An Overview of the

Vasicek Short Rate Model

Nicholas Burgess
[email protected]

Original Version: 9th November 2013

Last Update: 25th November 2017

Keywords: Vasicek Model, Short Rate Models, Bond Pricing

Abstract
The Vasicek model (1977) is one of the earliest stochastic models of the term structure of interest
rates. This model, though it has it’s shortcomings, has many advantages, such as analytical
tractability and mean reversion features, and may be viewed as a short rate model template.
Several short rate models have their foundations rooted in the Vasicek model. The classical
Hull-White model (1990a), for example, is an extension of the Vasicek model with time de-
pendent parameters1 .
In the work that follows we derive the short rate implied by the Vasicek model using the integ-
rating factor method and provide an overview of this method and it’s shorthand. Secondly we
consider the model dynamics and finally we apply the model to zero-coupon bond pricing2 and
provide a detailed derivation.
Finally in reviewing the Vasicek model we outline it’s disadvantages, consider other short rate
models and look at the Hull-White extension to this model. The aim of the paper is to provide
an overview of the Vasicek model and an introduction into short rate modelling.

1 Short Rates
The short rate rt is taken to be the continually compounded annualised interest rate at which
one can borrow or lend money for an infinitesimally short period at a future time t.
1
The Vasicek model is endogenous with time homogeneous parameters i.e. constant over time.
2
Likewise coupon bearing Bonds can be priced as series of zero-coupon bonds, see Jamshidian (1989)

Electronic copy available at: https://ssrn.com/abstract=2479671

Short rate models attempt to model the term structure of instantaneous forward interest rates3 .
Such models come in two flavours, namely endogenous equilibrium models, where the short
rate term structure is a model output and exogenous arbitrage-free models, where it is an ex-
plicit input. The later is preferred and better explains the interest rate dynamics observed in the
market.
In the models that follow the short rate is stochastic and corresponds to the instantaneous for-
ward rate. By modelling the evolution of the short rate we can determine at any future time t
the price zero-coupon bond Z(t, T ) maturing at time T as


 Z T  
Z(t, T ) = E exp − rs ds |Fs
s=t

which means that instantaneous forward rates can be specified as follows

∂
f (t, T ) = − ln(Z(t, T ))
∂T

2 Vasicek Short Rate Model

The Vasicek model describes short interest rates as having the following dynamics

drt = (θ − art )dt + σdBt (1)

which can also be represented as

drt = a(b − rt )dt + σdBt (2)

where
a = Speed of Mean Reversion, 0 ≤ a ≤ 1
b = Mean Reversion Level
rt = Short Rate at time, t
σ = Short Rate Volatility
Bt = Brownian Motion Process at time, t

and
θ = ab
The Vasicek Model’s drift term exhibits mean reversion, meaning that over time interest rates
will converge to the mean reversion level b with speed a. The b parameter can be thought of
as the long term interest rate level. When the short rate wanders above the long term rate it is
pulled down and likewise when it drifts below the long term rate it is pushed up.
3
In the discrete case the short rate would correspond to the borrowing or lending rate of a discrete period. For
example a discrete process with daily time steps the short rate would be a daily interest rate and may be taken to
be the liquid overnight interest rate.

Electronic copy available at: https://ssrn.com/abstract=2479671

Figure 1 demonstrates mean reversion manifesting itself as strong resistance and support from
which the short rate will bounce to return and converge to the long term interest rate level.

Figure 1: Vasicek Mean Reversion

The mean reversion for the Vasicek model is captured within the drift term drt = a(b − rt ),
where a and b are positive constants. Mathematically we demonstrate this below in table 1.
When the short rate rt drifts above (below) the long term rate b then the change in the Vasicek
process is negative (positive) forcing convergence to the long term interest rate level.

Short Rate, rt Rate Change, drt

rt > b drt = a(b − rt ) < 0

Short Rate is greater than Long Term Rate Adjust Short Rate Process Downwards

rt < b drt = a(b − rt ) > 0

Short Rate is less than Long Term Rate Adjust Short Rate Process Upwards

Table 1: Mean Reverting Drift: drt = a(b − rt )

The Vasicek model when used to evaluate yield curves can fit term structures that are increasing,
decreasing or humped, as shown in figure 2.

