A non-intrusive stratified resampler for multi-factor models: applications to pricing of Bermudan and Swing options

Abstract

We aim to solve dynamic programming equations (DPE) related to financial valuations in the energy market. On the underlying asset, we consider that a calibrated model is not available and a limited sample from its historical data is accessible. We look for a non-intrusive method solving the DPE with empirical regression techniques, by suitable resampling of the historical data (and therefore without calibration of the model). In the power market, the forward contracts are driven by hidden factors modeled by Markov processes. Even though the DPE solution depends on these hidden factors, we come up with a resampling scheme using only the historical data of a transformation of the observable forward prices. We show numerical experiments applied to Bermudan and Swing options on forward contracts

Avatar
Isaque Pimentel
Quantitative Analyst, Consultant

Ph.D. in Applied Mathematics interested in Quantitative Finance and Data Science.