Spurred by the oil price shocks in late 1973 and during the period 1979-80 a lot of attention was devoted to the analysis of energy demand. Numerous attempts were made to estimate price and income elasticities from time-series and cross-section samples. As a representative example of the problems encountered when microeconomic demand analysis is applied to energy data we present some results for Austria (Wohlgemuth 1992).
Figure 11.1 shows the final consumption of energy for Austria. Figure 11.2 contains for the same sample range real GDP and real energy price data. Visual comparison of these graphs suggests a pronounced impact of the energy price hikes on energy demand and a substantial decoupling of energy demand from real economic activity.
Specifying a log-linear demand model with partial adjustment we obtain the short-run and long-run income and price elasticites reported in Figures 11.3 and 11.4. The estimated coefficients represent moving window regressions over eleven years. Thus the estimates for 1977 are based on a sample ranging from 1967 to 1977 and the estimates for 1989 use data from 1979 to 1989.
As expected the long-run elasticities are larger than the short-run values. Striking, however, is the high sensitivity of the estimates with respect to variations of the sample period. A lot of experiments with other functional forms and improved dynamic specifications could not improve these obvious deficiencies of the estimated elasticities.
The time paths of the estimated elasticities suggest that, during the period of high energy prices, the absolute value of both price and income elasticities increased, but they sharply declined as real energy prices fell to pre-OPEC levels. Although these results question the assumption of structural stability of the postulated data-generating process we gain at least some insight into the kind of parameter variation that occurred within the sample period.
More disturbing are the results reported in Figure 11.5 which show the simulated forecasting bounds for an ex ante forecast of one to five years from 1984. The forecasting range is calculated for twice the simulated standard deviation of the prediction error. The empirical evidence of this experiment suggests that the forecasting power of this type of energy demand model is very limited.
Was this article helpful?