An exponential functional of random walks

Abstract: The aim of this paper is to investigate discrete approximations of the exponential functional $\int_0^{\infty} \exp(B(t) - \nu t) \di t$ of Brownian motion (which plays an important role in Asian options of financial mathematics) by the help of simple, symmetric random walks. In some applications the discrete model could be even more natural than the continuous one. The properties of the discrete exponential functional are rather different from the continuous one: typically its distribution is singular w.r.t. Lebesgue measure, all of its positive integer moments are finite and they characterize the distribution. On the other hand, using suitable random walk approximations to Brownian motion, the resulting discrete exponential functionals converge a.s. to the exponential functional of Brownian motion, hence their limit distribution is the same as in the continuous case, namely, the one of the reciprocal of a gamma random variable, so absolutely continuous w.r.t. Lebesgue measure. This way we give a new, elementary proof for an earlier result by Dufresne and Yor as well.