Interest Rate Theory

Aktuarieföreningen, Mars 2004

Tomas Björk

1
 A
Problems

1. How do you build a model to explain the

yield curve.

2. How do you build a model in order to price

derivatives?

3. How do you build a model to help you to

hedge your positions?

4. What is a good (parsimonious?) way to

describe the (partly observed) existing yield
curve?

2
 Explaining the yield curve

• Mixture economics–finance.

• Microeconomic equilibrium models. Macro-

economic models.

• Fundamental model in terms of a small

number of state variables with clear eco-
nomic interpretation.

• Model will typically not fit the observed

yield curve.

• Calculations done under the physical mea-

sure P.

“The more you fit, the less you explain.”

3
 Pricing derivatives and hedging

• Take the yield curve (cap curve, swap curve)

as unexplained observed market input.

• Try to obtain exact fit to market data.

• Compute prices of derivatives in terms of

market input data.

• Interpolation problem.

• Large number of state variables.

• Calculations done under the martingale

measure Q.

“The more you fit, the less you explain.”

4
 Describing the yield curve

• Ask Peter Alaton for help.

5
 Definitions

Bonds:
T -bond = zero coupon bond, paying 1$ at the
date of maturity T .

p(t, T ) = price, at t, of a T -bond.

p(T, T ) = 1.

Main Problem

• Investigate the term structure, i.e. how

prices of bonds with different dates of ma-
turity are related to each other.

• Compute arbitrage free prices of interest

rate derivatives (bond options, swaps, caps,
floors etc.)

6
 Interest Rates

Infinitesimal rates:
The forward rate at t with maturity T is de-
fined by
∂ ln p(t, T )
f (t, T ) = −
∂T
The short rate at t is defined by

r(t) = f (t, t)
The bank account is defined by

dB(t) = r(t)B(t)dt

Market rates:
The LIBOR forward rate at t for the time
period (S, T ] is defined by
1 p(t, T ) − p(t, S)
L(t, S, T ) = −
T −S p(t, T )

7
 Financial Markets

Price Process:

S(t) = [S0 (t), ..., SN (t)]

Si(t) = price of asset i at time t.

Definition: An arbitrage is a financial free

lunch.

Main Question: When is the market free of

arbitrage?

8
 Absence of Arbitrage

The market is arbitrage free

iff

There exists a probability measure Q ∼ P such

that all discounted price processes are
Q-martingales.

i.e.

S(t)
Z(t) = = [1, Z1(t), ..., ZN (t)]
S0(t)

is a Q martingale.

i.e.

E Q [ Zi(s)| Ft] = Zi(t), t≤s

9
 Basic Facts

• The market is arbitrage free ⇔ There exists

a martingale measure Q, equivalent to the
physical measure P .

• The market is complete ⇔ Q is unique.

• Every claim X must be priced by the for-

mula
 RT 
Π [t; X ] = E Q e− t r(s)ds
× X Ft
for some choice of Q.

• In a non-complete market, different choices

of Q will produce different prices for X.

• Bond prices are given by

 RT 
p(t, T ) = E Q e− t r(s)ds
Ft

10
 From P to Q

• Roughly speaking, P and Q are related by

the market price of risk λ.

• λ measures the aggregate risk aversion on

the market

11
 Arbitrage Theory for Interest
Rates

1. Model an arbitrage free bond market.

2. Take some vanilla products as input.

3. Price more exotic products in terms of the

model and the vanilla input.

12
 Model Classes

1. Models for infinitesimal rates.

• Low dimensional models. (Short rate

models etc)

• High dimensional models. (HJM etc)

2. Market Models.

• LIBOR market models.

• Swaption market models.

3. Other types.

• Markov functional models. (Hunt–Kennedy–

Pelsser).

• Hughston–Flesaker.

13
 Martingale Modelling

• Bond prices are NOT determined

by the P dynamics of r.

• All prices are determined by the Q-

dynamics of r.

• Model dr directly under Q!

Problem: Parameter estimation!

14
 Martingale pricing

Q-dynamics:
dr = µ(t, r)dt + σ(t, r)dW
 RT 
Π [t; X ] = E Q e− t r(s)ds
× X Ft
 RT 
p(t, T ) = E Q e− t r(s)ds
× 1 Ft

The Case X = Φ(r(T )):

The price is given by

Π [t; X ] = F (t, r(t))
(
Ft + µFr + 1
2 σ 2F − rF = 0,
rr
F (T, r) = Φ(r(T )).

(Term Structure Equation)

15
1. Vasiček
dr = (b − ar) dt + σdV,

2. Cox-Ingersoll-Ross
√
dr = (b − ar ) dt + σ rdV,

3. Dothan
dr = ardt + σrdV,

4. Black-Derman-Toy
dr = a(t)rdt + σ(t)rdV,

5. Ho-Lee
dr = a(t)dt + σdV,

6. Hull-White (extended Vasiček)

dr = {Φ(t) − ar} dt + σdV,

16
 Bond Options

European call on a T -bond with strike price K

and delivery date S.
X = max [hp(S, T ) − K, 0] i
T
X = max F (S, r(S)) − K, 0

1
FtT + µFrT + σ 2Frr
T − rF T = 0,
2
F T (T, r) = 1.

h i
T
Φ(r) = max F (S, r) − K, 0

1
Ft + µFr + σ 2Frr − rF = 0,
2
F (S, r) = Φ(r(S)).

Π [t; X ] = F (t, r(t))

17
 Affine Term Structures

Lots of equations!

Need analytic solutions.

