Exploratory RL Yields Closed-Form Trading Policies

On April 2, 2026, Yun Zhao posted an arXiv preprint that frames speculative trading as a sequential optimal-stopping problem within the exploratory reinforcement learning paradigm. The paper introduces a relaxed model using Cox-process stopping times, Shannon entropy regularization, and derives exploratory HJB equations with closed-form Gibbs optimal policies, proving error bounds and convergence. It also presents an RL algorithm evaluated on a pairs-trading example.