Basic Term Structure Modelling

1
Introduction to Single Factor
Models
2
Introduction
Interest rate models make different assumptions about
interest rates

Each model differs fundamentally in its implications for
valuing derivatives

Most interest rate models assume it rate is a continuous
random variable stochastic process continuous -time
stochastic process

What model should we use for interest rate

Cannot settle by theoretical reasoning
3
Things to Consider in a Model
Number of Factors: What is driving the yield curve other
maturities what is the sources of uncertainty

One Factor Models: Short rate only factor drives all other
rates other rates not randomly determined once short rate
specified all zero coupon bonds are driven by the same
source of uncertainty

Multi-factor Models: Short rate/long rate/central
tendency/stochastic volatility

Volatility-Level Effect: Rates |(+) volatility |(+)
Level Effect

4
Volatility/Level Effects

5
2
4
6
8
10
12
14
16
1992 1994 1996 1998 2000 2002 2004
UK
0
2
4
6
8
10
1992 1994 1996 1998 2000 2002 2004
US
0.0
0.5
1.0
1.5
2.0
2.5
3.0
1992 1994 1996 1998 2000 2002 2004
UK Volatility
0.0
0.4
0.8
1.2
1.6
2.0
1992 1994 1996 1998 2000 2002 2004
US Volatility
Figure 1.1: UK and US Interbank Interest Rates
Volatility/Level Effects

6
0
1
2
3
4
5
6
7
8
9
1992 1994 1996 1998 2000 2002 2004
JAPAN
0
2
4
6
8
10
12
1992 1994 1996 1998 2000 2002 2004
SINGAPORE
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1992 1994 1996 1998 2000 2002 2004
JAPAN Volatility
0
1
2
3
4
1992 1994 1996 1998 2000 2002 2004
SINGAPORE Volatility
Figure 1.2: JAPAN and SINGAPORE Interbank Interest Rates
Things to Consider in a Model
Stochastic Volatility
Jumps: Bank of England, Fed, BOJ
Tractability
Closed-Form Solutions

Understanding!!
Regulators: OCC, FSA
Clients
Management!!
7
How Do We Choose
No unique best model for all needs

Models are tools for doing jobs

What job am I wanting it to do

Empirical work can help us see how different models have
behaved in the past

Help with calibration parameters
8
Introductory Mathematics
9
Introductory Mathematics
Instantaneous change in R.V: x dx
Changes ~ Normal
Shock (Noise) make interest rate change
Wiener process or Brownian motion process a
stochastic process whose increments are independent,
stationary, and normally distributed

History:
1828: Robert Brown observed random behaviour
pollen particles in water looked at the trajectory of
particles in suspension under the microscope
observed constant irregular motion Brownian Motion

Louis Bachelier (1900, Ph.D thesis ) Brownian motion
first introduced into finance as a possible model for
stock market prices

10
Introductory Mathematics
Wiener (1923): Existence and construction of Brownian
motion established

Wiener process

Kiyoshi It (1951): Stochastic Differential Equations (SDE)

Samuelson (1965) early work options using Brownian
motion

Black and Scholes (1973) established Brownian motion as a
standard assumption in continuous time modelling


11
Building a Model of Interest Rates
Weiner Process

Instantaneous changes dz

z ~N(0,1)

Change in short rate for any two
different short intervals of time is
independent


( )
( )
Changes/jumps scaled volatility
s.d changes in
r dr t dz
r
dr t dz
o
o
=

=
stochastic process
12
Drift Element in Rates
Drift: Short rate

Rates: Up/down

Interest rate cycle

Expected direction change
( )

On average move in rate
function drift/random shocks
change in short rate
drift expected direction rate ch.
incremental change time
standard deviation ch. short rat
dr t dt dz
dr
dt
o o
o
o
= +
=
=
=
= e
"Volatility Term"
random process dz =
13
Mean Reversion
Mean Reversion Process: Allows drift to depend on level

Common characteristic - long run stable mean value exists

Rates above/below pull back to long run

Rates increase economy slows less demand loans etc
rates tend to fall

How do we add a term to capture this?






