Optimization of Futures Price Forecasting and Trading Strategies Based on Clustering and Multi-model Fusion

In this paper, a novel real-time trading system is presented that leverages market state awareness and multi-model fusion for enhanced futures price forecasting on 39 industrial futures listed in the China Commodity Exchange. First, we dynamically classify market states (volatility, trend strength, etc.) using K-means clustering, addressing the limitations of traditional methods in capturing market heterogeneity. Then, a synergistic forecasting approach is proposed, which combines Support Vector Regression (SVR) for smooth markets and multiple Transformer model variants in order to exploit diverse long-term dependencies for high volatility markets. Moreover, a Deep Q-Network (DQN) is used to dynamically integrate these models, optimizing forecasting accuracy in real-time and overcoming the limitations of static model switching. Experimental results show that our integrated approach significantly improves the performance and efficiency of futures price forecasting and trading with MSE of 0.0006 and R2 of 0.86.