Reinforcement Learning Models in Stock Trading

Abstract

Many researchers and professional stock traders have struggled with the specialty of figuring out stock prices. The study area of stock value prediction has piqued financial experts’ interest greatly. Many speculators are adept at predicting the stock market’s future direction, which allows for decent and profitable speculation. Brokers, speculators, and professional traders can provide crucial information on the stock market's future direction with the use of tremendous and strong prediction frameworks for the stock market. In this study, Reinforcement Learning (RL) models are shown to have the best predictive and trading signal accuracy for the stock market. The Preference Ranking Organization Method for Enrichment Evaluation (fuzzy PROMETHEE) multicriteria decision-making (MCDM) method was used to evaluate the RL models developed in this study. the following; accuracy, precision, consistency in making profits, simplicity in implementation, profit optimization rate, volatility rate/speed, reliability, and speed were employed to evaluate the performance of the models. The results from this study showed that with a net flow of 0.0823, DDQN was determined as the most favorable and preferred RL model in stock trading. DQN, Dueling QN, and CNN came second, third, and fourth, with net flows of 0.0364, −0.0142, and −0.0465, respectively. RNN-LSTM with a net flow of −0.0581 was the least preferred alternative. The obtained result illustrates the applicability and usage of the MCDM approach in model selection.