Climate Risks in Main Producing Areas and Realized Volatility in Agricultural Futures: Machine Learning Methods Based on High‐Frequency Data

This paper uses high‐frequency data of agricultural futures trading prices in the Chicago Mercantile Exchange, totaling about 1.2 million pieces of data, to conduct a study of climate risk and realized volatility based on machine learning methods. We carry out the following studies: first, we find that the nonlinear machine learning method has better applicability to the extended model of heterogeneous autoregressive model with the realized volatility model. Second, by analyzing the importance of the factors in the forecasting model, we find that compared with the global climate risk factor and most of the macroeconomic fundamentals, the climate risk factor of the main agricultural production region constructed in this paper contributes more to the forecasting of realized volatility. Third, based on the Shapley Additive exPlanations value analysis of the predictors, this paper examines the directional impact of climate risk factors and macroeconomic fundamentals on agricultural commodity volatility.