our procedure to create a new class of one factor models, the \  models." Unlike other
one-factor models, these models can match the implied volatility smiles of swaptions and caplets,
and thus enable us to largely eliminate smile error from our books. The   models are also
exactly solvable in that their transition densities (Green's functions) can be written exactly. For
these models we present accurate | but not exact | formulas for caplet and swaption prices, and
indicate how these closed form expressions can be used to eciently calibrate the models to exact
market prices.
2

Derivation.

Consider a continuous trading economy on the interval [0; Tmax]. To x notation, let f0(T ) be
today's instantaneous forward rate for date T for this economy. Then the current value of a zero
coupon bond which pays $1 at maturity T is
(2.1)

D(0; T )  e

RT
0

f0 (T 0 ) dT 0 ;

and the current discount factor from time t to T is D(t; T )  D(0; T )=D(0; t).
To model this economy, recall that arbitrage-free term structure models have three essential
elements. First is a set of stochastic processes which drive the evolution of interest rates. Second
is a numeraire, a tradable instrument with positive value and no cash ows. Commonly the money
market account or a specic zero coupon bond is used as the numeraire. The nal element is a
valuation formula which states that, in the absence of any cash ows, the value of any tradable
1

instrument1 is a Martingale when expressed in units of the numeraire [16, 17].

In a Markovian model, the entire term structure at any time t must be determined solely by the
values x1 ; x2 ; : : : ; xn of a nite set of random state variables X1 (t); X2 (t); : : : ; Xn (t). We assume
that these state variables evolve according to the drift-diusion processes
ci (t); Xi (0) = 0; i = 1; 2; : : : ; n:
dXi (t) = i (t; X ) dt + i (t; X ) dW

(2.2a)

c1 (t); W
c2 (t); : : : ; W
cn (t) represent n Brownian motions, which can be correlated
Here W
ci (t)dW
cj (t) = ij (t) dt:
dW

(2.2b)

To obtain a Markovian model, we need to choose a numeraire whose value N depends only on the
values x1 ; x2 ; : : : ; xn of the state variables: N  eH (t;x) . We stress that at this point, the function
H (t; x) is essentially arbitrary and we do not need to know which nancial instrument is represented
by the numeraire. Simply requiring N (t; x) to be a market instrument (with no cash ows) will
enable us to derive all interest rates.
To express the valuation formula, consider a tradable instrument which has the value V (T; X (T ))
at some date T and pays a cash ow C (t; X (t)). The value of the instrument expressed in units of
the numeraire is V (t; x)=N (t; x)  V (t; x)e H (t;x) . Requiring V (t; x)=N (t; x) to be a Martingale in
the absence of any cash ows then yields
(2.3) V (t; x)

= eH (t;x) Eb

H (T;X (T )) +

V (T; X (T ))e

Z T

C (t0 ; X (t0 ))e

H (t0 ;X (t0 )) dt0 

X (t) = x ;

where x = X (t) is the state of the economy at time t. Equation (2.3) gives the value of the
instrument at any time t < T . Note that Eb is the expected value under the probability measure 2
determined by (2.2).
Without loss of generality, we re-write the numeraire as

N (t; x) =

(2.4a)

1
eh(t;x)+A(t) ;
D(0; t)

where

h(t; 0)  0; A(0) = 0:

(2.4b)
Then the valuation formula becomes
n

(2.5)

V (t; x) = eh(t;x)+A(t) Eb V (T; X (T ))D(t; T )e

+

Z T

h(T;X (T )) A(T )

C (t0 ; X (t0 ))D(t; t0 )e

h(t0 ;X (t0 )) A(t0 ) dt0 

X (t) = x :

The valuation formula (2.5), with the state variables X (t) determined by the stochastic process
(2.2), denes a term structure model. As we shall see, if this model is consistent with the initial
term structure then it is arbitrage free. In particular, this model automatically satises the crucial
1 The term \tradable instruments" is used to refer to both fundamental instruments that are actively traded and
all 2derivatives and synthetic instrument that can be created.
Eq. (2.2) thus denes the risk neutral probability measure induced by the numeraire ( x), and does not
represent \real world" probabilities.
N t;

\forward rate drift restriction" [1] of HJM theory3 . We stress that apart from some mathematical
regularity conditions, the drift rates i (t; x), the volatilities i (t; x), and the function h(t; x) are
arbitrary. The remaining function A(t) will be determined by requiring the valuation formula (2.5)
to be consistent with the initial term structure.
To impose consistency, dene Z (t; x; T ) to be the value at t; x of a zero coupon bond which pays
$1 at maturity T . From (2.5),

Z (t; x; T ) = D(t; T )eh(t;x)+A(t)

(2.6a)
where

S (t; x; T )  Eb e

(2.6b)

h(T;X (T )) 

A(T ) S (t; x; T );

X (t) = x :

As we are taking X (0) = 0, consistency with the initial term structure requires Z (0; 0; T )  D(0; T ).
Since h(t; 0) = A(0) = 0, this requirement becomes


A(T ) = log S (0; 0; T ) = log Eb e

(2.7)

h(T;X (T )) 

X (0) = 0

for all T > 0:

Thus consistency with the initial term structure requires choosing A(T ) so that the expected value
of e h(T;X (T )) A(T ) is 1 for all T > 0.
Note that the current discount factors D(0; T ) do not appear in (2.7); choosing A(T) so that the
model is consistent with the initial term structure is independent of the actual initial term structure.
Once A(T ) has been selected according to (2.7), then if the initial term structure changes, all we
need do is replace the discount factors D (t; T ) with the new discount factors, and the model will be
consistent with the new initial term structure. This is useful during calibration since it separates
the volatility functions i (t; X ), i (t; X ), and h(t; X ) endogenous to the model from the current
discount factors D(t; T ), which are exogenous.
With the economy in state X (t) = xR at date t, the instantaneous forward rates f (t; x; T ) are
dened implicitly by Z (t; x; T ) = expf tT f (t; x; T 0) dT 0 g. From (2.6),
(2.8)

Z T

f (t; x; T 0 )dT 0 =

Z T

f0 (T 0 )dT 0 h(t; x) A(t) + A(T ) log S (t; x; T ):

Hence the forward rate curve at t; x is

f (t; x; T ) = f0 (T ) + A0 (T ) ST (t; x; T )=S (t; x; T );

where we are using subscripts to denote partial derivatives. The short rate r(t; x) is dened as
f (t; x; t), so we have
(2.10a)
r(t; x) = f0 (t) + A0 (t) ST (t; x; t)=S (t; x; t):
(2.9)

In Appendix A it is shown that this expression for the short rate is equivalent to
X
r(t; x) = f0 (t) + A0 (t) + ht (t; x) + i (t; x)hx (t; x)
(2.10b)



1X
 (t) (t; x)j (t; x) hxixj (t; x) hxi (t; x)hxj (t; x) :
2 ij ij i

3 This is not surprising: (2.2) and (2.5) self-consistently dene the probability measure of the risk neutral world
under a specic (if unknown) numeraire ( x); this then uniquely denes the risk neutral probability measure under
every other numeraire, including the money market numeraire; and HJM [1] showed that for any self-consistent riskneutral world, the instantaneous forward rates must satisfy the forward drift restriction under the probability measure
induced by the money market numeraire.
N t;

With A(t) dened by (2.7), the valuation formula (2.5) and random processes (2.2) uniquely
dene a term structure model which is consistent with the initial term structure. The instantaneous
forward interest rates and the short rate for this model are given by (2.9) and (2.10), respectively. At
this point it would be natural to use Martingale theory [16, 17] to show that this model is arbitrage
free. Instead, in Appendix A we show that this model ts within the HJM framework. There it is
found that using the money market as a numeraire, the process for the instantaneous forward rate,
F (t; T )  f (t; X (t); T ), satises
(2.11a)

dF (t; T ) =

ij

ij (t)aiT (t; X ; T )aj (t; X ; T ) dt

fi (t)
aiT (t; X ; T ) dW

in the risk neutral probability world, with

(2.11b)

ai (t; x; T ) = i (t; x)

Sxi (t; x; T )
S (t; x; T )

Sxi (t; x; t)
; i = 1; 2; : : : ; n:
S (t; x; t)

Inspection shows that the drift terms in (2.11a) satisfy the forward drift restriction of HJM theory
[1]. Consequently, any term structure model given by (2.5), (2.2), and (2.7) ts within the HJM
framework, and provided it satises the innocuous regularity conditions in [1], the model is arbitrage
free4 .
Equation (2.5) can be used to directly value any contingent claim whose \payo" V (T; X (T ))
and cash ow C (t; X (t)) can be expressed as explicit functions of the state variables. Since (2.9)
expresses the entire forward rate curve f (t; x; T ) explicitly, this includes all instruments whose payo
and cash ow depend only on the then current yield curve. This includes European, Bermudean,
and American swaptions, caps, and most yield curve options. However, these is no guarantee
that instruments whose payo or cash ow are path dependent (e.g., instruments with payments
determined by the average short rate over the preceding period) can be evaluated without adding
extra state variables or resorting to path dependent valuation techniques.
3

One Factor Models.

