Deep Learning Algorithms for Hedging Options

The purpose of this study is to test how modern machine learning methods are better or worse than classical statistical methods in finance. The basis for comparison is the problem of hedging options. To solve such a problem, a reinforcement learning model was developed, after which several steps were taken to improve the performance of this model. It was revealed that the final model can completely repeat the optimal hedging in the world where the underlying asset is distributed according to the Black-Scholes model. It is shown that in an unfamiliar and more complex world, the performance of the reinforcement learning model better than the basic Black-Scholes model. The proposed machine learning method for hedging does not rely on any information about the distribution of the underlying asset and, therefore, can be successfully applied in working with real data.