MScFE 622 Continuous-time Stochastic Processes - Module 7: Bibliography

Compiled Notes
Module 7
MScFE 622
Continuous-time Stochastic
Processes

Revised: 07/07/2020
MScFE 622 Continuous-time Stochastic Processes − Notes Module 7

Module 7: Interest Rate Modelling

Contents
Unit 1: Interest Rate Assets 3

Unit 2: Short Rate Models 5

Unit 3: Bond Pricing 8

Problem Set 10

2019 - WorldQuant University − All rights reserved. 1

MScFE 622 Continuous-time Stochastic Processes − Summary Module 7

Summary
This final module of the course presents commonly used interest rate models. Module 7 begins by
describing stochastic interest rate assets, and then explains the Ho-Lee, Vasicek, Cox-Ingersoll-Ross,
and Hull-White short rate models. The module concludes by presenting a derivation of the bond
pricing equation as well as some examples of bond prices using the short rate models previously
introduced.

2019 - WorldQuant University − All rights reserved. 2

MScFE 622 Continuous-time Stochastic Processes − Notes (1) Module 7: Unit 1

Unit 1: Interest Rate Assets

• In the last six modules we have assumed that interest rates are constant. This assumption is
not too harmful when dealing with equity derivatives since the impact of changes in interest
rates is minimal to equity derivatives as compared to changes in the underlying asset.
• This module is concerned with the pricing of interest rate related assets, so it is necessary to
consider the more realistic scenario of stochastic interest rates.
• There are various frameworks to model interest rates. In this course we are going to focus on
models for the so-called short rate. In these short rate models, the bank account B is assumed
to satisfy the following equation:
dBt = rt Bt dt, B0 = 1,
where rr is the instantaneous short rate, which is now assumed to be a stochastic process.
Solving this equation gives Rt
Bt = e 0 rs ds
for t ≥ 0, and Rt
ru du
Bt = Bs e s

for 0 ≤ s ≤ t. We will therefore spend time constructing models for the short rate r.
• We will fix a filtered probability space (Ω, F, F, P) and assume that all stochastic processes are
defined on this space and adapted to F.
• For 0 ≤ t ≤ T we define the stochastic discount factor between t and T as
Bt RT
D(t, T ) := = e− t rs ds .
BT
This represents the present value at time t of one unit of currency payable at time T . We call
T the maturity time and we have D(T, T ) = 1. This is again stochastic since the interest rate
is also stochastic.
• The first non-trivial interest rate asset is the zero-coupon bond. A zero-coupon bond is an asset
that pays one unit of currency at a future time T > 0, called the maturity. The time t price of
a zero coupon bond with maturity time T will be denoted by P (t, T ). Note that by definition
we have P (T, T ) = 1.
• Do not confuse D(t, T ) and P (t, T )! The discount factor, D(t, T ), is a random unknown quantity
at time t. It is calculated from the “true” but unknown evolution of rs for t ≤ s ≤ T . On the
other hand, the zero-coupon bond price, P (t, T ), is an asset price that is known at time t. We
will see later that it is calculated by averaging D(t, T ) over different paths or evolutions of r
between t and T . That is,
P (t, T ) = E∗ [D(t, T )|Ft ]
for some pricing measure P∗ .
• For a zero-coupon bond with price P (t, T ) we define the continuously compounded annual
(NACC) yield R(t, T ) as the constant NACC interest rate such that P (t, T ) is equal to the
discount value of one unit at time T , discounted using the constant rate R(t, T ). In symbols:
− ln P (t, T )
e−R(t,T )(T −t) = P (t, T ) =⇒ R(t, T ) = .
T −t

2019 - WorldQuant University − All rights reserved. 3

MScFE 622 Continuous-time Stochastic Processes − Notes (1) Module 7: Unit 1

• Similarly, we define L(t, T ) to be the corresponding simple yield:

1 1 − P (t, T )
= P (t, T ) =⇒ L(t, T ) = .
(1 + L(t, T )(T − t)) (T − t)P (t, T )

• For fixed t, the function T 7→ L(t, T ) is called the yield curve.

• In general, a coupon-bearing bond that pays coupons of C1 , . . . , Cn at times t ≤ T1 < . . . < Tn

and a nominal of N at maturity time Tn has a price of
n
X
Ck P (t, Tk ) + N P (t, Tn )
k=1

at time t. This is a result of linearity of expectation. So we can use P (t, T ) to discount

payments.

2019 - WorldQuant University − All rights reserved. 4

MScFE 622 Continuous-time Stochastic Processes − Notes (2) Module 7: Unit 2

Unit 2: Short Rate Models

• We now discuss a few widely known short rate models. For all the models, the dynamics of r
are presented in a risk neutral/pricing measure.