We now use the general framework for arbitrage-free Markovian models developed in Section 2 to
create a special class of one factor models. In Section 4 we will further specialize this class of models
to obtain the   models.
Although one factor models are usually expressed in terms of the money market numeraire using
the short rate r as the state variable, any such \short rate model" can be re-cast within the framework
of Section 2 by using a zero coupon bond as the numeraire. Conversely, any one factor model within
the framework of Section 2 can be transformed into a short rate model. These transformations are
carried out in Appendix B.
Within the framework of Section 2, a one factor model is dened by the stochastic process for
the state variable,
(3.1)

c (t); X (0) = 0;
dX (t) = (t; X ) dt + (t; X ) dW

We do not restate these conditions here; we refer the reader to [1]. Although these conditions are innocuous
mathematically, each points out a more stringent business condition. For example, the mathematical requirement
\market completeness" points out that risks must be hedgeable with liquid instruments with minimal transaction
costs.
4

and the numeraire

N (t; x) =

(3.2)

1
eh(t;x)+A(t);
D(0; t)

which creates the link between the state variable and nancial markets.
We argue that with our current knowledge about interest rates, we are not justied in using
both a complicated function h(t; x) in the numeraire and a complicated stochastic process for X (t).
Instead we should choose either a relatively simple function h(t; x) or a relatively simple stochastic
process. We arbitrarily choose to use a simple numeraire
1
(3.3)
N (t; x) =
e(t)x+A(t)
D(0; t)
for our model. The consistency condition (2.7) then becomes


A(t) = log Eb e

(3.4)

(t)X (t) 

X (0) = 0 :

Now consider the drift term, (t; X )dt. Suppose the model had, say, a positive drift so that X (t)
increased on average. Then A(t) would be negative and would eectively cancel out the average
eect of the drift term in the numeraire. Similarly, if the drift term was negative, A(t) would be
positive, again canceling out most of the drift. Indeed (3.4) can be written as


Eb e

(3.5)

(t)X (t) A(t) 

X (0) = 0 = 1:

Since A(t) cancels out the main eects of the drift term, any residual in uence of the drift term
(t; X )dt on the model would be quite subtle. Therefore, in the interests of simplicity, we take our
model to be driftless.
So consider the special class of one factor models
c (t); X (0) = 0;
dX (t) = (t; X ) dW

(3.6)

with the numeraire (3.3). As shown in Appendix C, this class of models includes the extended
Hull-White, the extended CIR, and \Black-Karasinski like" models. However, the Black-Karasinski
model itself cannot be included without adding a drift term to (3.6).
With the numeraire (3.3), the valuation formula becomes
n

(3.7)

V (t; x) = e(t)x+A(t) Eb V (T; X (T ))D(t; T )e

+

Z T

(T )X (T ) A(T )

C (t0 ; X (t0 ))D(t; t0 )e

(t0 )X (t0 ) A(t0 ) dt0 

so the value of a zero coupon bond becomes

Z (t; x; T ) = D(t; T )e

(3.8)
Here we have dened
(3.9a)

M (t; x; T ) = log Eb e

[(T ) (t)]x [A(T ) A(t)]+M (t;x;T )

(T )[X (T ) x] 

(3.9b)

X (t) = x = (T )x + log S (t; x; T );

and the consistency condition is now

A(T ) = M (0; 0; T ) for all T :

5

X (t ) = x ;

R
It is instructive to examine the forward rate curve. Since Z (t; x; T ) = expf tT f (t; x; T 0 ) dT 0g,
eq. (3.8) shows that if the economy is in state X (t) at time t, then the instantaneous forward rate
for date T would be
(3.10a)
f (t; X (t); T ) = f0 (T ) + 0 (T )X + A0 (T ) MT (t; X ; T ):

The corresponding short rate would be

(3.10b)

r(t; X (t)) = f0 (t) + 0 (t)X + A0 (t) MT (t; X ; t):

For the moment, let us neglect the last two terms in (3.10a) and (3.10b). Then the forward
rate curve is just the curve 0 (T )X added to the original forward curve f0 (T ). Note that 0 (T ) is a
`gearing' factor: if the state variable X (t) is shifted by an amount
, then the short rate would shift
by r(t) = 0 (t)
and the forward rates f (t; X ; T ) would shift by f = 0 (T )
= 0 (T )r=0 (t).
Since we expect shocks to aect short maturities more than longer maturities, we expect 0 (T ) to
be a decreasing function of T .
Note that eq. (3.6) implies that X (t) is a Martingale, so for all times T > t, the expected value
of X (T ) is just X (t). Thus, if X (t) was shifted by an amount
, then for all later times T the
expected value of X (T ) would also shift by
. Consequently, if a shock shifted the short rate by
r(t) = 0 (t)
, then for all later times T the expected value of the short rate r(T; X (T )) would shift
by 0 (T )r=0 (t). As T increases, the expected value of the short rate returns to the initial forward
curve f0 (T ) at a rate determined by how rapidly 0 (T ) decreases. So mean reversion of interest rates
is directly determined by the function (T ).
With f (t; X ; T ) = f0 (T ) + 0 (T )X + : : : , the distribution of forward rates is determined by the
distribution of X (t). The width of this distribution is determined mainly by the overall magnitude of
(t; x), and the shape of this distribution (i.e., the deviation from Gaussian or log normal behavior)
is determined mainly by the functional form of the dependence of (t; x) on x. As noted above,
mean reversion of interest rates is determined mainly by the function (T ). Since skews and smiles
(the changes in an option's implied volatility as the strike changes) depend mainly on the shape of
the distribution, and since the change in the implied volatility as the duration of the underlying
instrument changes depends mainly on mean reversion, this separation makes it easy to calibrate
these models to both at-the-money and o-market instruments of diering durations. This makes
these models very useful for pricing, hedging, and understanding instruments in the presence of
implied volatility skews and smiles.
The third term A0 (T ) in (3.10) embodies the consistency requirement, and is much smaller than
the rst two terms. Even though (3.6) implies that X (t) has mean zero, the convex relation between
bond prices and interest rates would cause bond prices to drift, on average, as the distribution of
X (t) spreads out. The term A0 (T ) cancels the expected value of this convexity eect. This is why
the \consistency" term A(T ) depends on the functions (T ) and (t; x), but not on the initial term
structure f0 (T ). The last term MT (t; x; T ) arises because as X (t) evolves away from X (0) = 0, the
expected value of the convexity eect changes.
We can clarify the relation between the distribution of X (T ) and the convexity terms A(T ) and
M (t; x; T ) by expanding (3.9) in powers of . Dene the moments
(3.11)

mk (t; x; T ) = Eb [X (T ) x]k  X (t) = x ; k = 1; 2; : : :

and note that m1 (t; x; T ) = 0. Expanding (3.9) yields

(3.12a)
M (t; x; T ) = 21 2 (T )m2(t; x; T )

 T )m3 (t; x; T ) + 241 4 (T )[m4 (t; x; T ) 3m22 (t; x; T )] + : : : ;

1 3(
6

(3.12b)
A(T ) = 21 2 (T )m2 (0; 0; T )

 T )m3(0; 0; T ) + 241 4 (T )[m4 (0; 0; T ) 3m22(0; 0; T )] + : : : :

