CQF Module 4: Probabilistic Methods for Interest Rates

CQF Module 4: Probabilistic Methods for

Interest Rates
CQF
1/89
CQF Module 4: Probabilistic Methods for Interest Rates
In this lecture...
... we will apply probabilistic and martingale methods to the
pricing of bonds and derivatives struck on bonds using short-term
rate models. We will see:

the pricing of interest rate products in a probabilistic setting;

the equivalent martingale measures;

the fundamental asset pricing formula for bonds;

application for popular interest rates models;

the dynamics of bond prices;

the forward measure;

the fundamental asset pricing formula for derivatives on

bonds;

rights and wrongs of short-term interest rate models;

2/89
CQF Module 4: Probabilistic Methods for Interest Rates
Introduction
3/89
CQF Module 4: Probabilistic Methods for Interest Rates
1. A Model for the Short-Term Rate
In the world of equity, a single class of models (the Geometric
Borwnian Motion and its extensions) dominates the landscape.
The situation is markedly dierent in the world of interest rates,
where several classes of models coexist:
1. Single factor short-term rate models, such as the Merton,
Vasicek, Cox-Ingersoll-Ross, Ho-Lee or Hull-White models;
2. Multifactor models such as the Brennan-Schwartz,
Longsta-Schwartz and Fong-Vasicek models;
3. Forward rate models such as the Heath-Jarrow-Morton model
and its modication by Brace, Gatarek and Musiela.
4/89
CQF Module 4: Probabilistic Methods for Interest Rates
Our focus today will be on single factor short-term rate models
whose dynamics under the physical measure P is of the form
dr (t) = (t, r
t
)dt + (t, r
t
)dX(t), r (0) = r (1)
where r (t) is the short-term rate at time t and X(t) is a Brownian
motion.
To simplify the notation, we will write this equation as
dr (t) =
t
dt +
t
dX(t), r (0) = r (2)
5/89
CQF Module 4: Probabilistic Methods for Interest Rates
The specication (2) is general enough to cover both
1. Equilibrium models such as the Vasicek or Cox-Ingersoll-Ross
models;
2. No arbitrage models such as Ho-Lee and Hull-White;
6/89
CQF Module 4: Probabilistic Methods for Interest Rates
Equipped with a short-term interest rate model, we can dene a
money market or money-in-the-bank process A(t) as:
A(t) = e

t
0
r (s)ds
(3)
7/89
CQF Module 4: Probabilistic Methods for Interest Rates
2. The Zero-Coupon Bond Market
A zero-coupon bond B(t, T) is a bond which pays 1 at maturity
time T.
We will dene the zero-coupon bond market as the set of all the
zero-coupon bonds B(t, T) for all t T T.
8/89
CQF Module 4: Probabilistic Methods for Interest Rates
Our denition of the zero-coupon bond market is clearly
unrealistic:

in reality, zero-coupon bonds are scarce and illiquid

investments;

we cannot expect to nd a zero-coupon bond maturing at any

time t T T.
9/89
CQF Module 4: Probabilistic Methods for Interest Rates
However, we will nd this representation convenient since
zero-coupon bonds are the building block of the xed income
market:

any bond can be decomposed as a sequence of coupons;

each coupon can be modelled as a zero-coupon bond.

10/89
CQF Module 4: Probabilistic Methods for Interest Rates
Another reason for dening the entire bond market rather than a
single security stems from the observation in Lecture 4.2 that since
the short-term rate cannot be traded directly, the only way to
hedge a bond is by trading another bond.
With this idea, we have made our way from the so-called complete
markets of Modules 2 and 3 to incomplete markets.
11/89
CQF Module 4: Probabilistic Methods for Interest Rates
3. Pricing Zero-Coupon Bonds
As in lecture 4.2, our starting point will be the valuation of
zero-coupon bonds.
12/89
CQF Module 4: Probabilistic Methods for Interest Rates
3.1. Equivalent Martingale Measure for the Zero-Coupon
Bond Market
Denition (Equivalent Martingale Measure for the
Zero-Coupon Bond Market)
An equivalent martingale measure for the zero-coupon bond
market is a measure Q
1. equivalent to the physical measure P;
2. such that for any maturity T [0, T] the process
Z

