FN6813 Notes

FN6813
Interest Rate Derivatives

Nicolas Privault
February 3, 2024 https://personal.ntu.edu.sg/nprivault/indext.html

i
This course deals with the stochastic modeling of interest rates and with the pricing of related
derivative products such as bonds, caps, and swaptions.
The modeling of short term interest rates and their use in bond pricing and covered in Chapter 1.
Forward and LIBOR/SOFR rates and their constructions in the HJM, two-factor and BGM models
are considered in Chapter 2.
A general construction of forward measures using the change of numéraire technique is given
in Chapter 3, and is applied to the pricing of interest rate derivatives such as caplets, caps and
swaptions in Chapter 4.
Finally, credit risk and stochastic default are considered in Chapter 5, with the pricing of
defaultable bonds.
This pdf file contains internal and external links, and 43 figures, including 9 animated figures,
e.g. Figures 1.16, 2.6, 2.9, 2.10, 2.17, that may require using Acrobat Reader for viewing on
the complete pdf file. It also includes 2 Python codes, 8 codes e.g. on pages 40, 110, and
49 exercises and problems with solutions. Supplementary exercises, problems and solutions are
available from the textbook Stochastic Interest Rate Modeling with Fixed Income Derivative
Pricing, World Scientific, 2021.
" February 3, 2024 FN6813 Interest Rate Derivatives - N. Privault

Contents
1 Short Rates and Bond Pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Short-Term Mean-Reverting Models 1
1.2 Calibration of the Vasicek Model 7
1.3 Zero-Coupon and Coupon Bonds 11
1.4 Bond Pricing PDE 14
Exercises 28
2 Forward Rate Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

2.1 Construction of Forward Rates 34
2.2 LIBOR and SOFR Swap Rates 44
2.3 The HJM Model 48
2.4 Yield Curve Modeling 52
2.5 Two-Factor Model 56
2.6 The BGM Model 59
Exercises 61
3 Change of Numéraire and Forward Measures . . . . . . . . . . . . . . . . . . 64

3.1 Notion of Numéraire 64
3.2 Change of Numéraire 67
3.3 Foreign Exchange 75
3.4 Pricing Exchange Options 83
3.5 Hedging by Change of Numéraire 85
Exercises 88

4 Pricing of Interest Rate Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.1 Forward Measures and Tenor Structure 93
4.2 Bond Options 97
4.3 Caplet Pricing 98
4.4 Forward Swap Measures 103
4.5 Swaption Pricing 105
Exercises 112
5 Reduced-Form Approach to Credit Risk . . . . . . . . . . . . . . . . . . . . . . . 123

5.1 Survival Probabilities 123
5.2 Stochastic Default 125
5.3 Defaultable Bonds 128
Exercises 131
Exercise Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

Chapter 1 134
Chapter 2 147
Chapter 3 153
Chapter 4 163
Chapter 5 180
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
Articles 184
Books 187
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
Author index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193

List of Figures
1.1 Short rate t 7→ rt in the Vasicek model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.2 CBOE 10 Year Treasury Note (TNX) yield . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Short rate t 7→ rt in the CIR model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Calibrated Vasicek simulation vs. market data . . . . . . . . . . . . . . . . . . . . . 10
1.5 Five-dollar 1875 Louisiana bond with 7.5% biannual coupons . . . . . . . . . . 11
1.6 Discrete-time coupon bond pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.7 Continuous-time coupon bond pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.8 Comparison of Monte Carlo and PDE solutions . . . . . . . . . . . . . . . . . . . . 20
1.9 Bond price t 7→ P(t, T ) vs. t 7→ e −r0 (T −t ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.10 Bond price t 7→ Pc (t, T ) with a 5% coupon rate . . . . . . . . . . . . . . . . . . . . 21
1.11 Bond price with coupon rate 6.25% . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.12 Orange Cnty Calif bond prices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.13 Orange Cnty Calif bond yields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.14 Fitting of a gamma probability density function . . . . . . . . . . . . . . . . . . . 26
1.15 Approximation of Dothan bond prices . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.16 Brownian bridge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.1 Graph of the spot forward rate S 7→ f (t,t, S) . . . . . . . . . . . . . . . . .. . . . . . 35

2.2 Indonesian government securities yield curve . . . . . . . . . . . . . . .. . . . . . 35
2.3 Forward rate process t 7→ f (t,t, T ) . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . 39
2.4 Instantaneous forward rate process t 7→ f (t, T ) . . . . . . . . . . . . . . .. . . . . . 40
2.5 Federal Reserve yield curves from 1982 to 2012 . . . . . . . . . . . . . .. . . . . . 40
2.6 European Central Bank yield curves* . . . . . . . . . . . . . . . . . . . . .. . . . . . 41
2.7 August 2019 Federal Reserve yield curve inversion* . . . . . . . . . .. . . . . . 42
2.8 Stochastic process of forward curves . . . . . . . . . . . . . . . . . . . . . .. . . . . . 48
2.9 Forward instantaneous curve in the Vasicek model* . . . . . . . . . .. . . . . . 51
2.10 Forward instantaneous curve x 7→ f (0, x) in the Vasicek model* . . . . . . 52
2.11 Short-term interest rate curve t 7→ rt in the Vasicek model . . . . .. . . . . . 52
2.12 Nelson-Siegel graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . 53

2.13 Svensson model graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.14 Fitting of a Svensson curve to market data . . . . . . . . . . . . . . . . . . . . . . 54
2.15 Graphs of forward rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.16 Forward instantaneous curve in the Vasicek model . . . . . . . . . . . . . . . . 55
2.17 ECB data vs. fitted yield curve* . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.18 Bond prices t 7→ P(t, T2 ), P(t, T2 ), P(t, T3 ) . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.19 Forward rates in a two-factor model . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.20 Evolution of instantaneous forward rates in a two-factor model . . . . . . 59
2.21 Roadmap of stochastic interest rate modeling . . . . . . . . . . . . . . . . . . . 61
3.1 Why change of numéraire? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

3.2 Overseas investment opportunity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.3 Evolution of exchange rate vs. interest rate . . . . . . . . . . . . . . . . . . . . . . . 78
4.1 Implied swaption volatilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
S.1 Log bond prices correlation graph in the two-factor model . . . . . . . . . . 152
List of Tables
2.1 Stochastic interest rate models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.1 Local vs. foreign exchange options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.1 Forward rates arranged according to a tenor structure . . . . . . . . . . . . . . 94

4.2 A list of numéraire processes and their applications . . . . . . . . . . . . . . . . 111
5.1 Mortality table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

5.2 Cumulative historic default rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
*Animated figures (work in Acrobat reader).

FN6813 Interest Rate Derivatives
1
1. Short Rates and Bond Pricing
Short-term rates, typically daily rates, are the interest rates applied to short-term lending between
financial institutions. The stochastic modeling of short-term interest rate processes is based on
the mean reversion property, as in the Vasicek, CIR, CEV, and affine-type models studied in this
chapter. The pricing of fixed income products, such as bonds, is considered in this framework using
probabilistic and PDE arguments.

Exercises 28
1.1 Short-Term Mean-Reverting Models

Money market accounts with price (At )t∈R+ can be defined from a short-term interest rate process
(rt )t∈R+ as
At +dt − At dAt dAt
= rt dt, = rt dt, = rt At , t ⩾ 0,
At At dt
with w t
At = A0 exp rs ds , t ⩾ 0.
0
As short-term interest rates behave differently from stock prices, they require the development of
specific models to account for properties such as positivity, boundedness, and return to equilibrium.

2 Chapter 1. Short Rates and Bond Pricing
Vašíček, 1977 model

The first model to capture the mean reversion property of interest rates, a property not possessed by
geometric Brownian motion, is the Vašíček, 1977 model, which is based on the Ornstein-Uhlenbeck
process. Here, the short-term interest rate process (rt )t∈R+ solves the equation
drt = (a − brt )dt + σ dBt , (1.1.1)
where a, σ ∈ R, b > 0, and (Bt )t∈R+ is a standard Brownian motion, with solution
a wt
rt = r0 e −bt + 1 − e −bt + σ e −(t−s)b dBs ,

t ⩾ 0, (1.1.2)
b 0
see Exercise 1.1. The probability distribution of rt is Gaussian at all times t, with mean
a
IE[rt ] = r0 e −bt + 1 − e −bt ,

b
and variance given from the Itô isometry as
h wt i
Var[rt ] = Var σ e −(t−s)b dBs
0
wt 2
= σ 2
e −(t−s)b ds
w0t
= σ 2
e −2bs ds
0
σ2
1 − e −2bt ,

= t ⩾ 0,
2b
i.e.
2

−bt a −bt σ −2bt
rt ≃ N r0 e + 1 − e

, 1− e , t > 0.
b 2b
In particular, the probability density function ft (x) of rt at time t > 0 is given by
√ 2 !
r0 e −bt + a 1 − e −bt /b − x

b/π
ft ( x ) = √ exp − , x ∈ R.
σ 1 − e −2bt σ 2 1 − e −2bt /b
In the long run,* i.e. as time t becomes large we have, assuming b > 0,
a σ2
lim IE[rt ] = and lim Var[rt ] = , (1.1.3)
t→∞ b t→∞ 2b
and this distribution converges to the Gaussian N (a/b, σ 2 /(2b)) distribution, which is also the
invariant (or stationary) distribution of (rt )t∈R+ , see Exercise 1.1. In addition, the process tends to
revert to its long term mean a/b = limt→∞ IE[rt ] which makes the average drift vanish, i.e.:
lim IE[a − brt ] = a − b lim IE[rt ] = 0.

t→∞ t→∞
Figure 1.1 presents a random simulation of t 7→ rt in the Vasicek model with r0 = 3%, and shows
the mean-reverting property of the process with respect to a/b = 2.5%.
*“But this long run is a misleading guide to current affairs. In the long run we are all dead.” Keynes, 1924, Ch. 3, p.
80.
February 3, 2024 FN6813 Interest Rate Derivatives - N. Privault "

8
7
rt (%)
6
5
4
3
a/b
2
1
0
-1
-2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t
Figure 1.1: Graph of the Vasicek short rate t 7→ rt with a = 0.025, b = 1, and σ = 0.1.
As can be checked from the simulation of Figure 1.1 the value of rt in the Vasicek model may
become negative due to its Gaussian distribution. Although real interest rates may sometimes fall
below zero,* this can be regarded as a potential drawback of the Vasicek model. The next code
provides a numerical solution of the stochastic differential equation (1.1.1) using the Euler method,
see Figure 1.1.
N=10000;t<-0:(N-1);dt<-1.0/N;nsim<-2; a=0.025;b=1;sigma=0.1;
dB <- matrix(rnorm(nsim*N,mean=0,sd=sqrt(dt)), nsim, N)
R <- matrix(0,nsim,N);R[,1]=0.03
dev.new(width=10,height=7);
for (i in 1:nsim){for (j in 2:N){R[i,j]=R[i,j-1]+(a-b*R[i,j-1])*dt+sigma*dB[i,j]}}
par(mar=c(0,1,1,1));par(oma=c(0,1,1,1));par(mgp=c(-5,1,1))
plot(t,R[1,],xlab = "Time",ylab = "",type = "l",ylim = c(R[1,1]-0.2,R[1,1]+0.2),col = 0,axes=FALSE)
axis(2, pos=0,las=1);for (i in 1:nsim){lines(t, R[i, ], xlab = "time", type = "l", col = i+0)}
abline(h=a/b,col="blue",lwd=3);abline(h=0)
√
We note that the process remains in the N a/b, σ 2 /2/b Gaussian distribution if it is started

from this distribution.
a=0.25;b=10;sigma=1; N=1000; t <- 0:N; dt <- 1.0/N; nsim=20;

dB <- matrix(rnorm( nsim * N, 0, sqrt(dt)), nsim, N); Y <- matrix(0, nsim, N+1)
for (i in 1:nsim){Y[i,1]=rnorm(1,a/b,sqrt(sigma**2/2/b));
for (j in 2:N){Y[i,j] = Y[i,j-1] +(a-b*Y[i,j-1])*dt +sigma*dB[i,j];}}
H<-hist(Y[,N],plot=FALSE,breaks=10); dev.new(width=16,height=7);
layout(matrix(c(1,2), nrow =1, byrow = TRUE));par(mar=c(2,2,2,0), oma = c(2, 2, 2, 2))
plot(t*dt, Y[1, ], xlab = "", ylab = "", type = "l", ylim = c(-1, 1), col = 0, xaxs='i',yaxs='i',las=1,
cex.axis=1.6)
for (i in 1:nsim){lines(t*dt, Y[i, ], type = "l", ylim = c(-1, 1), col = i,lwd=2)}
for (i in 1:nsim){points(0.999, Y[i,N], pch=1, lwd = 5, col = i)}
x <- seq(-2,2, length=100); px <- dnorm(x,a/b,sqrt(sigma**2/2/b));par(mar = c(2,2,2,2))
plot(NULL , xlab="", ylab="", xlim = c(0, max(px,H$density)), ylim = c(-2,2),axes=F)
rect(0, H$breaks[1:(length(H$breaks) - 1)], col=rainbow(20,start=0.08,end=0.6), H$density,
H$breaks[2:length(H$breaks)]); lines(px,x, lty=1, col="black",lwd=2)
* Eurozone interest rates turned negative in 2014.

1.0
0.5
0.0
−0.5
−1.0
0.0 0.2 0.4 0.6 0.8 1.0
Example - TNX yield
We consider the yield of the 10 Year Treasury Note on the Chicago Board Options Exchange
(CBOE), for later use in the calibration of the Vasicek model. Treasury notes usually have a
maturity between one and 10 years, whereas treasury bonds have maturities beyond 10 years).
library(quantmod)
getSymbols("^TNX",from="2012-01-01",to="2016-01-01",src="yahoo")
rate=Ad(`TNX`);rate<-rate[!is.na(rate)]
dev.new(width=10,height=7);chartSeries(rate,up.col="blue",theme="white")
n = length(!is.na(rate))
The next Figure 1.2 displays the yield of the 10 Year Treasury Note.
rate [2012−01−03/2015−12−31]
Last 2.269 3.0
2.5
2.0
1.5
Jan 03 Jul 02 Jan 02 Jul 01 Jan 02 Jul 01 Jan 02 Jul 01 Dec 31

2012 2012 2013 2013 2014 2014 2015 2015 2015
Figure 1.2: CBOE 10 Year Treasury Note (TNX) yield.
Cox-Ingersoll-Ross (CIR) model
The Cox, Ingersoll, and Ross, 1985 (CIR) model brings a solution to the positivity problem
encountered with the Vasicek model, by the use the nonlinear stochastic differential equation
√
drt = β (α − rt )dt + σ rt dBt , (1.1.4)

with α > 0, β > 0, σ > 0. The probability distribution of rt at time t > 0 admits the noncentral Chi
square probability density function given by
ft ( x ) (1.1.5)
−βt
! αβ /σ 2 −1/2 p !
2β 2β x + r0 e x 4β r0 x e −βt
= exp − I2αβ /σ 2 −1 ,
r0 e −βt

σ 2 1 − e −βt σ 2 1 − e −βt σ 2 1 − e −βt
x > 0, where
z λ (z2 /4)k
Iλ (z) := ∑ , z ∈ R,
2 k⩾0 k!Γ (λ + k + 1)
is the modified Bessel function of the first kind, see Lemma 9 in Feller, 1951 and Corollary 24
in Albanese and Lawi, 2005. Note that ft (x) is not defined at x = 0 if αβ /σ 2 − 1/2 < 0, i.e.
σ 2 > 2αβ , in which case the probability distribution of rt admits a point mass at x = 0. On the
other hand, rt remains almost surely strictly positive under the Feller condition 2αβ ⩾ σ 2 , cf. the
study of the associated probability density function in Lemma 4 of Feller, 1951 for α, β ∈ R.
Figure 1.3 presents a random simulation of t 7→ rt in the Cox, Ingersoll, and Ross, 1985 (CIR)
model in the case σ 2 > 2αβ , in which the process is mean reverting with respect to α = 2.5% and
has a nonzero probability of hitting 0.
7
rt (%)
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t
Figure 1.3: Graph of the CIR short rate t 7→ rt with α = 2.5%, β = 1, and σ = 1.3.
The next code provides a numerical solution of the stochastic differential equation (1.1.4) using
the Euler method, see Figure 1.3.
N=10000; t <- 0:(N-1); dt <- 1.0/N; nsim <- 2

a=0.025; b=1; sigma=0.1; sd=sqrt(sigma^2/2/b); R <- matrix(0,nsim,N);R[,1]=0.03
X <- matrix(rnorm(nsim*N,mean=0,sd=sqrt(dt)), nsim, N)
for (i in 1:nsim){for (j in 2:N){R[i,j]=max(0,R[i,j-1]+(a-b*R[i,j-1])*dt+sigma*sqrt(R[i,j-1])*X[i,j])}}
plot(t,R[1,],xlab="time",ylab="",type="l",ylim=c(0,R[1,1]+sd/5),col=0,axes=FALSE)
axis(2, pos=0)
for (i in 1:nsim){lines(t, R[i, ], xlab = "time", type = "l", col = i+8)}
abline(h=a/b,col="blue",lwd=3);abline(h=0)
In large time t → ∞, using the asymptotics
1 z λ
Iλ (z) ≃z→0 ,
Γ (λ + 1) 2

the probability density function (1.1.5) becomes the gamma density function
2αβ /σ 2
1 2β 2 2
f (x) = lim ft (x) = x−1+2αβ /σ e −2β x/σ , x > 0. (1.1.6)
t→∞ Γ(2αβ /σ 2 ) σ2
with shape parameter 2αβ /σ 2 and scale parameter σ 2 /(2β ), which is also the invariant distribu-
tion of rt .
The family of classical mean-reverting models also includes the Courtadon, 1982 model
drt = β (α − rt )dt + σ rt dBt ,
where α, β , σ are nonnegative, cf. Exercise 1.5, and the exponential Vasicek model
drt = rt (η − a log rt )dt + σ rt dBt ,
where a, η, σ > 0.
Constant Elasticity of Variance (CEV) model

Constant Elasticity of Variance models are designed to take into account nonconstant volatilities
that can vary as a power of the underlying asset price. The Marsh and Rosenfeld, 1983 short-term
interest rate model
γ−1
drt = (β rt + αrt )dt + σ rtγ/2 dBt , (1.1.7)
where α ∈ R, β , σ > 0 are constants and γ > 0 is the variance (or diffusion) elasticity coefficient,
covers most of the CEV models. Here, the elasticity coefficient is defined as ratio
dv2 (r )/v2 (r )
dr/r
between the relative change dv(r )/v(r ) in the variance v(r ) and the relative change dr/r in r.
Denoting by v2 (r ) := σ 2 rγ the variance coefficient in (1.1.7), constant elasticity refers to the
constant ratio
dv2 (r )/v2 (r ) r dv(r ) d log v(r ) d log rγ/2

=2 =2 =2 = γ.
dr/r v(r ) dr d log r d log r
For γ = 1, (1.1.7) yields the Cox, Ingersoll, and Ross, 1985 (CIR) equation
√
drt = (β + αrt )dt + σ rt dBt .
For β = 0 we get the standard CEV model
γ/2
drt = αrt dt + σ rt dBt ,
and for γ = 2 and β = 0 this yields the Dothan, 1978 model
drt = αrt dt + σ rt dBt ,
which is a version of geometric Brownian motion used for short-term interest rate modeling.

Time-dependent affine models

The class of short rate interest rate models admits a number of generalizations (see the references
quoted in the introduction of this chapter), including the class of affine models of the form
q
drt = (η (t ) + λ (t )rt )dt + δ (t ) + γ (t )rt dBt . (1.1.8)
Such models are called affine because the associated bonds can be priced using an affine PDE of
the type (1.4.9) below with solution of the form (1.4.10), as will be seen after Proposition 1.2.
The family of affine models also includes:

i) the Ho and Lee, 1986 model
drt = θ (t )dt + σ dBt ,
where θ (t ) is a deterministic function of time, as an extension of the Merton model drt =
θ dt + σ dBt ,
ii) the Hull and White, 1990 model
drt = (θ (t ) − α (t )rt )dt + σ (t )dBt
which is a time-dependent extension of the Vasicek model (1.1.1), with the explicit solution
rt w t rt wt rt
rt = r0 e − 0 α (τ )dτ + e − u α (τ )dτ θ (u)du + σ (u) e − u α (τ )dτ dBu ,
0 0
t ⩾ 0.
1.2 Calibration of the Vasicek Model

The Vasicek equation (1.1.1), i.e.
drt = (a − brt )dt + σ dBt ,
can be discretized according to a discrete-time sequence (tk )k⩾0 = (t0 ,t1 ,t2 , . . .) of time instants, as
rtk+1 − rtk = (a − brtk )∆t + σ Zk , k ⩾ 0,
where ∆t := tk+1 − tk and (Zk )k⩾0 is a Gaussian white noise with variance ∆t, i.e. a sequence of
independent, centered and identically distributed N (0, ∆t ) Gaussian random variables, which
yields
rtk+1 = rtk + (a − brtk )∆t + σ Zk = a∆t + (1 − b∆t )rtk + σ Zk , k ⩾ 0.
Based on a set (r̃tk )k=0,1,...,n of market data, we consider the quadratic residual
n−1
2
∑ (r̃t k +1
− a∆t − (1 − b∆t )r̃tk ) (1.2.1)
k =0
which represents the (squared) quadratic distance between the observed data sequence (r̃tk )k=1,2,...,n
and its predictions a∆t + (1 − b∆t )r̃tk k=0,1,...,n−1 .
In order to minimize the residual (1.2.1) over a and b we use Ordinary Least Square (OLS)
regression, and equate the following derivatives to zero. Namely, we have
∂ n−1
(r̃tl +1 − a∆t − (1 − b∆t )r̃tl )2
∂ a l∑
=0

!
n−1
= −2∆t −an∆t + ∑ (r̃tl +1 − (1 − b∆t )r̃tl )
l =0
= 0,
hence
1 n−1
a∆t = (r̃tl +1 − (1 − b∆t )r̃tl ) ,
n l∑
=0
and
∂ n−1
(r̃tk+1 − a∆t − (1 − b∆t )r̃tk )2
∂ b k∑
=0
n−1
= 2∆t ∑ r̃t k (−a∆t + r̃tk+1 − (1 − b∆t )r̃tk )
k =0
!
n−1
1 n−1
= 2∆t ∑ r̃t k
r̃tk+1 − (1 − b∆t )r̃tk − ∑ (r̃tl +1 − (1 − b∆t )r̃tl )
k =0 n l =0
!
n−1
∆t n−1 n−1
1 n−1
= 2∆t ∑ r̃tk r̃tk+1 − r̃tk r̃tl +1 − ∆t (1 − b∆t ) ( ) 2
−
∑ k n ∑ r̃tk r̃tl
r̃
n k,l∑
t
k =0 =0 k =0 k,l =0
= 0.
This leads to estimators for the parameters a and b, respectively as the empirical mean and
covariance of (r̃tk )k=0,1,...,n , i.e.
1 n−1

a ∆t = r̃ − ( 1 − b∆t ) r̃tk ,

n k∑
tk + 1

 b b

= 0







and








n−1
1 n−1



r̃ r̃ −
∑ tk tk+1 n ∑ r̃tk r̃tl+1




k =0 k,l =0
1−b b∆t = n−1 (1.2.2)

2 1 n−1
−




 ( t )
∑ k n ∑ r̃tk r̃tl
r̃


 k =0 k,l =0 ! !
n−1 n−1 n−1
1 1


∑ r̃tk − n ∑ r̃tl r̃tk+1 − n ∑ r̃tl+1





 k =0 l =0 l =0


 = !2 .
 n−1 n−1

 1


 ∑ r̃tk − n ∑ r̃tk
k =0 k =0
This also yields
σ 2 ∆t = Var[σ Zk ]
2
≃ IE r̃tk+1 − (1 − b∆t )r̃tk − a∆t , k ⩾ 0,
hence σ can be estimated as
1 n−1 2
σb 2 ∆t = ∑ b∆t ) − ab∆t .
r̃tk+1 − r̃tk (1 − b (1.2.3)
n k =0

Exercise. Show that (1.2.3) can be recovered by minimizing the residual

n−1 2 2
η 7→ ∑ b∆t ) − ab∆t
r̃tk+1 − r̃tk (1 − b − η∆t
k =0
as a function of η > 0, see also Exercise 1.3.
Time series modeling
Defining brtn := rtn − a/b, n ⩾ 0, we have
a
rtn+1 = rtn+1 −
b
b
a
= rtn − + (a − brtn )∆t + σ Zn
b
a a
= rtn − − b rtn − ∆t + σ Zn
b b
= b rtn ∆t + σ Zn
rtn − bb
= (1 − b∆t )b
rtn + σ Zn , n ⩾ 0.
In other words, the sequence (b
rtn )n⩾0 is modeled according to an autoregressive AR(1) time series
(Xn )n⩾0 with parameter α = 1 − b∆t, in which the current state Xn of the system is expressed as
the linear combination
Xn := σ Zn + αXn−1 , n ⩾ 1, (1.2.4)
where (Zn )n⩾1 another Gaussian white noise sequence with variance ∆t. This equation can be
solved recursively as the causal series
Xn = σ Zn + α (σ Zn−1 + αXn−2 ) = · · · = σ ∑ α k Zn−k ,
k⩾0
which converges when |α| < 1, i.e. |1 − b∆t| < 1, in which case the time series (Xn )n⩾0 is weakly
stationary, with
IE[Xn ] = σ ∑ α k IE[Zn−k ]
k⩾0
= σ IE[Z0 ] ∑ α k
k⩾0
σ
= IE[Z0 ]
1−α
= 0, n ⩾ 0.
The variance of Xn is given by
" #
Var[Xn ] = σ 2 Var ∑ α k Zn−k
k⩾0
= σ ∆t ∑ α
2 2k
k⩾0
= σ ∆t ∑ (1 − b∆t )2k
2
k⩾0
σ 2 ∆t
=
1 − (1 − b∆t )2
σ 2 ∆t
=
2b∆t − b2 (∆t )2
σ2
≃ , [∆t ≃ 0],
2b
which coincides with the variance (1.1.3) of the Vasicek process in the stationary regime.

Example - TNX yield calibration

The next code is estimating the parameters of the Vasicek model using the 10 Year Treasury
Note yield data of Figure 1.2, by implementing the formulas (1.2.2).
ratek=as.vector(rate);ratekplus1 <- c(ratek[-1],0)
oneminusbdt <- (sum(ratek*ratekplus1) - sum(ratek)*sum(ratekplus1)/n)/(sum(ratek*ratek) -
sum(ratek)*sum(ratek)/n)
adt <- sum(ratekplus1)/n-oneminusbdt*sum(ratek)/n;
sigmadt <- sqrt(sum((ratekplus1-oneminusbdt*ratek-adt)^2)/n)
Parameter estimation can also be implemented using the linear regression command
lm(c(diff(ratek)) ∼ ratek[1:length(ratek)-1])
in , which estimates the values of a∆t ≃ 0.017110 and −b∆t ≃ −0.007648 in the regression
rtk+1 − rtk = (a − brtk )∆t + σ Zk , k ⩾ 0,
Coefficients:
(Intercept) ratek[1:length(ratek) - 1]
0.017110 -0.007648
for (i in 1:100) {ar.sim<-arima.sim(model=list(ar=c(oneminusbdt)),n.start=100,n)

y=adt/oneminusbdt+sigmadt*ar.sim;y=y+ratek[1]-y[1]
time <- as.POSIXct(time(rate), format = "%Y-%m-%d")
sim_yield <- xts(x = y, order.by = time);
myPars <- chart_pars();myPars$cex<-1.4
myTheme <- chart_theme();myTheme$col$line.col <- "blue"
myTheme$rylab <- FALSE; dev.new(width=24,height=8)
par(mfrow=c(1,2));print(chart_Series(rate,theme=myTheme,pars=myPars))
graph <-chart_Series(sim_yield,theme=myTheme,pars=myPars); myylim <- graph$get_ylim()
myylim[[2]] <- structure(c(min(rate),max(rate)), fixed=TRUE)
graph$set_ylim(myylim); print(graph); Sys.sleep(1);dev.off()}
The above code is generating Vasicek random samples according to the AR(1) time series
(1.2.4), see Figure 1.4.
rate [2012−01−03/2015−12−31]
Last 2.269 3.0
2.5
2.0
1.5
Jan 03 Jul 02 Jan 02 Jul 01 Jan 02 Jul 01 Jan 02 Jul 01 Dec 31

2012 2012 2013 2013 2014 2014 2015 2015 2015
(a) CBOE TNX market yield. (b) Calibrated Vasicek sample path.
Figure 1.4: Calibrated Vasicek simulation vs. market data.
The package Sim.DiffProc can also be used to estimate the coefficients a∆t and b∆t.

install.packages("Sim.DiffProc");library(Sim.DiffProc)
fx <- expression( theta[1]-theta[2]*x ); gx <- expression( theta[3] )
fitsde(data = as.ts(ratek), drift = fx, diffusion = gx, start = list(theta1=0.1, theta2=0.1,
theta3=0.1),pmle="euler")
1.3 Zero-Coupon and Coupon Bonds

A zero-coupon bond is a contract priced P(t, T ) at time t < T to deliver the face value (or par
value) P(T , T ) = $1 at time T . In addition to its value at maturity, a bond may yield a periodic
coupon payment at regular time intervals until the maturity date.
Figure 1.5: Five-dollar 1875 Louisiana bond with 7.5% biannual coupons and maturity T = 1/1/1886.
The computation of the arbitrage-free price P0 (t, T ) of a zero-coupon bond based on an underlying
short-term interest rate process (rt )t∈R+ is a basic and important issue in interest rate modeling.
Constant short rate

In case the short-term interest rate is a constant rt = r, t ⩾ 0, a standard arbitrage argument shows
that the price P(t, T ) of the bond is given by
P(t, T ) = e −r(T −t ) , 0 ⩽ t ⩽ T.
Indeed, if P(t, T ) > e −r(T −t ) we could issue a bond at the price P(t, T ) and invest this amount at
the compounded risk free rate r, which would yield P(t, T ) e r(T −t ) > 1 at time T .
On the other hand, if P(t, T ) < e −r(T −t ) we could borrow P(t, T ) at the rate r to buy a bond priced
P(t, T ). At maturity time T we would receive $1 and refund only P(t, T ) e r(T −t ) < 1.
The price P(t, T ) = e −r(T −t ) of the bond is the value of P(t, T ) that makes the potential profit
P(t, T ) e r(T −t ) − 1 vanish for both traders.
Time-dependent deterministic short rates

Similarly to the above, when the short-term interest rate process (r (t ))t∈R+ is a deterministic
function of time, a similar argument shows that
rT
P(t, T ) = e − t r (s)ds
, 0 ⩽ t ⩽ T. (1.3.1)

Stochastic short rates

In case (rt )t∈R+ is an (Ft )t∈R+ -adapted random process the formula (1.3.1) is no longer valid as it
relies on future information, and we replace it with the averaged discounted payoff
h rT i
P(t, T ) = IE∗ e − t rs ds Ft , 0 ⩽ t ⩽ T, (1.3.2)
under a risk-neutral probability measure P∗ . It is natural to write P(t, T ) as a conditional expec-

tation under a martingale measure, as the use of conditional expectation
w helps to “filter out” the
T
(random/unknown) future information past time t contained in rs ds. The expression (1.3.2)
t rT
makes sense as the “best possible estimate” of the future quantity e − t rs ds
in mean-square sense,
given the information known up to time t.
Coupon bonds
Pricing bonds with nonzero coupon is not difficult since in general the amount and periodicity of
coupons are deterministic.* In the case of a succession of coupon payments c1 , c2 , . . . , cn at times
T1 , T2 , . . . , Tn ∈ (t, T ], another application of the above absence of arbitrage argument shows that
the price Pc (t, T ) of the coupon bond with discounted (deterministic) coupon payments is given by
the linear combination of zero-coupon bond prices
" #
n rT
rT

∗ − k ∗ −
Pc (t, T ) := IE ∑ ck e t rs ds
Ft + IE e t rs ds
Ft
k =1
n rT

∗ − k
= ∑ ck IE e t rs ds
Ft + P0 (t, T )
k =1
n
= P0 (t, T ) + ∑ ck P0 (t, Tk ), 0 ⩽ t ⩽ T1 , (1.3.3)
k =1
which represents the present value at time t of future $c1 , $c2 , . . . , $cn receipts respectively at times
T1 , T2 , . . . , Tn , in addition to a terminal $1 principal payment.
In the case of a constant coupon rate c paid at regular time intervals τ = Tk+1 − Tk , k =
0, 1, . . . , n − 1, with T0 = t, Tn = T , and a constant deterministic short rate r, we find
n
Pc (T0 , Tn ) = e −rnτ + c ∑ e −(Tk −T0 )r
k =1
n
= e −rnτ + c ∑ e −krτ
k =1
e −rτ − e −r(n+1)τ
= e −rnτ + c .
1 − e −rτ
In terms of the discrete-time interest rate r̃ := e rτ − 1, we also have
c r̃ − c
Pc (T0 , Tn ) = + .
r̃ (1 + r̃ )n r̃
* However, coupon default cannot be excluded.

1.8
~
c>r
~
c<r
1.6 ~
c=r
1.4
Bond price
1.2
0.8
0.6
0.4
0.2
0 2 4 6 8 10
t
Figure 1.6: Discrete-time coupon bond pricing.
In the case of a continuous-time coupon rate c > 0, the above discrete-time calculation (1.3.3) can
be reinterpreted as follows:
wT
Pc (t, T ) = P0 (t, T ) + c P0 (t, u)du (1.3.4)
t
wT
= P0 (t, T ) + c e −(u−t )r du
t
w T −t
= e −(T −t )r + c e −ru du
0
c
= e −(T −t )r + 1 − e −(T −t )r ,

r
c r − c −(T −t )r
= + e , 0 ⩽ t ⩽ T, (1.3.5)
r r
where the coupon bond price Pc (t, T ) solves the ordinary differential equation
dPc (t, T ) = (r − c) e −(T −t )r dt = −cdt + rPc (t, T )dt, 0 ⩽ t ⩽ T,
see also Figures 1.7 and 1.11 below.
1.8
c>r
c<r
1.6 c=r
1.4
Bond price
1.2
0.8
0.6
0.4
0.2
0 2 4 6 8 10
t
Figure 1.7: Continuous-time coupon bond pricing.
In what follows, we will mostly consider zero-coupon bonds priced as P(t, T ) = P0 (t, T ), 0 ⩽ t ⩽ T ,
in the setting of stochastic short rates.

Martingale property of discounted bond prices

The following proposition shows that Assumption 1 of Chapter 3 is satisfied, in other words, the
bond price process t 7→ P(t, T ) can be used as a numéraire.
Proposition 1.1 The discounted bond price process

rt
t 7→ Pe(t, T ) := e − 0 rs ds P(t, T )
is a martingale under P∗ .
Proof. By (1.3.2) we have
rt
Pe(t, T ) = e− 0 rs ds
P(t, T )
rt h rT i
= e − 0 rs ds IE∗ e − t rs ds Ft
h rt rT i
= IE∗ e − 0 rs ds e − t rs ds Ft
h rT i
= IE∗ e − 0 rs ds Ft , 0 ⩽ t ⩽ T,
and this suffices in order to conclude, since by the tower property of conditional expectations, any
process (Xt )t∈R+ of the form t 7→ Xt := IE∗ [F | Ft ], F ∈ L1 (Ω), is a martingale. In other words,
we have
r
T
∗ e ∗ ∗ − 0 rs ds
IE P(t, T ) Fu = IE IE e Ft Fu

r
T
∗ − 0 rs ds
= IE e Fu
= Pe(u, T ), 0 ⩽ u ⩽ t.
□
1.4 Bond Pricing PDE

We assume from now on that the underlying short rate process solves the stochastic differential
equation
drt = µ (t, rt )dt + σ (t, rt )dBt (1.4.1)
where (Bt )t∈R+ is a standard Brownian motion under P∗ . Note that specifying the dynamics of
(rt )t∈R+ under the historical probability measure P will also lead to a notion of market price of
risk (MPoR) for the modeling of short rates.
As all solutions of stochastic differential equations such as (1.4.1) have the Markov property, cf.
e.g. Theorem V-32 of Protter, 2004, the arbitrage-free price P(t, T ) can be rewritten as a function
F (t, rt ) of rt , i.e.
h rT i
P(t, T ) = IE∗ e − t rs ds Ft
h rT i
= IE∗ e − t rs ds rt
= F (t, rt ), (1.4.2)
and depends on (t, rt ) only, instead of depending on the whole information available in Ft up to
time t, meaning that the pricing problem can now be formulated as a search for the function F (t, x).

Proposition 1.2 (Bond pricing PDE). Consider a short rate (rt )t∈R+ modeled by a diffusion
equation of the form
drt = µ (t, rt )dt + σ (t, rt )dBt .
The bond pricing PDE for P(t, T ) = F (t, rt ) as in (1.4.2) is written as
∂F ∂F 1 ∂ 2F
xF (t, x) = (t, x) + µ (t, x) (t, x) + σ 2 (t, x) 2 (t, x), (1.4.3)
∂t ∂x 2 ∂x
t ⩾ 0, x ∈ R, subject to the terminal condition
F (T , x) = 1, x ∈ R. (1.4.4)
In addition, the bond price dynamics is given by
∂F
dP(t, T ) = rt P(t, T )dt + σ (t, rt ) (t, rt )dBt . (1.4.5)
∂x
Proof. By Itô’s formula, we have
rt rt rt
d e − 0 rs ds P(t, T ) = −rt e − 0 rs ds P(t, T )dt + e − 0 rs ds dP(t, T )
rt rt
= −rt e − 0 rs ds F (t, rt )dt + e − 0 rs ds
dF (t, rt )
rt rt ∂F
= −rt e − 0 rs ds F (t, rt )dt + e − 0 rs ds (t, rt )drt
∂x
1 rt ∂ 2F rt ∂F
+ e − 0 rs ds 2 (t, rt )(drt )2 + e − 0 rs ds (t, rt )dt
2 ∂x ∂t
rt rt ∂F
= −rt e − 0 rs ds F (t, rt )dt + e − 0 rs ds (t, rt )( µ (t, rt )dt + σ (t, rt )dBt )
∂x
rt ∂ 2F

− 0 rs ds 1 2 ∂F
+e σ (t, rt ) 2 (t, rt ) + (t, rt ) dt
2 ∂x ∂t
rt ∂F
= e − 0 rs ds σ (t, rt ) (t, rt )dBt
∂x
rt ∂ 2F

− 0 rs ds ∂F 1 2 ∂F
+e −rt F (t, rt ) + µ (t, rt ) (t, rt ) + σ (t, rt ) 2 (t, rt ) + (t, rt ) dt.
∂x 2 ∂x ∂t
(1.4.6)
rt
Given that t 7→ e − 0 rs ds P(t, T ) is a martingale, the above expression (1.4.6) should only contain
terms in dBt (cf. Corollary II-6-1, page 72 of Protter, 2004), and all terms in dt should vanish
inside (1.4.6). This leads to the identities

 ∂F 1 2 ∂ 2F ∂F

 rt F (t, rt ) = µ (t, rt ) (t, rt ) + σ (t, rt ) 2
(t, rt ) + (t, rt )
x 2 x


 ∂ ∂ ∂t
 rt

 rt ∂F
d e − 0 rs ds P(t, T ) = e − 0 rs ds σ (t, rt ) (t, rt )dBt ,

 (1.4.7a)
∂x
which lead to (1.4.3) and (1.4.5). Condition (1.4.4) is due to the fact that P(T , T ) = $1. □

By (1.4.7a), the proof of Proposition 1.2 also shows that
dP(t, T ) 1 rt rt
= d e 0 rs ds e − 0 rs ds P(t, T )
P(t, T ) P(t, T )
1 rt rt
= rt P(t, T )dt + e 0 rs ds d e − 0 rs ds P(t, T )
P(t, T )
1 rt rt
= rt dt + e 0 rs ds d e − 0 rs ds P(t, T )
P(t, T )
1 ∂F
= rt dt + (t, rt )σ (t, rt )dBt
F (t, rt ) ∂ x
∂
= rt dt + σ (t, rt ) log F (t, rt )dBt . (1.4.8)
∂x
In the case of an interest rate process modeled by (1.1.8), we have

q
µ (t, x) = η (t ) + λ (t )x, and σ (t, x) = δ (t ) + γ (t )x,
hence (1.4.3) yields the affine PDE
∂F ∂F 1 ∂ 2F
xF (t, x) = (t, x) + (η (t ) + λ (t )x) (t, x) + (δ (t ) + γ (t )x) 2 (t, x)
∂t ∂x 2 ∂x
(1.4.9)
with time-dependent coefficients, t ⩾ 0, x ∈ R.
Note that more generally, all affine short rate models as defined in Relation (1.1.8), including the
Vasicek model, will yield a bond pricing formula of the form
P(t, T ) = e A(T −t )+rt C(T −t ) , (1.4.10)
cf. e.g. § 3.2.4. in Brigo and Mercurio, 2006.
Vašíček, 1977 model

In the Vasicek case
drt = (a − brt )dt + σ dBt ,
the bond price takes the form
F (t, rt ) = P(t, T ) = e A(T −t )+rt C(T −t ) ,
where A(·) and C (·) are deterministic functions of time, see (1.4.16)-(1.4.17) below, and (1.4.8)
yields
dP(t, T )
= rt dt + σC (T − t )dBt (1.4.11)
P(t, T )
σ
1 − e −(T −t )b dBt ,

= rt dt −
b
since F (t, x) = e A(T −t )+xC(T −t ) .

Probabilistic solution of the Vasicek PDE

Next, we solve the PDE (1.4.3), written with µ (t, x) = a − bx and σ (t, x) = σ in the Vašíček, 1977
model
drt = (a − brt )dt + σ dBt (1.4.12)
as

∂F ∂F σ 2 ∂ 2F
 xF (t, x) = (t, x) + (a − bx) (t, x) + (t, x),


∂t ∂x 2 ∂ x2 (1.4.13)


 F (T , x) = 1.
For this, Proposition 1.3 relies on a direct computation of the conditional expectation
h rT i
F (t, rt ) = P(t, T ) = IE∗ e − t rs ds Ft . (1.4.14)
See also Exercise 1.7 for a closed-form bond pricing formula in the Cox, Ingersoll, and Ross, 1985
(CIR) model.
Proposition 1.3 The zero-coupon bond price in the Vasicek model (1.4.12) can be expressed as
P(t, T ) = e A(T −t )+rt C(T −t ) , 0 ⩽ t ⩽ T, (1.4.15)
where A(x) and C (x) are functions of time to maturity given by
1
1 − e −bx ,

C (x ) : = − (1.4.16)
b
and
4ab − 3σ 2 σ 2 − 2ab σ 2 − ab −bx σ 2 −2bx
A(x ) : = 3
+ x + e − 3e (1.4.17)
4b 2b2 b3 4b
a σ2 σ2 2

= − − (x + C (x)) − C (x), x ⩾ 0.
b 2b2 4b
Proof. Recall that in the Vasicek model (1.4.12), the short rate process (rt )t∈R+ solution of (1.4.12)
has the expression
wt a wt
rt = g(t ) + h(t, s)dBs = r0 e −bt + 1 − e −bt + σ e −(t−s)b dBs ,

0 b 0
see Exercise 1.1, where g and h are the deterministic functions

a
g(t ) := r0 e −bt + 1 − e −bt ,

t ⩾ 0,
b
and
h(t, s) := σ e −(t−s)b , 0 ⩽ s ⩽ t.
Using the fact that Wiener integrals are Gaussian random variables and the Gaussian moment
generating function, and exchanging the order of integration between ds and du over [t, T ] according
to the following picture:

u
t T
we have
h rT i
P(t, T ) = IE∗ e − t rs ds Ft
h rT rs i
= IE∗ e − t (g(s)+ 0 h(s,u)dBu )ds Ft
w h r r
T T s
i
= exp − g(s)ds IE∗ e − t 0 h(s,u)dBu ds Ft
t
w h r r
T T T i
= exp − g(s)ds IE∗ e − 0 Max(u,t ) h(s,u)dsdBu Ft
t
w wtwT h r r
T −
T T
h(s,u)dsdBu
i
= exp − g(s)ds − h(s, u)dsdBu IE∗ e t Max(u,t ) Ft
t 0 Max(u,t )
w w twT
h r r
T T T
i
= exp − g(s)ds − h(s, u)dsdBu IE∗ e − t u h(s,u)dsdBu Ft
t 0 t
w w twT
h r r
T T T
i
= exp − g(s)ds − h(s, u)dsdBu IE∗ e − t u h(s,u)dsdBu
t 0 t
wT wtwT 1wT wT
2 !
= exp − g(s)ds − h(s, u)dsdBu + h(s, u)ds du
t 0 t 2 t u
w wtwT
T
−bs a −bs
−(s−u)b
= exp − r0 e + 1 − e ds − σ e dsdBu
t b 0 t
σ 2 w T w T −(s−u)b
2 !
× exp e ds du
2 t u
w
w t −(t−u)b

T
−bs a −bs
σ −(T −t )b
= exp − r0 e + 1 − e ds − 1− e e dBu
t b b 0
σ 2 w T 2bu e −bu − e −bT

2 !
× exp e du
2 t b

rt 1 a
= exp − 1 − e −(T −t )b + 1 − e −(T −t )b r0 e −bt + 1 − e −bt
b b b
wT 2wT −bu − e −bT 2
!
a σ e
r0 e −bs + 1 − e −bs ds + e 2bu

× exp − du
t b 2 t b
= e A(T −t )+rt C(T −t ) , (1.4.18)
where A(x) and C (x) are the functions given by (1.4.16) and (1.4.17). □

Analytical solution of the Vasicek PDE
In order to solve the PDE (1.4.13) analytically, we may start by looking for a solution of the form
F (t, x) = e A(T −t )+xC(T −t ) , (1.4.19)
where A(·) and C (·) are functions to be determined under the conditions A(0) = 0 and C (0) = 0.
Substituting (1.4.19) into the PDE (1.4.3) with the Vasicek coefficients µ (t, x) = (a − bx) and
σ (t, x) = σ shows that
x e A(T −t )+xC(T −t ) = −(A′ (T − t ) + xC′ (T − t )) e A(T −t )+xC(T −t )

+(a − bx)C (T − t ) e A(T −t )+xC(T −t )
1
+ σ 2C2 (T − t ) e A(T −t )+xC(T −t ) ,
2
i.e.
1
x = −A′ (T − t ) − xC′ (T − t ) + (a − bx)C (T − t ) + σ 2C2 (T − t ).
2
By identification of terms for x = 0 and x ̸= 0, this yields the system of Riccati and linear differential
equations

σ2
 A′ (s) = aC (s) + C2 (s)


2

 C′ (s) = −1 − bC (s),

which can be solved to recover the above value of P(t, T ) = F (t, rt ) via
1
1 − e −bs

C (s) = −
b
and
wt
A(t ) = A(0) + A′ (s)ds
0
wt σ2 2

= aC (s) + C (s) ds
0 2
wt a

σ2

−bs −bs 2

= 1− e + 2 1− e ds
0 b 2b
awt σ2 w t 2
1 − e −bs ds + 2 1 − e −bs ds

=
b 0 2b 0
4ab − 3σ 2 σ 2 − 2ab σ 2 − ab −bt σ 2 −2bt
= + t + e − 3e , t ⩾ 0.
4b3 2b2 b3 4b
The next Figure 1.8 shows the output of the attached code for the Monte Carlo and analytical
estimation of Vasicek bond prices.

Bond prices estimates
1.0
0.9
0.8
0.7
Monte Carlo Estimation
Analytical PDE Solution
0.6
0.2 0.4 0.6 0.8 1.0
r0
(a) Short rate paths. (b) Comparison with the PDE solution.
Figure 1.8: Comparison of Monte Carlo and analytical PDE solutions.
Vasicek bond price simulations
In this section we consider again the Vasicek model, in which the short rate (rt )t∈R+ is solution
to (1.1.1). Figure 1.9 presents a random simulation of the zero-coupon bond price (1.4.15) in the
Vasicek model with σ = 10%, r0 = 2.96%, b = 0.5, and a = 0.025. The graph of the corresponding
deterministic zero-coupon bond price with r = r0 = 2.96% is also shown in Figure 1.9.
0.8
P(t,T)
0.6
0.4
0.2
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t
Figure 1.9: Graphs of t 7→ F (t, rt ) = P(t, T ) vs. t 7→ e −r0 (T −t ) .
Figure 1.10 presents a random simulation of the coupon bond price (1.3.4) in the Vasicek model with
σ = 2%, r0 = 3.5%, b = 0.5, a = 0.025, and coupon rate c = 5%. The graph of the corresponding
deterministic coupon bond price (1.3.5) with r = r0 = 3.5% is also shown in Figure 1.10.

1.8
1.7
1.6
1.5
Pc(t,T)
1.4
1.3
1.2
1.1
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t
Figure 1.10: Graph of t 7→ Pc (t, T ) for a bond with a 5% coupon rate.
Figure 1.11 presents market price data for a coupon bond with coupon rate c = 6.25%.
Figure 1.11: Bond price graph with maturity 01/18/08 and coupon rate 6.25%.
Zero-coupon bond price and yield data
The following zero-coupon public bond price data was downloaded from EMMA at the Municipal
Securities Rulemaking Board.
ORANGE CNTY CALIF PENSION OBLIG CAP APPREC-TAXABLE-REF-SER A (CA)

CUSIP: 68428LBB9
Dated Date: 06/12/1996 (June 12, 1996)
Maturity Date: 09/01/2016 (September 1st, 2016)
Interest Rate: 0.0 %
Principal Amount at Issuance: $26,056,000
Initial Offering Price: 19.465

library(quantmod);getwd()
bondprice <- read.table("bond_data_R.txt",col.names =
c("Date","HighPrice","LowPrice","HighYield","LowYield","Count","Amount"))
head(bondprice)
time <- as.POSIXct(bondprice$Date, format = "%Y-%m-%d")
price <- xts(x = bondprice$HighPrice, order.by = time)
yield <- xts(x = bondprice$HighYield, order.by = time)
dev.new(width=10,height=7);
chartSeries(price,up.col="blue",theme="white")
chartSeries(yield,up.col="blue",theme="white")
Date HighPrice LowPrice HighYield LowYield Count Amount

1 2016-01-13 99.082 98.982 1.666 1.501 2 20000
2 2015-12-29 99.183 99.183 1.250 1.250 1 10000
3 2015-12-21 97.952 97.952 3.014 3.014 1 10000
4 2015-12-17 99.141 98.550 2.123 1.251 5 610000
5 2015-12-07 98.770 98.770 1.714 1.714 2 10000
6 2015-12-04 98.363 98.118 2.628 2.280 2 10000
price [2005−01−26/2016−01−13]
100
Last 99.082
90
80
70
60
50
Jan 26 Feb 08 Mar 27 Feb 06 Aug 06 Nov 01 Nov 09 May 16 Dec 29

2005 2006 2007 2009 2010 2011 2012 2014 2015
Figure 1.12: Orange Cnty Calif bond prices.
The next Figure 1.13 plots the bond yield y(t, T ) defined as
log P(t, T )
y(t, T ) = − , or P(t, T ) = e −(T −t )y(t,T ) , 0 ⩽ t ⩽ T,
T −t
see also here for another source of bond price market data.

yield [2005−01−26/2016−01−13]
Last 1.666 8
Jan 26 Feb 08 Mar 27 Feb 06 Aug 06 Nov 01 Nov 09 May 16 Dec 29

2005 2006 2007 2009 2010 2011 2012 2014 2015
Figure 1.13: Orange Cnty Calif bond yields.
Bond pricing in the Dothan model

In the Dothan, 1978 model, the short-term interest rate process (rt )t∈R+ is modeled according to a
geometric Brownian motion
drt = µrt dt + σ rt dBt , (1.4.20)
where the volatility σ > 0 and the drift µ ∈ R are constant parameters and (Bt )t∈R+ is a standard
Brownian motion. In this model the short-term interest rate rt remains always positive, while the
proportional volatility term σ rt accounts for the sensitivity of the volatility of interest rate changes
to the level of the rate rt .
On the other hand, the Dothan model is the only lognormal short rate model that allows for an
analytical formula for the zero-coupon bond price
h rT i
P(t, T ) = IE∗ e − t rs ds Ft , 0 ⩽ t ⩽ T.
For convenience of notation we let p = 1 − 2µ/σ 2 and rewrite (1.4.20) as

σ2
drt = (1 − p) rt dt + σ rt dBt ,
2
with solution
2 t/2
rt = r0 e σ Bt −pσ , t ⩾ 0. (1.4.21)
By the Markov property of (rt )t∈R+ , the bond price P(t, T ) is a function F (t, rt ) of rt and time
t ∈ [0, T ]:
h rT i
P(t, T ) = F (t, rt ) = IE∗ e − t rs ds rt , 0 ⩽ t ⩽ T. (1.4.22)
By computation of the conditional expectation (1.4.22) we obtain the following result, cf. Proposi-
tion 1.2 of C. Pintoux and N. Privault, 2011.
Proposition 1.4 The zero-coupon bond price P(t, T ) = F (t, rt ) is given for all p ∈ R by
F (t, x) (1.4.23)

w ∞w ∞ 1 + z2
!
4z (T − t )σ 2 du dz

−σ 2 p2 (T −t )/8 −ux
=e e exp −2 θ , ,
0 0 σ 2u σ 2u 4 u z p+1
x > 0.
w T −t 2
Proof. By Proposition 2 in Yor, 1992, the probability distribution of the time integral e σ Bs −pσ s/2 ds
0
is given by
w
T −t
σ Bs −pσ 2 s/2
P e ds ∈ dy
0
w ∞ w t 2

= P e σ Bs −pσ s/2 ds ∈ dy, Bt − pσt/2 ∈ dz
−∞ 0
w
σ ∞ −pσ z/2−p2 σ 2t/8

1 + eσ z
σ z/2 2
4e σ t dy
= e exp −2 2
θ 2
, dz
2 −∞ σ y σ y 4 y
w∞
1 + z2

4z (T − t )σ 2

dz dy
−(T −t ) p2 σ 2 /8
= e exp −2 2 θ , , y > 0,
0 σ y σ y2 4 z +1 y
p
where the exchange of integrals is justified by the Fubini theorem and the nonnegativity of integrands.
Hence, by (1.4.21) we find
F (t, rt ) = P(t, T )
w
T
∗
= IE exp − rs ds Ft
t
wT
∗ σ (Bs −Bt )−σ 2 p(s−t )/2
= IE exp −rt e ds Ft
t
w
T
∗ σ (Bs −Bt )−σ 2 p(s−t )/2
= IE exp −x e ds
t
x=rt
w
T −t
∗ σ Bs −σ 2 ps/2
= IE exp −x e ds
0
x=rt
w∞ w
T −t

−rt y −pσ 2 s/2
= e P e σ Bs
ds ∈ dy
0 0
w∞ w∞
1 + z2

4z (T − t )σ 2

dz dy
−(T −t ) p2 σ 2 /8 −rt y
= e e exp −2 2 θ , .
0 0 σ y 2
σ y 4 z +1 y
p
□
The zero-coupon bond price P(t, T ) = F (t, rt ) in the Dothan model can also be written for all
p ∈ R as
√ !
(2x) p/2 w ∞ −( p2 +u2 )σ 2t/8 p u 2 8x
F (t, x) = 2 p
ue sinh(πu) Γ − + i Kiu du
2π σ 0 2 2 σ
√ !
(2x) p/2 2( p − 2k)+ σ 2 k(k−p)t/2 8x
+ p ∑ e Kp−2k , x > 0, t > 0,
σ k⩾0 k! ( p − k ) ! σ
cf. Corollary 2.2 of C. Pintoux and N. Privault, 2010, see also Privault and Uy, 2013 for numer-
ical computations. Zero-coupon bond prices in the Dothan model can also be computed by the
conditional expression
w w w
T ∞ T
IE exp − rt dt = IE exp − rt dt rT = z dP(rT ⩽ z), (1.4.24)
0 0 0

where rT has the lognormal distribution
2 T /2 1 2 2 2
dP(rT ⩽ z) = dP(r0 e σ BT −pσ ⩽ z) = √ e −( pσ T /2+log(z/r0 )) /(2σ T ) .
2
z 2πσ T
In Proposition 1.5 we note that the conditional Laplace transform
w
T
IE exp − rt dt rT = z
0
cf. (1.4.28) above, can be computed by a closed-form integral expression based on the modified
Bessel function of the second kind
zζ w ∞ z2

du
Kζ (z) := exp −u − , ζ ∈ R, z ∈ C, (1.4.25)
2ζ +1 0 4u uζ +1
cf. e.g. Watson, 1995 page 183, provided that the real part R (z2 ) of z2 ∈ C is positive.
Proposition 1.5 (Privault and Yu, 2016, Proposition 4.1). Taking r0 = 1, for all λ , z > 0 we have
wT 2
r
4 e −σ T /8

λ
IE exp −λ rs ds rT = z = 3/2 2 (1.4.26)
0 π σ p(z) T
√ p √
w∞
4πξ

K1 8λ 1 + 2 z cosh ξ + z/σ

2(π 2 −ξ 2 )/(σ 2 T )
× e sin sinh(ξ ) √ dξ .
σ 2T
p
0 1 + 2 z cosh ξ + z
Note however that the numerical evaluation of (1.4.26) can fail for small values of T > 0, and for
this reason the integral can be estimated by a gamma approximation as in (1.4.27) below. Under
the gamma approximation we can approximate the conditional bond price on the Dothan short rate
rt as
wT
−ν (z)
IE exp −λ rt dt rT = z ≃ (1 + λ θ (z)) ,
0
where, letting
wT
ΛT : = rt dt,
0
the parameters ν (z) and θ (z) are determined by conditional moment fitting to a gamma distribution,
as
Var[ΛT | ST = z] (IE[ΛT | ST = z])2 IE[ΛT | ST = z]
θ (z) : = , ν (z) : = = ,
IE[ΛT | ST = z] Var[ΛT | ST = z] θ
cf. Privault and Yu, 2016, which yields
wT w
∞
IE exp −λ rs ds ≃ (1 + λ θ (z))−ν (z) dP(rT ⩽ z). (1.4.27)
0 0

2.5
Exact expression
Gamma approximation
2
Probability density
1.5
0.5
0
0 0.5 1 1.5 2 2.5 3
x
Figure 1.14: Fitting of a gamma probability density function.