14
Mean Reversion

15
Interest
Rate
Mean Reversion
Time
High interest rate has
negative trend (drift)
Low interest rate has
positive trend (drift)
Interest
Rate
Interest
Rate
Mean Reversion
( ) ( ) ( )
( ) ( )
0; 0;
dr t r t dt dz
dr t r t dt dz
o | o
o |
o
| o
|
o
|
|
= + +
> <
| |
= + +
|
\ .
=
=
long - run mean of interest rate
Speed of mean reversion towards the long - run mean
16
Short Rate Models
17
Introduction
Proxy Choice of Short Rates

CKLS Model and Special Cases

CKLS Alternative Discrete Time Model and Results

International Evidence

Other Models Arbitrage Free Models


18
Choice of Proxy For Short
Rate
19
Choice of Proxy
Short Rate Proxy: What to use??
Duffee (1996, JF)
Instantaneous interest rate
Treasury-bills benchmark other debt instruments
EX: US 1-month risk-free proxy
Duffie idiosyncratic movements short-bills 2 months and less
show price movts that are unrelated to changes in other interest
rates either long term T-bills or similar-maturity money market
rates
Do not use 1-month as proxy
Overnight?: Spurious microstructure effects
EX: Short term S/D effects create excess volatility short end of curve
irrelevant for rest of curve


20
Choice of Proxy
Models default-free term structures
Not model yields on privately-issued instruments
But: Short-term instruments issued highly rated firms small
probability default
Proxy short-rate: London Inter Bank Offered Rate (LIBOR)
Others:
7-day rate: Eurodollar
Interbank; CD; Eurocurrency etc
1-month; 2-month; 3-month




21
CKLS Model and Special Cases
22
Equilibrium Models
Equilibrium Models:

Dynamics of curve - economic variables + interest rate process
do not incorporate market data

Derive closed form solutions equilibrium prices
bonds/interest rate derivatives

Assume: Functional form drift/interest rate volatility

Model generates family of curves



23
Equilibrium Models
Initial (Current) Term Structure (spot rate): Endogenous
specified term structure explained by model term
structure is an output

Constant Parameters
Parameters do not change day to day

No time dependent parameters

Time series data econometric methods
Estimate drift / mean reversion / volatility

24
Equilibrium Models
Equilibrium models do not take bond prices as given
These models are not calibrated to market data
Compute zero prices internally arbitrage free but not
externally arbitrage free
Thus these models do not guarantee that the prices of
zero-coupon bonds produced by the model will match the
prices observed in the market
Suggests possibility of an arbitrage opportunity provided
parameters used to build the model are correct so dealer
using the model could be offering arbitrage opportunities
Traders execute transactions at market prices which differ
from the dealers prices via model

25
Gaussian Models
Gaussian Models Normal Models
Assumes short-rate process follows normal distribution
Volatility independent of level of rates

Merton (1973)

Famous model of Vasicek (1977)

Known as Ornstein-Uhlenbeck Process
26
Negative Interest Rates
Normal Models have negative nominal interest rates

Positive probability that short rate take negative values

Negative rates - if you could borrow money pay back less
than borrowed accumulate infinite amount of wealth by
simply borrowing no matter what you did with the funds

Incorporation of mean reversion reduces chance of negative
interest rates over the long term



27
Negative Interest Rates
Model generated negative rates

Occurrence depends on initial interest rate + parameters of
model
EG: Generated sometime initial rate low Japan 2000 period;
vol set to market levels
EG: Japan 1997-1998 money market rates below 0.5%. At that
level even a vol value < 5% when plugged into Vasicek model
will imply a high probability of negative rates
Other models allow relationship between level of rates and
volatility of rates Level Effect




28
Merton (1973)






29
( )





dr t dt dW o o = +
| | |

Merton Model
Note : Arithmetric Brownian Motion representa
error term
Change Drift Volatility
short rate
tion
Vasicek (1977)