(t, T) =
B(t, T)
A(t)
(4)
is a martingale for all t [0, T].
13/89
CQF Module 4: Probabilistic Methods for Interest Rates
3.1.1 Equivalent Measure or Equivalent Measures?
First, Denition is consistent with the way we dened the
equivalent martingale measure for a stock in Lecture 3.3.
But the key dierence in the zero-coupon bond case is that the
equivalent martingale measure does not apply to a single security,
but to a continuum of securities.
14/89
CQF Module 4: Probabilistic Methods for Interest Rates
As a result, we should not expect to nd a unique martingale
measure to t Denition 14. This is a trademark of incomplete
markets: in some instances no martingale measure will exist while
in others a multiplicity (or even an innite number) of equivalent
martingale measures may exist.
In our case, the latter applies: we can nd multiple equivalent
martingale measures.
15/89
CQF Module 4: Probabilistic Methods for Interest Rates
3.1.2 Equivalent Martingale Measure and No-Arbitrage
As shown in Lecture 3.6, the existence of (at least) one equivalent
martingale measure implies an absence of arbitrage opportunity.
Our zero-coupon bond market is therefore arbitrage-free.
16/89
CQF Module 4: Probabilistic Methods for Interest Rates
3.2. Behaviour of the Short-Term Rate under the
Equivalent Martingale Measure
We consider an equivalent martingale measure Q

for which there

exists a process (t)

satisfying the Novikov condition

E

exp

1
2

T
0

2
s
ds

<

and which denes the Radon Nicodym derivative:

dQ

dP
=
t
= exp

1
2

t
0

2
s
ds

t
0

s
dX
s

, t [0,

T]
17/89
CQF Module 4: Probabilistic Methods for Interest Rates
By Girsanovs theorem, the process X

dened as
X

t
= X
t
+

t
0
(s)ds, t [0, T] (5)
is a standard Brownian Motion on (, F, Q

).
As a result, the dynamics of r (t) under Q

is:
dr (t) = (
t

t

t
) dt +
t
dX

(t), r (0) = r (6)

18/89
CQF Module 4: Probabilistic Methods for Interest Rates
3.3. The Fundamental Asset Pricing Formula for
Zero-Coupon Bonds
Applying the fundamental asset pricing formula (derived in Lecture
3.3) to the zero-coupon bond, we get
B(t, T) = A(t)E

B(T, T)
A(T)

F
t

(7)
19/89
CQF Module 4: Probabilistic Methods for Interest Rates
Since
B(T, T) = 1
and
A(T) = e

T
0
r (s)ds
then formula (7) simplies to
B(t, T) = E
Q

T
t
r (s)ds

F
t

20/89
CQF Module 4: Probabilistic Methods for Interest Rates
Key Fact (The Fundamental Asset Pricing Formula for
Zero-Coupon Bonds)
The price of a zero-coupon bond is given by the expectation:
B(t, T) = E
Q

T
t
r (s)ds

F
t

(8)
21/89
CQF Module 4: Probabilistic Methods for Interest Rates
3.4. Bond Pricing in Practice: Analytical Solutions
Analytical solutions exist for the most popular models such as the
Vacicek model or the CIR model.
In general, the best way to derive them is to apply Feynman-Kac
to the fundamental asset pricing formula to obtain a PDE, and
then solve that PDE
1
.
1
Although in the case of the Vaiscek model, there is an elegant and
insightful analytical solution through probabilities.
22/89
CQF Module 4: Probabilistic Methods for Interest Rates
By Feynman-Kac, the PDE associated with expectation (8) and
the Q

-process r (t) given by equation (6) is:

B
t
(t, r ) + (
t

t

t
)
B
r
(t, r (t)) +
1
2

2
(t, r )

2
B
r
2
(t, r (t))
r (t)B(t, r ) = 0
B(T, r (T)) = 1
We are now in familiar territory as this PDE was derived and then
solved in a few cases earlier in this module.
23/89
CQF Module 4: Probabilistic Methods for Interest Rates
3.5. Bond Pricing in Practice: Numerical Solutions
Given

a short-rate model with dynamics

dr (t) = (
t

t

t
) dt +
t
dX

(t), r (0) = r (9)

under the equivalent martingale measure Q

the fundamental asset pricing formula applied to zero-coupon

bonds (equation (8))
B(t, T) = E
Q

T
t
r (s)ds

F
t

it is not dicult to obtain a numerical solution using Monte Carlo

methods.
24/89
CQF Module 4: Probabilistic Methods for Interest Rates
The Monte-Carlo algorithm would look like this:

r (0) = r

For Simulation j = 1 to N,

For TimeStep t
i
, i = 1 to M

Simulate a N(0, 1) random variable Z

i
;

Simulate the short-term interest rate as

r (t
i
) = r (t
i 1
) +

(t
i 1
, r
t
i 1
)
(t
i 1
, r
t
i 1
)(t
i 1
)

(t
i
t
i 1
)
+(t
i 1
, r
t
i 1
)Z
i

t
i
t
i 1

Compute the value of the integral term

T
t
r (s)ds recursively
as:
Int = Int + r (t
i 1
)(t
i
t
i 1
)

Next TimeStep
25/89
CQF Module 4: Probabilistic Methods for Interest Rates

...

