Predicting open interest in thermal coal futures using machine learning

In this study, our objective is to tackle the open interest prediction problem by centering on the thermal coal futures traded on the Chinese Zhengzhou Commodity Exchange, using daily data from January 4, 2016, to December 31, 2020. The machine learning methodology we employ for prediction purposes is the Gaussian process regression model, tuned via cross validation and Bayesian optimizations that take into account multiple kernels, basis functions, and predictor standardization options. As a result, we develop a model configuration that produces quite accurate and stable predictions , with the root mean square error of 16048.2263 and the mean absolute error of 9163.3498 for the out-of-sample period from December 31, 2019 to December 31, 2020 as compared to the average open interest of 247968.6346. Our results have several key implications. Firstly, our results might be used to technical predictions not influenced by open interest. Additionally, they might be combined with other forecast results to create viewpoints on open interest trends to assess whether or not money flows into the contract are increasing or decreasing as part of understanding a contract’s liquidity and interest and carry out policy research based on these trends.