I have the following Schwartz model: $$dS_t=a(\mu-\ln S_t)S_tdt+\sigma S_tdW_t$$ $$X_t=\ln S_t$$ $$dX_t=a(\hat{\mu}-X_t)dt+\sigma dW_t$$ with $\hat{\mu}=\mu-\frac{\sigma^2}{2a}\sigma$ $$F_t(T)= \exp\left(e^{-a(T-t)}X_t+\hat{\mu}(1-e^{-a(T-t)})+\frac{\sigma^2}{4a}(1-e^{-2a(T-t)})\right)$$ and I want to derive the value of $d \ln F(t,T)$

I do the following:

Let $Y_t=\ln F(t,T)$

Then $dY=\frac{dF}{F}-\frac{1}{2}\frac{<dF,dF>}{F^2}$

We have $\frac{dF}{F}=e^{-a(T-t)} dX_t=e^{-a(T-t)}\left(a(\hat{\mu}-\ln S_t)dt+\sigma dW_t\right)$

and $\frac{<dF,dF>}{F^2}= e^{-2a(T-t)}\sigma^2dt$

So that $d\ln F(t,T)=e^{-a(T-t)}\left(a(\hat{\mu}-\ln S_t)dt+\sigma dW_t\right)-\frac{1}{2}e^{-2a(T-t)}\sigma^2dt$

However I believe that this is wrong as the answer given in correction is: $$ d\ln F(t,T) = e^{-a(T-t)}\sigma dW_t-\frac{1}{2}e^{-2a(T-t)}\sigma^2dt$$

This means that there is probably a mistake in my $\frac{dF}{F}$, can you help me?


There might be a misuse of your notations. You had better define another $g=g(t,x;T)$, where $t$ and $x$ are unknowns of $g$, while $T$ is some auxiliary parameter, such that $F_t(T)=g(t,X_t;T)$. The notation $F_t(T)$ appears somewhat confusing, because you might ignore the fact that it is also a function of $t$.

Therefore, define $$ g(t,x;T)=\exp\left(e^{-a\left(T-t\right)}x+\hat{\mu}\left(1-e^{-a\left(T-t\right)}\right)+\frac{\sigma^2}{4a}\left(1-e^{-2a\left(T-t\right)}\right)\right), $$ and it is obvious that $$ F_t(T)=g(t,X_t;T). $$

Now, note that $$ Y_t=\log F_t(T)=\log\left(g(t,X_t;T)\right)=\left(\log\circ g\right)(t,X_t;T), $$ where the composite function $\log\circ g$ depends on both $t$ and $x$. Thus It's formula should be $$ {\rm d}Y_t=\frac{\partial\left(\log\circ g\right)}{\partial t}(t,X_t;T){\rm d}t+\frac{\partial\left(\log\circ g\right)}{\partial x}(t,X_t;T){\rm d}X_t+\frac{1}{2}\frac{\partial^2\left(\log\circ g\right)}{\partial x^2}(t,X_t;T){\rm d}\left<X\right>_t. $$ It seems that you have left out the $$ \frac{\partial\left(\log\circ g\right)}{\partial t}(t,X_t;T){\rm d}t $$ term.

By the way, I am wondering why you chose to deal with $F_t(T)$ directly. Note that $$ \log F_t(T)=e^{-a\left(T-t\right)}X_t+\hat{\mu}\left(1-e^{-a\left(T-t\right)}\right)+\frac{\sigma^2}{4a}\left(1-e^{-2a\left(T-t\right)}\right). $$ If you choose to define $$ Z_t=\log F_t(T)=e^{-a\left(T-t\right)}X_t+\hat{\mu}\left(1-e^{-a\left(T-t\right)}\right)+\frac{\sigma^2}{4a}\left(1-e^{-2a\left(T-t\right)}\right), $$ then $$ {\rm d}\log F_t(T)={\rm d}Z_t={\rm d}\left(e^{-a\left(T-t\right)}X_t+\hat{\mu}\left(1-e^{-a\left(T-t\right)}\right)+\frac{\sigma^2}{4a}\left(1-e^{-2a\left(T-t\right)}\right)\right), $$ which is much easier to deal with.


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.