Akyapak, Omer FarukTunç, Ilhan2026-02-082026-02-0820259798331597276https://doi.org/10.1109/ASYU67174.2025.11208408https://hdl.handle.net/20.500.12885/53012025 Innovations in Intelligent Systems and Applications Conference, ASYU 2025 -- 2025-09-10 through 2025-09-12 -- Bursa -- 214381This study comparatively examines the potential of two reinforcement learning algorithms, Q-Learning and SARSA, in optimizing investment decisions in financial markets. Using the daily price data of Türk Traktör (TTRAK) stock for the year 2024 and Turkey's key macroeconomic indicators (interest rate, inflation, USD exchange rate, and gold price), a buy-sell-hold decision agent was trained. The agents learned situational information to make decisions aimed at optimizing investment portfolios. The results demonstrate that both algorithms are effective in generating investment strategies but exhibit distinct risk behaviors. Q-Learning produced higher yet more volatile returns, reflecting its aggressive, off-policy nature, while SARSA delivered more stable but conservative outcomes due to its on-policy approach. These findings emphasize the importance of aligning algorithm selection with investor profiles and highlight the benefits of incorporating macroeconomic factors into RL-based trading systems. The study underscores the value of integrating macroeconomic indicators with technical analysis in reinforcement learning-based trading systems to enhance decision-making quality. © 2025 IEEE.eninfo:eu-repo/semantics/closedAccessFinancial MarketsMacroeconomic IndicatorsQ-LearningReinforcement LearningSARSAStock TradingConference Object10.1109/ASYU67174.2025.112084082-s2.0-105022505889N/A