Understanding the complexity of futures markets investing in China: evidence from deep learning techniques

We examine the effectiveness of deep learning models in implementing the time-series momentum strategy in the Chinese futures market. Our empirical analysis shows that the long short-term memory (LSTM) model performs better than other machine learning methods in terms of profitability and risk management. Importantly, incorporating the Sharpe ratio into model training significantly increases returns while decreasing risks. Additionally, our findings indicate that considering momentum turning points and combining short- and long-term predictions further enhances the performance of the LSTM model.