1 3(
6

Our experience calibrating one factor models to US swaption and caplet markets shows that the
magnitude of the quadratic term 12 2 (T )m2 (t; x; T ) is roughly (T t)T 2  10 4, where time is measured in years. The cubic and quartic terms, which arise from the skewness and kurtosis of the distribution, are much smaller and have magnitudes of roughly (T t)2 T 3  10 8 and (T t)2 T 4  10 9,
respectively5. We conclude that it is safe to truncate after the quadratic term except for unusually
sensitive instruments, unusually long durations, or unusually competitive markets.
Truncating after the quadratic terms yields the forward rates

1 @  2
f (t; x; T ) = f0 (T ) + 0 (T )x +
 (T )[m2 (0; 0; T ) m2 (t; x; T )] + : : : :
2 @T
Consequently, the expected forward rate is



(3.14a)
Eb f (t; X (t); T )  X (0) = 0 = f0 (T ) + 0 (T )(T )m2 (0; 0; t) + : : : :

(3.13)

Note that m2 (0; 0; t) is the variance of X (t), not X (T ). The covariance between forward rates at
dierent maturities is



(3.14b)
Cov f (t; X (t); T1); f (t; X (t); T2)  X (0) = 0 = 0 (T1 )0 (T2 )m2 (0; 0; t) + : : : ;
showing that the two rates are perfectly correlated through this order.
Before continuing, let us brie y address the \unknown numeraire" in (3.2). By construction,
N (t; x) is always a valid numeraire6. Suppose we arbitrarily set (Tref ) = 0 for some date Tref .
Then (3.9) would imply that A(Tref ) = 0 and M (t; x; Tref ) = 0, so the value of the zero coupon
bond of maturity Tref would be D(t; Tref )e(t)+A(t) . See (3.8). So if we made this choice, then the
\unknown numeraire" N (t; x) would represent a zero coupon bond paying 1=D(0; Tref ) dollars at
maturity Tref | at least for times t  Tref . Although N (t; x) would remain a valid numeraire for
times t > Tref , it would no longer represent any simple security. In the same spirit, we could restrict
the (t) curve so that (t) ! 0 as t ! 1; then the numeraire would correspond to Flesaker and
Hughston's \absolute terminal measure" [15].
4

The

Models.

We now specialize to volatilities of the form (t; x) = (t)[1 + X (t)] , where  and  are constant,
and dene the   models by
c (t); X (0) = 0;
dX (t) = (t)[1 + X (t)] dW

(4.1a)
with the numeraire

N (t; x) =

(4.1b)

1
e(t)x+A(t):
D(0; t)

5 This is equivalent to the change in interest rates over a period
having a standard deviation of size
(100bps[

yrs]1=2 ), variance of size (10 4
yrs), kurtosis of size (1), and skewness of size (10 2 [
yrs]1=2 ).
6 Substituting (
)  ( ) and ( )  0 into the valuation formula (2.5) shows that for any final 0,
( ) is the value of the interest rate option which has no cash ow and has the \payo" ( final ( final )) at
final .
t

t=

t; x

N t; x

t=

t=

N t; x
T

C t; x

N T

;X T

>

Without loss of generality, we normalize (t) so that 0 (0) = 1.

A key advantage of   models is that there is a closed form solution for the transition density


p(t; x; t; x) dx = Prob x < X (t) < x + dx  X (t) = x :

(4.2)

The transition density p satises the backward equation

(4.3a)

pt + 12 2 (t)j1 + xj2 pxx = 0; t < t

(4.3b)

p ! (x x) as t ! t:

In terms of the new variables

 (t) =

(4.4)

Z t
0

j1 + xj1  ;
2 (t0 )dt0 ; y(x) =
 (1 )

this becomes

p +

(4.5a)

1
p
2 yy

p ! [ (1 )y]

(4.5b)

2

2 +1

py = 0;  < 

(y y) as 

! 

with  = 1=(2 2). In particular,  (t) is roughly the variance of X (t), which is around 10 4t in
US dollar markets, where t is measured in years.
Equation (4.5) can be solved by a combination of Laplace transform and Green's function techniques. For exponents 0 <  < 1=2 this yields
(4.6a)

y  1
p(t; x; t; x) = (y=y)
I  (yy=(  )) + 21 I (yy=(  )) e
2
 
(4.6b)

p(t; x; t; x) =

sin 
y
(y=y)
K (yy=(  ))e

  

(
y2 +y2 )=2(  )

(
y2 +y2 )=2(  )

dy
; for x > 1=
dx

dy
; for x < 1=
dx

Here

dy
= [(1 ) y] 2 +1 ;
dx
and K and I are the modied Bessel functions.
The   models have an articial barrier at x = 1= , where the volatility j1 + xj goes
to zero. For exponents  < 1=2, the volatility does not decay fast enough as y ! 0 to prevent
y from becoming negative. If we wished to avoid negative values of y, we could put a re ecting
boundary condition at y = 0; this would eliminate (4.6b) and change the Bessel functions in (4.6a)
from 21 (I  + I ) to I  .
2
2
The probability of y approaching zero is extremely small, of order e 1= (1 )  . This is much
too small to aect the pricing of realistic products. Since  and  can be chosen to reproduce the
volatility smile reasonably well, indicating that the model is reproducing the right shape of the risk
neutral probability distribution near the distribution's center, we argue that the benet of increased
(4.7)

accuracy near the distribution's center far outweighs the liability of having an articial barrier in
the extreme tails of the distribution. In fact, we do not expect any model to be valid in the extreme
tails of the distribution where there is no relevant market experience.
For exponents 1=2   < 1 we have

y
(4.8) p(t; x; t; x) = (y=y)
I (yy=(  ))e
  

(
y2 +y2 )=2(  )

dy
(; y2 =2(  ))
+
(x + 1= )
dx
( )

where (; ) is the incomplete gamma function. For these exponents the volatility declines fast
enough to prevent y from becoming negative, but it doesn't prevent y from reaching the origin. This
causes a very small, but nite, probability of y being exactly zero. As before, this probability is
much too small to aect the pricing of realistic deals.
For both sets of exponents, the transition probability can be expanded as
(4.9)

(y=y) 1=2
p(t; x; t; x) =
e
[2(  )]1=2

(
y y)2 =2(  )

dy 
1
dx

4 2 1
(  ) + : : :
8yy

for small values of the variance   .

Finally, for exponent  = 1 we need to re-dene y as

y (x ) = 

(4.10a)

log(1 + x):

Then the transition density is simply

(4.10b)

p(t; x; t; x) =

e y
e
[2(  )]1=2

(
y y+ (  )=2)2 =2(  )

With the transition density in hand, M (t; x; T ) can be found by integration,

(4.11a)

M (t; x; T ) = log

nZ

p(t; x; T; x)e

(T )(x x) dx


Close inspection of (4.5) and (4.11a) reveals that M (t; x; T ) is of the form
(4.11b)

M (t; x; T )  M (S; );

where
(4.11c)

) ;

S =  2 (  )=(1 + x)2(1

 = (T )(1 + x)=:

This is useful since for each exponent , the function M (S; ) can be calculated and tabulated once,
eliminating most of the computational burden of the   model. Once these tables have been
calculated, the values Z (t; x; T ) of zero coupons bonds, the forward rate curve f (t; x; T ), and the
short rate r(t; x) can be obtained directly from (3.8)-(3.10).
Three exponents merit special attention for their simplicity:  = 0,  = 1=2, and  = 1. In
Appendix C we analyze the   model for these exponents, allowing  to depend on time:
(4.12)

c (t):
dX (t) = (t)[1 +  (t)X (t)] dW

Here we quote only the results with  constant.

