Policy implications of generalised distributed lags and the estimated demand for petroleum products

In many recent analyses of energy consumption, the following functional form is assumed: Qt = BXt + [lambda]Qt-1 + et where Qt is a vector of consumption data, Xt is a matrix of explanatory economic variables, Qt-1 is a vector of consumption lagged one observation period, B and [lambda] are regression parameters to be estimated and et is a vector of appropriately distributed random disturbances. Although this specification is attractive in the sense that both short- and long-term elasticities can be obtained from it, it does require certain restrictive assumptions to be met in order for it to be theoretically valid. This paper tests those assumptions and finds them not met for certain petroleum products. An alternative specification is suggested and policy differences between the popular but incorrect specification and the suggested one are presented. These differences are shown to be substantial.