TIME SERIES MODELING OF THE LIVE HOG MARKET: FORECASTS AND ANALYSIS OF DYNAMICS

MICHAEL STEPHEN KAYLEN, Purdue University

Abstract

The three broad objectives addressed by this research were to: analyze the dynamic relationships between hog market related variables, develop forecasting models specifically for these variables, and develop techniques for building ARIMA and VAR forecasting models. The first objective was met using an unconstrained VAR model with six lags. The variables in the model included: sow farrowings, income, feed price, hog slaughter, hog price, retail beef price, and retail pork price. Nine deterministic variables were included, primarily to account for seasonality in the data. These variables also included a quadratic spline function to model the changing seasonality of sow farrowings due to the introductions of partial and total confinement systems in the late 1950s and early 1970s, respectively. Hypothesis tests confirmed the importance of modeling this changing seasonality, and the shapes of the graphs of the splines agreed with a priori expectations based on previous studies relating to the diffusion of new technologies. Statistical techniques used to analyze the dynamic relationships included: decomposition of the forecast error variance, impulse responses, and a historical decomposition of error variance. Results of the analyses suggested that feed prices continue to have large effects on sow farrowings and hog slaughter, despite the high proportion of total costs of production attributed to the fixed costs associated with total confinement systems. The large fixed costs probably explain why the largest responses of sow farrowings and hog slaughter to feed price shocks occur after lags of five to ten quarters. The analyses also suggested retail meat prices currently play only a small role in the live hog market. However, analysis of the episodic 1973 period showed retail pork prices (along with feed prices) accounted for a substantial portion of the volatility in live hog prices during this time. The imposition and subsequent lifting of the freeze on meat prices undoubtedly contributed to this effect. The forecasting models which were developed using techniques proposed in this research appeared to perform well in an out-of-sample forecasting evaluation and in relation to other forecasting models which have appeared in the literature. The two forecasting VAR models performed substantially better than the unconstrained VAR model. Since previous studies have suggested VAR models do not forecast well, the results suggest the techniques proposed in this research for building VAR forecasting models are significant contributions to the literature.

Degree

Ph.D.

Subject Area

Agricultural economics

COinS