On the use of multivariate cointegration analysis in residential energy demand modelling

The present paper shows how cointegration analysis within a multivariate framework may be applied for the estimation of energy demand elasticities in order to account for the non-stationarity of the time series used. The dynamic modelling approach followed is one based on general-to-specific modelling within a system. The case is for annual Austrian residential energy demand over the period 1970 to 1993. The explanatory variables used are real energy price, real disposable income, and the temperature variable heating degree days. The results indicate that there is one cointegrating vector only. The long-run energy demand elasticities derived are -0.02 for price, +1.13 for income, and +0.77 for temperature. The long-run system of energy demand makes sense from an economist's point of view and provides evidence that the aggregate price elasticity may be much lower than commonly assumed, and suggested by many other, more traditional empirical studies in this field. In the short-run analysis, the error-correction term turns out to play an important role, thereby clearly disqualifying a traditional VAR model in first differences investigated as a rival model.