Interest-Rate Models: Course Notes

Richard C. Stapleton
1
1
Manchester Business School
Interest-Rate 1
Spot-Rate Models
Normal Rate (Gaussian) Models
Vasicek (1977)
Hull and White (1994)
Lognormal Models
Black and Karasinski (1991) (BK)
Peterson, Stapleton and Subrah-
manyam (2003), 2-factor BK
Spot-rate Models
Assume a process for the spot short
rate
Derive bond prices, given the spot
rate process
Can be used to price derivatives
(caps, swaptions)
Interest-Rate 2
References
Vasicek, O., (1977), An Equilib-
rium Characterization of the Term
Structure, Journal of Financial Eco-
nomics, 5, 177188.
Hull, J., and A. White (1994), Nu-
merical Procedures for Implement-
ing Term Structure Models II: Two-
Factor Models, Journal of Derivatives,
2, 3748.
Black, F., and P. Karasinski (1991),
Bond and Option Pricing when Short
Rates are Lognormal, Financial An-
alysts Journal, 47, 5259
Peterson,S., Stapleton, R.C. and M.
Subrahmanyam (2003), A Multi-
Factor Spot-Rate Model for the Pric-
ing of Interest-rate Derivatives, Jour-
nal of Financial and Quantitative Analy-
sis
Interest-Rate 3
Lecture 1: The Gaussian Model
Vasicek-Hull and White type model
Assume that the short (- period)
rate follows a normal distribution,
mean-reverting process
Discrete time process
Under risk-neutral measure
Interest-Rate 4
Geometric Progressions
1.
S
1
= 1+(1k) +(1 k)
2
+... +(1 k)
n1
S
1
=
1 (1 k)
n
k
2.
S
2
= 1+(1k)
2
+(1k)
4
+...+(1k)
2(n1)
S
2
=
1 (1 k)
2n
1 (1 k)
2
Interest-Rate 5
Geometric Progressions
3.
S
3
= n+(n1)(1k)+(n2)(1k)
2
+...
+2(1 k)
(n2)
+ (1 k)
(n1)
S
3
=
1
k
_

_
n (1 k)
_

_
1 (1 k)
n
k
_

_
_

_
4.
S
4
= n+(n1)(1k)
2
+(n2)(1k)
4
+...
+2(1 k)
2(n2)
+ (1 k)
2(n1)
S
4
=
n (1 k)
2
_
_
1(1k)
2n
1(1k)
2
_
_
1 (1 k)
2
Interest-Rate 6
Short-Rate Model
r
t+1
r
t
= k(a r
t
) +
t+1
short rate of interest, r
t
long-term mean of short rate, a
rate of mean reversion, k, 0 < k < 1

t+1
, drawing from a normal distribu-
tion, E
t
(
t+1
) = 0, var
t
(
t+1
) =
2
.
Hence,
r
t+1
= ka + (1 k)r
t
+
t+1
(1)
Equation (1) holds for all t
Interest-Rate 7
Short-Rate Model
Example: Short rate:3-month Libor
t = 0
short rate, r
0
= 0.05
long term mean, a = 0.06
mean reversion, k = 0.25
variance,
2
= 0.0004
r
1
= 0.25(0.06) + 0.75(0.05) +
1
Interest-Rate 8
Mean and Variance of the Short-Rate
r
t+1
= ka + (1 k)r
t
+
t+1
(1)
r
t+2
= ka + (1 k)r
t+1
+
t+2
(2)
Substitute (1) in (2)
r
t+2
= ka+(1k)ka+(1k)
2
r
t
+(1k)
t+1
+
t+2
Mean of r
t+2
:
E
t
(r
t+2
) = ka + (1 k)ka + (1 k)
2
r
t
(3)
Variance of r
t+2
:
var
t
(r
t+2
) = (1 k)
2
var(
t+1
) + var(
t+2
) (4)
Exercise: Find the mean and variance
of r
t+3
Interest-Rate 9
Mean and Variance of r
t+n
r
t+n
= ka + (1 k)ka + (1 k)
2
ka + ... + (1 k)
n1
ka
+ (1 k)
n
r
t
+ (1 k)
n1

