Deep Reinforcement Learning Portfolio Optimization Using Macroeconomic Indicators and Market Data

This paper proposes a novel deep reinforcement learning (DRL) framework for portfolio optimization that integrates macroeconomic indicators with financial market data. Unlike traditional mean–variance and factor models, which provide static solutions, the proposed approach constructs an enriched state space combining asset returns, volatilities, and bond yields with macroeconomic variables such as consumer price index (CPI), GDP growth, and interest rates. The action space is defined as dynamic portfolio weight adjustments across multiple asset classes, while the reward function incorporates risk-adjusted measures to balance profitability and stability. An improved actor–critic algorithm, extending Proximal Policy Optimization (PPO) with a macro-factor attention mechanism, is employed to capture the influence of systemic drivers on asset performance. Experimental evaluation using multi-year market and macroeconomic data demonstrates that the proposed DRL method achieves superior cumulative returns, reduced volatility, and enhanced Sharpe ratios compared with benchmark models. These findings highlight the effectiveness of integrating macroeconomic information into reinforcement learning frameworks, providing a robust and adaptive pathway for intelligent financial decision-making.