RPR-2011-17 - ERIA

exist, at least the first column must contain non-zero elements. Thus, this cointegratingrelationship specified in Equation (4) represents the foundation of acomplete dynamic panel model and the regression which allows us to compare theimmediate and overall average elasticities of energy demand across countries.Finally, the data used in the above regression covered 71 countries and regionsfor the period 1965-2010. The data for energy consumption in each country and thatfor the real price of crude oil come from BP Statistical Review of World Energy (BP,2011). The data for population and GDP (calculated with the constant price andadjusted for purchasing power parity) come from the World Development Indicators(World Bank, 2012).4. Empirical Results4.1. Model Selection and Benchmark ResultsBefore estimating any relationship between energy consumption and itsexplanatory variables, one may need some identification strategy either fromeconomic or statistical perspectives.Specifically, it is assumed that all laggedindependent variables on the RHS of Equation (4) are exogenous so that their furtherlagged or differentiated items can be used as the instrument variables for GMMestimation. 3Based on Roodman (2006), we first differentiate the regression function(say, Equation (4)) to remove country specific effect ( u i) and thereafter produce anequation that is estimable by instrumental variables and use a Generalised Method ofMoments estimator for coefficients using lagged levels of the dependent variable andthe predetermined variables and differences of the strictly exogenous variables. Theresults from both a difference GMM and system-GMM estimation are compared toexamine the autocorrelation of the logged energy consumption (with Equation (5)).ln Citit1T ln C D u (5)t1itiit3 This assumption is only made for simplicity, and the results from the endogenous independentvariables are shown in Appendix A.23