9

When  = 0, the   model becomes the extended Hull-White model. In this case A(T ) and
M (t; x; T ) are given by

A(t) = 21 2 (T ) ; M (t; x; T ) = 12 2 (T )(  );

(4.13)
where

 =

(4.14)

Z T
0

2 (t0 )dt0 ;  =

Z t

2 (t0 )dt0 :

Equations (3.8)-(3.10) now yield explicit expressions for the value of zero coupon bonds, the forward
rate curve, and the short rate.
Similarly, A(T ) and M (t; x; T ) can be written explicitly when  = 1=2,
(1 + x)2 (T )(  )
2 (T )
; M (t; x; T ) =
;
2 + (T )
2 + (T )(  )
where  and  are given by (4.14). This again enables Z (t; x; T ), f (t; x; T ) and r(t; x) to be obtained
from (3.8)-(3.10). Besides deriving the corresponding formulas when  =  (t), in Appendix C we
nd that  (t) can be chosen so that this model is exactly the CIR model. Alternatively, one can
use the extra freedom of choosing  (t) to calibrate the model to a series of o-market instruments
as well as a series of at-the-money instruments, and thus reproduce the volatility smile.
The exponent  = 1 (\Black-Karasinski like" models) is more dicult. Although we have been
unable to derive explicit expressions for A(T ) and M (t; x; T ), the moments mk (t; x; T ) can be written
explicitly. When  is constant,

A(T ) =

(4.15)

(4.16a)
(4.16b)

m2 (t; x; T ) = [(1 + x)= ]2 [e

m3 (t; x; T ) = [(1 + x)= ]3 [e3
m4 (t; x; T ) = [(1 + x)= ]4 [e6

(4.16c)

2 (


)

2 (


2 (


)

4e3

)

3e
2 (


1];
2 (


 ) + 2];

 ) + 6e 2 (  )

3]:

We can use the rst few terms in expansion (3.12) for M (t; x; T ) and A(T ), and substitute these
expansions into (3.8) and (3.10) to obtain Z (t; x; T ), f (t; x; T ), and r(t; x). Even if only the second
moment is used, this should be accurate enough to price all but the most sensitive instruments.
4.1

European option prices.

A vanilla receiver swaption is a European option which gives the holder the right to receive a predetermined series of cash payments Ci on dates ti ; i = 1; 2; : : : ; n, in return for essentially7 paying
K on the settlement date ts . A payor swaption is a European option which essentially gives the
holder the right to receive the strike K at settlement in return for making a pre-determined series
of payments. Similarly, a caplet or oorlet is a European option which gives the holder the right to
exchange specic payments with the issuer.
Consider a receiver swaption with exercise date te . Since N (0; 0) = 1, the value of the option
today is
(4.17a)
7

V (0; 0) = D(0; te )

p(0; 0; te ; xe )Q(te ; xe )e

(te )xe A(te ) dx :

e

We are assuming that the oating leg re-values to at the settlement date.
K

10

Here

Q(te ; xe ) =

(4.17b)

nX

Ci Z (te ; xe ; ti ) KZ (te ; xe ; ts )

o+

is the value of the payo if the economy is in state X (te ) = xe at expiry. Substituting (3.8) for the
zero coupon bonds gives
(4.18)

V (0; 0) =

nX

p(0; 0; te ; xe )

Ci D(0; ti )e

i xe Ai +M (te ;xe ;ti )

KD(0; ts )e

s xe As +M (te ;xe ;ts )

o+

dxe ;

where i = (ti ), s = (ts ), etc. Since the transition densities p(0; 0; te; xe ) are known and the
function M (t; x; T ) is pre-tabulated, these options can be valued exactly with a single integration.
Let F0 be today's forward value of the cash ow at settlement,

F0 =

(4.19)

Ci D(ts ; ti ):

The implied price vol of the option8 is dened as the volatility B for which Black's formula,

V0 = D(0; ts )[F0 N (d1 ) K N (d2 )];

(4.20a)

d1;2 =

(4.20b)

log F0 =K  21 B2 te
p
;
B te

yields the correct price V0 given by (4.18).

In [18] singular perturbation techniques are used to obtain accurate approximations to the option
price (4.18), with the results stated in terms of implied price vols. There it is found that to leading
order the option price is
(4.21a)

B =

1
te

Z te
0

2 (t0 )dt0

1=2

1 (1 + x0 ) + : : :

where x0 is dened implicitly by

(4.21b)

F + K) =

1(
2 0

Here
(4.21c)

1 

Ci D(ts ; ti )e

(i s )x0

x
i D(ts ; ti )(i s )e i 0
i CP
:
i x0
i Ci D(ts ; ti )e

A much more accurate formula is quoted in Appendix F.

Equation (4.21a) illustrates the roles played by local volatility, mean reversion, and skew in
pricing. The local volatility (t) only enters (4.21a), which shows that the implied price vol B is
8

The price vol dened here should not be confused with the more commonly quoted rate vol.
11

proportional to the mean-square-average of the local volatility between today and the expiration
date.
Recall that 0 (t) is a decreasing function, and that \mean reversion" refers to how rapidly 0 (t)
decreases. See (3.10) et seq. As mean reversion increases, (ti ) (ts )  i s decreases, which
decreases the value of 1. Thus, as mean reversion increases, the price vol B of the swaption
decreases, as expected. Increasing the mean reversion also increases the importance of the earlier
payments Ci relative to the later payments.
Equations (4.19) and (4.21b) imply that x0 = 0 when the option is struck at-the-money (K = F0 ),
and that x0 is a decreasing function of the strike K . Thus, the price of an at-the-money swaption
is independent of  and , at least to within the approximations used to obtain (4.21). If  or  is
zero, the price vol is independent of the strike K ; as  and  increase from zero, the implied price
vol becomes a decreasing function of the strike K .
Besides giving insight into how the model parameters aect option prices, eq. (4.21) can be used
eectively in model calibration. Calibration is the process of choosing the model parameters (t)
and (t) to minimize or eliminate the errors between the model's predicted prices and the actual
prices of a set of standard market instruments. This is typically an iterative process, with most of the
computational time spent calculating derivatives of the predicted prices with respect to variations
in the model parameters. By using the approximate formula (4.21) to obtain a shrewd initial guess
for the model parameters, and then using the derivatives of (4.21) in place of the actual derivatives,
one can greatly speed up the calibration process.
5

Conclusions.

We used HJM theory to develop a framework for creating arbitrage-free Markovian term structure
models. This framework makes designing arbitrage-free interest rate models which match the initial
term structure a straight-forward task, freeing the designer to concentrate on imbuing the model
with less fundamental attributes. Not only are the resulting models Markovian, but they have no
more state variables than are needed to handle the random factors. This parsimony opens up new
methods for evaluating and using multi-factor models: trees, explicit and implicit nite dierence
schemes, and Green's function techniques can be used as well as Monte Carlo methods.
All arbitrage-free, Markovian term structure models we are aware of t within this framework.
Appendices B and C explicitly demonstrate that the extended Hull-White [8, 9], extended CIR
[10], Black-Karasinski [12], and Jamshidian's Brownian path independent models [14] t within
this framework, and Appendix D shows that Flesaker and Hughston's rational interest rate models
[15] also t within the framework. The separable n-factor models are more dicult since they
require n(n + 3)=2 state variables. Appendix E shows that these models can be constructed as
degenerate n(n + 3)=2 factor models. There we also create n-factor models which are analogues of
the separable models and are asymptotically equivalent to their separable counterparts in the limit
of small volatilities.
In Section 4 we used this framework to create the   models, a class of one factor models
dened by the process
(5.1)

c (t); X (0) = 0;
dX (t) = (t)[1 + X (t)] dW

and the numeraire

(5.2)

N (t; x) =

1
e(t)x+A(t):
D(0; t)
12

These models enable us to account for local volatility, mean reversion, and skew, and allow both
at-the-money and o-market instruments of varying maturities to be priced simultaneously. Their
analytic tractability is an added benet.
Describing local volatility, mean reversion, and skew is as much as can be expected from a one
factor model. However, some products are sensitive to risks which can not be described by a onefactor model. The two most common risks are sensitivity to rate decorrelation/curve exing and
forward/stochastic volatilities. Decorrelation risk occurs because forward rates become progressively
less correlated as the dierence between the maturity dates (and start dates) increases; this decorrelation tends to make the forward rate curve \ ex." In contrast, all forward rates in a one factor
model are essentially perfectly correlated. Forward volatility risk occurs in deals which contain an
\option on an option." Examples are captions (options on caps), and Bermudan and American
swaptions. In each case exercising the option involves either receiving or giving up an option, so
the exercise decision must balance the intrinsic value of the payo against the value of the option
received or lost, which is determined by the volatility environment that pertains at the exercise date.
The theoretical framework makes it easy to extend popular one factor models to include these
risks. For example, we can convert the   model to a stochastic volatility model by changing the
equation for the state variable to
(5.3a)

c1 (t); X (0) = 0;
dX (t) = (t)[1 +  (t)] Y (t)dW

(5.3b)

c2 (t); Y (0) = 1;
dY (t) = (t)Y (t)dW

and using the same numeraire as before,

1
e(t)x+A(t):
D(0; t)
Finally, note that the stochastic dierential equations for the state variables play only a peripheral
role in our theoretical framework; one could choose to directly model the risk neutral transition
densities p(t; x; T ; X ), dispensing with these equations entirely.

