A reinforcement learning approach for pricing derivatives

Analytical solutions to the derivatives pricing problem are known for only a small subset of derivatives and are usually based on strict assumptions. Practitioners will therefore frequently resort to numerical approximation techniques. In this paper, I will formulate a simple Markov decision process for which the optimal value function will, in a non-arbitrage world, be equivalent to a given derivative’s fair price function. This means that derivatives pricing can be understood as a reinforcement learning problem. In order to solve this problem I will propose a simplified version of the kernel-based reinforcement learning algorithm suggested in [4] and [6].