Issues and Problems in Applied Econometrics: Study of a Model of World Oil Supply and Demand
Gracki, Kirsten L.
MetadataShow full item record
The development of econometric equations that represent behavioral relations involves the complex interaction of several steps. Problems arise in each step because the equations are being designed to both represent economic behavior and statistically explain trends in the historical data. Economic theory does not specify all of the properties of the relation studied, the functional form of the relation, in particular, or the time lags involved. In estimating the numerical parameters of the relation, the regression analysis of time series data gives rise to problems of collinearity and autocorrelation. With the large and detailed database and English language regression software available, an equation must be chosen from among several estimated equations, each of which already meet theoretical and statistical acceptance criteria. Finally, the forecast behavior is an important characteristic of an equation to be used in a large structural model. According to McNown, the accuracy of forecasts "leaves much to be desired" (15, p. 359), and judgmental adjustments contribute substantially to improving forecast accuracy. Examination of Data Resources, Incorporated's (DRI) World Oil Model (WOM) reveals some of these issues in trend analysis and econometric analysis of behavioral patterns. Roger Brinner and Michael Dunn of DRI developed a preliminary econometric model of world oil demand in 1980 independent of the Energy Group at DRI. In September 1990, the WOM currently used by the Energy Group differed in many ways from the one developed in 1980. My research project was to modify the Energy Group's model in light of Brinner and Dunn's research. Brinner and Dunn investigated the functional form of oil demand reaction to changes in economic activity and real prices in the development of the preliminary model in 1980. A time lag scheme to represent the multiperiod demand response to a single price change was developed at this time. Collinearity problems led to the specification of oil consumption intensity (oil consumption per unit of economic activity) as a function of real prices. My research involved the modification of these behavioral equations to include oil consumption trends of the 1980's. This proved to be the most challenging task. In the Energy Group's model, endogenously determined forecast levels of oil consumption are adjusted with the addition of a series, CAD (consumption adjustment), to reproduce the forecast behavior of more detailed DRI energy models. A goal of my research project was to develop a better behavioral equation with different forecast behavior, so that the adjustment of the results would not need to be so large. This goal was accomplished only in the United States equation and the United Kingdom equation. The forecast levels generated by the other eight consumption equations, even lowered slightly with the inclusion of a time trend, are not satisfactory. Modeling oil consumption as a function of economic activity and real prices illustrated both the theoretical and practical problems involved in the econometric analysis of time series data.