The quantity θ (z) is also known in physics as the Fano factor or dispersion index that measures
the dispersion of the probability distribution of ΛT given that ST = z. Figures 1.15 shows that the
stratified gamma approximation (1.4.27) matches the Monte Carlo estimate, while the use of the
integral expressions (1.4.24) and (1.4.26) leads to numerical instabilities.
1
Stratified gamma
Monte Carlo
0.8 Integral expression
0.6
F(x,t)
0.4
0.2
0
0 1 2 3 4 5 6 7 8 9 T=10
t
Figure 1.15: Approximation of Dothan bond prices t 7→ F (t, x) with σ = 0.3 and T = 10.
Related computations for yield options in the Cox, Ingersoll, and Ross, 1985 (CIR) model can also
be found in Prayoga and Privault, 2017.
Path integrals in option pricing
Let ℏ denote the Planck constant, and let S(x(·)) denote the action functional given as
wT w T 1
2
S(x(·)) = L(x(s), ẋ(s), s)ds = m(ẋ(s)) −V (x(s)) ds,
t t 2
where L(x(s), ẋ(s), s) is the Lagrangian

1
L(x(s), ẋ(s), s) := m(ẋ(s))2 −V (x(s)).
2
In physics, the Feynman path integral
w
i

ψ (y,t ) := Dx(·) exp S(x(·))
x(t )=x, x(T )=y ℏ

solves the Schrödinger equation

∂ψ ℏ2 ∂ 2 ψ
iℏ (x,t ) = − (x,t ) + V (x(t ))ψ (x,t ).
∂t 2m ∂ x2
After the Wick rotation t 7→ −it, the function
w
1

φ (y,t ) := Dx(·) exp − S(x(·))
x(t )=x, x(T )=y ℏ
solves the heat equation
∂φ ℏ2 ∂ 2 φ
ℏ (x,t ) = − (x,t ) + V (x(t ))φ (x,t ).
∂t 2m ∂ x2
By reformulating the action functional S(x(·)) as
w T 1
2
S(x(·)) = m(ẋ(s)) + V (x(s)) ds
t 2
N
(x(ti ) − x(ti−1 ))2

≃ ∑ + V (x(ti−1 )) ∆ti ,
i=1 2(ti − ti−1 )2
we can rewrite the Euclidean path integral as
w
1

φ (y,t ) = Dx(·) exp − S(x(·))
x(t )=x, x(T )=y ℏ
w
!
1 N (x(ti ) − x(ti−1 ))2 1 N
= Dx(·) exp − ∑ − ∑ V (x(ti−1 ))
x(t )=x, x(T )=y 2ℏ i=1 2∆ti ℏ i=1

1 Tw
= IE∗ exp − V (Bs )ds Bt = x, BT = y .
ℏ t
This type of path integral computation
w
T
∗
φ (y,t ) = IE exp − V (Bs )ds Bt = x, BT = y . (1.4.28)
t
is particularly useful for bond pricing, as (1.4.28) can be interpreted as the price of a bond with
short-term interest rate process (rs )s∈R+ := (V (Bs )))s∈R+ conditionally to the value of the endpoint
BT = y, cf. (1.4.26) below. It can also be useful for exotic option pricing, and for risk management,
see e.g. Kakushadze, 2015. The path integral (1.4.28) can be estimated either by closed-form
expressions using Partial Differential Equations (PDEs) or probability densities, by approximations
such as (conditional) Moment matching, or by Monte Carlo estimation, from the paths of a Brownian
bridge as shown in Figure 1.16.
x
t T
Figure 1.16: Brownian bridge.

Exercises
Exercise 1.1 We consider the stochastic differential equation
drt = (a − brt )dt + σ dBt , (1.4.29)
where a, σ ∈ R, b > 0.
a) Show that the solution of (1.4.29) is
a wt
rt = r0 e −bt + 1 − e −bt + σ e −(t−s)b dBs ,

t ⩾ 0. (1.4.30)
b 0
b) Show that the Gaussian N (a/b, σ 2 /(2b)) distribution is the invariant (or stationary) distri-
bution of (rt )t∈R+ .
Exercise 1.2 (Brody, Hughston, and Meier, 2018) In the mean-reverting Vasicek model (1.4.30)
with b > 0, compute:
i) The asymptotic bond yield, or exponential long rate of interest
log P(t, T )
r∞ := − lim .
T →∞ T −t
ii) The long-bond return
P(t, T )
Lt := lim .
T →∞ P(0, T )
Hint: Start from log(P(t, T )/P(0, T )).
Exercise 1.3 Consider the Chan-Karolyi-Longstaff-Sanders (CKLS) interest rate model (Chan
et al., 1992) parametrized as
γ
drt = (a − brt )dt + σ rt dBt ,
and time-discretized as
rtk+1 = rtk + (a − brtk )∆t + σ rtγk Zk

= a∆t + (1 − b∆t )rtk + σ rtγk Zk , k ⩾ 0,
where ∆t := tk+1 − tk and (Zk )k⩾0 is an i.i.d. sequence of N (0, ∆t ) random variables. Assuming
that a, b, γ > 0 are known, find an unbiased estimator σb 2 for the variance coefficient σ 2 , based on a
market data set (r̃tk )k=0,1,...,n .
Exercise 1.4 Let (Bt )t∈R+ denote a standard Brownian motion started at 0 under the risk-neutral
probability measure P∗ . We consider a short-term interest rate process (rt )t∈R+ in a Ho-Lee model
with constant deterministic volatility, defined by
drt = adt + σ dBt , (1.4.31)
where a ∈ R and σ > 0. Let P(t, T ) denote the arbitrage-free price of a zero-coupon bond in this
model:
w
T
P(t, T ) = IE∗ exp − rs ds Ft , 0 ⩽ t ⩽ T. (1.4.32)
t

a) State the bond pricing PDE satisfied by the function F (t, x) defined via
w
T
∗
F (t, x) := IE exp − rs ds rt = x , 0 ⩽ t ⩽ T.
t
b) Compute the arbitrage-free price F (t, rt ) = P(t, T ) from its expression (1.4.32) as a condi-
tional expectation.
Hint: One
w T may use the integrationwbyT parts relation
Bs ds = T BT − tBt − sdBs
t t
wT
= (T − t )Bt + T (BT − Bt ) − sdBs
t
wT
= (T − t )Bt + (T − s)dBs ,
t
2 2
and the Gaussian moment generating function IE[ e λ X ] = e λ η /2 for X ≃ N (0, η 2 ).
c) Check that the function F (t, x) computed in Question (b)) does satisfy the PDE derived in
Question (a)).
Exercise 1.5 Consider the Courtadon, 1982 model
drt = β (α − rt )dt + σ rt dBt , (1.4.33)
where α, β , σ are nonnegative, which is a particular case of the Chan-Karolyi-Longstaff-Sanders

(CKLS) model (Chan et al., 1992) with γ = 1.
a) Show that the solution of (1.4.33) is given by
wt S
t
rt = αβ du + r0 St , t ⩾ 0, (1.4.34)
0 Su
where (St )t∈R+ is the geometric Brownian motion solution of dSt = −β St dt + σ St dBt with
S0 = 1.
b) Given that the discounted bond price process is a martingale, derive the bond pricing PDE
satisfied by the function F (t, x) such that
h rT i h rT i
F (t, rt ) = P(t, T ) = IE∗ e − t rs ds Ft = IE∗ e − t rs ds rt .
Exercise 1.6 Consider the Marsh and Rosenfeld, 1983 short-term interest rate model
γ−1 γ/2
drt = β rt + αrt dt + σ rt dBt
where α ∈ R and β , σ , γ > 0.

2−γ
a) Letting Rt := rt , t ⩾ 0, find the stochastic differential equation satisfied by (Rt )t∈R+ .
b) Given that the discounted bond price process is a martingale, derive the bond pricing PDE
satisfied by the function F (t, x) such that
h rT i h rT i
F (t, rt ) = P(t, T ) = IE∗ e − t rs ds Ft = IE∗ e − t rs ds rt .
Exercise 1.7 Consider the Cox, Ingersoll, and Ross, 1985 (CIR) process (rt )t∈R+ solution of
√
drt = −art dt + σ rt dBt ,
where a, σ > 0 are constants (Bt )t∈R+ is a standard Brownian motion started at 0.

a) Write down the bond pricing PDE for the function F (t, x) given by
w
T
∗
F (t, x) := IE exp − rs ds rt = x , 0 ⩽ t ⩽ T.
t
Hint: Use Itô calculus and the fact that the discounted bond price is a martingale.
b) Show that the PDE found in Question (a)) admits a solution of the form F (t, x) = e A(T −t )+xC(T −t )
where the functions A(s) and C (s) satisfy ordinary differential equations to be also written
down together with the values of A(0) and C (0).
Exercise 1.8 Convertible bonds. Consider an underlying asset price process (St )t∈R+ given by
(1)
dSt = rSt dt + σ St dBt ,
and a short-term interest rate process (rt )t∈R+ given by
(2)
drt = γ (t, rt )dt + η (t, rt )dBt ,
(1) (2)
where Bt t∈R+
and Bt t∈R+
are two correlated Brownian motions under the risk-neutral
( 1 ) (2)
probability measure P∗ , with dBt • dBt = ρdt. A convertible bond is made of a corporate bond
priced P(t, T ) at time t ∈ [0, T ], that can be exchanged into a quantity α > 0 of the underlying
company’s stock priced Sτ at a future time τ, whichever has a higher value, where α is a conversion
rate.
a) Find the payoff of the convertible bond at time τ.
b) Rewrite the convertible bond payoff at time τ as the linear combination of P(τ, T ) and a call
option payoff on Sτ , whose strike price is to be determined.
c) Write down the convertible bond price at time t ∈ [0, τ ] as a function C (t, St , rt ) of the
underlying asset price and interest rate, using a discounted conditional expectation, and show
that the discounted corporate bond price
rt
e− 0 rs ds
C (t, St , rt ), t ∈ [0, τ ],
is a martingale. r
t
d) Write down d e − 0 rs dsC (t, St , rt ) using the Itô formula and derive the pricing PDE satisfied

by the function C (t, x, y) together with its terminal condition.

e) Taking the bond price P(t, T ) as a numéraire, price the convertible bond as a European option
with strike price K = 1 on an underlying asset priced Zt := St /P(t, T ), t ∈ [0, τ ] under the
forward measure P b with maturity T .
f) Assuming the bond price dynamics
dP(t, T ) = rt P(t, T )dt + σB (t )P(t, T )dBt ,
determine the dynamics of the process (Zt )t∈R+ under the forward measure P.
b
g) Assuming that (Zt )t∈R+ can be modeled as a geometric Brownian motion, price the convert-
ible bond using the Black-Scholes formula.
Exercise 1.9 Bond duration. Compute the duration
1+r ∂
Dc (0, n) := − Pc (0, n)
Pc (0, n) ∂ r

of a discrete-time coupon bond priced as

n
1 1
Pc (0, n) = n
+ c ∑ k
(1 + r ) k =1 (1 + r )
c c 1
= + 1− ,
r r (1 + r )n
where r > 0, and c ⩾ 0 denotes the coupon rate. What happens when n becomes large?
Exercise 1.10 Consider a zero-coupon bond with prices P(1, 2) = 91.74% and P(0, 2) = 83.40%
at times t = 0 and t = 1.
a) Compute the corresponding yields y0,1 , y0,2 and y1,2 at times t = 0 and t = 1.
b) Assume that $0.1 coupons are paid at times t = 1 and t = 2. Price the corresponding coupon
bond at times t = 0 and t = 1 using the yields y0 and y1 .
Exercise 1.11 Consider the Vasicek process (rt )t∈R+ solution of the equation
drt = (a − brt )dt + σ dBt . (1.4.35)
a) Consider the discretization

√
rtk+1 := rtk + (a − brtk )∆t ± σ ∆t, k = 0, 1, 2, . . .
of the equation (1.4.35) with p(rt0 ) = p(rt1 ) = 1/2 and

√ √
IE[∆rt1 ] = (a − brt0 )∆t + σ p(rt0 ) ∆t − σ q(rt0 ) ∆t = (a − brt0 )∆t
and √ √
IE[∆rt2 ] = (a − brt1 )∆t + σ p(rt0 ) ∆t − σ q(rt0 ) ∆t = (a − brt1 )∆t.
Does this discretization lead to a (recombining) binomial tree?
b) Using the Girsanov Theorem, find a probability measure Q under which the process
(rt /σ )t∈[0,T ] with
drt a − brt
= dt + dBt
σ σ
is a standard Brownian motion.
rt
Hint: By the Girsanov Theorem, the process Xt = X0 + 0 us ds + Bt is a martingale under
the probability measure Q with Radon-Nikodym density
w
1wT

dQ T
2
= exp − ut dBt − (ut ) dt
dP 0 2 0
with respect to P.
c) Prove the Radon-Nikodym derivative approximation
√

1 a − brt
2T /∆T ∏ ± ∆t
0<t<T 2 2σ
w
1 wT

1 T 2
≃ exp (a − brt )drt − 2 (a − brt ) dt . (1.4.36)
σ2 0 2σ 0
d) Using (1.4.36), show that the Vasicek process can be discretized along the binomial tree

√ √
rt2 = rt1 + σ ∆t = rt0 + 2σ ∆t
p (r t 1 )
√
rt1 = rt0 + σ ∆t
)
p(r t 0
q(rt )
rt0 1 rt2 = rt0
p t1)
( r
q(r
t0 )
√
rt1 = rt0 − σ ∆t
q(r
t1 ) √ √
rt2 = rt1 − σ ∆t = rt0 − 2σ ∆t
with
IE[∆rt1 | rt0 ] = (a − brt0 )∆t and IE[∆rt2 | rt1 ] = (a − brt1 )∆t,
where ∆rt1 := rt1 − rt0 , ∆rt2 := rt2 − rt1 , and the probabilities p(rt0 ), q(rt0 ), p(rt1 ), q(rt1 ) will
be computed explicitly.
The use of binomial (or recombining) trees can make the implementation of the Monte Carlo
method easier as their size grows linearly instead of exponentially.
Exercise 1.12 Black-Derman-Toy model (Black, Derman, and Toy, 1990). Consider a one-step
interest rate model
√
in which the short-term
√
interest rate r0 on [0, 1] can turn into two possible values
r1u = r0 e µ∆t +σ ∆t and r1d = r0 e µ∆t−σ ∆t on the time interval [1, 2] with equal probabilities 1/2 at
time ∆t = 1 year and volatility σ = 22% per year, and a zero-coupon bonds with prices P(0, 1)
and P(0, 2) at time t = 0.
a) Write down the value of P(1, 2) using r1u and r1d .
b) Write down the value of P(0, 2) using r1u , r1d and r0 .
c) Estimate the value of r0 from the market price P(0, 1) = 91.74.
d) Estimate the values of r1u and r1d from the market price P(0, 2) = 83.40.
Exercise 1.13 Consider a yield curve ( f (t,t, T ))0⩽t⩽T and a bond paying coupons c1 , c2 , . . . , cn at
times T1 , T2 , . . . , Tn until maturity Tn , and priced as
n
P(t, Tn ) = ∑ ck e −(T −t ) f (t,t,T ) ,
k k
0 ⩽ t ⩽ T1 ,
k =1
where cn is inclusive of the last coupon payment and the nominal $1 value of the bond. Let fe(t,t, Tn )
denote the compounded yield to maturity defined by equating
n
˜
P(t, Tn ) = ∑ ck e −(T −t ) f (t,t,T ) ,
k n
0 ⩽ t ⩽ T1 , (1.4.37)
k =1
i.e. fe(t,t, Tn ) solves the equation

F t, fe(t,t, Tn ) = P(t, Tn ), 0 ⩽ t ⩽ T1 ,

33
with
n
F (t, x) := ∑ ck e −(T −t )x ,
k
0 ⩽ t ⩽ T1 .
k =1
The bond duration D(t, Tn ) is the relative sensitivity of P(t, Tn ) with respect to fe(t,t, Tn ), defined
as
1 ∂F e
D(t, Tn ) := − t, f (t,t, Tn ) , 0 ⩽ t ⩽ T1 .
P(t, Tn ) ∂ x
The bond convexity C (t, Tn ) is defined as
1 ∂ 2F e
C (t, Tn ) := t, f (t,t, Tn ) , 0 ⩽ t ⩽ T1 .
P(t, Tn ) ∂ x2
a) Compute the bond duration in case n = 1.

b) Show that the bond duration D(t, Tn ) can be interpreted as an average of times to maturity
weighted by the respective discounted bond payoffs.
c) Show that the bond convexity C (t, Tn ) satisfies
C (t, Tn ) = (D(t, Tn ))2 + (S(t, Tn ))2 ,
where
n
(S(t, Tn ))2 := ∑ wk (Tk − t − D(t, Tn ))2
k =1
measures the dispersion of the duration of the bond payoffs around the portfolio duration
D(t, Tn ).
d) Consider now the zero-coupon yield defined as

P(t,t + α (Tn − t )) = exp − α (Tn − t ) fα (t,t, Tn ) ,
where α ∈ (0, 1), i.e.
1
fα (t,t, Tn ) := − log P(t,t + α (Tn − t )), 0 ⩽ t ⩽ Tn .
α (Tn − t )
Compute the bond duration associated to the yield fα (t,t, Tn ) in affine bond pricing models
of the form
P(t, T ) = e A(T −t )+rt B(T −t ) , 0 ⩽ t ⩽ T.
e) (Wu, 2000) Compute the bond duration associated to the yield fα (t,t, Tn ) in the Vasicek
model, in which
1 − e −(T −t )b
B(T − t ) : = , 0 ⩽ t ⩽ T.
b

34 Chapter 2. Forward Rate Modeling
2. Forward Rate Modeling
Forward rates are interest rates used in Forward Rate Agreements (FRA) for financial transactions,
such as loans, that can take place at a future date. This chapter deals with the modeling of
forward rates and swap rates in the Heath-Jarrow-Morton (HJM) and Brace-Gatarek-Musiela
(BGM) models. It also includes a presentation of the Nelson and Siegel, 1987 and Svensson, 1994
curve parametrizations for yield curve fitting, and an introduction to two-factor interest rate models.

Exercises 61
2.1 Construction of Forward Rates

A forward interest rate contract (or Forward Rate Agreement, FRA) gives to its holder the possibility
to lock an interest rate denoted by f (t, T , S) at present time t for a loan to be delivered over a future
period of time [T , S], with t ⩽ T ⩽ S.
0 t T S
The rate f (t, T , S) is called a forward interest rate. When T = t, the spot forward rate f (t,t, S)
coincides with the yield, see Relation (2.1.3) below.

Figure 2.1 presents a typical yield curve on the LIBOR (London Interbank Offered Rate) market
with t =07 May 2003.
5 TimeSerieNb 505
Forward interest rate AsOfDate 7mai03
2D 2,55
1W 2,53
4.5 1M 2,56
2M 2,52
3M 2,48
4 1Y 2,34
2Y 2,49
3Y 2,79
4Y 3,07
3.5
%
5Y 3,31
6Y 3,52
7Y 3,71
8Y 3,88
3 9Y 4,02
10Y 4,14
11Y 4,23
2.5 12Y 4,33
13Y 4,4
14Y 4,47
15Y 4,54
2 20Y 4,74
0 5 10 15 20 25 30 25Y 4,83
30Y 4,86
Years
Figure 2.1: Graph of the spot forward rate S 7→ f (t,t, S).
Maturity transformation, i.e., the ability to transform short-term borrowing (debt with short maturi-
ties, such as deposits) into long term lending (credits with very long maturities, such as loans), is
among the roles of banks. Profitability is then dependent on the difference between long rates and
short rates.
Another example of market data is given in the next Figure 2.2, in which the red and blue curves
refer respectively to July 21 and 22 of year 2011.
Figure 2.2: Market example of yield curves, cf. (2.1.3).
Long maturities usually correspond to higher rates as they carry an increased risk. The dip observed
with short maturities can correspond to a lower motivation to lend/invest in the short-term.
Forward rates from bond prices

Let us determine the arbitrage or “fair” value of the forward interest rate f (t, T , S) by implementing
the Forward Rate Agreement using the instruments available in the market, which are bonds priced
at P(t, T ) for various maturity dates T > t.

The loan can be realized using the available instruments (here, bonds) on the market, by proceeding
in two steps:
1) At time t, borrow the amount P(t, S) by issuing (or short selling) one bond with maturity S,
which means refunding $1 at time S.
2) Since the money is only needed at time T , the rational investor will invest the amount P(t, S)
over the period [t, T ] by buying a (possibly fractional) quantity P(t, S)/P(t, T ) of a bond
with maturity T priced P(t, T ) at time t. This will yield the amount
P(t, S)
$1 ×
P(t, T )
at time T > 0.
As a consequence, the investor will actually receive P(t, S)/P(t, T ) at time T , to refund $1 at time
S.
The corresponding forward rate f (t, T , S) is then given by the relation
P(t, S)
exp ((S − T ) f (t, T , S)) = $1, 0 ⩽ t ⩽ T ⩽ S, (2.1.1)
P(t, T )
where we used exponential compounding, which leads to the following definition (2.1.2).
Definition 2.1 The forward rate f (t, T , S) at time t for a loan on [T , S] is given by
log P(t, T ) − log P(t, S)

f (t, T , S) = . (2.1.2)
S−T
The spot forward rate f (t,t, S) coincides with the yield y(t, S), with
log P(t, S)
f (t,t, S) = y(t, S) = − , or P(t, S) = e −(S−t ) f (t,t,S) , (2.1.3)
T −t
0 ⩽ t ⩽ S.
Instantaneous forward rates

Proposition 2.2 The instantaneous forward rate f (t, T ) = f (t, T , T ) is defined by taking the
limit of f (t, T , S) as S ↘ T , and satisfies
1 ∂P
f (t, T ) := lim f (t, T , S) = − (t, T ). (2.1.4)
S↘T P(t, T ) ∂ T
Proof. We have
f (t, T ) : = lim f (t, T , S)

S↘T
log P(t, S) − log P(t, T )
= − lim
S↘T S−T
log P(t, T + ε ) − log P(t, T )
= − lim
ε↘0 ε

∂
= − log P(t, T )
∂T
1 ∂P
= − (t, T ).
P(t, T ) ∂ T
□
The above equation (2.1.4) can be viewed as a differential equation to be solved for log P(t, T )
under the initial condition P(T , T ) = 1, which yields the following proposition.
Proposition 2.3 The bond price P(t, T ) can be recovered from the instantaneous forward rate
f (t, s) as
w
T
P(t, T ) = exp − f (t, s)ds , 0 ⩽ t ⩽ T. (2.1.5)
t
Proof. We check that
log P(t, T ) = log P(t, T ) − log P(t,t )

wT ∂
= log P(t, s)ds
t ∂s
wT
= − f (t, s)ds.
t
□
Proposition 2.3 also shows that
f (t,t,t ) = f (t,t )
∂ wT
= f (t, s)ds|T =t
∂T t
∂
= − log P(t, T )|T =t
∂T
1 ∂P
= − (t, T )|T =t
P(t, T ) |T =t ∂ T
1 ∂ ∗ h − r T rs ds i
= − IE e t Ft
P(T , T ) ∂ T |T =t
h rT i
= IE∗ rT e − t rs ds Ft
|T =t
= IE∗ [rt | Ft ]
= rt , (2.1.6)
i.e. the short rate rt can be recovered from the instantaneous forward rate as
rt = f (t,t ) = lim f (t, T ).

T ↘t
As a consequence of (2.1.1) and (2.1.5) the forward rate f (t, T , S) can be recovered from (2.1.2)
and the instantaneous forward rate f (t, s), as:
f (t, T , S) =
Sw− T wS
1 T
= − f (t, s)ds − f (t, s)ds
S−T t t

1 wS
= f (t, s)ds, 0 ⩽ t ⩽ T < S. (2.1.7)
S−T T
Similarly, as a consequence of (2.1.3) and (2.1.5) we have the next proposition.
Proposition 2.4 The spot forward rate or yield f (t,t, T ) can be written in terms of bond prices
as
log P(t, T ) 1 wT
f (t,t, T ) = − = f (t, s)ds, 0 ⩽ t < T. (2.1.8)
T −t T −t t
Differentiation with respect to T of the above relation shows that the yield f (t,t, T ) and the
instantaneous forward rate f (t, s) are linked by the relation
∂f 1 wT 1
(t,t, T ) = − f (t, s)ds + f (t, T ), 0 ⩽ t < T,
∂T (T − t )2 t T −t
from which it follows that
1 wT ∂f
f (t, T ) = f (t, s)ds + (T − t ) (t,t, T )
T −t t ∂T
∂f
= f (t,t, T ) + (T − t ) (t,t, T ), 0 ⩽ t < T.
∂T
Forward Vašíček, 1977 rates

In this section we consider the Vasicek model, in which the short rate process is the solution (1.1.2)
of (1.1.1) as illustrated in Figure 1.1.
In the Vasicek model, the forward rate is given by
log P(t, S) − log P(t, T )
f (t, T , S) = −
S−T
rt (C (S − t ) −C (T − t )) + A(S − t ) − A(T − t ))
= −
S−T
2
σ − 2ab
= − 2
2b
rt σ 2 − ab σ 2 −2(S−t )b

1 −(S−t )b −(T −t )b −2(T −t )b

− + e −e − 3 e −e ,
S−T b b3 4b
and the spot forward rate, or yield, satisfies
log P(t, T ) rt C (T − t ) + A(T − t )
f (t,t, T ) = − =−
T −t T −t
σ 2 − 2ab rt σ 2 − ab σ2

1 −(T −t )b −2(T −t )b

= − + + 1− e − 3 1− e ,
2b2 T −t b b3 4b
with the mean
IE[ f (t,t, T )]
σ 2 − 2ab IE[rt ] σ 2 − ab σ2

1 −(T −t )b −2(T −t )b

= − + + 1 − e − 1 − e
2b2 T −t b b3 4b3

σ 2 − 2ab σ 2 − ab

1 r0 −bt a −bt
1 − e −(T −t )b

= − 2
+ e + 2 1− e + 3
2b T −t b b b
σ 2
1 − e −2(T −t )b .

− 3
4b (T − t )
In this model, the forward rate t 7→ f (t,t, T ) can be represented as in the following Figure 2.3, with
a = 0.06, b = 0.1, σ = 0.1, r0 = %1 and T = 50.
1
E[f(t,t,T)]
f(t,t,T)
0.8
0.6
f(t,t,T)
0.4
0.2
0
0 5 10 15 20 25 30 35 40 45 50
t
Figure 2.3: Forward rate process t 7→ f (t,t, T ).
We note that the Vasicek forward rate curve t 7→ f (t,t, T ) appears flat for small values of t, i.e.
longer rates are more stable, while shorter rates show higher volatility or risk. Similar features can
be observed in Figure 2.4 for the instantaneous short rate given by
∂
f (t, T ) : = − log P(t, T ) (2.1.9)
∂T
a σ2 2
= rt e −(T −t )b + 1 − e −(T −t )b − 2 1 − e −(T −t )b ,
b 2b
from which the relation limT ↘t f (t, T ) = rt can be easily recovered. We can also evaluate the mean
a σ2 2
IE[ f (t, T )] = IE[rt ] e −(T −t )b +
1 − e −(T −t )b − 2 1 − e −(T −t )b
b 2b
a σ 2 2
= r0 e −bT + 1 − e −bT − 2 1 − e −(T −t )b .

b 2b
The instantaneous forward rate t 7→ f (t, T ) can be represented as in the following Figure 2.4, with
a = 0.06, b = 0.1, σ = 0.1 and r0 = %1.

0.7
E[f(t,T)]
0.6 f(t,T)
0.5
0.4
f(t,T)
0.3
0.2
0.1
0
0 5 10 15 20 25 30 35 40 45 50
t
Figure 2.4: Instantaneous forward rate process t 7→ f (t, T ).
Yield curve data

We refer to Chapter III-12 of Charpentier, 2014 on the package “YieldCurve” Guirreri, 2015 for
the following code and further details on yield curve and interest rate modeling using R.
install.packages("YieldCurve");require(YieldCurve);data(FedYieldCurve)
first(FedYieldCurve,'3 month');last(FedYieldCurve,'3 month')
mat.Fed=c(0.25,0.5,1,2,3,5,7,10);n=50
plot(mat.Fed, FedYieldCurve[n,], type="o",xlab="Maturities structure in years", ylab="Interest rates
values", col = "blue", lwd=3)
title(main=paste("Federal Reserve yield curve observed at",time(FedYieldCurve[n], sep=" ")))
grid()
The next Figure 2.5 is plotted using this code* which is adapted from
https://www.quantmod.com/examples/chartSeries3d/chartSeries3d.alpha.R
15%
14%
13%
12%
11%
10%
9%
8%
7%
6%
5%
Jan Jan 4%
1982 1984 Jan Jan 3%
1986 Jan
1988 Jan 2%
1990 Jan
1992 Jan 1%
1994 Jan
1996 Jan 0%
1998 Jan 10Y
7Y
2000 Jan 5Y
3Y
2002 Jan 2Y
2004 Jan 1Y
6M
2006 Jan 3M
2008 Jan
2010 Jan
2012
2012
Figure 2.5: Federal Reserve yield curves from 1982 to 2012.
European Central Bank (ECB) data can be similarly obtained by the next code.
* Click to open or download.

data(ECBYieldCurve);first(ECBYieldCurve,'3 month');last(ECBYieldCurve,'3 month')

mat.ECB<-c(3/12,0.5,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23, 24,25,26,27,28, 29,30)
dev.new(width=16,height=7)
for (n in 200:400) {
plot(mat.ECB, ECBYieldCurve[n,], type="o",xlab="Maturity structure in years", ylab="Interest rates
values",ylim=c(3.1,5.1),col="blue",lwd=2,cex.axis=1.5,cex.lab=1.5)
title(main=paste("European Central Bank yield curve observed at",time(ECBYieldCurve[n], sep=" ")))
grid();Sys.sleep(0.5)}
The next Figure 2.6 represents the output of the above script.
European Central Bank yield curve observed at 2008−07−23
5.0
4.5
Interest rates values
4.0
3.5
0 5 10 15 20 25 30
Maturity structure in years
Figure 2.6: European Central Bank yield curves.*
Yield curve inversion
Increasing yield curves are typical of economic expansion phases. Decreasing yield curves can
occur when central banks attempt to limit inflation by tightening interest rates, such as in the case
of an economic recession, see here.† In this case, uncertainty triggers increased investment in long
bonds whose rates tend to drop as a consequence, while reluctance to lend in the short term can
lead to higher short rates.
* The animation works in Acrobat Reader on the entire pdf file.

† Right-click to open or save the attachment.

Figure 2.7: August 2019 Federal Reserve yield curve inversion.*

The above Figure 2.7 illustrates a Federal Reserve (FED) yield curve inversions occurring in
February and August 2019.
LIBOR (London Interbank Offered) Rates

Recall that the forward rate f (t, T , S), 0 ⩽ t ⩽ T ⩽ S, is defined using exponential compounding,
from the relation

f (t, T , S) = . (2.1.10)
S−T
In order to compute swaption prices one prefers to use forward rates as defined on the London
InterBank Offered Rates (LIBOR) market instead of the standard forward rates given by (2.1.10).
Other types of LIBOR rates include EURIBOR (European Interbank Offered Rates), HIBOR (Hong
Kong Interbank Offered Rates), SHIBOR (Shanghai Interbank Offered Rates), SIBOR (Singapore
Interbank Offered Rates), TIBOR (Tokyo Interbank Offered Rates), etc. Most LIBOR rates have
been replaced by alternatives such as the Secured Overnight Financing Rate (SOFR) starting with
the end of year 2021, see below, page 43.
The forward LIBOR rate L(t, T , S) for a loan on [T , S] is defined by replacing exponential
compounding with linear compounding in the argument leading to (2.1.1), i.e. by replacing (2.1.10)
with the relation
P(t, T )
1 + (S − T )L(t, T , S) = , t ⩾ T, (2.1.11)
P(t, S)
which yields the following definition.

Definition 2.5 The forward LIBOR rate L(t, T , S) at time t for a loan on [T , S] is given by

1 P(t, T )
L(t, T , S) = −1 , 0 ⩽ t ⩽ T < S. (2.1.12)
S−T P(t, S)

Note that (2.1.12) above yields the same formula for the (LIBOR) instantaneous forward rate
L(t, T ) : = lim L(t, T , S)
S↘T
P(t, T ) − P(t, S)
= lim
S↘T (S − T )P(t, S)
P(t, T ) − P(t, T + ε )
= lim
ε↘0 εP(t, T + ε )
1 P(t, T ) − P(t, T + ε )
= lim
P(t, T ) ε↘0 ε
1 ∂P
= − (t, T )
P(t, T ) ∂ T
∂
= − log P(t, T )
∂T
= f (t, T ),
as in (2.1.4). In addition, Relation (2.1.12) shows that the LIBOR rate can be viewed as a forward
price Xbt = Xt /Nt with numéraire Nt = (S − T )P(t, S) and Xt = P(t, T ) − P(t, S), according to
Relation (3.2.4) of Chapter 3. As a consequence, from Proposition 3.4 we have the following result,
which uses the forward measure P b S defined by its Radon-Nikodym density
dP
bS 1 rS
− 0 rt dt
: = e , (2.1.13)
dP∗ P(0, S)
from the numéraire process Nt := P(t, S), t ∈ [0, S], see Definition 3.1.
Proposition 2.6 The (simply compounded) LIBOR forward rate (L(t, T , S))t∈[0,T ] is a martingale
under P
b S , i.e. we have
b S [L(T , T , S) | Ft ],
L(t, T , S) = IE 0 ⩽ t ⩽ T.
SOFR (Secured Overnight Financing) Rates

The repurchase agreement (“repo”) market is a market where government treasury securities can be
borrowed on the short term. The SOFR rate is a measure of the cost of borrowing which is estimated
using overnight activity on the repo market. In that sense, the SOFR, which is transaction-based,
differs from LIBOR which is relied on a survey of a panel of banks and subject to manipulation.
On the other hand, an important difference is that LIBOR rates are forward-looking using a term
structure, whereas SOFR rates are backward-looking.
rb ra
The next definition uses the integral convention a = − b , a < b.
Definition 2.7 The backward-looking bond price is defined for t ⩾ T as
h rT i h rt i rt
P(t, T ) = IE e − t ru du Ft = IE e T ru du Ft = e T ru du , t ⩾ T.
The forward SOFR rate R(t, T , S) for a loan on [T , S] is defined using linear compounding by the
same absence of arbitrage argument leading to (2.1.11), as
P(t, T )
1 + (S − T )R(t, T , S) = , 0 ⩽ T ⩽ t,
P(t, S)
which yields the following definition.

Definition 2.8 The forward SOFR rate R(t, T , S) at time t ∈ [T , S] for a loan on the time interval
[T , S] is given by

1 P(t, T )
R(t, T , S) = −1 , 0 ⩽ T ⩽ t ⩽ S. (2.1.14)
S−T P(t, S)
In particular, the spot Effective Federal Funds Rate (EFFR) is given for t = S as
1 r S ru du
R(S, T , S) = eT −1 .
S−T
The following proposition, see Rutkowski and Bickersteth, 2021, uses the forward S-measure P
bS
defined by its Radon-Nikodym density (2.1.13).
Proposition 2.9 The SOFR forward rate (R(t, T , S))t∈[T ,S] is a martingale under P
b S , i.e. we have
1 r S ru du

b S [R(S, T , S) | Ft ] = Ib
R(t, T , S) = IE ES eT − 1 Ft ,
S−T
T ⩽ t ⩽ S.
Proof. We have

1 P(t, T )
R(t, T , S) = −1
S−T P(t, S)
rt !
1 e T ru du
= −1
S−T P(t, S)

1 1
= IE[P(t, T ) | Ft ] − 1
S − T P(t, S)
1

1 h rt i
= IE e T r u du
Ft − 1
S − T P(t, S)
1

1 h rS rS i
− t ru du T ru du
= IE e e Ft − 1
S − T P(t, S)
1 b h r S ru du i
= IES e T Ft − 1
S−T
1 b
= IES [P(S, T ) | Ft ] − 1
S−T
= b S [R(S, T , S) | Ft ],
IE T ⩽ t ⩽ S.
2.2 LIBOR and SOFR Swap Rates

The first interest rate swap occurred in 1981 between the World Bank, which was interested in
borrowing German Marks and Swiss Francs, and IBM, which already had large amounts of those
currencies but needed to borrow U.S. dollars.

The vanilla interest rate swap makes it possible to exchange a sequence of variable LIBOR
rates L(t, Tk , Tk+1 ), k = 1, 2, . . . , n − 1, against a fixed rate κ over a succession of time intervals
[Ti , Ti+1 ), . . . , [T j−1 , T j ] defining a tenor structure, see Section 4.1 for details.
Making the agreement fair results into an exchange of cashflows
(Tk+1 − Tk )L(t, Tk , Tk+1 ) − (Tk+1 − Tk )κ ,
| {z } | {z }
floating leg fixed leg
at the dates Ti+1 , . . . , T j between the two parties, therefore generating a cumulative discounted cash
flow
j−1 r Tk+1
∑ e− t rs ds
(Tk+1 − Tk )(L(t, Tk , Tk+1 ) − κ ),
k =i
at time t = T0 , in which we used simple (or linear) interest rate compounding. This corresponds to
a payer swap in which the swap holder receives the floating leg and pays the fixed leg κ, whereas
the holder of a seller swap receives the fixed leg κ and pays the floating leg.
The above cash flow is used to make the contract fair, and it can be priced at time t as
" #
j−1 r Tk+1
IE∗ ∑ (Tk+1 − Tk ) e − t rs ds
(L(t, Tk , Tk+1 ) − κ ) Ft
k =i
j−1 r Tk+1

∗ −
= ∑ (Tk+1 − Tk )(L(t, Tk , Tk+1 ) − κ ) IE e t rs ds
Ft
k =i
j−1
= ∑ (Tk+1 − Tk )P(t, Tk+1 ) L(t, Tk , Tk+1 ) − κ . (2.2.1)
k =i
The swap rate S(t, Ti , T j ) is by definition the value of the rate κ that makes the contract fair by
making the above cash flow C (t ) vanish.
Definition 2.10 The LIBOR swap rate S(t, Ti , T j ) is the value of the break-even rate κ that
makes the contract fair by making the cash flow (2.2.1) vanish, i.e.
j−1
∑ (Tk+1 − Tk )P(t, Tk+1 ) L(t, Tk , Tk+1 ) − κ = 0. (2.2.2)
k =i
The next Proposition 2.11 makes use of the annuity numéraire

" #
j−1 r Tk+1
P(t, Ti , T j ) := IE∗ ∑ (Tk+1 − Tk ) e − t rs ds
Ft (2.2.3)
k =i
j−1 r Tk+1

∗ −
= ∑ (Tk+1 − Tk ) IE e t rs ds
Ft
k =i
j−1
= ∑ (Tk+1 − Tk )P(t, Tk+1 ), 0 ⩽ t ⩽ T2 ,
k =i
which represents the present value at time t of future $1 receipts at times Ti , . . . , T j , weighted by the
lengths Tk+1 − Tk of the time intervals (Tk , Tk+1 ], k = i, . . . , j − 1.
The time intervals (Tk+1 − Tk )k=i,..., j−1 in the definition (2.2.3) of the annuity numéraire can be
replaced by coupon payments (ck+1 )k=i,..., j−1 occurring at times (Tk+1 )k=i,..., j−1 , in which case the
annuity numéraire becomes
" #
j−1 r Tk+1
P(t, Ti , T j ) := IE∗ ∑ ck+1 e − t rs ds
Ft
k =i

j−1 r
T

∗ − t k+1 rs ds
= c
∑ k +1 IE e Ft
k =i
j−1
= ∑ ck+1 P(t, Tk+1 ), 0 ⩽ t ⩽ Ti , (2.2.4)
k =i
which represents the value at time t of the future coupon payments discounted according to the
bond prices (P(t, Tk+1 ))k=i,..., j−1 . This expression can also be used to define amortizing swaps in
which the value of the notional decreases over time, or accreting swaps in which the value of the
notional increases over time.
LIBOR Swap rates
The LIBOR swap rate S(t, Ti , T j ) is defined by solving Relation (2.2.2) for the forward rate
S(t, Tk , Tk+1 ), i.e.
j−1
∑ (Tk+1 − Tk )P(t, Tk+1 ) L(t, Tk , Tk+1 ) − S(t, Ti , T j ) = 0. (2.2.5)
k =i
Proposition 2.11 The LIBOR swap rate S(t, Ti , T j ) is given by
j−1
1
S(t, Ti , T j ) = (Tk+1 − Tk )P(t, Tk+1 )L(t, Tk , Tk+1 ), (2.2.6)
P(t, Ti , T j ) k∑=i
0 ⩽ t ⩽ Ti .
Proof. By definition, S(t, Ti , T j ) is the (fixed) break-even rate over [Ti , T j ] that will be agreed in
exchange for the family of forward rates L(t, Tk , Tk+1 ), k = i, . . . , j − 1, and it solves (2.2.5), i.e. we
have
j−1
∑ (Tk+1 − Tk )P(t, Tk+1 )L(t, Tk , Tk+1 ) − P(t, Ti , Tj )S(t, Ti , Tj )
k =i
j−1
= ∑ (Tk+1 − Tk )P(t, Tk+1 )L(t, Tk , Tk+1 )
k =i
j−1
−S(t, Ti , T j ) ∑ (Tk+1 − Tk )P(t, Tk+1 )
k =i
j−1
= ∑ (Tk+1 − Tk )P(t, Tk+1 )L(t, Tk , Tk+1 ) − S(t, Ti , Tj )P(t, Ti , Tj )
k =i
= 0,
which shows (2.2.6) by solving the above equation for S(t, Ti , T j ) . □
The LIBOR swap rate S(t, Ti , T j ) is defined by the same relation as (2.2.2), with the forward rate
L(t, Tk , Tk+1 ) replaced with the LIBOR rate L(t, Tk , Tk+1 ). In this case, using the Definition 2.1.12
of LIBOR rates we obtain the next corollary.

Corollary 2.12 The LIBOR swap rate S(t, Ti , T j ) is given by
P(t, Ti ) − P(t, T j )
S(t, Ti , T j ) = , 0 ⩽ t ⩽ Ti . (2.2.7)
P(t, Ti , T j )
Proof. By (2.2.6), (2.1.12) and a telescoping summation argument we have

j−1
1
S(t, Ti , T j ) = (Tk+1 − Tk )P(t, Tk+1 )L(t, Tk , Tk+1 )
j−1
1 P(t, Tk )
= P(t, Tk+1 ) −1
P(t, Ti , T j ) k∑=i P(t, Tk+1 )
j−1
1
= (P(t, Tk ) − P(t, Tk+1 ))
P(t, Ti ) − P(t, T j )
= . (2.2.8)
P(t, Ti , T j )
□
By (2.2.7), the bond prices P(t, Ti ) can be recovered from the values of the forward swap rates
S(t, Ti , T j ).
Clearly, a simple expression for the swap rate such as that of Corollary 2.12 cannot be obtained
using the standard (i.e. non-LIBOR) rates defined in (2.1.10). Similarly, it will not be available for
amortizing or accreting swaps because the telescoping summation argument does not apply to the
expression (2.2.4) of the annuity numéraire.
When n = 2, the LIBOR swap rate S(t, T1 , T2 ) coincides with the LIBOR rate L(t, T1 , T2 ), as from
(2.2.4) we have
P(t, T1 ) − P(t, T2 )
S(t, T1 , T2 ) = (2.2.9)
P(t, T1 , T2 )
P(t, T1 ) − P(t, T2 )
=
(T2 − T1 )P(t, T2 )
= L(t, T1 , T2 ).
Similarly to the case of LIBOR rates, Relation (2.2.7) shows that the LIBOR swap rate can be
viewed as a forward price with (annuity) numéraire Nt = P(t, Ti , T j ) and Xt = P(t, Ti ) − P(t, T j ).
Consequently the LIBOR swap rate (S(t, Ti , T j )t∈[T ,S] is a martingale under the forward measure Pb
defined from (3.2.1) by
dPb P(Ti , Ti , T j ) − r Ti rt dt
= e 0 .
dP∗ P(0, Ti , T j )
SOFR Swap rate

The expressions
j−1
1
S(t, Ti , T j ) = (Tk+1 − Tk )P(t, Tk+1 )R(t, Tk , Tk+1 ) (2.2.10)

and
P(t, Ti ) − P(t, T j )
S(t, Ti , T j ) = , Ti ⩽ t ⩽ T j , (2.2.11)
P(t, Ti , T j )
defining the SOFR swap rate S(t, Ti , T j ) are identical to the ones defining the LIBOR swap rate in
(2.2.6) and (2.2.7) by taking t ⩾ Ti in the case of the SOFR swap rate.
2.3 The HJM Model

In this section we turn to the modeling of instantaneous forward rate curves in the HJM Model.
From the beginning of this chapter we have started with the modeling of the short rate (rt )t∈R+ ,
followed by its consequences on the pricing of bonds P(t, T ) and on the expressions of the forward
rates f (t, T , S) and L(t, T , S).
In this section we choose a different starting point and consider the problem of directly modeling
the instantaneous forward rate f (t, T ). The graph given in Figure 2.8 presents a possible random
evolution of a forward interest rate curve using the Musiela convention, i.e. we will write
g(x) = f (t,t + x) = f (t, T ), (2.3.1)
under the substitution x = T − t, x ⩾ 0, and represent a sample of the instantaneous forward curve
x 7→ f (t,t + x) for each t ⩾ 0.
Forward rate
5
4.5
4
3.5
3 20
2.5
2 15
1.5
1 10
0.5 t
0
5 5
10
15 0
x 20
Figure 2.8: Stochastic process of forward curves.
Definition 2.13 In the Heath-Jarrow-Morton (HJM) model, the instantaneous forward rate
f (t, T ) is modeled under P∗ by a stochastic differential equation of the form
dt f (t, T ) = α (t, T )dt + σ (t, T )dBt , 0 ⩽ t ⩽ T, (2.3.2)
where t 7→ α (t, T ) and t 7→ σ (t, T ), 0 ⩽ t ⩽ T , are allowed to be random, (Ft )t∈[0,T ] -adapted,
processes.
In the above equation, the date T is fixed and the differential dt is with respect to the time variable t.
Under basic Markovianity assumptions, a HJM model with deterministic coefficients α (t, T )
and σ (t, T ) will yield a short rate process (rt )t∈R+ of the form
drt = (a(t ) − b(t )rt )dt + σ (t )dBt ,

see § 7.4 in Privault, 2021, which is the Hull and White, 1990 model, with the explicit solution
rt w t rt wt rt
rt = rs e − s b(τ )dτ + e − u b(τ )dτ a(u)du + σ (u) e − u b(τ )dτ dBu ,
s s
0 ⩽ s ⩽ t.
The HJM condition
How to “encode” absence of arbitrage in the defining HJM Equation (2.3.2) is an important question.
Recall that under absence of arbitrage, the bond price P(t, T ) has been constructed as
w w
T T
∗
P(t, T ) = IE exp − rs ds Ft = exp − f (t, s)ds , (2.3.3)
t t
cf. Proposition 2.3, hence the discounted bond price process is given by
wt w
t wT
t 7→ exp − rs ds P(t, T ) = exp − rs ds − f (t, s)ds (2.3.4)
0 0 t
is a martingale under P∗ by Proposition 1.1 and Relation (2.1.5) in Proposition 2.3. This shows
that P∗ is a risk-neutral probability measure, and by the first fundamental theorem of asset pricing
we conclude that the market is without arbitrage opportunities.
Proposition 2.14 (HJM Condition, Heath, Jarrow, and Morton, 1992). Under the condition
wT
α (t, T ) = σ (t, T ) σ (t, s)ds, 0 ⩽ t ⩽ T, (2.3.5)
t
which is known as the HJM absence of arbitrage condition, the discounted bond price process
(2.3.4) is a martingale, and the probability measure P∗ is risk-neutral.
Proof. Using the process (Xt )t∈[0,T ] defined as

wT
Xt := f (t, s)ds = − log P(t, T ), 0 ⩽ t ⩽ T,
t
such that P(t, T ) = e −Xt , we rewrite the spot forward rate, or yield
1 wT
f (t,t, T ) = f (t, s)ds,
T −t t
see (2.1.8), as
1 wT Xt
f (t,t, T ) = f (t, s)ds = , 0 ⩽ t ⩽ T,
T −t t T −t
where the dynamics of t 7→ f (t, s) is given by (2.3.2). We also use the extended Leibniz integral
rule wT wT wT
dt f (t, s)ds = − f (t,t )dt + dt f (t, s)ds = −rt dt + dt f (t, s)ds,
t t t
see (2.1.6). This identity can be checked in the particular case where f (t, s) = g(t )h(s) is a smooth
function that satisfies the separation of variables property, as
w wT
T
dt g(t )h(s)ds = dt g(t ) h(s)ds
t t

wT wT
= h(s)dsdg(t ) + g(t )dt h(s)ds
t t
w
T
= g′ (t ) h(s)ds dt − g(t )h(t )dt.
t
We have
wT
dt Xt = dt f (t, s)ds
t
wT
= − f (t,t )dt + dt f (t, s)ds
t
wT wT
= − f (t,t )dt + α (t, s)dsdt + σ (t, s)dsdBt
w t tw
T T
= −rt dt + α (t, s)ds dt + σ (t, s)ds dBt ,
t t
hence w 2
T
2
|dt Xt | = σ (t, s)ds dt.
t
By Itô’s calculus, we find
dt P(t, T ) = dt e −Xt
1
= − e −Xt dt Xt + e −Xt (dt Xt )2
2
w 2
1 T
= − e −Xt dt Xt + e −Xt σ (t, s)ds dt
2 t
wT wT
−Xt
= −e −rt dt + α (t, s)dsdt + σ (t, s)dsdBt
t t
1 −Xt w T
2
+ e σ (t, s)ds dt,
2 t
and the discounted bond price satisfies

wt
dt exp − rs ds P(t, T )
0
wt wt
= −rt exp − rs ds − Xt dt + exp − rs ds dt P(t, T )
w0t w 0t
= −rt exp − rs ds − Xt dt − exp − rs ds − Xt dt Xt
0 0
1 w t w
T
2
+ exp − rs ds − Xt σ (t, s)ds dt
2 0 t
wt
= −rt exp − rs ds − Xt dt
0
wt wT wT
− exp − rs ds − Xt −rt dt + α (t, s)dsdt + σ (t, s)dsdBt
0 t t
1 wt w T 2
+ exp − rs ds − Xt σ (t, s)ds dt
2 0 t
wt w T
= − exp − rs ds − Xt σ (t, s)dsdBt
0 t
wt wT 1 wT
2 !
− exp − rs ds − Xt α (t, s)ds − σ (t, s)ds dt.
0 t 2 t

Thus, the discounted bond price process

wt
t 7→ exp − rs ds P(t, T )
0
will be a martingale provided that
wT 1 wT
2
α (t, s)ds − σ (t, s)ds = 0, 0 ⩽ t ⩽ T. (2.3.6)
t 2 t
Differentiating the above relation with respect to T yields

wT
α (t, T ) = σ (t, T ) σ (t, s)ds,
t
which is in fact equivalent to (2.3.6). □
Forward Vasicek rates in the HJM model

The HJM coefficients in the Vasicek model are in fact deterministic, for example, taking a = 0, by
(2.1.9) we have
wT
dt f (t, T ) = σ 2 e −(T −t )b e (t−s)b dsdt + σ e −(T −t )b dBt ,
t
i.e.
wT 1 − e −(T −t )b
α (t, T ) = σ 2 e −(T −t )b e (t−s)b ds = σ 2 e −(T −t )b ,
t b
and σ (t, T ) = σ e −(T −t )b , and the HJM condition reads
wT wT
α (t, T ) = σ 2 e −(T −t )b e (t−s)b ds = σ (t, T ) σ (t, s)ds. (2.3.7)
t t
Random simulations of the Vasicek instantaneous forward rates are provided in Figures 2.9 and
2.10 using the Musiela convention (2.3.1).
7
6
rate %
5
4
3
2 40
1 30
0
0 20
5 t
10 10
15
20
x 25
30 0
Figure 2.9: Forward instantaneous curve (t, x) 7→ f (t,t + x) in the Vasicek model.*

5
rate (%)
0 5 10 15 20 25 30
x
Figure 2.10: Forward instantaneous curve x 7→ f (0, x) in the Vasicek model.*
For x = 0 the first “slice” of this surface is actually the short rate Vasicek process rt = f (t,t ) =
f (t,t + 0) which is represented in Figure 2.11 using another discretization.
6.5
5.5
5
rate (%)
4.5
3.5
3
0 1 2 3 4 5
t
Figure 2.11: Short-term interest rate curve t 7→ rt in the Vasicek model.
HJM-SOFR Model
In the HJM-SOFR model, the instantaneous forward rate f (t, T ) is extended to t > T by taking
dt f (t, T ) = 1[0,T ] (t )α (t, T )dt + 1[0,T ] (t )σ (t, T )dBt , t ⩾ T,
i.e.
f (t, T ) = f (T , T ) = rT , t ⩾ T,
see Lyashenko and Mercurio, 2020.
2.4 Yield Curve Modeling

Nelson-Siegel parametrization of instantaneous forward rates
In the Nelson and Siegel, 1987 parametrization the instantaneous forward rate curves are parametrized
by 4 coefficients z1 , z2 , z3 , z4 , as
g(x) = z1 + (z2 + z3 x) e −xz4 , x ⩾ 0.


An example of a graph obtained by the Nelson-Siegel parametrization is given in Figure 2.12, for
z1 = 1, z2 = −10, z3 = 100, z4 = 10.
4
z1+(z2+xz3)exp(-xz4)
-2
%
-4
-6
-8
-10
0 0.2 0.4 0.6 0.8 1
x
Figure 2.12: Graph of x 7→ g(x) in the Nelson-Siegel model.
Svensson parametrization of instantaneous forward rates
The Svensson, 1994 parametrization has the advantage to reproduce two humps instead of one, the
location and height of which can be chosen via 6 parameters z1 , z2 , z3 , z4 , z5 , z6 as
g(x) = z1 + (z2 + z3 x) e −xz4 + z5 x e −xz6 , x ⩾ 0.
A typical graph of a Svensson parametrization is given in Figure 2.13, for z1 = 6.6, z2 = −5,
z3 = −100, z4 = 10, z5 = −1/2, z6 = 1.
7
z1+(z2+z3x)exp(-xz4)+z5xexp(-z6x)
4
%
3
0
0 0.2 0.4 0.6 0.8 1
x
Figure 2.13: Graph of x 7→ g(x) in the Svensson model.
Figure 2.14 presents a fit of the market data of Figure 2.1 using a Svensson curve.

4.5
% 3.5
2.5
Market data
Svensson curve
2
0 5 10 15 20 25 30
years
Figure 2.14: Fitting of a Svensson curve to market data.
The attached IPython notebook can be run here or here to fit a Svensson curve to market data.
Vasicek parametrization
In the Vasicek model, the instantaneous forward rate process is given from (2.1.9) and (2.3.1) as
a σ2 a σ2 σ2

f (t, T ) = − 2 + rt − + 2 e −bx − 2 e −2bx , (2.4.1)
b 2b b b 2b
in the Musiela notation (x = T − t), and we have
σ2

∂f
(t, T ) = a − brt − (1 − e −(T −t )b ) e −(T −t )b .
∂T b
We check that the derivative ∂ f /∂ T vanishes when a − brt + a − σ 2 (1 − e −bx )/b = 0, i.e.
b
e −bx = 1 + (brt − a),
σ2
which admits at most one solution, provided that a > brt . As a consequence, the possible forward
curves in the Vasicek model are limited to one change of “regime” per curve, as illustrated in
Figure 2.15 for various values of rt , and in Figure 2.16.
0.09
0.08
0.07
0.06
0.05
0.04
0.03
0.02
0.01
0
0 5 10 15 20
Time to maturity
Figure 2.15: Graphs of forward rates with b = 0.16, a/b = 0.04, r0 = 2%, σ = 4.5%.
The next Figure 2.16 is also using the parameters b = 0.16, a/b = 0.04, r0 = 2%, and σ = 4.5%.