( ) ( ) ( )
( ) ( )
0; 0;
long-run mean of interest rate
Speed of mean reversion towards the long-run mean
More negative faster responds to deviations from the long-r
dr t r t dt dW
dr t r t dt dW
r
o | o o |
o
| o
|
o
|
|
|
= + + > <
| |
= + +
|
\ .
=
=
un mean
Note: <0 diversion from the mean
1
=
1
EX: 0.8 periods for short rate to revert towards long-run mean
0.8
o
|
|

=
Mean Time Lag
30
Cox-Ingersoll-Ross (1985)
CIR (1985) model actually described in academic circles in
1977!!
Conditional Heteroskedasticity
Volatility increase with the square root - level of the current
interest rate
When current interest rate goes to zero square root of zero
causes volatility to go to zero and the rate will be pulled up by
its drift
CIR model implied rate follow chi-squared distribution
skewed prob. distribution
CIR Model guarantees positive rates
Non-linear SDE

31
Cox-Ingersoll-Ross (1985)




( )
( )
( )
( )
2



Short rate is No negative rates
t t dr t r dt r dW o | o
_
= + +
| |

Speed of Volatility - Level
Mean Reve
Square Root Mo
rsion
d

el


32
Dothan (1978)






33
( ) ( )
- Log Normal Models short rate cannot go negative



dr t r t dW o =
|
Driftless Geometric Brownian Motion


Volatility - Level
Geometric Brownian Motion



( ) ( ) ( )



dr t r t dt r t dW | o = +
|


Volatility- Level
34
Brennan and Schwartz (1980)



( )
( )
( )
( )





t dt t dr t r r dW o | o = + +
|
Used for convertible bonds analysis
Volatility- Level

35
Cox-Ingersoll-Ross (1980)





( ) ( )
3
2






t dr t r dW o =
|


Volatility- Level
36
Constant Elasticity of Variance
(CEV)




( ) ( ) ( )



Constant elasticity of variance

dr t r t dt r t dW

| o

= +
|

Volatility - Level
37
Chan-Karolyi-Longstaff-Sanders
(1992,JF)
Many models used in the markets

Developed in 1970-1980s

Do not know to compare them all

Implied effect on prices

Chan, Karolyi, Longstaff and Sanders seminal paper Journal
of Finance known as CKLS model
38
Chan-Karolyi-Longstaff-Sanders
(1992,JF)




Introduced general interest model

It nested many models by imposing restrictions on drift and
mean reversion

Volatility-level to be free to power gamma
39
( ) ( )
{ }
( )
dr t r t dt r t dW

o | o = + +
CKLS and Nested Models
o |
Merton 0 0
Vasicek 0
CIRSR 1\2
Dothan 0 0 1
GBM 0 1
BS 1
CIRVR 0 0 3\2
CEV 0
40
CKLS Approximation
( ) ( )
{ }
( )

Only have discrete data map available discrete time data to
obtain parameters of the CTM
"Discretization bias"

1 1
0
1
2
1
dr t r t dt r t dW
r r r
t t
t t
E
t
E
t

o | o
o | c
c
c
(
(

(

= + +
= + +
+ +
=
+
+
"Forward Difference"
Variance same form c.t. model
2
2 2
r m
t tt

o
(
= =
41
CKLS Results
Use Generalized Method Moments (GMM)

Hansen (1980)

Find = 1.499 for monthly US 1-month T-Bill rates.