N (t; x) =

(5.3c)

Connection with HJM.

Here we show that the random process followed by the forward rate
(A.1)
F (t; T )  f (t; X (t); T ) = f0 (T ) + A0 (T ) ST (t; X (t); T )=S (t; X (t); T );
satises the forward rate drift restriction of HJM.
The backward Kolmogorov equation for (2.6b) implies that S (t; x; T ) satises the PDE
X
1X
(A.2a)
St + i (t; x)Sxi +
 (t) (t; x)j (t; x)Sxi xj = 0; 0 < t < T
2 ij ij i
i
(A.2b)

S (T; x; T ) = e

h(T;x) :

Applying Ito's lemma to (A.1), and using (A.2) to simplify the result, then yields
X
1X
ci (t):
(A.3) dF (t; T ) =
ij (t)i (t; X )j (t; X )(Sxi Sxj =S 2)T dt
i (t; X )(Sxi =S )T dW
2 ij
i
13

Equation (A.3) appears to violate the forward drift restriction of HJM theory [1]. However, this
restriction is phrased in terms of the risk neutral probability measure in the money market numeraire.
To switch to the money market numeraire, we apply the backwards Kolmogorov equation to the
valuation formula
n

(A.4)

V (t; x) = eh(t;x)+A(t) Eb V (T; X (T ))D(t; T )e

+

Z T

h(T;X (T )) A(T )

C (t0 ; X (t0 ))D(t; t0 )e

h(t0 ;X (t0 )) A(t0 ) dt0 

X (t) = x :

This shows that the value of a marketable instrument expressed in units of the numeraire,
(A.5a)

Vb (t; x)  V (t; x)e

H (t;x)

= V (t; x)e

Rt

0
0
0 f0 (T )dT

h(t;x) A(t) ;

satises the partial dierential equation

X
1X
(A.5b) Vbt + i (t; x)Vbxi +
ij (t)i (t; x)j (t; x)Vbxi xj = C (t; x)e
2
i
ij
Consequently, V (t; x) satises the partial dierential equation
X
1X
Vt + ~i (t; x)Vxi +
(A.6a)
 (t) (t; x)j (t; x)Vxi xj
2 ij ij i
i
where
(A.6b)
and
(A.6c)

~i (t; x) = i (t; x) i (t; x)

Rt

0
0
0 f0 (T )dT

h(t;x) A(t) :

r(t; x)V = C (t; x)

ij (t)j (t; x)hxj (t; x)

X
r(t; x) = f0 (t) + A0 (t) + ht (t; x) + i (t; x)hxi (t; x)

1X
+
 (t) (t; x)j (t; x)[hxi xj (t; x) hxi (t; x)hxj (t; x)]:
2 ij ij i

Expression (A.6c) for the short rate appears to be dierent from (2.10a). We can show that
these two expressions are equivalent by applying Ito's lemma to e h(T;X (T )) in (2.6b), taking the
expected value, and letting T ! t. This yields
(A.7a)

X
ST (t; x; t)
1X
= ht + i (t; x)hxi +
 (t) (t; x)j (t; x)[hxi xj
S (t; x; t)
2 ij ij i
i

as is needed for (A.6c) to agree with (2.10a). For future reference note that
(A.7b)

Sxi (t; x; t)
= hxi (t; x):
S (t; x; t)

Dene the money market numeraire B (t) by

(A.8)

dB (t)
= r(t; X (t)) dt:
B (t)
14

hxi hxj ];

Inspection of (A.6a) reveals that it is the backwards Kolmogorov equation for

V (t; x) = Ee

(A.9)

B (t)
V (T; X (T )) +
B (T )

Z T


o
B (t) 0
0 )) dt0  X (t) = x ;
C
(
t
;
X
(
t
B (t0 )

where Ee refers to the expected value in a probability measure for which

(A.10a)

fi (t); i = 1; 2; : : : ; n:
dXi (t) = ~i (t; X (t)) dt + i (t; X (t)) dW

f1 (t); W
f2 (t); : : : ; W
fn (t) are Brownian motions with the same correlation structure as before:
Here W
fi (t)dW
fj (t) = ij (t) dt:
dW

(A.10b)

Clearly, (A.9) is the valuation formula using the money market as the numeraire.
Comparing (A.10) with (2.2) shows that changing from N (t; x) to the money market numeraire
is equivalent to using Girsanov's theorem to change the probability measure so that
(A.11)

fi (t) = dW
ci (t) +
dW

ij (t)j (t; X )hxj (t; X ) dt; i = 1; 2; : : : ; n;

are n Brownian motions with the same correlation structure as before. Substituting (A.11) into the
forward rate process (A.3) now yields
(A.12a)

dF (t; T ) =

ij

ij (t)aiT (t; X ; T )aj (t; X ; T ) dt

with

ai (t; x; T ) =

(A.12b)

S (t; x; T )
i (t; x) xi
S (t; x; T )

fi (t)
aiT (t; X ; T ) dW

Sxi (t; x; t)
;
S (t; x; t)

where we have used (A.7b) to replace hxi (t; x). Inspection of the drift term in (A.12a) shows that
the forward rate process satises the HJM restriction. See eq.18 of [1].
B

Short rate models.

Arbitrage free short rate models are expressed in the money market numeraire,
(B.1a)

V (t; r) =

Ee

RT

0 0
t R(t )dt V (T; R(T )) +

Z T

C (t0 ; R(t0 ))e


00 00
t R(t )dt dt0 

R t0

R(t) = r ;

and assume that the risk neutral process for the short rate R(t) is of the form
(B.1b)

f (t)
dR(t) = [(t) + (t; R)] dt + s(t; R) dW

under this numeraire. Here the function (t) must be selected to match the initial term structure.
With some ingenuity, (t) can be found analytically for some models; otherwise a forward induction
scheme can be used [14].

15

B.1

Recasting short rate models.

We rst recast these models in the framework of Section 2 by choosing a zero coupon bond as the
numeraire. Equations (B.1a) and (B.1b) imply that V (t; r) satises

Vt + [(t) + (t; r)]Vr + 12 s2 (t; r)Vrr

(B.2)

rV = C (t; r):

Let us select a maturity Tref , and dene the zero coupon bond



Z (t; r; Tref ) = Ee e

(B.3)

R Tref

R(t0 )dt0 

R(t) = r :

Then Z solves the partial dierential equation

Zt + [(t) + (t; r)]Zr + 21 s2 (t; r)Zrr

(B.4a)

rZ = 0; t < Tref

Z (Tref ; r; Tref ) = 1:

(B.4b)

Suppose we denominate the value V (t; r) of tradable instruments in units of Z (t; r; Tref ),

Vb (t; r) =

(B.5)

V (t; r)
:
Z (t; r; Tref )

Substituting (B.5) into (B.2) then yields

Vbt + [(t) + b(t; r)]Vbr + 21 s2 (t; r)Vbrr =

(B.6a)

C (t; r)
;
Z (t; r; Tref )

where

b(t; r) = (t; r) + s2 (t; r)

(B.6b)

Zr (t; r; Tref )
:
Z (t; r; Tref )

This is equivalent to
(B.7a)

V (t; r) = Z (t; r; Tref )Eb

V (T; R(T ))
+
Z (T; R(T ); Tref )

Z T


o
C (t0 ; R(t0 ))
dt0  R(t) = r
0
0
Z (t ; R(t ); Tref )

where R(t) evolves according to

c (t);
dR(t) = [(t) + b(t; r)] dt + s(t; r) dW

(B.7b)

as can be seen by applying the backwards Kolmogorov equation to (B.7a). Note that this argument
represents an extension of Jamshidian's Brownian path independent models [14].
B.2

Recasting one factor models as short rate models.

