Several other models with a mean-reverting process like in Vasicek and CIR have also

been developed. Usually the difference is only in modelling the second term describing
uncertainty of future developments. In Vasicek the coefficient of the Wiener-process
(explained in the next section) is simply s and thus independent of r. In CIR it is
a
s r . Courtadon (1982) models it as s r, Chan et al. (1992) as s r and Duffie &

Kan (1993) as s 0 + s 1r . Ait-Sahalia (1996) finds that the linearity of the drift term
a ( g - r )dt is actually the main source of misspecification and not the second term as
thought before. To overcome the problem of linearity it is possible to add jumps to the
process, which means unfortunately that closed form solutions of the bond prices
become unattainable. (Copeland et al, 2005, p. 268-269)

Duffie (2000) states several negative aspects of these equilibrium models. For one he
explains that the models cannot simultaneously allow for negative correlations between
state variables and guarantee that interest rates would be positive. They can´t also
capture the nonlinearities in the data in a satisfying manor. Ahn, Dittmar and Gallant
(2002) created a model to overcome these drawbacks. It specifies yields as quadratic
functions of the state variables. In the Black-Karasinski (1991) model the state variable
is identified as log r to avoid the problem of negative interest rates (Leippold & Wiener,
2000, p. 2). (Copeland et al, 2005, p. 269)

2.2.4 Vasicek interest rate model

The Vasicek model uses a mean-reverting stochastic process to model the evolution of
the short-term interest rate. Mean reversion is one of the key innovations of the model
and this feature of interest rates can also be justified with economic arguments. High
interest rates tend to cause the economy to slow down and borrowers require less
funds. This causes the rates to decline to the equilibrium long-term mean. In the
opposite situation when the rates are low, funds are of high demand on the part of the
borrowers so rates tend to increase again towards the long-term mean. (Zeytun &
Gupta, 2007, p. 2)

9
The model assumes that the current short interest rate is known for sure and the
subsequent values of it follow this stochastic differential equation

drt = a (g - rt ) dt + s dzt (2)

which is interpreted as,

ò dr = ò a (g - r )dt + ò s dz
t t t

(Vasicek, 1977, p. 9)

r(t) is a continuous function of time (no jumps) and follows a Markovian process, which
means that the system has no memory. That is, future developments of the short rate
are independent of past movements. Since this is a continuous Markovian process it´s
called a diffusion process. The model assumes the market to be efficient. (Svoboda,
2003, p. 5)

In the formula g is the long-term mean of the short-term interest rate r (the shortest
possible - usually understood as instantaneous). In some advanced models the long-
term mean isn´t constant but a function of time. a is the speed of adjustment with
which r closes on the long-term mean g . If r > g then the coefficient a > 0 makes the
drift-term negative and thus the rate will be pulled back down towards g . The opposite
happens when r < g . (Zeytun & Gupta, 2007, p. 2) Because there is only one factor, all
interest rates are ultimately dependent on the shortest interest rate r. All interest rate

models are additive and that is why the t+ D t interest rate is of type: rt + D r. This
characteristic will be used in section 5, where we develop a program for the evolution of
the short rate.

The second term aims to capture the instantaneous volatility caused by possibly infinite

number of unpredictable factors. The symbol s is the volatility and zt is a Wiener-

process, which means that the value zT at time T conditional on its value zt is normally

distributed with a mean zt and variance T-t, where T is larger than t (James & Webber,

10
2002, p. 54). The market price of risk is assumed to be a constant and usually negative
or zero to guarantee a positive premium for bond prices (Vasicek, 1977, p. 7 & 9). In

order to construct the term structure (a function of rt ) we only need the long-term mean,

the speed of adjustment and the standard deviation. This procedure and a program that
uses it, is described more thoroughly in section 6.

