Bayesian Symbolic Regression and Other Similar Methods as a Tool for Forecasting Commodities Prices

Bayesian Symbolic Regression (BSR) is used to predict spot prices of 56 commodities. BSR is a certain improvement over the symbolic regression technique based on genetic programming. Besides, there has been limited applications of the symbolic regression to forecasting prices in economics and finance. Contrary to prior simulations of BSR with synthetic data, this study discusses an application to the real-world data derived from commodities markets. In particular, forecasting one month ahead spot prices of 56 commodities. Indeed, BSR presents valuable capabilities for addressing the complexities associated with variable selection in econometric modelling. It is expected to also handle also some other challenges smoothly. Therefore, this study is carefully tailored to deal with commodity markets time-series data. Moreover, several alternative techniques are also tested, i.e., the symbolic regression with genetic programming, Dynamic Model Averaging, LASSO and RIDGE regressions, time-varying parameters regression, ARIMA, and no-change method, etc. In particular, the main aim is to focus on forecast accuracy. The obtained outcomes can give valuable insights for both researchers and practitioners interested in implementing BSR in econometric and financial projects in the future.