We present a new and flexible computational approach to derivative hedging. The method is based on the use of least squares to compute option sensitivities. This methodology can be readily applied to a huge class of derivative contracts under the assumptions of market completeness and continuous Markov price processes.
We illustrate this technique with a series of examples. In particular, we test it on plain vanilla and exotic options with both european and american exercise style, written on a single underlying asset with either constant or stochastic volatility. We show that the numerical accuracy that is achieved is always greater than or comparable with the best simulation and semianalytic techniques.