نوع مقاله : مقاله پژوهشی
نویسندگان
1 دانشگاه یزد- دانشکده مهندسی کامپیوتر
2 دانشگاه یزد، دانشکده مهندسی کامپیوتر
چکیده
کلیدواژهها
عنوان مقاله [English]
نویسنده [English]
Due to the volatility and uncertainty inherent in the stock market, devising an optimal trading strategy is a complex endeavor. Given the non-repetitive nature of trading circumstances, learning through interactions becomes imperative. Reinforcement learning emerges as an interactive learning approach capable of adjusting system parameters based solely on a scalar efficiency signal. This paper introduces a methodology wherein the states of the system are defined by the time step, the total number of shares purchased thus far, the standard deviation of stock prices from the beginning to the current step, and the difference between the current price and the price at the previous step. By defining a suitable reinforcement signal, the paper employs one of the most popular reinforcement learning algorithms, Q-learning, to approximate state-action value functions. The stock market is simulated using a set of equations, and the proposed method is applied. Performance evaluation is conducted by comparing the proposed model against mean reversion trading strategy across 5000 simulated markets. The experimental results demonstrate that the trading strategy derived from the Q-model not only yields lower average cost but also exhibits greater reliability compared to mean reversion strategy.
کلیدواژهها [English]