Predicting Scrap Steel Prices Through Machine Learning for South China

Governments and investors have traditionally placed reliance on estimates of price time series of many different types of commodities. This study uses time series data from 08/23/2013 to 04/15/2021 to investigate the challenging job of predicting prices of scrap steel, which are issued for the South China market on a daily basis. Previous research has not given enough weight to predictions of this crucial commodity price indicator. Here, price forecasts are generated using Gaussian process regression algorithms built using cross-validation processes and Bayesian optimization methods. This empirical forecast framework provides reasonably accurate price estimates over the out-of-sample period of 09/17/2019–04/15/2021, with a relative root mean square error of 0.3287%. Price research models can be employed by governments and investors to make well-informed decisions on regional scrap steel markets.