HEDGING BARRIER OPTIONS USING REINFORCEMENT LEARNING

We examine the use of reinforcement learning (RL) to hedge barrier options. We find that, when the hedger's objective is to minimize value at risk or conditional value at risk, RL is an attractive alternative to traditional hedging approaches. RL requires an assumption about the stochastic process followed by the underlying asset during the life of the exotic option, but our tests show that the results from using RL are fairly robust to this assumption. We do not consider transaction costs in this research. However, we show that RL involves less trading than traditional hedging approaches. As a result, the existence of transaction costs can be expected to increase the attractiveness of RL.