We now consider the general class of one factor models,

(B.8a)

c (t)
dX (t) = (t; X (t)) dt + (t; X (t)) dW

under the numeraire

(B.8b)

N (t; x) =

1 h(t;x)+A(t)
e
:
D(0; t)
16

To match the initial term-structure, A(t) must be chosen as

h(t;X (t)) 

A(t) = log Eb e

(B.8c)

X (0) = 0 :

Once this is done, the short rate is given by

(B.9)
r(t; x) = f0 (t) + A0 (t) + ht (t; x) + (t; x)hx (t; x) + 21 2 (t; x)[hxx (t; x) h2x (t; x)]:
See (A.6c)
To re-write this as a short-rate model requires switching to the money market numeraire. Using
(A.11), the risk neutral process for X (t) is
f (t)
dX (t) = [(t; X ) 2 (t; X )hx(t; X )] dt + (t; X ) dW

(B.10)

under the money market numeraire. Applying Ito's lemma to (B.9) now shows that the process for
the short rate, R(t)  r(t; X (t)), is
f (t):
dR(t) = [rt + ( 2 hx )rx + 21 2 rxx ] dt + rx dW

(B.11)

We see that expressing an arbitrary one factor model (B.8a)-(B.8c) as a short rate model requires
inverting the relationship R = r(t; x) given in (B.9) to obtain x = x(t; R), and then substituting this
into (B.11) to obtain the drift and volatility terms as functions of t and R. This is a tedious (and
needless) process, but it simplies signicantly for several classes of models. For example, consider
the class of models examined in Section 3. For these models (t; x)  0 and h(t; x)  (t)x, so the
corresponding short rate models are

dR(t) = [rt

(B.12a)

f (t);
(t)2 (t; x)rx + 21 2 (t; x)rxx ] dt + (t; x)rx dW

where

r(t; x) = f0 (t) + A0 (t) + 0 (t)x

(B.12b)
C

 t)2 (t; x):

1 2
2 (

Special exponents.

The 

 models are dened by the numeraire

(C.1a)

N (t; x) =

1 (t)x+A(t)
e
D(0; t)

and the process

(C.1b)

c (t);
dX (t) = (t)[1 +  (t)X (t)] dW

where A(t) = M (0; 0; T ). Here (t), (t), and  (t) are arbitrary model parameters that can be used
to calibrate the model to the prices of market instruments. We now consider the exponents  = 0,
 = 1=2, and  = 1 and solve for
(C.2)

M (t; x; T ) = log Eb e

(T )[X (T ) x] 

X (t) = x :

The value Z (t; x; T ) of zero coupon bonds, the instantaneous forward rates f (t; x; T ), and the short
rate r(t; x) can then be read o of (3.8) and (3.10).
17

C.1

Extended Hull-White (

).

= 0

When  = 0, the transition density p(t; x; t; x) is Gaussian,

2

e (x x) =2(  )
;
p(t; x; t; x) =
[2(  )]1=2

(C.3)
where
(C.4)

 =

Z t
0

 (t0 ) dt0 ;  =
2

Z t
0

2 (t0 ) dt0 :

Integrating to compute the expected value in (C.2) yields

(C.5)

M (t; x; T ) = 21 2 (T )(  ); A(T ) = 12 2 (T ) :

To show that this is identical to the extended Hull-White model, note that (3.10) yields
(C.6)
r(t; x) = f0 (t) + 0 (t)x + 0 (t)(t) (t):
Equation (B.12) now shows that the equivalent short rate model is
(C.7a)

f (t);
dR(t) = [(t) (t)R(t)] dt + s(t) dW

under the money market numeraire, where

(C.7b)
(t) = 00 (t)=0 (t); s(t) = 0 (t)(t); (t) = f00 (t) + (t)f0 (t) + [0 (t)]2  (t):
The extended Hull-White model is usually expressed as the short rate process (C.7a), where (t)
and s(t) are arbitrary model parameters and (t) is chosen as
(C.8)

(t) = f 0 (t) + (t)f0 (t) +

Z t

s2 (t0 )e

R
2 tt0 (t00 )dt00

dt0

to match the initial term structure. Note that mean reversion is expressed through the parameter
(t) in the extended Hull-White model, while it is expressed through the \gearing factor" (t) in
the equivalent   model.
C.2

CIR-like models (

When  = 21 the state variable obeys

(C.9)

1 ).
2

c (t):
dX (t) = (t)[1 +  (t)X (t)]1=2 dW

Consequently, the transition density p(t; x; t; x) satises the forward Kolmogorov equation
(C.10a)

pt = 12 2 (t) ([1 +  (t)x]p)xx ; t > t

(C.10b)

p ! (x x) as t ! t:
18

To obtain M (t; x; T ) we take the two-sided Laplace transform

Z 1
(C.11)
P (t; x; t; ) 
e x p(t; x; t; x) dx;
1
which yields

Pt = 21 2 (t)2 fP

(C.12a)

x

P =e

(C.12b)

 (t)P g ; t > t
at t = t:

This is a rst order hyperbolic equation which can be solved by the method of characteristics. We
obtain
Z t
n
o
x[B (t) B (t)]
22 (t0 ) dt0
2

(C.13a)
log P (t; x; t; ) = x + 
+
0 2
2 + [B (t) B (t)]
t (2 + [B (t) B (t )])
where

B (t) =

(C.13b)

Z t
0

2 (t0 ) (t0 ) dt0 :

Consequently, M (t; x; T ) and A(T ) are

(C.14a)
(C.14b)

x[B (T ) B (t)]
+
M (t; x; T ) =  (T )
2 + (T )[B (T ) B (t)]
2

A(T ) = 2 (T )

Z T
0

Z T

o
22 (t0 ) dt0
;
(2 + (T )[B (T ) B (t0 )])2

22 (t0 ) dt0
:
(2 + (T )[B (T ) B (t0 )])2

The values of zero coupon bonds, the forward rate curve, and the short rate can now be obtained
directly from (3.8) and (3.10). In particular, the short rate is

r(t; x) = f0 (t) + K (t)[x + C (t)]

(C.15a)
with
(C.15b)

K (t) = 0 (t)

 t) (t) (t); C (t) = (t)

1 2
2 (

Z t
0

2 (t0 ) dt0
:
(1 + (t)[B (t) B (t0 )]=2)3

Let us cast this model as the equivalent short rate model. Using (B.12) shows that this model is
(C.16a)

f (t);
dR(t) = [(t) (t)R(t)] dt + s(t) d(t) + R(t) dW

under the money market numeraire, where

(C.16b)
(t) = [K 0(t) (t)2 (t) (t)]=K (t);
(C.16c)

(t) = f00 (t) + (t)f0 (t) + K (t)C 0 (t) (t)2 (t)K (t)[1  (t)C (t)];
19

(C.16d)

s(t) = (t) K (t) (t); d(t) = K (t)= (t) f0 (t) K (t)C (t):

The extended CIR model is usually expressed as

f (t):
dR(t) = [(t) (t)R(t)] dt + s(t) R(t) dW

(C.17)

Here (t) and s(t) are arbitrary model parameters, and (t) must be chosen to t the initial term
structure. To state this requirement explicitly, recall that the value of a zero coupon bond is
RT

0 0 
t R(t )dt 

Z (t; r; T ) = Ee e

(C.18)

R (t ) = r :

Equivalently, Z (t; r; T ) is the solution of the backwards equation

(C.19a)

Zt + [(t) (t)r]Zr + 12 s2 (t)rZrr

rZ = 0; t < T

with

Z (T; r; T ) = 1:

(C.19b)
The solution of (C.19) is [7]

Z (t; r; T ) = e

(C.20a)

B (t;T )r A(t;T )

where B (t; T ) satises the Ricatti equation

Bt = (t)B + 12 s2 (t)B 2

(C.20b)

1; t < T

with the boundary condition B (T; T ) = 0, and

A(t; T ) =

(C.20c)

Z T

(t0 )B (t0 ; T ) dt0 :

Matching the initial term structure requires Z (0; 0; T ) = D(0; T )  e

(t) to be chosen so that
(C.21)

Z T
0

(t0 )B (t0 ; T ) dt0 =

Z T
0

RT
0

f0 (T 0 )dT 0 ,

which requires

f0 (T 0) dT 0 B (0; T )f0(0):

We see that matching the extended CIR model to the initial term structure requires solving
a Ricatti equation to obtain the zero coupon bond values. This can be done eciently, but it is
dicult to understand the qualitative impact of yield curve shifts on pricing.
Comparing (C.16) and (C.17) shows that the extended CIR model is a special case of the  
model with  = 1=2 and with  (t) chosen so that d(t) = 0. Unsurprisingly, making this choice of
 (t) also requires solving a Ricatti equation.
A simpler alternative is to use the  = 1=2 model without requiring d(t)  0. The  = 1=2 model
then gives the zero coupon bond values directly, without solving a Ricatti equation. In addition,
 (t) can be used to match the implied volatility smile by calibrating the model to the prices of
o-market instruments. Although the CIR model is more aesthetically appealing, with its natural
barrier occuring at R(t) = 0 and its simple relation between the volatility and the short rate, we
stress that neither model can be expected to be valid near R(t) = 0. Since the barrier in the  = 1=2
model is irrelevant for pricing commonly traded US instruments, it seems better to use  (t) to match
the implied volatility smile.
20

C.3

Black-Karasinski-like models (

).