Ho-Lee Model
• In the Ho-Lee Model the risk neutral dynamics of r are
drt = θ(t) dt + σ dWt
where W is a Brownian motion, θ(t) is deterministic and σ is a constant.
• We can easily solve the SDE to get
Z t
rt = r0 + θ(s) ds + σWt .
0

Hence rt is normally distributed with

Z t
E (rt ) = r0 + θ(s) ds and Var (rt ) = σ 2 t.
0

• The fact that rt is normally distributed implies that there is a positive probability of rt being
negative, no matter what the parameters are. This is not a desirable property of the model if
we assume that interest rates are positive.

Vasicek Model
• In this model the short rate has the following risk neutral dynamics:
drt = (θ − αrt ) dt + σ dWt ,
where θ, α and σ are constants and W is a Brownian motion. Another common formulation of
this model (and the one used in the lecture videos) is
drt = α (θ − rt ) dt + σ dWt .
Students are expected to be familiar with both expressions, that is why both formulations are
introduced. However, we will always refer to the first formulation of the model in the rest of
the notes and exercises.
• To solve for rt we apply Ito’s formula to f (t, rt ) = eαt rt to get
df (t, rt ) = αeαt rt dt + eαt drt = θeαt dt + σeαt dWt ,
which implies that
Z t Z t Z t
αt αs αs θ αt
σeαs dWs .


e rt = r0 + θe ds + σe dWs = r0 + e −1 +
0 0 α 0

Therefore Z t
−αt θ
1 − e−αt + σeα(s−t) dWs .


rt = r0 e +
α 0

2019 - WorldQuant University − All rights reserved. 5

MScFE 622 Continuous-time Stochastic Processes − Notes (2) Module 7: Unit 2

• We can see again that rt is normally distributed with parameters:

Z t
−αt θ −αt
σ 2 e2α(s−t) ds.


E (rt ) = r0 e + 1−e and Var (rt ) =
α 0

Again, the rate can go negative.

• In this model, the rate r is mean reverting with a long run mean of θ/α. This is a desirable
property since it is in line with empirical observations.

Cox-Ingersoll-Ross Model
• In the Cox-Ingersoll-Ross (CIR) model, the risk neutral dynamics of r are
√
drt = (θ − αrt ) dt + σ rt dWt ,

where α, θ and σ are constants. This process is often called the square root process. Again, as
with the Vasicek model, an alternative formulation is
√
drt = α (θ − rt ) dt + σ rt dWt .

However, the first one will be the default.

• This model can be considered an improvement to the Vasicek model and it avoids negative
rates. If 2θ ≥ σ 2 , then rt never touches zero.

• The distribution of rt is given by the non-central chi-squared distribution. We will not show
the derivation here, but we note that
θ σ 2 −αt
θσ 2
2
E (rt ) = r0 e−αt + 1 − e−αt and Var (rt ) = r0 − e−2αt + 2 1 − e−αt .


e
α α 2α

Hull-White Model
• One of the major drawbacks of the last two models presented is that, due to only having three
parameters, they fail to reproduce observed bond prices (i.e., they fail to ‘fit the market’).

• The Hull˙White model is an extension to the Vasicek model that allows θ to be a deterministic
function of time, rather than a constant. This flexibility ensures that the model can fit the
market-observed bond prices.

• The dynamics of r in the Hull-White model are

drt = (θ(t) − αrt ) dt + σ dWt ,

where θ(t) is a deterministic function of time and α and σ are constants. In some specifications
of the model, α is also a deterministic function of t. There is also an alternative formulation as

drt = α (θ(t) − rt ) dt + σ dWt ,

which is sometimes used in the lecture videos but not in the notes and quizzes. You are expected
to be familiar with both.

2019 - WorldQuant University − All rights reserved. 6

MScFE 622 Continuous-time Stochastic Processes − Notes (2) Module 7: Unit 2

• The Hull-White model is similar to the Vasicek model and the derivation of the distributions
are almost identical. To solve for rt , first we apply Ito’s formula to f (t, rt ) = eαt rt to get

df (t, rt ) = αeαt rt dr + eαt drt = θ(t)eαt dt + σeαt dWt ,

which implies that Z t Z t

αt αs
e rt = r0 + θ(s)e ds + σeαs dWs .
0 0
Therefore Z t Z t
−αt α(s−t)
rt = r0 e + θ(s)e ds + σeα(s−t) dWs .
0 0

• We can see again that rt is normally distributed with parameters:

Z t Z t
−αt α(s−t)
E (rt ) = r0 e + θ(s)e ds and Var (rt ) = σ 2 e2α(s−t) ds.
0 0

Again, the rate can go negative.