0.08
0.06
0.04
0.02
0
0 20
2 15
4
x 6 10
5 t
8
10 0
Figure 2.16: Forward instantaneous curve (t, x) 7→ f (t,t + x) in the Vasicek model.
One may think of constructing an instantaneous forward rate process taking values in the Svensson
space, however this type of modeling is not consistent with absence of arbitrage, and it can be
proved that the HJM curves cannot live in the Nelson-Siegel or Svensson spaces, see §3.5 of Björk,
2004. In other words, it can be shown that the forward yield curves produced by the Vasicek model
are included neither in the Nelson-Siegel space, nor in the Svensson space. In addition, the Vasicek
yield curves do not appear to correctly model the market forward curves cf. also Figure 2.1 above.
Another way to deal with the curve fitting problem is to use deterministic shifts for the fitting of
one forward curve, such as the initial curve at t = 0, cf. e.g. § 6.3 in Privault, 2021.
Fitting the Nelson-Siegel and Svensson models to yield curve data

Recall that in the Nelson-Siegel parametrization the instantaneous forward rate curves are parametrized
by four coefficients z1 , z2 , z3 , z4 , as
f (t,t + x) = z1 + (z2 + z3 x) e −xz4 , x ⩾ 0. (2.4.2)
Taking x = T − t, the yield f (t,t, T ) is given as

1 wT
f (t,t, T ) = f (t, s)ds
T −t t
1wx
= f (t,t + y)dy
x 0 w
z2 x −yz4 z3 w x −yz4
= z1 + e dy + ye dy
x 0 x 0
1 − e −xz4 1 − e −xz4 + x e −xz4
= z1 + z2 + z3 .
xz4 xz4
The yield f (t,t, T ) can be reparametrized as
β2 −x/λ
f (t,t + x) = z1 + (z2 + z3 x) e −xz4 = β0 + β1 e −x/λ + xe , x ⩾ 0,
λ
see Charpentier, 2014, with β0 = z1 , β1 = z2 , β2 = z3 /z4 , λ = 1/z4 .
require(YieldCurve);data(ECBYieldCurve)
mat.ECB<-c(3/12,0.5,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23, 24,25,26,27,28,29,30)
first(ECBYieldCurve, '1 month');Nelson.Siegel(first(ECBYieldCurve, '1 month'), mat.ECB)

for (n in seq(from=70, to=290, by=10)) {

ECB.NS <- Nelson.Siegel(ECBYieldCurve[n,], mat.ECB)
ECB.S <- Svensson(ECBYieldCurve[n,], mat.ECB)
ECB.NS.yield.curve <- NSrates(ECB.NS, mat.ECB)
ECB.S.yield.curve <- Srates(ECB.S, mat.ECB,"Spot")
plot(mat.ECB, as.numeric(ECBYieldCurve[n,]), type="o", lty=1, col=1,ylab="Interest rates",
xlab="Maturity in years", ylim=c(3.2,4.8),cex.lab=1.6,cex.axis=1.6)
lines(mat.ECB, as.numeric(ECB.NS.yield.curve), type="l", lty=3,col=2,lwd=2)
lines(mat.ECB, as.numeric(ECB.S.yield.curve), type="l", lty=2,col=6,lwd=2)
title(main=paste("ECB yield curve observed at",time(ECBYieldCurve[n], sep=" "),"vs fitted yield curve"))
legend('bottomright', legend=c("ECB data","Nelson-Siegel","Svensson"),col=c(1,2,6), lty=1, bg='gray90')
grid();}
ECB yield curve observed at 2008−02−17 vs fitted yield curve
● ● ●
● ● ●
● ●
● ●
● ●
●
4.5
●
●
●
●
●
●
Interest rates
●
●
●
4.0
●
●
●
● ●
●
3.5
●
●
ECB data
● ● Nelson−Siegel
Svensson
0 5 10 15 20 25 30
Maturity in years
Figure 2.17: ECB data vs. fitted yield curve.*
2.5 Two-Factor Model
The correlation problem is another issue of concern when using the affine models considered so
far, see (1.1.8) and (1.4.10). Let us compare three bond price simulations with maturity T1 = 10,
T2 = 20, and T3 = 30 based on the same Brownian path, as given in Figure 2.18. Clearly, the bond
prices
F (rt , Ti ) = P(t, Ti ) = e A(t,Ti )rt C(t,Ti ) , 0 ⩽ t ⩽ Ti , i = 1, 2,
with maturities T1 and T2 are linked by the relation

P(t, T2 ) = P(t, T1 ) exp A(t, T2 ) − A(t, T1 ) + rt (C (t, T2 ) −C (t, T1 )) , (2.5.1)
meaning that bond prices with different maturities could be deduced from each other, which is
unrealistic.

1
0.9
0.8
0.7
0.6
0.5
0.4
P(t,T1)
P(t,T2)
P(t,T3)
0.3
0 5 10 15 20 25 30
t
Figure 2.18: Graph of t 7→ P(t, T1 ), P(t, T2 ), P(t, T3 ).
In affine short rate models, by (2.5.1), log P(t, T1 ) and log P(t, T2 ) are linked by the affine relation-
ship
log P(t, T2 ) = log P(t, T1 ) + A(t, T2 ) − A(t, T1 ) + rt (C (t, T2 ) −C (t, T1 ))

log P(t, T1 ) − A(t, T1 )
= log P(t, T1 ) + A(t, T2 ) − A(t, T1 ) + (C (t, T2 ) −C (t, T1 ))
C (t, T1 )

C (t, T2 ) −C (t, T1 ) C (t, T2 )
= 1+ log P(t, T1 ) + A(t, T2 ) − A(t, T1 )
A(t, T1 ) C (t, T1 )
with constant coefficients, which yields the perfect correlation or anticorrelation
Cor(log P(t, T1 ), log P(t, T2 )) = ±1,
depending on the sign of the coefficient 1 + (C (t, T2 ) − C (t, T1 ))/A(t, T1 ), cf. § 6.4 in Privault,
2021,
A solution to the correlation problem is to consider a two-factor model based on two state
processes (Xt )t∈R+ , (Yt )t∈R+ which are solution of

(1)
 dXt = µ1 (t, Xt )dt + σ1 (t, Xt )dBt ,

(2.5.2)
 (2)
dYt = µ2 (t,Yt )dt + σ2 (t,Yt )dBt ,

(1) (2)
where Bt t∈R+
, Bt t∈R+
are correlated Brownian motion, with
(1) (2)
Cov Bs , Bt = ρ min(s,t ), s,t ⩾ 0, (2.5.3)
and
(1) (2)
dBt • dBt = ρdt, (2.5.4)
(1) (2)
for some correlation parameter ρ ∈ [−1, 1]. In practice, Bt t∈R and Bt t∈R can be con-
+ +
(1) (2)
structed from two independent Brownian motions Wt t∈R and Wt t∈R , by letting
+ +

(1) (1)
 Bt = Wt ,

 (2) (1) p (2)

Bt = ρWt + 1 − ρ 2Wt , t ⩾ 0,


and Relations (2.5.3) and (2.5.4) are easily satisfied from this construction.
In two-factor models one chooses to build the short-term interest rate rt via
rt := Xt + Yt , t ⩾ 0.
By the previous standard arbitrage arguments we define the price of a bond with maturity T as
w
T
P(t, T ) : = IE∗ exp − rs ds Ft
t
w
T
∗
= IE exp − rs ds Xt , Yt
t
w
T
= IE∗ exp − (Xs + Ys )ds Xt , Yt
t
= F (t, Xt ,Yt ), (2.5.5)

since the couple (Xt ,Yt )t∈R+ is Markovian. Applying the Itô formula with two variables to
w
T
∗
t 7→ F (t, Xt ,Yt ) = P(t, T ) = IE exp − rs ds Ft ,
t
and using the fact that the discounted process

rt
w
T
− ∗
t 7→ e 0 rs ds P(t, T ) = IE exp − rs ds Ft
0
is an Ft -martingale under P∗ , we can derive the PDE

∂F ∂F
−(x + y)F (t, x, y) + µ1 (t, x) (t, x, y) + µ2 (t, y) (t, x, y)
∂x ∂y
1 2 ∂ 2F 1 2 ∂ 2F
+ σ1 (t, x) 2 (t, x, y) + σ2 (t, y) 2 (t, x, y)
2 ∂x 2 ∂y
2
∂ F ∂F
+ρσ1 (t, x)σ2 (t, y) (t, x, y) + (t, x, y) = 0, (2.5.6)
∂ x∂ y ∂t
on R2 for the bond price P(t, T ). In the Vasicek model

(1)
 dXt = −aXt dt + σ dBt ,

 (2)
dYt = −bYt dt + ηdBt ,

this yields the solution F (t, x, y) of (2.5.6) as
P(t, T ) = F (t, Xt ,Yt ) = F1 (t, Xt )F2 (t,Yt ) exp (ρU (t, T )) , (2.5.7)
where F1 (t, Xt ) and F2 (t,Yt ) are the bond prices associated to Xt and Yt in the Vasicek model, and
!
ση e −(T −t )a − 1 e −(T −t )b − 1 e −(a+b)(T −t ) − 1
U (t, T ) := T −t + + −
ab a b a+b
(1) (2)
is a correlation term which vanishes when Bt t∈R and Bt t∈R are independent, i.e. when
+ +
ρ = 0, cf. Ch. 4, Appendix A in Brigo and Mercurio, 2006, § 6.5 of Privault, 2021.

Partial differentiation of log P(t, T ) with respect to T leads to the instantaneous forward rate
ση
1 − e −(T −t )a 1 − e −(T −t )b ,

f (t, T ) = f1 (t, T ) + f2 (t, T ) − ρ (2.5.8)
ab
where f1 (t, T ), f2 (t, T ) are the instantaneous forward rates corresponding to Xt and Yt respectively,
cf. § 6.5 of Privault, 2021.
An example of a forward rate curve obtained in this way is given in Figure 2.19.
2.4
2.3
2.2
2.1
%
1.9
1.8
0 5 10 15 20 25 30 35 40
T
Figure 2.19: Graph of forward rates in a two-factor model.
Next, in Figure 2.20 we present a graph of the evolution of forward curves in a two-factor model.
0.24
0.235
0.23
0.225
0.22
0.215
1.4
1.2
1
0.8 8
7
t 0.6 6
5
0.4 4
0.2 3
2
1 x
0 0
Figure 2.20: Random evolution of instantaneous forward rates in a two-factor model.
2.6 The BGM Model

The models (HJM, affine, etc.) considered in the previous chapter suffer from various drawbacks
such as nonpositivity of interest rates in Vasicek model, and lack of closed-form solutions in more
complex models. The Brace, Gatarek, and Musiela, 1997 (BGM) model has the advantage of
yielding positive interest rates, and to permit to derive explicit formulas for the computation of
prices for interest rate derivatives such as interest rate caps and swaptions on the LIBOR market.
In the BGM model we consider two bond prices P(t, T1 ), P(t, T2 ) with maturities T1 , T2 , and the
forward probability measure P
b 2 defined as
r T2
dP
b2 e − 0 rs ds
= ,
dP∗ P(0, T2 )

with numéraire P(t, T2 ), cf. (3.2.10). The forward LIBOR rate L(t, T1 , T2 ) is modeled as a driftless
geometric Brownian motion under P b 2 , i.e.
dL(t, T1 , T2 )
= γ1 (t )dBt , (2.6.1)
L(t, T1 , T2 )
0 ⩽ t ⩽ T1 , for some deterministic volatility function of time γ1 (t ), with solution
w
1wu

u
L(u, T1 , T2 ) = L(t, T1 , T2 ) exp γ1 (s)dBs − |γ1 |2 (s)ds ,
t 2 t
i.e. for u = T1 ,
w
1 w T1

T1
2
L(T1 , T1 , T2 ) = L(t, T1 , T2 ) exp γ1 (s)dBs − |γ1 | (s)ds .
t 2 t
Since L(t, T1 , T2 ) is a geometric Brownian motion under P

b 2 , standard caplets can be priced at time
t ∈ [0, T1 ] from the Black-Scholes formula.
In Table 2.1 we summarize some stochastic models used for interest rates.
Model
Short rate rt Mean reverting SDEs
Instantaneous forward rate f (t, s) HJM model
Forward rate f (t, T , S) BGM model
Table 2.1: Stochastic interest rate models.
The following Graph 2.21 summarizes the notions introduced in this chapter.

Bondh price 2
rT i Bond price
P (t, T ) = IE∗ e − t rs ds | Ft P (t, T ) = e −(T −t)f (t,t,T )
Swap rate
S(t, T1 , Tn ) = P (t,T 1 )−P (t,Tn )
P (t,T1 ,Tn )
Short rate1 rt
LIBOR rate3
L(t, T, S) = P(S−T
(t,T )−P (t,S)
)P (t,S)
Forward rate3
Bond price
rT f (t, T, S) = log P (t,TS−T
)−log P (t,S)
P (t, T ) = e − t f (t,s)ds
Instantaneous forward rate4

f (t, T ) = L(t, T ) = − ∂ log∂T
P (t,T )
Short rate Spot forward rate (yield)

rt = f (t, t) = f (t, t, t) rT
f (t, t, T ) = t f (t, s)ds/(T − t)
Instantaneous forward rate4

f (t, T ) = L(t, T ) = limS&T f (t, T, S)
= limS&T L(t, T, S)
1
Can be modeled by Vasiçek and other short rate models
2
Can be modeled from dP (t, T )/P (t, T ).
3
Can be modeled in the BGM model
4
Can be modeled in the HJM model
Figure 2.21: Roadmap of stochastic interest rate modeling.
Exercises
h rT i
Exercise 2.1 We consider a bond with maturity T , priced P(t, T ) = IE∗ e − t rs ds Ft at time
t ∈ [0, T ].
a) Using the forward measure P b with numéraire Nt = P(t, T ), apply the change of numéraire
∂P
formula (3.2.9) to compute the derivative (t, T ).
∂T
b) Using Relation (2.1.5), find an expression of the instantaneous forward rate f (t, T ) using the
short rate rT and the forward expectation Ib
E.
c) Show that the instantaneous forward rate ( f (t, T ))t∈[0,T ] is a martingale under the forward
measure P.b
Exercise 2.2 Consider a tenor structure {T1 , T2 } and a bond with maturity T2 and price given at
time t ∈ [0, T2 ] by
w
T2
P(t, T2 ) = exp − f (t, s)ds , t ∈ [0, T2 ],
t

where the instantaneous yield curve f (t, s) is parametrized as
f (t, s) = r1 1[0,T1 ] (s) + r2 1[T1 ,T2 ] (s), t ⩽ s ⩽ T2 .
Find a formula to estimate the values of r1 and r2 from the data of P(0, T2 ) and P(T1 , T2 ).
Same question when f (t, s) is parametrized as
f (t, s) = r1 s1[0,T1 ] (s) + (r1 T1 + (s − T1 )r2 )1[T1 ,T2 ] (s), t ⩽ s ⩽ T2 .
Exercise 2.3 Consider a short rate process (rt )t∈R+ of the form rt = h(t ) + Xt , where h(t ) is a
deterministic function of time and (Xt )R+ is a Vasicek process started at X0 = 0.
a) Compute the price P(0, T ) at time t = 0 of a bond with maturity T , using h(t ) and the
function A(T ) defined in (1.4.17) for the pricing of Vasicek bonds.
b) Show how the function h(t ) can be estimated from the market data of the initial instantaneous
forward rate curve f (0,t ).
Exercise 2.4
a) Given two LIBOR spot rates R(t,t, T ) and R(t,t, S), express the LIBOR forward rate
R(t, T , S) in terms of R(t,t, T ) and R(t,t, S).
b) Assuming t = 0, T = 1 year, S = 2 years, R(0, 0, T ) = 2%, R(0, 0, S) = 2.5% would you
sign a LIBOR forward rate agreement at t = 0 with rate R(0, T , S) over [T , S] if you believe
that R(T , T , S) will remain at the level R(T , T , S) = R(0, 0, T ) = 2%?
Exercise 2.5 (Exercise 1.4 continued).

a) Compute the forward rate f (t, T , S) in the Ho-Lee model (1.4.31) with constant deterministic
volatility.
In the next questions we take a = 0.
b) Compute the instantaneous forward rate f (t, T ) in this model.
c) Derive the stochastic equation satisfied by the instantaneous forward rate f (t, T ).
d) Check that the HJM absence of arbitrage condition is satisfied in this equation.
Exercise 2.6 Consider the two-factor Vasicek model


(1)
 dXt = −bXt dt + σ dBt ,

 (2)
dYt = −bYt dt + σ dBt ,

(1) (2) (1) (2)

where Bt t∈R , Bt t∈R are correlated Brownian motion such that dBt • dBt = ρdt, for
+ +
ρ ∈ [−1, 1].
a) Write down the expressions of the short rates Xt and Yt .
Hint: They can be found in Section 1.1.
b) Compute the variances Var[Xt ], Var[Yt ], and the covariance Cov(Xt ,Yt ).
Hint: The expressions of Var[Xt ] and Var[Yt ] can be found in Section 1.1.
c) Compute the covariance Cov(log P(t, T1 ), log P(t, T2 )) for the two-factor bond prices
P(t, T1 ) = F1 (t, Xt , T1 )F2 (t,Yt , T1 ) e ρU (t,T1 )
and
P(t, T2 ) = F1 (t, Xt , T2 )F2 (t,Yt , T2 ) e ρU (t,T2 ) ,

63
where
log F1 (t, x, T ) = C1T + xAT1 and log F2 (t, x) = C2T + xAT2 .
Hint: We have Cov(X + Y , Z ) = Cov(X, Z ) + Cov(Y , Z ) and Cov(c, X ) = 0 when c is a
constant.
Exercise 2.7 Stochastic string model (Santa-Clara and Sornette, 2001). Consider an instantaneous
forward rate f (t, x) solution of
dt f (t, x) = αx2 dt + σ dt B(t, x), (2.6.2)
with a flat initial curve f (0, x) = r, where x represents the time to maturity, and (B(t, x))(t,x)∈R2+ is
a standard Brownian sheet with covariance
IE[B(s, x)B(t, y)] = (min(s,t ))(min(x, y)), s,t, x, y ⩾ 0, (2.6.3)
and initial conditions B(t, 0) = B(0, x) = 0 for all t, x ⩾ 0.

a) Solve the equation (2.6.2) for f (t, x).
b) Compute the short-term interest rate rt = f (t, 0).
c) Compute the value at time t ∈ [0, T ] of the bond price
w
T −t
P(t, T ) = exp − f (t, x)dx
0
with maturity T .
w 2
T −t
d) Compute the variance IE B(t, x)dx of the centered Gaussian random variable
0
r T −t
0 B(t, x)dx.
e) Compute the expected value IE∗ [P(t, T )].
f) Find the value of α such thatthe discounted bond price
α w T −t
−rt 3
e P(t, T ) = exp −rT − t (T − t ) − σ B(t, x)dx , t ∈ [0, T ].
3 0
satisfies IE∗ [P(t, T )] = e −(T −t )r . w
T
∗ +
g) Compute the bond option price IE exp − rs ds (P(T , S) − K ) by the Black-Scholes
0
formula, knowing that for any centered Gaussian random variable X ≃ N (0, v2 ) with
variance v2 we have
IE[(x e m+X − K )+ ]
2 /2
= x e m+v Φ(v + (m + log(x/K ))/v) − KΦ((m + log(x/K ))/v).

64 Chapter 3. Change of Numéraire and Forward Measures
3. Change of Numéraire and Forward Measures
Change of numéraire is a powerful technique for the pricing of options under random discount
factors by the use of forward measures. It has applications to the pricing of interest rate derivatives
and other types of options, including exchange options (Margrabe formula) and foreign exchange
options (Garman-Kohlagen formula). The computation of self-financing hedging strategies by
change of numéraire is treated in Section 3.5, and the change of numéraire technique will be applied
to the pricing of interest rate derivatives such as bond options and swaptions in Chapter 4.

Exercises 88
3.1 Notion of Numéraire

A numéraire is any strictly positive (Ft )t∈R+ -adapted stochastic process (Nt )t∈R+ that can be
taken as a unit of reference when pricing an asset or a claim.
In general, the price St of an asset, when quoted in terms of the numéraire Nt , is given by
St
Sbt := , t ⩾ 0.
Nt
Deterministic numéraire transformations are easy to handle, as change of numéraire by a constant
factor is a formal algebraic transformation that does not involve any risk. This can be the case for

example when a currency is pegged to another currency,* for example the exchange rate of the
French franc to the Euro was locked at €1 = FRF 6.55957 on 31 December 1998.
On the other hand, a random numéraire may involve new risks, and can allow for arbitrage
opportunities.
Examples of numéraire processes (Nt )t∈R+ include:

- Money market account.
Given (rt )t∈R+ a possibly random, time-dependent and (Ft )t∈R+ -adapted risk-free interest
rate process, let†
w t
Nt := exp rs ds .
0
In this case,
St rt
Sbt = = e − 0 rs ds St , t ⩾ 0,
Nt
represents the discounted price of the asset at time 0.
- Currency exchange rates
In this case, Nt := Rt denotes e.g. the EUR/SGD (EURSGD=X) exchange rate from a foreign
currency (e.g. EUR) to a domestic currency (e.g. SGD), i.e. one unit of foreign currency
(EUR) corresponds to Rt units of local currency (SGD). Let
St
Sbt := , t ⩾ 0,
Rt
denote the price of a local (SG) asset quoted in units of the foreign currency (EUR). For
example, if Rt = 1.63 and St = S$1, then
St 1
Sbt = = × $1 ≃ €0.61,
Rt 1.63
and 1/Rt is the domestic SGD/EUR exchange rate. A question of interest is whether a local
asset $St , discounted according to a foreign risk-free rate r f and priced in foreign currency as
ft St f
e −r = e −r t Sbt ,
Rt
can be a martingale on the foreign market.
* Major currencies have started floating against each other since 1973, following the end of the system of fixed
exchanged rates agreed upon at the Bretton Woods Conference, July 1-22, 1944.
†“Anyone who believes exponential growth can go on forever in a finite world is either a madman or an economist”,
K.E. Boulding, 1973, page 248.

My foreign currency account St grew by 5%

My foreign currency account St grew by 5%
this year.
this year.
The foreign exchange rate dropped by 10%.
Q: Did I achieve a positive return?
Q: Did I achieve a positive return?
A:
A:
(a) Scenario A. (b) Scenario B.
Figure 3.1: Why change of numéraire?
- Forward numéraire.
The price P(t, T ) of a bond paying P(T , T ) = $1 at maturity T can be taken as numéraire.
In this case, we have
h rT i
Nt := P(t, T ) = IE∗ e − t rs ds Ft , 0 ⩽ t ⩽ T.
Recall that rt h rT i
t 7→ e − 0 rs ds P(t, T ) = IE∗ e − 0 rs ds Ft , 0 ⩽ t ⩽ T,
is an Ft - martingale.
- Annuity numéraire.
Processes of the form

n
Nt = P(t, T0 , Tn ) := ∑ (Tk − Tk−1 )P(t, Tk ), 0 ⩽ t ⩽ T0 ,
k =1
where P(t, T1 ), P(t, T2 ), . . . , P(t, Tn ) are bond prices with maturities T1 < T2 < · · · < Tn ar-
ranged according to a tenor structure.
rt f
- Combinations of the above: for example a foreign money market account e 0 rs ds Rt , expressed in
f

local (or domestic) units of currency, where rt t∈R represents a short-term interest rate on
+
the foreign market.
When the numéraire is a random process, the pricing of a claim whose value has been transformed
under change of numéraire, e.g. under a change of currency, has to take into account the risks
existing on the foreign market.
In particular, in order to perform a fair pricing, one has to determine a probability measure under
which the transformed (or forward, or deflated) process Sbt = St /Nt will be a martingale, see Sec-
tion 3.3 for details.
rt
For example in case Nt := e 0 rs ds is the money market account, the risk-neutral probability measure
P∗ is a measure under which the discounted price process
St rt
Sbt = = e − 0 rs ds St , t ⩾ 0,
Nt

is a martingale. In the next section, we will see that this property can be extended to any type of
numéraire.
See Exercises 3.5 and 3.6 for other examples of numéraires.
3.2 Change of Numéraire

In this section we review the pricing of options by a change of measure associated to a numéraire
Nt , cf. e.g. Geman, El Karoui, and Rochet, 1995 and references therein.
Most of the results of this chapter rely on the following assumption, which expresses absence
of arbitrage. In the foreign exchange setting where Nt = Rt , this condition states that the price of
one unit of foreign currency is a martingale when quoted and discounted in the domestic currency.
Assumption 1. The discounted numéraire

rt
t 7→ Mt := e − 0 rs ds Nt
is an Ft -martingale under the risk-neutral probability measure P∗ .
Definition 3.1 Given (Nt )t∈[0,T ] a numéraire process, the associated forward measure P
b is
defined by its Radon-Nikodym density
dP
b MT rT
− 0 rs ds NT
: = = e . (3.2.1)
dP∗ M0 N0
w t
In particular, we note that P
b = P∗ when Nt := exp rs ds is the money market account. Recall
0
also that Relation (3.2.1) rewrites as
MT ∗ rT NT
dP
b= dP = e − 0 rs ds dP∗ ,
M0 N0
which is equivalent to stating that
w
E[F ] =
Ib F ( ω ) dP
b (ω )
Ω
w rT NT
= F (ω ) e − 0 rs ds dP∗ (ω )
Ω N0
rT

∗ N
− 0 rs ds T
= IE F e ,
N0
for all integrable FT -measurable random variables F. More generally, by (3.2.1) and the fact that
the process rt
t 7→ Mt := e − 0 rs ds Nt
is an Ft -martingale under P∗ by Assumption 1, we find that
" #
P rT

∗ d b
∗ NT − 0 rs ds
IE Ft = IE e Ft
dP∗ N0

MT
= IE∗ Ft
M0
Mt
=
M0

Nt − r t rs ds
= e 0 , 0 ⩽ t ⩽ T. (3.2.2)
N0
In Proposition 3.5 we will show, as a consequence of next Lemma 3.2 below, that for any integrable
random claim payoff C we have
h rT i
IE∗ C e − t rs ds NT Ft = Nt Ib
E[C | Ft ], 0 ⩽ t ⩽ T.
Similarly to the above, the Radon-Nikodym density dP

b |F /dP∗ of P
t |Ft
b |F with respect to P∗
t |Ft
satisfies the relation
w
E[F | Ft ] =
Ib F ( ω ) dP
b |F (ω )
Ω t
w
rs ds dP|Ft
rT b
= F (ω ) e − 0 dP∗|Ft (ω )
Ω dP∗|Ft
" #
∗ dP
b |F
= IE F ∗ Ft ,
t
dP|Ft
for all integrable FT -measurable random variables F, 0 ⩽ t ⩽ T . Note that (3.2.2), which is
Ft -measurable, should not be confused with (3.2.3), which is FT -measurable.
Lemma 3.2 We have
dP
b |F
t MT rT
− t rs ds NT
= = e , 0 ⩽ t ⩽ T. (3.2.3)
dP∗|Ft Mt Nt
Proof. The proof of (3.2.3) relies on an abstract version of the Bayes formula. For all bounded
Ft -measurable random variable G, by (3.2.2) and the tower property of conditional expectations,
we have
rT

∗ − 0 rs ds NT

IE GX = IE GX e
b b b
N0
Nt − rs ds ∗ b − r T rs ds NT
r

t
∗
= IE G e 0 IE X e t Ft
N0 Nt
" " # #
d Pb rT NT
= IE∗ G IE∗ Ft IE∗ Xb e − t rs ds Ft
dP∗ Nt
" #
P rT

d b N T
= IE∗ G ∗ IE∗ Xb e − t rs ds Ft
dP Nt
rT

= IEb G IE∗ Xb e − t rs ds NT Ft ,
Nt
for all bounded random variables X,

b which shows that
rT

b Xb | Ft = IE∗ Xb e − t rs ds NT Ft ,

IE
Nt
i.e. (3.2.3) holds. □

rt
We note that in case the numéraire Nt = e 0 rs ds is equal to the money market account we simply
have P
b = P∗ .

Definition 3.3 Given (Xt )t∈R+ an asset price process, we define the process of forward (or
deflated) prices
Xt
Xbt := , 0 ⩽ t ⩽ T. (3.2.4)
Nt

The process Xbt t∈R+
epresents the values at times t of Xt , expressed in units of the numéraire Nt .
In the sequel, it will be useful to determine the dynamics of Xbt t∈R under the forward measure P.

b
rt +
The next proposition shows in particular that the process e 0 rs ds /Nt is an Ft -martingale
t∈R+
under P.
b
Proposition 3.4 Let (Xt )t∈R+ denote a continuous (Ft )t∈R+ -adapted asset price process such
that rt
t 7→ e − 0 rs ds Xt , t ⩾ 0,
is a martingale under P∗ . Then, under change of numéraire,
the deflated process Xbt t∈[0,T ] = (Xt /Nt )t∈[0,T ] of forward prices is an Ft -martingale under

P,
b
is integrable under P.

provided that Xbt t∈[0,T ]
b
Proof. We show that

Xt Xs
IE
b Fs = , 0 ⩽ s ⩽ t. (3.2.5)
Nt Ns
Namely, for all bounded Fs -measurable random variables G we note that using (3.2.2) under
Assumption 1 we have
" #
P

Xt Xt d b
Ib
E G = IE∗ G
Nt Nt dP∗
" " ##
Xt dPb
= IE∗ IE∗ G Ft
Nt dP∗
" " ##
∗ Xt ∗ dP b
= IE G IE Ft
Nt dP∗
rt

∗ − 0 ru du Xt
= IE G e (3.2.6)
N0
rs

∗ − 0 ru du Xs
= IE G e (3.2.7)
N0
" " ##
∗ Xs ∗ dP b
= IE G IE Fs
Ns dP∗
" " ##
Xs d P
b
= IE∗ IE∗ G Fs
Ns dP∗

70Chapter 3. Change of Numéraire and Forward Measures
" #
∗ Xs dP b
= IE G
Ns dP∗

Xs
= IE G
b , 0 ⩽ s ⩽ t,
Ns
where from (3.2.6) to (3.2.7)we used the fact that

rt
t 7→ e − 0 rs ds Xt
is an Ft -martingale under P∗ . Finally, the identity

Xt Xs
IE GXt = IE G
b b b = IE G
b = IbE GXbs , 0 ⩽ s ⩽ t,
Nt Ns
for all bounded Fs -measurable G, implies (3.2.5). □
Pricing using change of numéraire

The change of numéraire technique is especially useful for pricing under random interest rates, in
which case an expectation of the form
h rT i
IE∗ e − t rs dsC Ft
becomes a path integral, see e.g. Dash, 2004 for an account of path integral methods in quantitative
finance. The next proposition is the basic result of this section,
rT
it provides a way to price an option
− t rs ds
with arbitrary payoff C under a random discount factor e by use of the forward measure. It
will be applied in Chapter 4 to the pricing of bond options and caplets, cf. Propositions 4.1, 4.3 and
4.5 below.
Proposition 3.5 An option with integrable claim payoff C ∈ L1 (Ω, P∗ , FT ) is priced at time t
as
rT

∗
h
−
i C
IE e t C Ft = Nt IE
rs ds b Ft , 0 ⩽ t ⩽ T, (3.2.8)
NT
provided that C/NT ∈ L1 Ω, P,

b FT .

Proof. By Relation (3.2.3) in Lemma 3.2 we have

" #
h rT i d P
b |F N
t
IE∗ e − t rs dsC Ft = IE∗ t
C Ft
dP∗|Ft NT
" #
d P
b |F C
= Nt IE∗ t
Ft
dP∗|Ft NT
w dP b |F C
= Nt t
dP∗|Ft
Ω dP∗ NT
|Ft
w C
= Nt dPb |F
Ω NT t

b C Ft ,
= Nt IE 0 ⩽ t ⩽ T.
NT

Equivalently, we can write

" #
C dP

C b |F
Nt Ib
E Ft = Nt IE∗ ∗
t
Ft
NT NT dP|Ft
h rT i
= IE∗ e − t rs dsC Ft , 0 ⩽ t ⩽ T.
□
Each application of the change of numéraire formula (3.2.8) will require to:
a) pick a suitable numéraire (Nt )t∈R+ satisfying Assumption 1,
b) make sure that the ratio C/NT takes a sufficiently simple form,
c) use the Girsanov theorem in order to determine the dynamics of asset prices under the new
probability measure P,
b
so as to compute the expectation under Pb on the right-hand side of (3.2.8).
Next, we consider further examples of numéraires and associated examples of option prices.
Examples:
a) Money market account.
rt
Take Nt := e 0 rs ds , where (rt )t∈R+ is a possibly random and time-dependent risk-free interest
rate. In this case, Assumption 1 is clearly satisfied, we have P b = P∗ and dP∗ /dP, b and
(3.2.8) simply reads
h rT i rt h rT i
IE∗ e − t rs dsC Ft = e 0 rs ds IE∗ e − 0 rs dsC Ft , 0 ⩽ t ⩽ T,
which yields no particular information.
b) Forward numéraire.
Here, Nt := P(t, T ) is the priceP(t, T ) of a bond

maturing at time T , 0 ⩽ t ⩽ T , and the
rt
discounted bond price process e− 0 rs ds P(t, T ) is an Ft -martingale under P∗ , i.e.
t∈[0,T ]
Assumption 1 is satisfied and Nt = P(t, T ) can be taken as numéraire. In this case, (3.2.8)
shows that a random claim payoff C can be priced as
h rT i
IE∗ e − t rs dsC Ft = P(t, T )Ib
E[C | Ft ], 0 ⩽ t ⩽ T, (3.2.9)
since NT = P(T , T ) = 1, where the forward measure P

b satisfies
rT
dP
b rT
− 0 rs ds P(T , T ) e − 0 rs ds
= e = (3.2.10)
dP∗ P(0, T ) P(0, T )
by (3.2.1).
c) Annuity numéraires.
We take
n
Nt := ∑ (Tk − Tk−1 )P(t, Tk )
k =1

where P(t, T1 ), . . . , P(t, Tn ) are bond prices with maturities T1 < T2 < · · · < Tn . Here, (3.2.8)
h that a swaption on the cash flow P(T , Tn )i− P(T , T1 ) − κNT can be priced as
shows
rT
IE∗ e − t (P(T , Tn ) − P(T , T1 ) − κNT )+ Ft
rs ds
" + #
P ( T , Tn ) − P ( T , T1 )
= Nt IEb −κ Ft ,
NT
0 ⩽ t ⩽ T < T1 , where (P(T , Tn ) − P(T , T1 ))/NT becomes a swap rate, cf. (2.2.7) in
Proposition 2.12 and Section 4.5.
Girsanov theorem
We refer to e.g. Theorem III-35 page 132 of Protter, 2004 for the following version of the Girsanov
Theorem.
b is equivalenta to P∗ with Radon-Nikodym density dP/dP
Theorem 3.6 Assume that P b ∗ , and
let (Wt )t∈[0,T ] be a standard Brownian motion under P∗ . Then, letting
dP
b
∗
Φt := IE Ft , 0 ⩽ t ⩽ T, (3.2.11)
dP∗

the process W
bt
t∈[0,T ]
defined by
bt := dWt − 1
dW dΦt • dWt , 0 ⩽ t ⩽ T, (3.2.12)
Φt
is a standard Brownian motion under P.

b
a This means that the Radon-Nikodym densities dP/dP
b ∗ and dP∗ /dP
b exist and are strictly positive with P∗ and
P-probability one, respectively.
b
In case the martingale (Φt )t∈[0,T ] takes the form

w
1wt

t
Φt = exp − ψs dWs − 2
|ψs | ds , 0 ⩽ t ⩽ T,
0 2 0
i.e.
dΦt = −ψt Φt dWt , 0 ⩽ t ⩽ T,
the Itô multiplication table shows that Relation (3.2.12) reads
dWbt = dWt − 1 dΦt • dWt

Φt
1
= dWt − (−ψt Φt dWt ) • dWt
Φt
= dWt + ψt dt, 0 ⩽ t ⩽ T,
rt
and shows that the shifted process Wbt
t∈[0,T ]
= Wt + 0 ψs ds t∈[0,T ] is a standard Brownian motion
under P,
b which is consistent with the Girsanov Theorem. The next result is another application of
the Girsanov Theorem.

Proposition 3.7 The process W
bt
t∈[0,T ]
defined by
bt := dWt − 1
dW dNt • dWt , 0 ⩽ t ⩽ T, (3.2.13)
Nt


b
Proof. Relation (3.2.2) shows that Φt defined in (3.2.11) satisfies
dP
b
Φt = IE∗ F t
dP∗
rT

∗ NT − 0 rs ds
= IE e Ft
N0
Nt − r t rs ds
= e 0 , 0 ⩽ t ⩽ T,
N0
hence
Nt − r t rs ds

dΦt = d e 0
N0
rt

− Nt
= −Φt rt dt + e 0 rs ds d
N0
Φt
= −Φt rt dt + dNt ,
Nt
which, by (3.2.12), yields
1
dW
bt = dWt − dΦt • dWt
Φt
Φt

1
= dWt − −Φt rt dt + dNt • dWt
Φt Nt
1
= dWt − dNt • dWt ,
Nt
which is (3.2.13), from Relation (3.2.12) and the Itô multiplication table. □
The next Proposition 3.8 is consistent with the statement of Proposition 3.4, and in addition it
specifies the dynamics of Xbt t∈R under P

b using the Girsanov Theorem 3.7. As a consequence,
+
we have the next proposition, see Exercise 3.1 for another calculation based on geometric Brownian
motion, and Exercise 3.10 for an extension to correlated Brownian motions.
Proposition 3.8 Assume that (Xt )t∈R+ and (Nt )t∈R+ satisfy the stochastic differential equations
dXt = rt Xt dt + σtX Xt dWt , and dNt = rt Nt dt + σtN Nt dWt , (3.2.14)
where (σtX )t∈R+ and (σtN )t∈R+ are (Ft )t∈R+ -adapted volatility processes and (Wt )t∈R+ is a
standard Brownian motion under P∗ . Then the forward (or deflated) process Xbt t∈[0,T ] =

(Xt /Nt )t∈[0,T ] satisfies
d Xbt = σtX − σtN Xbt dW

bt , (3.2.15)

hence (Xbt )t∈[0,T ] is given by the driftless geometric Brownian motion

w
1wt X

t
Xbt = Xb0 exp (σsX − σsN )dW
bs − (σs − σsN )2 ds , 0 ⩽ t ⩽ T.
0 2 0
Proof. First, we note that by (3.2.13) and (3.2.14),

1
bt = dWt −
dW dNt • dWt = dWt − σtN dt, t ⩾ 0,
Nt
is a standard Brownian motion under P. b Next, by Itô’s calculus and the Itô multiplication table and
(3.2.14) we have

1 1 1
d = − 2 dNt + 3 dNt • dNt
Nt Nt Nt
1 |σ N |2
= − 2 rt Nt dt + σtN Nt dWt + t dt
Nt Nt
N 2
1
= − 2 rt Nt dt + σtN Nt dW bt + σtN dt + |σt | dt

Nt Nt
1
rt dt + σtN dW

= − bt , (3.2.16)
Nt
hence

Xt
d Xt = d
b
Nt

dXt 1 1
= + Xt d + dXt • d
Nt Nt Nt
1 Xt
rt Xt dt + σtX Xt dWt − rt dt + σtN dWt − |σtN |2 dt

=
Nt Nt
1
rt Xt dt + σtX Xt dWt · rt dt + σtN dWt − |σtN |2 dt

−
Nt
1 Xt
rt Xt dt + σtX Xt dWt − rt dt + σtN dWt

=
Nt Nt
Xt N 2 Xt X N
+ |σt | dt − σt σt dt
Nt Nt
Xt X Xt Xt |σ N |2
= σt dWt − σtN dWt − σtX σtN dt + Xt t dt
Nt Nt Nt Nt
Xt X N X N N 2

= σ dWt − σt dWt − σt σt dt + |σt | dt
Nt t
= Xbt σtX − σtN dWt − Xbt σtX − σtN σtN dt

= Xbt σtX − σtN dW

bt ,
bt = dWt − σtN dt, 0 ⩽ t ⩽ T .
since dW □
We end this section with some comments on inverse changes of measure.
Inverse changes of measure
In the next proposition we compute the conditional inverse Radon-Nikodym density dP∗ /dP,
b see
also (3.2.2).

Proposition 3.9 We have

∗
dP N0 w t
Ib
E Ft = exp rs ds , 0 ⩽ t ⩽ T, (3.2.17)
dP
b Nt 0
and the process w

M0 N0 t
t 7→ = exp rs ds , 0 ⩽ t ⩽ T,
Mt Nt 0
is an Ft -martingale under P.
b
Proof. For all bounded and Ft -measurable random variables F we have,

dP∗

IE F
b = IE∗ [F ]
dPb

∗ Nt
= IE F
Nt
w
∗ NT T
= IE F exp − rs ds
Nt t

N0
w t
= IEb F exp rs ds .
Nt 0
□
By (3.2.16) we also have

1 w t 1 w t
d exp rs ds = − exp rs ds σtN dW
bt ,
Nt 0 Nt 0
which recovers the second part of Proposition 3.9, i.e. the martingale property of
M0 1 w t
t 7→ = exp rs ds
Mt Nt 0
under P.
b
3.3 Foreign Exchange

Currency exchange is a typical application of change of numéraire, that illustrates the absence of
arbitrage principle.
Let Rt denote the foreign exchange rate, i.e. Rt is the (possibly fractional) quantity of local
currency that correspond to one unit of foreign currency, while 1/Rt represents the quantity of
foreign currency that correspond to a unit of local currency.
Rt t rf Discount Rt ( rf − rl )t
Local $1 e e
R0 R0
f
1 et r
Foreign
R0 R0
Consider an investor that intends to exploit an “overseas investment opportunity” by

a) at time 0, changing one unit of local currency into 1/R0 units of foreign currency,
f
b) investing 1/R0 on the foreign market at the rate rf , which will yield the amount e t r /R0 at
time t > 0,
f f
c) changing back e t r /R0 into a quantity e t r Rt /R0 = Nt /R0 of local currency.
Figure 3.2: Overseas investment opportunity.*

f f
In other words, the foreign money market account e t r is valued e t r Rt on the local (or domestic)
market, and its discounted value on the local market is
l f
e −t r +t r Rt , t > 0.
f
The outcome of this investment will be obtained by a martingale comparison of e t r Rt /R0 to the
l
amount e t r that could have been obtained by investing on the local market.
Taking
f
Nt := e t r Rt , t ⩾ 0, (3.3.1)
as numéraire, absence of arbitrage is expressed by Assumption 1, which states that the discounted
numéraire process
l l f
t 7→ e − r t Nt = e −t (r −r ) Rt
is an Ft -martingale under P∗ .
Next, we find a characterization of this arbitrage condition using the model parameters
r, r f , µ, by modeling the foreign exchange rates Rt according to a geometric Brownian motion
(3.3.2).
Proposition 3.10 Assume that the foreign exchange rate Rt satisfies a stochastic differential
equation of the form
dRt = µRt dt + σ Rt dWt , (3.3.2)
where (Wt )t∈R+ is a standard Brownian motion under P∗ . Under the absence of arbitrage
Assumption 1 for the numéraire (3.3.1), we have
µ = rl − rf , (3.3.3)
hence the exchange rate process satisfies
dRt = rl − rf Rt dt + σ Rt dWt .

(3.3.4)
* For illustration purposes only. Not an advertisement.

under P∗ .
Proof. The equation (3.3.2) has solution
2 t/2
Rt = R0 e µt +σWt −σ , t ⩾ 0,
f
hence the discounted value of the foreign money market account e t r on the local market is
l l f f l 2 t/2
e −t r Nt = e −t r +t r Rt = R0 e (r −r +µ )t +σWt −σ , t ⩾ 0.
l f l
Under the absence of arbitrage Assumption 1, the process e −(r −r )t Rt = e −t r Nt should be an
Ft -martingale under P∗ , and this holds provided that rf − r + µ = 0, which yields (3.3.3) and
(3.3.4). □
As a consequence of Proposition 3.10, under absence of arbitrage a local investor who buys a unit
of foreign currency in the hope of a higher return rf >> r will have to face a lower (or even more
negative) drift
µ = rl − rf << 0
in his exchange rate Rt . The drift µ = rl − rf is also called the cost of carrying the foreign currency.
l l
The local money market account Xt := e t r is valued e t r /Rt on the foreign market, and its
discounted value at time t ⩾ 0 on the foreign market is
l f
e (r −r )t Xt
= = Xbt (3.3.5)
Rt Nt
1 (rl −rf )t−µt−σWt +σ 2t/2
= e
R0
1 (rl −rf )t−µt−σ Wbt −σ 2t/2
= e ,
R0
where
1
dW
bt = dWt − dNt • dWt
Nt
1
= dWt − dRt • dWt
Rt
= dWt − σ dt, t ⩾ 0,
is a standard Brownian motion under P b by (3.2.13). Under absence of arbitrage, the process
l −r f )t
e − ( r Rt is an Ft -martingale under P∗ and (3.3.5) is an Ft -martingale under P
b by Proposi-
tion 3.4, which recovers (3.3.3).
library(quantmod)
getSymbols("EURTRY=X",src = "yahoo",from = "2018-01-01",to = "2021-12-31")
getSymbols("INTDSRTRM193N",src = "FRED")
Interestrate<-`INTDSRTRM193N`["2018-01-01::2021-31-12"]
EURTRY<-Ad(`EURTRY=X`); myPars <- chart_pars();myPars$cex<-1.2
Cumulative<-cumprod(1+Interestrate/100/12)
normalizedfxrate<-1+(as.numeric(last(Cumulative))-1)*(EURTRY-as.numeric(EURTRY[1]))/
(as.numeric(last(EURTRY))-as.numeric(EURTRY[1]))
myTheme <- chart_theme();myTheme$col$line.col <- "blue"
dev.new(width=16,height=8)
chart_Series(Cumulative,pars=myPars, theme = myTheme)
add_TA(normalizedfxrate, col='black', lw =2, on = 1)
add_TA(Interestrate, col='purple', lw =2)

The above code plots an evolution of currency exchange rates compared with the evolution of
interest rates, as shown in Figure 3.3.
Cumulative 2018−01−01 / 2022−02−01

2.1 2.1
2.0 2.0
1.9 1.9
1.8 1.8
1.7 1.7
1.6 1.6
1.5 1.5
1.4 1.4
1.3 1.3
1.2 1.2
1.1 1.1
1.0
20 1.0
20
Interestrate 14.75
18 18
16 16
14 14
12 12
10 10
8 2018 2019 2020 2021 2022 8
Jan Jul Jan Jul Jan Jul Jan Jul Dec

2018 2018 2019 2019 2020 2020 2021 2021 2021
Figure 3.3: Evolution of exchange rate vs. interest rate.*
Proposition 3.11 Under the absence of arbitrage condition (3.3.3), the inverse exchange rate
1/Rt satisfies
rf − rl

1 σ b
d = dt − dWt, (3.3.6)
Rt Rt Rt
under P
b , where (Rt )t∈R is given by (3.3.4).
+
Proof. By (3.3.3), the exchange rate 1/Rt is written using Itô’s calculus as

1 1 1
d = − 2 ( µRt dt + σ Rt dWt ) + 3 σ 2 Rt2 dt
Rt Rt Rt
µ −σ 2 σ
= − dt − dWt
Rt Rt
µ σ b
= − dt − dWt
Rt Rt
f
r −r l σ b
= dt − dW t,
Rt Rt
bt )t∈R+ is a standard Brownian motion under P.

where (W b □
Consequently, under absence of arbitrage, a foreign investor who buys a unit of the local currency
in the hope of a higher return r >> rf will have to face a lower (or even more negative) drift
−µ = rf − r in his exchange rate 1/Rt as written in (3.3.6) under P.
b
Foreign exchange options

We now price a foreign exchange call option with payoff (RT − κ )+ under P∗ on the exchange
rate RT by the Black-Scholes formula as in the next proposition, also known as the Garman and
Kohlhagen, 1983 formula. The foreign exchange call option is designed for a local buyer of foreign
currency.

Proposition 3.12 (Garman and Kohlhagen, 1983 formula for call options). Consider the ex-
change rate process (Rt )t∈R+ given by (3.3.4). The price of the foreign exchange call option on
RT with maturity T and strike price κ > 0 is given in local currency units as
l f l
e −(T −t ) r IE∗ (RT − κ )+ Ft = e −(T −t ) r Rt Φ+ (t, Rt ) − κ e −(T −t ) r Φ− (t, Rt ),

(3.3.7)
0 ⩽ t ⩽ T , where
!
log(x/κ ) + (T − t ) rl − rf + σ 2 /2
Φ+ (t, x) = Φ √ ,
σ T −t
and !
log(x/κ ) + (T − t ) rl − rf − σ 2 /2
Φ− (t, x) = Φ √ .
σ T −t
Proof. As a consequence of (3.3.4), we find the numéraire dynamics

f
dNt = d e t r Rt

f f
= rf e t r Rt dt + e t r dRt
f f
= rf e t r Rt dt + σ e t r Rt dWt
= rf Nt dt + σ Nt dWt .
Hence, a standard application of the Black-Scholes formula yields
l l f +
e −(T −t ) r IE∗ (RT − κ )+ Ft = e −(T −t ) r IE∗ e −T r NT − κ Ft

l f f +
= e −(T −t ) r e −T r IE∗ NT − κ e T r | Ft

f
!
−T rf log(Nt e −T r /κ ) + (r + σ 2 /2)(T − t )
= e Nt Φ √
σ T −t
−Trf /κ ) + r l − σ 2 /2 (T − t )
!!
f l log ( N t e
−κ e T r −(T −t ) r Φ √
σ T −t
l − r f + σ 2 /2
!
f log ( Rt /κ ) + ( T − t ) r
= e −T r Nt Φ √
σ T −t
l − r f − σ 2 /2
!!
f l log ( Rt /κ ) + ( T − t ) r
−κ e T r −(T −t ) r Φ √
σ T −t
f l
= e −(T −t ) r Rt Φ+ (t, Rt ) − κ e −(T −t ) r Φ− (t, Rt ).
A similar conclusion can be reached by directly applying (3.3.4). □
Similarly, from (3.3.6) rewritten as
l
! l l
et r et r
f et r b
d =r dt − σ dWt ,
Rt Rt Rt
a foreign exchange put option with payoff (1/κ − 1/RT )+ can be priced under P b in a Black-
l
Scholes model by taking e t r /Rt as underlying asset price, rf as risk-free interest rate, and −σ

as volatility parameter. The foreign exchange put option is designed for the foreign seller of
local currency, see for example the buy back guarantee* which is a typical example of a foreign
exchange put option.
Proposition 3.13 (Garman and Kohlhagen, 1983 formula for put options). Consider the exchange
rate process (Rt )t∈R+ given by (3.3.4). The price of the foreign exchange put option on RT with
maturity T and strike price 1/κ > 0 is given in foreign currency units as
1 +

−(T −t ) rf b 1
e IE − Ft (3.3.8)
κ RT
f l
e −(T −t ) r e −(T −t ) r

1 1
= Φ− t, − Φ+ t, ,
κ Rt Rt Rt
0 ⩽ t ⩽ T , where
!
log(κx) + (T − t ) rf − rl + σ 2 /2
Φ+ (t, x) := Φ − √ ,
σ T −t
and !
log(κx) + (T − t ) rf − rl − σ 2 /2
Φ− (t, x) := Φ − √ .
σ T −t
Proof. The Black-Scholes formula yields

" + # " l l
!+ #
−(T −t ) rf b 1 1 eT r
−(T −t ) rf −T rl b eT r
e IE − Ft = e e IE − Ft
κ RT κ RT
l
e −(T −t ) r

1 −(T −t ) rf 1 1
= e Φ− t, − Φ+ t, ,
κ Rt Rt Rt
which is the symmetric of (3.3.7) by exchanging Rt with 1/Rt , and r with r f . □
Call/put duality for foreign exchange options

f
Let Nt = e t r Rt , where Rt is an exchange rate with respect to a foreign currency and r f is the
foreign market interest rate.
Proposition 3.14 The foreign exchange call and put options on the local and foreign markets
are linked by the call/put duality relation
" + #
−(T −t ) rl ∗ −(T −t ) rf b 1 1
+
Ft = κRt e Ft ,

e IE (RT − κ ) IE − (3.3.9)
κ RT
between a put option with strike price 1/κ and a (possibly fractional) quantity 1/(κRt ) of call
option(s) with strike price κ.
* Right-click to open or save the attachment.

Proof. By application of change of numéraire from Proposition 3.5 and (3.2.8) we have

1 + 1 −(T −t ) rl ∗
Ft = IE (RT − κ )+ Ft ,

IE
b f ( RT − κ ) e
e T r RT Nt
hence
" + #
−(T −t ) rf b 1 1 −(T −t ) rf b 1
e IE − Ft = e IE (RT − κ )+ Ft
κ RT κRT

1 t rf b 1
= e IE T rf (RT − κ )+ Ft
κ e RT
1 t rf −(T −t ) rl ∗
IE (RT − κ )+ Ft

= e
κNt
1 −(T −t ) rl ∗
IE (RT − κ )+ Ft .

= e
κRt
In the Black-Scholes case, the duality (3.3.9) can be directly checked by verifying that (3.3.8)
coincides with
1 −(T −t ) rl ∗
IE (RT − κ )+ Ft

e
κRt
1 −(T −t ) rl −T rf ∗ T rf f +
e RT − κ e T r Ft

= e e IE
κRt
1 −(T −t ) rl −T rf ∗ f +
IE NT − κ e T r Ft

= e e
κRt
1 f l
e −(T −t ) r Rt Φc+ (t, Rt ) − κ e −(T −t ) r Φc− (t, Rt )

=
κRt
l
1 −(T −t ) rf c e −(T −t ) r c
= e Φ+ (t, Rt ) − Φ− (t, Rt )
κ Rt
l
e −(T −t ) r

1 −(T −t ) rf 1 1
= e Φ− t, − Φ+ t, ,
κ Rt Rt Rt
where
!
log(x/κ ) + (T − t ) rl − rf + σ 2 /2
Φc+ (t, x) := Φ √ ,
σ T −t
and
!
log(x/κ ) + (T − t ) rl − rf − σ 2 /2
Φc− (t, x) := Φ √ .
σ T −t

“Local” market “Foreign” market
Measure P∗ P
b
l f
Discount factor t 7→ e −r t t 7→ e −r t
l f
l l f Xt e (r −r )t
Martingale t 7→ e −t r Nt = e −t (r −r ) Rt t 7→ = Xbt =
Nt Rt
" + #
−(T −t ) rl −(T −t ) rf 1 1
IE∗ (RT − κ )+ Ft Ft

Option e e Ib
E −
κ RT
Application Local purchase of foreign currency Foreign selling of local currency
Table 3.1: Local vs. foreign exchange options.
Example - Buy back guarantee

The put option priced
1 +

−(T −t ) rf b 1
e IE − Ft
κ RT
l
e −(T −t ) r

1 −(T −t ) rf 1 1
= e Φ− t, − Φ+ t,
κ Rt Rt Rt
on the foreign market corresponds to a buy back guarantee* in currency exchange. In the case of
an option “at the money” with κ = Rt and rl = rf ≃ 0, we find
" + # √ √
1 1 1 σ T −t σ T −t
IE
b − Ft = × Φ −Φ −
Rt RT Rt 2 2
√
1 σ T −t
= × 2Φ −1 .
Rt 2
For example, let Rt denote the USD/EUR (USDEUR=X) exchange rate from a foreign currency
(USD) to a local currency (EUR), i.e. one unit of the foreign currency (USD) corresponds to
Rt = 1/1.23 units of local currency (EUR). Taking T − t = 30 days and σ = 10%, we find that the
foreign currency put option allowing the foreign sale of one EURO back into USDs is priced at the
money in USD as
" + #
1 1 p
Ft

Ib
E − = 1.23 2Φ 0.05 × 31/365 − 1
Rt RT
= 1.23(2 × 0.505813 − 1)
= $0.01429998
* Right-click to open or save the attachment.

per USD, or €0.011626 per exchanged unit of EURO. Based on a displayed option price of €4.5 and
in order to make the contract fair, this would translate into an average amount of 4.5/0.011626 ≃
€387 exchanged at the counter by customers subscribing to the buy back guarantee.
3.4 Pricing Exchange Options

Based on Proposition 3.4, we model the process Xbt of forward prices as a continuous martingale
under P,
b written as
d Xbt = σbt dW
bt , t ⩾ 0, (3.4.1)
is a standard Brownian motion under P bt )t∈R+ is an (Ft )t∈R+ -adapted

where W bt b and (σ
t∈R+

stochastic volatility process. More precisely, we assume that Xbt has the dynamics
t∈R+

d Xbt = σbt Xbt dW
bt , (3.4.2)
where x 7→ σbt (x) is a local volatility function which is Lipschitz in x, uniformly in t ⩾ 0. The
Markov property of the diffusion process Xbt t∈R , cf. Theorem V-6-32 of Protter, 2004, shows
+
E gb XbT Ft can be

that when gb is a deterministic payoff function, the conditional expectation Ib
written using a (measurable) function Cb(t, x) of t and Xbt , as
b gb XbT Ft = Cb t, Xbt ,

IE 0 ⩽ t ⩽ T.

Consequently, a vanilla option with claim payoff C := NT gb XbT can be priced from Proposition 3.5
as
h rT i
IE∗ e − t rs ds NT gb XbT Ft = Nt Ib E gb XbT Ft

= Nt Cb t, Xbt , 0 ⩽ t ⩽ T . (3.4.3)
In the next Proposition 3.15 we state the Margrabe, 1978 formula for the pricing of exchange options
by the zero interest rate Black-Scholes formula. It will be applied in particular in Proposition 4.3
below for the pricing of bond options. Here, (Nt )t∈R+ denotes any numéraire process satisfying
Assumption 1.

