Modelling soybean prices in a changing policy environment

Accurate forecasts of commodity prices are an important ingredient in the policy formation process. A commodity price forecasting procedure used routinely by the US Department of Agriculture in their policy and market analysis activities is a simple, linear, reduced-form regression model that predicts season-average farm prices (SAFP) using policy variables and the ratio of total ending stocks to use. This approach is extended to the soybean SAFP to estimate a benchmark model using annual data. Also several specification issues related to this estimation framework are addressed. Evaluation suggests that the standard forecasting procedure may be affected by the fact that the ratio of stocks to use is endogenous to prices. In addition, important structural changes are revealed in these relationships over time. A model is then considered that allows parameters to shift gradually. Improvements in the accuracy of model forecasts allowed by this parameter switching technique are identified and discussed. In addition, the exact nature of the structural shifts is evaluated using dynamic impulse response functions.