Compute the value B(j ) of the zero-coupon bond under

simulation j :
B(j ) = exp {Int}

Next Simulation

Compute the value of the bond as

B =
1
M
M

j =1
B(j )
Done!
26/89
CQF Module 4: Probabilistic Methods for Interest Rates
4. Dynamics of the Zero-Coupon Bond Price
Later in this session, we will consider the pricing of derivatives
struck on a zero-coupon bond. To price such derivatives, we will
need to understand no only the dynamics of the short-term rate
but also the dynamics of the zero-coupon bond prices.
Our starting point will be somewhat unusual. We will look at the
process
M(t) = Z

(t, T)
t
=
B(t, T)
A(t)

t
where
t
is the Radon Nikodym derivative.
27/89
CQF Module 4: Probabilistic Methods for Interest Rates
The reason for considering this unusual process M(t) is that
Fact
If a process Y(t) is a martingale under Q

and
t
=
Q

P
, then the
process M(t) = Y(t)
t
is a martingale under P.
28/89
CQF Module 4: Probabilistic Methods for Interest Rates
Since Z

(t, T) is a martingale under Q

(by denition of Q

),
M(t) = Z

(t, T)
t
is a martingale under P. Hence, by the
Martingale Representation Theorem, there exists a process such
that M(t) can be represented as
M(t) = M(0) +

t
0
(s)dX(t)
= Z

(0, T) +

t
0
(s)dX(t)
and hence
dM(t) = (t)dX(t)
29/89
CQF Module 4: Probabilistic Methods for Interest Rates
However, we still do not know how Z

(t, T) behaves.
Recalling that M(t) = Z

(t, T)
t
, we can express Z

(t, T) as
Z

(t, T) = M(t)
1
t
and in particular,
dZ

(t, T) = d(M(t)
1
t
) (10)
30/89
CQF Module 4: Probabilistic Methods for Interest Rates
The next step is to apply the Ito product rule to (10).
Since
d
t
=
t

t
dX(t)
=
t

dX

(t)
t
dt

then
d
1
t
=
t

1
t
dX

(t)
=
t

1
t
(dX(t) +
t
dt)
...
31/89
CQF Module 4: Probabilistic Methods for Interest Rates
... and we can nally derive the dynamics of Z

(t, T) under Q

:
dZ

(t, T) = d(M(t)
1
t
)
= M(t)d
1
t
+ dM(t)
1
t
+
t

1
t

t
dt
...
32/89
CQF Module 4: Probabilistic Methods for Interest Rates
And therefore,
dZ

(t, T) = M(t)

1
t

t
dX

(t)

+
t

1
t

dX

(t)
t
dt

+
t

1
t

t
dt
=
1
t
(
t
+ M(t)
t
) dX

(t)
33/89
CQF Module 4: Probabilistic Methods for Interest Rates
Recalling that Z

(t, T) =
B(t,T)
A(t)
, we can nally derive the
dynamics of B(t, T) under Q

:
dB(t, T) = d(Z

(t, T)A(t))
= Z

(t, T)dA(t) + dZ

(t, T)A(t)
= rZ

(t, T)A(t)dt +

1
t

t
+ Z

(t, T)
t

A(t)dX

(t)
...
34/89
CQF Module 4: Probabilistic Methods for Interest Rates
Thus,
dB(t, T) = rB(t, T)dt +

1
t

t
A(t) + B(t, T)
t

dX

(t)
= rB(t, T)dt +

t
Z

(t, T)
t
B(t, T) + B(t, T)
t

dX

(t)
= rB(t, T)dt +

t
M(t)
+
t

B(t, T)dX

(t)
...
35/89
CQF Module 4: Probabilistic Methods for Interest Rates
And hence,
dB(t, T) = rB(t, T)dt + b

