Interest Rate Models: Vasicek

Peter Carr Bloomberg LP and Courant Institute, NYU

Based on Notes by Robert Kohn, Courant Institute, NYU

Continuous Time Finance

Lecture 10
Wednesday, March 30th, 2005

The Short Rate Dynamics

The Vasicek model describes the short rates Q dynamics by the following SDE:
drt = ( art) dt + dwt

(1)

where , a > 0, and are constants.

An explicit formula for rt: Start with:
d(eatrt) = eat drt + aeatrt dt = eat dt + eat dwt,
so:
at

e rt = r0 +

as

easdws.

e ds +
0

Simplifying:

rt = r0eat + (1 eat) +
a

Z
0

ea(ts) dws.

(2)

 Recall

rt = r0eat + (1 eat) +
a

ea(ts) dws.

(3)

As the starting time is arbitrary:

rt = rsea(ts) + (1 ea(ts)) +
a

ea(t ) dw .

(4)

(??) implies that rt is Gaussian at each t, with

expectation:

E Q[rt] = r0eat + (1 eat), and

a
variance:
Q

" Z

Var [rt] = E

2 #

t
a(ts)

dw(s)

2a(ts)

e
0

2
ds = (1e2at).
2a

Dynamics of Bond Price

We now show that the bond price is lognormally distributed in the Vasicek model:
By definition of the risk-neutral measure Q, the zero coupon bond price is:
i
h RT
Q
t r(s) ds
Pt(T ) = E e
| Ft .

(5)

From (??), (interchanging t, s)

Pt(T ) = A(t, T )eB(t,T )rt

(6)

ea(st) ds,

B(t, T ) =

and

i
R
R
tT { a (1ea(st) )+ ts ea(s ) dw( )} ds

A(t, T ) = E e

A(t, T ), B(t, T ) deterministic, rt Gaussian Pt(T ) lognormal.

Explicit Bond Pricing Formula

Can evaluate A(t, T ), B(t, T ) - see Lamberton & Lapeyre, pages 128-129.
Alternative approach: use Pt(T ) = V (t, rt), where V (t, r) solves BVP consisting of
PDE:
Vt + ( ar)Vr + 12 2Vrr = rV
subject to the final-time condition V (T, r) = 1 for all r.
Guess a solution of the form:
V (t, r; T ) = A(t, T )eB(t,T )r .

 Considered as functions of t, A(t, T ) and B(t, T ) solve the ODEs:

At AB + 12 2AB 2 = 0 and Bt aB + 1 = 0
subject to:
A(T, T ) = 1 and B(T, T ) = 0.
Get:

1
B(t, T ) = (1 ea(T t))
a

and:


A(t, T ) = exp

a 2a2


2 2
(B(t, T ) T + t) B (t, T ) .
4a

Term Structure and Volatility

Only three parameters special term structure.
P0 (T )
By definition, the initial instantaneous forward rate curve f0(T ) = lnT
.

After some calculations, in the Vasicek model, one has:




aT
r0
2 (1 eaT )2.
f0(T ) = + e
a
a
2a
Volatility of f , (t, T ) is defined by
dft(T ) = (stuff) dt + (t, T ) dwt.
ln Pt(T ) = ln A(t, T ) B(t, T )rt, implies
ft(T ) = T ln A(t, T ) + T B(t, T )rt,
Itos formula gives:
(t, T ) = T B(t, T ) = ea(T t).
7

Validity of Blacks Formula

We now show that Pt(T ) is lognormal under the forward-risk-neutral measure QT .

(The measure under which tradeables normalized by P (t, T ) are martingales.)
Already know that Pt(T ) is lognormal under the risk-neutral measure Q, but here
were interested in a different numeraire.
Change-of-numeraire in the one-factor setting:
The risk-neutral measure is associated with the risk-free money-market account
as numeraire (by definition dt = rtt dt with 0 = 1).
Say N is another numeraire, and Q is the associated equivalent martingale measure.
Only positive tradeables can be numeraires, so the risk-neutral process for N is
dNt = rtNt dt + tN Nt dwt
where tN is in general stochastic and w is a Q standard Brownian motion.
8

 Itos formula gives:



t
d
Nt

= t d(Nt1) + Nt1 dt

After some algebra:



t
d
Nt

t
Nt


=

t
t N 2
(t ) dt tN dwt.
Nt
Nt

is a Q-martingale, i.e.


t
d
Nt


=

t N
dwt
Nt t

where w is a Q-Brownian motion.

Therefore:
dwt = tN dt + dwt.

 What is the SDE for the short rate in the Vasicek model under the forward-risk-neutral
measure QT ?
Numeraire is Pt(T ) = A(t, T )eB(t,T )rt
Ito the (uusal lognormal) volatility of Pt(T ) is B(t, T ).
The preceding calculation gives:
dwt = B(t, T ) dt + dwt.
Conclusion:
drt = ( art) dt + dwt = [ art 2B(t, T )] dt + dwt,
where w is a QT standard Brownian motion.
This SDE shows that short rates are normal and bond prices are lognormal, as under
the risk-neutral measure Q.

10