Proposition 3.15 (Margrabe, 1978 formula). Assume that σ bt Xbt = σb (t )Xbt , i.e. the martingale
Xbt t∈[0,T ] is a (driftless) geometric Brownian motion under P

b with deterministic volatility
(σb (t ))t∈[0,T ] . Then we have
h rT i
IE∗ e − t rs ds (XT − κNT )+ Ft = Xt Φ0+ t, Xbt − κNt Φ0− t, Xbt ,

(3.4.4)
t ∈ [0, T ], where

log(x/κ ) v(t, T ) log(x/κ ) v(t, T )
Φ0+ (t, x) =Φ + , Φ0− (t, x) =Φ − , (3.4.5)
v(t, T ) 2 v(t, T ) 2

wT
and v2 (t, T ) = b 2 (s)ds.
σ
t
Proof. Taking g(x) := (x − κ )+ in (3.4.3), the call option with payoff

+
(XT − κNT )+ = NT XbT − κ
w
1wT
+
T
2
= NT Xt exp
b σb (t )dWt −
b |σ
b (t )| dt − κ ,
t 2 t
and floating strike price κNT is priced by (3.4.3) as
h rT i h rT + i
IE∗ e − t rs ds (XT − κNT )+ Ft = IE∗ e − t rs ds NT XbT − κ Ft
+
Ft

= Nt IbE XbT − κ

= Nt Cb t, Xbt ,

where the function Cb t, Xbt is given by the Black-Scholes formula
Cb(t, x) = xΦ0+ (t, x) − κΦ0− (t, x),

with zero interest rate, since Xbt t∈[0,T ] is a driftless geometric Brownian motion which is an
F -martingale under P, b and XbT is a lognormal random variable with variance coefficient v2 (t, T ) =
w Tt
b 2 (s)ds. Hence we have
σ
t
h rT i
IE∗ e − t rs ds (XT − κNT )+ Ft = Nt Cb t, Xbt

= Nt Xbt Φ0+ t, Xbt − κNt Φ0− t, Xbt ,

t ⩾ 0. □
In particular, from Proposition 3.8 and (3.2.15), we can take σb (t ) = σtX − σtN when (σtX )t∈R+ and
(σtN )t∈R+ are deterministic.
Examples:
rT
a) When the short rate process (r (t ))t∈[0,T ] is a deterministic function of time and Nt = e t r(s)ds ,
0 ⩽ t ⩽ T , we have P b = P∗ and Proposition 3.15 yields the Merton, 1973 “zero interest rate”
version
r T of the Black-Scholes formula
− t r (s)ds ∗
IE (XT − κ ) Ft
+

e
rT rT rT
= Xt Φ0+ t, e t r(s)ds Xt − κ e − t r(s)ds Φ0− t, e t r(s)ds Xt ,
where Φ0+ and Φ0− are defined in (3.4.5) and (Xt )t∈R+ satisfies the equation
dXt d Xbt
= r (t )dt + σb (t )dWt , i.e. = σb (t )dWt , 0 ⩽ t ⩽ T.
Xt Xbt
b) In the case of pricing under a forward numéraire, i.e. when Nt = P(t, T ), t ∈ [0, T ], we get
h rT i
IE∗ e − t rs ds (XT − κ )+ Ft = Xt Φ0+ t, Xbt − κP(t, T )Φ0− t, Xbt ,

0 ⩽ t ⩽ T , since NT = P(T , T ) = 1. In particular, when Xt = P(t, S) the above formula

allows us to price a bond call option on P(T , S) as
h rT i
+
IE∗ e − t rs ds (P(T , S) − κ ) Ft = P(t, S)Φ0+ t, Xbt − κP(t, T )Φ0− t, Xbt ,

0 ⩽ t ⩽ T , provided that the martingale Xbt = P(t, S)/P(t, T ) under P

b is given by a geometric
Brownian motion, cf. Section 4.2.

3.5 Hedging by Change of Numéraire

In this section we reconsider and extend the standard Black-Scholes self-financing hedging strategy.
For this, we use the stochastic integral representation of the forward claim payoffs and change of
numéraire in order to compute self-financing portfolio strategies. Our hedging rt
portfolios will be
built on the assets (Xt , Nt ), not on Xt and the money market account Bt = e 0 ds , extending the
rs
classical hedging portfolios that are available in from the Black-Scholes formula, using a technique
from Jamshidian, 1996, cf. also Privault and Teng, 2012.
Consider a claim with random payoff C, typically an interest rate derivative, cf. Chapter 4.
Assume that the forward claim payoff C/NT ∈ L2 (Ω) has the stochastic integral representation
w
C C T
C :=
b = IE
b + φbt d Xbt , (3.5.1)
NT NT 0
where Xbt t∈[0,T ] is given by (3.4.1) and φbt t∈[0,T ] is a square-integrable adapted process under P,

b
from which it follows that the forward claim price
Vt 1 h rT i
C

Vbt := = IE∗ e − t rs dsC Ft = IbE Ft , 0 ⩽ t ⩽ T,
Nt Nt NT
is an Ft -martingale under P,
b that can be decomposed as
w
C t
Vt = IE C | Ft = IE

b b b b + φbs d Xbs , 0 ⩽ t ⩽ T. (3.5.2)
NT 0
The next Proposition 3.16 extends the argument of Jamshidian, 1996 to the general framework of
pricing using change of numéraire.r Note that this result differs from the standard formula that uses
t
the money market account Bt = e 0 rs ds for hedging instead of Nt , cf. e.g. Geman, El Karoui, and
Rochet, 1995 pages 453-454.
bt := Vbt − Xbt φbt , with φbt defined in (3.5.2), 0 ⩽ t ⩽ T , the portfolio

Proposition 3.16 Letting η
allocation
φbt , η
bt t∈[0,T ]
with value
Vt = φbt Xt + η
bt Nt , 0 ⩽ t ⩽ T,
is self-financing in the sense that
dVt = φbt dXt + η

bt dNt ,
and it hedges the claim payoff C, i.e.

h rT i
bt Nt = IE∗ e − t rs dsC Ft ,
Vt = φbt Xt + η 0 ⩽ t ⩽ T. (3.5.3)

Proof. In order to check that the portfolio allocation φbt , η
bt t∈[0,T ] hedges the claim payoff C it
suffices to check that (3.5.3) holds since by (3.2.8) the price Vt at time t ∈ [0, T ] of the hedging
portfolio satisfies
h rT i
Vt = Nt Vbt = IE∗ e − t rs dsC Ft , 0 ⩽ t ⩽ T.


Next, we show that the portfolio allocation φbt , η
bt t∈[0,T ] is self-financing. By numéraire invariance,
cf. e.g. page 184 of Protter, 2001, we have, using the relation dVbt = φbt d Xbt from (3.5.2),

dVt = d Nt Vbt
= Vbt dNt + Nt dVbt + dNt • dVbt
= Vbt dNt + Nt φbt d Xbt + φbt dNt • d Xbt

= φbt Xbt dNt + Nt φbt d Xbt + φbt dNt • d Xbt + Vbt − φbt Xbt dNt

= φbt d Nt Xbt + η bt dNt
= φbt dXt + η bt dNt .
□
We now consider
an application to the forward Delta hedging of European-type options with payoff
C = NT gb XT where gb : R −→ R and Xt t∈R has the Markov property as in (3.4.2), where
b b
+
σb : R+ × R is a deterministic function. Assuming that the function Cb(t, x) defined by
b gb XbT Ft = Cb t, Xbt

Vbt := IE
is C 2 on R+ , we have the following corollary of Proposition 3.16, which extends the Black-Scholes
Delta hedging technique to the general change of numéraire setup.
∂ Cb b
bt = Cb t, Xbt − Xbt
Corollary 3.17 Letting η t, Xt , 0 ⩽ t ⩽ T , the portfolio allocation
∂x
∂ Cb
t, Xbt , η
bt
∂x t∈[0,T ]
with value
∂ Cb b
Vt = η
bt Nt + Xt t, Xt , t ⩾ 0,
∂x

is self-financing and hedges the claim payoff C = NT gb XbT .
Proof. This result follows directly from Proposition 3.16 by noting that by Itô’s formula, and the
martingale property of Vbt under P
b the stochastic integral representation (3.5.2) is given by
VbT = Cb

= gb XbT
w T ∂ Cb w T ∂ Cb
= Cb(0, Xb0 ) + t, Xbt dt + t, Xbt d Xbt
0 ∂t 0 ∂x
w 2
1 T ∂ C b 2
b
+ t, Xt |σ bt | dt
2 0 ∂ x2
w T ∂ Cb
= Cb(0, Xb0 ) + t, Xbt d Xbt
w0 ∂ x
b C +
T
= IE φbt d Xbt , 0 ⩽ t ⩽ T,
NT 0
hence
∂ Cb b
φbt = t, Xt , 0 ⩽ t ⩽ T.
∂x
□

In the case of an exchange option with payoff function

+
C = (XT − κNT )+ = NT XbT − κ
under P

on the geometric Brownian motion Xbt t∈[0,T ]
b with

σbt Xbt = σ
b (t )Xbt , (3.5.4)

where σ b (t ) t∈[0,T ] is a deterministic volatility function of time, we have the following corollary
on the hedging of exchange options based on the Margrabe, 1978 formula (3.4.4).
Corollary 3.18 The decomposition

h rT i
IE∗ e − t rs ds (XT − κNT )+ Ft = Xt Φ0+ t, Xbt − κNt Φ0− t, Xbt

yields a self-financing portfolio allocation Φ0+ t, Xbt , −κΦ0− t, Xbt t∈[0,T ] in the assets (Xt , Nt ),

that hedges the claim payoff C = (XT − κNT )+ .
Proof. We apply Corollary 3.17 and the relation
∂ Cb
(t, x) = Φ0+ (t, x), x ∈ R,
∂x
for the function Cb(t, x) = xΦ0+ (t, x) − κΦ0− (t, x). □
Note that the Delta hedging method requires the computation of the function Cb(t, x) and that of the
associated finite differences, and may not apply to path-dependent claims.
Examples:
rT
a) When the short rate process (r (t ))t∈[0,T ] is a deterministic function of time and Nt = e t r (s)ds ,
Corollary
3.18 yields the
rT
usual Black-Scholes hedging strategy
Φ+ t, Xbt , −κ e 0 r ( s ) ds
Φ− (t, Xt )

t∈[0,T ]
rT rT rT
= Φ0+ t, e t r(s)ds Xbt , −κ e 0 r(s)ds Φ0− t, e t r(s)ds Xt t∈[0,T ] ,

rt
in the assets Xt , e 0 r(s)ds , that hedges the claim payoff C = (XT − κ )+ , with

rT 2
!
log(x/κ ) + t r (s)ds + (T − t )σ /2
Φ+ (t, x) := Φ √ ,
σ T −t
and rT !
2
log(x/κ ) + t r (s)ds − (T − t )σ /2
Φ− (t, x) := Φ √ .
σ T −t
b) In case Nt = P(t, T ) and Xt = P(t, S), 0 ⩽ t ⩽ T < S, Corollary 3.18 shows that when
Xbt t∈[0,T ] is modeled as the geometric Brownian motion (3.5.4) under P,

b the bond call
option with payoff (P(T , S) − κ )+ can be hedged as
h rT i
+
IE∗ e − t rs ds (P(T , S) − κ ) Ft = P(t, S)Φ+ t, Xbt − κP(t, T )Φ− t, Xbt


by the self-financing portfolio allocation

Φ+ t, Xbt , −κΦ− t, Xbt t∈[0,T ]

in the assets (P(t, S), P(t, T )), i.e. one needs to hold the quantity Φ+ t, Xbt of the bond

maturing at time S, and to short a quantity κΦ− t, Xbt of the bond maturing at time T .
Exercises
Exercise 3.1 Let (Bt )t∈R+ be a standard Brownian motion started at 0 under the risk-neutral
probability measure P∗ . Consider a numéraire (Nt )t∈R+ given by
2 t/2
Nt := N0 e ηBt −η , t ⩾ 0,
and a risky asset price process (Xt )t∈R+ given by
2 t/2
Xt := X0 e σ Bt −σ , t ⩾ 0,
in a market with risk-free interest rate r = 0. Let P b denote the forward measure relative to the
numéraire (Nt )t∈R+ , under which the process Xt := Xt /Nt of forward prices is known to be a
b
martingale.
a) Using the Itô formula,
compute
X t
d Xbt = d
Nt
X0 2 2
d e (σ −η )Bt −(σ −η )t/2 .

=
N0
b) Explain why the exchange option price IE∗ [(XT − λ NT )+ ] at time 0 has the Black-Scholes
form
IE∗ [(XT − λ NT )+ ] (3.5.5)
√ ! √ !
log Xb0 /λ b T
σ log Xb0 /λ b T
σ
= X0 Φ √ + − λ N0 Φ √ − .
σb T 2 b T
σ 2
Hints:
(i) Use the change of numéraire identity
+
IE∗ [(XT − λ NT )+ ] = N0 Ib

E XbT − λ .
(ii) The forward price Xbt is a martingale under the forward measure P b relative to the
numéraire (Nt )t∈R+ .
c) Give the value of σb in terms of σ and η.
(1) (2)
Exercise 3.2 Let Bt t∈R and Bt t∈R be correlated standard Brownian motions started at 0
+ +
(1) (2)
under the risk-neutral probability measure P∗ , with correlation Corr Bs , Bt = ρ min(s,t ), i.e.
(1) (2) (1) (2)
dBt • dBt = ρdt. Consider two asset prices St t∈R and St t∈R given by the geometric
+ +
Brownian motions
(1) (1) (1) (2) (2) (2)
−σ 2 t/2 −η 2 t/2
St := S0 e rt +σ Bt , and St := S0 e rt +ηBt , t ⩾ 0.
(2)
Let P
b 2 denote the forward measure with numéraire (Nt )t∈R := St
+ t∈R+
and Radon-Nikodym
density
(2)
dP
b2
−rT ST
(2)
ηBT −η 2 T /2
= e = e .
dP∗ S
(2)
0

(1) (2) (1)
a) Using the Girsanov Theorem 3.7, determine the shifts Bbt t∈R and Bbt t∈R of Bt t∈R
+ + +
(2)
and Bt t∈R which are standard Brownian motions under P2 . b
+
b) Using the Itô formula, compute
!
(1)
( 1 ) S
d Sbt = d t(2)
St
(1)
S0 (1) (2)
−ηBt −(σ 2 −η 2 )t/2
d e σ Bt

= (2)
,
S0
(1) (2)
and write the answer in terms of the martingales d Bbt and d Bbt .
c) Using change of numéraire, explain why the exchange option price
(1) (2) +
e −rT IE∗ ST − λ ST
at time 0 has the Black-Scholes form √ !

(1)
(1) (2) + (1) log Sb0 /λ σb T
e −rT IE∗ ST − λ ST = S0 Φ √ +
σb T 2
(1) √ !
(2) log Sb0 /λ σb T
−λ S0 Φ √ − ,
b T
σ 2
where the value of σb can be expressed in terms of σ and η.
Exercise 3.3 Consider two zero-coupon bond prices of the form P(t, T ) = F (t, rt ) and P(t, S) =
G(t, rt ), where (rt )t∈R+ is a short-term interest rate process. Taking Nt := P(t, T ) as a numéraire
defining the forward measure P, b compute the dynamics of (P(t, S))t∈[0,T ] under P
b using a standard
under P.

Brownian motion W bt
t∈[0,T ]
b
Exercise 3.4 Forward contracts. Using a change of numéraire argument for the numéraire
Nt := P(t, T ), t ∈ [0, T ], compute the price at time t ∈ [0, T ] of a forward (or future) contract with
payoff P(T , S) − K in a bond market with short-term interest rate (rt )t∈R+ . How would you hedge
this forward contract?
Exercise 3.5 (Question 2.7 page 17 of Downes, Joshi, and Denson, 2008). Consider a price
process (St )t∈R+ given by dSt = rSt dt + σ St dBt under the risk-neutral probability measure P∗ ,
where r ∈ R and σ > 0, and the option with payoff
ST (ST − K )+ = Max(ST (ST − K ), 0)
at maturity T .
a) Show that the option payoff can be rewritten as
(ST (ST − K ))+ = NT (ST − K )+

for a suitable choice of numéraire process (Nt )t∈[0,T ] .
b) Rewrite the option price e −(T −t )r IE∗ (ST (ST − K ))+ | Ft using a forward measure P

b and
a change of numéraire argument.
c) Find the dynamics of (St )t∈R+ under the forward measure P. b
d) Price the option with payoff
ST (ST − K )+ = Max(ST (ST − K ), 0)
at time t ∈ [0, T ] using the Black-Scholes formula.

Exercise 3.6 Consider the risk asset with dynamics dSt = rSt dt + σ St dWt with constant interest
rate r ∈ R and volatility σ > 0 under the risk-neutral measure P∗ . The power call option with
payoff (STn − K n )+ , n ⩾ 1, is priced at time t ∈ [0, T ] as
+
e −(T −t )r IE∗ STn − K n Ft

= e −(T −t )r IE∗ STn 1{ST ⩾K} Ft − K n e −(T −t )r IE∗ 1{ST ⩾K} Ft

under the risk-neutral measure P∗ .

dP
b
a) Write down the density using the numéraire process
dP∗
2 t/2−(n−1)rt
Nt := Stn e −(n−1)nσ , t ∈ [0, T ].
b) Construct a standard Brownianmotion W bt under P.
b
− ( T −t ) r IE ST 1{ST ⩾K} Ft using change of numéraire.
∗ n

c) Compute the term e
Hint: We have
P∗ (ST ⩾ K | Ft ) = IE∗ 1{ST ⩾K} Ft

log(St /K ) + (rl − σ 2 /2)(T − t )

= Φ √ .
σ T −t
d) Price the power call option with payoff (STn − K n )+ at time t ∈ [0, T ].
Exercise 3.7 Bond options. Consider two bonds with maturities T and S, with prices P(t, T ) and
P(t, S) given by
dP(t, T )
= rt dt + ζtT dWt ,
P(t, T )
and
dP(t, S)
= rt dt + ζtS dWt ,
P(t, S)
where (ζ T (s))s∈[0,T ] and (ζ S (s))s∈[0,S] are deterministic volatility functions of time.
a) Show, using Itô’s formula, that

P(t, S) P(t, S) S
d = (ζ (t ) − ζ T (t ))dWbt ,
P(t, T ) P(t, T )

where W bt
t∈R+
b
b) Show that
w
1wT S

P(t, S) T
S T T 2
P(T , S ) = exp (ζ (s) − ζ (s))dWs −
b |ζ (s) − ζ (s)| ds .
P(t, T ) t 2 t
Let P
b denote the forward measure associated to the numéraire
Nt := P(t, T ), 0 ⩽ t ⩽ T.
c) Show that for all S, T > 0 the price at time t
h rT i
IE∗ e − t rs ds (P(T , S) − κ )+ Ft
of a bondhcall option on P(T , S) with payoff

rT i (P(T , S) − κ )+ is equal to
IE∗ e − (P(T , S) − κ )+ Ft
t rs ds
(3.5.6)

v 1 P(t, S) v 1 P(t, S)
= P(t, S)Φ + log − κP(t, T )Φ − + log ,
2 v κP(t, T ) 2 v κP(t, T )
where wT
v2 = |ζ S (s) − ζ T (s)|2 ds.
t

d) Compute the self-financing hedging strategy that hedges the bond option using a portfolio
based on the assets P(t, T ) and P(t, S).
Exercise 3.8 Consider two risky assets S1 and S2 modeled by the geometric Brownian motions
S1 (t ) = e σ1Wt +µt and S2 (t ) = e σ2Wt +µt , t ⩾ 0, (3.5.7)
where (Wt )t∈R+ is a standard Brownian motion under P.
a) Find a condition on r, µ and σ2 so that the discounted price process e −rt S2 (t ) is a martingale
under P.
b) Assume that r − µ = σ22 /2, and let
2 2
Xt = e (σ2 −σ1 )t/2 S1 (t ), t ⩾ 0.
Show that the discounted process e −rt Xt is a martingale under P.
c) Taking Nt = S2 (t ) as numéraire, show that the forward process Xb (t ) = Xt /Nt is a martingale
under the forward measure P b defined by the Radon-Nikodym density
dP
b NT
= e −rT .
dP N0
Recall that
Wbt := Wt − σ2t
b
d) Using the relation
( S (

T ) − S ( T )) +
−rT ∗ + 1 2
e IE [(S1 (T ) − S2 (T )) ] = N0 Ib
E ,
NT
compute the price
e −rT IE∗ [(S1 (T ) − S2 (T ))+ ]
of the exchange option on the assets S1 and S2 .
Exercise 3.9 Compute the price e −(T −t )r IE∗ 1{RT ⩾κ} Ft at time t ∈ [0, T ] of a cash-or-nothing

“binary” foreign exchange call option with maturity T and strike price κ on the foreign exchange
rate process (Rt )t∈R+ given by
dRt = (rl − rf )Rt dt + σ Rt dWt ,
where (Wt )t∈R+ is a standard Brownian motion under P∗ .
Hint: We have the relation
!
∗ µ − log(κ/x)
P xe X +µ
⩾κ =Φ

p
Var[X ]
for X ≃ N (0, Var[X ]) a centered Gaussian random variable.
Exercise 3.10 Extension of Proposition 3.8 to correlated Brownian motions. Assume that (St )t∈R+
and (Nt )t∈R+ satisfy the stochastic differential equations
dSt = rt St dt + σtS St dWtS , and dNt = ηt Nt dt + σtN Nt dWtN ,
where (WtS )t∈R+ and (WtN )t∈R+ have the correlation
dWtS • dWtN = ρdt,
where ρ ∈ [−1, 1].

Pricing of Interest Rate Derivatives
92
a) Show that (WtN )t∈R+ can be written as
p
WtN = ρWtS + 1 − ρ 2Wt , t ⩾ 0,
where (Wt )t∈R+ is a standard Brownian motion under P∗ , independent of (WtS )t∈R+ .
b) Letting Xt = St /Nt , show that dXt can be written as
dXt = (rt − ηt + (σtN )2 − ρσtN σtS )Xt dt + σ

bt Xt dWtX ,
where (WtX )t∈R+ is a standard Brownian motion under P∗ and σbt is to be computed.
Exercise 3.11 Quanto options (Exercise 9.5 in Shreve, 2004). Consider an asset priced St at time t,
with
dSt = rSt dt + σ S St dWtS ,
and an exchange rate (Rt )t∈R+ given by
dRt = (r − rf )Rt dt + σ R Rt dWtR ,
from (3.3.3) in Proposition 3.10, where (WtR )t∈R+ is written as

p
WtR = ρWtS + 1 − ρ 2Wt , t ⩾ 0,
where (Wt )t∈R+ is a standard Brownian motion under P∗ , independent of (WtS )t∈R+ , i.e., we have
dWtR • dWtS = ρdt,
where ρ is a correlation coefficient.

a) Let
a = rl − rf + ρσ R σ S − (σ R )2
and Xt = e at St /Rt , t ⩾ 0, and show by Exercise 3.10 that dXt can be written as
b Xt dWtX ,
dXt = rXt dt + σ
where (WtX )t∈R+ is a standard Brownian motion under P∗ and σb is to be determined.

b) Compute the price
+
−(T −t )r ∗ ST
e IE −κ Ft
RT
of the quanto option at time t ∈ [0, T ].

93
4. Pricing of Interest Rate Derivatives
Interest rate derivatives are option contracts whose payoffs can be based on fixed-income securities
such as bonds, or on cash flows exchanged in e.g. interest rate swaps. In this chapter we consider
the pricing and hedging of interest rate and fixed income derivatives such as bond options, caplets,
caps and swaptions, using the change of numéraire technique and forward measures.
4.2 Bond Options 97
Exercises 112
4.1 Forward Measures and Tenor Structure

The maturity dates are arranged according to a discrete tenor structure
{0 = T0 < T1 < T2 < · · · < Tn }.
A sample of forward interest rate curve data is given in Table 4.1, which contains the values
of (T1 , T2 , . . . , T23 ) and of { f (t,t + Ti ,t + Ti + δ )}i=1,2,...,23 , with t = 07/05/2003 and δ = six
months.

94 Chapter 4. Pricing of Interest Rate Derivatives
Maturity 2D 1W 1M 2M 3M 1Y 2Y 3Y 4Y 5Y 6Y 7Y
Rate (%) 2.55 2.53 2.56 2.52 2.48 2.34 2.49 2.79 3.07 3.31 3.52 3.71
Maturity 8Y 9Y 10Y 11Y 12Y 13Y 14Y 15Y 20Y 25Y 30Y
Rate (%) 3.88 4.02 4.14 4.23 4.33 4.40 4.47 4.54 4.74 4.83 4.86
Table 4.1: Forward rates arranged according to a tenor structure.
Recall that by definition of P(t, Ti ) and absence of arbitrage the discounted bond price process
rt
t 7→ e − 0 rs ds P(t, Ti ), 0 ⩽ t ⩽ Ti ,
is an Ft -martingale under the probability measure P∗ = P, hence it satisfies the Assumption 1 on
page 67 for i = 1, 2, . . . , n. As a consequence the bond price process can be taken as a numéraire
(i)
Nt := P(t, Ti ), 0 ⩽ t ⩽ Ti ,
in the definition
dP
bi 1 rT
− 0 i rs ds
= e (4.1.1)
dP∗ P(0, Ti )
of the forward measure P b i , see Definition 3.1. The following proposition will allow us to price
contingent claims using the forward measure P b i , it is a direct consequence of Proposition 3.5,
noting that here we have P(Ti , Ti ) = 1.
Proposition 4.1 For all sufficiently integrable random variables C we have

h rT i
i
IE∗ C e − t rs ds Ft = P(t, Ti )IE
b i [C | Ft ], 0 ⩽ t ⩽ Ti , i = 1, 2, . . . , n. (4.1.2)
Recall that by Proposition 3.4, the deflated process

P(t, T j )
t 7→ , 0 ⩽ t ⩽ min(Ti , T j ),
P(t, Ti )
is an Ft -martingale under P
b i for all Ti , T j ⩾ 0, i, j = 1, 2, . . . , n.
In the sequel we assume as in (1.4.8) that the dynamics of the bond price P(t, Ti ) is given by
dP(t, Ti )
= rt dt + ζi (t )dWt , i = 1, 2, . . . , n, (4.1.3)
P(t, Ti )
see e.g. (1.4.11) in the Vasicek case, where (Wt )t∈R+ is a standard Brownian motion under P∗ and
(rt )t∈R+ and (ζi (t ))t∈R+ are adapted processes with respect to the filtration (Ft )t∈R+ generated
by (Wt )t∈R+ , i.e.
w wt 1wt

t
2
P(t, Ti ) = P(0, Ti ) exp rs ds + ζi (s)dWs − |ζi (s)| ds ,
0 0 2 0
0 ⩽ t ⩽ Ti , i = 1, 2, . . . , n.
Forward Brownian motions

Proposition 4.2 For all i = 1, 2, . . . , n, the process

wt
bt(i) := Wt −
W ζi (s)ds, 0 ⩽ t ⩽ Ti , (4.1.4)
0
is a standard Brownian motion under the forward measure P

b i.
Proof. The Girsanov Proposition 3.7 applied to the numéraire

(i)
Nt := P(t, Ti ), 0 ⩽ t ⩽ Ti ,
as in (3.2.13), shows that
(i) 1 (i)
dW
bt := dWt − (i)
dNt • dWt
Nt
1
= dWt − dP(t, Ti ) • dWt
P(t, Ti )
1
= dWt − (P(t, Ti )rt dt + ζi (t )P(t, Ti )dWt ) • dWt
P(t, Ti )
= dWt − ζi (t )dt,

b i for all i = 1, 2, . . . , n. □
We have
(i)
dW
bt = dWt − ζi (t )dt, i = 1, 2, . . . , n, (4.1.5)
and
( j) (i)
dW
bt = dWt − ζ j (t )dt = dW
bt + (ζi (t ) − ζ j (t ))dt, i, j = 1, 2, . . . , n,
bt( j) )t∈R+ has drift (ζi (t ) − ζ j (t ))t∈R+ under P
which shows that (W b i.
Bond price dynamics under the forward measure

In order to apply Proposition 4.1 and to compute the price
h r Ti i
IE∗ e − t rs dsC Ft = P(t, Ti )Ib
Ei [C | Ft ],
of a random claim payoff C, it can be useful to determine the dynamics of the underlying variables rt ,
f (t, T , S), and P(t, T ) via their stochastic differential equations written under the forward measure
Pi .
b
As a consequence of Proposition 4.2 and (4.1.3), the dynamics of t 7→ P(t, T j ) under P b i is given by
dP(t, T j ) bt(i) ,
= rt dt + ζi (t )ζ j (t )dt + ζ j (t )dW i, j = 1, 2, . . . , n, (4.1.6)
P(t, T j )
(i)
bt )t∈R+ is a standard Brownian motion under P
where (W b i , and we have
P(t, T j )
w wt 1wt

t
= P(0, T j ) exp rs ds + ζ j (s)dWs − |ζ j (s)|2 ds [under P∗ ]
0 0 2 0

w wt 1wt

t ( j)
= P(0, T j ) exp rs ds + ζ j (s)dWs +
b 2
|ζ j (s)| ds [under P b j]
0 0 2 0
w wt wt wt
t ( i ) 1
= P(0, T j ) exp bs + ζ j (s)ζi (s)ds −
rs ds + ζ j (s)dW |ζ j (s)| ds [under P
2 b i]
0 0 0 2 0
w wt 1wt 1wt

t (i) 2 2
= P(0, T j ) exp rs ds + ζ j (s)dWs −
b |ζ j (s) − ζi (s)| ds + |ζi (s)| ds ,
0 0 2 0 2 0
t ∈ [0, T j ], i, j = 1, 2, . . . , n. Consequently, the forward price P(t, T j )/P(t, Ti ) can be written as
P(t, T j )
P(t, Ti )
w
1wt

P(0, T j ) t ( j)
= exp (ζ j (s) − ζi (s))dWs +
b |ζ j (s) − ζi (s)| ds , [under P
2 b j]
P(0, Ti ) 0 2 0
w wt
P(0, T j ) t
bs(i) − 1 |ζi (s) − ζ j (s)|2 ds , [under P
= exp (ζ j (s) − ζi (s))dW b i]
P(0, Ti ) 0 2 0
(4.1.7)
t ∈ [0, min(Ti , T j )], i, j = 1, 2, . . . , n, which also follows from Proposition 3.8.
Short rate dynamics under the forward measure

In case the short rate process (rt )t∈R+ is given as the (Markovian) solution to the stochastic
differential equation
drt = µ (t, rt )dt + σ (t, rt )dWt ,
by (4.1.5) its dynamics will be given under P

b i by
bt(i)

drt = µ (t, rt )dt + σ (t, rt ) ζi (t )dt + dW
bt(i) .
= µ (t, rt )dt + σ (t, rt )ζi (t )dt + σ (t, rt )dW (4.1.8)
In the case of the Vašíček, 1977 model, by (1.4.11) we have
drt = (a − brt )dt + σ dWt ,
and
σ
1 − e −b(Ti −t ) ,

ζi (t ) = − 0 ⩽ t ⩽ Ti ,
b
hence from (4.1.8) we have
(i) σ
1 − e −b(Ti −t ) dt,

dW
bt = dWt − ζi (t )dt = dWt + (4.1.9)
b
and
σ2 bt(i)
1 − e −b(Ti −t ) dt + σ dW

drt = (a − brt )dt − (4.1.10)
b
and we obtain
dP(t, Ti ) σ2 2 σ (i)
= rt dt + 2 1 − e −b(Ti −t ) dt − 1 − e −b(Ti −t ) dW
bt ,
P(t, Ti ) b b
from (1.4.11).

4.2 Bond Options 97
4.2 Bond Options

The next proposition can be obtained as an application of the Margrabe formula (3.4.4) of Propo-
(i) (i)
sition 3.15 by taking Xt = P(t, T j ), Nt = P(t, Ti ), and Xbt = Xt /Nt = P(t, T j )/P(t, Ti ). In the
Vasicek model, this formula has been first obtained in Jamshidian, 1989.
We work with a standard Brownian motion (Wt )t∈R+ under P∗ , generating the filtration (Ft )t∈R+ ,
and an (Ft )t∈R+ -adapted short rate process (rt )t∈R+ .
Proposition 4.3 Let 0 ⩽ Ti ⩽ T j and assume as in (1.4.8) that the dynamics of the bond prices
P(t, Ti ), P(t, T j ) under P∗ are given by
dP(t, Ti ) dP(t, T j )
= rt dt + ζi (t )dWt , = rt dt + ζ j (t )dWt ,
P(t, Ti ) P(t, T j )
where (ζi (t ))t∈R+ and (ζ j (t ))t∈R+ are deterministic volatility functions. Then, the price of a
bond call option on P(Ti , T j ) with payoff
C := (P(Ti , T j ) − κ )+
can be written as
h r Ti i
IE∗ e − t rs ds (P(Ti , T j ) − κ )+ Ft (4.2.1)

v(t, Ti ) 1 P(t, T j )
= P(t, T j )Φ + log
2 v(t, Ti ) κP(t, Ti )

v(t, Ti ) 1 P(t, T j )
−κP(t, Ti )Φ − + log ,
2 v(t, Ti ) κP(t, Ti )
w Ti
where v2 (t, Ti ) := |ζi (s) − ζ j (s)|2 ds and
t
1 w x −y2 /2
Φ (x ) : = √ e dy, x ∈ R,
2π −∞
is the Gaussian cumulative distribution function.
(i)
Proof. First, we note that using Nt := P(t, Ti ) as a numéraire the price of a bond call option on
P(Ti , T j ) with payoff F = (P(Ti , T j ) − κ )+ can be written from Proposition 3.5 using the forward
measure P b i , or directly by (3.2.9), as
h r Ti i
IE∗ e − t rs ds (P(Ti , T j ) − κ )+ Ft = P(t, Ti )Ib
Ei (P(Ti , T j ) − κ )+ Ft .

(4.2.2)
Next, by (4.1.7) or by solving (3.2.15) in Proposition 3.8 we can write P(Ti , T j ) as the geometric
Brownian motion
P(Ti , T j )
P(Ti , T j ) =
P(Ti , Ti )
w wT
P(t, T j ) Ti
bs(i) − 1 i |ζi (s) − ζ j (s)|2 ds ,
= exp (ζ j (s) − ζi (s))dW
P(t, Ti ) t 2 t

under the forward measure P b i , and rewrite (4.2.2) as

h r Ti i
IE∗ e − t rs ds (P(Ti , T j ) − κ )+ Ft
" + #
P ( t, T ) rT ( i ) 1
rT 2
j i i
= P(t, Ti )IE e t (ζ j (s)−ζi (s))dWs − 2 t |ζi (s)−ζ j (s)| ds − κ Ft
bi b
P(t, Ti )
rT rT +

(i)
= IE b i P(t, T j ) e t i (ζ j (s)−ζi (s))dWbs − 12 t i |ζi (s)−ζ j (s)|2 ds − κP(t, Ti ) Ft .
Since (ζi (s))s∈[0,Ti ] and (ζ j (s))s∈[0,Tj ] in (4.1.3) are deterministic volatility functions, P(Ti , T j ) is a
lognormal random variable given Ft under P b i and as in Proposition 3.15 we can price the bond
option by the zero-rate Black-Scholes formula
√
Bl P(t, T j ), κP(t, Ti ), v(t, Ti )/ Ti − t, 0, Ti − t
with underlying asset price P(t, T j ), strike level κP(t, Ti ), volatility parameter
sr
Ti 2
v(t, T ) t |ζi (s) − ζ j (s)| ds
√ i = ,
Ti − t Ti − t
time to maturity Ti − t, and zero interest rate, which yields (4.2.1). □
Note that from Corollary 3.17 the decomposition (4.2.1) gives the self-financing portfolio in the
assets P(t, Ti ) and P(t, T j ) for the claim with payoff (P(Ti , T j ) − κ )+ .
In the
Vasicek case
the above bond option price could also be computed from the joint distribution
rT
of rT , t rs ds , which is Gaussian, or from the dynamics (4.1.6)-(4.1.10) of P(t, T ) and rt under
P
b i , see Kim, 2002 and § 8.3 of Privault, 2021.
4.3 Caplet Pricing

An interest rate caplet is an option contract that offers protection against the fluctuations of a
variable (or floating) rate with respect to a fixed rate κ. The payoff of a LIBOR caplet on the yield
(or spot forward rate) L(Ti , Ti , Ti+1 ) with strike level κ can be written as
(L(Ti , Ti , Ti+1 ) − κ )+ ,
and priced at time t ∈ [0, Ti ] from Proposition 3.5 using the forward measure P
b i+1 as
h r Ti+1 i
IE∗ e − t rs ds (L(Ti , Ti , Ti+1 ) − κ )+ Ft (4.3.1)
b i+1 (L(Ti , Ti , Ti+1 ) − κ )+ | Ft ,

= P(t, Ti+1 )IE
(i+1)
by taking Nt = P(t, Ti+1 ) as a numéraire.
Proposition 4.4 The LIBOR rate

1 P(t, Ti )
L(t, Ti , Ti+1 ) := −1 , 0 ⩽ t ⩽ Ti < Ti+1 ,
Ti+1 − Ti P(t, Ti+1 )
is a martingale under the forward measure P

b i+1 defined in (4.1.1).
Proof. The LIBOR rate L(t, Ti , Ti+1 ) is a deflated process according to the forward numéraire
process (P(t, Ti+1 ))t∈[0,Ti+1 ] . Therefore, by Proposition 3.4 it is a martingale under P
b i+1 . □

The caplet on L(Ti , Ti , Ti+1 ) can be priced at time t ∈ [0, Ti ] as

r
T
∗ − t i+1 rs ds
IE e (L(Ti , Ti , Ti+1 ) − κ ) Ft
+
(4.3.2)
" + #
rT

∗ − t i+1 rs ds 1 P(t, Ti )
= IE e −1 −κ Ft ,
Ti+1 − Ti P(t, Ti+1 )
where the discount factor is counted from the settlement date Ti+1 . The next pricing formula (4.3.4)
allows us to price and hedge a caplet using a portfolio based on the bonds P(t, Ti ) and P(t, Ti+1 ), cf.
(4.3.8) below, when L(t, Ti , Ti+1 ) is modeled in the BGM model of Section 2.6.
Proposition 4.5 (Black LIBOR caplet formula). Assume that L(t, Ti , Ti+1 ) is modeled in the
BGM model as
dL(t, Ti , Ti+1 ) bti+1 ,
= γi (t )dW (4.3.3)
L(t, Ti , Ti+1 )
0 ⩽ t ⩽ Ti , i = 1, 2, . . . , n − 1, where γi (t ) is a deterministic volatility function of time t ∈ [0, Ti ],

i = 1, 2, . . . , n − 1. The caplet on L(Ti , Ti , Ti+1 ) with strike level κ is priced at time t ∈ [0, Ti ] as
h r Ti+1 i
(Ti+1 − Ti ) IE∗ e − t rs ds (L(Ti , Ti , Ti+1 ) − κ )+ Ft (4.3.4)
= (P(t, Ti ) − P(t, Ti+1 ))Φ(d+ (t, Ti )) − κ (Ti+1 − Ti )P(t, Ti+1 )Φ(d− (t, Ti )),
0 ⩽ t ⩽ Ti , where
log(L(t, Ti , Ti+1 )/κ ) + (Ti − t )σi2 (t, Ti )/2

d+ (t, Ti ) = √ , (4.3.5)
σi (t, Ti ) Ti − t
and
log(L(t, Ti , Ti+1 )/κ ) − (Ti − t )σi2 (t, Ti )/2
d− (t, Ti ) = √ , (4.3.6)
σi (t, Ti ) Ti − t
and
1 w Ti 2
|σi (t, Ti )|2 = |γi | (s)ds. (4.3.7)
Ti − t t
Proof. Taking P(t, Ti+1 ) as a numéraire, the forward price
P(t, Ti )
Xbt := = 1 + (Ti+1 − Ti )L(Ti , Ti , Ti+1 )
P(t, Ti+1 )
and the forward LIBOR rate process (L(t, Ti , Ti+1 )t∈[0,Ti ] are martingales under P b i+1 by Proposi-
tion 4.4, i = 1, 2, . . . , n − 1. More precisely, by (4.3.3) we have
w
1 w Ti

Ti
i+1 2
L(Ti , Ti , Ti+1 ) = L(t, Ti , Ti+1 ) exp γi (s)dWs −
b |γi (s)| ds ,
t 2 t
0 ⩽ t ⩽ Ti , i.e. t 7→ L(t, Ti , Ti+1 ) is a geometric Brownian motion with time-dependent volatility

b i+1 . Hence by (4.3.1), since N (i+1) = 1, we have

γi (t ) under P Ti+1
h r Ti+1 i
IE∗ e − t rs ds (L(Ti , Ti , Ti+1 ) − κ )+ Ft
b i+1 (L(Ti , Ti , Ti+1 ) − κ )+ Ft

= P(t, Ti+1 )IE
= P(t, Ti+1 ) (L(t, Ti , Ti+1 )Φ(d+ (t, Ti )) − κΦ(d− (t, Ti )))
= P(t, Ti+1 )Bl(L(t, Ti , Ti+1 ), κ, σi (t, Ti ), 0, Ti − t ),
t ∈ [0, Ti ], where
Bl(x, κ, σ , 0, τ ) = xΦ(d+ (t, Ti )) − κΦ(d− (t, Ti ))
is the zero-interest rate Black-Scholes function, with
1 w Ti 2
|σi (t, Ti )|2 = |γi | (s)ds.
Ti − t t
Therefore, we obtain
h r Ti+1 i
(Ti+1 − Ti ) IE∗ e − t rs ds (L(Ti , Ti , Ti+1 ) − κ )+ Ft
= (Ti+1 − Ti )P(t, Ti+1 )L(t, Ti , Ti+1 )Φ(d+ (t, Ti )) − (Ti+1 − Ti )κP(t, Ti+1 )Φ(d− (t, Ti ))

P(t, Ti )
= P(t, Ti+1 ) − 1 Φ(d+ (t, Ti ))
P(t, Ti+1 )
−κ (Ti+1 − Ti )P(t, Ti+1 )Φ(d− (t, Ti )),
which yields (4.3.4). □

In addition, from Corollary 3.17 we obtain the self-financing portfolio strategy
(Φ(d+ (t, Ti )), −Φ(d+ (t, Ti )) − κ (Ti+1 − Ti )Φ(d− (t, Ti ))) (4.3.8)
in the bonds priced (P(t, Ti ), P(t, Ti+1 )) with maturities Ti and Ti+1 , cf. Corollary 3.18 and Privault
and Teng, 2012.
The formula (4.3.4) can be applied to options on underlying futures or forward contracts on
commodities whose prices are modeled according to (4.3.3), as in the next corollary.
Corollary 4.6 (Black, 1976 formula). Let L(t, Ti , Ti+1 ) be modeled as in (4.3.3) and let the
bond price P(t, Ti+1 ) be given as P(t, Ti+1 ) = e −(Ti+1 −t )r . Then, (4.3.4) becomes
e −(Ti+1 −t )r L(t, Ti , Ti+1 )Φ(d+ (t, Ti )) − κ e −(Ti+1 −t )r Φ(d− (t, Ti )),
0 ⩽ t ⩽ Ti .
Floorlet pricing
The floorlet on L(Ti , Ti , Ti+1 ) with strike level κ is a contract with payoff (κ − L(Ti , Ti , Ti+1 ))+ .
Floorlets are analog to put options and can be similarly priced by the call/put parity in the Black-
Scholes formula.
Proposition 4.7 Assume that L(t, Ti , Ti+1 ) is modeled in the BGM model as in (4.3.3). The
floorlet on L(Ti , Ti , Ti+1 ) with strike level κ is priced at time t ∈ [0, Ti ] as

h r Ti+1 i
(Ti+1 − Ti ) IE∗ e − t rs ds (κ − L(Ti , Ti , Ti+1 ))+ Ft (4.3.9)
= κ (Ti+1 − Ti )P(t, Ti+1 )Φ − d− (t, Ti ) − (P(t, Ti ) − P(t, Ti+1 ))Φ − d+ (t, Ti ) ,

0 ⩽ t ⩽ Ti , where d+ (t, Ti ), d− (t, Ti ) and |σi (t, Ti )|2 are defined in (4.3.5)-(4.3.7).
Proof. We have
h r Ti+1 i
(Ti+1 − Ti ) IE∗ e − t rs ds (κ − L(Ti , Ti , Ti+1 ))+ Ft
b i+1 (κ − L(Ti , Ti , Ti+1 ))+ | Ft

= (Ti+1 − Ti )P(t, Ti+1 )IE
= (Ti+1 − Ti )P(t, Ti+1 ) κΦ − d− (t, Ti ) − L(t, Ti , Ti+1 )Φ − d+ (t, Ti )

= (Ti+1 − Ti )P(t, Ti+1 )κΦ − d− (t, Ti ) − (P(t, Ti ) − P(t, Ti+1 ))Φ − d+ (t, Ti ) ,

0 ⩽ t ⩽ Ti . □
Cap pricing
More generally, one can consider interest rate caps that are relative to a given tenor structure
{T1 , T2 , . . . , Tn }, with discounted payoff
j−1 r Tk+1
∑ (Tk+1 − Tk ) e − t rs ds
(L(Tk , Tk , Tk+1 ) − κ )+ .
k =i
Pricing formulas for interest rate caps are easily deduced from analog formulas for caplets, since
the payoff of a cap can be decomposed into a sum of caplet payoffs. Thus, the cap price at time
t ∈ [0, Ti ] is given by
" #
j−1 r Tk+1
IE∗ ∑ (Tk+1 − Tk ) e − t rs ds
(L(Tk , Tk , Tk+1 ) − κ )+ Ft
k =i
j−1 r Tk+1

∗ −
(L(Tk , Tk , Tk+1 ) − κ ) +
Ft
k =i
j−1
(L(Tk , Tk , Tk+1 ) − κ )+ Ft .

= ∑ (Tk+1 − Tk )P(t, Tk+1 )IEb k+1
k =i
(4.3.10)
In the BGM model (4.3.3) the interest rate cap with payoff
j−1
∑ (Tk+1 − Tk )(L(Tk , Tk , Tk+1 ) − κ )+
k =i
can be priced at time t ∈ [0, T1 ] by the Black formula

j−1
∑ (Tk+1 − Tk )P(t, Tk+1 )Bl(L(t, Tk , Tk+1 ), κ, σk (t, Tk ), 0, Tk − t ),
k =i
where
1 w Tk
|σk (t, Tk )|2 = |γk |2 (s)ds.
Tk − t t

SOFR Caplets
The backward-looking SOFR caplet has payoff (R(S, T , S) − K )+ , which is known only at time S.
By the Jensen, 1906 inequality we note the relation
IES [(R(S, T , S) − K )+ | Ft ] = IES [IES [(R(S, T , S) − K )+ | FT ] | Ft ]

⩾ IES [(IES [R(S, T , S) | FT ] − K )+ | Ft ]
= IES [(R(T , T , S) − K )+ | Ft ]
= IES [(L(T , T , S) − K )+ | Ft ],
hence the backward-looking SOFR caplet is more expensive than the forward-looking LIBOR
caplet. The caplet on the SOFR rate R(Ti+1 , Ti , Ti+1 ) with payoff (R(Ti+1 , Ti , Ti+1 ) − κ )+ and strike
level κ can be priced at time t ∈ [0, Ti ] with a discount factor counted from the settlement date Ti+1
from Proposition 3.5 as
h r Ti+1 i
∗ − t
IE e rs ds
(R(Ti+1 , Ti , Ti+1 ) − κ ) Ft
+
(4.3.11)
" #
r Ti+1
+
1 P(t, Ti )
= IE∗ e − t rs ds −1 −κ Ft
Ti+1 − Ti P(t, Ti+1 )
b i+1 (R(Ti+1 , Ti , Ti+1 ) − κ )+ | Ft ,

= P(t, Ti+1 )IE
(i+1)
by taking Nt := P(t, Ti+1 ) as a numéraire and using the forward measure P
b i+1 .
Proposition 4.8 The SOFR rate

1 P(t, Ti )
R(t, Ti , Ti+1 ) := −1 , 0 ⩽ Ti ⩽ t ⩽ Ti+1 ,
Ti+1 − Ti P(t, Ti+1 )
is a martingale under the forward measure P

b i+1 .
Proof. The SOFR rate R(t, Ti , Ti+1 ) is a deflated process according to the forward numéraire
process (P(t, Ti+1 ))t∈[0,Ti+1 ] . Therefore, it is a martingale under P
b i+1 by Proposition 3.4. □
The next pricing formula (4.3.13) allows us to price and hedge a caplet using a portfolio based on
the bonds P(t, Ti ) and P(t, Ti+1 ), cf. (4.3.14) below, when R(t, Ti , Ti+1 ) is modeled in the BGM
model.
Proposition 4.9 (Black SOFR caplet formula). Assume that R(t, Ti , Ti+1 ) is modeled in the
BGM model as
dR(t, Ti , Ti+1 ) bti+1 ,
= γi (t )dW (4.3.12)
R(t, Ti , Ti+1 )
0 ⩽ t ⩽ Ti+1 , i = 1, 2, . . . , n − 1, where γi (t ) is a deterministic volatility function of time t ∈

[0, Ti+1 ], i = 1, 2, . . . , n − 1. The caplet on R(Ti+1 , Ti , Ti+1 ) with strike level κ > 0 is priced at
time t ∈ [0, Ti+1 ] as
h r Ti+1 i
(Ti+1 − Ti ) IE∗ e − t rs ds (R(Ti+1 , Ti , Ti+1 ) − κ )+ Ft (4.3.13)
= (P(t, Ti ) − P(t, Ti+1 ))Φ(d+ (t, Ti+1 )) − κ (Ti+1 − Ti )P(t, Ti+1 )Φ(d− (t, Ti+1 )),

0 ⩽ t ⩽ Ti+1 , where
log(R(t, Ti , Ti+1 )/κ ) + (Ti+1 − t )σi2 (t, Ti+1 )/2

d+ (t, Ti+1 ) = √ ,
σi (t, Ti+1 ) Ti+1 − t
and
log(R(t, Ti , Ti+1 )/κ ) − (Ti+1 − t )σi2 (t, Ti+1 )/2
d− (t, Ti+1 ) = √ ,
σi (t, Ti+1 ) Ti+1 − t
and
1 w Ti+1
|σi (t, Ti+1 )|2 = |γi |2 (s)ds.
Ti+1 − t t
Proof. The forward price
P(t, Ti )
Xbt := = 1 + (Ti+1 − Ti )R(Ti+1 , Ti , Ti+1 )
P(t, Ti+1 )
and the SOFR rate process (R(t, Ti , Ti+1 )t∈[0,Ti+1 ] are martingales under P b i+1 by Proposition 4.8,
i = 1, 2, . . . , n − 1, and
w wT
bsi+1 − 1 i+1 |γi (s)|2 ds ,
Ti+1
R(Ti+1 , Ti , Ti+1 ) = R(t, Ti , Ti+1 ) exp γi (s)dW
t 2 t
0 ⩽ t ⩽ Ti+1 , where t 7→ R(t, Ti , Ti+1 ) is a geometric Brownian motion under P

b i+1 (4.3.12). Hence
by (4.3.11) we have
h r Ti+1 i
IE∗ e − t rs ds (R(Ti+1 , Ti , Ti+1 ) − κ )+ Ft
b i+1 (R(Ti+1 , Ti , Ti+1 ) − κ )+ Ft

= P(t, Ti+1 )IE
= P(t, Ti+1 ) (R(t, Ti , Ti+1 )Φ(d+ (t, Ti+1 )) − κΦ(d− (t, Ti+1 )))
= P(t, Ti+1 )Bl(R(t, Ti , Ti+1 ), κ, σi (t, Ti+1 ), 0, Ti+1 − t ),
t ∈ [0, Ti+1 ], with
1 w Ti+1
|σi (t, Ti+1 )|2 = |γi |2 (s)ds.
Ti+1 − t t
□
In addition, we obtain the self-financing portfolio strategy
(Φ(d+ (t, Ti+1 )), −Φ(d+ (t, Ti+1 )) − κ (Ti+1 − Ti )Φ(d− (t, Ti+1 ))) (4.3.14)
in the bonds priced (P(t, Ti ), P(t, Ti+1 )), t ∈ [0, Ti+1 ], with maturities Ti and Ti+1 .
4.4 Forward Swap Measures

In this section we introduce the forward swap (or annuity) measures, or annuity measures, to be
used for the pricing of swaptions, and we study their properties. We start with the definition of the
annuity numéraire
j−1
(i, j )
Nt := P(t, Ti , T j ) = ∑ (Tk+1 − Tk )P(t, Tk+1 ), 0 ⩽ t ⩽ Ti , (4.4.1)
k =i

with in particular, when j = i + 1,
P(t, Ti , Ti+1 ) = (Ti+1 − Ti )P(t, Ti+1 ), 0 ⩽ t ⩽ Ti .
1 ⩽ i < n. The annuity numéraire can be also used to price a bond ladder. It satisfies
rt
the following
− 0 rs ds
martingale property, which can be proved by linearity and the fact that t 7→ e P(t, Tk ) is a
martingale for all k = 1, 2, . . . , n, under Assumption 1.
Remark 4.10 The discounted annuity numéraire
rt rt j−1
t 7→ e − 0 rs ds P(t, Ti , T j ) = e − 0 rs ds
∑ (Tk+1 − Tk )P(t, Tk+1 ), 0 ⩽ t ⩽ Ti ,
k =i
is a martingale under P∗ .
The forward swap measure P

b i, j is defined, according to Definition 3.1, by
(i, j )
dP
b i, j rT N
− 0 i rs ds Ti
rT
− 0 i rs ds P(Ti , Ti , T j )
: = e = e , (4.4.2)
dP∗ N
(i, j ) P(0, Ti , T j )
0
1 ⩽ i < j ⩽ n.
Remark 4.11 By (3.2.2) we have
" #
dP
b i, j 1 h r Ti i
IE∗ Ft = IE∗ e − 0 rs ds P(Ti , Ti , T j ) Ft
dP∗ P(0, Ti , T j )
" #
r Ti j−1
1
= IE∗ e − 0 rs ds ∑ (Tk+1 − Tk )P(Ti , Tk+1 ) Ft
P(0, Ti , T j ) k =i
j−1 h r Ti
1 ∗ − 0 rs ds
i
= ( T − T ) I
E e P ( T , T ) F
P(0, Ti , T j ) k∑
k +1 k i k +1 t
=i
rt j−1
1
= e − 0 rs ds ∑ (Tk+1 − Tk )P(t, Tk+1 )
P(0, Ti , T j ) k =i
rt P(t, Ti , T j )
= e− 0 rs ds ,
P(0, Ti , T j )
0 ⩽ t ⩽ Ti , see Remark 4.10, and
dP
b i, j|F rT
i P(Ti , Ti , T j )
∗
t
= e − t rs ds , 0 ⩽ t ⩽ Ti+1 , (4.4.3)
dP|Ft P(t, Ti , T j )
by Relation (3.2.3) in Lemma 3.2.
Proposition 4.12 The LIBOR swap rate
P(t, Ti ) − P(t, T j )
S(t, Ti , T j ) = , 0 ⩽ t ⩽ Ti ,
P(t, Ti , T j )
see Corollary 2.12, is a martingale under the forward swap measure P

b i, j .

Proof. We use the fact that the deflated process

P(t, Tk )
t 7→ , i, j, k = 1, 2, . . . , n,
P(t, Ti , T j )
is an Ft -martingale under P
b i, j by Proposition 3.4. □
The following pricing formula is then stated for a given integrable claim with payoff of the form
P(Ti , Ti , T j )F, using the forward swap measure Pb i, j :
" #
P
r
Ti d b i, j|F
IE∗ e − t rs ds P(Ti , Ti , T j )F Ft = P(t, Ti , T j ) IE∗ F t
Ft
dP∗|Ft
= P(t, Ti , T j )IbEi, j [F | Ft ], (4.4.4)
after applying (4.4.2) and (4.4.3) on the last line, or Proposition 3.5.
4.5 Swaption Pricing
Definition 4.13 A payer (or call) swaption gives the option, but not the obligation, to enter an
interest rate swap as payer of a fixed rate κ and as receiver of floating LIBOR rates L(Ti , Tk , Tk+1 )
at time Tk+1 , k = i, . . . , j − 1, and has the payoff
!+
j−1 r Tk+1
∑ (Tk+1 − Tk ) IE∗ e − Ti rs ds
FTi (L(Ti , Tk , Tk+1 ) − κ )
k =i
!+
j−1
= ∑ (Tk+1 − Tk )P(Ti , Tk+1 )(L(Ti , Tk , Tk+1 ) − κ ) (4.5.1)
k =i
at time Ti .
This swaption can be priced at time t ∈ [0, Ti ] under the risk-neutral probability measure P∗ as
" !+ #
rT j−1
i
IE∗ e − t rs ds
∑ (Tk+1 − Tk )P(Ti , Tk+1 )(L(Ti , Tk , Tk+1 ) − κ ) Ft , (4.5.2)
k =i
t ∈ [0, Ti ]. When j = i + 1, the swaption price (4.5.2) coincides with the price at time t of a caplet
on [Ti , Ti+1 ] up to a factor δi := Ti+1 − Ti , since
h r Ti i
+
IE∗ e − t rs ds ((Ti+1 − Ti )P(Ti , Ti+1 )(L(Ti , Ti , Ti+1 ) − κ )) Ft
h r Ti i
= (Ti+1 − Ti ) IE∗ e − t rs ds P(Ti , Ti+1 ) (L(Ti , Ti , Ti+1 ) − κ )+ Ft
r rT
Ti i+1
= (Ti+1 − Ti ) IE∗ e − t rs ds IE∗ e − Ti rs ds FTi (L(Ti , Ti , Ti+1 ) − κ )+ Ft
r
T
r Ti+1
∗ ∗ − t i rs ds − Ti rs ds +
= (Ti+1 − Ti ) IE IE e e (L(Ti , Ti , Ti+1 ) − κ ) FTi Ft
h r Ti+1 i
= (Ti+1 − Ti ) IE∗ e − t rs ds (L(Ti , Ti , Ti+1 ) − κ )+ Ft , (4.5.3)
0 ⩽ t ⩽ Ti , which coincides with the caplet price (4.3.1) up to the factor Ti+1 − Ti . Unlike in the
case of interest rate caps, the sum in (4.5.2) cannot be taken out of the positive part. Nevertheless,
the price of the swaption can be bounded as in the next proposition.

