Development and Assessment of Trading Strategies for the A-Share Market Based on PSO-Enhanced LSTM Models

This study proposes a hybrid trading strategy framework for China’s A-share market by integrating Particle Swarm Optimization (PSO) with Long Short-Term Memory (LSTM) networks. The PSO algorithm optimizes LSTM hyperparameters to enhance predictive accuracy and model generalization. Based on model forecasts, four trading strategies were developed and evaluated-ranging from a basic binary decision rule to more advanced approaches incorporating trend confirmation, dynamic position sizing, and risk management techniques. Backtesting on historical CSI 300 data reveals that the basic strategy underperforms the buy-and-hold benchmark, with negative returns and high drawdowns. Incremental refinements improve risk-adjusted metrics but fail to achieve profitability. The results highlight a critical gap between predictive performance and real-world trading success, emphasizing the need for sophisticated, context-aware decision rules and robust execution mechanisms. This research contributes to the field by demonstrating both the potential and limitations of PSO-enhanced LSTM models in financial forecasting and by offering a structured path toward developing adaptive, AI-driven trading systems for emerging markets.