3
 Figure 2: Vasicek Yield Curve

3 Vasicek Solution for the Short Rate

To solve the SDE described in (1) for rt we use the the Integrating Factor Method, outlined in
detail in section 4 below, whereby (1) is rearranged into the form

dXt + Pt Xt dt = dBt (3)

R
Pt dt
The Integrating Factor It is set to be It = e . Multiplying both sides of (3) by It and
integrating leads to an expression for rt .
Applying the Integrating Factor method to the Vasicek process we obtain

drt + art dt = θdt + σdBt (4)

R
adt
Setting It = e = eat , since a is a constant and multiplying both sides by It gives

I dr + I ar dt = It θdt + It σdBt (5)

|t t {z t t }
d(It rt )

The Integrating Factor Method says that d(It X) + It Pt Xdt can be reduced to d(It Xt ), which
can be confirmed by applying Itô’s Lemma. Applied to (5) the left-hand side becomes

d(It rt ) = It θdt + It σdBt

4
 d(eat rt ) = eat θdt + eat σdBt

Integrating over ( s, t ), where 0 ≤ s ≤ t yields

Z t Z t
at as au
e rt − e rs = e du + σ eau dBu (6)
u=s u=s

Z t
θ
at as
e rt − e rs = (eat − eas ) + σ eau dBu
a u=s

Collecting rt terms we obtain

Z t
−a(t−s) θ
rt = e rs + (1 − e−a(t−s) ) + σ e−a(t−u) dBu (7)
| a{z } | u=s
{z }
Drif t Dif f usion

This is exactly what is required, namely an expression for the short rate at a future time t.

4 Solving SDEs using the Integrating Factor Method

First Order SDEs, much like First Order ODEs, can be solved using the Integrating Factor
Method outlined below.

4.1 Integrating Factor Method

1. Collect terms for dXt and dt and rearrange the SDE into the form
dXt + Pt Xt dt = dBt

2. Set the Integrating Factor R

Pt dt
It = e

3. Multiply both sides by It giving

It dXt + It Pt Xt dt = It dBt

4. From Itô’s Lemma the left-hand side becomes

It dXt + It Pt Xt dt = d(It Xt )

5. This leads to
d(It Xt ) = It dBt

6. Simple integration can now be used to obtain a result for Xt

5
4.2 Integrating Factor Shorthand
It is common practice within academic literature to jump straight to step 5 above when solving
first order SDEs, which for the Vasicek Model would be as follows
drt = (θ − art )dt + σdBt
Using Integrating Factor, It = eat we arrive at
d(eat rt ) = θeat dt + σeat dBt
Integrating over ( s, t ), where 0 ≤ s ≤ t
Z t
at as θ at
e − eas + σ eau dBu


e rt − e rs =
a u=s

Rearranging for rt
Z t
−a(t−s) θ
1 − e−a(t−s) + σ e−a(t−u) dBu


rt = e rs +
a u=s

5 Vasicek Model Dynamics

Inspecting (7) we observe that the expression for rt is made up of two deterministic terms
representing the drift and a stochastic integral representing the diffusion of our process, from
which we can deduce the first and second conditional moments.
From the drift terms
θ
E[rt |Fs ] = e−a(t−s) rs + (1 − e−a(t−s) ) (8)
a
and from diffusion term
 Z t 
−a(t−u)
V ar(rt |Fs ) = V ar σ e dBu
u=s
" Z 2 # 2
t (9)
 Z t
−a(t−u) −a(t−u)
=E σ e dBu |Fs −E σ e dBu |Fs
u=s u=s
| {z }
=0

The second term is zero, since the expectation of a stochastic integral is zero by definition
" Z 2 #
t
V ar(rt |Fs ) = E σ e−a(t−u) dBu |Fs
u=s
Z t
= σ2 e−2a(t−u) du
u=s (10)
−2a(t−u) t
 
e
= σ2
2a u=s
2
σ
1 − e−2a(t−s)