t+1
+ (1 k)
n2

t+2
+ ... +
t+n
Hence
E
t
(r
t+n
) = ka + (1 k)ka + (1 k)
2
ka + ...
+ (1 k)
n1
ka + (1 k)
n
r
t
= ka[1 + (1 k) + (1 k)
2
+ ... + (1 k)
n1
Variance of r
t+n
:
var
t
(r
t+n
) = (1 k)
2(n1)
var(
t+1
)
+ (1 k)
2(n2)
var(
t+2
)
+ ... + var(
t+n
)
and if
var(
t+1
) = var(
t+2
) = ... = var(
t+n
) =
2
var
t
(r
t+n
) =
2
[(1k)
2(n1)
+(1k)
2(n2)
+...+1]
Interest-Rate 10
Mean and Variance of r
t+n
Using the results from Geometric Pro-
gression:
Mean of r
t+n
:
E
t
(r
t+n
) = a[1 (1 k)
n
] + (1 k)
n
r
t
Variance of r
t+n
:
var
t
(r
t+n
) =
2
_

_
1 (1 k)
2n
1 (1 k)
2
_

_
Annualised st.dev.
std
t
(r
t+n
) =
_
var
t
(r
t+n
)/n
Interest-Rate 11
Long-Bond Prices and Yields
Some key results:
1. Mean of a lognormal variable:
Let x be normally distributed, with
(,
2
) then
E(e
bx
) = e
b+0.5(b
2

2
)
2. Forward price of a bond:
The forward price of a 1-period bond
is
B
t,t+,t++1
= E
t
[B
t+,t++1
]
3. The price of a long-term bond
B
t,t+n
= B
t,t+1
B
t,t+1,t+2
...B
t+,t++1
...B
t+n1,t+n
Interest-Rate 12
Long-Bond Prices and Yields
B
t,t+,t++1
= E
t
[e
r
t+
]
B
t,t+,t++1
= e

t+
+
var
t
(r
t+
)
2

t+
= E
t
[ln r
t+
]
Substitute in the equation for B
t,t+n
to
obtain the long bond price.
Bond yields:
Dene the yield by y
t+n
in
B
t,t+n
= e
y
t+n
n
and solve for
y
t+n
=
ln(B
t,t+n
)
n
Interest-Rate 13
Long-Bond Prices and Yields
The price of a long-term bond
B
t,t+3
= B
t,t+1
B
t,t+1,t+2
B
t,t+2,t+3
= e
r
t
E
t
[e
r
t+1
]E
t
[e
r
t+2
]
= e
r
t
e

t+1
+
var
t
(r
t+1
)
2
e

t+2
+
var
t
(r
t+2
)
2
B
t,t+3
= e
[r
t
+ka+(1k)r
t
+ka+(1k)ka+(1k)
2
r
t
]
. e
1
2
[2
2
+(1k)
2

2
]
= e
[r
t
[1+(1k)+(1k)
2
]
e
[2ka+(1k)ka]
. e
1
2
[2
2
+(1k)
2

2
]
Interest-Rate 14
Long-Bond Prices and Yields
B
t,t+n+1
= e
[r
t
[1+(1k)+...+(1k)
n
]
. e
[nka+(n1)(1k)ka+...+(1k)
n1
ka]
. e
1
2
[n
2
+(n1)(1k)
2

2
+...(1k)
2(n1)

2
]
Using geometric progressions 3 and 4:
B
t,t+n+1
= e
r
t
__
1(1k)
n+1
k
__
. e
a
_
n(1k)
_
1(1k)
n
k
__
. e

