Multi-Agent Deep Reinforcement Learning (MADRL) frameworks have increasingly shown profitable algorithmic trading from stock market data. MADRL frameworks are typically simplified with a single final agent acting on the environment, where sub-agents are specialized with distinct reward function with the same stock market data processed on intra-day time intervals, limiting the MADRL framework’s focus on stock price movement. To explore this problem, a new sub-agent with specialized macroeconomic input data processed on varying timescales is incorporated into a MADRL framework with a focus on algorithmic trading improvement performance. This study derives its architecture from the MADRL architecture on the Multi-Agent Double-Deep Q-Network (MADDQN) framework, which uses TimesNet in its sub-agent architecture. A macroeconomic-based sub-agent has been added to the framework. An experimental research strategy was implemented where each sub-agent, a baseline MADRL, and an optimized MADRL framework were statistically tested across eight metrics. The results demonstrate that incorporating a macroeconomic agent can enhance trading performance, improving key metrics such as the Sharpe Ratio, Sortino Ratio, and Cumulative Return. This indicates that macroeconomic data, when integrated into a multi-agent system, improves risk management and trading accuracy, thereby optimizing overall trading outcomes. However, the improvement varies depending on the stocks.