Data-driven hedging with generative models

Abstract We propose a nonparametric data-driven methodology for hedging using generative models. In contrast with model-based hedging approaches relying on sensitivity analysis of model pricing functions, our approach uses a conditional generative model trained on market data to simulate realistic market scenarios given current market conditions and computes hedge ratios which minimize risk across these scenarios. The approach incorporates transaction costs, leads to an optimal selection of hedging instruments, and adapts to market conditions. We illustrate the effectiveness of this methodology for hedging option portfolios using VolGAN, a generative model for implied volatility surfaces. The out-of-sample performance of the method matches and improves over delta and delta-vega hedging, without retraining the model for more than 4 years after the training period.