The yield curves start at the current level of instantaneous interest rate r(t) and approach
the following infinite yield

s2
R¥ = g - (3)
2a 2
The model can produce upward or downward sloping or humped yield curves. If

s2
r (t ) £ R¥ - (4)
4a 2
the yield curve is monotonically increasing, that is, upward sloping. If
s2 s2
R¥ + ³ r ( t ) ³ R¥ - (5)
2a 2 4a 2
the yield curve is humped. The last scenario is when

s2
r (t ) ³ R¥ + 2 (6)
2a
and this produces a monotonically decreasing curve. (Vasicek, 1977, p. 10)

The model can be extended into a multifactor model simply by taking other factors into
the process and handling them in the same way as the short-term interest rate is
handled (mean-reverting process). This was done by Langetieg in 1980. Actually the
famous one-factor model is simply a special case. In the single factor model the short
rate Rt equals the state variable but in the multifactor model the short rate is the sum of

independent state variables yit

k
Rt = å yit (7)
i =1

where, k is the number of the factors. (Piazzesi, 2003, p. 32)

11
The discount bond price has a well-known closed form solution, which is, when
generalized into the multifactor form, the following
k
Pt = Õ Ait e- Bit rit (8)
i =1

where,
æ ( B - t )(a i (a ig - qis i ) - s i2 / 2) s i2 Bit 2 ö
Ait = exp ç it - ÷ (9)
è a i2 4a i ø

and,
1 - exp(-a i t )
Bit = (10)
ai

In the equation Ait we can see the market risk premium q for the first time. This is set to

zero in this paper since we want to model the interest rate process and the term
structure in a risk-neutral world where bond and fixed income derivatives are priced. It´s
good to remind that the real world development of r(t) does not matter in the pricing
process. (Hull, 2003, p. 537)

2.2.5 Drawbacks of the single factor Vasicek model

We must however acknowledge that the single factor model used here, although
empirically very tractable, is also subject to some criticism. The dependence on a single
factor greatly limits the possible shapes of the yield curve and often leads into situations,
where the theoretical yield curve does not correspond to the market yield curve. In fact
this is the case nearly always. Another drawback is the undesired property that the
yields of all maturities are perfectly correlated. This is an unrealistic assumption about
the behaviour of yields. (Käppi, 1997, p. 35)

The most problematic point in the Vasicek model is that it can produce negative interest
rates. This isn´t a big problem with real interest rates since they often can be negative
but nominal ones are unlikely to be negative. In the long run it is useful to point out that
negativity isn´t a large problem usually since the distribution of the short rate is

12
 s2
stationary and Gaussian with expected rate and variance (when a> 0) of g and
2a
respectively. (Xie, 2006, p. 1; Brigo et al, 2007, p. 29) These values can be derived from

Er (T ) = g + ( r (t ) - g )e-a (T -t ) (11)
When T→∞ the second term approaches 0.
And,

s2
var r (T ) = (1 - e-2a (T -t ) ) (12)
2a
-2a (T - t )
When T→∞ the coefficient (1 - e ) approaches 1. (Vasicek, 1977, p. 9)

Another drawback is also that it hasn´t passed many empirical studies. For example
Murto (1992, p. 33-34) rejects it in his study about interest rate models using data from
the Finnish money market because of fitting errors. Käppi also finds that the single factor
Vasicek model fits the data poorly but multifactor versions do better (Käppi, 1997, p. 61).
Despite the drawbacks, the Vasicek model is very useful and widely used in practice.
(Benninga & Wiener, 1998, p. 6)

Like with all interest rate models the data is a major source of misspecification. The
sample should cover well over ten years in order to catch the effects of mean reversion.
This is a serious problem since some datasets simply don´t have such a long history.
For example the LIBOR swap data popularly used in estimations is too short due to the
fact that the market hasn´t existed that long. Regime changes, like the creation of the
European Central Bank and the harmonization of short-term interest rates across
European countries, have also ruined many national samples. (Kim & Orphanides, 2005,
p. 5-6)

13