2
2
_

_
n(1k)
2
_
1(1k)
2n
1(1k)
2
_
1(1k)
2
_

_
Interest-Rate 15
Calibrating the Model to the Current
Term Structure
Hull and White suggest the following
generalisation:
r
t+1
r
t
= k(a
t
r
t
) +
t+1
This is equivalent to
r
t+1
= r
t
+
t
+ k(a r
t
) +
t+1
where
t
is a time dependent drift.
With this adjustment, the model can
be tted exactly to the current term
structure of bond forward prices.
Interest-Rate 16
A Two-factor Extension
1-factor model does not reect term-
structure of volatility
2-factor model has a stochastic cen-
tral tendency
Let
r
t+1
r
t
= k
1
(a
t
r
t
) + y
t
+
t+1
y
t
y
t1
= k
2
y
t1
+
t
k
1
and k
2
are rates of mean reversion

t+1
and
t
are a drawings from nor-
mal distributions (0,
1
), (0,
2
)
Interest-Rate 17
A Two-factor Extension
After successive substitution:
r
t+3
= (1 k
1
)
3
r
t
+ k
1
[a
t+2
+ (1 k
1
)a
t+1
+ (1 k
1
)
2
a
t
]
+ y
t
[(1 k
1
)
2
+ (1 k
1
)(1 k
2
) + (1 k
2
)
2
]
+ (1 k
1
)
2

t+1
+ (1 k
1
)
t+2
+
t+3
+ (1 k
2
)
t+1
+
t+2
If var(
t+i
) =
2
1
and var(
t+i
) =
2
2
,
var(r
t+n
) =
2
1
_

_
1 (1 k
1
)
2n
1 (1 k
1
)
2
_

_
+
2
2
_

_
1 (1 k
2
)
2(n1)
1 (1 k
2
)
2
_

_
Annualised st.dev.
std
t
(r
t+n
) =
_
var
t
(r
t+n
)/n
Interest-Rate 18
Gaussian Model: Advantages and
Disadvantages
Simple model
Relatively easy to program (us-
ing a binomial tree)
Can capture the current term struc-
ture
Two-factor model can reect term
structure of volatilities
Yields analytical (zero-coupon) bond
prices
Closed form expression for bond
price
Bond prices are lognormal
Options on bonds priced with BS
But, short rates may not normally
distributed
Two-factor model may not yield re-
alistic swaption prices
Interest-Rate 19
Lecture 2: Lognormal Models
Types of Lognormal Models
One-Factor BK spot-rate model
Two-factor PSS spot-rate model
Forward rate models
One-Factor BK model
Short rate follows a mean revert-
ing, lognormal process
Under risk-neutral measure
Constructed using HSS (1995) method
Two-Factor PSS model
Short rate follows a mean revert-
ing, lognormal process
With stochastic central tendency
Constructed using PSS (2003) method
Interest-Rate 20
Short-Rate Lognormal Model
Assume that:
ln r
t+1
ln r
t
= k(a
t
ln r
t
) +
t+1
short rate of interest, r
t
long-term mean of log of short rate, a
t
rate of mean reversion, k, 0 < k < 1

t+1
, drawing from a normal distribu-
tion, E
t
(
t+1
) = 0, var
t
(
t+1
) =
2
.
Hence,
ln r
t+1
= ka
t
+ (1 k)ln r
t
+
t+1
(1)
Equation (1) holds for all t
Interest-Rate 21
Mean and Variance of ln r
t+n
(a
t
= a)
Mean of ln r
t+n
:
E
t
(ln r
t+n
) = a[1 (1 k)
n
] + (1 k)
n
ln r
t
Variance of ln r
t+n
:
var
t
(ln r
t+n
) =
2
_

_
1 (1 k)
2n
1 (1 k)
2
_

_
Annualised volatility
std
t
(ln r
t+n
) =
_
var
t
(ln r
t+n
)/n
Mean of r
t+n
:
E
t
(r
t+n
) = e
E
t
(ln r
t+n
)+
var
t
(ln r
t+n
)
2
Interest-Rate 22
Two-Factor BK Model
ln r
t+1
ln r
t
= k
1
(a
t
ln r
t
) + ln y
t
+
t+1
ln y
t
ln y
t1
= k
2
ln y
t1
+
t
y
t
is a premium factor
a shock to the futures rate
Interest-Rate 23
Mean and Volatility of r
t+n
Since r
t+n
is lognormal, the mean is
given by
E
t
(r
t+n
) = e
E
t
(ln r
t+n
)+
var
t
(ln r
t+n
)
2
Example: 2-Factor BK model
t = 0
short rate, r
0
= 0.05
long term mean, a = 0.06
mean reversion, k
1
= 0.15

