A Multi-Scaling Reinforcement Learning Trading System Based on Multi-Scaling Convolutional Neural Networks

Advancements in machine learning have led to an increased interest applying deep reinforcement techniques investment decision-making problems. Despite this, existing approaches often rely solely on single-scaling daily data, neglecting the importance of multi-scaling information, such as weekly or monthly processes. To address this limitation, a convolutional neural network for learning-based stock trading, termed SARSA (state, action, reward, state, action), is proposed. Our method utilizes obtain features and financial data automatically. This involves using with several filter sizes perform extraction temporal features. Multiple-scaling feature mining allows agents operate over longer time scaling, identifying low positions line avoiding fluctuations during continuous declines. mimics human approach considering information at varying spatial scaling trading. We further enhance network’s robustness by adding average pooling layer backbone network, reducing overfitting. State, on-policy method, generates dynamic trading strategies that combine across different while dangerous strategies. evaluate effectiveness our proposed four real-world datasets (Dow Jones, NASDAQ, General Electric, AAPLE) spanning from 1 January 2007 31 December 2020, demonstrate its superior profits compared baseline methods. In addition, we various comparative ablation tests order superiority architecture. Through these experiments, module yields better results module.