New RL Algorithm Solves Continuous-Time Optimal Stopping Problems

This research introduces CARLOS (Continuous-time Adaptive Reinforcement Learning for Optimal Stopping), a new algorithm designed to solve optimal stopping problems with continuous time resolution. Traditional simulation-based methods often rely on discretizing the stopping decision, which can lead to inaccuracies or computational inefficiencies. CARLOS addresses these limitations by employing an aggregate deep neural network (ADNN) to learn a joint space-time decision boundary. The algorithm progressively refines its timing-value estimates by starting with a coarse time grid and gradually increasing the frequency of stopping opportunities. An adaptive sampling strategy further concentrates training efforts near the optimal stopping boundary. Benchmarking results demonstrate that CARLOS outperforms existing Bermudan option solvers, achieving prices closer to the theoretical American upper bound. Furthermore, it exhibits high computational efficiency compared to non-RL alternatives, making it a powerful tool for financial modeling and other applications requiring precise optimal stopping decisions.