1
= 0.2
mean reversion, k
2
= 0.2

2
= 0.15

t
= 0
E
0
(r
5
) = e
2.894+
0.168
2
2
= 0.0602
Volatility = 18.32%
Interest-Rate 24
Implementing the BK and 2-Factor
BK Models
One-Factor BK spot-rate model
Fitting the mean of the process
Using futures Libor quotes
Iterative calibration to forward
bond prices
Calibrating to cap volatilities
Generalise model using (t)
Recombining tree using HSS (1995)
Two-factor BK model
PSS implementation
Recombining tree in two dimen-
sions
G2++ model, Brigo and Mercu-
rio
Interest-Rate 25
HSS Implementing the BK Model
HSS (1995) build a binomial approx-
imation to a lognormal process
A re-combining tree with mean re-
version and time-dependent voatil-
ity
Conditional probabilities in the tree
Method can be applied to interest
rate process
Interest-Rate 26
HSS Implemention of the BK Model
HSS Method
First build a process for x
t
, where
E
0
(x
t
) = 1
Assume mean reversion and volatil-
ity same as in the interest-rate pro-
cess

ln x
t+1
= a
x,t
+ (1 k)ln x
t
+
t+1
a
x,t
= E
0
[ln x
t+1
] (1 k)E
0
[ln x
t
]
=
var
0
[ln x
t+1
]
2
(1 k)
var
0
[ln x
t
]
2
The conditional probabilities, q
t
de-
pend on a
x,t
and k
Scale the process using futures rates
to obtain approximation to
ln r
t+1
=
t
+ (1 k)ln r
t
+
t+1
Interest-Rate 27
HSS Implemention of the BK Model:
an Example
Futures rates
h
0,1
= 5.0%
h
0,2
= 5.0%
h
0,3
= 5.0%
h
0,4
= 5.0%
Volatility (constant): 10%
Mean reversion k = 0.2
Cap Vols
t = 1 : 0.1
t = 2 : 0.0906
t = 3 : 0.0827
t = 4 : 0.0760
Interest-Rate 28
Implemention of the 2-Factor BK
Model
The PSS (2002) method
First build a process for x
t
, y
t
, where
E
0
(x
t
) = 1, E
0
(y
t
) = 1
Assume mean reversion and volatil-
ity same as in the interest-rate pro-
cess

ln x
t+1
= a
x,t
+ (1 k
1
)ln x
t
+ ln y
t
+
t+1
ln y
t+1
= a
y,t
+ (1 k
2
)ln y
t
+
t+1
The conditional probabilities, q
x,t
de-
pend on a
x,t
, k
1
, and ln y
t
Scale the process using futures rates
to obtain approximation to
ln r
t+1
=
t
+ (1 k
1
)ln r
t
+
t+1
Interest-Rate 29
Implemention of the 2-Factor BK
Model
The G2++ method
Brigo and Mercurio suggest a transfor-
mation of the (r
t
, y
t
) process:
r
t+1
r
t
=
t
k
1
r
t
+ y
t
+
t+1
y
t+1
y
t
= k
2
y
t
+
t+1
Dene
x
t
= r
t
+
y
t
k
2
k
1

t
=
y
t
k
1
k
2
Then:
x
t
x
t1
=
t
k
1
x
t
+
t+1

t1
= k
2

t
+

t+1
k
1
k
2
,

t+1
=
t+1
+

t+1
k
1
k
2
Interest-Rate 30
Implemention of the 2-Factor BK
Model
The G2++ method
cov(
t+1
,
t+1
) =
1
k
2
k
1
var(
t+1
)
=
1
k
2
k
1

2
2
Interest-Rate 31
2-Factor BK Model
Conclusions
BK model is too simple
LMM is complex and lacking intu-
ition
2-factor BK model is a good com-
promise
Captures termstructure of cap volatil-
ities
Generate scenarios for risk manage-
ment
Valuation of Bermudan swaptions,
exotics
3-factor extension to capture swap-
tion vols