Proposition 4.14 The payer swaption price (4.5.2) can be upper bounded by the interest rate
cap price (4.3.10) as
" !+ #
rT j−1
i
IE∗ e − t rs ds
∑ (Tk+1 − Tk )P(Ti , Tk+1 )(L(Ti , Tk , Tk+1 ) − κ ) Ft
k =i
j−1 r Tk+1

∗ − +
⩽ ∑ (Tk+1 − Tk ) IE e t rs ds
(L(Ti , Tk , Tk+1 ) − κ ) Ft ,
k =i
0 ⩽ t ⩽ Ti .
Proof. Due to the inequality
(x1 + x2 + · · · + xm )+ ⩽ x1+ + x2+ + · · · + xm+ , x1 , x2 , . . . , xm ∈ R,
we have
" !+ #
rT j−1
i
IE∗ e − t rs ds
k =i
" #
rT j−1
∑ (Tk+1 − Tk )P(Ti , Tk+1 ) (L(Ti , Tk , Tk+1 ) − κ )+
i
⩽ IE∗ e − t rs ds
Ft
k =i
j−1 h r Ti i
∗ − t rs ds +
= ∑ ( Tk + 1 − Tk ) IE e P ( Ti , Tk + 1 ) ( L ( Ti , Tk , Tk + 1 ) − κ ) Ft
k =i
j−1 rT
r Tk+1
∗ − i ∗ − rs ds +
IE e Ti
FTi (L(Ti , Tk , Tk+1 ) − κ ) Ft
k =i
j−1 r
T

∗ − t k+1 rs ds ∗ +
= ∑ (Tk+1 − Tk ) IE IE e (L(Ti , Tk , Tk+1 ) − κ ) FTi Ft
k =i
j−1 r
T

∗ − t k+1 rs ds +
= ( T
∑ k +1 k − T ) IE e L (
( i k k +1
T , T , T ) − κ ) Ft
k =i
" #
j−1 r Tk+1
= IE∗ ∑ (Tk+1 − Tk ) e − t rs ds
(L(Ti , Tk , Tk+1 ) − κ )+ Ft ,
k =i
0 ⩽ t ⩽ Ti . □
The payoff of the payer swaption can be rewritten as in the following lemma which is a direct
consequence of the definition of the swap rate S(Ti , Ti , T j ), see Proposition 2.11 and Corollary 2.12.
Lemma 4.15 The payer swaption payoff (4.5.1) at time Ti with swap rate κ = S(t, T j , T j ) can be
rewritten as
!+
j−1
∑ (Tk+1 − Tk )P(Ti , Tk+1 )(L(Ti , Tk , Tk+1 ) − κ )
k =i
= (P(Ti , Ti ) − P(Ti , T j ) − κP(Ti , Ti , T j ))+ (4.5.4)
+
= P(Ti , Ti , T j ) (S(Ti , Ti , T j ) − κ ) . (4.5.5)
Proof. The relation
j−1
∑ (Tk+1 − Tk )P(t, Tk+1 )(L(t, Tk , Tk+1 ) − S(t, Ti , Tj )) = 0
k =i

that defines the forward swap rate S(t, Ti , T j ) shows that

j−1
∑ (Tk+1 − Tk )P(t, Tk+1 )L(t, Tk , Tk+1 )
k =i
j−1
= S(t, Ti , T j ) ∑ (Tk+1 − Tk )P(t, Tk+1 )
k =i
= P(t, Ti , T j )S(t, Ti , T j )
= P(t, Ti ) − P(t, T j )
as in the proof of Corollary 2.12, hence by the definition (4.4.1) of P(t, Ti , T j ) we have
j−1
∑ (Tk+1 − Tk )P(t, Tk+1 )(L(t, Tk , Tk+1 ) − κ )
k =i
= P(t, Ti ) − P(t, T j ) − κP(t, Ti , T j )
= P(t, Ti , T j ) (S(t, Ti , T j ) − κ ) ,
and for t = Ti we get
!+
j−1
∑ (Tk+1 − Tk )P(Ti , Tk+1 )(L(Ti , Tk , Tk+1 ) − κ )
k =i
= P(Ti , Ti , T j ) (S(Ti , Ti , T j ) − κ )+ .
□
The next proposition simply states that a payer swaption on the LIBOR rate can be priced as a Euro-
pean call option on the swap rate S(Ti , Ti , T j ) under the forward swap measure P
b i, j .
Proposition 4.16 The price (4.5.2) of the payer swaption with payoff
!+
j−1
∑ (Tk+1 − Tk )P(Ti , Tk+1 )(L(Ti , Tk , Tk+1 ) − κ ) (4.5.6)
k =i
on the LIBOR market can be written under the forward swap measure P b i, j as the European call
price
b i, j (S(Ti , Ti , T j ) − κ )+ Ft , 0 ⩽ t ⩽ Ti ,

P(t, Ti , T j )IE
on the swap rate S(Ti , Ti , T j ).
Proof. As a consequence of (4.4.4) and Lemma 4.15, we find

" !+ #
rT j−1
i
IE∗ e − t rs ds
k =i
h r Ti i
= IE∗ e − t rs ds (P(Ti , Ti ) − P(Ti , T j ) − κP(Ti , Ti , T j ))+ Ft (4.5.7)
h r Ti i
= IE∗ e − t rs ds P(Ti , Ti , T j ) (S(Ti , Ti , T j ) − κ )+ Ft
" #
d P
b i, j|F
= P(t, Ti , T j ) IE∗ t
(S(Ti , Ti , T j ) − κ )+ Ft
dP∗|Ft
b i, j S(Ti , Ti , T j ) − κ + Ft .

= P(t, Ti , T j )IE (4.5.8)
□

In the next Proposition 4.17 we price the payer swaption with payoff (4.5.6) or equivalently (4.5.5),
by modeling the swap rate (S(t, Ti , T j ))0⩽t⩽Ti using standard Brownian motion (W bti, j )0⩽t⩽Ti under
the swap forward measure P b i, j . See Exercise 4 for swaption pricing without the Black-Scholes
formula.
Proposition 4.17 (Black swaption formula for payer swaptions). Assume that the LIBOR swap
rate (2.2.7) is modeled as a geometric Brownian motion under P
b i, j , i.e.
dS(t, Ti , T j ) = S(t, Ti , T j )σ bti, j ,

bi, j (t )dW (4.5.9)

where σbi, j (t ) t∈R is a deterministic volatility function of time. Then, the payer swaption with
+
payoff
(P(T , Ti ) − P(T , T j ) − κP(Ti , Ti , T j ))+ = P(Ti , Ti , T j ) (S(Ti , Ti , T j ) − κ )+
can be priced using the Black-Scholes call formula as

h r Ti i
+
IE∗ e − t rs ds P(Ti , Ti , T j ) (S(Ti , Ti , T j ) − κ ) Ft
= (P(t, Ti ) − P(t, T j ))Φ(d+ (t, Ti ))
j−1
−κΦ(d− (t, Ti )) ∑ (Tk+1 − Tk )P(t, Tk+1 ),
k =i
t ∈ [0, Ti ], where
log(S(t, Ti , T j )/κ ) + σi,2j (t, Ti )(Ti − t )/2

d+ (t, Ti ) = √ , (4.5.10)
σi, j (t, Ti ) Ti − t
and
log(S(t, Ti , T j )/κ ) − σi,2j (t, Ti )(Ti − t )/2
d− (t, Ti ) = √ , (4.5.11)
σi, j (t, Ti ) Ti − t
and
1 w Ti
|σi, j (t, Ti )|2 = |σbi, j (s)|2 ds, 0 ⩽ t ⩽ Ti . (4.5.12)
Ti − t t
Proof. Since S(t, Ti , T j ) is a geometric Brownian motion with volatility function (σb (t ))t∈R+ under
P
b i, j , by (4.5.4)-(4.5.5) in Lemma 4.15 or (4.5.7)-(4.5.8) we have
h r Ti i
IE∗ e − t rs ds P(Ti , Ti , T j ) (S(Ti , Ti , T j ) − κ )+ Ft
h rT i
= IE∗ e − t rs ds (P(T , Ti ) − P(T , T j ) − κP(Ti , Ti , T j ))+ Ft
b i, j (S(Ti , Ti , T j ) − κ )+ Ft

= P(t, Ti , T j )IE
= P(t, Ti , T j )Bl(S(t, Ti , T j ), κ, σi, j (t, Ti ), 0, Ti − t )
= P(t, Ti , T j ) (S(t, Ti , T j )Φ+ (t, S(t, Ti , T j )) − κΦ− (t, S(t, Ti , T j )))
= P(t, Ti ) − P(t, T j ) Φ+ (t, S(t, Ti , T j )) − κP(t, Ti , T j )Φ− (t, S(t, Ti , T j ))

= P(t, Ti ) − P(t, T j ) Φ+ (t, S(t, Ti , T j ))


j−1
−κΦ− (t, S(t, Ti , T j )) ∑ (Tk+1 − Tk )P(t, Tk+1 ).
k =i
In addition, the hedging strategy
(Φ+ (t, S(t, Ti , T j )), −κΦ− (t, S(t, Ti , T j ))(Ti+1 − Ti ), . . .

. . . , −κΦ− (t, S(t, Ti , T j ))(T j−1 − T j−2 ), −Φ+ (t, S(t, Ti , T j )))
based on the assets (P(t, Ti ), . . . , P(t, T j )) is self-financing by Corollary 3.18, see also Privault
and Teng, 2012. Similarly to the above, a receiver (or put) swaption gives the option, but not
the obligation, to enter an interest rate swap as receiver of a fixed rate κ and as payer of floating
LIBOR rates L(Ti , Tk , Tk+1 ) at times Ti+1 , . . . , T j , and can be priced as in the next proposition.
Proposition 4.18 (Black swaption formula for receiver swaptions). Assume that the LIBOR
swap rate (2.2.7) is modeled as the geometric Brownian motion (4.5.9) under the forward swap
measure Pb i, j . Then, the receiver swaption with payoff
+
κP(Ti , Ti , T j ) − P(T , Ti ) − P(T , T j ) = P(Ti , Ti , T j ) (κ − S(Ti , Ti , T j ))+
can be priced using the Black-Scholes put formula as

h r Ti i
IE∗ e − t rs ds P(Ti , Ti , T j ) (κ − S(Ti , Ti , T j ))+ Ft
j−1
= κΦ(−d− (t, Ti )) ∑ (Tk+1 − Tk )P(t, Tk+1 )
k =i
−(P(t, Ti ) − P(t, T j ))Φ(−d+ (t, Ti )),
where d+ (t, Ti ), and d− (t, Ti ) and |σi, j (t, Ti )|2 are defined in (4.5.10)-(4.5.12).
When the SOFR swap rate (2.2.11) is modeled as a geometric Brownian motion under P
b i, j as in
(4.5.9), SOFR swaptions are priced in the same way as LIBOR swaptions.
Swaption prices can also be computed by an approximation formula, from the exact dynamics of
the swap rate S(t, Ti , T j ) under the forward swap measure P
b i, j , based on the bond price dynamics of
the form (4.1.3), cf. Schoenmakers, 2005, page 17.
Swaption volatilities can be estimated from swaption prices as implied volatilities from the
Black pricing formula:

0.2
0.18
0.16
0.14
0.12
9
8
0.1 7
6
5
0.08 4 i
0 3
1 2 2
3 4 5 6 1
7 8 0
j 9
Figure 4.1: Implied swaption volatilities.
Implied swaption volatilities can then be used to calibrate the BGM model, cf. Schoenmakers,
2005, Privault and Wei, 2009, or § 9.5 of Privault, 2021.
LIBOR-SOFR Swaps
We consider the swap contract with payoff
j−1
∑ (Tk+1 − Tk )(R(Tk+1 , Tk , Tk+1 ) − L(Tk , Tk , Tk+1 )),
k =i
for the exchange of a backward-looking SOFR rate R(Tk+1 , Tk , Tk+1 ) with the forward-looking
LIBOR rate L(Tk , Tk , Tk+1 ) over the time period [Tk , Tk+1 ]. The price of this interest rate swap
vanishes at any time t ∈ [0, T1 ], as
r
Tk+1
(Tk+1 − Tk ) IE e − t rs ds (R(Tk+1 , Tk , Tk+1 ) − L(Tk , Tk , Tk+1 )) Ft
= (Tk+1 − Tk )P(t, Tk+1 ) IEk+1 [R(Tk+1 , Tk , Tk+1 ) − L(Tk , Tk , Tk+1 ) | Ft ]

= (Tk+1 − Tk )P(t, Tk+1 )(R(t, Tk , Tk+1 ) − L(t, Tk , Tk+1 )
= 0, 0 ⩽ t ⩽ Tk .
see Mercurio, 2018. On the other hand, for any i = 1, . . . , n, we also have
r
T
− t k+1 rs ds
(Tk+1 − Tk ) IE e (R(Tk+1 , Tk , Tk+1 ) − L(Tk , Tk , Tk+1 )) FTi
= (Tk+1 − Tk )P(Ti , Tk+1 ) IEk+1 [R(Tk+1 , Tk , Tk+1 ) − L(Tk , Tk , Tk+1 ) | FTk ]

= (Tk+1 − Tk )P(Ti , Tk+1 )(R(Ti , Tk , Tk+1 ) − L(Ti , Tk , Tk+1 )
= 0.
Bermudan swaption pricing in Quantlib

The Bermudan swaption on the tenor structure {Ti , . . . , T j } is priced as the supremum
" !+ #
rT j−1
l
Sup IE∗ e − t rs ds
δk P(Tl , Tk+1 )(L(Tl , Tk , Tk+1 ) − κ )
∑ Ft
l∈{i,..., j−1} k =l
h rT i
IE∗ e − t rs ds (P(Tl , Tl ) − P(Tl , T j ) − κP(Tl , Tl , T j ))+
l
= Sup Ft
l∈{i,..., j−1}

h rT i
l
= Sup IE∗ e − t rs ds P(Tl , Tl , T j )(S(Tl , Tl , T j ) − κ )+ Ft ,
l∈{i,..., j−1}
where the supremum is over all stopping times taking values in {Ti , . . . , T j }.
Bermudan swaptions can be priced using this Rcode* in (R)quantlib, with the following output:
Summary of pricing results for Bermudan Swaption
Price (in bp) of Bermudan swaption is 24.92137

Strike is NULL (ATM strike is 0.05 )
Model used is: Hull-White using analytic formulas
Calibrated model parameters are:
a = 0.04641
sigma = 0.005869
This modified code* can be used in particular the pricing of ordinary swaptions, with the output:
Summary of pricing results for Bermudan Swaption
Price (in bp) of Bermudan swaption is 22.45436

Strike is NULL (ATM strike is 0.05 )
Model used is: Hull-White using analytic formulas
Calibrated model parameters are:
a = 0.07107
sigma = 0.006018
Table 4.2 summarizes some possible uses of change of numéraire in option pricing.
Application Asset price Numéraire process Option payoff Forward measure P

b Deflated process Option price Change of numéraire formula
rt dP
b rt h rT i rt h rT i
Risk-neutral pricing St Nt = e 0 rs ds C =1 Set = e − 0 rs ds St IE∗ e − t rs dsC Ft e 0 rs ds IE∗ e − 0 rs dsC Ft
dP∗
dP
b rT NT St h rT i +
(ST − κNT )+ = e − 0 rs ds IE∗ e − t rs ds (ST − κNT )+ Ft Ft

Exchange option St Nt Xbt = E XbT − κ
Nt Ib
dP∗ N0 Nt
2 t/2 dP
b ST +
St = S0 e rt +σWt −σ ST ( ST − K ) + = e −(T −t )r IE∗ e −(T −t )r ST ST − K Ft E[(ST − K )+ | Ft ]

Exotic Nt = St Xbt = 1 St Ib
dP∗ S0
" #
dP κ +

f f
b f RT f f
e r t Rt Nt = e r t Rt (RT − κ )+ = e (r −r)T e r T IE∗ e −(T −t )r (RT − κ ) | Ft e r t Rt Ib

Foreign exchange Xbt = 1 E 1− Rt
dP∗ R0 RT
rT
dP
bT e − 0 rs ds P(t, S) h rT i
Bond option P(t, S) Nt = P(t, T ) (P(T , S ) − K ) + = Xbt = IE∗ e − t rs ds (P(T , S) − K )+ Ft ET [(P(T , S) − K )+ | Ft ]
P(t, T )Ib
dP∗ P(0, T ) P(t, T )
rS
dP e − 0 rs ds h rS

bS 1 P(t, T ) i
Caplets and caps P(t, T ) Nt = P(t, S) (S − T )(L(T , T , S) − κ )+ = L(t, T , S) = −1 (S − T ) IE∗ e − t rs ds (L(T , T , S) − K )+ Ft (S − T )P(t, S)IbES [(L(T , T , S) − K )+ | Ft ]
dP∗ P(0, S) S−T P(t, S)
dPb 1,n r T1 P(T1 , T1 , Tn ) P(t, T1 ) − P(t, Tn ) h r T1 i

(P(T1 , T1 ) − P(T1 , Tn ) − κP(T1 , T1 , Tn ))+ = e − 0 rs ds IE∗ e − t rs ds (P(T1 , T1 ) − P(T1 , Tn ) − κP(T1 , T1 , Tn ))+ Ft E1,n (S(T1 , T1 , Tn ) − κ )+ Ft

Swaption P(t, T1 ), P(t, Tn ) Nt = P(t, T1 , Tn ) S(t, T1 , Tn ) = P(t, T1 , Tn )Ib
dP∗ P(0, T1 , Tn ) P(t, T1 , Tn )
dP
e −(T −t )r IE∗ STn 1{ST ⩾K} Ft − K n e −(T −t )r IE∗ 1{ST ⩾K} Ft E 1{ST ⩾K} Ft
2 t/2 2 t/2−(n−1)rt
b 2 2 2 T /2+(n−1)rT
St = S0 e rt +σWt −σ Nt = Stn e −(n−1)nσ (STn − K n )+ = e nσWT −(nσ ) T /2 Xbt = Stn−1 e (n−1)nσ t/2+(n−1)rt e (n−1)nσ

Power options Nt Ib
dP∗
−K n e −(T −t )r IE∗ 1{ST ⩾K} Ft

Table 4.2: A list of numéraire processes and their applications.


Exercises
Exercise 4.1 Consider a floorlet on a three-month LIBOR rate in nine month’s time, with a notional
principal amount of $10, 000 per interest rate percentage point. The term structure is flat at 3.95%
per year with discrete compounding, the volatility of the forward LIBOR rate in nine months is
10%, and the floor rate is 4.5%.
a) What are the key assumptions on the LIBOR rate in nine month in order to apply Black’s
formula to price this floorlet?
b) Compute the price of this floorlet using Black’s formula as an application of Proposition 4.7
and (4.3.9), using the functions Φ(d+ ) and Φ(d− ).
Exercise 4.2 Consider a payer swaption giving its holder the right, but not the obligation, to
enter into a 3-year annual pay swap in four years, where a fixed rate of 5% will be paid and the
LIBOR rate will be received. Assume that the yield curve is flat at 5% with continuous annual
compounding and the volatility of the swap rate is 20%. The notional principal is $100,000 per
interest rate percentage point.
a) What are the key assumptions in order to apply Black’s formula to value this swaption?
b) Compute the price of this swaption using Black’s formula as an application of Proposi-
tion 4.17.
Exercise 4.3 Consider a receiver swaption which is giving its holder the right, but not the obliga-
tion, to enter into a 2-year annual pay swap in three years, where a fixed rate of 5% will be received
and the LIBOR rate will be paid. Assume that the yield curve is flat at 2% with continuous annual
compounding and the volatility of the swap rate is 10%. The notional principal is $10,000 per
percentage point, and the swaption price is quoted in basis points. Write down the expression of the
price of this swaption using Black’s formula.
Exercise 4.4 Consider two bonds with maturities T1 and T2 , T1 < T2 , which follow the stochastic
differential equations
dP(t, T1 ) = rt P(t, T1 )dt + ζ1 (t )P(t, T1 )dWt
and
dP(t, T2 ) = rt P(t, T2 )dt + ζ2 (t )P(t, T2 )dWt .
a) Using Itô calculus, show that the forward process P(t, T2 )/P(t, T1 ) is a driftless geometric
Brownian motion driven h by dWt := dWt − ζ1 (t )dt underithe T1 -forward measure P.
b b
rT
1
b) Compute the price IE∗ e − t (K − P(T1 , T2 ))+ Ft of a bond put option at time t ∈
rs ds
[0, T1 ] using change of numéraire and the Black-Scholes formula.
and variance v given Ft , we have:

Hint: Given X a Gaussian random variable with mean m 2

+ 1
IE κ − e X | Ft = κΦ − (m − log κ )

(4.5.13)
v

2 1
− e m+v /2 Φ − (m + v2 − log κ ) .
v
Exercise 4.5 Given two bonds with maturities T , S and prices P(t, T ), P(t, S), consider the LIBOR
rate
P(t, T ) − P(t, S)
L(t, T , S) :=
(S − T )P(t, S)

at time t ∈ [0, T ], modeled as
dL(t, T , S) = µt L(t, T , S)dt + σ L(t, T , S)dWt , 0 ⩽ t ⩽ T, (4.5.14)
where (Wt )t∈[0,T ] is a standard Brownian motion under the risk-neutral probability measure P∗ ,
σ > 0 is a constant, and ( µt )t∈[0,T ] is an adapted process. Let
h rS i
F (t ) := IE∗ e − t rs ds (κ − L(T , T , S))+ Ft
denote the price at time t of a floorlet option with strike level κ, maturity T , and payment date S.
a) Rewrite the value of F (t ) using the forward measure P b S with maturity S.
b) What is the dynamics of L(t, T , S) under the forward measure P b S?
c) Write down the value of F (t ) using the Black-Scholes formula.
Hint: Given X a centered Gaussian random variable with variance v2 , we have

2 /2
IE∗ [(κ − e m+X )+ ] = κΦ(−(m − log κ )/v) − e m+v Φ(−v − (m − log κ )/v),
where Φ denotes the Gaussian cumulative distribution function.
Exercise 4.6 Jamshidian’s trick (Jamshidian, 1989). Consider a family (P(t, Tl ))l =i,..., j of bond
prices defined from a short rate process (rt )t∈R+ . We assume that the bond prices are functions
P(Ti , Tl +1 ) = Fl +1 (Ti , rTi ) of rTi that are increasing in the variable rTi , for all l = i, i + 1, . . . , j − 1.
a) Compute the price P(t, Ti , T j ) of the annuity numéraire paying coupons ci+1 , . . . , c j at times
Ti+1 , . . . , T j in terms of the bond prices
P(t, Ti+1 ), . . . , P(t, T j ).
b) Show that the payoff

+
P(Ti , Ti ) − P(Ti , T j ) − κP(Ti , Ti , T j )
of a European swaption can be rewrittten as
!+
j−1
1−κ ∑ c̃l +1 P(Ti , Tl +1 ) ,
l =i
by writing c̃l in terms of cl , l = i + 1, . . . , j.

c) Assuming that the bond prices are functions P(Ti , Tl +1 ) = Fl (Ti , rTi ) of rTi that are increasing
in the variable rTi , for all l = i, . . . , j − 1, show, choosing γκ such that
j−1
κ ∑ cl +1 Fl +1 (Ti , γκ ) = 1,
l =i
that the European swaption with payoff

!+
+ j−1
P(Ti , Ti ) − P(Ti , T j ) − κP(Ti , Ti , T j ) = 1−κ ∑ cl +1 P(Ti , Tl +1 ) ,
l =i
where c j contains the final coupon payment, can be priced as a weighted sum of bond put
options under the forward measure P b i with numéraire Nt(i) := P(t, Ti ).

Exercise 4.7 Path freezing. Consider n bonds with prices (P(t, Ti ))i=1,...,n and the bond option
with payoff
!+
n
∑ ci P(T0 , Ti ) − κP(T0 , T1 ) = P(T0 , T1 ) (XT0 − κ )+ ,
i=2
where Nt := P(t, T1 ) is taken as numéraire and

n n
1
Xt := ∑ ci P(t, Ti ) = ∑ ci Pb(t, Ti ), 0 ⩽ t ⩽ T1 .
P(t, T1 ) i=2 i=2
with Pb(t, Ti ) := P(t, Ti )/P(t, T1 ), i = 2, 3, . . . , n.

a) Assuming that the deflated bond price (Pb(t, Ti ))t∈[0,Ti ] has the (martingale) dynamics d Pb(t, Ti ) =
bt under the forward measure P
σi (t )Pb(t, Ti )dW b 1 , where (σi (t ))t∈R is a deterministic func-
+
tion, write down the dynamics of Xt as dXt = σt Xt dW bt , where σt is to be computed explicitly.
b) Approximating (Pb(t, Ti ))t∈[0,Ti ] by Pb(0, Ti ) and (P(t, T2 , Tn ))t∈[0,T2 ] by P(0, T2 , Tn ), find a
deterministic approximation σ b (t ) of σt , and deduce an expression of the option price
" !+ #
r T1 n
IE∗ e − rs ds
= P(0, T1 )IbE (XT0 − κ )+

0
∑ ci P(T0 , Ti ) − κP(T0 , T1 )
i=2
using the Black-Scholes formula.
Hint: Given X a centered Gaussian random variable with variance v2 , we have:

2 +
IE x e X−v /2 − κ

= xΦ(v/2 + (log(x/κ ))/v) − κΦ(−v/2 + (log(x/κ ))/v).
Exercise 4.8 (Exercise 1.4 continued). We work in the short rate model
drt = σ dBt ,
where (Bt )t∈R+ is a standard Brownian motion under P∗ , and P

b 2 is the forward measure defined by
dP
b2 1 rT
− 0 2 rs ds
= e .
dP∗ P(0, T2 )
a) State the expressions of ζ1 (t ) and ζ2 (t ) in
dP(t, Ti )
= rt dt + ζi (t )dBt , i = 1, 2,
P(t, Ti )
and the dynamics of the P(t, T1 )/P(t, T2 ) under P b 2 , where P(t, T1 ) and P(t, T2 ) are bond
prices with maturities T1 and T2 .
Hint: Use Exercise 1.4 and the relation (1.4.8).
b) State the expression of the forward rate f (t, T1 , T2 ).
c) Compute the dynamics of f (t, T1 , T2 ) under the forward measure P b 2 with
dP
b2 1 rT
− 0 2 rs ds
= e .
dP∗ P(0, T2 )

d) Compute the price

h r T2 i
(T2 − T1 ) IE∗ e − t rs ds ( f (T1 , T1 , T2 ) − κ )+ Ft
of an interest rate cap at time t ∈ [0, T1 ], using the expectation under the forward measure P
b 2.
e) Compute the dynamics of the swap rate process
P(t, T1 ) − P(t, T2 )
S(t, T1 , T2 ) = , t ∈ [0, T1 ],
(T2 − T1 )P(t, T2 )
under P
b 2.
f) Using (4.5.3), compute the swaption price
h r T1 i
(T2 − T1 ) IE∗ e − t rs ds P(T1 , T2 )(S(T1 , T1 , T2 ) − κ )+ Ft
on the swap rate S(T1 , T1 , T2 ) using the expectation under the forward swap measure P
b 1,2 .
Exercise 4.9 Consider three zero-coupon bonds P(t, T1 ), P(t, T2 ) and P(t, T3 ) with maturities
T1 = δ , T2 = 2δ and T3 = 3δ respectively, and the forward LIBOR L(t, T1 , T2 ) and L(t, T2 , T3 )
defined by

1 P(t, Ti )
L(t, Ti , Ti+1 ) = −1 , i = 1, 2.
δ P(t, Ti+1 )
Assume that L(t, T1 , T2 ) and L(t, T2 , T3 ) are modeled in the BGM model by
dL(t, T1 , T2 ) bt(2) ,
= e −at dW 0 ⩽ t ⩽ T1 , (4.5.15)
L(t, T1 , T2 )
(2)
and L(t, T2 , T3 ) = b, 0 ⩽ t ⩽ T2 , for some constants a, b > 0, where W
bt is a standard Brownian
motion under the forward measure P2 defined by
b
r T2
dP
b2 e − 0 rs ds
= .
dP∗ P(0, T2 )
a) Compute L(t, T1 , T2 ), 0 ⩽ t ⩽ T2 by solving Equation (4.5.15).

b) Show that the price at time t ∈ [0, T1 ] of the caplet with strike level κ can be written as
h r T2 i
IE∗ e − t rs ds (L(T1 , T1 , T2 ) − κ )+ Ft = P(t, T2 )Ib
E2 (L(T1 , T1 , T2 ) − κ )+ | Ft ,

where IEb 2 denotes the expectation under the forward measure P b 2.

c) Using the hint below, compute the price at time t of the caplet with strike level κ on
L(T1 , T1 , T2 ).
d) Compute
P(t, T1 ) P(t, T3 )
, 0 ⩽ t ⩽ T1 , and , 0 ⩽ t ⩽ T2 ,
P(t, T1 , T3 ) P(t, T1 , T3 )
in terms of b and L(t, T1 , T2 ), where P(t, T1 , T3 ) is the annuity numéraire
P(t, T1 , T3 ) = δ P(t, T2 ) + δ P(t, T3 ), 0 ⩽ t ⩽ T2 .

e) Compute the dynamics of the swap rate
P(t, T1 ) − P(t, T3 )
t 7→ S(t, T1 , T3 ) = , 0 ⩽ t ⩽ T1 ,
P(t, T1 , T3 )
i.e. show that we have

(2)
dS(t, T1 , T3 ) = σ1,3 (t )S(t, T1 , T3 )dW
bt ,
where σ1,3 (t ) is a stochastic process to be determined.

f) Using hthe Black-Scholes
rT
formula, compute an approximation
i of the swaption price
∗ − t 1 rs ds
IE e P(T1 , T1 , T3 )(S(T1 , T1 , T3 ) − κ ) Ft
+
b 2 (S(T1 , T1 , T3 ) − κ )+ | Ft ,

= P(t, T1 , T3 )IE
at time t ∈ [0, T1 ]. You will need to approximate σ1,3 (s), s ⩾ t, by “freezing” all random
terms at time t.

2
IE∗ ( e m+X − κ )+ = e m+v /2 Φ(v + (m − log κ )/v) − κΦ((m − log κ )/v),

Exercise 4.10 Bond option hedging. Consider a portfolio allocation (ξtT , ξtS )t∈[0,T ] made of two
bonds with maturities T , S, and value
Vt = ξtT P(t, T ) + ξtS P(t, S), 0 ⩽ t ⩽ T,
at time t. We assume that the portfolio is self-financing, i.e.
dVt = ξtT dP(t, T ) + ξtS dP(t, S), 0 ⩽ t ⩽ T, (4.5.16)
and that it hedges the claim payoff (P(T , S) − κ )+ , so that

h rT i
Vt = IE∗ e − t rs ds (P(T , S) − κ )+ Ft
= P(t, T ) IET (P(T , S) − κ )+ Ft ,

0 ⩽ t ⩽ T.
a) Show thathwe rhave

T
i
IE∗ e − t rs ds (P(T , S) − K )+ Ft
wt wt
= P(0, T ) IET (P(T , S) − K )+ + ξsT dP(s, T ) + ξsS dP(s, S).

0 0
b) Show
rt
that under the self-financing condition (4.5.16), the deflated portfolio value Vet =
− 0 rs ds
e Vt satisfies
dVet = ξtT d Pe(t, T ) + ξtS d Pe(t, S),
where rt
Pe(t, T ) := e − 0 rs ds P(t, T ), t ∈ [0, T ],
and rt
Pe(t, S) := e − 0 rs ds P(t, S), t ∈ [0, S],
denote the discounted bond prices.

c) From now on we work in the framework of Proposition 4.3, and we let the function C (x, v)
be defined by
C (Xt , v(t, T )) := IET (P(T , S) − K )+ Ft ,

where Xt is the forward price Xt := P(t, S)/P(t, T ), t ∈ [0, T ], and

wT 2
v2 (t, T ) := σsS − σsT ds.
t
Show that
IET (P(T , S) − K )+ Ft = IET (P(T , S) − K )+

w t ∂C
+ (Xu , v(u, T ))dXu , t ⩾ 0.
0 ∂x
Hint: Use the martingale property and the Itô formula.
d) Show that the deflated portfolio value Vbt = Vt /P(t, T ) satisfies
∂C
dVbt = (Xt , v(t, T ))dXt
∂x
P(t, S) ∂C
= (Xt , v(t, T ))(σtS − σtT )d BbtT .
P(t, T ) ∂ x
e) Show that
∂C
dVt = P(t, S) (Xt , v(t, T ))(σtS − σtT )dBt + Vbt dP(t, T ).
∂x
f) Show that
∂C
dVet = Pe(t, S) (Xt , v(t, T ))(σtS − σtT )dBt + Vbt d Pe(t, T ).
∂x
g) Compute the hedging strategy (ξtT , ξtS )t∈[0,T ] of this bond option.
h) Show that
log(x/K ) + τv2 /2

∂C
(x, v) = Φ √ ,
∂x τv
and compute the hedging strategy (ξtT , ξtS )t∈[0,T ] in terms of the normal cumulative distribu-
tion function Φ.
Exercise 4.11 Consider a LIBOR rate L(t, T , S), t ∈ [0, T ], modeled as dL(t, T , S) = µt L(t, T , S)dt +
σ (t )L(t, T , S)dWt , 0 ⩽ t ⩽ T , where (Wt )t∈[0,T ] is a standard Brownian motion under the risk-neutral
probability measure P∗ , ( µt )t∈[0,T ] is an adapted process, and σ (t ) > 0 is a deterministic volatility
function of time t.
a) What is the dynamics of L(t, T , S) under the forward measure P b with numéraire Nt := P(t, S)?
b) Rewrite the price
h rS i
IE∗ e − t rs ds φ (L(T , T , S)) Ft (4.5.17)
at time t ∈ [0, T ] of an option with payoff function φ using the forward measure P.
b
c) Write down the above option price (4.5.17) using an integral.
Exercise 4.12 Given n bonds with maturities T1 , T2 , . . . , Tn , consider the annuity numéraire
j−1
P(t, Ti , T j ) = ∑ (Tk+1 − Tk )P(t, Tk+1 )
k =i
and the swap rate

P(t, Ti ) − P(t, T j )
S(t, Ti , T j ) =
P(t, Ti , T j )

at time t ∈ [0, Ti ], modeled as

dS(t, Ti , T j ) = µt S(t, Ti , T j )dt + σ S(t, Ti , T j )dWt , 0 ⩽ t ⩽ Ti , (4.5.18)
where (Wt )t∈[0,Ti ] is a standard Brownian motion under the risk-neutral probability measure P∗ ,
( µt )t∈[0,T ] is an adapted process and σ > 0 is a constant. Let
h r Ti i
IE∗ e − t rs ds P(Ti , Ti , T j )φ (S(Ti , Ti , T j )) Ft (4.5.19)
at time t ∈ [0, Ti ] of an option with payoff function φ .

a) Rewrite the option price (4.5.19) at time t ∈ [0, Ti ] using the forward swap measure P b i, j
defined from the annuity numéraire P(t, Ti , T j ).
b) What is the dynamics of S(t, Ti , T j ) under the forward swap measure P b i, j ?
c) Write down the above option price (4.5.17) using a Gaussian integral.
d) Apply the above to the computation at time t ∈ [0, Ti ] of the put swaption price
h r Ti i
IE∗ e − t rs ds P(Ti , Ti , T j )(κ − S(Ti , Ti , T j ))+ Ft
with strike level κ, using the Black-Scholes formula.

2 /2
IE[(κ − e m+X )+ ] = κΦ(−(m − log κ )/v) − e m+v Φ(−v − (m − log κ )/v),
Exercise 4.13 Consider a bond market with two bonds with maturities T1 , T2 , whose prices
P(t, T1 ), P(t, T2 ) at time t are given by
dP(t, T1 ) dP(t, T2 )
= rt dt + ζ1 (t )dBt , = rt dt + ζ2 (t )dBt ,
P(t, T1 ) P(t, T2 )
where (rt )t∈R+ is a short-term interest rate process, (Bt )t∈R+ is a standard Brownian motion gener-
ating a filtration (Ft )t∈R+ , and ζ1 (t ), ζ2 (t ) are volatility processes. The LIBOR rate L(t, T1 , T2 ) is
defined by
P(t, T1 ) − P(t, T2 )
L(t, T1 , T2 ) = .
P(t, T2 )
Recall that a caplet on the LIBOR market can be priced at time t ∈ [0, T1 ] as
h r T2 i
IE e − t rs ds (L(T1 , T1 , T2 ) − κ )+ Ft (4.5.20)
b L(T1 , T1 , T2 ) − κ + Ft ,

= P(t, T2 )IE
under the forward measure P
b defined by
dP
b rT
− 0 1 rs ds P(T1 , T2 )
= e ,
dP∗ P(0, T2 )
under which
wt
Bbt := Bt − ζ2 (s)ds, t ∈ R+ , (4.5.21)
0
is a standard Brownian motion.
In what follows we let Lt = L(t, T1 , T2 ) for simplicity of notation.

a) Using Itô calculus, show that the LIBOR rate satisfies
dLt = Lt σ (t )d Bbt , 0 ⩽ t ⩽ T1 , (4.5.22)
where the LIBOR rate volatility is given by

P(t, T1 )(ζ1 (t ) − ζ2 (t ))
σ (t ) = .
P(t, T1 ) − P(t, T2 )
b) Solve the equation (4.5.22) on the interval [t, T1 ], and compute LT1 from the initial condition
Lt .
c) Assuming that σ (t ) in (4.5.22) is a deterministic volatility function of time t ∈ [0, T1 ], show
that the price
E (LT1 − κ )+ Ft

P(t, T2 )Ib
w T1
of the caplet can be written as P(t, T2 )C (Lt , v(t, T1 )), where v2 (t, T1 ) = |σ (s)|2 ds, and
t
C (t, v(t, T1 )) is a function of Lt and v(t, T1 ).
(1) (2)
d) Consider a portfolio allocation (ξt , ξt )t∈[0,T1 ] made of bonds with maturities T1 , T2 and
value
(1) (2)
Vt = ξt P(t, T1 ) + ξt P(t, T2 ),
at time t ∈ [0, T1 ]. We assume that the portfolio is self-financing, i.e.
(1) (2)
dVt = ξt dP(t, T1 ) + ξt dP(t, T2 ), 0 ⩽ t ⩽ T1 , (4.5.23)
and that it hedgeshthe claim payoff (LT1 − κ )+ , so thati

rT
1
Vt = IE e − t (P(T1 , T2 )(LT1 − κ ))+ Ft
rs ds
b (LT1 − κ )+ Ft ,

= P(t, T2 )IE
0 ⩽ t ⩽ Th1 . Show
rT
that we have i
− t 1 rs ds
+
IE e P(T1 , T2 )(LT1 − κ ) Ft
+ w t (1) wt
(2)
= P(0, T2 )IE b LT1 − κ + ξs dP(s, T1 ) + ξs dP(s, T1 ),
0 0
0 ⩽ t ⩽ T1 .
e) Show
rt
that under the self-financing condition (4.5.23), the discounted portfolio value Vet =
e − 0 rs dsVt satisfies
(1) (2)
dVet = ξt d Pe(t, T1 ) + ξt d Pe(t, T2 ),
rt rt
where Pe(t, T1 ) := e − 0 rs ds P(t, T1 ) and Pe(t, T2 ) := e − 0 rs ds P(t, T2 ) denote the discounted
bond prices.
f) Show that
+ w t ∂C
b LT1 − κ + Ft = Ib

IE E LT1 − κ + (Lu , v(u, T1 ))dLu ,
0 ∂x
and that the deflated portfolio value Vbt = Vt /P(t, T2 ) satisfies

∂C ∂C
dVbt = (Lt , v(t, T1 ))dLt = σ (t )Lt (Lt , v(t, T1 ))d Bbt .
∂x ∂x
Hint: use the martingale property and the Itô formula.
g) Show that
∂C
dVt = (P(t, T1 ) − P(t, T2 )) (Lt , v(t, T1 ))σ (t )dBt + Vbt dP(t, T2 ).
∂x

h) Show that
∂C
dVet = (Lt , v(t, T1 ))d (Pe(t, T1 ) − Pe(t, T2 ))
∂ x
∂C
+ Vt − Lt
b (Lt , v(t, T1 )) d Pe(t, T2 ),
∂x
(1) (2)
and deduce the values of the hedging portfolio allocation (ξt , ξt )t∈R+ .
Problem 4.14 Consider a bond market with tenor structure {Ti , . . . , T j } and j − i + 1 bonds with
maturities Ti , . . . , T j , whose prices P(t, Ti ), . . . P(t, T j ) at time t are given by
dP(t, Tk )
= rt dt + ζk (t )dBt , k = i, . . . , j,
P(t, Tk )
where (rt )t∈R+ is a short-term interest rate process and (Bt )t∈R+ denotes a standard Brownian
motion generating a filtration (Ft )t∈R+ , and ζi (t ), . . . , ζ j (t ) are volatility processes.
The swap rate S(t, Ti , T j ) is defined by

P(t, Ti ) − P(t, T j )
S(t, Ti , T j ) = ,
P(t, Ti , T j )
where
j−1
P(t, Ti , T j ) = ∑ (Tk+1 − Tk )P(t, Tk+1 )
k =i
is the annuity numéraire. Recall that a swaption on the LIBOR market can be priced at time
t ∈ [0, Ti ] as
" !+ #
rT j−1
i
IE∗ e − t rs ds
∑ (Tk+1 − Tk )P(Ti , Tk+1 )(S(Ti , Tk , Tk+1 ) − κ ) Ft
k =i
= P(t, Ti , T j ) IEi, j (S(Ti , Ti , T j ) − κ )+ Ft ,

(4.5.24)
under the forward swap measure P

b i, j defined by
dP
b i, j r Ti P(Ti , Ti , T j )
∗
= e − 0 rs ds , 1 ⩽ i < j ⩽ n,
dP P(0, Ti , T j )
under which
j−1
P(t, Tk+1 )
Bbti, j := Bt − ∑ (Tk+1 − Tk ) ζk+1 (t )dt (4.5.25)
k =i P(t, Ti , T j )
is a standard Brownian motion. Recall that the swap rate can be modeled as
dS(t, Ti , T j ) = S(t, Ti , T j )σi, j (t )d Bbti, j , 0 ⩽ t ⩽ Ti , (4.5.26)
where the swap rate volatilities are given by
j−1
P(t, Tl +1 )
σi, j (t ) = ∑ (Tl +1 − Tl ) P(t, Ti , Tj ) (ζi (t ) − ζl +1 (t )) (4.5.27)
l =i
P(t, T j )
+ (ζi (t ) − ζ j (t ))
P(t, Ti ) − P(t, T j )
1 ⩽ i, j ⩽ n, cf. e.g. Proposition 8.12 in Privault, 2021. In what follows we denote St = S(t, Ti , T j )
for simplicity of notation.

a) Solve the equation (4.5.26) on the interval [t, Ti ], and compute S(Ti , Ti , T j ) from the initial
condition S(t, Ti , T j ).
b) Assuming that σi, j (t ) is a deterministic volatility function of time t ∈ [0, Ti ] for 1 ⩽ i, j ⩽ n,
show that the price (4.5.8) of the swaption can be written as
P(t, Ti , T j )C (St , v(t, Ti )),
where w Ti
v2 (t, Ti ) := |σi, j (s)|2 ds,
t
and C (x, v) is a function to be specified using the Black-Scholes formula Bl(x, K, σ , r, τ ),
with the relation
IE[(x e m+X − K )+ ] = Φ(v + (m + log(x/K ))/v) − KΦ((m + log(x/K ))/v),
where X is a centered Gaussian random variable with variance v2 .

(i) ( j)
c) Consider a portfolio allocation (ξt , . . . , ξt )t∈[0,Ti ] made of bonds with maturities Ti , . . . , T j
and value
j
(k )
Vt = ∑ ξt P(t, Tk ),
k =i
at time t ∈ [0, Ti ]. We assume that the portfolio is self-financing, i.e.

j
(k )
dVt = ∑ ξt dP(t, Tk ), 0 ⩽ t ⩽ Ti , (4.5.28)
k =i
and that it hedges the claim payoff (S(Ti , Ti , T j ) − κ )+ , so that

" !+ #
rT j−1
i
Vt = IE∗ e − t rs ds
k =i
= P(t, Ti , T j ) IEi, j (S(Ti , Ti , T j ) − κ )+ Ft ,

0⩽t ⩽
" Ti . Show that !+ #
rT j−1
∗ − i
IE e t rs ds
k =i
j wt
(k )
= P(0, Ti , T j ) IEi, j (S(Ti , Ti , T j ) − κ )+ + ∑ ξs dP(s, Ti ),

0
k =i
0 ⩽ t ⩽ Ti .
d) Show
rt
that under the self-financing condition (4.5.28), the discounted portfolio value Vet =
e − 0 rs dsVt satisfies
j
(k )
dVet = ∑ ξt d Pe(t, Tk ),
k =i
rt
where Pe(t, Tk ) = e − 0 rs ds P(t, Tk ), k = i, i + 1 . . . , j, denote the discounted bond prices.
e) Show that
IEi, j (S(Ti , Ti , T j ) − κ )+ Ft

w t ∂C
= IEi, j (S(Ti , Ti , T j ) − κ )+ +

(Su , v(u, Ti ))dSu .
0 ∂x
Hint: use the martingale property and the Itô formula.
f) Show that the deflated portfolio value Vbt = Vt /P(t, Ti , T j ) satisfies
∂C ∂C
dVbt = (St , v(t, Ti ))dSt = St (St , v(t, Ti ))σti, j d Bbti, j .
∂x ∂x

Default Risk in Bond Markets
122
g) Show that
∂C
dVt = (P(t, Ti ) − P(t, T j )) (St , v(t, Ti ))σti, j dBt + Vbt dP(t, Ti , T j ).
∂x
h) Show that
j−1
∂C
dVt = St ζi (t ) (St , v(t, Ti )) ∑ (Tk+1 − Tk )P(t, Tk+1 )dBt
∂x k =i
j−1
∂C
+(Vbt − St (St , v(t, Ti ))) ∑ (Tk+1 − Tk )P(t, Tk+1 )ζk+1 (t )dBt
∂x k =i
∂C
+ (St , v(t, Ti ))P(t, T j )(ζi (t ) − ζ j (t ))dBt .
∂x
i) Show that
∂C
dVet = (St , v(t, Ti ))d (Pe(t, Ti ) − Pe(t, T j ))
∂x
∂C
+(Vbt − St (St , v(t, Ti )))d Pe(t, Ti , T j ).
∂x
j) Show that
∂C log(x/K ) v(t, Ti )
(x, v(t, Ti )) = Φ + .
∂x v(t, Ti ) 2
k) Show that we have
log ( St /K ) v (t, T i )
dVet = Φ + d (Pe(t, Ti ) − Pe(t, T j ))
v(t, Ti ) 2

log(St /K ) v(t, Ti )
−κΦ − d Pe(t, Ti , T j ).
v(t, Ti ) 2
l) Show that the hedging strategy is given by

(i) log(St /K ) v(t, Ti )
ξt = Φ + ,
v(t, Ti ) 2

( j) log(St /K ) v(t, Ti ) log(St /K ) v(t, Ti )
ξt = −Φ + − κ (T j − T j−1 )Φ − ,
v(t, Ti ) 2 v(t, Ti ) 2
and
(k ) log(St /K ) v(t, Ti )
ξt = −κ (Tk+1 − Tk )Φ − , i ⩽ k ⩽ j − 2.
v(t, Ti ) 2

123
5. Reduced-Form Approach to Credit Risk
In this chapter, credit risk is estimated by modeling default probabilities using stochastic failure
rate processes. In addition, information on default events is incorporated to the model by the use of
exogeneous random variables and enlargement of filtrations. This is in contrast to the structural
approach to credit risk, in which bankruptcy is modeled from a firm’s asset value. Applications are
given to the pricing of default bonds.
5.1 Survival Probabilities 123

Exercises 131
5.1 Survival Probabilities

The reduced-form approach to credit risk relies on the concept of survival probability, defined as
the probability P(τ > t ) that a random system with lifetime τ survives at least over t years, t > 0.
Assuming that survival probabilities P(τ > t ) are strictly positive for all t > 0, we can compute the
conditional probability for that system to survive up to time T , given that it was still functioning at
time t ∈ [0, T ], as
P(τ > T and τ > t ) P(τ > T )

P(τ > T | τ > t ) = = , 0 ⩽ t ⩽ T,
P(τ > t ) P(τ > t )
with
P(τ ⩽ T | τ > t ) = 1 − P(τ > T | τ > t )

P(τ > t ) − P(τ > T )
=
P(τ > t )

124 Chapter 5. Reduced-Form Approach to Credit Risk
P(τ ⩽ T ) − P(τ ⩽ t )
=
P(τ > t )
P(t < τ ⩽ T )
= , 0 ⩽ t ⩽ T. (5.1.1)
P(τ > t )
Such survival probabilities are typically found in life (or mortality) tables:
Age t P(τ ⩽ t + 1 | τ > t )

20 0.0894%
30 0.1008%
40 0.2038%
50 0.4458%
60 0.9827%
Table 5.1: Mortality table.
The corresponding conditional survival probability distribution can be computed as follows:
P(τ ∈ dx | τ > t ) = P(x < τ ⩽ x + dx | τ > t )

= P(τ ⩽ x + dx | τ > t ) − P(τ ⩽ x | τ > t )
P(τ ⩽ x + dx) − P(τ ⩽ x)
=
P(τ > t )
1
= dP(τ ⩽ x)
P(τ > t )
1
= − dP(τ > x), x > t.
P(τ > t )
Proposition 5.1 The failure rate function, defined as
P(τ ⩽ t + dt | τ > t )
λ (t ) := ,
dt
satisfies
wt
P(τ > t ) = exp − λ (u)du , t ⩾ 0. (5.1.2)
0
Proof. By (5.1.1), we have
P(τ ⩽ t + dt | τ > t )
λ (t ) :=
dt
1 P(t < τ ⩽ t + dt )
=
P(τ > t ) dt
1 P(τ > t ) − P(τ > t + dt )
=
P(τ > t ) dt
d
= − log P(τ > t )
dt
1 d
= − P(τ > t ), t > 0,
P(τ > t ) dt

and the differential equation
d
P(τ > t ) = −λ (t )P(τ > t ),
dt
which can be solved as in (5.1.2) under the initial condition P(τ > 0) = 1. □
Proposition 5.1 allows us to rewrite the (conditional) survival probability as
w
P(τ > T )

T
P(τ > T | τ > t ) = = exp − λ (u)du , 0 ⩽ t ⩽ T,
P(τ > t ) t
with
P(τ > t + h | τ > t ) = e −λ (t )h ≃ 1 − λ (t )h, [h ↘ 0],
and
P(τ ⩽ t + h | τ > t ) = 1 − e −λ (t )h ≃ λ (t )h, [h ↘ 0],
as h tends to 0. When the failure rate λ (t ) = λ > 0 is a constant function of time, Relation (5.1.2)
shows that
P(τ > T ) = e −λ T , T ⩾ 0,
i.e. τ has the exponential distribution with parameter λ . Note that given (τn )n⩾1 a sequence of i.i.d.
exponentially distributed random variables, letting
Tn = τ1 + τ2 + · · · + τn , n ⩾ 1,
defines the sequence of jump times of a standard Poisson process with intensity λ > 0.
5.2 Stochastic Default

When the random time τ is a stopping time with respect to (Ft )t∈R+ we have
{τ > t} ∈ Ft , t ⩾ 0,
i.e. the knowledge of whether default or bankruptcy has already occurred at time t is contained in
Ft , t ∈ R+ , cf. e.g. Section 14.3 of Privault, 2022. As a consequence, we can write
P(τ > t | Ft ) = IE 1{τ>t} | Ft = 1{τ>t} ,

t ⩾ 0.
In what follows we will not assume that τ is an Ft -stopping time, and by analogy with (5.1.2) we
will write P(τ > t | Ft ) as
wt
P(τ > t | Ft ) = exp − λu du , t ⩾ 0, (5.2.1)
0
where the failure rate function (λt )t∈R+ is modeled as a random process adapted to a filtration
(Ft )t∈R+ .
The process (λt )t∈R+ can also be chosen among the classical mean-reverting diffusion processes,
including jump-diffusion processes. In Lando, 1998, the process (λt )t∈R+ is constructed as
λt := h(Xt ), t ∈ R+ , where h is a nonnegative function and (Xt )t∈R+ is a stochastic process
generating the filtration (Ft )t∈R+ .

The default time τ is then defined as

n wt o
τ := inf t ∈ R+ : h(Xu )du ⩾ L ,
0
where L is an exponentially distributed random variable with parameter µ > 0 and distribution
function P(L > x) = e −µx , x ⩾ 0, independent of (Ft )t∈R+ . In this case, as τ is not an (Ft )t∈R+ -
stopping time, we have
w t
P(τ > t | Ft ) = P h(Xu )du < L Ft
0
wt
= exp −µ h(Xu )du
0
wt
= exp −µ λu du , t ⩾ 0.
0
Definition 5.2 Let (Gt )t∈R+ be the filtration defined by G∞ := F∞ ∨ σ (τ ) and
Gt := B ∈ G∞ : ∃A ∈ Ft such that A ∩ {τ > t} = B ∩ {τ > t} ,

(5.2.2)
with Ft ⊂ Gt , t ⩾ 0.
In other words, Gt contains insider information on whether default at time τ has occurred or not
before time t, and τ is a (Gt )t∈R+ -stopping time. Note that this information on τ may not be
available to a generic user who has only access to the smaller filtration (Ft )t∈R+ . The next key
Lemma 5.3, see Lando, 1998, Guo, Jarrow, and Menn, 2007, allows us to price a contingent claim
r (Gt )t∈R
given the information in the larger filtration + , by only using information in (Ft )t∈R+ and
T
factoring in the default rate factor exp − t λu du .
Lemma 5.3 (Guo, Jarrow, and Menn, 2007, Theorem 1) For any FT -measurable integrable
random variable F, we have
IE F 1{τ>T } | Gt = 1{τ>t} IE FP(τ > T | τ > t ) | Ft

w
T
= 1{τ>t} IE F exp − λu du Ft , 0 ⩽ t ⩽ T.
t
Proof. By (5.2.1) we have

rT w
P(τ > T | FT ) e − 0 λu du

T
= rt = exp − λu du ,
P(τ > t | Ft ) e − 0 λu du t
hence, since F is FT -measurable,

w
P(τ > T | FT )

T
1{τ>t} IE F exp − λu du Ft = 1{τ>t} IE F Ft
t P(τ > t | Ft )
1{τ>t}
IE F IE[1{τ>T } | FT ] | Ft

=
P(τ > t | Ft )
1{τ>t}
IE IE[F 1{τ>T } | FT ] | Ft

=
P(τ > t | Ft )
IE F 1{τ>T } Ft

= 1{τ>t}
P(τ > t | Ft )
= 1{τ>t} IE F 1{τ>T } Gt


= IE F 1{τ>T } Gt ,

0 ⩽ t ⩽ T.
In the last step of the above argument, we used the key relation
h i 1{τ>t}
1{τ>t} IE F 1{τ>T } Gt = IE F 1{τ>T } | Ft ,

P(τ > t | Ft )
cf. Relation (75.2) in § XX-75 page 186 of Dellacherie, Maisonneuve, and Meyer, 1992, Theo-
rem VI-3-14 page 371 of Protter, 2004, and Lemma 3.1 of Elliott, Jeanblanc, and Yor, 2000, under
the conditional probability measure P|Ft , 0 ⩽ t ⩽ T . Indeed, according to (5.2.2), for any B ∈ Gt
we have, for some event A ∈ Ft ,
IE 1B 1{τ>t} F 1{τ>T } = IE 1B∩{τ>t} F 1{τ>T }

= IE 1A∩{τ>t} F 1{τ>T }

= IE 1A 1{τ>t} F 1{τ>T }

IE[1{τ>t} | Ft ]

= IE 1A 1{τ>t} F 1{τ>T }
P(τ > t | Ft )
IE[1A 1{τ>t} | Ft ]

= IE F 1{τ>T }
P(τ > t | Ft )
IE[1A 1{τ>t} | Ft ]

IE F 1{τ>T } | Ft

= IE
P(τ > t | Ft )
1A 1{τ>t}

IE F 1{τ>T } | Ft

= IE
P(τ > t | Ft )
IE[1A 1{τ>t} | Ft ]

IE F 1{τ>T } | Ft

= IE
P(τ > t | Ft )
1A 1{τ>t}

IE F 1{τ>T } | Ft

= IE
P(τ > t | Ft )
1B 1{τ>t}

IE F 1{τ>T } | Ft ,

= IE
P(τ > t | Ft )
hence by a standard characterization of conditional expectations, we have
1{τ>t}
1{τ>t} F 1{τ>T } | Gt = IE F 1{τ>T } | Ft

IE
P(τ > t | Ft )
□
Taking F = 1 in Lemma 5.3 allows one to write the survival probability up to time T , given the
information known up to time t, as
P(τ > T | Gt ) = IE 1{τ>T } | Gt

(5.2.3)
w
T
= 1{τ>t} IE exp − λu du Ft , 0 ⩽ t ⩽ T .
t
In particular, applying Lemma 5.3 for t = T and F = 1 shows that
1{τ>t} | Gt = 1{τ>t} ,

IE
which shows that {τ > t} ∈ Gt for all t > 0, and recovers the fact that τ is a (Gt )t∈R+ -stopping time,
while in general, τ is not (Ft )t∈R+ -stopping time.

The computation of P(τ > T | Gt ) according to (5.2.3) is then similar to that of a bond price, by
considering the failure rate λ (t ) as a “virtual” short-term interest rate. In particular the failure rate
λ (t, T ) can be modeled in the HJM framework, cf. e.g. Chapter 18.3 of Privault, 2022, and
w
T
P(τ > T | Gt ) = IE exp − λ (t, u)du Ft
t
can then be computed by applying HJM bond pricing techniques.

