Modelling Primary Energy Consumption
under Model Uncertainty
Zsuzsanna Csereklyei, Department of Economics, ViennaUniversity of Economics and Business (WU),
Tel:+43/6643238603, E-mail:
Stefan Humer, Department of Economics, ViennaUniversity of Economics and Business (WU),
Tel:+43/1313364509, E-mail:
Overview
The causality between energy consumption and real income in developed countries has been an important research topic in recent years. Rising concerns about climate change and global warming increase the pressure on policy makers to take action against greenhouse gas emissions. We work with a panel dataset of 64 countries between 1965-2009 to establish the worldwide long-term link between energy consumption and other key macroeconomic variables like real income, population, capital stock and technology. Our goal is, while accounting for model uncertainty, to build a reliable energy model that could be used by policy makers to assess the impact of changing technology, income and population on future energy demand.
Methods
We deploy panel unit root and cointegration tests and different panel error correction models to estimate the long run determinants of primary energy consumption worldwide, but also in geopolitical and GDPpc clusters. We work with three basic panel error correction models normalized on the natural logarithm of primary energy consumption. The cointegrating equations are estimated with full information maximum likelihood methodology. The model specifications are as follows:
(1)The EC model of primary energy consumption on GDP and on population (E.YL)
(2) The EC model of primary energy consumption on population and physical capital (E.LK)
(3) The EC model of primary energy consumption on population, physical capital and technology (E.LKA)
Country fixed effects and fixed effects for the different decades are included. As significant differences in the patterns of energy consumption may be attributable to the different levels of economic development and different geopolitical developments, we decided take into account these effects. We created regional groups including North America, Western Europe, Eastern Europe Eurasia, Asia Pacific, South America, and Middle East & Africa, while the gross domestic product per capita groups consist of low income (0-10K$), low-middle income (10-20K$), middle-high income (20-35K$), high income (>35K$) countries. Thus, we have estimated for each country 3 x 3 models, the three basic models from above, and three attribute models, including the world, geopolitical and wealth dummies.
In this area of the literature, uncertainty about the choice of a statistical model is rarely considered. Whenever one model is opted over reasonable alternative models to represent knowledge about a specific process, model uncertainty occurs. We argue that inference on the basis of a single model without taking model uncertainty into account can lead to biased conclusions. Consequently, we tackle this problem by applying simple model-averaging techniques to the above described panel error correction models. We attempt to prove, that an averaged model will have a significantly better out-of-sample predicting ability than any single individual model.
Model averaging is performed in this paper by running two separate weighting procedures. In the first case, the weights are assigned based on the out-of-sample predicting ability of the nine models, which we test with the help of rolling forecasts. The second procedure is based on the residuals, or the in-sample-fit of the models. After the weights are assigned to the different models, we conduct another out-of-sample rolling forecast to determine with the help of the Diebold-Mariano test, whether the averaged models have a significantly better predicting ability.
Results
Analysing the results of the panel error correction models, we acknowledge that the error correction adjustment parameter is always negative and highly significant, indicating that the estimated relationships tend to a long-run equilibrium. We find that there is a positive long-term relationship running from physical capital, GDP, and population to primary energy consumption. We observe however a negative long-term relationship between total factor productivity and primary energy usage. Significant differences arise in the magnitude of the cointegrating coefficients, when we allow for differences in geopolitics and wealth levels. The averaged coefficient estimates indicate that population increase could be the major driver of energy consumption in the future, with real income and physical capital playing a propelling, but somewhat smaller role. Technology improvements will certainly contribute to the reduction of energy consumption levels.
Table 1: Coefficient estimates obtained by model averaging
In comparing the out-of-sample predicting ability of our models, we find the average models to have the highest long-term predictive power.
Conclusions
Establishing the determinants of long-term primary energy consumption with high predictive long-term accuracy is a very challenging but crucially important task. We have, while seeing the inherent uncertainty connected to modelling in general, attempted to decrease it. To our knowledge, not only is this the largest panel cointegration study on the subject, but also no other study has applied model averaging techniques to a large number of panel error correction models before. This allows forconsiderable flexibility for policy makers, who can utilize both the regional and wealth level results of specific regions for future predictions, and also the optimal coefficient estimates, which we obtained by model averaging.
Although we find population and income to be the major drivers of energy consumption, examining the total factor productivity variable, we see that our world, but also the average models exhibit a negative and slightly significant relationship between energy use and technology. Implicitly, a higher technology component in economic growth reduces primary energy consumption. This might be attributable to the structure and development stage of the economies, to the higher efficiency in energy use by the industry and the households, as well as to the higher part of knowledge based and non--energy intensive production in GDP, including the de-industrialisation in developed countries.
Presently highly growing countries like India, China, the Caspian area and Africa will have to deal with the questions of infrastructure and industry expansion, therefore will experience enhanced energy needs. It is important that fast growing countries should anticipate increased energy demand, and implement energy efficient technologies from the beginning on. Therefore, it would be advisable to invest as much as possible in energy research at universities and research institutes or to support energy efficiencies at industry level.