I would like to establish the equations of forward libors under risk neutral measure. Here is how I do it, and what I get :

Under the $P_{T_j} $ measure, forward Libor $L_j$ is martingale. Thus:

$$ dL_j = L_j \times \sigma_j(t) \times dW^{T_j} $$

The change of numéraire implies that:

$$ \frac{dQ^{T_j}}{dQ^*} = \frac{P(t,T)\times\exp{\left(-\int_0^T{r(s)ds}\right)}}{P(0,T)} $$

Under $Q^*$, $P(t,T)$ discounted is martingale, which means that:

$$ \frac{dP(t,T)}{P(t,T)} = r(t) dt + \eta(t) dW^* $$

Solving this gives:

$$ P(t,T) = P(0,T) \times \exp\left(\int_0^T\left(r(s)-\frac{1}{2} \eta(s)^2\right)ds + \int_0^T{\eta(s)}dW^{T_j}\right) $$


$$ \frac{dQ^{T_j}}{dQ^*} = \exp\left(-\int_0^T{\frac{1}{2}} \eta(s)^2ds + \int_0^T{\eta(s)}dW^{T_j}\right) $$

Girsanov shows that:

$$ dW^{T_j} = dW^* - \eta(t) dt $$

Under $Q^*$:

$$ \frac{dL_j}{L_j} = \alpha(t) dt + \sigma_j(t) dW^{*} $$

Writing $P(t,T) $ as $\exp(-\int_t^T{f(t,s)ds}) = \exp(Y_t)$:

$$ \frac{dP(t,T)}{P(t,T)} = dY_t +\frac{1}{2}<Y_t>dt $$

with $ dY_t = f(t,t) dt -\int_t^T{\alpha(t) dt ds } - \int_t^T{\sigma(t)dt dW_s}$. Identifying, I conclude that:

$$ \eta(t) = \sigma(t) $$


$$ dL_j(t) = -L_j(t) \sigma_j(t) \int_t^{T_j}{\sigma_j(s)ds} dt + L_j\sigma_j(t) dW^*.$$

Is it correct ?

  • $\begingroup$ It is not clear why $dY_t = f(t,t) dt -\int_t^T{\alpha(t) dt ds } - \int_t^T{\sigma(t)dt dW_s}$. Note that $f(t, T)$ is the instantaneous forward rate, while $L_j$ is a forward rate over a given time period, for example, $[T_{j-1}, \, T_j]$. $\endgroup$ – Gordon Sep 7 '18 at 18:02

We assume that, under the $T_j$-forward probability measure $P_{T_j}$, \begin{align*} \frac{dP(t, T_j)}{P(t, T_j)} = \mu_P(t, T_j) dt + \sigma_P(t, T_j) dW_t^{T_j}, \end{align*} where $\mu_P(t, T_j)$ and $\sigma_P(t, T_j)$ are the respective drift and volatility functions. Let $Q$ be the risk-neutral probability measure. Then \begin{align*} \frac{dQ}{dP_{T_j}}\big|_t &= \frac{e^{\int_0^t r_s ds}P(0, T_j)}{P(t, T_j)}\\ &=e^{\int_0^t \big(r_s -\mu_P(s, T_j)+\frac{1}{2} \sigma_P(s, T_j)^2 \big) ds - \int_0^t \sigma_P(s, T_j) dW_s^{T_j}}. \end{align*} Since $\frac{dQ}{dP_{T_j}}\big|_t$ is a martingale under $P_{T_j}$, \begin{align*} \int_0^t \Big(r_s -\mu_P(s, T_j)+\frac{1}{2} \sigma_P(s, T_j)^2 \Big) ds = -\frac{1}{2}\int_0^t \sigma_P(s, T_j)^2 ds. \end{align*} That is, \begin{align*} \mu_P(t, T_j) = r_t + \sigma_P(t, T_j)^2, \end{align*} and \begin{align*} \frac{dQ}{dP_{T_j}}\big|_t &= e^{-\frac{1}{2} \int_0^t\sigma_P(s, T_j)^2 ds - \int_0^t \sigma_P(s, T_j) dW_s^{T_j}}. \end{align*} Then, under the risk-neutral probability measure $Q$, $\{W_t, \, t \ge 0\}$, where, for $t \ge 0$, \begin{align*} W_t = W_t^{T_j} + \int_0^t \sigma_P(s, T_j) ds, \end{align*} is a standard Brownian motion. Moreover, \begin{align*} \frac{dL_j}{L_j} = -\sigma_j(t) \sigma_P(t, T_j) dt + \sigma_j(t) d W_t. \end{align*}

| improve this answer | |

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.