= 1

When  = 1 we have been unable to nd the Laplace transform of the transition density, and thus
M (t; x; T ). However, we can nd the moments mk (t; x; T ) explicitly, and then use the expansions
(C.22a)

M (t; x; T ) = 12 2 (T )m2 (t; x; T )

(C.22b)

A(T ) = 12 2 (T )m2 (0; 0; T )

 T )m3 (t; x; T ) + : : : ;

1 3
6 (

 T )m3(0; 0; T ) + : : : :

1 3
6 (

Equations (3.8) and (3.10) then determine the zero coupon values, the forward rate curve, and the
short rate as before. Although these expressions are only approximate, using the rst term suces
to price all but the most sensitive instruments.
With the exponent  = 1,
c (T );
dX (T ) = (T )[1 +  (T )X (T )] dW

(C.23)

and applying Ito's lemma yields

(C.24)

d[X (T ) x]k = 12 k(k 1)2 (T )[1 +  (T )X ]2 [X x]k 2 dT

c (T ):
+ k(T )[1 +  (T )X ][X x]k 1 dW

Taking the expected value then yields equations for the moments mk (t; x; T ):


dmk 1
= 2 k(k 1)2 (T ) (1 + x)2 mk 2 + 2 (1 + x)mk 1 +  2 mk :
dT
See (3.11). Clearly at T = t we have m0 (t; x; t) = 1 and mk (t; x; t) = 0 for all k > 1. Solving these
equations successively yields

(C.25)

m0 (t; x; T )  1; m1 (t; x; T )  0;

(C.26a)
(C.26b)

(C.26c)

m2 (t; x; T ) =
m3 (t; x; T ) = 6

Z T

Z T Z t0

nZ T

 (t0 )[1 +  (t0 )x]2 exp

2

t0

2 (T 0) 2 (T 0 )dT 0 dt0 ;

2 (t0 )2 (t00 ) (t0 )[1 +  (t0 )x][1 +  (t00 )x]2

exp

n Z T

t0

 (T 0 ) 2 (T 0) dT 0 +
2

Z t0

t00

2 (T 0) 2 (T 0 ) dT 0 dt00 dt0 :

As before, eq. (B.12) can be used to obtain the equivalent short rate model. Even though the
distribution of X (t) is roughly lognormal, the short rate model is not the Black-Karasinski model
regardless of how  (t) is chosen. The Black-Karasinski model requires adding a drift term to the
stochastic dierential equation for X (t).

21

Rational interest rate models.

Consider the interest rate model dened by the valuation formula

Z T
n
o
V (T; X (T ))
C (t0 ; X (t0 )) 0 
V (t; x) = N (t; x)Eb
(D.1a)
+
dt
X
(
t
)
=
x
;

N (T; X (T )) t N (t0 ; X (t0 ))
with the numeraire

N (t; x) =

(D.1b)

1
;
D(0; t) + b(t)
x

and assume that the state variables X (t) evolve according to

ci (t); Xi (0) = 0; i = 1; 2; : : : ; n
dXi (t) = i (t; X ) dW

(D.1c)

under this numeraire in the risk neutral world. Noting that X (t) is a Martingale, we nd that the
zero coupon bond prices are
o
D(0; T ) + b(T )
x
D(0; T ) + b(T )
X (T ) 
:
 X (t) = x =
D(0; t) + b(t)
x
D(0; t) + b(t)
x
Consequently, the instantaneous forward rate curve is
DT (0; T ) + b0 (T )
x
f (t; x; T ) =
(D.3)
:
D(0; T ) + b(T )
x

Z (t; x; T ) = Eb

(D.2)

Flesaker and Hughston's rational lognormal model [15] is the special case i (t; x) = i (t)(1 + xi ).
These models have the advantage that all forward ratesPf (t; x; T ) are guaranteed to remain positive
provided that b0i (T )  0 for each i and DT (0; T ) < b0i (T ) for all T . Moreover, the one factor
version of this model has closed-form prices for swaptions and caplets. See [15].
E

Separable models.

The general n-factor HJM model for the forward rate is

(E.1)

dF (t; !(t); T ) =

ij

ij (t)aiT (t; !(t); T )aj (t; !(t); T ) dt

fi (t)
aiT (t; !(t); T ) dW

in the risk neutral world under the money market numeraire:

(E.2)

V (t) = Ee e

RT

00 00 00
t R(t ;! )dt V (T; ! (T )) +

Z T

C (t0 ; !(t0 ))e

R t0

00 00 00
t R(t ;! )dt

aiT ,

dt0

Ft :

Here !(t) indicates that the volatilities

payos, and cash ows can depend on all measurable
events that can be resolved by time t.
The separable models [4, 5, 6] are derived by presuming that the volatilities have the form
(E.3a)
ai (t; !(t); T ) = 0 (T )i (t; !(t)); i = 1; 2; : : : ; n
T

so that
(E.3b)

ai (t; !(t); T ) = [i (T ) i (t)]i (t; !(t)); i = 1; 2 : : : ; n:

22

Then the stochastic process for the forward rates becomes

dF (t; !(t); T ) =

(E.4)

ij

0i (T )[j (T ) j (t)]ij (t)i (t; !(t))j (t; !(t)) dt

X

fi (t):
0i (T )i (t; !(t)) dW

Integrating over t now yields

(E.5)

F (t; !; T ) = f0 (T ) +

0i (T )Xi (t; !) +

ij

0i (T )j (T )Yij (t; !)

where
(E.6a) Xi (t; !) =

(E.6b)

Z t
0

XZ t

fi (t0 )
i (t0 ; !0 ) dW

Yij (t; !) =

Z t
0

j (t0 )ij (t0 )i (t0 ; !0 )j (t0 ; !0 ) dt0 ; i = 1; : : : ; n;

ij (t0 )i (t0 ; !0 )j (t0 ; !0) dt0 ; i; j = 1; : : : ; n:

One now assumes that the coecients i (t; !(t)) depend on the path taken by the Brownian
motion only through the value of the state variables X (t), Y (t) at time t. This is realistic since
(E.5) expresses the entire term structure in terms of these variables. With this assumption, (E.6)
becomes


fi (t) +
(E.7a) dXi (t) = i (t; X ; Y ) dW

(E.7b)

j (t)ij (t)j (t; X ; Y ) dt ; Xi (0) = 0; i = 1; : : : ; n

dYij (t) = ij (t)i (t; X ; Y )j (t; X ; Y ) dt; Y ij (0) = 0; i; j = 1; : : : ; n;

and the instantaneous forward rate curve is

(E.8)

f (t; x; y; T ) = f0 (T ) +

0i (T )xi +

ij

0i (T )j (T )yij ;

where xi = Xi (t) and yij = Yij (t) are the values of the state variables at time t.
Under the money market numeraire, the general separable model is dened by the forward rate
curve (E.8) and the stochastic processes (E.7). We can put this model into the context of Section 2
by switching to the numeraire
(E.9)

N (t; x; y) =

nX
o
1
1X
exp
i (t)xi +
i (t)j (t)yij :
D(0; t)
2 ij
i

Under this numeraire the value of a tradable instrument is

(E.10)