The computation of expectations given Gt as in Lemma 5.3 can be useful for pricing under
insider trading, in which the insider has access to the augmented filtration Gt while the ordinary
trader has only access to Ft , therefore generating two different prices IE∗ [F | Ft ] and IE∗ [F | Gt ] for
the same claim payoff F under the same risk-neutral probability measure P∗ . This leads to the issue
of computing the dynamics of the underlying asset price by decomposing it using a Ft -martingale
vs. a Gt -martingale instead of using different forward measures as in e.g. § 19.1 of Privault, 2022.
This can be obtained by the technique of enlargement of filtration, cf. Jeulin, 1980, Jacod, 1985,
Yor, 1985, Elliott and Jeanblanc, 1999.
5.3 Defaultable Bonds

Bond pricing models are generally based on the terminal condition P(T , T ) = $1 according to
which the bond payoff at maturity is always equal to $1, and default does not occurs. In this chapter
we allow for the possibility of default at a random time τ, in which case the terminal payoff of a
bond is allowed to vanish at maturity.
The price Pd (t, T ) at time t of a default bond with maturity T , (random) default time τ and
(possibly random) recovery rate ξ ∈ [0, 1] is given by
w
T
Pd (t, T ) = IE 1{τ>T } exp −
∗
ru du Gt
t
w
T
+ IE ξ 1{τ⩽T } exp − ru du
∗
Gt , 0 ⩽ t ⩽ T.
t
Proposition 5.4 The default bond with maturity T and default time τ can be priced at time
t ∈ [0, T ] as
w
T
Pd (t, T ) = 1{τ>t} IE exp − (ru + λu )du
∗
Ft
t
w
T
∗
Gt , 0 ⩽ t ⩽ T .
t
r
T
Proof. We take F = exp − t ru du in Lemma 5.3, which shows that
w w
T T
IE∗ 1{τ>T } exp − ru du Gt = 1{τ>t} IE∗ exp − (ru + λu )du Ft ,
t t
cf. e.g. Lando, 1998, Duffie and Singleton, 2003, Guo, Jarrow, and Menn, 2007. □
In the case of complete default (zero-recovery), we have ξ = 0 and
w
T
Pd (t, T ) = 1{τ>t} IE∗ exp − (rs + λs )ds Ft , 0 ⩽ t ⩽ T . (5.3.1)
t

From the above expression (5.3.1) we note that the effect of the presence of a default time τ is to
decrease the bond price, which can be viewed as an increase of the short rate by the amount λu . In
a simple setting where the interest rate r > 0 and failure rate λ > 0 are constant, the default bond
price becomes
Pd (t, T ) = 1{τ>t} e −(r+λ )(T −t ) , 0 ⩽ t ⩽ T.
In this case, the failure rate λ can be estimated at time t ∈ [0, T ] from a default bond price Pd (t, T )
and a non-default bond price P(t, T ) = e −(T −t )r as
1 P(t, T )
λ= log .
T −t Pd (t, T )
Finally, from e.g. Proposition 19.1 in Privault, 2022 the bond price (5.3.1) can also be expressed
under the forward measure P b with maturity T , as
w
T
Pd (t, T ) = 1{τ>t} IE exp − (rs + λs )ds
∗
Ft
t
w w
T T
= 1{τ>t} IE exp − rs ds
∗
Ft IE exp −
b λs ds Ft
t t
= 1{τ>t} Nt P
b (τ > T | Gt ),
where (Nt )t∈R+ is the numéraire process

w
T
∗
Nt := P(t, T ) = IE exp − rs ds Ft , 0 ⩽ t ⩽ T,
t
and by (5.2.3), w
T
P(τ > T | Gt ) = 1{τ>t} IE exp −
b b λs ds Ft
t
is the survival probability under the forward measure P

b defined as
dP
b NT − r T rt dt
:= e 0 ,
dP N0
see Chen and Huang, 2001 and Chen, Cheng, et al., 2008.
Estimating the default rates
Recall that the price of a default bond with maturity T , (random) default time τ and (possibly
random) recovery rate ξ ∈ [0, 1] is given by
w
T
Pd (t, T ) = 1{τ>t} IE exp − (ru + λu )du
∗
Ft
t
w
T
∗
Gt , 0 ⩽ t ⩽ T,
t
where ξ denotes the recovery rate. We consider a simplified deterministic step function model with
zero recovery rate and tenor structure
{t = T0 < T1 < · · · < Tn = T },
where
n−1 n−1
r (t ) = ∑ rl 1(Tl ,Tl+1 ] (t ) and λ (t ) = ∑ λl 1(T ,T
l l +1 ]
(t ), t ⩾ 0. (5.3.2)
l =0 l =0

i) Estimating the default rates from default bond prices.
From Proposition 5.4, we have

w Tk
Pd (t, Tk ) = 1{τ>t} exp − (r (u) + λ (u))du
t
!
k−1
= 1{τ>t} exp − ∑ (rl + λl )(Tl +1 − Tl ) ,
l =0
k = 1, 2, . . . , n, from which we can infer
1 Pd (t, Tk )
λk = −rk + log > 0, k = 0, 1, . . . , n − 1.
Tk+1 − Tk Pd (t, Tk+1 )
ii) Estimating (implied) default probabilities P∗ (τ < T | Gt ) from default rates.
Based on the expression

P∗ (τ > T | Gt ) = IE∗ 1{τ>T } | Gt

(5.3.3)
w
T
= 1{τ>t} IE exp − λu du
∗
Ft , 0 ⩽ t ⩽ T,
t
of the survival probability up to time T , see (5.2.1), and given the information known up to
time t, in terms of the hazard rate process(λu )u∈R+ adapted to a filtration (Ft )t∈R+ , we find
wT
P(τ > T | GTk ) = 1{τ>Tk } exp − λu du
Tk
!
n−1
= 1{τ>t} exp − ∑ λl (Tl +1 − Tl ) , k = 0, 1, . . . , n − 1,
l =k
where
Gt = Ft ∨ σ ({τ ⩽ u} : 0 ⩽ u ⩽ t ), t ⩾ 0,
i.e. Gt contains the additional information on whether default at time τ has occurred or not
before time t.
In Table 5.2, bond ratings are determined according to hazard (or failure) rate thresholds.
Bond Credit Moody’s S&P

Ratings Municipal Corporate Municipal Corporate
Aaa/AAAs 0.00 0.52 0.00 0.60
Aa/AA 0.06 0.52 0.00 1.50
A/A 0.03 1.29 0.23 2.91
Baa/BBB 0.13 4.64 0.32 10.29
Ba/BB 2.65 19.12 1.74 29.93
B/B 11.86 43.34 8.48 53.72
Caa-C/CCC-C 16.58 69.18 44.81 69.19
Investment Grade 0.07 2.09 0.20 4.14
Non-Invest. Grade 4.29 31.37 7.37 42.35
All 0.10 9.70 0.29 12.98
Table 5.2: Cumulative historic default rates (in percentage).*
* Sources: Moody’s, S&P.

Exercises
Exercise 5.1 Consider a standard zero-coupon bond with constant yield r > 0 and a defaultable
(risky) bond with constant yield rd and default probability α ∈ (0, 1). Find a relation between r, rd ,
α and the bond maturity T .
Exercise 5.2 A standard zero-coupon bond with constant yield r > 0 and maturity T is priced
P(t, T ) = e −(T −t )r at time t ∈ [0, T ]. Assume that the company can get bankrupt at a random time
t + τ, and default on its final $1 payment if τ < T − t.
a) Explain why the defaultable bond price Pd (t, T ) can be expressed as
Pd (t, T ) = e −(T −t )r IE∗ 1{τ>T −t} .

(5.3.4)
b) Assuming that the default time τ is exponentially distributed with parameter λ > 0, compute
the default bond price Pd (t, T ) using (5.3.4).
c) Find a formula that can estimate the parameter λ from the risk-free rate r and the market data
PM (t, T ) of the defaultable bond price at time t ∈ [0, T ].
Exercise 5.3 Consider an interest rate process (rt )t∈R+ and a default rate process (λt )t∈R+ ,
modeled according to the Vasicek processes

 drt = −art dt + σ dBt(1) ,
 dλ = −bλ dt + ηdB(2) ,
t t t
(1) (2)
where Bt t∈R+
and Bt t∈R are standard (Ft )t∈R+ -Brownian motions with correlation ρ ∈
+
(1) (2)
[−1, 1],and dBt • dBt = ρdt.
a) Taking r0 := 0, show that we have
wT wT
(1)
rs ds = C (a,t, T )rt + σ C (a, s, T )dBs ,
t t
and wT wT
(2)
λs ds = C (b,t, T )λt + η C (b, s, T )dBs ,
t t
where
1
C (a,t, T ) = − ( e −(T −t )a − 1).
a
b) Show that the random variable
wT wT
rs ds + λs ds
t t
is has a Gaussian distribution, and compute its conditional mean

w wT
T
∗
IE rs ds + λs ds Ft
t t
and variance w
T wT
Var rs ds + λs ds Ft ,
t t
conditionally to Ft .

Exercise 5.4 (Exercise 5.3 continued). Consider a (random) default time τ with cumulative
distribution function wt
P(τ > t | Ft ) = exp − λu du , t ⩾ 0,
0
where λt is a (random) default rate process which is adapted to the filtration (Ft )t∈R+ . Recall that
the probability of survival up to time T , given the information known up to time t, is given by
w
T
P(τ > T | Gt ) = 1{τ>t} IE exp − ∗
λu du Ft ,
t
where Gt = Ft ∨ σ ({τ < u} : 0 ⩽ u ⩽ t ), t ∈ R+ , is the filtration defined by adding the default

time information to the history (Ft )t∈R+ . In this framework, the price P(t, T ) of defaultable bond
with maturity T , short-term interest rate rt and (random) default time τ is given by
w
T
P(t, T ) = IE 1{τ>T } exp −
∗
ru du Gt (5.3.5)
t
w
T
= 1{τ>t} IE exp − (ru + λu )du
∗
Ft .
t
a) Give a justification for the fact that

w
T
∗
IE exp − (ru + λu )du Ft
t
can be written as a function F (t, rt , λt ) of t, rt and λt , t ∈ [0, T ].

b) Show that
wt w
T

∗
t 7−→ exp − (rs + λs )ds IE exp − (ru + λu )du Ft
0 t
is an Ft -martingale under P.
c) Use the Itô formula with two variables to derive a PDE on R2 for the function F (t, x, y).
d) Compute P(t, T ) from its expression (5.3.5) as a conditional expectation.
e) Show that the solution F (t, x, y) to the 2-dimensional PDE of Question (c)) is
F (t, x, y) = exp (−C (a,t, T )x −C (b,t, T )y)
2w
η2 w T 2

σ T
2
× exp C (a, s, T )ds + C (b, s, T )ds
2 t 2 t
wT
× exp ρσ η C (a, s, T )C (b, s, T )ds .
t
f) Show that the defaultable bond price P(t, T ) can also be written as
w
T
∗
P(t, T ) = e U (t,T )
P(τ > T | Gt ) IE exp − rs ds Ft ,
t
where
ση
U (t, T ) = ρ (T − t −C (a,t, T ) −C (b,t, T ) + C (a + b,t, T )) .
ab
g) By partial differentiation of log P(t, T ) with respect to T , compute the corresponding instan-
taneous short rate
∂
f (t, T ) = − log P(t, T ).
∂T

FN6813 Interest Rate Derivatives 133
h) Show that P(τ > T | Gt ) can be written using an HJM type default rate as
w
T
P(τ > T | Gt ) = 1{τ>t} exp − f2 (t, u)du ,
t
where
η2 2
f2 (t, u) = λt e −(u−t )b − C (b,t, u).
2
(1)
i) Show how the result of Question (f)) can be simplified when the processes Bt t∈R+
and
(2)
Bt t∈R are independent.
+

134 Exercise Solutions
Exercise Solutions
Chapter 1
Exercise 1.1
a) We have wt
a
= r0 d e −bt + d 1 − e −bt + σ d e −bt e bs dBs

drt
b w
0
w
t t
= −br0 e −bt dt + a e −bt dt + σ e −bt d e bs dBs + σ e bs dBs d e −bt
0 0
wt
= −br0 e −bt dt + a e −bt dt + σ e −bt e bt dBt − σ b e bs dBs e −bt dt
0
wt
−bt −bt
= −br0 e dt + a e dt + σ dBt − σ b e dBs e −bt dt bs
0 a
= −br0 e dt + a e dt + σ dBt − b rt − r0 e −bt − 1 − e −bt dt
−bt −bt
b
= (a − brt )dt + σ dBt ,
which shows that rt solves (1.4.29).
b) We note that wt
a
rt = r0 e −bt + 1 − e −bt + σ e −(t−u)b dBu

b 0
a
= r0 e −bs e −(t−s)b + e −(t−s)b 1 − e −bs

b ws wt
a −(t−s)b
+ σ e −(t−s)b e −(s−u)b dBu + σ e −(t−u)b dBu

+ 1− e
b 0
wt
s
−(t−s)b a −(t−s)b
−(t−u)b
= rs e + 1− e +σ e dBu , 0 ⩽ s ⩽ t.
b s
Hence, assuming that rs has the N (a/b, σ 2 /(2b)) distribution, the distribution of rt is
Gaussian with mean
a
IE[rt ] = e −(t−s)b IE[rs ] + 1 − e (t−s)b

b
a −(t−s)b a
+ 1 − e (t−s)b

= e
b b
a
= ,
b

and variance wt
h a i
Var[rt ] = Var rs e −(t−s)b + 1 − e (t−s)b + σ e −(t−u)b dBu

bw s
h t i
−(t−s)b −(t−u)b
= Var rs e +σ e dBu
s
h w t i
= Var rs e −(t−s)b + Var σ e −(t−u)b dBu

s
hw t i
−2(t−s)b 2
e −(t−u)b dBu

= e Var rs + σ Var
s
σ 2 wt
= e −2(t−s)b + σ 2 e −2(t−u)b du
2b s
σ 2 w t−s
= e −2(t−s)b + σ 2 e −2bu du
2b 0
σ2
= , t ⩾ 0.
2b
Exercise 1.2
a) The zero-coupon bond price P(t, T ) in the Vasicek model is given by
log P(t, T ) = A(T − t ) + rt C (T − t ), 0 ⩽ t ⩽ T,
where
1
1 − e −(T −t )b ,

C (T − t ) : = −
b
and
4ab − 3σ 2 σ 2 − 2ab
A(T − t ) : = + (T − t )
4b3 2b2
σ 2 − ab −(T −t )b σ 2 −2(T −t )b
+ e − 3e 0 ⩽ t ⩽ T. (A.1)
b3 4b
Since limT →∞ C (T − t )/(T − t ) = 0 and
lim A(T − t )/(T − t ) = (σ 2 − 2ab)/(2b2 ),

T →∞
we find
log P(t, T ) σ 2 − 2ab a σ 2
r∞ = − lim =− = − 2.
T →∞ T −t 2b2 b 2b
b) We have
P(t, T )
log = log P(t, T ) − log P(0, T )
P(0, T )
= A(T − t ) − A(T ) + rt C (T − t ) − r0C (T )
σ 2 − 2ab
2
σ 2 −(T −t )b

−(T −t )b σ − ab
= −t −e + 3e
2b2 b3 4b
2 2

σ σ − ab rt r0
+ e −bT − 1 − e −(T −t )b + 1 − e −bT ,

3
− 3
4b b b b
hence
a σ2

P(t, T ) rt − r0 rt − r0
lim log = − 2 t− =− + r∞t,
T →∞ P(0, T ) b 2b b b
and*
P(t, T ) 2 2
lim log = e −(rt −r0 )/b+t (a/b−σ /(2b )) = e −(rt −r0 )/b+r∞t ,
T →∞ P(0, T )
* The log function is continuous on (0, ∞).

which shows that the yield of the long-bond return is the asymptotic bond yield r∞ .
Remark: By Relations (1.4.15)-(1.4.17), the Vasicek bond price P(t, T ) can be rewritten in
terms of the asymptotic bond yield r∞ as
2C2 (T −t ) / (4b)
P(t, T ) = e −(T −t )r∞ +(rt −r∞ )C(T −t )−σ , 0 ⩽ t ⩽ T,
see e.g. Relation (3.12) in Brody, Hughston, and Meier, 2018.
Exercise 1.3 An estimator of σ can be obtained from the orthogonality relation

n−1 2 n−1
− σ 2 ∆t (r̃tl )2γ = σ 2 ∑ (r̃tl )2γ (Zl )2 − ∆t

∑ r̃tl +1 − a∆t − (1 − b∆t )r̃tl
l =0 l =0
≃ 0,
which is due to the independence of ttl and Zl , l = 0, . . . , n − 1, and yields
n−1 2
∑ r̃tl +1 − a∆t − (1 − b∆t )r̃tl
l =0
b2 =
σ .
n−1
∆t ∑ (r̃t ) l
2γ
l =0
Regarding the estimation of γ, we can combine the above relation with the second orthogonality
relation
n−1 2
∑ r̃tl+1 − a∆t − (1 − b∆t )r̃tl − σ 2 ∆t (r̃tl )2γ r̃tl

l =0
n−1
= σ 2 ∑ (r̃tl )2γ +1 (Zl )2 − ∆t

l =0
≃ 0,
cf. § 2.2 of Faff and Gray, 2006. One may also attempt to minimize the residual
n−1 2 2
∑ r̃tl +1 − a∆t − (1 − b∆t )r̃tl − σ 2 ∆t (r̃tl )2γ
l =0
by equating the following derivatives to zero, as

∂ n−1 2 2
∂ σ l =0
n−1 2
∑ (r̃t )2γ − σ 2 ∆t (r̃tl )2γ

= −4σ l
r̃tl +1 − a∆t − (1 − b∆t )r̃tl
l =0
= 0,
hence
n−1 2 n−1
∑ (r̃tl )2γ r̃tl+1 − a∆t − (1 − b∆t )r̃tl − σ 2 ∆t ∑ (r̃t )4γ = 0,
l
l =0 l =0
which yields
n−1 2
∑ (r̃t )2γ
l
r̃tl +1 − a∆t − (1 − b∆t )r̃tl
l =0
σb 2 = n−1
. (A.2)
∆t ∑ (r̃t )
l
4γ
l =0

We also have
n−1 2
∂ 2
∂γ l =0
n−1 2
= −4σ 2 ∆t ∑ (r̃t )2γ − σ 2 ∆t (r̃tl )2γ log r̃tl

l
r̃tl +1 − a∆t − (1 − b∆t )r̃tl
l =0
= 0,
which yields
n−1 2
∑ (r̃t )2γ
l
r̃tl +1 − a∆t − (1 − b∆t )r̃tl log r̃tl
l =0
σb 2 = n−1
, (A.3)
∆t ∑ (r̃t ) l
4γ
log r̃tl
l =0
and shows that γ can be estimated by matching Relations (A.2) and (A.3), i.e.
n−1 2
∑ (r̃t )2γl
r̃tl +1 − a∆t − (1 − b∆t )r̃tl
l =0
n−1
∑ (r̃t )4γ
l
l =0
n−1 2
∑ (r̃t )2γ l
r̃tl +1 − a∆t − (1 − b∆t )r̃tl log r̃tl
l =0
= n−1
.
4γ
∑ (r̃t ) l
log r̃tl
l =0
Remarks.
i) Estimators of a and b can be obtained by minimizing the residual
n−1
∑ (r̃t l +1
− a∆t − (1 − b∆t )r̃tl )2
l =0
as in the Vašíček, 1977 model, i.e. from the equations

n−1
∑ (r̃t l +1
− a∆t − (1 − b∆t )r̃tl ) = 0
l =0
and
n−1
∑ (r̃t l +1
− a∆t − (1 − b∆t )r̃tl ) r̃tl = 0.
l =0
ii) Instead of using the (generalised) method of moments, parameter estimation for stochastic
differential equations can be achieved by maximum likelihood estimation, see e.g. Lindström,
2007 and references therein.
Exercise 1.4
a) We have rt = r0 + at + σ Bt , and
F (t, rt ) = F (t, r0 + at + σ Bt ),

hence by Proposition 1.2 the PDE satisfied by F (t, x) is

∂F ∂F 1 ∂ 2F
−xF (t, x) + (t, x) + a (t, x) + σ 2 2 (t, x) = 0, (A.4)
∂t ∂x 2 ∂x
with terminal condition F (T , x) = 1.
wT
b) Using the relation rt = r0 + at + σ Bt and the fact that the stochastic integral (T − s)dBs
t
is independent of Ft , we have
wT
∗
F (t, rt ) = IE exp − rs ds Ft
t
wT wT
∗
= IE exp −r0 (T − t ) − a sds − σ Bs ds Ft
t t
wT
∗ −r0 (T −t )−a(T 2 −t 2 )/2
= IE e exp −(T − t )σ Bt − σ (T − s)dBs Ft
t
wT
−r0 (T −t )−a(T 2 −t 2 )/2−(T −t )σ Bt ∗
= e IE exp −σ (T − s)dBs Ft
t
wT
−r0 (T −t )−a(T −t )(T +t )/2−(T −t )σ Bt ∗
= e IE exp −σ (T − s)dBs
t
σ2 w T

2 2
= exp −(T − t )rt − a(T − t ) /2 + (T − s) ds
2 t
= exp −(T − t )rt − a(T − t )2 /2 + (T − t )3 σ 2 /6 ,

hence F (t, x) = exp −(T − t )x − a(T − t )2 /2 + (T − t )3 σ 2 /6 .

Note that the PDE (A.4) can also be solved by looking for a solution of the form F (t, x) =
e A(T −t )+xC(T −t ) , in which case one would find A(s) = −as2 /2 + σ 2 s3 /6 and C (s) = −s.
c) We check that the function F (t, x) of Question (b)) satisfies the PDE (A.4) of Question (a)),
F (T , x) = 1 and 2
since
σ
−xF (t, x) + x + a(T − t ) − (T − t )2 F (t, x) − a(T − t )F (t, x)
2
σ 2
+ (T − t )2 F (t, x) = 0.
2
Exercise 1.5 wt
a) We check from (1.4.34) and the differentiation rule d f (u)du = f (t )dt that
w 0
t 1
drt = αβ d St du + r0 dSt
0 Su
wt 1 wt 1
= αβ St d du + αβ dudSt + r0 dSt
0 Su 0 Su
St wt S dSt
t
= αβ dt + αβ du + r0 dSt
St 0 Su St
dSt
= αβ dt + (rt − r0 St ) + r0 dSt
St
dSt
= αβ dt + rt
St
= αβ dt + rt (−β dt + σ dBt )
= β (α − rt )dt + σ dBt , t ⩾ 0.
b) Taking µ (t, x) := β (α − x) and σ (t, x) = σ x, by Itô’s formula we have
rt rt rt

rt rt
= −rt e − 0 rs ds F (t, rt )dt + e − 0 rs ds
dF (t, rt )
rt rt
∂F
= −rt e − 0 rs ds F (t, rt )dt + e − 0 rs ds (t, rt )( µ (t, rt )dt + σ (t, rt )dBt )
∂x
rt ∂ 2F

− 0 rs ds 1 2 ∂F
+e σ (t, rt ) 2 (t, rt ) + (t, rt ) dt
2 ∂x ∂t
rt ∂F
= e − 0 rs ds σ (t, rt ) (t, rt )dBt
∂x
rt ∂ 2F

− 0 rs ds ∂F 1 2 ∂F
+e −rt F (t, rt ) + µ (t, rt ) (t, rt ) + σ (t, rt ) 2 (t, rt ) + (t, rt ) dt.
∂x 2 ∂x ∂t
(A.5)
rt
Given that t 7−→ e − 0 rs ds P(t, T ) is a martingale, the above expression (A.5) should only
contain terms in dBt , and all terms in dt should vanish therein. This leads us to the identities


rt F (t, rt )

∂F 1 ∂ 2F ∂F


 = µ (t, rt ) (t, rt ) + σ 2 (t, rt ) 2 (t, rt ) + (t, rt ) (A.6a)


∂x 2 ∂x ∂t



 rt rt
d e − 0 rs ds P(t, T ) = e − 0 rs ds σ (t, rt ) ∂ F (t, rt )dBt ,



∂x
and in particular to the bond pricing PDE
∂F 1 ∂ 2F ∂F
xF (t, x) = β (α − x) (t, x) + σ 2 x2 2 (t, x) + (t, x).
∂x 2 ∂x ∂t
Exercise 1.6
a) Applying the Itô formula
1 ′′
d f (rt ) = f ′ (rt )drt + f (rt )(drt )2
2
to the function f (x) = x2−γ with
f ′ (x) = (2 − γ )x1−γ and f ′′ (x) = (2 − γ )(1 − γ )x−γ ,
we have
dRt = drt2−γ
= d f (rt )
1 ′′
= f ′ (rt )drt + f (rt )(drt )2
2
f ′ (rt ) (β rt + αrt )dt + σ rt dBt drt +
γ−1 γ/2
=
1 ′′ γ−1 γ/2 2
f (rt ) (β rt + αrt )dt + σ rt dBt
2
σ 2 ′′
f ′ (rt ) (β rt + αrt )dt + σ rt dBt drt +
γ−1 γ/2 γ
= f (rt )rt dt
2
σ2
= (2 − γ )rt1−γ (β rtγ−1 + αrt )dt + σ rtγ/2 dBt + (2 − γ )(1 − γ )rtγ rt−γ dt
2
σ 2
= (2 − γ )(β + αrt2−γ )dt + (2 − γ )(1 − γ )dt + σ (2 − γ )rt1−γ/2 dBt
2

σ2 √
= (2 − γ ) β + (1 − γ ) + αRt dt + (2 − γ )σ Rt dBt .
2
2−γ
We conclude that the process Rt = rt follows the CIR equation
√
dRt = b(a − Rt )dt + η Rt dBt
2−γ
with initial condition R0 = r0 and coefficients
σ2

1
b = (2 − γ )α, a= β + (1 − γ ) , and η = (2 − γ )σ .
α 2
b) By Itô’s formula and the relation P(t, T ) = F (t, rt ), t ∈ [0, T ], we have
rt rt rt
rt rt
= −rt e − 0 rs ds F (t, rt )dt + e − 0 rs ds dF (t, rt )
rt rt ∂F rt ∂F
= −rt e − 0 rs ds F (t, rt )dt + e − 0 rs ds (t, rt )dt + e − 0 rs ds (t, rt )drt
∂t ∂x
rt 1 ∂ 2F
+ e− 0 rs ds (t, rt )(drt )2
2 ∂ x2
rt rt ∂F −(1−γ )
= −rt e − F (t, rt )dt + e − 0 rs ds
0 rs ds (t, rt )((β rt + αrt )dt + σ rtγ/2 dBt )
∂x
rt ∂ 2F

∂F 1
+ e − 0 rs ds (t, rt ) + σ 2 rtγ 2 (t, rt ) dt
∂t 2 ∂x
rt ∂F
= e − 0 rs ds σ rtγ/2 (t, rt )dBt
∂x
rt

−(1−γ ) ∂F
+ e − 0 rs ds −rt F (t, rt ) + (β rt + αrt ) (t, rt )
∂x
2

1 ∂ F ∂F
+ σ 2 rtγ 2 (t, rt ) + (t, rt ) dt. (A.7)
2 ∂x ∂t
rt
Given that t 7−→ e − 0 rs ds P(t, T ) is a martingale, the above expression (A.7) should only
contain terms in dBt and all terms in dt should vanish inside (A.7). This leads to the identities

 −(1−γ ) ∂F 1 2 γ ∂ 2F ∂F

 rt F (t, rt ) = ( β rt + αrt ) (t, rt ) + σ rt (t, rt ) + (t, rt )
∂x 2 ∂ x2


 ∂t
 rt

 rt ∂F
 d e − 0 rs ds P(t, T ) = σ e − 0 rs ds rtγ/2 (t, rt )dBt ,


∂x
and to the PDE
∂F ∂F σ 2 ∂ 2F
xF (t, x) = (t, x) + (β x−(1−γ ) + αx) (t, x) + xγ 2 (t, x).
∂t ∂x 2 ∂x
Exercise 1.7 rt
− 0 rs ds
a) The discounted bond prices process e F (t, rt ) is a martingale, and we have
rt
d e− 0 rs ds
F (t, rt )
rt σ 2 ∂ 2F

∂F ∂F
= e − 0 rs ds −rt F (t, rt )dt + (t, rt )dt + (t, rt )drt + (t, rt )( drt ) 2
∂t ∂x 2 ∂ x2

rt

− 0 rs ds ∂F ∂F √
= e −rt F (t, rt )dt + (t, rt )dt + (t, rt )(−art dt + σ rt dBt )
∂t ∂x
rt σ 2 ∂ 2F
+rt e − 0 rs ds (t, rt )dt,
2 ∂ x2
hence F (t, x) satisfies the affine PDE
∂F ∂F σ 2 ∂ 2F
−xF (t, x) + (t, x) − ax (t, x) + x (t, x) = 0. (A.9)
∂t ∂x 2 ∂ x2
b) Plugging F (t, x) = e A(T −t )+xC(T −t ) into the PDE (A.9) shows that
σ 2x 2

A(T −t )+xC (T −t ) ′ ′
e −x − A (T − t ) − xC (T − t ) − axC (T − t ) + C (T − t )
2
= 0.
Taking successively x = 0 and x = 1 in the above relation then yields the two equations
A′ (T − t ) = 0,




σ2
 −1 −C′ (T − t ) − aC (T − t ) + C2 (T − t ) = 0.


2
Remark: The initial condition A(0) = 0 shows that A(s) = 1, and it can be shown from the
condition C (0) = 0 that
2(1 − e γ (T −t ) )
C (T − t ) = , t ∈ [0, T ],
2γ + (a + γ )( e γ (T −t ) − 1)
√
with γ = a2 + 2σ 2 , see e.g. Eq. (3.25) page 66 of Brigo and Mercurio, 2006.
Exercise 1.8
a) The payoff of the convertible bond is given by Max(αSτ , P(τ, T )).
b) We have
Max(αSτ , P(τ, T )) = P(τ, T )1{αSτ ⩽P(τ,T ))} + αSτ 1{αSτ >P(τ,T ))}
= P(τ, T ) + (αSτ − P(τ, T ))1{αSτ >P(τ,T ))}
= P(τ, T ) + (αSτ − P(τ, T ))+
= P(τ, T ) + α (Sτ − P(τ, T )/α )+ ,
where the latter European call option payoff has the strike price K := P(τ, T )/α.
c) From the Markov property applied at time t ∈ [0, τ ], we will write the corporate bond price
as a function C (t, St , rt ) of the underlying asset price and interest rate, hence we have
h rτ i
C (t, St , rt ) = IE∗ e − t rs ds Max(αSτ , P(τ, T )) Ft .
The martingale
rt
property follows fromrt
the equalities
h rτ i
− 0 rs ds − 0 rs ds ∗
e C (t, St , rt ) = e IE e − t rs ds Max(αSτ , P(τ, T )) Ft
h rτ i
= IE∗ e − 0 rs ds Max(αSτ , P(τ, T )) Ft .
d) We have r
t

− 0 rs ds
d e C (t, St , rt )
rt rt ∂C
= −rt e − 0 rs dsC (t, St , rt )dt + e − 0 rs ds (t, St , rt )dt
∂t

rt ∂C (1)
+ e− 0 rs ds
(t, St , rt )(rSt dt + σ St dBt )
∂x
rt
− 0 rs ds ∂C (2)
+e (t, St , rt )(γ (t, rt )dt + η (t, St )dBt )
∂y
rt σ 2 ∂ 2C rt 1 ∂ 2C
+ e − 0 rs ds St2 2 (t, St , rt )dt + e − 0 rs ds η 2 (t, rt ) (t, St , rt )dt
2 ∂x 2 ∂ y2
rt ∂ 2C
+ρσ St η (t, rt ) e − 0 rs ds (t, St , rt )dt. (A.10)
rt
∂ x∂ y
The martingale property of e − 0 rs dsC (t, St , rt ) t∈R shows that the sum of terms in factor

+
of dt vanishes in the above Relation (A.10), which yields the PDE
∂C ∂C ∂C
0 = − yC (t, x, y) +
(t, x, y)dt + ry (t, x, y) + γ (t, y) (t, x, y)
∂t ∂x ∂y
σ 2 2
∂ C 2
1∂ C ∂ 2C
+ x2 2 (t, x, y) + η 2 (t, y) (t, x, y ) + ρσ xη (t, y ) (t, x, y),
2 ∂x 2 ∂ y2 ∂ x∂ y
with the terminal condition
C (τ, x, y) = Max(αx, F (τ, y)), where F (τ, rτ ) = P(τ, T )
is the bond pricing function.

e) The convertible
h r τ bond can be priced as i
IE e − t rs ds Max(αSτ , P(τ, T )) Ft
∗
h rτ i h rτ i
= IE∗ e − t rs ds P(τ, T ) Ft + α IE∗ e − t rs ds (Sτ − P(τ, T )/α )+ Ft
b (Sτ /P(τ, T ) − 1/α )+ Ft .

= P(t, T ) + αP(t, T )IE (A.11)
f) By Proposition 3.8 we find
dZt = (σ − σB (t ))Zt dW
bt ,
bt )t∈R+ is a standard Brownian motion under the forward measure P.

where (W b
g) By modeling (Zt )t∈R+ as the geometric Brownian motion
dZt = σ (t )Zt dW
bt ,
Relation (A.11) shows that the convertible bond is priced as
P(t, T ) + αSt Φ(d+ ) − P(t, T )Φ(d− ),
where
v2 (t, τ )

1 St
d+ = log + ,
v(t, T ) P(t, T ) 2
v2 (t, τ )

1 St
d− = log − ,
v(t, T ) P(t, T ) 2
rT
and v2 (t, T ) = t σ 2 (s, T )ds, 0 < t < T .
Exercise 1.9 We have

∂ ∂ 1 c 1
Pc (0, n) = + 1 −
∂r ∂r (1 + r )n r (1 + r )n


n c 1 nc
= − +
− 2 1− + ,
(1 + r ) n 1 r (1 + r )n r (1 + r )n+1
hence
1+r ∂
Dc (0, n) = − Pc (0, n)
Pc (0, n) ∂ r

n (1 + r )c 1 nc
− − 1− +
(1 + r )n r2 (1 + r )n r (1 + r )n
= −
1 c 1
+ 1−
(1 + r )n r (1 + r )n
1+r
−nr − (c((1 + r )n − 1)) + nc
= − r
r + c ((1 + r )n − 1)
1+r
1 + r − nr − (r + c((1 + r )n − 1)) + nc
= − r
r + c ((1 + r )n − 1)
1+r 1 + r + n(c − r )
= −
r r + c ((1 + r )n − 1)
c ((1 + r )n − 1)

(1 − c/r )n 1+r
= + ,
1 + c ((1 + r )n − 1) /r r r + c ((1 + r )n − 1)
with D0 (0, n) = n. We note that

1+r 1 + r + n(c − r ) 1
lim Dc (0, n) = lim − = 1+ .
n→∞ n→∞ r r + c ((1 + r )n − 1) r
When n becomes large, the duration (or relative sensitivity) of the bond price converges to 1 + 1/r
whenever the (nonnegative) coupon amount c is nonzero, otherwise the bond duration of Pc (0, n)
is n. In particular, the presence of a nonzero coupon makes the duration (or relative sensitivity)
of the bond price bounded as n increases, whereas the duration n of P0 (0, n) goes to infinity as n
increases.
As a consequence, the presence of the coupon tends to put an upper limit the risk and sensitivity
of bond prices with respect to the market interest rate r as n becomes large, which can be used for
bond immunization. Note that the duration Dc (0, n) can also be written as the relative average
!
n
1 n k
Dc (0, n) = +c ∑
Pc (0, n) (1 + r )n k =1 (1 + r )
k
of zero coupon bond durations, weighted by their respective zero-coupon prices.
Exercise 1.10
a) We have
1


 y0,1 = − log P(0, T1 ) = 9.53%,


 T1
 1
y0,2 = − log P(0, T2 ) = 9.1%,
 T2
1 P(0, T2 )


 y1,2 = − = 8.6%,

 log
T2 − T1 P(T1 , T2 )
with T1 = 1 and T2 = 2.

b) We have
Pc (1, 2) = ($1 + $0.1) × P0 (1, 2) = ($1 + $0.1) × e −(T2 −T2 )y1,2 = $1.00914,
and
Pc (0, 2) = ($1 + $0.1) × P0 (0, 2) + $0.1 × P0 (0, 1)
= ($1 + $0.1) × e −(T2 −T2 )y0,2 + $0.1 × e −(T2 −T2 )y0,1
= $1.00831.
Exercise 1.11 √
a) The discretization rtk+1 := rtk + (a − brtk )∆t ± σ ∆t does not lead to a binomial tree because
rt2 can be obtained in four different ways
√ from rt0 , as
rt2 = rt1 (1 − b∆t ) + a∆t ± σ ∆t
 √ √

 (rt0 (1 − b∆t ) + a∆t + σ ∆t )(1 − b∆t ) + a∆t + σ ∆t


√ √


 (rt0 (1 − b∆t ) + a∆t + σ ∆t )(1 − b∆t ) + a∆t − σ ∆t



= √ √
rt0 (1 − b∆t ) + a∆t − σ ∆t )(1 − b∆t ) + a∆t + σ ∆t





(

√ √


(rt0 (1 − b∆t ) + a∆t − σ ∆t )(1 − b∆t ) + a∆t − σ ∆t.


b) By the Girsanov Theorem, the process (rt /σ )t∈[0,T ] with
drt a − brt
= dt + dBt
σ σ
is a standard Brownian motion under the probability measure Q withRadon-Nikodym density
dQ 1 wT 1 wT
= exp − (a − brt )dBt − 2 (a − brt )2 dt
dP σ 0 2σ 0
1 wT 1 wT

2
≃ exp − 2 (a − brt )(drt − (a − brt )dt ) − 2 (a − brt ) dt
σ 0 2σ 0
1 wT 1 wT

= exp − 2 (a − brt )drt + 2 (a − brt )2 dt .
σ 0 2σ 0 √
In other words, if we generate (rt /σ )t∈[0,T ] and the increments σ −1 drt ≃ ± ∆t as a standard
Brownian motion under Q, then, under the probability measure P such that
w
1 wT

dP 1 T 2
= exp (a − brt )drt − 2 (a − brt ) dt ,
dQ σ2 0 2σ 0
the process
drt a − brt
dBt = − dt
σ σ
will be a standard Brownian motion under P, and the samples
drt = (a − brt )dt + σ dBt
of (rt )t∈[0,T ] will be distributed as a Vasicek process under P.

√
incrementσ drt under Qby ± ∆t, we have
c) Approximating thestandard Brownian −1
1 a − brt √ a − brt √
2T /∆T ∏ ± ∆t = ∏ 1 ± ∆t
0<t<T 2 2σ 0<t<T σ
!
a − brt √

= exp log ∏ 1 ± ∆t
0<t<T σ

!
a − brt √

= exp ∑ log 1 ± ∆t
0<t<T σ
!
a − brt √ 1 (a − brt )2
≃ exp ∑ σ ± ∆t − 2 ∑ σ2
∆t
0<t<T 0<t<T
w
1 wT

1 T 2
≃ exp (a − brt )drt − 2 (a − brt ) dt
σ2 0 2σ 0
dP
= .
dQ
d) We check that
IE[∆rt1 | rt0 ] = (a − brt0 )∆t
√ √
= p(rt0 )σ ∆t − (1 − p(rt0 ))σ ∆t
√ √
= σ p(rt0 ) ∆t − σ q(rt0 ) ∆t,
with
1 a − brt0 √ 1 a − brt0 √
p(rt0 ) = + ∆t and q(rt0 ) = − ∆t.
2 2σ 2 2σ
Similarly, we have
IE[∆rt2 | rt1 ] = (a − brt1 )∆t
√ √
= p(rt1 )σ ∆t − (1 − p(rt1 ))σ ∆t
√ √
= σ p(rt1 ) ∆t − σ q(rt1 ) ∆t,
with
1 a − brt1 √ 1 a − brt1 √
p(rt1 ) = + ∆t, q(rt1 ) = − ∆t.
2 2σ 2 2σ
Exercise 1.12
a) We have
100 100 100
P(1, 2) = IE∗ = + .
1 + r1 2(1 + r1u ) 2(1 + r1d )
b) We have
100 100
P(0, 2) = u + .
2(1 + r0 )(1 + r1 ) 2(1 + r0 )(1 + r1d )
c) We have P(0, 1) = 91.74 = 100/(1 + r0 ), hence
100 − P(0, 1)
r0 = = 100/91.74 − 1 = 0.090037061 ≃ 9%.
P(0, 1)
d) We have
P(0, 1) P(0, 1)
83.40 = P(0, 2) = +
2(1 + r1u ) 2(1 + r1d )
√
and r1u /r1d = e 2σ ∆t , hence
P(0, 1) P(0, 1)
83.40 = P(0, 2) = √ +
2(1 + r1d e 2σ ∆t ) 2(1 + r1d )
or
√ √ √

2σ ∆t P(0, 1) P(0, 1)
(r1d )2 + 2r1d e σ ∆t cosh σ ∆t 1 −

e +1− = 0,
2P(0, 2) P(0, 2)
and
√ √ P(0, 1)

∆t
r1d = e −σ cosh σ ∆t −1
2P(0, 2)

s 
2
√

P(0, 1) P(0, 1)
− 1 cosh2 σ ∆t +

± − 1
2P(0, 2) P(0, 2)
= 0.078684844 ≃ 7.87%,
and √
∆t
r1u = r1d e 2σ
√ √ P(0, 1)

σ ∆t
= e cosh σ ∆t −1
2P(0, 2)
s 
2
√

P(0, 1) P ( 0, 1 )
− 1 cosh2 σ ∆t +

± − 1
2P(0, 2) P(0, 2)
= 0.122174525 ≃ 12.22%.
We also find
√ r1d √ r1u

1 1
µ= σ ∆t + log = −σ ∆t + log = 0.085229181 ≃ 8.52%.
∆t r0 ∆t r0
Exercise 1.13
a) When n = 1 the relation (1.4.37) shows that fe(t,t, T1 ) = f (t,t, T1 ) with F (t, x) = c1 e −(T1 −1)x
and P(t, T1 ) = c1 e f (t,t,T1 ) , hence
1 ∂F
D(t, T1 ) := − (t, f (t,t, T1 )) = T1 − t, 0 ⩽ t ⩽ T1 .
P(t, T1 ) ∂ x
b) In general, we have
1 ∂F e
D(t, Tn ) = − t, f (t,t, Tn )
P(t, Tn ) ∂ x
n
1 ˜
= ∑ (Tk − t )ck e −(Tk −t ) f (t,t,Tn )
P(t, Tn ) k=1
n
= ∑ (Tk − t )wk ,
k =1
where
ck ˜
wk := e −(Tk −t ) f (t,t,Tn )
P(t, Tn )
˜
ck e −(Tk −t ) f (t,t,Tn )
= n
∑ cl e −(T −t ) f (t,t,T )
l l
l =1
˜
ck e −(Tk −t ) f (t,t,Tn )
= n , k = 1, 2, . . . , n,
−(Tl −t ) f˜(t,t,Tn )
∑ cl e
l =1
and the weights w1 , w2 , . . . , wn satisfy
n
∑ wk = 1.
k =1
c) We have
1 ∂ 2F e
C (t, Tn ) = 2
t, f (t,t, Tn )
P(t, Tn ) ∂ x
n
= ∑ (Tk − t )2 wk
k =1

n n
= ∑ (Tk − t − D(t, Tn ))2 wk + 2D(t, Tn ) ∑ (Tk − t )wk − (D(t, Tn ))2
k =1 k =1
n
= (D(t, Tn ))2 + ∑ (Tk − t − D(t, Tn ))2 wk
k =1
= (D(t, Tn )) + (S(t, Tn ))2 ,
2
where
n
(S(t, Tn ))2 := ∑ (Tk − t − D(t, Tn ))2 wk .
k =1
d) We have
n
1
D(t, Tn ) = ∑ ck B(Tk − t ) e A(Tk −t )+B(Tk −t ) fα (t,t,Tn )
P(t, Tn ) k=1
n
1
= c B(Tk − t ) e A(T −t )+B(T −t ) f
(t,t,T ) ∑ k
k k α (t,t,Tn )
e A(Tn −t )+B(Tn −t ) fα n
k =1
n
= ∑ ck B(Tk − t ) e A(T −t )−A(T −t )+(B(T −t )−B(T −t )) f
k n k n α (t,t,Tn )
.
k =1
e) We have
1 n

e −(Tn −t )b − e −(Tk −t )b fα (t,t,Tn )/b
1 − e −(Tk −t )b ck e A(Tk −t )−A(Tn −t )+

D(t, Tn ) = ∑
b k =1
1 n ( e −(Tn −t )b − e −(Tk −t )b )
−(Tk −t )b
A(T −t )−A(T −t )
= 1− e ck e k n
(P(t,t + α (Tn − t )) (Tn −t )αb .
b k∑
=1
Chapter 2
Exercise 2.1
a) By partial differentiation with respect to T under the expectation Ib
E, we have
∂P ∂ ∗ h − r T rs ds i
(t, T ) = IE e t Ft
∂T ∂ Th irT
= IE∗ − rT e − t rs ds
Ft
= −P(t, T )IE b [rT | Ft ].
b) As a consequence of Question (a)), we find
1 ∂P
f (t, T ) = − (t, T ) = IbE[rT | Ft ], 0 ⩽ t ⩽ T, (A.12)
P(t, T ) ∂ T
see Relation (22) page 10 of Mamon, 2004.

c) The martingale property of ( f (t, T ))t∈[0,T ] under the forward measure Ib
E follows from
Relation (A.12) and the tower property of conditional expectations.
Remark. In the Vasicek model, by (1.1.2) and (4.1.9) we have
a wT
rT = r0 e −bT + (1 − e −bT ) + σ e −(T −s)b dWs
b 0
a wT
= r0 e −bT + (1 − e −bT ) + σ e −(T −s)b dW
bs
b 0
σ 2 w T −(T −s)b
− e (1 − e −(T −s)b )ds
b 0
a wT
= r0 e −bT + (1 − e −bT ) + σ e −(T −s)b dW
bs
b 0

σ 2 w T −(T −s)b σ 2 w T −2(T −s)b

− e ds + e ds,
b 0 b 0
hence
a wt
E[rT | Ft ] = r0 e −bT + 1 − e −bT + σ e −(T −s)b dW

Ib bs
b 0
σ 2 w T −(T −s)b σ 2 w T −2(T −s)b
− e ds + e ds
b 0 b w0
a t
= r0 e −bT + 1 − e −bT + σ e −(T −s)b dWs

b 0
σ 2 w t −(T −s)b
1 − e −(T −s)b ds

+ e
b 0
σ 2 w T −(T −s)b σ 2 w T −2(T −s)b
− e ds + e ds
b 0 b 0
a a
= r0 e −bT + 1 − e −bT + e −(T −t )b rt − r0 e −bt − 1 − e −bt

b b
σ 2 w t σ 2 w t
+ e −(T −s)b ds − e −2(T −s)b ds
b 0 b 0
σ 2 w T −(T −s)b σ 2 w T −2(T −s)b
− e ds + e ds
b 0 b 0
a a σ 2 w T −(T −s)b σ 2 w T −2(T −s)b
= + e −(T −t )b rt − − e ds + e ds
b b b t b t
a a σ 2 w T −t −bs σ 2 w T −t −2bs
= + e −(T −t )b rt − − e ds + e ds
b b b 0 b 0
a a σ2 σ2
+ e −(T −t )b rt − − 2 1 − e −(T −t )b + 2 1 − e −2(T −t )b

=
b b b b
a σ2 2 2

a σ σ
= − + e −(T −t )b rt − + 2 − 2 e −2(T −t )b ,
b 2b2 b b b
which recovers (2.4.1).

w
T2
P(0, T2 ) = exp − f (t, s)ds = e −r1 T1 −r2 (T2 −T1 ) , t ∈ [0, T2 ],
0
and w
T2
P(T1 , T2 ) = exp − f (t, s)ds = e −r2 (T2 −T1 ) , t ∈ [0, T2 ],
T1
from which we deduce

1
r2 = − log P(T1 , T2 ),
T2 − T1
and
T2 − T1 1
r1 = −r2 − log P(0, T2 )
T1 T1
1 1
= log P(T1 , T2 ) − log P(0, T2 )
T1 T1
1 P(0, T2 )
= − log .
T1 P(T1 , T2 )
Exercise 2.3

a) In the Vašíček, 1977 model,

by w(1.4.15) we have
T
P(t, T ) = IE exp − rs ds
t
w wT
T
= IE exp − h(s)ds − Xs ds
t t
w w
T T
= exp − h(s)ds IE exp − Xs ds
t t
w
T
= exp − h(s)ds + A(T − t ) + Xt C (T − t ) , 0 ⩽ t ⩽ T ,
t
w
T
hence, since X0 = 0 we find P(0, T ) = exp − h(s)ds + A(T ) .
0
b) By the identification w
T
P(t, T ) = exp − h(s)ds + A(T − t ) + Xt C (T − t )
t
w
T
= exp − f (t, s)ds ,
t
we find wT wT
h(s)ds = f (t, s)ds + A(T − t ) + Xt C (T − t ),
t t
and by differentiation with respect ot T this yields
h(T ) = f (t, T ) + A′ (T − t ) + Xt C′ (T − t ), 0 ⩽ t ⩽ T,
where
4ab − 3σ 2 σ 2 − 2ab σ 2 − ab −b(T −t ) σ 2 −2b(T −t )
A(T − t ) = + ( T − t ) + e − 3e .
4b3 2b2 b3 4b
Given an initial market data curve f M (0, T ), the matching f M (0, T ) = f (0, T ) can be
achieved at time t = 0 by letting
σ 2 − 2ab σ 2 − ab −bT σ 2 −2bT

h(T ) := f M (0, T ) + A′ (T ) = f M (0, T ) + − e + e ,
2b2 b2 2b2
T > 0. Note however that in general, at time t ∈ (0, T ] we will have
h(T ) = f (t, T ) + A′ (T − t ) + Xt C′ (T − t ) = f M (0, T ) + A′ (T ),
and the relation
f (t, T ) = f M (0, T ) + A′ (T ) − A′ (T − t ) − Xt C′ (T − t ), t ∈ [0, T ],
will allow us to match market data at time t = 0 only, i.e. for the initial curve. In any case,
model calibration is to be done at time t = 0.
Exercise 2.4
a) From the definition
1 1
L(t,t, T ) = −1
T −t P(t, T )
we have
1
P(t, T ) = .
1 + (T − t )L(t,t, T )

and similarly
1
P(t, S) = .
1 + (S − t )L(t,t, S)
Hence we get
1 P(t, T )
L(t, T , S) = −1
S − T P(t, S)

1 1 + (S − t )L(t,t, S)
= −1
S − T 1 + (T − t )L(t,t, T )

1 (S − t )L(t,t, S) − (T − t )L(t,t, T )
=
S−T 1 + (T − t )L(t,t, T )
b) When T = 1 year and L(0, 0, T ) = 2%, L(0, 0, 2T ) = 2.5%, we find

1 2T L(0, 0, 2T ) − T L(0, 0, T ) 2 × 0.025 − 0.02
L(t, T , S) = = = 2.94%,
T 1 + T L(0, 0, T ) 1 + 0.02
so we would prefer paying the spot rate at L(T , T , 2T ) = 2% instead of locking a forward
contract with rate L(0, T , 2T ) = 2.94%.

a) By Definition 2.1, we have
1
f (t, T , S) = (log P(t, T ) − log P(t, S))
S−T
σ2 σ2

1
= −(T − t )rt + (T − t )3 − −(S − t )rt + (S − t )3
S−T 6 6
1 σ 2
(T − t )3 − (S − t )3 .

= rt +
S−T 6
b) We have
f (t, T ) = lim f (t, T , S)
S↘T
∂
= − log P(t, T )
∂T
σ2

∂ 3
= − −(T − t )rt + (T − t )
∂T 6
σ2
= rt − (T − t )2 .
2
c) We have
dt f (t, T ) = (T − t )σ 2 dt + σ dBt .
rT
d) The HJM condition (2.3.5) is satisfied since the drift of dt f (t, T ) equals σ t σ ds.
Exercise 2.6
a) We have
wt
(1)
Xt = X0 e −bt + σ e −(t−s)b dBs
0
and wt
(2)
Yt = Y0 e −bt + σ e −(t−s)b dBs , t ∈ R+ ,
0
see (1.1.2).

b) We have
σ2
Var[Xt ] = Var[Yt ] = (1 − e −2bt ), t ∈ R+ ,
2b
see page 2, and therefore wt wt

(1) (2)
Cov(Xt ,Yt ) = Cov X0 e −bt + σ e −(t−s)b dBs ,Y0 e −bt + σ e −(t−s)b dBs
0
w t wt 0
( 1 ) ( 2 )
= σ 2 Cov e −(t−s)b dBs , e −(t−s)b dBs
0 0
hw t wt i
( 1 ) (2)
= σ 2 IE e −(t−s)b dBs e −(t−s)b dBs
0 0
wt
= ρσ 2 e −2(t−s)b ds
0
σ2
1 − e −2bt , t ∈ R+ .

= ρ
2b
c) We have
Cov(log P(t, T1 ), log P(t, T2 ))

= Cov log F1 (t, Xt , T1 )F2 (t,Yt , T2 ) e ρU (t,T1 ) , log F1 (t, Xt , T1 )F2 (t,Yt , T2 ) e ρU (t,T2 )

= Cov C1T1 + Xt AT11 + C2T1 + Yt AT21 ,C1T2 + Xt AT12 + C2T2 + Yt AT22

= Cov Xt AT11 + Yt AT21 , Xt AT12 + Yt AT22

= AT11 AT12 Var[Xt ] + AT21 AT22 Var[Yt ] + AT11 AT22 + AT12 AT21 Cov(Xt ,Yt )

= AT11 AT12 + AT21 AT22 + ρ AT11 AT22 + AT12 AT21 Var[Xt ].

When Cov(Xt ,Yt ) = ρ Var[Xt ] = ρ Var[Yt ], we find the correlation

Cov(log P(t, T1 ), log P(t, T2 )) = p
Var[log P(t, T1 )] Var[log P(t, T2 )]
A1 A1 + A2 A2 + ρ AT11 AT22 + AT12 AT21
T1 T2 T1 T2
= q 2 2
AT11 + AT21 + ρ AT11 AT21 + AT11 AT21

1
×q .
T2 2 T2 2
A1 + A2 + ρ AT12 AT22 + AT12 AT22
When ρ = 1, we find
AT11 AT12 + AT21 AT22 + AT11 AT22 + AT12 AT21
= q 2 2 q 2 2
AT11 + AT21 + AT11 AT21 + AT11 AT21 AT12 + AT22 + AT12 AT22 + AT12 AT22
AT11 + AT21 AT12 + AT22

=
AT11 + AT21 AT12 + AT22
= ±1.
For example, if AT11 = 4, AT12 = 1, AT21 = 1 and AT22 = 4, we find
8 + 17ρ
Cov(log P(t, T1 ), log P(t, T2 )) = .
17 + 8ρ

1
0.5
Correlation
-0.5
-1
-1 -0.5 0 0.5 1
ρ
Figure S.1: Log bond prices correlation graph in the two-factor model.
Exercise 2.7
a) We have
wt wt
f (t, x) = f (0, x) + α s2 ds + σ ds B(s, x) = r + αtx2 + σ B(t, x).
0 0
b) We have
rt = f (t, 0) = r + B(t, 0) = r.
c) We have w
T
P(t, T ) = exp − f (t, s)ds
t
w T −t w T −t
2
= exp −(T − t )r − αt s ds − σ B(t, x)dx
0 0

α w T −t
3
= exp −(T − t )r − t (T − t ) − σ B(t, x)dx , t ∈ [0, T ].
3 0
d) Using (2.6.3),
" we have #
w T −t 2 w T −t w T −t
IE B(t, x)dx = IE[B(t, x)B(t, y)]dxdy
0 0 0
w T −t w T −t
= t min(x, y)dxdy
0 0
w T −t w y
1
xdxdy = t (T − t )3 .
= 2t
0 0 3
e) By Question (d)) we have
α w T −t
3
IE[P(t, T )] = IE exp −(T − t )r − t (T − t ) − σ B(t, x)dx
3 0
α w T −t
3
= exp −(T − t )r − t (T − t ) IE exp −σ B(t, x)dx
3 0
2 w
α σ T −t
3
= exp −(T − t )r − t (T − t ) exp Var B(t, x)dx
3 2 0
σ2

α 3 3
= exp −(T − t )r − t (T − t ) + t (T − t ) , 0 ⩽ t ⩽ T.
3 6
f) By Question (e)) we check that the required relation is satisfied if
α σ2
− t (T − t )3 + t (T − t )3 = 0,
3 6

i.e. α = σ 2 /2.
Remark: In order to derive an analog of the HJM absence of arbitrage condition in this stochas-
tic string model, one would have to check whether the discounted bond price e −rt P(t, T ) can
be a martingale by doing stochastic calculus with respect to the Brownian sheet B(t, x).
g) We have w
T
+
IE exp − rs ds (P(T , S) − K )
0
w S−T
" + #
−rT α 3
= e IE exp −(S − T )r − T (S − T ) + σ B(T , x)dx − K
3 0
= e −rT IE (x e m+X − K )+ ,

where x = e −(S−T )r , m = −αT (S − T )3 /3, and

w S−T
X =σ B(T , x)dx ≃ N (0, σ 2t (T − t )3 /3).
0
2
Given the wα = σ /2, this yields
relation
T
IE exp − rs ds (P(T , S) − K )+
0
!
q
log ( e −(S−T )r /K )
= e −rS Φ σ T (S − T )3 /12 + p
σ T (S − T )3 /3
!
q
log ( e −(S−T )r /K )
−K e −rT Φ −σ T (S − T )3 /12 + p
σ T (S − T )3 /3
!
log( e −(S−T )r /K )
q
= P(0, S)Φ σ T (S − T ) /12 + p
3
σ T (S − T )3 /3
!
q
log ( e −(S−T )r /K )
−KP(0, T )Φ −σ T (S − T )3 /12 + p .
σ T (S − T )3 /3
Chapter 3
Exercise 3.1
a) We have
Xt
d Xbt = d
Nt
X0 (σ −η )Bt −(σ 2 −η 2 )t/2
= d e
N0
X0 2 2
= (σ − η ) e (σ −η )Bt −(σ −η )t/2 dBt
N0
X0 2 2
+ (σ − η )2 e (σ −η )Bt −(σ −η )t/2 dt
2N0
X0 2 2
− (σ 2 − η 2 ) e (σ −η )Bt −(σ −η )t/2 dt
2N0
Xt Xt Xt
= − (σ 2 − η 2 )dt + (σ − η )dBt + (σ − η )2 dt
2Nt Nt 2Nt
Xt Xt
= − η (σ − η )dt + (σ − η )dBt
Nt Nt

Xt
= (σ − η )(dBt − ηdt )
Nt
Xt
= (σ − η ) d Bbt = (σ − η )Xbt d Bbt ,
Nt
where d Bt = dBt − ηdt is a standard Brownian motion under P.
b b
b) By change of numéraire, we have

+ N0
(XT − λ NT ) = N0 IbE (XbT − λ )+ .
+

IE[(XT − λ NT ) ] = IE
b
NT
Next, by the result of Question (a)), Xbt is a driftless geometric Brownian motion with volatility
σ − η under P, b hence we have
√ ! √ !
log ( X /λ ) σ T log ( X /λ ) σ T
b [(XbT − λ )+ ] = Xb0 Φ √0 −λΦ √0
b b b b
IE + − ,
b T
σ 2 b T
σ 2
by the Black-Scholes formula with zero interest rate and volatility parameter σb = σ − η. By
multiplication by N0 and the relation X0 = N0 X+0 we obtain (3.5.5), i.e.
b
+

IE (XT − λ NT ) = N0 IE (XT − λ )
b b
= N0 Xb0 Φ(d+ ) − λ N0 Φ(d − )
= X0 Φ(d+ ) − λ N0 Φ(d − ).
c) We have σb = σ − η.
Exercise 3.2
a) By the Girsanov Theorem 3.7, the processes
(1) (1) 1 (1) (1) 1 (2) (1) (1)

d Bbt = dBt − dNt • dBt = dBt − (2) dSt • dBt = dBt − ηρdt,
Nt S t
and
(2) (2) 1 (2) (2) 1 (2) (2) (2)
d Bbt = dBt − dNt • dBt = dBt − (2) dSt • dBt = dBt − ηdt
Nt S t
are both standard Brownian motions (and martingales) under P

b 2.
b) We have !
(1)
( 1 ) St
d Sbt = d (2)
St
(1)
S0 (1) (2)
−ηBt −(σ 2 −η 2 )t/2
d e σ Bt

= (2)
S0
(1)
S0 (1) (2)
−ηBt −(σ 2 −η 2 )t/2
= (2)
e σ Bt
S0
σ2 η2 σ2 − η2

(1) (2)
× σ dBt + dt − ηdBt + dt − dt − σ ηρdt
2 2 2
(1) (1) (2)
= Sbt σ d Bbt − ηd Bbt .
(1)
c) We note that the driftless geometric Brownian motion (Sbt )t∈R can be written as
(1) (1) b
d Sbt = σb Sbt dW t,

where (Wbt )t∈R is a standard Brownian motion under P b 2 . In order to determine σ

b we note
that
σb 2 dt = σ b 2 dW bt • dW bt
( 1 ) (1)
d Sbt d Sbt
= (1)
•
(1)
Sb Sbt
t
(1) (2) (1) (2)
= σ dBt − ηd Bbt · σ dBt − ηd Bbt
= (σ 2 + η 2 − 2σ ηρ )dt,
hence σb 2 = σ 2 + η 2 − 2σ ηρ. We conclude by applying the change of numéraire formula
(1) (2) + (2) (1) +
e −rT IE∗

ST − λ ST = S0 IbE SbT − λ
(1)
and the Black-Scholes formula to the the driftless geometric Brownian motion (Sbt )t∈R .
Exercise 3.3 We have Nt = P(t, T ) and from (1.4.8) and the relations P(t, T ) = F (t, rt ) and
P(t, S) = G(t, rt ) we find

dP(t, S) ∂
= rt dt + σ (t, rt ) log G(t, rt )dWt ,


 P(t, S) ∂x


dNt dP(t, T )

 ∂
= = rt dt + σ (t, rt ) log F (t, rt )dWt .