(t, T)B(t, T)dX

(t)
= B(t, T)

rdt + b

(t, T)dX

(t)

where
b

(t, T) =

t
M(t)
+
t
36/89
CQF Module 4: Probabilistic Methods for Interest Rates
To conclude,
Key Fact (Dynamics of the Zero-Coupon Bond Price under
Q

)
The dynamics of a zero-coupon bond B(t, T) under the equivalent
martingale measure Q

is given by
dB(t, T)
B(t, T)
= r (t)dt + b

(t, T)dX

(t), B(T, T) = 1 (11)

where b

is the volatility of the zero-coupon bond. Therefore

B(t, T) = B(0, T)A(t) exp

1
2

t
0

(s, T)

2
ds
+

t
0
b

(s, T)dX

(s)

(12)
37/89
CQF Module 4: Probabilistic Methods for Interest Rates
5. The Market Price of Interest Risk
Under an equivalent martingale measure Q

, the dynamics of a
zero-coupon bond is given by
dB(t, T)
B(t, T)
= r (t)dt + b

(t, T)dX

, B(T, T) = 1
As a result, the dynamics of a zero-coupon bond under the actual
probability measure P is
dB(t, T)
B(t, T)
=

r (t) + (t)b

(t, T)

dt + b

(t, T)dX(t), B(T, T) = 1

38/89
CQF Module 4: Probabilistic Methods for Interest Rates
The process (t) through which we dened the relation between
the physical measure P and the equivalent martingale measure Q

has an economic meaning: it is the market price of risk.

The market price of risk represents the compensation paid by the
market to an investor per unit of risk.
In our framework, the risk is represented by the volatility of the
zero-coupon bond, b(t, T), and the total compensation for risk
that an investor should expect is therefore equal to (t)b

(t, T).
39/89
CQF Module 4: Probabilistic Methods for Interest Rates
This observation is consistent with the complete market case we
explored in Lecture 3.3. In the stock option case, the market price
of risk was also the process =
r

.
The key dierence between the complete stock market and the
incomplete bond market is that while in the stock market the
process (and hence the market price of risk) was uniquely
determined, the process is not unique in the bond market.
To formalize,
40/89
CQF Module 4: Probabilistic Methods for Interest Rates
Key Fact
The relationship between the physical measure P and an equivalent
martingale Q

measure is established by the market price of risk

which acts as the change of measure process.
In complete markets the equivalent martingale measure is unique
and so is the market price of risk.
In incomplete markets, we may have several equivalent martingale
measures, each with its own market price of risk.
41/89
CQF Module 4: Probabilistic Methods for Interest Rates
6. Pricing Bond Derivatives
Fixed income markets present a wide diversity of instruments:
bonds, of course, but also forwards, options, caps and oors,
numerous swaps and swaptions, structure notes... and this is
without mentioning the interconnection between xed income
markets and credit market or between xed income products and
ination-linked products.
All this to say that pricing zero-coupon bond is simply the starting
point of any serious attempt to price xed income products. In this
section, we will start to expand our horizons by considering the
pricing of derivatives on zero-coupon bonds.
42/89
CQF Module 4: Probabilistic Methods for Interest Rates
To alleviate the notation, we will drop the superscript and simply
refer to the Q-measure to indicate one of the equivalent martingale
measures. Consequently, we will denote by X
Q
(t) the Q-standard
Brownian motion and express the bond price dynamics as
dB(t, T)
B(t, T)
= r (t)dt + b(t, T)dX
Q
(t), B(T, T) = 1
and
B(t, T) = B(0, T)A(t) exp

1
2

t
0
(b(s, T))
2
ds
+

t
0
b(s, T)dX
Q
(s)

43/89
CQF Module 4: Probabilistic Methods for Interest Rates
6.1. Applying the Fundamental Asset Pricing Formula
The fundamental asset pricing formula tells us that the time t
price of a contingent claim paying some (random) amount Y at
time T is given by
V(t) = A(t)E
Q

Y
A(T)

F
t

In particular, the value of a zero-coupon bond maturing at time T

is given by
B(t, T) = A(t)E
Q

1
A(T)

F
t

44/89
CQF Module 4: Probabilistic Methods for Interest Rates
Now, what would happen if we wanted to price a call option C(t)
on a zero-coupon bond maturing at time U? The call option has
strike K and expiry T < U.
45/89
CQF Module 4: Probabilistic Methods for Interest Rates
Applying the fundamental asset pricing formula, we would get
C(t) = A(t)E
Q

(B(T, U) K)
+
A(T)