=
2a
6
Note we dropped the expectation operator E[.] whilst applying Itô Isometry, where dBu2 = du,
since there is no longer any source of randomness.
Now since rt is a Brownian Motion we know it is normally distributed as follows
rt ∼ N (E[rt |Fs ], V ar (rt |Fs ))
2 (11)
 
−a(t−s) θ −a(t−s) σ −2a(t−s)


rt ∼ N e rs + (1 − e , 1−e
a 2a

In the limit as t → ∞ we observe

θ
lim E[rt |Fs ] =
t→∞ a
and
σ2
lim V ar(rt |Fs ) =
t→∞ 2a
Hence in the limiting case the Vasicek model has the following dynamics
θ σ2
 
rt ∼ N , (12)
a, 2a

θ


It can be seen that the expected value tends to , which equals b, namely the long term rate
a 
2
or mean reversion level and the variance tends to σ2a , which is the model variance scaled by
the speed of mean reversion.

6 Zero Coupon Bond Pricing

In what follows we describe how to price a zero-coupon bond as at a future time t, which enables
us to price a variety of linear and non-linear products. We also draw the reader’s attention to
the fact that coupon bearing bonds, and for that matter any cashflow, can be decomposed into a
series of zero-coupon bonds, see Jamshidian (1989).
To evaluate the price at a future time t of a zero-coupon bond Z(t,T) maturing at time T, where
0 ≤ t ≤ T , we consider the expected value of the short rate over time
RT
Z(t, T ) = E[e− u=t ru du
|Ft ] (13)

We know from the model dynamics that the process is normally distributed and recall from the
definition of the normal moment generating function
1
E(eX ) = eE[X]+ 2 V ar(X) (14)

Applying the normal moment generating Function to (13)

RT RT
du|Ft ]+ 12 V ar(−
Z(t, T ) = eE[− u=t ru u=t ru du|Ft )
(15)

7
Evaluating the E[.] and V ar(.) terms will lead to an expression for the bond price of the form
Z(t, T ) = A(t, T )e−rt B(t,T ) (16)

To derive a closed form solution to equation (15) we decompose the problem. Firstly we evalu-
ate the integral, secondly the expectation term and finally the variance term.

6.1 Zero Coupon Bond, Integral Term

Evaluating the integral term of (15) using the short rate rt derived in (7) and paying attention to
the relabelling of integration bounds within (7) from ( s, t ) to ( t, T )
Z T Z T
θ T
Z Z T Z u
−a(u−t) −a(u−t)
e−a(u−s) dBs du


ru du = rt e du + 1−e du + σ
u=t u=t a u=t u=t s=t
−a(u−t) T −a(u−t) T Z T Z u
   
rt e θ e
= − + u+ +σ e−a(u−s) dBs du
a u=t a a u=t s=t
 −a(T −t)
   u=t −a(T −t)  Z T Z u
1−e θ 1−e
= rt + (T − t) − +σ e−a(u−s) dBs du
a a a u=t s=t
(17)

6.2 Zero Coupon Bond, Expectation Term

The derivation of the conditional expectation term within (15) follows from (17). The expecta-
tion of the stochastic integral is zero and therefore after observing the negative sign within the
expectation
 Z T
1 − e−a(T −t) 1 − e−a(T −t)
     
θ
E − ru du|Ft = −rt − (T − t) − (18)
u=t a a a

6.3 Zero Coupon Bond, Variance Term

Likewise the conditional variance term required within (15) follows from (17), where the vari-
ance is determined from the diffusion term, namely the stochastic integral, since the two de-
terministic drift terms make no contribution to the variance. This gives
 Z T   Z T Z u 
−a(u−s)
V ar − ru du|Ft = V ar −σ e dBs du
u=t u=t s=t

Knowing V ar(X) = E[X 2 ] − E[X]2 and that V ar(−X) = V ar(X)

 Z T  Z T 
V ar − ru du|Ft = V ar ru du
u=t u=t
" Z Z
T u 2 #  Z T Z u 2
−a(u−s) −a(u−s)
=E σ e dBs du −E σ e dBs du
u=t s=t u=t s=t
| {z }
=0

