Perbandingan Metode Q-Learning Dan SARSA Dalam Optimasi Prediksi Tren Saham Pada Indeks Harga Saham Gabungan (IDX)

Abstract

This study evaluates the performance of Reinforcement Learning algorithms, namely Q-Learning and SARSA, in generating automated trading strategies for the Indonesian stock market. The research is motivated by the high volatility and uncertainty of stock price movements, which require adaptive decision-making methods. The dataset consists of historical stock price data from five companies listed on the Indonesia Stock Exchange (BBCA, BBRI, TLKM, UNVR, and ASII), obtained using the yfinance library and simulated through 1,000 trading episodes. Performance evaluation was conducted based on reward trends, equity curve patterns, and final performance statistics to assess learning effectiveness under different market conditions. The results indicate that Q-Learning performs better on stocks with strong price momentum due to its more aggressive exploration behavior, while SARSA provides more stable performance in highly volatile markets owing to its conservative on-policy approach. Overall, neither algorithm demonstrates absolute dominance; instead, each offers distinct advantages depending on stock characteristics and risk profiles. These findings highlight the potential of Reinforcement Learning for the development of algorithmic trading strategies in the Indonesian stock market.