The Vasicek model assumes that the short rate $y(t)$ evolves according to an Ornstein-Uhlenbeck process, capturing mean-reverting behavior. Its dynamics are given by:
$ dy= \alpha(\bar y- y) \, dt + \sigma \,dB_t$
where:
- $\alpha>0$ is the speed of the mean reversion.
- $\bar y$ is the long-term mean (level) to which $y$ reverts.
- $\sigma$ is the volatility, constant in the basic Vasicek model.
- $B_t$ is a standard [[Brownian Motion]] (under some measure, which will be risk-neutral $\mathbb Q$ once we have adjusted the drift).
## Market Price of Risk
To price interest-rate-dependent instruments, we often move from the real-world measure, to the risk-neutral measure $\mathbb Q$. This change in measure requires adjusting for the drift of $y(t)$ by the [[Bond Pricing One Factor Model#Market Price of Risk|Market Price of Risk]] $\eta$.
A common assumption is that $\eta$ is either constant or linear in the short-rate $y$.
$ \eta = c_0 +c_1 y$
Under the framework of a [[Bond Pricing One Factor Model]], the drift term in the PDE can be written generally as $f(t, y)$. From the definition, we have:
$
\begin{align}
f(t,y)&= b \eta -a \\
f(t,y)&=\sigma (c_0+c_1y)-\alpha (\bar y -y)
\end{align}
$
To simplify notation, we can change variables for $\alpha$ and $\bar y$:
$ \alpha^\prime = \alpha+ \sigma c_1, \quad \bar y^\prime = \frac{1}{\alpha^\prime}(\alpha \bar y - \sigma c_0)$
Which leads to:
$ \begin{align}
f(t,y) &= \sigma c_0 + \sigma c_1y - \alpha \bar y + \alpha y \\
&=\sigma c_0 + y(\sigma c_1+ \alpha)- \alpha \bar y \\
&=y\alpha^\prime + \sigma c_0 - \alpha \bar y \\
&=y\alpha^\prime - \bar y^\prime \alpha^\prime \\
&=\alpha^\prime(y-\bar y) \\
-f(t,y) &=\alpha^\prime(\bar y-y)
\end{align}$
>[!note:]
>So under the risk-neutral measure $\mathbb Q$, the mean reversion speed becomes $\alpha^\prime$, and the long-term mean becomes $\bar y ^\prime$.
## Linking $f(t,y)$ and the Model Parameters
In the [[Bond Pricing One Factor Model]] the PDE for a zero coupon bond $V(t,y)$ has the following form, where $f(t,y)$ is the drift term that depends on time and on the short-rate.
$
\left(\frac{\partial V}{\partial t} + \frac{b^2}{2}\frac{\partial^2 V}{\partial y^2}-Vy\right) + \underbrace{\alpha^\prime (\bar y^\prime -y)}_{-f(t,y)}*\frac{\partial V}{\partial y}= 0
$
>[!note:]
>Here $b=\sigma$ given the form of $dy$, where $\sigma$ is the pre-factor of $dB_t$.
This PDE can be solved analytically for plain-vanilla zero-coupon bonds. We can calibrate $\alpha^\prime, \bar y^\prime$ and $\sigma$ directly to market data (i.e. bond yields) and solve the PDE for $V$.
However, when pricing path-dependent derivatives, a Monte Carlo approach (simulating many paths of $y(t)$ is often necessary, which requires the integrated SDE form of $y(t)$.
## Integrating the Short-Rate SDE
Recall the original Ornstein-Uhlenbeck–type SDE:
$ dy= \alpha(\bar y- y) \, dt + \sigma \,dB_t$
Integrating $dy$ is not straight-forward as $y$ appears inside its own drift term. A standard approach for linear SDEs like this is the *exponential trick*, which involves a change of variable.
## Exponential Trick
We introduce a new process $z(t)$ via:
$ y(t) = F(t,z(t))= e^{-\alpha t} z(t)$
Since $y$ is a function of time $t$ and the stochastic variable $z$ we can apply [[Itô Processes and Itô's Lemma#Itô's Lemma for $F(t, X_t)$|Itô's Lemma]].