F
t

= A(t)

E
Q

B(T, U)1
{B(T,U)>K}
A(T)

F
t

KE
Q

1
{B(T,U)>K}
A(T)

F
t

...
46/89
CQF Module 4: Probabilistic Methods for Interest Rates
C(t) = A(t)

E
Q

A(T)E
Q

1
A(U)

F
T

1
{B(T,U)>K}
A(T)

F
t

KE
Q

1
{B(T,U)>K}
A(T)

F
t

= A(t)

E
Q

1
{B(T,U)>K}
A(U)

F
t

KE
Q

1
{B(T,U)>K}
A(T)

F
t

47/89
CQF Module 4: Probabilistic Methods for Interest Rates
And thats it. We cannot go any further.
To go any further, we would need to know at time t the joint
distribution of B(T, U), A(U) and A(T). This is unlikely, unless
we make very constraining assumptions.
This approach appears to lead to a dead end.
48/89
CQF Module 4: Probabilistic Methods for Interest Rates
One way out of this situation would be to look for a measure P
T
such that the expectation in the fundamental asset pricing formula
would be a sole function of the derivative payo (B(T, U) K)
+
.
This idea implies that rather than having the classic formula
V(t) = A(t)E
Q

(B(T, U) K)
+
A(T)

F
t

we would have the modied formula

V(t) = B(t, T)E
P
T

(B(T, U) K)
+
|F
t

(13)
49/89
CQF Module 4: Probabilistic Methods for Interest Rates
To be in a position to use this modied formula, we must
answer 2 questions:
1. We do not know what P
T
is. In fact, we do not even know if
P
T
exists.
2. Given information up to time t, what would B(T, U) be equal
to?
50/89
CQF Module 4: Probabilistic Methods for Interest Rates
Lets start with the second, and easiest, question. If we are at time
t and we would like to know the price at some future time T of a
bond maturing at time U, we would use the forward price for that
bond.
This leads us to a possible answer to our rst question. To dene
P
T
, we could look for a forward martingale measure, that is, an
equivalent martingale measure dened with respect to forward
prices.
Hence, to use the modied fundamental asset pricing formula,
we need to know a little bit about forwards.
51/89
CQF Module 4: Probabilistic Methods for Interest Rates
6.2. Forward Contracts and Forward Prices
Forward contracts are OTC derivatives securities in which the
long party has the obligation to buy an agreed upon quantity of an
underlying asset (securities, commodities or others) at an agreed
upon time and at an agreed upon price called the forward price.
Forward contracts are symmetrical contracts. Therefore, the
obligations of the short party mirror those of the long party. The
contract is settled at maturity and typically no cash ow is
exchanged in the meantime. As they are OTC derivatives, forward
contracts are subject to counterparty risk.
52/89
CQF Module 4: Probabilistic Methods for Interest Rates
How do we price a forward contract?
Lets say that we want to enter into a (long) forward contract on a
nancial instrument (stock, bond, currency...) whose value at time
t is Y(t). The forward matures at time T.
Based on our denition of forward contracts, the payo G(T, Y
T
)
of the contract is
G(T, Y
T
) = Y(T) F
Y
(t, T)
where F
Y
(t, T) is the forward price of Y determined at time t for
delivery at time T
53/89
CQF Module 4: Probabilistic Methods for Interest Rates
Plugging this into the fundamental asset pricing formula, we see
that the time t T value V() of a forward contract entered
into at time t is equal to
V() = A()E
Q

Y(T) F
Y
(t, T)
A(T)

This formula can be simplied by noting that F

Y
(t, T) is a
constant:
V() = A()

E
Q

Y(T)
A(T)

F
Y
(t, T)E
Q

1
A(T)

and now we know the value of a forward contract for any time
t T.
54/89
CQF Module 4: Probabilistic Methods for Interest Rates
How do we know the forward price?
The forward price F
Y
(t, T) was originally set at time t so that the
value of the forward contract at time t is 0. Hence,
V(t) = A(t)

E
Q

Y(T)
A(T)

F
t

F
Y
(t, T)E
Q

1
A(T)

F
t

= 0
Rearranging,
F
Y
(t, T) =
E
Q

Y(T)
A(T)

F
t

E
Q

1
A(T)

F
t

=:
W(t)
B(t, T)
(14)
55/89
CQF Module 4: Probabilistic Methods for Interest Rates
where
W(t) = A(t)E
Q

Y(T)
A(T)