8
Observe that the expected value of the second stochastic integral is zero by definition and there-
fore
2
" Z Z #
 Z T  T u
−a(u−s)
V ar − ru du|Ft =E σ e dBs du
u=t u=t s=t

Using Fubini’s Theorem to switch the integration order allows us reduce the remaining stochastic
integrand to a deterministic term, since via Itô’s Isometry dBs2 = ds. Careful attention is paid
to the upper bound s = u within the inner integral which requires a bounds transformation4 and
this gives
 2 
 Z T   Z T Z T  
V ar − ru du|Ft = E σ e−a(u−s) du dBs  (19)
  

u=t
| s=t{z u=s}
  
BoundsChange

The square term eliminates the source of randomness, since dBs2 becomes ds, and hence the
expectation operator is no longer required
 Z T  " Z Z
T T 2 #
V ar − ru du|Ft = E σ 2 e−a(u−s) du ds
u=t s=t u=s
2 (20)
Z T Z T
−a(u−s)
= σ2 e du ds
s=t u=s

Continuing with simple integration leads to

T T 2
1 − e−a(T −s)
 Z  Z 
2
V ar − ru du|Ft =σ ds
u=t s=t a
Z T 
1 − 2e−a(T −s) + e−2a(T −s)

2
=σ ds
s=t a2
T
σ2 2e−a(T −s) e−2a(T −s)

= 2 s− +
a a 2a s=t

giving
T
σ2 1 − e−a(T −t) 1 − e−2a(T −t)
 Z      
V ar − ru du|Ft = 2 (T − t) − 2 + (21)
u=t a a 2a
4
Fubini’s Theorem allows the change of integration order for double and multiple stochastic integrals, however
when the integration bounds are a function of one-another a bounds transformation is required.

9
6.4 Fubini’s Theorem and Change of Integration Bounds

Fubini’s Theorem provides the mechanism to alternate the order of integration of double and
higher order stochastic integrals. This provides flexibility to decide which stochastic integral to
process first, which was helpful in determining the zero-coupon bond variance term in (19). It
allowed us to first apply Itô’s Isometry to square the stochastic term dBs making it deterministic
ds and thus simplifying the calculation.
Applying Fubini’s Theorem is straightforward when we are dealing with constant integration
limits, however some thought is required if integration bounds are a function of one-another. In
such cases we follow the below steps.

Bounds Transformations, Step by Step

Consider the following double integral where we wish to apply Fubini’s Theorem to reverse the
integration order.
Z x=b Z y=f (x)
[.] dy dx
x=a y=c

1. Plot the area represented by the original integration bounds, ignoring the function being
integrated.

2. First start with the inner integral and chart the area bounded by the limits of the inner
integral. Always plot functional bounds on the y-axis. In this case we plot y = f (x) on
the y-axis.

3. Next chart the area bounded by the outer integral’s limits on the same chart, giving the
net area captured by the limits of the inner and outer integrals.

4. Replot the same area, reflected in the y = x axis, this is equivalent switching the integ-
ration order. Again we do this in stages, first for the new inner and then the new outer
integral, inverting any functional limits.

5. Swap the axis of the original chart and visualise the region captured by the original integ-
ration limits reflected in the y = x axis.

6. Next plot the new y limits that capture the same area as the original chart, but reflected in
y = x axis. Remember to invert any functional limits, in this case x = f −1 (y).

7. Repeat the process for the new x limits, again inverting any functional limits.

8. Finally read the new integration bounds from the resulting bounds chart, in the correct
order, starting with the inner then outer integral.

10
Bounds Transformations, Example

Next we outline the above step by step process providing a graphical representation of the
bounds in equation (19), namely Z T Z u
[.] ds du
u=t s=t

Following the steps outlined above, we plot the area captured by the original integral. First by
charting the inner integral limits in s on the y-axis, to allow us to plot s = u, then the outer
integral limits in u on the x-axis. It can be seen that the bounds capture the shaded area in figure
4 below.