Models with > 1 have a better fit

Models with < 1 have a worse fit

42
Volatility
43
Test Monetary Policy
44
Test Monetary Policy
Federal Reserve Experiment: October 1979-September 1982 -
FED announced focus on monetary aggregates
Less on interest rate levels in combating historically high
inflation. Early 1980: Imposition of credit controls sharp
drop in interest rates Feb-May then lifted rates | winter
levels end of summer
They test changes of US monetary policy on the interest rate
process
CKLS: No evidence of structural shift in interest rate process
Look data late 1970s/early1980s - Level/volatility appear
elevated



45
CKLS Revisited
Alternative Discrete Time Model
46
CKLS Revisited
Another approach to obtaining the parameters of the CKLS
model

Nowman (1997, Journal of Finance)

Alternative discrete model

Reduces bias




47
CKLS Revisited
( )
( )
( )
( )
( )
( )
(1)
Assume over the interval 0 , satisifies the SDE
1 (2)
dr t r t dt r t dZ
,T r t
dr t r t dt r t dZ

o | o

o | o

`
)
(
(


| |

| ` `
\ .
)

)
= + +
'
= + +
CKLS model
48
Assumptions
Volatility changes beginning of unit observation period then
remains constant

Beginning of each data observation period

Quarter, Month, Week, Day: Remains constant through the
period until the next period

Assumed Mean Reversion: Continuously changing, |r(t),

49
CKLS Revisited
( )
asiticity heterosced
2
2
1 1
2
2
2
2
n correlatio serial no t s 0
0
1
(3) 1 1
= =
= =
=
=
+ + =
)
`

|
.
|

\
|
|
|
.
|

\
|
|
|
.
|

\
|
|
|
.
|

\
|
|
|
.
|

\
|
|
.
|

\
|
|
|
.
|

\
|
|
.
|

\
|
tt
m t r e
t
E
t
s
E
t
E
,...,T t
t
e t r e t r

|
|
o
q
q q
q
q
|
|
o |
50
Likelihood Function
( )

|
|
|
.
|

\
|
|
.
|

\
|
|
|
.
|

\
|
)
`

|
.
|

\
|
|
|
.
|

\
|
|
|
.
|

\
|
|
.
|

\
|
|
|
.
|

\
|
|
.
|

\
|
'

=
= =
= + + =
t
tt
m
t tt
m
t
r p
tt
m t r e
t
E
,...,T t
t
e t r e t r
q q t

|
|
o
q
q
|
|
o |
2
1
2
1
exp
2 1
2
2 1
2
: p.d.f.is
2
2
1 1
2
2
2
2
1 1 1
: Start with
51
Likelihood Function


=

)

'
+
=
=

=

)

'

=
=

T
t
t
tt
m
t
tt
m
T
t
LF
T
t
t
tt
m
t
tt
m
T
t
T
LF
T
1
2
1
2
1
log log 2
constant) a (excluding 2
1
2
1
2
1
2
1
log
2
1
2 log
2
log
: is obs for logLF
q q
q q t
52
Likelihood Function

=
+ =
= =
(
(
(
(

|
.
|

\
|
|
|
|
.
|

\
|
T
t
tt
m
t

tt
m L
LF L
1
2
2
log 2
2
, , , , log 2
Let
u
o | o u
53
Likelihood Function
This is the likelihood
function L to estimate 0

We minimize this function
with respect to unknown
parameters. The minimized
value (or -1 times min
value= max value) of L then
gives us the maximum
likelihood estimates (MLE)


.
2

(
(

= o | o u
54
Empirical Results
Monthly 1-Month Interbank Rate

CKLSs monthly 1-Month US T-bill

Found for UK that = 0.2898

Found for US that = 1.3610

Weak mean reversions


55
International Evidence
56
Tse (1995, JIMF)
Tse (1995)

CKLS Model

Money market 3m rates

France, Holland, USA: High Elasticity

Australia, Belgium, Germany, Japan: Medium

Canada, Italy, Switzerland, UK: Low elasticity
57
Tse (1995, JIMF)
Country
Au 0.6763
Be 0.7579
Ca -0.36
Fr 1.6289
Ge 0.7141
Ho 1.5997
It 0.2519
Ja 0.6187
Sw 0.0424
UK 0.1132
USA 1.7283
58
International Evidence on
Volatility/Level Effects Japan
Application to 7d,1m, 2m, 3m, 6m, 9m, 12m Japanese Offshore Rates:
Hiraki and Takezawa (1997, Economic Letters)