P(t, T )

Nt ∂x
By the Girsanov Theorem (3.2.12) we also have
bt = dWt − dNt ∂
dW • dW = dW − σ (t, r ) log F (t, rt )dt,
t t t
Nt ∂x
hence
dP(t, S) ∂ ∂ ∂
= rt dt + σ 2 (t, rt ) log F (t, rt ) log G(t, rt )dt + σ (t, rt ) log G(t, rt )dW
bt .
P(t, S) ∂x ∂x ∂x
Using the relation P(t, S) = G(t, rt ) we can also write
∂ ∂ ∂
dP(t, S) = rt P(t, S)dt + σ 2 (t, rt ) log F (t, rt ) G(t, rt )dt + σ (t, rt ) G(t, rt )dW
bt .
∂x ∂x ∂x
Exercise 3.4 Forward contract. Taking Nt := P(t, T ), t ∈ [0, T ], we have

w
T (P(T , S ) − K )
IE∗ exp − rs ds (P(T , S) − K ) Ft = Nt IbE Ft
t NT

= P(t, T )IE b (P(T , S) − K ) Ft
P(T , T )
= P(t, T )IE b [P(T , S) − K | Ft ]
b [P(T , S) | Ft ] − KP(t, T )
= P(t, T )IE

P(T , S )
= P(t, T )IE
b | Ft − KP(t, T )
P(T , T )
P(t, S)
= P(t, T ) − KP(t, T )
P(t, T )

= P(t, S) − KP(t, T ),
since
P(t, T ) P(t, S)
t 7−→ =
Nt P(t, T )
is a martingale under the forward measure P.
b The corresponding (static) hedging strategy is given
by buying one bond with maturity S and by short selling K units of the bond with maturity T .
Remark: The above result can also be obtained by a direct argument using the tower property of
conditional expectations:
w
T
∗
IE exp − rs ds (P(T , S) − K ) Ft
t
w w
T S
∗ ∗
= IE exp − rs ds IE exp − rs ds FT − K Ft
t T
w w
T S
= IE∗ exp − rs ds IE∗ exp − rs ds − K FT Ft
t T
w w
S T
∗ ∗
= IE IE exp − rs ds − K exp − rs ds FT Ft
t t
w w
S T
= IE∗ exp − rs ds − K exp − rs ds Ft
t t
= P(t, S) − KP(t, T ), t ∈ [0, T ].
Exercise 3.5
a) We choose Nt := St as numéraire because this allows us to write the option payoff as
(ST (ST − K ))+ = NT (ST − K )+ . In this case, the forward measure P
b satisfies
dP
b NT ST
= e −rT = e −rT ,
dP N0 S0
or
dP
b |F NT ST
t
= e −(T −t )r = e −(T −t )r , 0 ⩽ t ⩽ T.
dP|Ft Nt St
b) By the change of numéraire formula of Proposition 3.5, the option price becomes
−(T −t )r ∗
IE (ST (ST − K ))+ Ft = IE∗ e −(T −t )r NT (ST − K )+ Ft

e
= Nt IbE (ST − K )+ Ft

= St IbE (ST − K )+ Ft .

(A.13)
c) In order to compute (A.13) it remains to determine the dynamics of (St )t∈R+ under P. Since
b
2
St = S0 e σ Bt +rt−σ t/2 , we have
dP
b ST 2
= e −rT = e σ BT −σ T /2 ,
dP S0
hence by the Girsanov Theorem, Bbt := Bt − σt is a standard Brownian motion under P,

b with
rT +σ BT −σ 2 T /2
ST = S0 e
2 )T +σ B
bT −σ 2 T /2
= S0 e ( r + σ
2 )(T −t )+(B
bT −Bbt )σ −(T −t )σ 2 /2
= St e (r+σ , 0 ⩽ t ⩽ T.

d) According to the above, (A.13) becomes

+ +
e −(T −t )r IE∗ ST (ST − K ) Ft = St Ib Ft

E ST − K
b S0 e rT +σ 2 T +σ BbT −σ 2 T /2 − K + Ft

= St IE
b St e (r+σ 2 )(T −t )+(BbT −Bbt )σ −(T −t )σ 2 /2 − K + Ft

= St IE
2
= St e (T −t )(r+σ ) Bl(St , K, r + σ 2 , σ , T − t ), 0 ⩽ t ⩽ T,
since the Black-Scholes formula with interest rate r + σ 2 reads
2
b St e (r+σ 2 )(T −t )+(BbT −Bbt )σ −(T −t )σ 2 /2 − K + Ft
e −(T −t )(r+σ ) IE

= Bl(St , K, r + σ 2 , σ , T − t ), 0 ⩽ t ⩽ T.
Remarks:
i) The option price can be rewritten using other Black-Scholes parametrizations, such as for
example
2
St Bl St e (T −t )(r+σ ) , K, 0, σ , T − t ,

or
2 2
St e (T −t )(r+σ ) Bl St , K e −(T −t )(r+σ ) , 0, σ , T − t ,

however we prefer to choose the simplest possibility.

ii) Deflated (or forward) processes such as St /N1 = 1 or
e −(T −t )r ∗
IE (ST (ST − K ))+ Ft = Ib
E (ST − K )+ Ft ,

0 ⩽ t ⩽ T,
Nt
are martingales under the forward measure P. b

iii) This option can also be priced via an integral calculation instead of using change of numéraire,
as follows:
e −(T −t )r IE∗ [ST (ST − K )+ | Ft ]
2
= e −(T −t )r IE∗ St e (T −t )r+(BT −Bt )σ −(T −t )σ /2

2
×(St e (T −t )r+(BT −Bt )σ −(T −t )σ /2 − K )+ Ft

2 2
= St e −(T −t )σ /2 IE∗ (St e (T −t )r+2(BT −Bt )σ −(T −t )σ /2 − K e (BT −Bt )σ )+ Ft

2 2
= St e −(T −t )σ /2 IE∗ (x e (T −t )r+2(BT −Bt )σ −(T −t )σ /2 − K e (BT −Bt )σ )+ x=S

t
−(T −t )σ 2 /2 ∗
m(x)+2X X +

= St e IE ( e − K e ) x=S , 0 ⩽ t ⩽ T,
t
where X ≃ N (0, v ) with v = (T −t )σ and m(x) = (T −t )r − (T −t )σ 2 /2 + log x. Next,
2 2 2
we note that
+ 1 w∞ + 2 2
IE e m+2X − K e X e m+2x − K e x e −x /(2v ) dx

=√
2πv −∞2
1 w∞ 2 2
=√ ( e m+2x − K e x ) e −x /(2v ) dx
2πv2 −m+log K
em w ∞ 2 2 K w∞ 2 2
=√ e 2x−x /(2v ) dx − √ e x−x /(2v ) dx
2πv2 −m+log K 2πv2 −m+log K
e m+2v ∞
2 w 2 2 2 K e v /2 w ∞
2
2 2 2
=√ e −(2v −x) /(2v ) dx − √ e −(v −x) /(2v ) dx
2πv 2 −m + log K 2π −m + log K
m + 2v 2 w v 2 /2 w
e ∞
−x2 /(2v2 ) Ke ∞ 2 2
=√ 2 −m+log K
e dx − √ 2 −m+log K
e −x /(2v ) dx
2πv 2 −2v 2πv 2 −v
2 w
K e v /2 w ∞
m + 2v 2
e ∞
−x2 /2 2
= √ 2 −m+log K ) /v
e dx − √ 2 −m+log K ) /v
e −x /2 dx
2π ( −2v 2π ( −v


m+2v2 m − log K v2 /2 m − log K
=e Φ 2v + −K e Φ v+ ,
v v
hence +
e −(T −t )r IE∗ ST ST − K Ft

2 +
= St e −(T −t )σ /2 IE∗ e m(x)+2X − K e X

x=St
(T − t )(r + σ ) + (T − t )σ 2 /2 + log(St /K )
2

2
= St2 e (T −t )(r+σ ) Φ √
σ T −t
(T − t )(r + σ ) − (T − t )σ 2 /2 + log(St /K )
2

−KSt Φ √
σ T −t
2
= St e (T −t )(r+σ ) Bl(St , K, r + σ 2 , σ , T − t ), 0 ⩽ t ⩽ T.
Exercise 3.6
2
a) Knowing that St = S0 e σWt +rt−σ t/2 , we check that the discounted numéraire process
2
e −rt Nt := Stn e −(n−1)nσ t/2−nrt
2 t/2 2 t/2−nrt
= S0n e nσWt +nrt−nσ e −(n−1)nσ
2
= S0 e nσWt −(nσ ) t/2 , 0 ⩽ t ⩽ T,
is a martingale under P∗ , and
dP
b
−rT NT STn −nσ 2 T −nrT 2
∗
= e = n e = e nσWT −(nσ ) T /2 . (A.14)
dP N0 S0
bt := Wt − σ nt is a standard
b) From Equation (A.14) and the Girsanov theorem, the process W
Brownian motion under P, and we have
b
2
St = S0 e rt +σWt −σ t/2
2 t/2
= S0 e rt +σ (Wt +nσt )−σ
b
2 )t +σ W
bt +σ 2 t/2
= S0 e (r+nσ , t ⩾ 0.
c) We have
e −(T −t )r IE∗ STn 1{ST ⩾K} Ft = IE∗ e −(T −t )r STn 1{ST ⩾K} Ft

= e (n−1)nσ T /2+(n−1)rT IE∗ NT 1{ST ⩾K} Ft

2
= e (n−1)nσ T /2+(n−1)rT Nt IbE 1{ST ⩾K} Ft

2
2 T /2+(n−1)rT
= e (n−1)nσ Nt P
b (ST ⩾ K | Ft )
log(St /K ) + (r + (n − 1/2)σ 2 )(T − t )

2
= e (n−1)nσ T /2+(n−1)rT Nt Φ √
σ T −t
) + (r + (n − 1/2)σ 2 )(T − t )

2
n (n−1)nσ (T −t )/2+(n−1)r (T −t ) log ( St /K
= St e Φ √ ,
σ T −t
0 ⩽ t ⩽ T , n ⩾ 0.
d) The power call option with payoff (STn − K n )+ is priced at time t ∈ [0, T ] as
+
e −(T −t )r IE∗ STn − K n Ft

log(St /K ) + (r + (n − 1/2)σ 2 )(T − t )

2
= Stn e (n−1)nσ (T −t )/2+(n−1)r(T −t ) Φ √
σ T −t
2

n −(T −t )r log(St /K ) + (r − σ /2)(T − t )
−K e Φ √ , n ⩾ 1.
σ T −t

Exercise 3.7 Bond options.

a) Itô’s formula
yields
P(t, S) P(t, S) S
ζ (t ) − ζ T (t ) (dWt − ζ T (t )dt )

d =
P(t, T ) P(t, T )
P(t, S) S
ζ (t ) − ζ T (t ) dW

= bt , (A.15)
P(t, T )
where (W b by the Girsanov Theorem.
b) From (A.15) or (4.1.7) we have
w
1wt S

P(t, S) P(0, S) t
S T
T 2
= exp ζ (s) − ζ (s) dWs − b ζ (s) − ζ (s) ds ,
P(t, T ) P(0, T ) 0 2 0
hence
w
1wu S

P(u, S) P(t, S) u
S T
T 2
= exp ζ (s) − ζ (s) dWs −
b ζ (s) − ζ (s) ds ,
P(u, T ) P(t, T ) t 2 t
t ∈ [0, u], and for u = T this yields
w
1wT S

P(t, S) T
S T
T 2
P(T , S ) = exp ζ (s) − ζ (s) dWs −
b ζ (s) − ζ (s) ds ,
P(t, T ) t 2 t
since P(T , T ) = 1. Let P
b denote the forward measure associated to the numéraire
Nt := P(t, T ), 0 ⩽ t ⩽ T.
c) For all Sh⩾ Tr > 0 we have
T
i
IE e − t rs ds (P(T , S) − K )+ Ft
wT
" + #
P (t, S ) 1 2
= P(t, T )IE b exp X − ζ S (s) − ζ T (s) ds − K Ft
P(t, T ) 2 t
b e X +m(t,T ,S) − K + Ft ,

= P(t, T )IE
where X is a centered Gaussian random variable with variance
wT 2
v2 (t, T , S) = ζ S (s) − ζ T (s) ds
t
given Ft , and
1 P(t, S)
m(t, T , S) = − v2 (t, T , S) + log .
2 P(t, T )
Recall that when X is a centered Gaussian random variable with variance v2 , the expectation
of ( e m+X − K )+ is given, as in the standard Black-Scholes formula, by
2 /2
IE[( e m+X − K )+ ] = e m+v Φ(v + (m − log K )/v) − KΦ((m − log K )/v),
where wz dy
2 /2
Φ (z) = e −y√ , z ∈ R,
−∞ 2π
denotes the Gaussian cumulative distribution function and for simplicity of notation we
dropped the indices t, T , S in m(t, T , S) and v2 (t, T , S).
Consequently
h r Twe have i
IE e − t rs ds (P(T , S) − K )+ Ft

v 1 P(t, S) v 1 P(t, S)
= P(t, S)Φ + log − KP(t, T )Φ − + log .
2 v KP(t, T ) 2 v KP(t, T )

d) The self-financing hedging strategy that hedges the bond option is obtained by holding a
(possibly fractional) quantity

v 1 P(t, S)
Φ + log
2 v KP(t, T )
of the bond with maturity S, and by shorting a quantity

v 1 P(t, S)
KΦ − + log
2 v KP(t, T )
of the bond with maturity T .
Exercise 3.8
a) The process
e −rt S2 (t ) = S2 (0) e σ2Wt +(µ−r)t
is a martingale if
1
r − µ = σ22 .
2
b) We note that
2
e −rt Xt = e −rt e (r−µ )t−σ1 t/2 S1 (t )
2 2
= e −rt e (σ2 −σ1 )t/2 S1 (t )
2
= e −µt−σ1 t/2 S1 (t )
2
= S1 (0) e µt−σ1 t/2 e σ1Wt +µt
2
= S1 (0) e σ1Wt −σ1 t/2
is a martingale, where
2 2 2
Xt = e (r−µ )t−σ1 t/2 S1 (t ) = e (σ2 −σ1 )t/2 S1 (t ).
c) By (3.5.7) we have
Xt
Xb (t ) =
Nt
2 2 S1 (t )
= e (σ2 −σ1 )t/2
S2 (t )
S1 (0) (σ22 −σ12 )t/2+(σ1 −σ2 )Wt
= e
S2 ( 0 )
S1 (0) (σ22 −σ12 )t/2+(σ1 −σ2 )Wbt +σ2 (σ1 −σ2 )t
= e
S2 ( 0 )
S1 (0) (σ1 −σ2 )Wbt +σ2 σ1t−(σ22 +σ12 )t/2
= e
S2 ( 0 )
S1 (0) (σ1 −σ2 )Wbt −(σ1 −σ2 )2t/2
= e ,
S2 ( 0 )
where
bt := Wt − σ2t
W
b defined by
dP
b r T NT
= e − 0 rs ds
dP N0
S2 ( T )
= e −rT
S2 ( 0 )

= e −rT e σ2WT +µT

= e σ2WT +(µ−r)T
2
= e σ2WT −σ2 t/2 .
2 2
d) Given that Xt = e (σ2 −σ1 )t/2 S1 (t ) and Xb (t ) = Xt /Nt = Xt /S2 (t ), we have
2 2
e −rT IE[(S1 (T ) − κS2 (T ))+ ] = e −rT IE[( e −(σ2 −σ1 )T /2 XT − κS2 (T ))+ ]
2 2 2 2
= e −rT e −(σ2 −σ1 )T /2 IE[(XT − κ e (σ2 −σ1 )T /2 S2 (T ))+ ]
2 2
b [(XbT − κ e (σ22 −σ12 )T /2 )+ ]
= S2 (0) e −(σ2 −σ1 )T /2 IE
2 2
b [(Xb0 e (σ1 −σ2 )WbT −(σ1 −σ2 )2 T /2 − κ e (σ22 −σ12 )T /2 )+ ]
= S2 (0) e −(σ2 −σ1 )T /2 IE
2 2 2 2
= S2 (0) e −(σ2 −σ1 )T /2 Xb0 Φ0+ (T , Xb0 ) − κ e (σ2 −σ1 )T /2 Φ0− (T , Xb0 )

2 2
= S2 (0) e −(σ2 −σ1 )T /2 Xb0 Φ0+ (T , Xb0 )
2 2 2 2
−κS2 (0) e −(σ2 −σ1 )T /2 e (σ2 −σ1 )T /2 Φ0− (T , Xb0 )
2 2
= X0 e −(σ2 −σ1 )T /2 Φ0+ (T , Xb0 ) − κS2 (0)Φ0− (T , Xb0 )
2 2
= S1 (0) e −(σ2 −σ1 )T /2 Φ0+ (T , Xb0 ) − κS2 (0)Φ0− (T , Xb0 ),
where
(σ1 − σ2 )2 − (σ22 − σ12 ) √

log(x/κ )
Φ + (T , x ) = Φ
0
√ + T
|σ1 − σ2 | T 2|σ1 − σ2 |
 √

log(x/κ )
 Φ
 √ + σ1 T , σ1 > σ2 ,
|σ1 − σ2 | T



=
√

log(x/κ )

 Φ

√ − σ1 T , σ1 < σ2 ,


|σ1 − σ2 | T
and
(σ1 − σ2 )2 + (σ22 − σ12 ) √

log(x/κ )
Φ− (T , x ) = Φ
0
√ − T
|σ1 − σ2 | T 2|σ1 − σ2 |
 √

log(x/κ )

 Φ √ + σ2 T , σ1 > σ2 ,
|σ1 − σ2 | T



=
√

log(x/κ )

 Φ

√ − σ2 T , σ1 < σ2 ,


|σ1 − σ2 | T
if σ1 ̸= σ2 . In case σ1 = σ2 , we find
e −rT IE[(S1 (T ) − κS2 (T ))+ ] = e −rT IE[S1 (T )(1 − κS2 (0)/S1 (0))+ ]
= (1 − κS2 (0)/S1 (0))+ e −rT IE[S1 (T )]
= (S1 (0) − κS2 (0))1{S1 (0)>κS2 (0)} .
e −(T −t )r IE∗ 1{RT ⩾κ} Rt = e −(T −t )r P∗ RT ⩾ κ Rt

f 2 /2
e −(T −t )r P∗ Rt e (WT −Wt )σ +(r−r )(T −t )−(T −t )σ

= ⩾ κ Rt
f 2
e −(T −t )r P∗ x e (WT −Wt )σ +(r−r )(T −t )−(T −t )σ /2 ⩾ κ x=R

=
t
f )(T − t ) − (T − t )σ 2 /2 − log(κ/R )

−(T −t )r ( r − r t
= e Φ √ ,
σ T −t
after applying the hint provided, with
η 2 : = (T − t )σ 2 and µ := (r − rf )(T − t ) − (T − t )σ 2 /2.

Remark: Binary options are often proposed at the money, i.e. κ = Rt , with a short time to maturity,
for example the small value
T − t ≃ 30 seconds = 0.000000951 = 9.51 × 10−7 year−1 ,
in which case we have

r − rf σ √

−(T −t )r ∗
1{RT ⩾κ} Rt −(T −t )r
Φ

e IE = e − T −t
σ 2
≃ Φ (0)
1
= .
2
Taking for example r − rf = 0.02 = 2% and σ = 0.3 = 30%, we have
r − rf σ √

0.02 0.3 p
− T −t = − 9.51 × 10−7 = −0.000081279
σ 2 0.3 2
and
e −(T −t )r IE∗ 1{RT ⩾κ} Rt e −(T −t )r Φ(−0.000081279)

=
= e −r×0.000000951 × 0.499968
= 0.49996801
1
≃ ,
2
with r = 0.02 = 2%.
Exercise 3.10
a) It suffices to check that the definition of (WtN )t∈R+ implies the correlation identity dWtS •
dWtN = ρdt by Itô’s calculus.
b) We let q
σbt = (σtS )2 − 2ρσtR σtS + (σtR )2
and
σtS − ρσtN p σN
dWtX = dWtS − 1 − ρ 2 t dWt , t ⩾ 0,
σbt σbt
which defines a standard Brownian motion under P∗ due to the definition of σ
bt .
Exercise 3.11 p
a) We have σb = (σ S )2 − 2ρσ R σ S + (σ R )2 .
b) Letting Xet"= e −rt Xt = e (a−r#)t St /Rt , t ⩾ 0, we have
+
∗ ST h + i
IE −κ Ft = e −aT IE∗ XT − e aT κ Ft
RT
+
−(a−r )T ∗
= e IE XT − e
e (a−r )T
κ Ft
(r − a + σb 2 /2)(T − t )

−(a−r )T 1 St
= e Xt Φ
e √ + √ log
b T −t
σ σb T − t κRt
2

(r − a − σ /2)(T − t ) 1 St
−κ e (a−r)T Φ √ + √
b
log
σb T − t σb T − t κRt

(r − a + σb 2 /2)(T − t )

St (r−a)(T −t ) 1 St
= e Φ √ + √ log
Rt σb T −t b T −t
σ κRt
2

(r − a − σb /2)(T − t ) 1 St
−κΦ √ + √ log ,
b T −t
σ σb T − t κRt
hence the price of "the quanto option is#
+
−(T −t )r ∗ ST
e IE −κ Ft
RT
(r − a + σb 2 /2)(T − t )

St −a(T −t ) 1 St
= e Φ √ + √ log
Rt b T −t
σ b T −t
σ κRt
2

−r (T −t ) (r − a − σb /2)(T − t ) 1 St
−κ e Φ √ + √ log .
σb T −t σb T − t κRt
Chapter 4
Exercise 4.1
a) We price the floorlet at t = 0, with T1 = 9 months, T2 = 1 year, κ = 4.5%. The LIBOR
rate (L(t, T1 , T2 ))t∈[0,T1 ] is modeled as a driftless geometric Brownian motion with volatility
b = σ1,2 (t ) = 0.1 under the forward measure P2 . The discount factors are given
coefficient σ
by
P(0, T1 ) = e −9r/12 ≃ 0.970809519
and
P(0, T2 ) = e −r ≃ 0.961269954,
with r = 3.95%.
b) By (4.3.9),
h the price of the floorlet is
r T2 i
IE∗ e − 0(κ − L(T1 , T1 , T2 ))+
rs ds
= P(0, T2 ) κΦ − d− (T1 ) − L(0, T1 , T2 )Φ − d+ (T1 ) ,

(A.16)
where
log(L(0, T1 , T2 )/κ ) + σ 2 T1 /2
d+ (T1 ) = √ ,
σ1 T1
and
log(L(0, T1 , T2 )/κ ) − σ 2 T1 /2
d− (T1 ) = √ ,
σ T1
are given in Proposition 4.5, and the LIBOR rate L(0, T1 , T2 ) is given by
P(0, T1 ) − P(0, T2 )
L(0, T1 , T2 ) =
(T2 − T1 )P(0, T2 )
e −3r/4 − e −r
=
0.25 e −r
= 4( e r/4 − 1)
≃ 3.9695675%.
Hence, we have
log(0.039695675/0.045) + (0.1)2 × 0.75/2
d+ (T1 ) = √ ≃ −1.404927033,
0.1 × 0.75
and
log(0.039695675/0.045) − (0.1)2 × 0.75/2
d− (T1 ) = √ ≃ −1.491529573,
0.1 × 0.75

hence h r T2 i
IE∗ e − 0 rs ds (κ − L(T1 , T1 , T2 ))+
= 0.961269954 × (κΦ(1.491529573) − L(0, T1 , T2 ) × Φ(1.404927033))
= 0.961269954 × (0.045 × 0.932089 − 0.039695675 × 0.919979)
≃ 0.52147141%.
Finally, we need to multiply (A.16) by the notional principal amount of $1 million per interest
rate percentage point, i.e. $10, 000 per percentage point or $100 per basis point, which yields
$5214.71.
Exercise 4.2
a) We price the swaption at t = 0, with T1 = 4 years, T2 = 5 years, T3 = 6 years, T4 = 7
years, κ = 5%, and the swap rate (S(t, T1 , T4 ))t∈[0,T1 ] is modeled as a driftless geometric
Brownian motion with volatility coefficient σ b = σ1,4 (t ) = 0.2 under the forward swap
measure P1.4 . The discount factors are given by P(0, T1 ) = e −4r , P(0, T2 ) = e −5r , P(0, T3 ) =
e −6r , P(0, T4 ) = e −7r , where r = 5%.
b) By Proposition 4.17 the price of the swaption is
(P(0, T1 ) − P(0, T4 ))Φ(d+ (T1 − t ))
−κΦ(d− (T1 ))(P(0, T2 ) + P(0, T3 ) + P(0, T4 )),
where d+ (T1 ) and d− (T1 ) are given in Proposition 4.17, and the LIBOR swap rate S(0, T1 , T4 )
is given by
P(0, T1 ) − P(0, T4 )
S(0, T1 , T4 ) =
P(0, T1 , T4 )
P(0, T1 ) − P(0, T4 )
=
P(0, T2 ) + P(0, T3 ) + P(0, T4 )
e −4r − e −7r
=
e −5r + e −6r + e −7r
e 3r − 1
=
e 2r + e r + 1
e 0.15 − 1
=
e 0.1 + e 0.05 + 1
= 0.051271096.
By Proposition 4.17 we also have
log(0.051271096/0.05) + (0.2)2 × 4/2

d+ (T1 ) = √ = 0.526161682,
0.2 4
and
log(0.051271096/0.05) − (0.2)2 2 × 4/2
d− (T1 ) = √ = 0.005214714,
0.2 4
Hence, the price of the swaption is given by
( e −4r − e −7r )Φ(0.526161682) (A.17)
−5r −6r −7r
−κΦ(0.005214714)( e +e +e )
= (0.818730753 − 0.70468809) × 0.700612
−0.05 × 0.50208 × (0.818730753 + 0.740818221 + 0.70468809)
= 2.3058251%.
Finally, we need to multiply (A.17) by the notional principal amount of $10 million, i.e.
$100, 000 by interest percentage point, or $1, 000 by basis point, which yields $230, 582.51.

Exercise 4.3 Taking t = 0, we have T1 = 3, T2 = 4 and T3 = 5. The LIBOR swap rate S(t, T1 , T3 ) is
modeled as a driftless geometric Brownian motion with volatility σ = 0.1 under the forward swap
measure Pb i, j . The receiver swaption is priced using the Black-Scholes formula as
h r T1 i
IE∗ e − t rs ds P(T1 , T1 , T3 ) (κ − S(T1 , T1 , T3 ))+ Ft
2
= κΦ(−d− (T1 − t )) ∑ (Tk+1 − Tk )P(t, Tk+1 )
k =1
−(P(t, T1 ) − P(t, T3 ))Φ(−d+ (T1 − t )),
where κ = 5%, r = 2% and P(t, T1 ) = e −3r = 0.9417, P(t, T2 ) = e −4r = 0.9231, P(t, T3 ) =
e −5r = 0.9048. Hence,
P(t, T1 , T3 ) = P(t, T2 ) + P(t, T3 ) = 0.92311 + 0.90483 = 1.82794
and
P(t, T1 ) − P(t, T3 ) 0.9417 − 0.9048
S(t, T1 , T3 ) = = = 0.02018.
P(t, T1 , T3 ) 1.82794
We also have
log(S(t, T1 , T3 )/κ ) + σ 2 (T1 − t )/2
d+ (T1 − t ) = √
σ T1 − t
log(2.018/5) + 0.12 × 3/2
= √ = −5.1518
0.1 3
and √
d− (T1 − t ) = d+ (T1 − t ) − σ T1 − t = −5.3250,
hence
h r T1 i
+
IE∗ e − t rs ds P(T1 , T1 , T3 ) (κ − S(T1 , T1 , T3 )) Ft
= 0.05 × 1.82794 × Φ(5.3250) − (0.9417 − 0.9048) × Φ(5.1518)
= 0.05 × 1.82794 × 0.9999999 − (0.9417 − 0.9048) × 0.9999999
= 0.054496
= 5.4496%
= 544.96 bp,
which yields $54, 496 after multiplication by the $10, 000 notional principal.
Exercise 4.4
a) We have

P(t, T2 ) dP(t, T2 ) 1 1
d = + P(t, T2 )d + dP(t, T2 ) • d
P(t, T1 ) P(t, T1 ) P(t, T1 ) P(t, T1 )

dP(t, T2 ) dP(t, T1 ) dP(t, T1 ) dP(t, T1 )
•
= + P(t, T2 ) − 2
+
P(t, T1 ) (P(t, T1 )) (P(t, T1 ))3
dP(t, T1 ) • dP(t, T2 )
−
(P(t, T1 ))2
1
= rt P(t, T2 )dt + ζ2 (t )P(t, T2 )dWt
P(t, T1 )

P(t, T2 )
− 2
rt P(t, T1 )dt + ζ1 (t )P(t, T1 )dWt
(P(t, T1 ))
P(t, T2 )
(rt P(t, T1 )dt + ζ1 (t )P(t, T1 )dWt )2

+ 3
(P(t, T1 ))
1
− ( rt P (t, T1 ) dt + ζ1 ( t ) P (t, T1 ) dW t ) • (r P(t, T )dt + ζ (t )P(t, T )dW )
t 2 2 2 t
(P(t, T1 ))2
P(t, T2 ) P(t, T2 )
= ζ2 (t ) dWt − ζ1 (t ) dWt
P(t, T1 ) P(t, T1 )
P(t, T2 ) P(t, T2 )
+ (ζ1 (t ))2 dt − ζ1 (t )ζ2 (t ) dt
P(t, T1 ) P(t, T1 )
P(t, T2 ) P(t, T2 )
=− ζ1 (t )(ζ2 (t ) − ζ1 (t ))dt + (ζ2 (t ) − ζ1 (t ))dWt
P(t, T1 ) P(t, T1 )
P(t, T2 )
= (ζ2 (t ) − ζ1 (t ))(dWt − ζ1 (t )dt )
P(t, T1 )
P(t, T2 ) b P(t, T2 ) b
= (ζ2 (t ) − ζ1 (t )) dWt = (ζ2 (t ) − ζ1 (t )) dWt ,
P(t, T1 ) P(t, T1 )
where dWbt = dWt − ζ1 (t )dt is a standard Brownian motion under the T1 -forward measure P. b
b) From Question (a)) or (4.1.7) we have
P(T1 , T2 )
P(T1 , T2 ) =
P(T1 , T1 )
w
1 w T1

P(t, T2 ) T1
2
= exp (ζ2 (s) − ζ1 (s))dWs −
b (ζ2 (s) − ζ1 (s)) ds
P(t, T1 ) t 2 t
P(t, T2 ) X−v2 /2
= e ,
P(t, T1 )
w T1
where X is a centered Gaussian random variable with variance v2 = (ζ2 (s) − ζ1 (s))2 ds,
t
independent of Ft under P. b Hence by the hint (4.5.13) with x := P(t, T2 )/P(t, T1 ) and
κ := K/x,
h we
rT
find i
∗ − 0 1 rs ds
(K − P(T1 , T2 ))+ Ft = P(t, T1 )IbE (K − P(T1 , T2 ))+ | Ft

IE e

v 1 K P(t, T2 ) v 1 K
= P(t, T1 ) KΦ + log − Φ − + log
2 v x P(t, T1 ) 2 v x

v 1 K v 1 K
= KP(t, T1 )Φ + log − P(t, T2 )Φ − + log .
2 v x 2 v x
Exercise 4.5
a) The forward measure P
b S is defined from the numéraire Nt := P(t, S) and this gives
E[(κ − L(T , T , S))+ | Ft ].

Ft = P(t, S)Ib
b) The LIBOR rate L(t, T , S) is a driftless geometric Brownian motion with volatility σ under
the forward measure P b S . Indeed, the LIBOR rate L(t, T , S) can be written as the forward
price L(t, T , S) = Xbt = Xt /Nt wherer Xt = (P(t, T ) − P(rt, S))/(S − T ) and Nt = P(t, S).
t t
Since both discounted bond prices e − 0 rs ds P(t, T ) and e − 0 rs ds P(t, S) are martingales under
P∗ , the same is true of Xt . Hence L(t, T , S) = Xt /Nt becomes a martingale under the forward
measure P b S by Proposition 3.4, and computing its dynamics under P b S amounts to removing
any “dt” term in (4.5.14) after rewriting the equation in terms of the standard Brownian

bt )t∈R+ under P
motion (W b S , i.e. we have
dL(t, T , S) = σ L(t, T , S)dW

bt ,
2 t/2
which is solved as L(t, T , S) = L(0, T , S) e σ Wt −σ , 0 ⩽ t ⩽ T.
b
c) We find
Ft = P(t, S)IE b [(κ − L(T , T , S))+ | Ft ]
= P(t, S)IE b [(κ − L(t, T , S) e −(T −t )σ 2 /2+(WbT −Wbt )σ )+ | Ft ]
= P(t, S)(κΦ(−d− (T − t )) − Xbt Φ(−d+ (T − t )))
= κP(t, S)Φ(−d− (T − t )) − P(t, S)L(t, T , S)Φ(−d+ (T − t ))
= κP(t, S)Φ(−d− (T − t )) − (P(t, T ) − P(t, S))Φ(−d+ (T − t ))/(S − T ),
2
where e = L(t, T , S) e −(T −t )σ /2 , v2 = (T − t )σ 2 , and
m
√
log(L(t, T , S)/κ ) σ T − t
d+ (T − t ) = √ + ,
σ T −t 2
and √
log(L(t, T , S)/κ ) σ T − t
d− ( T − t ) = √ − ,
σ T −t 2
because L(t, T , S) is a driftless geometric Brownian motion with volatility σ under the for-
ward measure P b S.
Exercise 4.6
a) We have
j−1
P(Ti , Ti , T j ) = ∑ cl +1 P(Ti , Tl +1 ).
l =i
b) It suffices to let c̃l = 1, l = i + 1, . . . , j − 1. and c̃ j = c j + 1/κ.

c) The swaptionh rcan be priced as
T + i
∗ − t i rs ds
IE e P(Ti , Ti ) − P(Ti , T j ) − κP(Ti , Ti , T j ) Ft
" !+ #
rT j−1
i
= IE∗ e − t rs ds
1−κ ∑ cl +1 P(Ti , Tl +1 ) Ft
l =i
" !+ #
rT j−1 j−1
i
= κ IE∗ e − t rs ds
∑ cl +1 Fl +1 (Ti , γκ ) − ∑ cl +1 Fl +1 (Ti , rT ) i Ft
l =i l =i
" !+ #
rT j−1
i
∑ cl +1 (Fl +1 (Ti , γκ ) − Fl +1 (Ti , rT )) i Ft
l =i
" #
rT j−1
1{rTi ⩽γκ } ∑ cl +1 (Fl +1 (Ti , γκ ) − Fl +1 (Ti , rTi )) Ft
i
l =i
" #
rT j−1
∑ cl +1 (Fl +1 (Ti , γκ ) − Fl +1 (Ti , rT ))+
i
i Ft
l =i
j−1 h r Ti i
∗ − t rs ds +
= κ c
∑ l +1 IE e ( F ( T
l +1 i κ, γ ) − P ( T , T
i l +1 )) Ft
l =i
j−1
(Fl +1 (Ti , γκ ) − P(Ti , Tl +1 ))+ Ft ,

= κ ∑ cl +1 P(t, Ti )IEb i
l =i

which is a weighted sum of bond put option prices with strike prices Fl +1 (Ti , γκ ), l =
i, i + 1, . . . , j − 1.
Exercise 4.7
a) We have
n n
dXt = ∑ ci d Pb(t, Ti ) = ∑ ci σi (t )Pb(t, Ti )dWbt = σt Xt dWbt ,
i=2 i=2
from which we obtain
n n
1 n ∑ ci σi (t )Pb(t, Ti ) ∑ ci σi (t )P(t, Ti )
σt =
Xt ∑ ci σi (t )Pb(t, Ti ) = i=2 n = i=2
n .
i=2
∑ ci Pb(t, Ti ) ∑ ci P(t, Ti )
i=2 i=2
b) Approximating the random process (σt )t∈R+ by the deterministic function of time
n n
1 n
∑ ci σi (t )Pb(0, Ti ) ∑ ci σi (t )P(0, Ti )
i=2 i=2
σb (t ) := ci σi (t )Pb(0, Ti ) = = ,
X0 i∑
=2
n n
∑ ci Pb(0, Ti ) ∑ ci P(0, Ti )
i=2 i=2
we find
b (XT0 − κ )+

P(0, T1 )IE
r T0 rT +

(1)
|σb (t )|2 dWt − 21 0 0 |σb (t )|2 dt
≃ P(0, T1 )IE X0 e
b 0 −κ
r w
1 T0 2
= P(0, T1 )Blcall X0 , κ, |σ
b (t )| dt, 0, T
T0 0
= P(0, T1 ) X0 Φ(v/2 + (log(X0 /κ ))/v) − κΦ(−v/2 + (log(X0 /κ ))/v)

!!
n n
v 1 ci P(0, Ti )
= ∑ ci P(0, Ti )Φ − log ∑
i=2 2 v i=2 κP(0, T1 )
!!
n
v 1 ci P(0, Ti )
−κP(0, T1 )Φ − − log ∑ ,
2 v i=2 κP(0, T1 )
rw
T0
with v := |σb (t )|2 dt.
0
Exercise 4.8
a) We have
dP(t, Ti )
= rt dt + ζ (i) (t )dBt , i = 1, 2,
P(t, Ti )
and w wT 1 w T (i)

T
(i) 2
P(T , Ti ) = P(t, Ti ) exp rs ds + ζ (s)dBs − |ζ (s)| ds ,
t t 2 t
0 ⩽ t ⩽ T ⩽ Ti , i = 1, 2, hence
wT wT 1 w T (i)
log P(T , Ti ) = log P(t, Ti ) + rs ds + ζ (i) (s)dBs − |ζ (s)|2 ds,
t t 2 t
0 ⩽ t ⩽ T ⩽ Ti , i = 1, 2, and
1
d log P(t, Ti ) = rt dt + ζ (i) (t )dBt − |ζ (i) (t )|2 dt, i = 1, 2.
2

In the present model, we have

drt = σ dBt ,
where (Bt )t∈R+ is a standard Brownian motion under P, by the solution of Exercise 1.4 and
(1.4.8) we have
ζ (i) (t ) = −(Ti − t )σ , 0 ⩽ t ⩽ Ti , i = 1, 2.
Letting
(i)
dBt = dBt − ζ (i) (t )dt,
defines a standard Brownianmotion under Pi , i = 1, 2, and we have

P(T , T1 ) P(t, T1 ) wT 1 w T (1)

(1) (2) 2 (2) 2
= exp (ζ (s) − ζ (s))dBs − (|ζ (s)| − |ζ (s)| )ds
P(T , T2 ) P(t, T2 ) t 2 t
w
1 w T (1)

P(t, T1 ) T
(1) (2) (2) (2) 2
= exp (ζ (s) − ζ (s))dBs − (ζ (s) − ζ (s)) ds ,
P(t, T2 ) t 2 t
which is an Ft -martingale under wP2 and under P1,2 , and
1 w T (1)

P(T , T2 ) P(t, T2 ) T
(1) (2) (1) (2) 2
= exp − (ζ (s) − ζ (s))dBs − (ζ (s) − ζ (s)) ds ,
P(T , T1 ) P(t, T1 ) t 2 t
which is an Ft -martingale under P1 .
b) We have
1
f (t, T1 , T2 ) = − (log P(t, T2 ) − log P(t, T1 ))
T2 − T1
1 σ2
= rt + ((T1 − t )3 − (T2 − t )3 ).
T2 − T1 6
c) We have
1 P(t, T2 )
d f (t, T1 , T2 ) = − d log
T2 − T1 P(t, T1 )

1 1
= − (ζ (2) (t ) − ζ (1) (t ))dBt − (|ζ (2) (t )|2 − |ζ (1) (t )|2 )dt
T2 − T1 2

1 (2) (1) (2) (2) 1 (2) 2 (1) 2
= − (ζ (t ) − ζ (t ))(dBt + ζ (t )dt ) − (|ζ (t )| − |ζ (t )| )dt
T2 − T1 2

1 (2) (1) (2) 1 (2) (1) 2
= − (ζ (t ) − ζ (t ))dBt − (ζ (t ) − ζ (t )) dt .
T2 − T1 2
d) We have
1 P(T , T2 )
f (T , T1 , T2 ) = − log
T2 − T1 P(T , T1 )
w
1 T
(2) (1) 1 (2) 2 (1) 2
= f (t, T1 , T2 ) − (ζ (s) − ζ (s))dBs − (|ζ (s)| − |ζ (s)| )ds
T2 − T1 t 2
w
1 w T (2)

1 T
(2) (1) (2) (1) 2
= f (t, T1 , T2 ) − (ζ (s) − ζ (s))dBs − (ζ (s) − ζ (s)) ds
T2 − T1 t 2 t
w
1 w T (2)

1 T
(2) (1) (1) (1) 2
= f (t, T1 , T2 ) − (ζ (s) − ζ (s))dBs + (ζ (s) − ζ (s)) ds .
T2 − T1 t 2 t
Hence f (T , T1 , T2 ) has a Gaussian distribution given Ft with conditional mean
1 w T (2)
m1 := f (t, T1 , T2 ) − (ζ (s) − ζ (1) (s))2 ds
2 t
under P1 , resp.
1 w T (2)
m2 := f (t, T1 , T2 ) + (ζ (s) − ζ (1) (s))2 ds
2 t

under P2 , and variance
1 wT
v2 = (ζ (2) (s) − ζ (1) (s))2 ds.
(T2 − T1 )2 t
Hence we have h r T2 i
(T2 − T1 ) IE∗ e − t rs ds ( f (T1 , T1 , T2 ) − κ )+ Ft
= (T2 − T1 )P(t, T2 ) IE2 ( f (T1 , T1 , T2 ) − κ )+ Ft

= (T2 − T1 )P(t, T2 ) IE2 (m2 + X − κ )+ Ft

v −(κ−m2 )2 /(2v2 )
= (T2 − T1 )P(t, T2 ) √ e + (m2 − κ )Φ((m2 − κ )/v) .
2π
e) We have
L(T , T1 , T2 ) = S(T , T1 , T2 )

1 P(T , T1 )
= −1
T2 − T1 P(T , T2 )
1
=
T2 − T1
w
1 w T (1)

P(t, T1 ) T
(1) (2) 2 (2) 2
× exp (ζ (s) − ζ (s))dBs − (|ζ (s)| − |ζ (s)| )ds − 1
P(t, T2 ) t 2 t
1
=
T2 − T1
w
1 w T (1)

P(t, T1 ) T
(1) (2) (2) (2) 2
× exp (ζ (s) − ζ (s))dBs − (ζ (s) − ζ (s)) ds − 1
P(t, T2 ) t 2 t
1
=
T2 − T1
w
1 w T (1)

P(t, T1 ) T
(1) (2) (1) (2) 2
× exp (ζ (s) − ζ (s))dBs + (ζ (s) − ζ (s)) ds − 1 ,
P(t, T2 ) t 2 t
and, by Itô calculus,
1 P(t, T1 )
dS(t, T1 , T2 ) = d
T2 − T1 P(t, T2 )

1 P(t, T1 ) 1
= (ζ (1) (t ) − ζ (2) (t ))dBt + (ζ (1) (t ) − ζ (2) (t ))2 dt
T2 − T1 P(t, T2 ) 2

1
− (|ζ (1) (t )|2 − |ζ (2) (t )|2 )dt
2

1
= + S(t, T1 , T2 ) (ζ (1) (t ) − ζ (2) (t ))dBt + ζ (2) (t )(ζ (2) (t ) − ζ (1) (t ))dt )dt
T2 − T1

1 (1)

= + S(t, T1 , T2 ) (ζ (1) (t ) − ζ (2) (t ))dBt + (|ζ (2) (t )|2 − |ζ (1) (t )|2 )dt
T2 − T1

1 (2)
= + S(t, T1 , T2 ) (ζ (1) (t ) − ζ (2) (t ))dBt , t ∈ [0, T1 ],
T2 − T1
1
hence t 7→ T2 −T 1
+ S(t, T1 , T2 ) is a geometric Brownian motion, with
1
+ S(T , T1 , T2 )
T2 − T1

1
= + S(t, T1 , T2 )
T2 − T1
w
1 w T (1)

T (2)
× exp (ζ (1) (s) − ζ (2) (s))dBs − (ζ (s) − ζ (2) (s))2 ds ,
t 2 t

0 ⩽ t ⩽ T ⩽ T1 .
f) We have h r T2 i
(T2 − T1 ) IE∗ e − t rs ds (L(T1 , T1 , T2 ) − κ )+ Ft
h r T1 i
= (T2 − T1 ) IE∗ e − t rs ds P(T1 , T2 )(L(T1 , T1 , T2 ) − κ )+ Ft
= P(t, T1 , T2 ) IE1,2 (S(T1 , T1 , T2 ) − κ )+ Ft .

The forward measure P2 is defined by

P(t, T2 ) − r t rs ds

∗ dP2
IE Ft = e 0 , 0 ⩽ t ⩽ T2 ,
dP P(0, T2 )
and the forward swap measure is defined by
P(t, T2 ) − r t rs ds

∗ dP1,2
IE Ft = e 0 , 0 ⩽ t ⩽ T1 ,
dP P(0, T2 )
(2)
hence P2 and P1,2 coincide up to time T1 and Bt t∈[0,T ] is a standard Brownian motion
1
until time T1 under P2 and under P1,2 , consequently under P1,2 we have
L(T , T1 , T2 ) = S(T , T1 , T2 )
r
(2) 1 r T

1 1 T (1) (2) (1) (2) 2
= − + + S(t, T1 , T2 ) e t (ζ (s)−ζ (s))dBs − 2 t (ζ (s)−ζ (s)) ds ,
T2 − T1 T2 − T1
has same distribution as

1 P(t, T1 ) X−Var [X ]/2
e −1 ,
T2 − T1 P(t, T2 )
where X is a centered Gaussian random variable with variance
w T1
(ζ (1) (s) − ζ (2) (s))2 ds
t
given Ft . Hence, we h have
rT i
2
(T2 − T1 ) IE e − t rs ds (L(T1 , T1 , T2 ) − κ )+ Ft
∗
= P(t, T1 , T2 )
r T1 (1)
t (ζ (s) − ζ (2) (s))2 ds

1 1
×Bl + S(t, T1 , T2 ), ,κ + , T1 − t .
T2 − T1 T1 − t T2 − T1
Exercise 4.9
a) We have
rt (2) r t 2
L(t, T1 , T2 ) = L(0, T1 , T2 ) e 0 γ1 (s)dWs − 0 |γ1 (s)| ds/2 , 0 ⩽ t ⩽ T1 ,
and L(t, T2 , T3 ) = b. Note that we also have P(t, T2 )/P(t, T3 ) = 1 + δ b, hence P b2 = P
b3 =
P1,2 up to time T1 .
b
b) We use change of numéraire under the forward measure P2 .
c) We have h r
T2
i
IE∗ e − t rs ds (L(T1 , T1 , T2 ) − κ )+ Ft
b 2 (L(T1 , T1 , T2 ) − κ )+ Ft

= P(t, T2 )IE
h rT (2) r T
i
= P(t, T2 )IE b 2 (L(t, T1 , T2 ) e t 1 γ1 (s)dWs − t 1 |γ1 (s)|2 ds/2 − κ )+ Ft
= P(t, T2 )Bl(L(t, T1 , T2 ), κ, σ1 (t ), 0, T1 − t ),
where
1 w T1
σ12 (t ) = |γ1 (s)|2 ds.
T1 − t t

d) We have
P(t, T1 ) P(t, T1 )
=
P(t, T1 , T3 ) δ P(t, T2 ) + δ P(t, T3 )
P(t, T1 ) 1
=
δ P(t, T2 ) 1 + P(t, T3 )/P(t, T2 )
1+δb
= (1 + δ L(t, T1 , T2 )), 0 ⩽ t ⩽ T1 ,
δ (δ b + 2)
and
P(t, T3 ) P(t, T3 )
=
P(t, T1 , T3 ) P(t, T2 ) + P(t, T3 )
1
=
1 + P(t, T2 )/P(t, T3 )
1 1
= , 0 ⩽ t ⩽ T2 .
δ 2+δb
e) We have
P(t, T1 ) P(t, T3 )
S(t, T1 , T3 ) = −
P(t, T1 , T3 ) P(t, T1 , T3 )
1+δb 1
= (1 + δ L(t, T1 , T2 )) −
δ (2 + δ b) δ (2 + δ b)
1
= (b + (1 + δ b)L(t, T1 , T2 )), 0 ⩽ t ⩽ T2 .
2+δb
We have
1+δb (2)
dS(t, T1 , T3 ) = L(t, T1 , T2 )γ1 (t )dWt
2
+ δ b
b (2)
= S(t, T1 , T3 ) − γ1 (t )dWt
2+δb
(2)
= S(t, T1 , T3 )σ1,3 (t )dWt , 0 ⩽ t ⩽ T2 ,
with
b
σ1,3 (t ) = 1− γ1 (t )
S(t, T1 , T3 )(2 + δ b)

b
= 1− γ1 (t )
b + (1 + δ b)L(t, T1 , T2 )
(1 + δ b)L(t, T1 , T2 )
= γ1 (t )
b + (1 + δ b)L(t, T1 , T2 )
(1 + δ b)L(t, T1 , T2 )
= γ1 (t ).
(2 + δ b)S(t, T1 , T3 )
f) The process W (2) t∈R is a standard Brownian motion under P b2 = P

b 1,3 and
+
b 1,3 (S(T1 , T1 , T3 ) − κ )+ | Ft

P(t, T1 , T3 )IE
= P(t, T2 )Bl(S(t, T1 , T2 ), κ, σe1,3 (t ), 0, T1 − t ),
where |σe1,3 (t )|2 is the approximation of the volatility
1 w T1 1 w T1 (1 + δ b)L(s, T1 , T2 ) 2

|σ1,3 (s)|2 ds = γ1 (s)ds
T1 − t t T1 − t t (2 + δ b)S(s, T1 , T3 )
obtained by freezing the random component of σ1,3 (s) at time t, i.e.
(1 + δ b)L(t, T1 , T2 ) 2 w T1

2 1
σe1,3 (t ) = |γ1 (s)|2 ds.
T1 − t (2 + δ b)S(t, T1 , T3 ) t
Exercise 4.10 Bond option hedging.

a) We have h wT
rT i
IE∗ e − t rs ds (P(T , S) − κ )+ Ft = VT = V0 + dVt
0
w t wt
= P(0, T ) IET (P(T , S) − κ )+ + ξsT dP(s, T ) + ξsS dP(s, S).