F
t

(15)
that is W(t) is the value of a claim paying Y(T) at time T.
56/89
CQF Module 4: Probabilistic Methods for Interest Rates
6.3. The Forward Martingale Measure
We will dene the forward martingale measure, or simply forward
measure, via the Radon-Nikodym derivative
t
dened as

t
=
dP
T
dQ
= E
Q

A(0)B(T, T)
A(T)B(0, T)

F
t

57/89
CQF Module 4: Probabilistic Methods for Interest Rates
Thus,

t
=
1
B(0, T)
E
Q

1
A(T)

F
t

=
1
A(t)B(0, T)

A(t)E
Q

1
A(T)

F
t

58/89
CQF Module 4: Probabilistic Methods for Interest Rates
And hence,

t
=
A
1
(t)B(t, T)
B(0, T)
59/89
CQF Module 4: Probabilistic Methods for Interest Rates
The strange formula

t
=
A
1
(t)B(t, T)
B(0, T)
can be interpreted as the ratio of two trades.
60/89
CQF Module 4: Probabilistic Methods for Interest Rates
The denominator represents

The purchase at time 0 of a bond maturing at time T. This

trade costs B(0, T) at time 0 and pays 1 at time T;
The numerator is

the purchase at time t of a zero-coupon bond maturing at

time T, for a price B(t, T). This bond pays 1 at time T;

the time 0 value of this trade is therefore equal to

1
A(t)
B(t, T);
61/89
CQF Module 4: Probabilistic Methods for Interest Rates
Since both trade will end up paying 1 at time T, then to prevent
arbitrage, they should have on average the same value at time 0.
Restating this observation in more mathematical terms:
To prevent arbitrage,
t
=
A
1
(t)B(t,T)
B(0,T)
must be a martingale.
62/89
CQF Module 4: Probabilistic Methods for Interest Rates
We must emphasize the on average. Indeed, the short-term
interest rate r (t) is stochastic and therefore, both

A(t), and

B(t, T)
are not know with certainty at time 0.
63/89
CQF Module 4: Probabilistic Methods for Interest Rates
Lets check that this is indeed a measure by verifying that the
Radon-Nikodym derivative
t
is indeed an exponential martingale.
We know that the time t value of a zero-coupon bond is given by
formula (12)
B(t, T) = B(0, T)A(t) exp

1
2

t
0
(b(s, T))
2
ds +

t
0
b(s, T)dX
Q
s

and
A(t) = e

t
0
r (s)ds
64/89
CQF Module 4: Probabilistic Methods for Interest Rates
Therefore,

t
=
A
1
(t)B(t, T)
B(0, T)
= exp

1
2

t
0
(b(s, T))
2
ds +

t
0
b(s, T)dX
Q
s

As long as the bond volatility b(s, T) remains nite,

t
is indeed
an exponential martingale.
As a result, the forward measure P
T
is well-dened.
65/89
CQF Module 4: Probabilistic Methods for Interest Rates
Moreover, by Girsanovs theorem, the process X
T
dened as
X
T
t
= X
Q
t

t
0
b(s, T)ds, t [0, T]
is a standard Brownian Motion under the forward measure P
T
.
X
T
is called the forward Brownian motion.
66/89
CQF Module 4: Probabilistic Methods for Interest Rates
6.4. Pricing a Derivative Under the Forward Measure
We now have a candidate measure P
T
. But before we can use the
forward asset pricing formula
V(t) = B(t, T)E
P
T
[Y|F
t
]
we need to make sure that it will give the same result as the
classic fundamental asset pricing formula
V(t) = A(t)E
Q

Y
A(T)

F
t

which we know to be correct.

67/89
CQF Module 4: Probabilistic Methods for Interest Rates
We will start with what we know: the fundamental asset pricing
formula
V(t) = A(t)E
Q

Y
A(T)

F
t

68/89
CQF Module 4: Probabilistic Methods for Interest Rates
Noting that

T
=
A
1
(T)B(T, T)
B(0, T)
=
A
1
(T)
B(0, T)
(16)
the asset pricing formula can be rewritten as
V(t) = A(t)E
Q

B(0, T)A(T)
Y
T
A(T)

