Pseudo-Model-Free Hedging for Variable Annuities via Deep Reinforcement Learning

Dynamic hedging for variable annuities has been studied widely in the literature. In general, when the financial and actuarial market models are complicated, closed-form solutions are usually unavailable or computationally expensive. In recent years, machine learning techniques have been utilized to facilitate the valuation and hedging of variable annuities. In this talk, we shall implement deep reinforcement learning, an emerging field of machine learning, to hedge the variable annuities. Instead of being given the financial and actuarial market models, the insurance company learns the optimal hedging strategies. The company trades in the market, receives reward signals, and revises the hedging strategy dynamically. We shall demonstrate novel results, comparing with the traditional Deltas and state-of-the-art supervised learning method.