V (t; x; y) = N (t; x; y)Eb

V (T; X (T ); Y (T ))
+
N (T; X (T ); Y (T ))

Z T

23

o
C (t0 ; X (t0 ); Y (t0 )) 0 
dt
X
(
t
)
=
x
;
Y
(
t
)
=
y
:

N (t0 ; X (t0 ); Y (t0 ))

Reversing the steps in Appendix A, we nd that the state variables evolve according to
(E.11a)
(E.11b)

ci (t); Xi (0) = 0; i = 1; : : : ; n
dXi (t) = i (t; X ; Y ) dW

dYij (t) = ij (t)i (t; X ; Y )j (t; X ; Y ) dt; Yij (0) = 0; i; j = 1; : : : ; n;

under the new numeraire. Finally, integrating the forward rate curve (E.8) with respect to the
maturity T yields the zero coupon bond prices:
n X
o
1X
(E.12) Z (t; x; y; T ) = D(t; T ) exp
[i (T ) i (t)]xi
[i (T )j (T ) i (t)j (t)]yij :
2 ij
i
Note that Z (0; 0; 0; T ) = D(0; T ), so the separable models are automatically consistent with the
initial term structure.
Since Yij (t) = Yji (t), the n-factor separable model generally requires n(n + 3)=2 state variables.
As the random processes Yij (t) are not being driven directly by Brownian motions, we can only view
the separable models within the framework of Section 2 by considering (E.11) as a highly degenerate
n(n + 3)=2 factor model.
The zero coupon bond prices in (E.12) show that the extra state variables yij are needed to
account for convexity eects in an arbitrage free manner. Since convexity eects are quite small,
ci (t),
and since (E.11b) shows that the Yij (t) are only driven indirectly by the stochastic processes dW
the distributions of Yij (t) are highly peaked about their means, which are near zero. Specically,
as the volatilities i decrease, the means of the variables Yij (t) scale like 2 t and the variances of
Yij (t) scale like 6 t3 . This suggests that the convexity eects can be accounted for by replacing the
distributions of Yij (t) by -functions, selecting the position of the -functions to make the theory
arbitrage free. Carrying this idea through leads to the arbitrage-free n-state variable model dened
by the valuation formula
Z T
n
o
V (T; X (T ))
C (t0 ; X (t0 )) 0 
V (t; x) = N (t; x)Eb
(E.13a)
+
dt
X
(
t
)
=
x
;

N (T; X (T )) t N (t0 ; X (t0 ))
the numeraire
1 (t)
x+A(t)
N (t; x) =
(E.13b)
e
;
D(0; t)
and the risk neutral processes
(E.13c)

ci (t); Xi (0) = 0; i = 1; : : : ; n
dXi (t) = i (t; X ) dW

under this numeraire. Here A(T ) is dened by

(E.13d)

A(T ) = log Eb e (T )
X (T )  X (0) = 0 :

This model is the multi-factor generalization of the one-factor models in Section 3. Following
the analysis there shows that the zero coupon bond prices are
(E.14a)
Z (t; x; T ) = D(t; T )e [(T ) (t)]
x A(T )+A(t)+M (t;x;T ) ;
where
(E.14b)

M (t; x; T ) = log Eb e (T )
[X (T ) x]  X (t) = x :

24

Consequently the instantaneous forward rates are

(E.14c)
f (t; x; T ) = f0 (T ) + 0 (T )
x + A0 (T ) MT (t; x; T ):
Typically 0i (T )i is around 10 2 in US dollar markets. At these volatilities the n-state variable
model (E.13) is virtually identical to the analogous separable model in (E.9)-(E.11). Although the
n-state variable model has the disadvantage of requiring computation of M (t; x; T ) to obtain the
zero coupon bond prices, this is usually outweighed by having many fewer state variables than the
separable models.
F

Approximate prices of European options.

As in Section 4.1, let te be the expiration date of a European option to receive the cash ows Ci on
dates ti ; i = 1; 2; : : : ; n in return for paying the strike K on the settlement date ts . Also dene

F0 =

(F.1)

Ci D(ts ; ti ); e 

Z te
0

2 (t0 )dt0

as before. To express the option prices succinctly, dene t0 by the relation

Z t0

(F.2)

dene x by the implicit relation

(F.3)

F + K) =

1
2( 0

2 (t0 )dt0 = e =2;

Ci D(ts ; ti )e

[i s ]x Ai +As +M (t0 ;x ;ti ) M (t0 ;x ;ts )

and dene k by
(F.4)

k =

i Ci D(ts ; ti )[i

s

Mx (t0 ; x ; ti ) + Mx(t0 ; x ; ts )]k e

i x Ai +M (t0 ;x ;ti )
i Ci D(ts ; ti )e

i x Ai +M (t0 ;x ;ti )

Then the implied price vol of the option is [18]

n

(F.5)

(2 ) 2
3
22 
+
2
3
(1 + x )2
1
21
o
2
(1 + )
3 2
3
22 
+
+
3
+
:
:
:
:
(1 + x )2 (1 + x )1 1
21


B = 1 (e =te )1=2 (1 + x ) 1 + 241 (1 + x )2 e 21

+

(F0 K )2  2
2
621(F0 + K )2 1

This expression yields prices that are usually accurate to within a tenth of a basis point, although
the error can be several times larger in unusual situations. See [18].
Acknowledgements

We gratefully acknowledge the assistance of A. Berner, Olivier Van Eyseren, our colleagues on
Paribas' New York Swaps desk and Paribas' London FIRST team.

25

References

[1] D. Heath, R. Jarrow, and A. Morton, \Bond pricing and the term structure of interest rates:
A new methodology," Econometrica 60 (1992), pp. 77-105
[2] D. Heath, R. Jarrow, and A. Morton, \Bond pricing and the term structure of interest rates:
A discrete time approximation," J. Fin. Quant. Anal. 25 (1990), pp. 419-440
[3] R. Rebonato, Interest-Rate Option Models (Wiley, New York, 1996)
[4] A. Li, P. Ritchken, and L. Sankarasubramanian, \Lattice models for pricing American interest
rate claims," J. Finance 50 (1995), pp. 719-737
[5] P. Ritchken and L. Sankarasubramanian, \Volatility structures of forward rates and the dynamics of the term structure," Math. Finance 5 (1995), pp. 55-72
[6] O. Cheyette, \Term structure dynamics and mortgage valuation," J. Fixed Income (1992), pp.
28-41
[7] J. C. Hull, Options, Futures, and Other Derivatives (Prentice-Hall, Upper Saddle River, 1997)
[8] J. Hull and A. White, \Pricing interest rate derivative securities," Rev. Fin. Studies 3 (1990),
pp. 573-592
[9] J. Hull and A. White, \Ecient procedures for valuing European and American path-dependent
securities," Jour. Fin. Quan. Anal. 25 (1990), pp. 87-100
[10] J. C. Cox, J. E. Ingersoll, and S. A. Ross, \A theory of the term structure of interest rates,"
Econ. 53 (1985), pp. 385-407
[11] F. A. Longsta and E. S. Schwartz, \Interest rate volatility and the term structure: a two-factor
general equilibrium model," Jour. Fin. 47 (1992), pp. 1259-1282
[12] F. Black and P. Karasinski, \Bond and option pricing when short rates are lognormal," Financial
Analysts J. (1991), pp. 52-59
[13] T. S. Y. Ho and S. B. Lee, \Term structure movements and pricing interest rate contingent
claims," J. Fin. 41 (1986), pp. 1011-1028
[14] F. Jamshidian, \Forward induction and construction of yield curve diusion models," J. Fixed
Income (1991), pp. 62-74
[15] B. Flesaker and L. P. Hughston, \Positive interest," Vasicek
(Risk Publications, London, 1996)

, ed. Lane Hughston

and Beyond

[16] J. M. Harrison and S. R. Pliska, \Martingales and stochastic integrals in the theory of continuous
trading," Stoch. Proc. Applications 11 (1981), pp. 215-260
[17] J. M. Harrison and D. Kreps, \Martingales and arbitrage in multiperiod securities markets," J.
Econ. Theory 20 (1979), pp. 381-408
[18] P. S. Hagan and D. E. Woodward, \Swaption and caplet prices for one factor models,"
preparation

26

in