We have an Affine Term Structure if

F (t, r; T ) = eA(t,T )−B(t,T )r ,

where A and B are deterministic functions.

Problem: How do we specify µ and σ in order

to have an ATS?

18
Proposition: Assume that µ and σ are of the
form

µ(t, r) = α(t)r + β(t),

σ 2 (t, r) = γ(t)r + δ(t).

Then the model admits an affine term struc-

ture
F (t, r; T ) = eA(t,T )−B(t,T )r ,
where A and B satisfy the system

(
Bt (t, T ) = −α(t)B(t, T ) + 1 2
2 γ(t)B (t, T ) − 1,
B(T ; T ) = 0.
(
At(t, T ) = β(t)B(t, T ) − 1
2 δ(t)B 2 (t, T ),
A(T ; T ) = 0.

19
 Parameter Estimation

Suppose that we have chosen a specific model,

e.g. H-W . How do we estimate the parameters
a, b, σ?

Naive answer:
Use standard methods from statistical theory.

NONSENSE!!

20
• The parameters are Q-parameters.

• Our observations are not under Q, but un-

der P .

• Standard statistical techniques can not be

used.

• We need to know the market price of risk

(λ).

• Who determines λ?

• The Market!

• We must get price information from the

market in order to estimate parameters.

21
 Inverting the Yield Curve
Q-dynamics with parameter vector α:
dr = µ(t, r; α)dt + σ(t, r; α)dV

Theoretical term structure

{p(0, T ; α); T ≥ 0}

Observed term structure:

{p? (0, T ); T ≥ 0} .

Want: A model such that theoretical prices

fit the observed prices of today, i.e. choose
parameter vector α such that
p(0, T ; α) ≈ {p?(0, T ); ∀T ≥ 0}

Number of equations = ∞ (one for each T ).

Number of unknowns =dim(α)

Need: Infinite dimensional parameter vector.

22
 Hull-White

Q-dynamics:

dr = {Φ(t) − ar} dt + σdV (t),

p(t, T ) = eA(t,T )−B(t,T )r(t),

1n −a(T −t)
o
B(t, T ) = 1−e
a

The instantaneous forward rate at T , con-

tracted at t is given by

∂ log p(t, T )
f (t, T ) = − .
∂T

Fit the observed forward rate curve!

23
Result: The Hull-White model can be fitted
exactly to any observed initial term structure.
The calibrated model takes the form

p(0, T )
p(t, T ) = × eC(t,r(t))
p(0, t)

where C is given by

σ2 2  
? −2aT
B(t, T )f (0, t)− 2 B (t, T ) 1 − e −B(t, T )r(t)
2a

Analytical formulas for bond-options.

24
 Heath-Jarrow-Morton

Idea: Model the dynamics for the entire yield

curve.

The yield curve itself (rather than the short

rate r) is the explanatory variable.

Model forward rates. Use observed yield curve

as boundary value.

Q-dynamics:

df (t, T ) = α(t, T )dt + σ(t, T )dV (t),

f (0, T ) = f ? (0, T ).

One SDE for every fixed maturity time T .

25
Theorem: (HJM drift Condition) Assume Q
dynamics of the form

df (t, T ) = α(t, T )dt + σ(t, T )dV (t),

f (0, T ) = f ? (0, T ).
Then the folloowing drift condition must hold.

Z T
α(t, T ) = σ(t, T ) σ(t, s)ds.
t

Moral: Volatility can be specified freely. The

forward rate drift term is then uniquely deter-
mined.

26
 Low Dimensional Models

• Historically the first to appear.

• Typical state variables are short rate, mean

reversion level, volatility, long rate etc.

• PDE techniques.

• Pro:
Low dimensional Markov processes. Ana-
lytical formulas.

• Con:
Hard to fit. Hard to model forward rate
volatilities in a flexible way.

27
High Dimensional Models (HJM)

• Model all infinitesimal forward rates simul-

taneously.

• Infinitely many state variables (one for each

forward rate).

• Numerical techniques.

• Pro:
Automatic fit to initial yield curve. Very
easy to model forward rate volatilities.

• Con:
The short rate is typically not Markov. Few
analytical formulas.

28
Problems with all “infinitesimal
models”

• The infinitesimal short rate and forward

rates can never be observed in real life.

• Market rates are often awkward in terms

of infinitesimal rates.

• Not consistent with market practice of us-

ing Black-76.

• The parameters in the models are hard to

interpret in economic terms.

• Hard to fit cap and/or swaption curves.

29
 Irritating Theoretical Problem

• There exists naive versions of Black-76 to

price caps, floors and swaptions.

• The logic behind these formulas typically

contains the following assumptions

– The relevant interst rates are stochas-

tic.

– The same rates are deterministic.

• Despite this logical disaster, the market

happily continues to use Black-76.

• If you cannot beat them: join them.

• Try to build a logically coherent theory whih

produces Black-76 type formulas.

30
 Market Models

• Model (lognormally) market interest rates

directly.

– LIBOR rates.

– Par swap rates.

• Will produce Black-76 type formulas for

caps/floors (LIBOR models), and swap-
tions (swap market models).

• Easy to fit to cap curve (LIBOR market

models) and swap curve (swap market mod-
els)

• The parameters in the models are easy to

interpret in economic terms (for example
LIBOR volatilities).

31
 Problems with Market Models

• LIBOR models are not consistent with swap

market models.

• Easy to fit to caps, floors and swaptions,

but for exotic products, Monte Carlo is
needded.

• Need to simulate high dimensional SDEs.

• Quarterly LIBOR model is not consistent

with semiannual LIBOR model etc.

32