0 0
b) We have rt
dVet = d e − 0 rs dsVt
rt rt
= −rt e − Vt dt + e −
0 rs ds 0 rs ds dVt
rt
= −rt e − 0 rs ds (ξtT P(t, T )
rt rt
+ξtS P(t, S))dt + e − 0 rs ds ξtT dP(t, T ) + e − 0 rs ds ξtS dP(t, S)
= ξtT d Pe(t, T ) + ξtS d Pe(t, S).
c) By Itô’s formula we have
IET (P(T , S) − κ )+ Ft = C (Xt , v(t, T ))

w t ∂C
= C (X0 , v(0, T )) + (Xs , v(s, T ))dXs
0 ∂x
w t ∂C
= IET (P(T , S) − κ )+ +

(Xs , v(s, T ))dXs ,
0 ∂x
since the process
t 7→ IET (P(T , S) − κ )+ Ft

is a martingale under P.
e
d) We have
V t
dVbt = d
P(t, T )
= d IET (P(T , S) − κ )+ Ft

∂C
= (Xt , v(t, T ))dXt
∂x
P(t, S) ∂C
= (Xt , v(t, T ))(σtS − σtT )dBtT .
P(t, T ) ∂ x
e) We have
P(t, T )
dVt = d
Vbt
= P(t, T )dVbt + Vbt dP(t, T ) + dVbt • dP(t, T )
∂C
= P(t, S) (Xt , v(t, T ))(σtS − σtT )dBtT + Vbt dP(t, T )
∂x
∂C
+P(t, S) (Xt , v(t, T ))(σtS − σtT )σtT dt
∂x
∂C
= P(t, S) (Xt , v(t, T ))(σtS − σtT )dBt + Vbt dP(t, T ).
∂x
f) We have rt
dVet = d ( e − 0 rs dsVt )
rt rt
= −rt e − Vt dt + e −
0 rs ds 0 rs ds dVt
∂C
= Pe(t, S) (Xt , v(t, T ))(σtS − σtT )dBt + Vbt d Pe(t, T ).
∂x
g) We have
∂C
dVet = Pe(t, S) (Xt , v(t, T ))(σtS − σtT )dBt + Vbt d Pe(t, T )
∂x
∂C
= (Xt , v(t, T ))d Pe(t, S)
∂x

P(t, S) ∂C
− (Xt , v(t, T ))d Pe(t, T ) + Vbt d Pe(t, T )
P(t, T ) ∂ x

P ( t, S ) ∂C
= Vbt − (Xt , v(t, T )) d Pe(t, T )
P(t, T ) ∂ x
∂C
+ (Xt , v(t, T ))d Pe(t, S),
∂x
hence the delta hedging strategy (ξtT , ξtS )t∈[0,T ] of the bond option is given by
P(t, S) ∂C
ξtT = Vbt − (Xt , v(t, T ))
P(t, T ) ∂ x
P(t, S) ∂C
= C (Xt , v(t, T )) − (Xt , v(t, T )),
P(t, T ) ∂ x
and
∂C
ξtS = (Xt , v(t, T )), 0 ⩽ t ⩽ T.
∂x
h) We have
∂C ∂ v 1 x v 1 x
(x, v) = xΦ + log − κΦ − + log
∂x ∂x 2 v κ 2 v κ

∂ v 1 x ∂ v 1 x v 1 x
= x Φ + log − κ Φ − + log +Φ + log
∂x 2 v κ ∂x 2 v κ 2 v κ
1 v 1 x 2 1 v 1 x 2
e − 2 ( 2 + v log κ ) e − 2 (− 2 + v log κ )

v 1 x
= x √ −κ √ +Φ + log
2πvx 2πvx 2 v κ
2

log(x/κ ) + v /2
= Φ .
v
As a consequence, we get
P(t, S) ∂C
ξtT = C (Xt , v(t, T )) − (Xt , v(t, T ))
P(t, T ) ∂ x
2
P(t, S) v (t, T )/2 + log Xt
= Φ
P(t, T ) v(t, T )

v(t, T ) 1 P(t, S)
−κΦ − + log
2 v(t, T ) κP(t, T )
2

P(t, S) log(Xt /κ ) + v (t, T )/2
− Φ √
P(t, T ) T −t
2

log(Xt /κ ) − v (t, T )/2
= −κΦ ,
v(t, T )
and
log(Xt /κ ) + v2 (t, T )/2

∂C
S
ξt = (Xt , v(t, T )) = Φ ,
∂x v(t, T )
t ∈ [0, T ], and the hedging
h r T strategy is given by i
VT = IE e − t rs ds (P(T , S) − κ )+ Ft
∗
wt wt
= V0 + ξsT dP(s, T ) + ξsS dP(s, S)
0 0
w t log(X /κ ) − v2 (t, T )/2
t
= V0 − κ Φ dP(s, T )
0 v(t, T )
w t log(X /κ ) + v2 (t, T )/2
t
+ Φ dP(s, S).
0 v(t, T )
Consequently, the bond option can be hedged by shortselling a bond with maturity T for the

amount
log(Xt /κ ) − v2 (t, T )/2

κΦ ,
v(t, T )
and by holding a bond with maturity S for the amount
log(Xt /κ ) + v2 (t, T )/2

Φ .
v(t, T )
Exercise 4.11
a) The LIBOR rate L(t, T , S) is a driftless geometric Brownian motion with deterministic volatil-
ity function σ (t ) under the forward measure Pb S.
Explanation: The LIBOR rate L(t, T , S) can be written as the forward price L(t, T , S) =
Xbt = Xt /Nt where X = (P(t, T ) − Pr(t, S))/(S − T ) and Nt = P(t, S). Since both discounted
rt t t
bond prices e − 0 rs ds P(t, T ) and e − 0 rs ds P(t, S) are martingales under P∗ , the same is true
of Xt . Hence L(t, T , S) = Xt /Nt becomes a martingale under the forward measure P b S by
Proposition 3.4, and computing its dynamics under P b S amounts to removing any “dt” term
in the original SDE defining L(t, T , S), i.e. we find
dL(t, T , S) = σ (t )L(t, T , S)dW

bt , 0 ⩽ t ⩽ T,
hence w t wt
L(t, T , S) = L(0, T , S) exp bs −
σ ( s ) dW σ 2 (s)ds/2 ,
0 0

where (W b S.
b) Choosinghthe annuity numéraire Nt =i P(t, S), we
h have
rS rS i
IE∗ e − t rs ds
φ (L(T , T , S)) Ft = IE∗ e − t rs ds NS φ (L(T , T , S)) Ft
= Nt IbE φ (L(T , T , S)) Ft

= P(t, S)IbE[φ (L(T , T , S)) | Ft ].

c) Given the solution w
T
wT
bs − 2
L(T , T , S) = L(0, T , S) exp σ ( s ) dW
σ (s)ds/2
0 0
w wT
T
2
= L(t, T , S) exp σ (s)dWs −
b σ (s)ds/2 ,
t t
we find
b [φ (L(T , T , S)) | Ft ]
P(t, S)IE
rT r
bs − T σ 2 (s)ds/2
h i
= P(t, S)IE φ L(t, T , S) e
b t σ ( s ) dW t Ft
w∞ 2 2 2 dx
φ L(t, T , S) e x−η /2 e −x /(2η ) p

= P(t, S) ,
−∞ 2πη 2
wT wT
because bs is a centered Gaussian variable with variance η 2 :=
σ ( s ) dW σ 2 (s)ds, inde-
t t
pendent of Ft under the forward measure P. b
Exercise 4.12
h the annuity numéraire Nt = P(t, Ti , Tij ), we have
a) Choosing
rT
i
IE∗ e − t rs ds
P(Ti , Ti , T j )φ (S(Ti , Ti , T j )) Ft


P(Ti , Ti , T j )
= Nt IEi, j b φ (S(Ti , Ti , T j )) Ft
NTi
= P(t, Ti , T j )IE b i, j [φ (S(Ti , Ti , T j )) | Ft ].
b) Since S(t, Ti , T j ) is a forward price for the numéraire P(t, Ti , T j ), it is a martingale under the
forward swap measure P b i, j and we have
bti, j ,
S(t, Ti , T j ) = σ S(t, Ti , T j )dW 0 ⩽ t ⩽ Ti ,
where (Wbti, j )t∈R+ is a standard Brownian motion under the forward swap measure P
b i, j .
c) Given the solution
2 T /2 2 /2
S(Ti , Ti , T j ) = S(0, Ti , T j ) e σ WTi −σ i
= S(t, Ti , T j ) e (WTi −Wt )σ −(Ti −t )σ
b b b
of (4.5.18), we find
b i, j [φ (S(Ti , Ti , T j )) | Ft ]
P(t, Ti , T j )IE
h i
= P(t, Ti , T j )IE b i, j φ S(t, Ti , T j ) e (WbTi −Wbt )σ −(Ti −t )σ 2 /2 Ft
w∞ 2
2 dx
= P(t, Ti , T j ) φ S(t, Ti , T j ) e σ x−(Ti −t )σ /2 e −x /(2(Ti −t )) p ,
−∞ 2π (Ti − t )
because W bTi − W bt is a centered Gaussian variable with variance Ti − t, independent of Ft
under the forward measure P b i, j .
d) We find
b i, j (κ − S(Ti , Ti , T j ))+ | Ft

P(t, Ti , T j )IE
b i, j (κ − S(t, Ti , T j ) e −(Ti −t )σ 2 /2+(WbTi −Wbt )σ )+ | Ft

= P(t, Ti , T j )IE
= P(t, Ti , T j )(κΦ(−d− (Ti − t )) − Xbt Φ(−d+ (Ti − t )))
= P(t, Ti , T j )κΦ(−d− (Ti − t )) − P(t, Ti , T j )S(t, Ti , T j )Φ(−d+ (Ti − t ))
= P(t, Ti , T j )κΦ(−d− (Ti − t )) − (P(t, Ti ) − P(t, T j ))Φ(−d+ (Ti − t )),
2 /2
where e m = S(t, Ti , T j ) e −(T −t )σ , v2 = (T − t )σ 2 , and
√
log(S(t, Ti , T j )/κ ) σ Ti − t
d+ (Ti − t ) = √ + ,
σ Ti − t 2
and √
log(S(t, Ti , T j )/κ ) σ Ti − t
d− (Ti − t ) = √ − ,
σ Ti − t 2
because S(t, Ti , T j ) is a driftless geometric Brownian motion with volatility σ under the
forward measure P b i, j .
Exercise 4.13
a) Apply the Itô formula to d (P(t, T1 )/P(t, T2 ).
b) We have w
1 w T1

T1
2
LT1 = Lt exp σ (s)d Bs −
b |σ (s)| ds .
t 2 t
c) We have
b (LT1 − κ )+ Ft

P(t, T2 )IE
r rT +
T
= P(t, T2 )IEb Lt e t 1 σ (s)d Bbs − 12 t 1 |σ (s)|2 ds − κ Ft

√
= P(t, T2 )Bl v(t, T1 )/ T1 − t, κ, 0, T1 − t

log(x/K ) v(t, T1 ) log(x/K ) v(t, T1 )
= P(t, T2 ) Lt Φ + − κΦ − .
v(t, T1 ) 2 v(t, T1 ) 2
d) Integrate the self-financing condition (4.5.23) between 0 and t.
e) We have rt
rt rt
= −rt e − 0 rs dsVt dt + e −
dVt 0 rs ds
rt rt
( 1 ) (2)
= −rt e 0 s ξt P(t, T1 ) − rt e − 0 rs ds ξt P(t, T2 )dt
− r ds
rt rt
(1) (2)
+ e − 0 rs ds ξt dP(t, T1 ) + e − 0 rs ds ξt dP(t, T2 )
(1) (2)
= ξt d Pe(t, T1 ) + ξt d Pe(t, T2 ), 0 ⩽ t ⩽ T1 .
since
d Pe(t, T1 ) d Pe(t, T2 )
= ζ1 (t )dt, = ζ2 (t )dt.
Pe(t, T1 ) Pe(t, T2 )
f) We apply the Itô formula and the fact that
E (LT1 − κ )+ Ft

t 7→ Ib
and (Lt )t∈R+ are both martingales under P. b Next we use the fact that
E (LT1 − κ )+ Ft ,

Vbt = Ib
and apply the result of Question (f)).

g) Apply the Itô rule to Vt = P(t, T2 )Vbt using Relation (4.5.21) and the result of Question (f)).
h) We have
∂C
dVt = (Lt , v(t, T1 ))P(t, T1 )(ζ1 (t ) − ζ2 (t ))dBt + Vbt dP(t, T2 )
∂x
∂C
= (Lt , v(t, T1 ))P(t, T1 )(ζ1 (t ) − ζ2 (t ))dBt + Vbt ζ2 (t )P(t, T2 )dBt
∂x
∂C
= (1 + Lt ) (Lt , v(t, T1 ))P(t, T2 )(ζ1 (t ) − ζ2 (t ))dBt + Vbt ζ2 (t )P(t, T2 )dBt
∂x
∂C ∂C
= Lt ζ1 (t ) (Lt , v(t, T1 ))P(t, T2 )dBt − Lt (Lt , v(t, T1 ))P(t, T2 )ζ2 (t )dBt
∂x ∂x
∂C
+ (Lt , v(t, T1 ))P(t, T2 )(ζ1 (t ) − ζ2 (t ))dBt + Vbt ζ2 (t )P(t, T2 )dBt
∂x
∂C ∂C
= Lt (Lt , v(t, T1 ))P(t, T2 )ζ1 (t )dBt + Vt − Lt
b (Lt , v(t, T1 )) P(t, T2 )ζ2 (t )dBt
∂x ∂x
∂C
+ (Lt , v(t, T1 ))P(t, T2 )(ζ1 (t ) − ζ2 (t ))dBt ,
∂x
hence
∂C
dVet = Lt ζ1 (t ) (Lt , v(t, T1 ))Pe(t, T2 )dBt
∂x
∂C
+ Vbt − Lt (Lt , v(t, T1 )) Pe(t, T2 )ζ2 (t )dBt
∂x
∂C
+ (Lt , v(t, T1 ))Pe(t, T2 )(ζ1 (t ) − ζ2 (t ))dBt
∂x
∂C
= (Pe(t, T1 ) − Pe(t, T2 ))ζ1 (t ) (Lt , v(t, T1 ))dBt
x
∂
∂C
+ Vbt − Lt (Lt , v(t, T1 )) d Pe(t, T2 )
∂x

∂C
+ (Lt , v(t, T1 ))Pe(t, T2 )(ζ1 (t ) − ζ2 (t ))dBt
∂x
∂C
= (ζ1 (t )Pe(t, T1 ) − ζ2 (t )Pe(t, T2 )) (Lt , v(t, T1 ))dBt
∂x
∂C
+ Vbt − Lt (Lt , v(t, T1 )) d Pe(t, T2 )
∂x

∂C ∂C
= (Lt , v(t, T1 ))d (P(t, T1 ) − P(t, T2 )) + Vt − Lt
e e b (Lt , v(t, T1 )) d Pe(t, T2 ).
∂x ∂x
Problem 4.14
a) We have
w
1 w Ti

Ti
2
S(Ti , Ti , T j ) = S(t, Ti , T j ) exp σi, j (s)dBi,s j − |σi, j | (s)ds .
t 2 t
b) We have
P(t, Ti , T j ) IEi, j (S(Ti , Ti , T j ) − κ )+ Ft

rT i, j 1 r Ti
+
i 2
= P(t, Ti , T j ) IEi, j S(t, Ti , T j ) e t σi, j (s)dBs − 2 t |σi, j | (s)ds − κ Ft
√
= P(t, Ti , T j )Bl v(t, Ti )/ Ti − t, κ, 0, Ti − t
= P(t, Ti , T j )

log(x/K ) v(t, Ti ) log(x/K ) v(t, Ti )
× S(t, Ti , T j )Φ + − κΦ − ,
v(t, Ti ) 2 v(t, Ti ) 2
where w Ti
v2 (t, Ti ) = |σi, j |2 (s)ds.
t
c) Integrate the self-financing condition (4.5.28) between 0 and t.

d) We have rt
rt rt
= −rt e − 0 rs dsVt dt + e − 0 rs ds dVt
rt j rt j
− 0 rs ds
(k ) − 0 rs ds
(k )
= −rt e ∑ ξt P(t, Tk )dt + e ∑ ξt dP(t, Tk )
k =i k =i
j
(k )
= ∑ ξt d Pe(t, Tk ), 0 ⩽ t ⩽ Ti .
k =i
since
d Pe(t, Tk )
= ζk (t )dt, k = i, . . . , j.
Pe(t, Tk )
e) We apply the Itô formula and the fact that
t 7→ IEi, j (S(Ti , Ti , T j ) − κ )+ Ft

and (St )t∈R+ are both martingales under Pi, j .

f) Use the fact that
Vbt = IEi, j (S(Ti , Ti , T j ) − κ )+ Ft ,

and apply the result of Question (e)).

g) Apply the Itô rule to Vt = P(t, Ti , T j )Vbt using Relation (4.5.25) and the result of Question (f)).

h) From the expression (4.5.27) of the swap rate volatilities we have

∂C
dVt = St (St , v(t, Ti ))
∂x !
j−1
× ∑ (Tk+1 − Tk )P(t, Tk+1 )(ζi (t ) − ζk+1 (t )) + P(t, Tj )(ζi (t ) − ζ j (t )) dBt
k =i
+Vbt dP(t, Ti , T j )
∂C
= St (St , v(t, Ti ))
∂x !
j−1
× ∑ (Tk+1 − Tk )P(t, Tk+1 )(ζi (t ) − ζk+1 (t )) + P(t, Tj )(ζi (t ) − ζ j (t )) dBt
k =i
j−1
+Vbt ∑ (Tk+1 − Tk )ζk+1 (t )P(t, Tk+1 )dBt
k =i
j−1
∂C
= St (St , v(t, Ti )) ∑ (Tk+1 − Tk )P(t, Tk+1 )(ζi (t ) − ζk+1 (t ))dBt
∂x k =i
∂C
+ (St , v(t, Ti ))P(t, T j )(ζi (t ) − ζ j (t ))dBt
∂x
j−1
k =i
j−1
∂C
= St ζi (t ) (St , v(t, Ti )) ∑ (Tk+1 − Tk )P(t, Tk+1 )dBt
∂x k =i
j−1
∂C
−St (St , v(t, Ti )) ∑ (Tk+1 − Tk )P(t, Tk+1 )ζk+1 (t )dBt
∂x k =i
∂C
+ (St , v(t, Ti ))P(t, T j )(ζi (t ) − ζ j (t ))dBt
∂x
j−1
k =i
j−1
∂C
= St ζi (t ) (St , v(t, Ti )) ∑ (Tk+1 − Tk )P(t, Tk+1 )dBt
∂x k =i
j−1
∂C
+ Vt − St
b (St , v(t, Ti )) ∑ (Tk+1 − Tk )P(t, Tk+1 )ζk+1 (t )dBt
∂x k =i
∂C
+ (St , v(t, Ti ))P(t, T j )(ζi (t ) − ζ j (t ))dBt .
∂x
i) We have
j−1
∂C
dVet = St ζi (t ) (St , v(t, Ti )) ∑ (Tk+1 − Tk )Pe(t, Tk+1 )dBt
∂x k =i
j−1
∂C
+ Vbt − St (St , v(t, Ti )) ∑ (Tk+1 − Tk )Pe(t, Tk+1 )ζk+1 (t )dBt
∂x k =i
∂C
+ (St , v(t, Ti ))Pe(t, T j )(ζi (t ) − ζ j (t ))dBt
∂x
∂C
= (Pe(t, Ti ) − Pe(t, T j ))ζi (t ) (St , v(t, Ti ))dBt
∂x


∂C
+ Vt − St
b (St , v(t, Ti )) d Pe(t, Ti , T j )
∂x
∂C
+ (St , v(t, Ti ))Pe(t, T j )(ζi (t ) − ζ j (t ))dBt
∂x
∂C
= (ζi (t )Pe(t, Ti ) − ζ j (t )Pe(t, T j )) (St , v(t, Ti ))dBt
∂x
∂C
+ Vbt − St (St , v(t, Ti )) d Pe(t, Ti , T j )
∂x
∂C
= (St , v(t, Ti ))d (Pe(t, Ti ) − Pe(t, T j ))
∂ x
∂C
+ Vbt − St (St , v(t, Ti )) d Pe(t, Ti , T j ).
∂x
j) We have
∂C ∂ v 1 x v 1 x
(x, τ, v) = xΦ + log − κΦ − + log
∂x ∂x 2 v κ 2 v κ

∂ v 1 x ∂ v 1 x v 1 x
= x Φ + log − κ Φ − + log +Φ + log
∂x 2 v κ ∂x 2 v κ 2 v κ
1 −1 2 κ −1 2
= √ e −(v/2+v log(x/κ )) /2 − √ e −(−v/2+v log(x/κ )) /2
v 2π vx 2π

log(x/κ ) v log(x/κ ) v
+Φ + =Φ + .
v 2 v 2
k) We have
∂C
dVet = (St , v(t, Ti ))d (Pe(t, Ti ) − Pe(t, T j ))
∂ x
∂C
+ Vbt − St (St , v(t, Ti )) d Pe(t, Ti , T j )
∂x

= Φ + d (Pe(t, Ti ) − Pe(t, T j ))
v(t, Ti ) 2

−κΦ − d Pe(t, Ti , T j ).
v(t, Ti ) 2
l) We compare the results of Questions (d)) and (k)).
Chapter 5
Exercise 5.1 By absence of arbitrage we have (1 − α ) e rd T = e rT , hence α = 1 − e (r−rd )T .
Exercise 5.2
a) The bond payoff 1{τ>T −t} is discounted according to the risk-free rate, before taking expec-
tation.
b) We have IE 1{τ>T −t} = e −λ (T −t ) , hence Pd (t, T ) = e −(λ +r)(T −t ) .

c) We have PM (t, T ) = e −(λ +r)(T −t ) , hence λ = −r − T 1−t log PM (t, T ).
Exercise 5.3
a) We have wt
(1)
rt = −a rs ds + σ Bt , t ⩾ 0,
0
hence w
t 1 (1)
rs ds = σ Bt − rt
0 a

σ (1) w t −(t−s)a (1)
= B − e dBs
aw t 0
σ t (1)
= (1 − e −(t−s)a )dBs ,
a 0
and w wT wt
T
rs ds = rs ds − rs ds
t 0 0
σ wT −(T −s)a (1) σ wt (1)
= (1 − e )dBs − (1 − e −(t−s)a )dBs
a 0 a 0
σ w t −(T −s)a wT
−(t−s)a (1) −(T −s)a (1)
= − (e −e )dBs + ( e − 1)dBs
a 0 t
σ wt σ w T −(T −s)a
(1) (1)
= − ( e −(T −t )a − 1) e −(t−s)a dBs − (e − 1)dBs
a 0 a t
1 σ w T −(T −s)a (1)
= − ( e −(T −t )a − 1)rt − (e − 1)dBs .
a a t
The answer for λt is similar.
b) As a consequence of the answer to the previous question, we have
w wT
T
IE rs ds + λs ds Ft = C (a,t, T )rt + C (b,t, T )λt ,
t t
and w wT
T
Var rs ds + λs ds Ft
t t
w w
T T
= Var rs ds Ft + Var λs ds Ft
t t
w wT
T
+2 Cov Xs ds, Ys ds Ft
t t
σ2 w T
= ( e −(T −s)a − 1)2 ds
a2 t
σ η w T −(T −s)a
+2ρ (e − 1)( e −(T −s)b − 1)ds
ab t
η 2 w T −(T −s)b
+ 2 (e − 1)2 ds
b t
wT wT
= σ 2 C2 (a, s, T )ds + 2ρσ η C (a, s, T )C (b, s, T )ds
t t
wT
+η 2 C2 (b, sT )ds,
t
from the Itô isometry.

a) We use the fact that (rt , λt )t∈[0,T ] is a Markov process.
b) We use the tower property of the conditional expectation given Ft .
c) Writing F (t, rt , λt ) = P(t, T ), we have
rt
d e − 0 (rs +λs )ds P(t, T )
rt rt
= −(rt + λt ) e − 0 (rs +λs )ds P(t, T )dt + e − 0 (rs +λs )ds dP(t, T )
rt rt
= −(rt + λt ) e − 0 (rs +λs )ds P(t, T )dt + e − 0 (rs +λs )ds
dF (t, rt , λt )
rt rt∂F
= −(rt + λt ) e − 0 (rs +λs )ds P(t, T )dt + e − 0 (rs +λs )ds (t, rt , λt )drt
∂x
rt ∂F 1 rt ∂ 2F
+ e − 0 (rs +λs )ds (t, rt , λt )dλt + e − 0 (rs +λs )ds 2 (t, rt , λt )σ12 (t, rt )dt
∂y 2 ∂x

1 rt ∂ 2F
+ e − 0 (rs +λs )ds 2 (t, rt , λt )σ22 (t, λt )dt
2 ∂y
rt ∂ 2F rt ∂F
+ e − 0 (rs +λs )ds ρ (t, rt , λt )σ1 (t, rt )σ2 (t, λt )dt + e − 0 (rs +λs )ds (t, rt , λt )dt
∂ x∂ y ∂t
rt ∂F rt ∂F
(1) (2)
= e − 0 (rs +λs )ds (t, rt , λt )σ1 (t, rt )dBt + e − 0 (rs +λs )ds (t, rt , λt )σ2 (t, λt )dBt
∂x ∂y
rt

∂ F
+ e − 0 (rs +λs )ds −(rt + λt )P(t, T ) + (t, rt , λt ) µ1 (t, rt )
∂x
∂F 1 ∂ 2F 2 1 ∂ 2F
+ (t, rt , λt ) µ2 (t, λt ) + ( t, rt , λt ) σ1 (t, rt ) + (t, rt , λt )σ22 (t, λt )
∂y 2 ∂ x2 2 ∂ y2
∂ 2F

∂F
+ρ (t, rt , λt )σ1 (t, rt )σ2 (t, λt ) + (t, rt , λt ) dt,
∂ x∂ y ∂t
hence the bond pricing PDE is
∂F
− (x + y)F (t, x, y) + µ1 (t, x) (t, x, y)
∂x
∂F 1 ∂ 2F
+ µ2 (t, y) (t, x, y) + σ12 (t, x) 2 (t, x, y)
∂y 2 ∂x
1 2
∂ F ∂ 2F ∂F
+ σ22 (t, y) 2 (t, x, y) + ρσ1 (t, x)σ2 (t, y) (t, x, y) + (t, rt , λt ) = 0.
2 ∂y ∂ x∂ y ∂t
d) We have w

T wT
P(t, T ) = 1{τ>t} IE exp − rs ds − λs ds Ft
t t
w w
T T
= 1{τ>t} exp − IE rs ds Ft − IE λs ds Ft
t t
w wT
1 T
× exp Var rs ds + λs ds Ft
2 t t
= 1{τ>t} exp (−C (a,t, T )rt −C (b,t, T )λt )

2w
η2 w T 2

σ T
2 −(T −s)b
× exp C (a, s, T )ds + C (b, s, T ) e ds
2 t 2 t
wT
× exp ρσ η C (a, s, T )C (b, s, T )ds .
t
e) This is a direct consequence of the answers to Questions (c)) and d).
f) The above analysis shows that w
T
P(τ > T | Gt ) = 1{τ>t} IE exp − λs ds Ft
t
η2 w T 2

= 1{τ>t} exp −C (b,t, T )λt + C (b, s, T )ds ,
2 t
for a = 0 and
wT
σ2 w T 2

IE exp − rs ds Ft = exp −C (a,t, T )rt + C (a, s, T )ds ,
t 2 t
for b = 0, and this implies
wT
Uρ (t, T ) = exp ρσ η C (a, s, T )C (b, s, T )ds
t
ση
= exp ρ (T − t −C (a,t, T ) −C (b,t, T ) + C (a + b,t, T )) .
ab

g) We have
∂
f (t, T ) = −1{τ>t} log P(t, T )
∂T
σ2 2 η2 2

= 1{τ>t} rt e −(T −t )a
− C (a,t, T ) + λt e −(T −t )b
− C (b,t, T )
2 2
−1{τ>t} ρσ ηC (a,t, T )C (b,t, T ).
h) We use the relation w
T
P(τ > T | Gt ) = 1{τ>t} IE exp − λs ds Ft
t
η2 w T 2

= 1{τ>t} exp −C (b,t, T )λt + C (b, s, T )ds
2 t
rT
= 1{τ>t} e − t f2 (t,u)du ,
where f2 (t, T ) is the Vasicek forward rate corresponding to λt , i.e.
η2 2
f2 (t, u) = λt e −(u−t )b − C (b,t, u).
2
i) In this case we have ρ = 0 and
w
T
P(t, T ) = P(τ > T | Gt ) IE exp − rs ds Ft ,
t
since Uρ (t, T ) = 0.

Bibliography
Articles
[AL05] C. Albanese and S. Lawi. “Laplace transforms for integrals of Markov processes”. In:
Markov Process. Related Fields 11.4 (2005), pages 677–724 (Cited on page 5).
[Bla76] F. Black. “The pricing of commodity contracts”. In: J. of Financial Economics 3
(1976), pages 167–179 (Cited on page 100).
[BDT90] F. Black, E. Derman, and W. Toy. “A One-Factor Model of Interest Rates and Its
Application to Treasury Bond Options”. In: Financial Analysts Journal 46.1 (1990),
pages 24–32 (Cited on page 32).
[BGM97] A. Brace, D. Gatarek, and M. Musiela. “The market model of interest rate dynamics”.
In: Math. Finance 7.2 (1997), pages 127–155 (Cited on page 59).
[BHM18] D.C. Brody, L.P. Hughston, and D.M. Meier. “Lévy-Vasicek Models and the Long-
Bond Return Process”. In: Int. J. Theor. Appl. Finance 21.3 (2018), page 1850026
(Cited on pages 28 and 136).
[CN10] C. Pintoux and N. Privault. “A direct solution to the Fokker-Planck equation for expo-
nential Brownian functionals”. In: Analysis and Applications 8.3 (2010), pages 287–
304 (Cited on page 24).
[CN11] C. Pintoux and N. Privault. “The Dothan pricing model revisited”. In: Math. Finance
21 (2011), pages 355–363 (Cited on page 23).
[Cha+92] K.C. Chan et al. “An Empirical Comparison of Alternative Models of the Short-Term
Interest Rate”. In: The Journal of Finance 47.3 (1992). Papers and Proceedings of the
Fifty-Second Annual Meeting of the American Finance Association, New Orleans,
Louisiana, pages 1209–1227 (Cited on pages 28 and 29).
[Che+08] R.-R. Chen, X. Cheng, et al. “An Explicit, Multi-Factor Credit Default Swap Pricing
Model with Correlated Factors”. In: Journal of Financial and Quantitative Analysis
43.1 (2008), pages 123–160 (Cited on page 129).

[Cou82] G. Courtadon. “The Pricing of Options on Default-Free Bonds”. In: The Journal of
Financial and Quantitative Analysis 17.1 (1982), pages 75–100 (Cited on pages 6 and
29).
[CIR85] J.C. Cox, J.E. Ingersoll, and S.A. Ross. “A Theory of the Term Structure of Interest
Rates”. In: Econometrica 53 (1985), pages 385–407 (Cited on pages 4, 5, 6, 17,
26, and 29).
[Dot78] L.U. Dothan. “On the term structure of interest rates”. In: Jour. of Fin. Ec. 6 (1978),
pages 59–69 (Cited on pages 6 and 23).
[EJ99] R.J. Elliott and M. Jeanblanc. “Incomplete markets with jumps and informed agents”.
In: Math. Methods Oper. Res. 50.3 (1999), pages 475–492 (Cited on page 128).
[EJY00] R.J. Elliott, M. Jeanblanc, and M. Yor. “On models of default risk”. In: Math. Finance
10.2 (2000), pages 179–195 (Cited on page 127).
[FG06] R. Faff and P. Gray. “On the estimation and comparison of short-rate models using the
generalised method of moments”. In: Journal of Banking and Finance 30 (11 2006),
[Fel51] W. Feller. “Two singular diffusion problems”. In: Ann. of Math. (2) 54 (1951),
[GK83] M.B. Garman and S.W. Kohlhagen. “Foreign Currency Option Values”. In: J. Interna-
tional Money and Finance 2 (1983), pages 231–237 (Cited on pages 78, 79, and
80).
[GER95] H. Geman, N. El Karoui, and J.-C. Rochet. “Changes of numéraire, changes of
probability measure and option pricing”. In: J. Appl. Probab. 32.2 (1995), pages 443–
458 (Cited on pages 67 and 85).
[HJM92] D. Heath, R. Jarrow, and A. Morton. “Bond pricing and the term structure of interest
rates: a new methodology”. In: Econometrica 60 (1992), pages 77–105 (Cited on
page 49).
[HL86] S.Y. Ho and S.B. Lee. “Term structure movements and pricing interest rate contingent
claims”. In: Journal of Finance 41 (5 1986), pages 1011–1029 (Cited on page 7).
[HW90] J. Hull and A. White. “Pricing interest rate derivative securities”. In: The Review of
Financial Studies 3 (1990), pages 537–592 (Cited on pages 7 and 49).
[Jam89] F. Jamshidian. “An exact bond option formula”. In: The Journal of Finance XLIV.1
(1989), pages 205–209 (Cited on pages 97 and 113).
[Jam96] F. Jamshidian. “Sorting out swaptions”. In: Risk Magazine 9.3 (1996), pages 59–60
(Cited on page 85).
[Jen06] J.L.W.V. Jensen. “Sur les fonctions convexes et les inégalités entre les valeurs moyennes”.
In: Acta Math. 30 (1906), pages 175–193 (Cited on page 102).
[Kak15] Z. Kakushadze. “Path integral and asset pricing”. In: Quant. Finance 15.11 (2015),
[Kim02] Y.-J. Kim. “Option Pricing under Stochastic Interest Rates: An Empirical Investiga-
tion”. In: Asia-Pacific Financial Markets 9 (1 2002), pages 23–44 (Cited on page 98).
[Lan98] D. Lando. “On Cox processes and credit risky securities”. In: Review of Derivative
Research 2 (1998), pages 99–120 (Cited on pages 125, 126, and 128).

186 Bibliography
[Lin07] E. Lindström. “Estimating parameters in diffusion processes using an approximate

maximum likelihood approach”. In: Annals of Operations Research 151 (2007),
[Mam04] R.S. Mamon. “Three ways to solve for bond prices in the Vasicek model”. In: Journal
of Applied Mathematics and Decision Sciences 8.1 (2004), pages 1–14 (Cited on
page 147).
[Mar78] W. Margrabe. “The value of an option to exchange one asset for another”. In: The
Journal of Finance XXXIII.1 (1978), pages 177–186 (Cited on pages 83 and 87).
[MR83] T.A. Marsh and E.R. Rosenfeld. “Stochastic Processes for Interest Rates and Equi-
librium Bond Prices”. In: The Journal of Finance 38.2 (1983). Papers and Proceed-
ings Forty-First Annual Meeting American Finance Association New York, N.Y.,
pages 635–646 (Cited on pages 6 and 29).
[Mer73] R. Merton. “Theory of rational option pricing”. In: Bell Journal of Economics 4.1
(1973), pages 141–183 (Cited on page 84).
[NS87] C.R. Nelson and A.F. Siegel. “Parsimonious Modeling of Yield Curves”. In: Journal
of Business 60 (1987), pages 473–489 (Cited on pages 34 and 52).
[PP17] A. Prayoga and N. Privault. “Pricing CIR yield options by conditional moment match-
ing”. In: Asia-Pacific Financial Markets 24 (2017), pages 19–38 (Cited on page 26).
[PT12] N. Privault and T.-R. Teng. “Risk-neutral hedging of interest rate derivatives”. In: Risk
and Decision Analysis 3 (2012), pages 201–209 (Cited on pages 85, 100, and 109).
[PU13] N. Privault and W.T. Uy. “Monte Carlo Computation of the Laplace Transform of
Exponential Brownian Functionals”. In: Methodol. Comput. Appl. Probab. 15.3 (2013),
[PW09] N. Privault and X. Wei. “Calibration of the LIBOR market model - implementation
in PREMIA”. In: Bankers, Markets & Investors 99 (2009), pages 20–28 (Cited on
page 110).
[PY16] N. Privault and J.D. Yu. “Stratified approximations for the pricing of options on
average”. In: Journal of Computational Finance 19.4 (2016), pages 95–113 (Cited on
page 25).
[Pro01] P. Protter. “A partial introduction to financial asset pricing theory”. In: Stochastic
Process. Appl. 91.2 (2001), pages 169–203 (Cited on page 86).
[SS01] P. Santa-Clara and D. Sornette. “The Dynamics of the Forward Interest Rate Curve
with Stochastic String Shocks”. In: The Review of Financial Studies 14.1 (2001),
[Vaš77] O. Vašíček. “An equilibrium characterisation of the term structure”. In: Journal of
Financial Economics 5 (1977), pages 177–188 (Cited on pages 2, 16, 17, 38, 96,
137, and 149).
[Wu00] X. Wu. “A New Stochastic Duration Based on the Vasicek and CIR Term Structure
Theories”. In: Journal of Business Finance and Accounting 27 (2000) (Cited on
page 33).
[Yor92] M. Yor. “On some exponential functionals of Brownian motion”. In: Adv. in Appl.
Probab. 24.3 (1992), pages 509–531 (Cited on page 24).

Books
[Bou73] K.E. Boulding. In “Energy Reorganization Act of 1973. Hearings, Ninety-third
Congress, first session, on H.R. 11510”. Washington: U.S. Government Printing
Office, 1973, pages iv+422 (Cited on page 65).
[BM06] D. Brigo and F. Mercurio. Interest rate models—theory and practice. Second. Springer
Finance. Berlin: Springer-Verlag, 2006, pages lvi+981 (Cited on pages 16, 58, and
141).
[Cha14] A. Charpentier, editor. Computational Actuarial Science with R. The R Series. USA:
Chapman & Hall/CRC, 2014 (Cited on pages 40 and 55).
[Das04] J.W. Dash. Quantitative finance and risk management. River Edge, NJ: World Scien-
tific Publishing Co. Inc., 2004, pages xx+781 (Cited on page 70).
[DMM92] C. Dellacherie, B. Maisonneuve, and P.A. Meyer. Probabilités et Potentiel. Volume 5.
Hermann, 1992. Chapter XVII-XXIV (Cited on page 127).
[DJD08] A. Downes, M.S. Joshi, and N. Denson. Quant Job Interview Questions and Answers.
first. CreateSpace Independent Publishing Platform, 2008, pages x+316 (Cited on
page 89).
[DS03] D. Duffie and K.J. Singleton. Credit risk. Pricing, measurement, and management.
Princeton Series in Finance. Princeton, NJ: Princeton University Press, 2003, pages xvi+396
(Cited on page 128).
[Jeu80] Th. Jeulin. Semi-martingales et grossissement d’une filtration. Volume 833. Lecture
Notes in Mathematics. Springer Verlag, 1980 (Cited on page 128).
[Key24] J.M. Keynes. A Tract on Monetary Reform. London: MacMillan & Co., 1924 (Cited
on page 2).
[Pri21] N. Privault. Stochastic Interest Rate Modeling With Fixed Income Derivative Pricing
(3rd edition). Advanced Series on Statistical Science & Applied Probability. 372 pp.
Singapore: World Scientific Publishing Co., 2021 (Cited on pages 49, 55, 57, 58,
59, 98, 110, and 120).
[Pri22] N. Privault. Introduction to Stochastic Finance with Market Examples (2nd edition).
Financial Mathematics Series. Chapman & Hall/CRC, 2022, pages x+662 (Cited on
pages 125, 128, and 129).
[Pro04] P. Protter. Stochastic integration and differential equations. second. Volume 21.
Stochastic Modelling and Applied Probability. Berlin: Springer-Verlag, 2004, pages xiv+419
(Cited on pages 14, 15, 72, 83, and 127).
[Sch05] J. Schoenmakers. Robust LIBOR modelling and pricing of derivative products. Chap-
man & Hall/CRC Financial Mathematics Series. Chapman & Hall/CRC, Boca Raton,
FL, 2005, pages xvi+202 (Cited on pages 109 and 110).
[Shr04] S.E. Shreve. Stochastic calculus for finance. II. Springer Finance. Continuous-time
models. New York: Springer-Verlag, 2004, pages xx+550 (Cited on page 92).
[Wat95] G. N. Watson. A treatise on the theory of Bessel functions. Reprint of the second
(1944) edition. Cambridge: Cambridge University Press, 1995, pages viii+804 (Cited
on page 25).

Index
Symbols B
code . . . . . . . . . . . . . 3, 5, 10, 19, 40, 77, 110 backward-looking bond price . . . . . . . . . . . . . 43

package basis point . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
quantmod . . . . . . . . . . . . . . . . . . . . . . . . 9, 40 BDT model. . . . . . . . . . . . . . . . . . . . . . . . . . . . .32
RQuantLib . . . . . . . . . . . . . . . . . . . . . . . . 110 Bermudan swaption . . . . . . . . . . . . . . . . . . . . 110
Sim.DiffProc . . . . . . . . . . . . . . . . . . . . . . . 10 Bessel function . . . . . . . . . . . . . . . . . . . . . . . 5, 25
YieldCurve . . . . . . . . . . . . . . . . . . . . . . . . . 40 BGM model . . . . . . . . . . . . . . . . . . . . . . . . 59, 99
binary option . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
A Black
(1976) formula . . . . . . . . . . . . . . . . . . . . 100
absence of arbitrage . . . . . . . . . . . . . . . . . . . . . 12 LIBOR caplet formula . . . . . . . . . . . . . . . 99
abstract Bayes formula . . . . . . . . . . . . . . . . . . 68 SOFR caplet formula . . . . . . . . . . . . . . . 102
accreting swap . . . . . . . . . . . . . . . . . . . . . . . . . . 46 SOFR swaption formula . . . . . . . . . . . . 109
adapted process . . . . . . . . . . . . . . . . . . . . . . . . 126 swaption formula . . . . . . . . . . . . . . . . . . 108
affine model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Black-Derman-Toy model . . . . . . . . . . . . . . . 32
affine PDE . . . . . . . . . . . . . . . . . . . . . . . . . . 7, 142 Black-Scholes
amortizing swap . . . . . . . . . . . . . . . . . . . . . . . . 46 formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
annuity call options, 100, 103
measure . . . . . . . . . . . . . . . . . . 104, 119, 121 put options, 100
numéraire . . . . . . . . . . . . . . . . . . . . 103, 118 bond
annuity numéraire . . . . . . . . . . . . . . . . . . . . . . . 45 convertible . . . . . . . . . . . . . . . . . . . . . . . . . 30
approximation convexity . . . . . . . . . . . . . . . . . . . . . . . . . . 33
gamma . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 corporate . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
arbitrage defaultable . . . . . . . . . . . . . . . . . . . . . . . . 129
absence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 duration . . . . . . . . . . . . . . . . . . . . . . . . 31, 33
asset pricing immunization . . . . . . . . . . . . . . . . . . . . . 144
first theorem ladder . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
continuous time, 49 option . . . . . . . . . . . . . . . . . . . . . 90, 97, 160

INDEX 189
pricing defaultable bonds . . . . . . . . . . . . . . . . . . 132, 133

PDE, 14, 58 deflated price . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
pricing PDE . . . . . . . . . . . . . . . . . . . . . . . 183 Delta
yield . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 hedging . . . . . . . . . . . . . . . . . . . . . . . . . 86, 87
zero-coupon . . . . . . . . . . . . . . . . . . . . . . . . 11 derivatives
break-even fixed income . . . . . . . . . . . . . . . . . . . . . . . 93
rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45, 46 interest rate . . . . . . . . . . . . . . . . . . . . . . . . 93
Bretton Woods . . . . . . . . . . . . . . . . . . . . . . . . . . 65 diffusion
Brownian elasticity . . . . . . . . . . . . . . . . . . . . . . . . . 6, 29
bridge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 digital option . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
buy back guarantee . . . . . . . . . . . . . . . . . . . . . . 82 dispersion index . . . . . . . . . . . . . . . . . . . . . . . . 26
distribution
C invariant . . . . . . . . . . . . . . . . . . . . . . . . . . 2, 6
lognormal . . . . . . . . . . . . . . . . . . . . . . . . . . 25
call stationary . . . . . . . . . . . . . . . . . . . . . . . . . 2, 6
swaption . . . . . . . . . . . . . . . . . . . . . . . . . . 107 dollar value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
call-put parity . . . . . . . . . . . . . . . . . . . . . . . . . . 80 Dothan model . . . . . . . . . . . . . . . . . . . . . . . . 6, 23
cap pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 duration . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31, 33
caplet pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
cash-or-nothing option . . . . . . . . . . . . . . . . . . . 91 E
CBOE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
CEV model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 ECB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
change of numéraire . . . . . . . . . . . . . . . . . 67, 80 elasticity of diffusion . . . . . . . . . . . . . . . . . . 6, 29
Chi square distribution . . . . . . . . . . . . . . . . . . . . 5 enlargement of filtration . . . . . . . . . . . . . . . . 129
Chicago Board Options Exchange . . . . . . . . . 4 Euclidean path integral . . . . . . . . . . . . . . . . . . 27
CIR model . . . . . . . . . . . . . . . . . . . . . . . . . . . 4, 29 EURIBOR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
CKLS model . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 exchange option . . . . . . . . . . . . . . . . . . . . . . . . 83
compounded yield to maturity . . . . . . . . . . . . 32 exponential
compounding distribution . . . . . . . . . . . . . . . . . . . . . . . . 126
linear . . . . . . . . . . . . . . . . . . . . . . . . . . . 42, 43 Vasicek model . . . . . . . . . . . . . . . . . . . . . . . 6
conditional
survival probability . . . . . . . . . . . . . . . . 124 F
conversion rate . . . . . . . . . . . . . . . . . . . . . . . . . 30
convertible bond . . . . . . . . . . . . . . . . . . . . . . . . 30 face value . . . . . . . . . . . . . . . . . . . . . . . . . . . 11, 32
convexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 failure rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
corporate bond . . . . . . . . . . . . . . . . . . . . . . . . . . 30 Fano factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
correlation FED . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
perfect . . . . . . . . . . . . . . . . . . . . . . . . . 57, 63 Feller condition . . . . . . . . . . . . . . . . . . . . . . . . . . 5
problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 filtration
cost of carry . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 enlargement of . . . . . . . . . . . . . . . . . . . . 129
coupon first theorem of asset pricing . . . . . . . . . . . . . 49
bond . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 fixed
rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 income . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Courtadon model . . . . . . . . . . . . . . . . . . . . . 6, 29 derivatives, 93
credit default leg . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
reduced-form approach . . . . . . . . . . . . . 124 rate . . . . . . . . . . . . . . . . . . . . . . . . . . . 98, 102
floating
D leg . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
rate . . . . . . . . . . . . . . . . . . . . . . . . . . . 98, 102
default rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126 floorlet . . . . . . . . . . . . . . . . . . . . . . . . . . . 100, 114

190 INDEX
foreign exchange . . . . . . . . . . . . . . . . . . . . . . . . 75 Courtadon, 6, 29

option . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 Cox-Ingersoll-Ross, 4
forward Dothan, 6, 23
contract . . . . . . . . . . . . . . . . . . . . 34, 89, 156 exponential Vasicek, 6
measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 Ho-Lee, 7
price . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 Hull-White, 7, 49
rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 Marsh-Rosenfeld, 6, 29
agreement, 34 Vasicek, 2, 7
spot, 34–36, 98, 102 short term . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
swap, 44 invariant distribution . . . . . . . . . . . . . . . . . . . 2, 6
swap rate . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 IPython notebook . . . . . . . . . . . . . . . . . . . . . . . 54
forward measure . . . . . . . . . . . . . . . . . . . . . . . . 67
forward swap J
measure . . . . . . . . . . . . . . . . . . 104, 119, 121
FRA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 Jamshidian’s trick . . . . . . . . . . . . . . . . . . . . . . 114
K
G
key lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
gamma
approximation . . . . . . . . . . . . . . . . . . . . . . 25
L
Garman-Kohlagen formula . . . . . . . . . . . . . . . 78
Gaussian Lagrangian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
cumulative distribution function . . . . . . 97 least square regression . . . . . . . . . . . . . . . . . . . . 7
Girsanov Theorem . . . . . . . . . . . . . . . . . . . . . . 72 leg
guarantee fixed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
buy back . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 floating . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Leibniz integral rule . . . . . . . . . . . . . . . . . . . . . 49
H LIBOR
model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
hedge and forget . . . . . . . . . . . . . . . . . . . . . . . 157 rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
hedging swap rate . . . . . . . . . . . . . 45, 105, 107, 108
change of numéraire . . . . . . . . . . . . . . . . 85 LIBOR-SOFR swap . . . . . . . . . . . . . . . . . . . . 110
static . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 Lipschitz function . . . . . . . . . . . . . . . . . . . . . . . 83
HIBOR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 lognormal
HJM distribution . . . . . . . . . . . . . . . . . . . . . . . . . 25
condition . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
model . . . . . . . . . . . . . . . . . . . . . . . . . 48, 129 M
SOFR model . . . . . . . . . . . . . . . . . . . . . . . 52
Ho-Lee model . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Macaulay duration . . . . . . . . . . . . . . . . . . . . . . 33
Hull-White model . . . . . . . . . . . . . . . . . . . . 7, 49 Margrabe formula . . . . . . . . . . . . . . . . . . . . . . . 83
market
I price of risk . . . . . . . . . . . . . . . . . . . . . . . . 14
Markov property . . . . . . . . . . . . . . . . . . . . 83, 86
immunization . . . . . . . . . . . . . . . . . . . . . . . . . . 144 Marsh-Rosenfeld model . . . . . . . . . . . . . . . 6, 29
instantaneous forward rate . . . . . . . . . . . . . . . 36 maturity
interest rate transformation . . . . . . . . . . . . . . . . . . . . . . 35
derivatives . . . . . . . . . . . . . . . . . . . . . . . . . 93 mean
model reversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
affine, 6 model
Constant Elasticity of Variance, 6 affine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

INDEX 191
BDT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 PDE
CEV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 affine . . . . . . . . . . . . . . . . . . . . . . . . . . . 7, 142
CIR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4, 29 perfect correlation . . . . . . . . . . . . . . . . . . . 57, 63
CKLS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 Planck constant . . . . . . . . . . . . . . . . . . . . . . . . . 26
Courtadon . . . . . . . . . . . . . . . . . . . . . . . 6, 29 Poisson
Dothan . . . . . . . . . . . . . . . . . . . . . . . . . . 6, 23 process . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
exponential Vasicek . . . . . . . . . . . . . . . . . . 6 power option . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
Ho-Lee . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 principal amount . . . . . . . . . . . . . . . . . . . . . . . 113
Hull-White . . . . . . . . . . . . . . . . . . . . . . . 7, 49 put
Marsh-Rosenfeld . . . . . . . . . . . . . . . . . 6, 29 swaption . . . . . . . . . . . . . . . . . . . . . . . . . . 109
Vasicek . . . . . . . . . . . . . . . . . . . . . . . . . . . 2, 7 Python code . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
modified
Bessel function . . . . . . . . . . . . . . . . . . . 5, 25 Q
duration . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
MPoR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 Quantlib . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
Musiela notation . . . . . . . . . . . . . . . . . . . . . . . . 48 quantmod ( package) . . . . . . . . . . . . . . . 9, 40
N R
Nelson-Siegel . . . . . . . . . . . . . . . . . . . . . . . 52, 55 Radon-Nikodym density . . . . . . . . . . . . . . . . . 67

nominal value . . . . . . . . . . . . . . . . . . . . . . . . . . 32 rate
noncentral Chi square . . . . . . . . . . . . . . . . . . . . 5 default . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
notional . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 forward . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
principal . . . . . . . . . . . . . . . . . . . . . 113, 165 forward swap . . . . . . . . . . . . . . . . . . . . . . . 44
numéraire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 instantaneous forward . . . . . . . . . . . . . . . 36
annuity . . . . . . . . . . . . . . . . . . . . . . . . 45, 103 LIBOR . . . . . . . . . . . . . . . . . . . . . . . . . 42, 46
invariance . . . . . . . . . . . . . . . . . . . . . . . . . . 85 swap, 45
numéraire invariance . . . . . . . . . . . . . . . . . . . . 86 LIBOR swap . . . . . . . . . . . . . 105, 107, 108
SOFR swap . . . . . . . . . . . . . . . . . . . . . . . 109
O swap. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .44
receiver swaption . . . . . . . . . . . . . . . . . . . . . . 109
OLS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 recovery rate . . . . . . . . . . . . . . . . . . . . . . 129, 130
option reduced-form approach . . . . . . . . . . . . . . . . . 124
binary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .91 repo market . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
cash-or-nothing . . . . . . . . . . . . . . . . . . . . . 91 repurchase agreement . . . . . . . . . . . . . . . . . . . 43
digital. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .91 Riccati equation . . . . . . . . . . . . . . . . . . . . . . . . 19
Ornstein-Uhlenbeck process . . . . . . . . . . . . . . . 2 risk
market price . . . . . . . . . . . . . . . . . . . . . . . . 14
P RQuantLib . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
par value . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11, 32 S

parity
call-put . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 self-financing portfolio
path change of numéraire . . . . . . . . . . . . . . . . 86
freezing . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 seller swap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
integral . . . . . . . . . . . . . . . . . . . . . . . . . 26, 70 SHIBOR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Euclidean, 27 short rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
payer SIBOR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
swap. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .45 Sim.DiffProc . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
swaption . . . . . . . . . . . . . . . . . . . . . . . . . . 105 SOFR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

192 INDEX
Black caplet formula . . . . . . . . . . . . . . . 102 Z

caplet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
forward rate . . . . . . . . . . . . . . . . . . . . . . . . 44 zero-
HJM model . . . . . . . . . . . . . . . . . . . . . . . . 52 coupon bond. . . . . . . . . . . . . . . . . . . . . . . .11
swap rate . . . . . . . . . . . . . . . . . . . . . . 47, 109
swaption . . . . . . . . . . . . . . . . . . . . . . . . . . 109
spot forward rate . . . . . . . . . . . . 34–36, 98, 102
static hedging . . . . . . . . . . . . . . . . . . . . . . . . . . 157
stationary distribution . . . . . . . . . . . . . . . . . . 2, 6
stochastic
default . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
stopping time . . . . . . . . . . . . . . . . . . . . . . . . . . 126
string model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
survival probability . . . . . . . . . . . . . . . . . . . . 124
Svensson parametrization . . . . . . . . . . . . . . . . 53
swap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
amortizing . . . . . . . . . . . . . . . . . . . . . . . . . 46
forward measure . . . . . . . . . . 104, 119, 121
payer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44, 46
seller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
swaption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
Bermudan . . . . . . . . . . . . . . . . . . . . . . . . 110
telescoping sum . . . . . . . . . . . . . . . . . . . . . . . . . 47
tenor structure . . . . . . . . . . . . . . . 45, 66, 93, 130
theorem
asset pricing . . . . . . . . . . . . . . . . . . . . . . . . 49
Girsanov . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
TIBOR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
tower property . . . . . . . . . . . . . 14, 68, 148, 182
treasury note . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
two-factor model . . . . . . . . . . . . . . . . . . . . . . . . 58
variable rate . . . . . . . . . . . . . . . . . . . . . . . . 98, 102

Vasicek model . . . . . . . . . . . . . . . . . . . . . . . . . 2, 7
yield . . . . . . . . . . . . . . . . . . . . . . . 34, 36, 98, 102

bond . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
compounded to maturity . . . . . . . . . . . . . 32
curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
data, 40
inversion, 41
YieldCurve ( package). . . . . . . . . . . . . . . . . 40

INDEX 193
Author index
Albanese, C. 5 Jacod, J. 129

Jamshidian, F. 85, 97, 114
Bickersteth, M. 44 Jarrow, R. 49, 127
Björk, T. 55 Jeanblanc, M. 128, 129
Black, F. 32, 99, 102, 108 Jensen, J. 102
Boulding, K.E. 65 Jeulin, Th. 129
Brace, A. 59 Joshi, M.S. 89
Brigo, D. 16, 58, 142
Brody, D.C. 28 Kakushadze, Z. 27
Karolyi, G.A. 28
Chan, K.C. 28 Keynes, J.M. 2
Charpentier, A. 40 Kim, Y.-J. 98
Chen, R.R. 130 Kohlhagen, S.W. 78
Cheng, X. 130 Lando, D. 126
Courtadon, G. 6, 29 Lawi, S. 5
Cox, J.C. 4 Lee, S.B. 7
Lindström, E. 138
Dash, J. 70
Liu, B. 130
Dellacherie, C. 128
Longstaff, F.A. 28
Denson, N. 89
Lyashenko, A. 52
Derman, E. 32
Dothan, L.U. 6, 23 Maisonneuve, B. 128
Downes, A. 89 Mamon, R.S. 148
Margrabe, W. 83
El Karoui, N. 67, 85 Marsh, T.A. 6, 29
Elliott, R.J. 128, 129 Meier, D.M. 28
Menn, C. 127
Fabozzi, F. 130 Mercurio, F. 16, 58, 110, 142
Faff, R. 137 Merton, R.C. 84
Feller, W. 5 Meyer, P.A. 128
Morton, A. 49
Garman, M.B. 78
Musiela, M. 48, 59
Gatarek, D. 59
Geman, H. 67, 85 Nelson, C.R. 52
Gray, P. 137
Pintoux, C. 23, 24
Guirreri, S. 40
Prayoga, A. 26
Guo, X. 127
Protter, P. 14, 15, 72, 83, 86, 128
Heath, D. 49 Rochet, J.-C. 67, 85
Ho, S.Y. 7 Rosenfeld, E.R. 6, 29
Huang, J.Z. 130 Ross, S.A. 4
Hughston, L.P. 28 Rutkowski, M. 44
Hull, J. 7
Sanders, A.B. 28
Ingersoll, J.E. 4 Santa-Clara, P. 63

Schoenmakers, J. 109, 110 Vašíček, O. 2, 17
Shreve, S. 92
Siegel, A.F. 52
Wei, X. 110
Sornette, D. 63
White, A. 7
Svensson, L.E.O. 53
Wu, X. 33
Teng, T.-R. 85, 100, 109
Toy, B. 32
Yor, M. 24, 128, 129
Uy, W.I. 24 Yu, J.D. 25

FN6813 Notes

Uploaded by

Copyright:

Available Formats

FN6813 Notes

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

FN6813 Notes

Uploaded by

Copyright:

Available Formats

FN6813

Interest Rate Derivatives

February 3, 2024 https://personal.ntu.edu.sg/nprivault/indext.html

" February 3, 2024 FN6813 Interest Rate Derivatives - N. Privault

1 Short Rates and Bond Pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

2 Forward Rate Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

3 Change of Numéraire and Forward Measures . . . . . . . . . . . . . . . . . . 64

" February 3, 2024 FN6813 Interest Rate Derivatives - N. Privault

5 Reduced-Form Approach to Credit Risk . . . . . . . . . . . . . . . . . . . . . . . 123

Exercise Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

Author index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193

1.1 Short rate t 7→ rt in the Vasicek model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2.1 Graph of the spot forward rate S 7→ f (t,t, S) . . . . . . . . . . . . . . . . .. . . . . . 35

" February 3, 2024 FN6813 Interest Rate Derivatives - N. Privault

3.1 Why change of numéraire? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

4.1 Implied swaption volatilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

2.1 Stochastic interest rate models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

3.1 Local vs. foreign exchange options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

4.1 Forward rates arranged according to a tenor structure . . . . . . . . . . . . . . 94

5.1 Mortality table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

*Animated figures (work in Acrobat reader).

" February 3, 2024 FN6813 Interest Rate Derivatives - N. Privault

1. Short Rates and Bond Pricing

1.1 Short-Term Mean-Reverting Models 1

1.1 Short-Term Mean-Reverting Models

" February 3, 2024 FN6813 Interest Rate Derivatives - N. Privault

Vašíček, 1977 model

drt = (a − brt )dt + σ dBt , (1.1.1)

lim IE[a − brt ] = a − b lim IE[rt ] = 0.

February 3, 2024 FN6813 Interest Rate Derivatives - N. Privault "

from this distribution.

a=0.25;b=10;sigma=1; N=1000; t <- 0:N; dt <- 1.0/N; nsim=20;

* Eurozone interest rates turned negative in 2014.

" February 3, 2024 FN6813 Interest Rate Derivatives - N. Privault

Example - TNX yield

Jan 03 Jul 02 Jan 02 Jul 01 Jan 02 Jul 01 Jan 02 Jul 01 Dec 31

Figure 1.2: CBOE 10 Year Treasury Note (TNX) yield.

Cox-Ingersoll-Ross (CIR) model

February 3, 2024 FN6813 Interest Rate Derivatives - N. Privault "

N=10000; t <- 0:(N-1); dt <- 1.0/N; nsim <- 2

In large time t → ∞, using the asymptotics

" February 3, 2024 FN6813 Interest Rate Derivatives - N. Privault

drt = β (α − rt )dt + σ rt dBt ,

drt = rt (η − a log rt )dt + σ rt dBt ,

Constant Elasticity of Variance (CEV) model

dv2 (r )/v2 (r ) r dv(r ) d log v(r ) d log rγ/2

For β = 0 we get the standard CEV model

and for γ = 2 and β = 0 this yields the Dothan, 1978 model

drt = αrt dt + σ rt dBt ,

February 3, 2024 FN6813 Interest Rate Derivatives - N. Privault "

Time-dependent affine models

The family of affine models also includes:

drt = (θ (t ) − α (t )rt )dt + σ (t )dBt

1.2 Calibration of the Vasicek Model

drt = (a − brt )dt + σ dBt ,

rtk+1 − rtk = (a − brtk )∆t + σ Zk , k ⩾ 0,

" February 3, 2024 FN6813 Interest Rate Derivatives - N. Privault

This also yields

hence σ can be estimated as

February 3, 2024 FN6813 Interest Rate Derivatives - N. Privault "