Hedging using simulation: a least squares approach

This paper presents a new and flexible computational approach to derivative hedging. It is based on the use of least squares regression in order to compute the hedging portfolio. This nonparametric methodology can be readily applied to any derivative contract written on a single underlying risky asset in a complete market with continuous Markov price paths. We illustrate this technique computing sensitivities on plain vanilla and exotic options with both European and American exercise style. The achieved numerical accuracy is always comparable with the best simulation and semianalytic techniques presented in the literature.