• In this model, the rate r is mean reverting with a long run mean of θ(t)/α that keeps changing.
This is a desirable property since it is in line with empirical observations.

2019 - WorldQuant University − All rights reserved. 7

MScFE 622 Continuous-time Stochastic Processes − Notes (3) Module 7: Unit 3

Unit 3: Bond Pricing

• We now consider the pricing of bonds under each of the four models presented above.

• First, if H is a derivative that expires at time T , then its price at time t < T is given by
E∗ (D(t, T )H|Ft ), where P∗ is an ELMM for the discounted bond prices. This implies that the
bond prices must satisfy the equation

P (t, T ) = E∗ (D(t, T )|Ft )

for t < T since P (T, T ) = 1. Indeed, since P (t, T )/Bt is a martingale and P (T, T ) = 1, we have
   
P (t, T ) ∗ P (T, T ) ∗ 1
=E |Ft = E |Ft .
Bt BT BT
Multiplying both sides by Bt and noting that Bt is Ft -measurable gives the result.

• Now suppose that the risk-neutral dynamics of the short rate r as follows:

drt = µ(t, rt ) dt + σ(t, rt ) dWt .

Also suppose that the zero-coupon bond prices are of the form

P (t, T ) = F T (t, rt )

for some sufficiently smooth function F T . Then through some no-arbitrage considerations, we
can show that F T satisfies the following PDE:

∂F T ∂F T 1 ∂ 2F T
+ µ(t, r) + σ 2 (t, r) 2
− rF T = 0, F T (T, r) = 1.
∂t ∂r 2 ∂r
This PDE is called the fundamental term structure PDE.

• The short rate models we considered above have the special property that the bond prices are
of the form
P (t, T ) = F T (t, rt ) = eA(t,T )−B(t,T )rt
for some sufficiently smooth functions A and B. Such models are called affine term-structure
models.

• Let us now substitute the new expression for F T into the PDE to obtain the corresponding
equations satisfied by A and B. We get
 
∂A 1 2 2 ∂B
− µ(t, r)B + σ (t, r)B − 1 + r = 0.
∂t 2 ∂t

Now suppose that µ and σ 2 can be expressed as

µ(t, r) = a(t)r + b(t) and σ 2 (t, r) = c(t)r + d(t).

Then the equation becomes

 
∂A 1 ∂B 1
− b(t)B + d(t)B 2 = 1+ 2
+ a(t)B − d(t)B r.
∂t 2 ∂t 2

2019 - WorldQuant University − All rights reserved. 8

MScFE 622 Continuous-time Stochastic Processes − Notes (3) Module 7: Unit 3

Since the LHS is independent of r, while the RHS is dependent on r, both sides must be equal
to 0. This gives the following system of differential equations for A and B:

 ∂A = b(t)B − 1 d(t)B 2

A(T, T ) = 0
2
∂t
∂B

 = −a(t)B + 12 c(t)B 2 − 1 B(T, T ) = 0.
∂t

• Let us now solve the differential equations to find the bond price for some of the four models.

• First we start with the Ho-Lee model. Note that for this model we have

a(t) = c(t) = 0, and b(t) = θ(t), d(t) = σ 2 .

Thus the system of equations is


 ∂A = θ(t)B − 1 σ 2 B 2

A(T, T ) = 0
2
∂t
∂B

 = −1 B(T, T ) = 0.
∂t
We can easily solve the second equation for B to get B(t, T ) = T − t. Substituting this to the
first equation and integrating we get
Z T
1
A(t, T ) = − θ(s)(T − s) ds + σ 2 (T − t)3 .
t 6

• Next is the Vasicek model. We have

a(t) = −α, b(t) = θ, c(t) = 0 and d(t) = σ 2

hence the differential equations are


 ∂A = θB − 1 σ 2 B 2

A(T, T ) = 0
2
∂t
∂B

 = αB − 1 B(T, T ) = 0.
∂t
The second equation is linear with constant coefficients, hence it can solved by multiplying
both sides by an integrating factor. The solutions are:
1 2


1 (B(t, T ) − (T − t)) αθ − σ σ 2 B 2 (t, T )
1 − e−α(T −t) , and A(t, T ) = 2


B(t, T ) = − .
α α2 4α

• Finally we look at the CIR model. We have


 ∂A = θB

A(T, T ) = 0
∂t
∂B

 = αB + 12 σ 2 B 2 − 1 B(T, T ) = 0.
∂t
The solution to these equations is lengthy and we will not show it here. Refer to Brigo and
Mercurio’s Interest Rate Modelling for a discussion.

