Empirical Deep Hedging

The existing delta hedging strategies are model-based: either the strategies are explicitly based on a given option pricing model or reinforcement learning is used to train an agent based on synthetic data generated by a given volatility model. In this paper, we empirically train an agent for the hedging of derivative securities in a model-free manner using deep reinforcement learning (DRL) with continuous actions. The training data consists of intra-day option price observations on S&P500 index over six years, and top of that, we use other data periods for validation and testing. We have two important results. First, a DRL agent trained using synthetic data generated from a calibrated stochastic volatility model outperforms the practitioners delta hedging strategy. Second, and more importantly, we find that an empirically trained DRL agent has superior performance, even if no specifications about volatility models were provided in the DRL procedure. This implies that DRL can capture the dynamics of S&P500 and self-learn how to hedge actual options efficiently