Figure 3: Original Inner bounds Figure 4: Original Outer bounds

Next we follow the steps outlined above for the reflection in y = x. We reflect the image from
figure 4 in the y = x axis, here the s = u axis. We swap the x and y-axis ( here s and u ) and
invert all functional limits, so in this case s = u becomes u = s.
We infer the limits that create the area required first starting with the new inner integral and then
the outer one. This allows us to observe the bounds transformation and read off the correspond-
ing values, see figure 6 below. Note we have deduced the s = T via reflection in s = u.
This leads to the bounds transformation as shown in equation (19)
Z T Z u Z T Z T
[.] ds du −→ [.] du ds
u=t s=t s=t u=s

6.5 Zero Coupon Bond Pricing

Returning to the Zero Coupon Bond Pricing formula (13) having derived all the terms required
RT RT
du|Ft ]+ 21 V ar(−
Z(t, T ) = eE[− u=t ru u=t ru du|Ft )

11
 Figure 5: Transformed Inner bounds Figure 6: Transformed Outer bounds

Substituting the conditional expectation (18) and variance (21)

          
−a(T −t) −a(T −t) σ2 −a(T −t) −2a(T −t)
−rt 1−e a − aθ (T −t)− 1−e a + (T −t)−2 1−e a + 1−e 2a
2a2
Z(t, T ) = e
(22)
Factorize the common factor in (22) by setting

1 − e−a(T −t)
 
B(t, T ) = (23)
a

Substituting (23) into (22)

   
σ2 −2a(T −t)
−rt B(t,T )− aθ ((T −t)−B(t,T ))+ (T −t)−2B(t,T )+ 1−e 2a
2a2
Z(t, T ) = e (24)

 
1−e−2a(T −t)
Evaluating 2a
in terms of B(t, T )

1 − e−a(T −t)
 
B(t, T ) =
a
−a(T −t)
2 1 − 2e + e−2a(T −t)
⇒ B(t, T ) =
a2
−2a(T −t)
⇔e = a B(t, T ) − 1 + 2e−a(T −t)
2 2

12
Therefore
1 − e−2a(T −t) 1 − a2 B(t, T )2 + 1 − 2e−a(T −t)
   
=
2a 2a
2 − 2e−a(T −t)
 2
a B(t, T )2
  
= −
2a 2a
−a(T −t)
  2 (25)
a B(t, T )2
 
1−e
= −
a 2a
2
 
aB(t, T )
= B(t, T ) −
2

Substituting (25) into (24)

   
σ2 aB(t,T )2
−rt B(t,T )− aθ ((T −t)−B(t,T ))+ (T −t)−2B(t,T )+B(t,T )−
2a2 2
Z(t, T ) = e
    (26)
σ2 aB(t,T )2
−rt B(t,T )− aθ ((T −t)−B(t,T ))+ (T −t)−B(t,T )−
2a2 2
=e

Collecting (T − t) and B(t, T ) terms

   2 
σ2 σ aB(t,T )2

−rt B(t,T )−((T −t)−B(t,T )) aθ + −
2a2 2
4a
Z(t, T ) = e
   2 
σ2 σ B(t,T )2

−rt B(t,T )+(B(t,T )−(T −t)) aθ − 2 − 4a (27)
2a
=e
   2 
2 σ B(t,T )2

(B(t,T )−(T −t)) aθ − σ 2 −
= e−rt B(t,T ) e 2a 4a

Hence we have a solution for Z(t, T ) of the desired functional form, namely

Z(t, T ) = A(t, T )e−rt B(t,T ) (28)

where    2 
2 σ B(t,T )2

θ
(B(t,T )−(T −t)) a
− σ2 − 4a
2a
A(t, T ) = e
and
1 − e−a(T −t)
 
B(t, T ) =
a

7 Disadvantages of the Vasicek Model

The Vasicek Model has several disadvantages which we list below.

1. It is endogenous with time homogeneous parameters that do not vary over time. As such
short rate term structure is an output not an input.

13
 2. Vasicek short rates can be negative for certain combinations of model parameters. Al-
though negative interest rates have indeed been observed in Non-Deliverable Forwards,
in Switzerland & Japan. This is neither typical nor as frequent an observation as implied
by this model.
3. There is no term structure of volatility. Volatility is assumed constant5 .
4. Some observed short rate term structures are difficult and sometimes impossible to fit
with this model.
5. It is a one factor model. It assumes that short rates are 100% correlated and can only cater
for parallel shifts in the yield curve.

