Deep RL Portfolio Paper: SAC Beats Buy-and-Hold Only in Euro Stoxx 50

A new academic paper (arXiv:2605.17307) from Kamil Kashif and Robert Ślepaczuk explores the use of deep reinforcement learning for portfolio management across global equity markets. The framework employs the Soft Actor-Critic (SAC) algorithm to learn continuous portfolio weights within a Markov Decision Process. The reward function incorporates transaction costs, turnover penalties, and diversification constraints. Five model configurations were compared, varying reward formulation, policy structure (flat vs. hierarchical Dirichlet), and temporal encoder (LSTM vs. Transformer). The models were evaluated via walk-forward optimization across sixteen out-of-sample folds spanning 2003-2026 on the Nasdaq-100, Nikkei 225, and Euro Stoxx 50.

Results were mixed: RL strategies achieved competitive risk-adjusted performance primarily in the Euro Stoxx 50, where statistically significant abnormal returns were observed after HAC-robust inference. However, the central hypothesis was only partially confirmed—no strategy delivered statistically significant excess returns relative to Buy and Hold across all markets. Regime analysis reveals that RL adds the most value during periods of elevated uncertainty (e.g., financial crises, COVID volatility). Ensemble aggregation across markets improved risk-adjusted performance, confirming the benefits of geographic diversification. The study provides a rigorous, reproducible benchmark for RL in quantitative finance, with 67 pages of detailed analysis.