$ dy \equiv dF(t,z) = \frac{\partial F}{\partial t}dt+\frac{\partial F}{\partial z}dz+\frac{b^2}{2}\frac{\partial^2 F}{\partial z^2}dz$
The partial derivatives are as follows:
$ \frac{\partial F}{\partial t} = - \alpha e^{-\alpha t} z, \quad \frac{\partial F}{\partial z} = e^{-\alpha t}, \quad \frac{\partial^2 F}{\partial z^2}= 0$
Hence we solve Itô's Lemma:
$ dy=-\alpha e^{-\alpha t} z\, dt+e^{-\alpha t} \, dz$
No we can equate this with the original SDE:
$
\begin{align}
-\alpha e^{-\alpha t} z\, dt+e^{-\alpha t} \, dz &=\alpha(\bar y-y)dt + \sigma \, dB \tag{1}\\[2pt]
-\alpha e^{-\alpha t} z\, dt+e^{-\alpha t} \, dz &= \alpha \bar y \,dt -\alpha e^{-\alpha t}z \,dt + \sigma \, dB \tag{2}\\[2pt]
e^{-\alpha t} \, dz &= \alpha \bar y \,dt + \sigma \, dB \tag{3}\\[2pt]
dz &= e^{\alpha t}(\alpha \bar y \,dt + \sigma \, dB) \tag{4}
\end{align}
$
where:
- (2) Multiply out the right hand side $\alpha(\bar y -y) dt$.
- (3) Recognize that the term $(-\alpha e^{-\alpha t}z\, dt)$ cancels out on both sides.
- (4) Divide by $(e^{-\alpha t})$ which is the same as multiplying by $(e^{\alpha t})$.
Thus, the *mean-reverting drift* $\alpha(\bar{y}-y)$ transforms into a *simpler drift* term $\alpha\bar{y}$ for the $z$-process multiplied by the exponential factor $e^{\alpha t}$.
$ \boxed{dz = e^{\alpha t}(\alpha \bar y \,dt + \sigma \, dB) \vphantom{\frac{1}{1}}}$
## Integrating $z(t)$ and Recovering $y(t)$
We can now integrate $dz$ from $0$ to $t$:
$ z(t) - z(0)=\int_0^t \alpha \bar y e^{\alpha s} \, ds + \sigma \int_0^t e^{\alpha s} dB_s$
The *first term* is a standard (deterministic) integral, where we can pull the constants $\alpha, \bar y$ out of the integral and integrate the exponential function.
$ \int_0^t \alpha \bar y e^{\alpha s} ds= \alpha \bar y \int_0^t e^{\alpha s} ds=\alpha \bar y \left[\frac{1}{\alpha}e^{\alpha s} \right]_0^t =\alpha \bar y \left(\frac{1}{\alpha}e^{\alpha t}-\frac{1}{\alpha}\right) = \bar y(e^{\alpha t}-1)$
The *second term* is an Itô integral, which has a [[Gaussian Distribution|Normal Distribution]] with zero mean. Hence we can write the overall integral as:
$ z(t) = z(0)+ \bar y ( e^{\alpha t}-1)+ \sigma \int_0^t e^{\alpha s} dB_s$
Since $y(t) = e^{-\alpha t}z(t)$, we can substitute $z(t) = e^{\alpha t} y(t)$.
$ \begin{align}
e^{\alpha t} y(t) &= e^{\alpha 0} y(0)+ \bar y e^{\alpha t}-\bar y+ \sigma \int_0^t e^{\alpha s} dB_s \\[2pt]
y(t) &= y(0)\,e^{-\alpha t}+ \bar y -e^{-\alpha t}\bar y+ \sigma e^{-\alpha t} \int_0^t e^{\alpha s} dB_s
\end{align}$
Finally we obtain the closed-form solution for $y(t)$.
$ \boxed{y(t) = y(0)\,e^{-\alpha t}+ \bar y\left(1-e^{-\alpha t}\right)+ \sigma e^{-\alpha t} \int_0^t e^{\alpha s} dB_s \vphantom{\frac{\int}{\int}}}$