F
t

= A(t)B(0, T)E
Q

Y
T

F
t

69/89
CQF Module 4: Probabilistic Methods for Interest Rates
It turns out that we can write
2
E
Q
[Y
T
|F
t
] = E
T
[Y|F
t
] E
Q
[
T
|F
t
]
Therefore,
V(t) = A(t)B(0, T)E
T
[Y|F
t
] E
Q
[
T
|F
t
]
2
This is due to a rather advanced stochastic analysis result which is an
extension of Bayes formula dealing with change of measure and conditional
expectations.
70/89
CQF Module 4: Probabilistic Methods for Interest Rates
Now,
T
is a martingale under Q. Hence
E
Q
[
T
|F
t
] = (t)
=
A
1
(t)B(t, T)
B(0, T)
...
71/89
CQF Module 4: Probabilistic Methods for Interest Rates
... and we can nally conclude that
V(t) = B(t, T)E
P
T
[Y|F
t
]
72/89
CQF Module 4: Probabilistic Methods for Interest Rates
Key Fact
We want to price a European derivative expiring at time T on a
zero-coupon bond maturing at time U, with T < U. The payo of
the derivative is G(B(T, U)).
Under the forward (martingale) measure P
T
, the value of this
derivative is given by the forward asset pricing formula
V(t) = B(t, T)E
P
T
[G(B(T, U))|F
t
] (17)
The forward (martingale) measure P
T
is dened in terms of the
equivalent martingale measure Q via the Radon-Nikodym derivative

t
=
dP
T
dQ
=
A
1
(t)B(t, T)
B(0, T)
(18)
73/89
CQF Module 4: Probabilistic Methods for Interest Rates
6.5 Pricing a Call on a Zero-Coupon Bond
In the case of a call on a zero-coupon bond, a Black-Scholes-type
formula exists.
To see this, we will need to express the call payo at time T, not
in terms of the zero-coupon bond, but in terms of a forward on the
zero-coupon bond
3
as
(B(T, U) K)
+
= (F
B
(T, T, U) K)
+
(19)
where F
B
(t, T, U) is the forward price at time t for settlement at
time T of a zero-coupon bond maturing at time U > T. Note that
F(T, T, U) is simply the instantaneous forward price at time T,
which is equal to the spot price B(T, U).
3
After all, this should be more consistent with the forward measure-based
pricing framework we have developed.
74/89
CQF Module 4: Probabilistic Methods for Interest Rates
Applying formula (14), we deduce that the zero-coupon forward
price F
B
(t, T, U) is given by
F
B
(t, T, U) =
E
Q

B(T,U)
A(T)

F
t

E
Q

1
A(T)

F
t

=
A(t)E
Q

B(T,U)
A(T)

F
t

A(t)E
Q

1
A(T)

F
t

=
B(t, U)
B(t, T)
(20)
75/89
CQF Module 4: Probabilistic Methods for Interest Rates
As a result, the Q-dynamics of the forward price is given by
F
B
(t, T, U) = F
B
(0, T, U) exp

1
2

t
0
(b(s, U))
2
(b(s, T))
2
ds
+

t
0
(b(s, U) b(s, T)) dX
Q
(s)

= F
B
(0, T, U) exp

t
0
b(s, T) (b(s, U) b(s, T)) ds

exp

1
2

t
0
(b(s, U) b(s, T))
2
ds
+

t
0
(b(s, U) b(s, T)) dX
Q
(s)

(21)
76/89
CQF Module 4: Probabilistic Methods for Interest Rates
and hence
dF
B
(t, T, U)
F
B
(t, T, U)
= b(s, T) (b(t, U) b(t, T)) dt
+(b(t, U) b(t, T)) dX
Q
(t) (22)
77/89
CQF Module 4: Probabilistic Methods for Interest Rates
Deriving the dynamics of F
B
(t, T, U) under the Q-measure is a
promising start. But it is not enough.
As we are going to price the call option using the forward asset
pricing formula, we need to know the dynamics of the forward price
F
B
(t, T, U) under the forward measure P
T
.
78/89
CQF Module 4: Probabilistic Methods for Interest Rates
Recalling that
X
T
t
= X
Q
t

t
0
b(s, T)ds, t [0, T]
is a standard Brownian Motion under the forward measure P
T
, we
immediately get
dF
B
(t, T, U)
F
B
(t, T, U)
= (b(s, U) b(s, T)) dX
T
(s) (23)
and therefore
F
B
(t, T, U) = F
B
(0, T, U) exp

T
0
(b(s, U) b(s, T)) dX
T
(s)

1
2

T
0
(b(s, U) b(s, T))
2
ds

(24)
which implies that the forward price is a martingale under the
forward measure.
79/89
CQF Module 4: Probabilistic Methods for Interest Rates
We now have all we need to solve the Call option pricing problem
using the forward asset pricing formula
C(t) = B(t, T)E
P
T

