Suppose that the short rate follows the process $$dr(t) = a(t, r(t))dt + \sigma(t, r(t))dW(t)$$

If $B(t) = exp(-\int_0^t r(u) d u)$, can one still write the differential $dB(t)$ a-la-Ito? Thanks.


I assume you are interested in the bond price. Let $B_t=B(t,r_t)=\mathbb{E}^\mathbb{Q}[\exp\left(-\int_t^T r_u\mathrm{d}u\right)\mid\mathcal{F}_t]$ be the time $t$ price of a default-free zero-coupon bond maturing at $T$.

That is, by the way, very different to $\exp\left(-\int_0^t r(u)\mathrm{d}u\right)$ which relates to a money-market account (bank account, savings account) $M_t=\exp\left(\int_0^t r(u)\mathrm{d}u\right)$, where $\mathrm{d}M_t=r_tM_t\mathrm{d}t$. The bond price has to have an expectation around the exponential! Remember that the bond price, at time $t$, is a real number, the value of your bank account at time $t$ is a random variable!

A priori, all you can do with $B_t=B(t,r_t)$ is to write \begin{align*} \mathrm{d}B_t = \underbrace{\left(\frac{\partial B_t}{\partial t} + a(t,r_t)\frac{\partial B_t}{\partial r_t}+\frac{1}{2}\sigma^2(t,r_t)\frac{\partial^2B_t}{\partial r_t^2}\right)}_{\mu_B(t,r_t)}\mathrm{d}t+\underbrace{\sigma(t,r_t)\frac{\partial B_t}{\partial r_t}}_{\sigma_B(t,r_t)}\mathrm{d}W_t \end{align*} That is what Itô's Lemma gives you.

Allow me to make two points:

  • Using a dynamic hedge and the Black-Scholes line of argument, you can find a second order PDE for the bond price. You need to be careful though because the ``underlying'', the short rate, is not a traded asset.

  • Many popular short rate models (e.g. Vasicek, Hull-White, CIR) are affine term-structure (ATS) models meaning that $B_t=e^{A_t+C_tr_t}$, where $A_t$, $C_t$ are deterministic functions of time. Then, of course, \begin{align*} \frac{\partial B_t}{\partial t} &= \left(\frac{\partial A_t}{\partial t}+\frac{\partial C_t}{\partial t}r_t\right)B_t \\ \frac{\partial B_t}{\partial r_t} &= C_tB_t,\\ \frac{\partial^2 B_t}{\partial r_t^2} &= C_t^2B_t. \end{align*} This allows you to simplify the equation for $\mathrm{d}B_t$, \begin{align*} \mathrm{d}B_t = \left(\frac{\partial A_t}{\partial t}+\frac{\partial C_t}{\partial t}r_t + a(t,r_t)C_t+\frac{1}{2}\sigma^2(t,r_t)C_t^2\right)B_t\mathrm{d}t+\Big(\sigma(t,r_t)C_t\Big)B_t\mathrm{d}W_t \end{align*}

  • $\begingroup$ I have a problem with the formula you propose: my understanding, is that Ito can be applied to something like $f\circ (id \times r)$, i.e. compute the differential of f(t, r(t)). Now, the function B seems to depend on both t and $r(s)$ for all $s$ between $0$ and $t$... so it is not really clear to me what does $\partial_r B$ mean... $\endgroup$ – desmodena Apr 30 '20 at 16:59
  • $\begingroup$ @mallapazzo are things clearer now? The bond price is a function of time and the short rate, i.e. $B_t=B(t,r_t)$. And we can apply Itô's Lemma to this function in the usual way. $\endgroup$ – Kevin Apr 30 '20 at 18:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.