Japan Weekly
7d = 0.313
1m = 0.392
2m = 0.437
3m = 0.367
6m = 0.438
9m = 0.419
12m = 0.343
Conclusions:
Range: 0.31-0.44


59
Arbitrage-Free Models
60
Arbitrage-Free Models
Arbitrage-Free Models (No-Arbitrage Models): Designed to
be exactly consistent with todays observed term structure of
interest rates Term Structure Consistent Models
Starts:
Observed market price set financial instruments (cash market
instruments; interest rate derivatives)
Benchmark Instrument (reference set) fairly priced
Assume: Random process drift/volatility



61
Arbitrage-Free Models
Assume parameter drift
Computational procedure compute term structure (spot
curve)

So that - valuation process generates observed market prices
for benchmark instruments
Model bond prices match market prices - advantage
Use: Term structure estimated + volatility

Then: Non-benchmark instruments valued theoretically

62
Arbitrage-Free Models
Todays term structure of interest rates in an input to the
model

Observed bond yields equal to computed models bond yields
Arbitrage-Free Model: Consistent with currently observed
zero-coupon curve

Drift: Short rate function of time
Why: Future path of short rate dependent on initial yield
curve

63
Ho and Lee Model (1986, JF)
( ) ( )
First authors develop model consisitent with any specified
initial yield curve
No mean reversion/vol independent rate normal model
Continuous time limit of the model is:

s
dr t dt dW t u o
o

= +
=
( )
.d of the short rate is constant vol/level indep normal model
t a function chosen to ensure that the model fits
the initial term structure
It give the average direction that moves at time
(i
r t
u

=
ndependent of short rate)

64
Ho and Lee Model (1986)
Easy to Implement: Lot of work on analytical bond (options)
prices
Interest Rates: Normally distributed
Uses: Information current term structure
Model fits current term structure
Two parameters needed
Model has no mean reversion realistic ?
Same volatility all spot rates restrictive ?
Negative yields possible
Also since no mean reversion it has no limiting stationary
state
65
Hull and White Model (1990, RFS)
Hull and White Model (HW)
Sometime called Extended Vasicek Model with time dependent
drift
Normal Model
Uses Vasiceks model to obtain a theoretical yield curve and fit it to
the observed current market term structure curve
+ time dependent drift
Time-dependent drift is a drift rate whose value is dependent on
the time period used to calculate it based on historical movt up
until now
Model is popular since can calculate a theoretical curve that is
identical to yields observed in the market and that can be used to
price bonds and bond derivatives



66
Hull and White Model (1990) (HW)
( )
( )
speed of mean reversion
time-dependent drift
t
dr r dt dW
t
|
o o
o
o
|
o
| |
|
|
|
\ .
= +
=
=
67
Black-Derman-Toy (1990, Financial Analysts
Journal)
( )
( )
( )
( )
( )
( )
( )
( ) ( )
BDT model
BDT model incorporates two independent functions of time
t and t -chosen so
mean revertin
that the model fits th
g
e
Inr t
t
dInr t t Inr t dt t dW
t
o
u o
o
u o
(
(
(
(
(


'
= + +
term structure of
spot interest ra
Therefore it uses additional market information. It focuses on the volatility curve,
the volatilities of yields of zero-coupon bonds of different ma
and the tes term structure of spot rate volatilities
turities.
(1) No analytic solutions for prices of bonds
(2) Complex drift - numerical fitting to observed current int. rate
Drift - not computed analytically
(3) Short rate volatility linked to mean
Note:
reversion
68
Black and Karasinski (1991 Financial
Analysts Journal) (BK)
( ) ( )
( )
BK model incorporates 3 independent functions of time
Allows a third initial condition to be incorporated
Mean reverting log-normal short rate model
-
The model compute
dInr t t Inr dt t dW u o o
(
(

= +
d using numerical methods
69
Conclusions
Single factor models

International volatility-level effects

Different type of models








70