Forecasting Base Metals Prices: A Comparison of Various Bayesian-Based Methods

This paper addresses the topic of forecasting base metal prices index using advanced Bayesian methods, emphasising Bayesian dynamic mixture models.Original schemes were expanded by certain modifications.A broad set of macroeconomic indicators, such as interest rates, industrial production, economic activity, market stress indices, others commodities prices, exchange rates and information from stock markets, etc. were taken as potential predictors.Models were recursively estimated, taking under consideration possible discrepancy between released and revised data, carefully simulating real-time forecasting conditions.Dynamic Model Averaging was found to provide the highest accuracy of predictions compared to competing models.The forecasts were significantly more accurate than the ARIMA method or the no-change method.Among the dynamic mixture variants, model selection appeared to offer the best performance.The Clark-West test for nested models confirmed that forecast combination schemes lead to significant forecast accuracy improvements.Sector companies' stock prices and particular exchange rates were found to be the important base metals price predictors.