Improving prediction accuracy in agricultural markets through the CIMA-AttGRU model

In the Chinese futures market, agricultural product futures play a crucial role. While previous studies have primarily relied on historical price data and fundamental financial indicators of agricultural product futures, there is a growing recognition of the value that lies within the vast amounts of textual data generated in the financial domain. Our study specifically focuses on the limitations of existing methods in capturing the complex relationships and rich semantic information embedded in these textual sources. This article designs a CIMA AttGRU (CIMA-AttGRU) model for soybean futures, which is a forecasting method for the agricultural product market. This model uniquely integrates Collective Intrinsic Mode Analysis (CIMA) with an Attention-Gated Recurrent Unit (AttGRU), leveraging the strengths of both techniques to enhance predictive accuracy and adaptability. The rationale behind employing the CIMA-AttGRU model lies in its ability to effectively tackle the inherent challenges of financial market analysis. By incorporating CIMA, the model adeptly filters out market noise, directly addressing the challenge of high volatility. Additionally, with its attention mechanism, the CIMA-AttGRU targets the issue of non-linear patterns by allowing dynamic adjustment to temporal dependencies, offering differential learning capabilities crucial for capturing the nuanced fluctuations in futures prices. Complementing the CIMA and AttGRU, the integration of Class-wise Adversarial Domain Adaptation (CADA) further refines the model's robustness, addressing the critical challenge of domain adaptivity. This aspect is particularly vital for edamame futures, where price determinants can vary significantly over time and across regions. Our empirical results demonstrate a significant improvement in forecasting precision, with the CIMA-AttGRU model achieving a Mean Absolute Error (MAE) reduction of 15% and a Mean Squared Error (MSE) reduction of 20% compared to conventional models. This superior performance, especially in terms of prediction accuracy and handling market fluctuations, highlights the improve of the model compared to existing methods and has made significant explorations in agricultural market forecasting.