Reinforcement Learning Based Dynamic Asset Allocation with Technical and Macro-Economic Analysis

This study proposes a reinforcement learning based dynamic asset allocation framework that incorporates both market-based technical indicators and macroeconomic variables. The model aims to learn an optimal investment policy that maximizes cumulative portfolio returns while adapting to changing market regimes and economic conditions. We construct a multi-asset portfolio including equities, government bonds, corporate bonds, and cash equivalents, and train the reinforcement learning agent using Proximal Policy Optimization algorithms within a Markov Decision Process framework. Extensive ablation studies reveal that the inclusion of select macroeconomic variables enhances both portfolio returns and downside risk control. Regime-specific analyses confirm that macro-informed models outperform baseline strategies in market downturns while maintaining competitive performance during bull and bear markets. This study bridges the gap between the literature in the field of computer science and financial economics, offering an empirically validated, end-to-end decision-making tool for regime-aware, multi-asset portfolio management.