2019 - WorldQuant University − All rights reserved. 9

MScFE 622 Continuous-time Stochastic Processes − Problem Set Module 7

Problem Set

Problem 1. Assume that r evolves according to the Vasicek model with

r0 = 0.1, θ = 0.15, α = 0.2, and σ = 0.5.

What is E(r5 )?

Solution: From the lecture notes, we know that:

θ
E (rt ) = r0 e−αt + 1 − e−αt


α
Thus, to compute E (rt ) we have that,

θ 0.15
E (rt ) = r0 e−αt + 1 − e−αt = 0.1e−0.2∗5 + 1 − e−0.2∗5 = 0.511



α 0.2

Problem 2. Assume that r evolves according to the Vasicek model with

r0 = 0.1, θ = 0.15, α = 0.2, and σ = 0.5.

If the zero-coupon bond price P (1, 3) under this model is eA(1,3)−B(1,3)r1 , then B(1, 3) is equal
to

Solution: Next is the Vasicek model. We have

a(t) = −α, b(t) = θ, c(t) = 0 and d(t) = σ 2

hence the differential equations are


 ∂A = θB − 1 σ 2 B 2

A(T, T ) = 0
2
∂t
∂B

 = αB − 1 B(T, T ) = 0.
∂t
The second equation is linear with constant coefficients, hence it can solved by multiplying
both sides by an integrating factor. The solutions are:
(B(t, T ) − (T − t)) αθ − 21 σ 2


1 −α(T −t)

σ 2 B 2 (t, T )
B(t, T ) = 1−e , and A(t, T ) = − .
α α2 4α

In our case, we only care about B, and then,

1
1 − e−0.2(3−1) = 1.6484


B(1, 3) =
0.2

2019 - WorldQuant University − All rights reserved. 10

MScFE 622 Continuous-time Stochastic Processes − Problem Set Module 7

Rt
Problem 3. Assuming that 0
rs ds ∼ N (0.01t, 0.2t), then compute the value of E(D(0, 1)).

Solution: We know from the lecture notes that,

Bt RT
D(t, T ) := = e− t rs ds .
BT
And, as the exponent is a normal distribution, we can apply the following,

2 /2
E (D(0, 1)) = e−µ+σ = e−0.01+(0.2)/2 = e0.09

Problem 4. Assume that r evolves according to the Ho-Lee model with

r0 = 0.06, θ(t) = 0.003, and σ = 0.25.

If the zero-coupon bond price P (1, 3) under this model is eA(1,3)−B(1,3)r1 , then compute the value
of B(1, 3).

Solution: From the lecture notes we know that the Ho-Lee model we have

a(t) = c(t) = 0, and b(t) = θ(t), d(t) = σ 2 .

Thus the system of equations is


 ∂A = θ(t)B − 1 σ 2 B 2

A(T, T ) = 0
2
∂t
∂B

 = −1 B(T, T ) = 0.
∂t
We can easily solve the second equation for B to get B(t, T ) = T − t. Substituting this to the
first equation and integrating we also can get
Z T
1
A(t, T ) = − θ(s)(T − s) ds + σ 2 (T − t)3 .
t 6

In our case the solution is straightforward, B(1, 3) = 3 − 1 = 2.

Problem 5. Assume that r evolves according to the Vasicek model with

r0 = 0.06, θ = 0.09, α = 0.11, and σ = 0.25.

If the zero-coupon bond price P (2, 3) under this model is eA(2,3)−B(2,3)r2 , then B(2, 3) is equal
to

2019 - WorldQuant University − All rights reserved. 11

MScFE 622 Continuous-time Stochastic Processes − Problem Set Module 7

Solution: Next is the Vasicek model. We have

a(t) = −α, b(t) = θ, c(t) = 0 and d(t) = σ 2

hence the differential equations are


 ∂A = θB − 1 σ 2 B 2

A(T, T ) = 0
2
∂t
∂B

 = αB − 1 B(T, T ) = 0.
∂t
The second equation is linear with constant coefficients, hence it can solved by multiplying
both sides by an integrating factor. The solutions are:
1 2


1 (B(t, T ) − (T − t)) αθ − σ σ 2 B 2 (t, T )
1 − e−α(T −t) , and A(t, T ) = 2


B(t, T ) = − .
α α2 4α

In our case, we only care about B, and then,

1
1 − e−0.11(3−2) = 0.94696


B(2, 3) =
0.11

2019 - WorldQuant University − All rights reserved. 12