In the described research three agricultural commodities (i.e., wheat, corn and soybean) spot prices were analyzed. In particular, one-month ahead forecasts were built with techniques like dynamic model averaging (DMA), the median probability model and Bayesian model averaging. The common features of these methods are time-varying parameters approach toward estimation of regression coefficients and dealing with model uncertainty. In other words, starting with multiple potentially important explanatory variables, various component linear regression models can be constructed. Then, from these models an averaged forecast can be constructed. Moreover, the mentioned techniques can be easily modified from model averaging into a model selection approach. Considering as benchmark models, time-varying parameters regression with all considered potential price drivers, historical average, ARIMA (Auto-Regressive Integrated Moving Average) and the naïve forecast models, the Diebold–Mariano test suggested that DMA is an interesting alternative model, if forecast accuracy is the aim. Secondly, the interpretation of time-varying weights ascribed to component models containing a given variable suggested that economic development of emerging BRIC economies (Brazil, Russia, India and China) is recently one of the most important drivers of agricultural commodities prices. The analysis was made on the monthly data between 1976 and 2016. The initial price drivers were various fundamental, macroeconomic and financial factors.
Forecast accuracy and economic gains from Bayesian model averaging using time-varying weights
