Dynamic Econometric Modeling of the U.S. Wheat Grain Market

Structural-time series models have not gained much ground in commodity market modeling despite the overwhelming popularity of time series approaches in forecasting and dynamic analyses. This dissertation contributes by applying developments in seasonal cointegration and structural-time series analysis (e.g., Zellner and Palm (1974); Hsiao (1997); Lee (1992); Franses and Kunst (1999); Ghysels and Osborn, 2001) to the study of agricultural commodity markets. The focus is on three research themes. The first theme investigates the role of cointegration and seasonal cointegration for market data, an issue considered timely because most applications assume deterministic seasonal components. The second issue breaks new ground in agricultural commodity modeling by introducing a new dynamic simultaneous equation model (DSEM) that accounts for seasonal cointegration. Lastly, the research compares the out-of-sample forecasting performance and impulse responses of four multi-equation models for the U.S. wheat market. The forecasting comparisons apply recent developments on testing for differences in mean-squared-errors.The study adopts a structural model for the U.S. wheat market and estimates four econometric specifications: a vector error-correction model without seasonal cointegration (VECM), a VECM with seasonal cointegration (SVECM), a DSEM with cointegration (CDSEM), and a DSEM with seasonal cointegration (SCDSEM). The conclusions may be summarized as follows. First, quarterly data in the U.S. wheat market (1975:03-1999:04) have seasonal unit roots, therefore, a VECM or DSEM should be specified. Second, in a forecasting context, seasonally cointegrated VECMs perform uniformly better that their nonseasonal counterpart. DSEM with seasonal cointegration, however, perform better than VECMs at longer forecast horizons. Lastly, the impulse response analysis and dynamic multiplier comparisons lead to one salient conclusion, omission of seasonal cointegration components when significant generates unexpected response functions and dynamic multipliers.Of particular interest for future research is an assessment of the small sample properties of impulse response functions for structural-time series models with seasonal cointegration. From a more pure economic perspective, a similar structural-time series analysis to other agricultural markets seems timely given the new finding that these models may outperform other multiple time series models that are often used in empirical work.