A Study on the Hedging Effect of Compound Index Weather Futures on Yield Fluctuations of Agricultural Products Based on Machine Learning—A Case Study of Soybeans in Harbin

Weather derivatives are priced based on weather indices. A single weather index reflects limited weather information, while a composite weather index provides a more comprehensive picture of the real weather conditions. In this paper, we selected meteorological data from Harbin, Heilongjiang, and synthesized a composite weather index (CWD) by using the PCA dimensionality reduction technique. In this paper, machine learning models such as Ordinary Least Squares (OLS), Multilayer Perceptron (MLP), Time-Domain Convolutional Neural Networks (TCNs), and Convolutional Long-Short Memory neural network (CNN-LSTM) are used to fit the variance of CWD index, respectively, and CWD futures are priced based on them. In this paper, the efficiency of CWD futures in hedging soybean yield volatility is calculated under a variance minimization strategy. The empirical results show that among the four weather futures pricing models covered in this paper, CWD futures based on the TCN algorithm have the highest pricing efficiency, and CWD futures based on the deep learning algorithm have higher pricing efficiency than CWD futures based on other algorithms. CWD futures can effectively hedge the quantity risk of soybean production, and CWD futures have a better hedging effect than single weather index futures.