9
$\begingroup$

I'm looking for a more general Feynman-Kac formula that works in the case of jump-diffusion processes. I know that, given a pure diffusion process like $$dS_t=\mu_tdt+\sigma_tdW_t,$$ if $u(t,s)$ satisfies the PDE $$f_t(t,s)+\mu_tf_s(t,s)+\frac{\sigma_t^2}{2}f_{ss}(t,s)-V(s)f(t,s)=0$$ with terminal condition $f(T,s)=H(S_T)$ then $u$ is of the form $$u(t,s)=\mathbb{E}\big[e^{-\int_t^T{V(S_x)dx}}H(S_T)|S_t=s\big].$$ Is it possible to extend such a result when the process dynamics are given by $$dS_t=\mu_tdt+\sigma_tdW_t+\gamma_tdN_t$$ with $N$ a Poisson process independent fom $W$?

$\endgroup$

1 Answer 1

7
$\begingroup$

Hi it is possible to get some Feynman-Kac formula in this case. The proof only use the martingale property and Itô's formula for jump-diffusion processes.

So let's have $X$ s.t. (I took the compensated version of your sde):

$dX_t=[\mu(t,X_t)+\lambda(t)\gamma(t,X_t)]dt + \sigma(t,X_t)dW_t+ \gamma(t,X_{t-})d\tilde{N}_t$ where $\tilde{N}_t$ is a compensated Poisson process of intensity $\lambda(t)$.

Please notice the explicit dependence in $t$ and $X_t$ of the above equation that is necessary to have Markov property for the solution which is necessary for the Feynman-Kac theorem to apply.

Now let's us be given $F(t,X_t)=e^{-\int_s^t V(X_r)dr}u(t,X_t)=e^{-IV(s,t)}.u(t,X_t)$ and apply Itô to this formula. You get :

$$dY_t=dF(t,X_t)=e^{-IV(s,t)}\Big(\big(\partial_t u(t,X_t)+\lambda(t)[u(t,X_t+\gamma(t,X_t))-u(t,X_t)]+\mu(t,X_t)\partial_x u(t,X_t)+\frac{\sigma^2(t,X_t)\partial_{xx}u(t,X_t)}{2}-V(t,X_t).u(t,X_t)\big)dt+ (u(t,X_t+\gamma(t,X_t))-u(t,X_t))d\tilde{N}_t+(\sigma(t,X_t)\partial_{x}u(t,X_t))dW_t\Big)$$

Now if the $dt$ term is null then $Y_t$ is martingale and for $t=T$ : $$Y_s=F(s,X_s=x)=E[Y_T|X_s=x]=E[e^{-\int_s^T V(X_r)dr}H(X_T)|X_s=x]$$

So in this case the PIDE that solves the Feynman-Kac formula is : $$\partial_t u(t,X_t)+\mu(t,x)\partial_x u(t,x)+\frac{\sigma^2(t,x)}{2}\partial_{xx}u(t,x)+\lambda(t)[u(t,x+\gamma(t,x))-u(t,x)]=V(t,x)u(t,x) $$

With final condition $u(T,x)=H(x)$

Best regards

$\endgroup$
1
  • $\begingroup$ @ Laura: You can extend the result if $N$ is a compound Poisson process, and most likely for general Lévy process. Regards. $\endgroup$
    – The Bridge
    Aug 28, 2013 at 14:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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