= Brownian representation replication in a local volatility market
{c}
{title2=$\pi_t=h_t/(\widetilde S_t\sigma(t,S_t))$}
Work in the usual <natural Brownian filtration>, with positive <bank account> $B$, and suppose the discounted stock satisfies $d\widetilde S_t=\widetilde S_t\sigma(t,S_t)dW_t^Q$. Assume this diffusion coefficient is nonzero for $dt\,dQ$-almost every $(t,\omega)$. Let a terminal claim $H$ have square-integrable discounted payoff, and represent its price <martingale> as
$$
M_t=\mathbb E_Q[H/B_T\mid\mathcal F_t],\qquad dM_t=h_t\,dW_t^Q.
$$
The stock holding
$$
\pi_t=\frac{h_t}{\widetilde S_t\sigma(t,S_t)}
$$
and bank holding $M_t-\pi_t\widetilde S_t$ replicate it: discounted wealth is $M$ and $\pi\,d\widetilde S=dM$. This stochastic integral is well-defined through the square-integrability of $h$, even if the holdings themselves are not bounded.
For $H\ge0$, wealth is nonnegative, hence admissible. Every competing nonnegative-wealth <self-financing portfolio> has discounted wealth a <supermartingale>, so it cannot replicate $H$ with less initial capital. The nonzero-exposure hypothesis is essential; a division formula alone does not prove completeness at an absorbing zero.
Back to article page