Stochastic Calculus

Lesson 11 of 11

Feynman-Kac

Why an equation with nothing random in it computes an expected payoff: a fair game can’t drift.

The idea

Let dX=β dt+γ dWdX = \beta\,dt + \gamma\,dW and define the expected payoff g(t,x)=Et,x[h(XT)]g(t, x) = E^{t,x}[h(X_T)]. Then gg solves

gt+βgx+12γ2gxx=0,g(T,x)=h(x).g_t + \beta g_x + \tfrac12 \gamma^2 g_{xx} = 0, \qquad g(T, x) = h(x).

The proof takes three steps. 1. g(t,Xt)g(t, X_t) is a martingale, by the tower property. 2. Itô splits dgdg into a dtdt term and a dWdW term. 3. A martingale can’t drift, so the dtdt term is zero, and that zero is the PDE.