8 Other Short Rate Models

Other short rate models were developed, both endogenous and exogenous to address the short-
comings of the Vasicek and other preceding short rate models. Consequently there are several
short rate models which have similar but different functional forms, some of which are listed in
table 2.
Model Descritpion
√
drt = (θ − art )dt + rt σdBt Cox, Ingersoll and Ross (1985)
√
Mean reverting drift and rt scaled diffusion

drt = θt dt + σdBt Ho-Lee (1986)

Time dependent drift with no mean reversion

drt = (θt − at rt )dt + σt dBt Hull-White (1990)

Mean reverting drift and time dependent parameters

dln(rt ) = θt dt + σdBt Black, Derman and Toy (1990)

Log normal with time dependent drift but no mean reversion

drt = (θt − aln(rt ))dt + σt dBt Black Karasinski (1991)

Mean reverting drift and log normal

Table 2: Short Rate Models

The Vasicek model provided a foundation for these models. For example scaling the Vasicek
√
diffusion term by rt gives the Cox, Ingersoll and Ross model, setting the Vasicek a parameter
and making θ time dependent to zero leads to the Ho-Lee Model. Likewise making the all the
Vasicek parameters time dependent leads to the Hull-White (1990a) model.
5
Despite the model volatility parameter being constant, the model dynamics in equation (11) show the variance
2
is monotonically decreasing with time and equal to σ2a 1 − e−2a(t−s) .

14
9 The Extended Vasicek / Hull-White Model

The Hull-White (1990a) Model below, is also known as the Extended Vasicek Model. It is in
essence the very same model, but with time dependent parameters.

drt = (θt − at rt )dt + σt dBt

The Hull-White Model appears in it’s most generic case with the θt , at and σt parameters being
deterministic and time dependent, however it is not unusual to see the Hull White Model with
any combination of these parameters fixed. The Hull-White model can be implemented as a
series of discrete Vasicek models. The extended Vasicicek model attempts to deal with the
disadvantages highlighted in section 7.
Hull & White developed this model further into a two factor model. This was due to one
factor models only catering for short rates that are 100% correlated. This in turn only allows
for parallel moves resulting yield curves. The two factor Hull-White model has a stochastic
mean reversion term, giving the model richer explanatory power. The 2 factor model process is
outlined below.
dr(t) = (θ(t) + u(t) − ar(t))dt + σ1 dB1 (t)
du(t) = −bu(t) + σ2 dB2 (t)

with
dB1 (t)dB2 (t) = ρdt

Conclusion

In summary the short rate is the instantaneous forward rate at which one can borrow or lend
money for an infinitesimally short period at a future time t. Short rate models can be endogenous
or exogenous and the term structure of rates can be a model output or input respectively. The
later is preferred and a better fit of market rates.
The Vasicek model and it’s mean reversion feature were examined, as well as the possible
functional forms of the resulting yield curve. The short rate was derived from the Vasicek
SDE using the integrating factor method and model dynamics were considered with particular
interest in the limiting case.
The closed form solution for zero-coupon bond prices was derived and the procedure for the
change of integration bounds when applying Fubini’s theorem outlined. Finally the model
shortcomings were discussed and other short rate models considered, including the Hull-White
extension to the Vasicek model, which were aimed at addressing these shortcomings.
It is hoped that this paper has provided a helpful overview of the Vasicek and short rate models.

15
Appendix

This paper comes supplied with an excel support file6 , a snapshot of which is below.

Figure 7: Excel Support File

References

Hull J and White A (1990a) ”Pricing Interest Rate Derivative Securities” Review of Financial
Studies 3(4), 573-92
Hull J and White A (1993) ”One-Factor Interest Rate Models and the Valuation of Interest Rate
Derivative Securities” Journal of Finance and Quantitative Analysis 28, 235-254
Jamshidian, F (1989) ”An Exact Bond Option Formula,” Journal of Finance 44, 205-9
Vasicek, O (1977) ”An Equilibrium Characterization of the Termstructure,” Journal of Financial
Economics 5, 177-88

6
Kindly email the author to request a copy.

16