Research on reinforcement learning stock trading strategy based on cost-aware information selection

Hu, Meiqi, Zhang, Zuoquan
Journal of Asset Management · Springer Nature · 2026-09-12 · DOI: 10.1057/s41260-026-00476-w

Abstract

We propose a cost-aware information-selective Markov decision process (IS-MDP) framework based on deep reinforcement learning, which integrates information acquisition into the Markov decision process (MDP) and jointly optimizes transactions and information selection. IS-MDP compresses the state through standardized three-dimensional core state, adopts layered action space, and uses multi-objective rewards to balance benefits, costs and risks. In the empirical study, we conducted a retrospective test of the CSI 300 Index under five market systems, as well as external experimental verification of Euro Stoxx 50 and S&P 500 markets. These experimental results show that the CSI 300 Index is more stable in bear market and risk control; Ablation tests confirmed the key role of factor formation and adaptive selection. IS-MDP provides a practical method for resource constrained algorithmic trading.

Read at publisher →

Similar articles