REINFORCEMENT LEARNING FOR FINANCIAL STOCHASTIC CONTROL: OPTIMAL MARKET MAKING WITH REBATE AND OPTIMAL LIQUIDATION WITH HIDDEN ORDER

Abstract

This thesis studies two important aspects of liquidity innovation, i.e., rebate to market makers and use of hidden orders through the lens of financial stochastic control. To obtain the solutions, modern financial stochastic control problems often involve cumbersome numerical solutions given the complex dynamics, which promotes the need of developing scalable and accurate algorithms for solving real problems. Among others, reinforcement learning has demonstrated initial success as documented in the literatures. This thesis proposes the Hamiltonian-Guided-Value-Function Approximation methods to solve the HJB equations associated to the two topics.