(F
B
(T, T, U) K)
+
|F
t

(25)
80/89
CQF Module 4: Probabilistic Methods for Interest Rates
Key Fact
The time t price of a European Call expiring at time T and with
strike K, written on a zero-coupon bond maturing at time U > T
is given by the following Black-Scholes type of formula:
C(t) = B(t, U)N [d
1
(B(t, U), t, T)]
KB(t, T)N [d
2
(B(t, U), t, T)] (26)
where
d
1
(b, t, T) =
ln

b
K

ln B(t, T) +
1
2

U
(t, T)

U
(t, T)
d
2
(b, t, T) = d
1

U
(t, T)

2
U
(t, T) =

T
t
(b(s, U) b(s, T))
2
ds
81/89
CQF Module 4: Probabilistic Methods for Interest Rates
7. What Should We Make of this Entire Approach?
82/89
CQF Module 4: Probabilistic Methods for Interest Rates
7.1. Whats Unsatisfactory?
First, when we look at specic interest rate models such as the
CIR or the Vasicek models, the probabilistic approach does not
bring us any new insights. Whats worse, we have to use
Feynman-Kac to transform the problem into a PDE problem if we
want to solve it analytically...
Second, if we want the bond price to have a lognormal-type
behaviour in a short rate model, the bond volatility function
b(t, T) will unfortunately be a fudge function.
This unpleasantness will however motivate us to turn our models
around and specify a bond dynamics rst, and then deduce a
dynamics of interest rates. This approach forms the base of
forward rate models such as the HJM class of models.
83/89
CQF Module 4: Probabilistic Methods for Interest Rates
7.2. What is Satisfactory?
The forward measure.
The existence of the forward measure and the critical role played
by forwards in the pricing of bond derivatives also provides a
powerful motivation for looking at the term structure of forward
(as opposed to spot) rates (see the HJM class of models).
Moreover, the interpretation of the Radon-Nikodym derivative

t
=
A
1
(t)B(t,T)
B(0,T)
as an intertemporal no-arbitrage condition is
particularly intuitive and elegant.
84/89
CQF Module 4: Probabilistic Methods for Interest Rates
Another aspect is worth noting, as long as the bond price follows a
geometric dynamics, then irrespective of the specic interest rate
model we chose, the value of a bond derivative will always be of
the same form.
This raises an intriguing question: if we assume a geometric
dynamics for the bond price, how many interest rate models do we
have access to? Many, or few?
This question can be reformulated in a slightly more mathematical
language as: what is the most general interest rate model we can
nd such that the bond price follows a geometric dynamics?
85/89
CQF Module 4: Probabilistic Methods for Interest Rates
7.3. Where to Next? Forward Rate Models
The answer to this question, and next chapter in the development
of interest rate models is the derivation of models of the forward
rate dynamics. This critical step was achieved by Heath, Jarrow
and Morton (1992) and then further developed by Brace, Gatarek
and Musiela (1997).
86/89
CQF Module 4: Probabilistic Methods for Interest Rates
The key attraction of forward rate models is

they start from a (nice) geometric dynamics for the

zero-coupon bond price and then deduce the behaviour of the
term structure of forward rates;

they are a meta-model which encompasses all existing

interest rate models;

as such, you can use them to price or manage the risk of

anything, from vanilla derivatives (which typically only require
a short-term interest rate model) to complex xed income
portfolios (which are heavily dependent on an accurate
modelling of the term structure).
87/89
CQF Module 4: Probabilistic Methods for Interest Rates
The key problems related to forward rate models:

non-Markov. All the mathematics we have are based on

Markov models. Hence, we need to choose our parameters
carefully and make assumptions to ensure that the forward
rate models we work with are indeed Markov;

mathematically sophisticated, sometimes too sophisticated for

the applications at hand (such as pricing vanilla derivatives);

meta-model: no clear indication of which form to use and

when to use it.
88/89
CQF Module 4: Probabilistic Methods for Interest Rates
In this lecture, we have seen...

the pricing of interest rate products in a probabilistic setting;

the equivalent martingale measures;

the fundamental asset pricing formula for bonds;

application for popular interest rates models;

the dynamics of bond prices;

the forward measure;

the fundamental asset pricing formula for derivatives on

bonds;

right and wrongs of this approach to short-term interest rate

models;
89/89