강화학습 기반 주식 자동 매매 모델 전략 제안

This work proposes an effective learning strategy by developing a stock auto-sell system based on reinforcement learning and comparing and analyzing it under various conditions. Based on PG, DQN, and A2C reinforcement learning techniques, reinforcement learning was implemented by determining the behavior of buying/selling/ viewing networks at the end of the day and providing compensation. The experiment was conducted largely by dividing four conditions. We propose an effective reinforcement learning strategy through performance comparison by reinforcement learning technique, model stability comparison by variability in learning period, and performance comparison experiment by length of learning period of model.