Mobile Edition
Print
Get the Mag
Weekly Updates1. INTRODUCTION
The sharp increase in energy demand observed at the beginning of the new millennium was mostly attributed to the rapid growth of China, India and other emerging economies. This was a stark reminder of how crucial energy is as less developed economies--often based on agriculture--gradually become industrial economies. During this process the energy intensity of each additional unit of output increases. This process of development, the climb up of the 'energy ladder', is familiar to the energy literature: industrialized countries have followed similar paths in the process of their development.
Observing the historical behavior of these economies reveals that countries, at the end of their climb, reach very different energy intensity levels. This is due to a number of factors. Some are predetermined by their geographical location, size, and climatic exposure, which all have a strong influence on energy intensity. But factors such as industrial structure, efficiency and mass mobility solutions are significant and are the results of explicit economic and policy choices made by countries. Therefore, understanding the way in which energy demand has responded to economic growth and price variations in developing countries can provide some foresight on where they will reach by the end of their climb. Existing empirical literature on the relationship between economic growth and energy demand addresses some of these issues for OECD countries, or specific developing regions; there are fewer works that attempt to draw a broader picture of the experience of developing countries using advanced econometric tools.
This paper contributes to this literature by expanding the geographic coverage, using previously unavailable end-use price data and including the detail of three sectors. It uses recently developed methodologies that control for unobserved common dynamic effects including but not limited to technological change, and tests which of these methodologies fits the data best. By adding the detail of sector- and income-specific regressions, the analysis identifies common patterns as well as important differences across both sectors and income levels in the developing world.
2. ENERGY DEMAND, ECONOMIC GROWTH AND TECHNOLOGICAL CHANGE
There is a large literature that looks at the relationship between energy demand and economic growth (Stevens, 2000). Models in which economic growth and prices are the sole determinants of energy demand have been widely adopted, as they offer some important advantages: they are replicable across countries and produce elasticity estimates comparable over time and space.
Most studies in the 1980s and 1990s relied on time series data to provide evidence of the change in income elasticity of energy demand over different levels of economic development both in industrial countries (Eden et al., 1981 for the US; Humphrey and Stanislaw, 1979 for the UK) and developing countries (Bhatia, 1987; Hall, 1991; Galli, 1998). Cross sectional studies became more popular as data became available (Dahl and McDonald, 1998; Dahl, 1994; Paga and Gurer, 1996; Ang, 1987; Schipper and Meyers, 1992).
During the past decade, there has been significant effort to use modern econometric techniques to estimate the elasticity of energy demand to income and price (Galli, 1998; Judson et al., 1999; Medlock and Soligo, 2001). These studies focus on different regions and rely on different specifications--some allowing for important elements of non-linearity -, often making use of panel data sets. The paper by Medlock and Soligo (2001), in particular, uses panel data for 28 countries, a country-specific fixed effects estimator and a non-linear specification for income. They find strong evidence of non-linearity in the relationship between income and energy demand in three sectors, which reflects the effects of the structural change of the economy and technology adoption during the course of development. The study is only partially useful in terms of estimating this relationship in developing countries, as its results are based on a mix of poorer and wealthier countries.
Galli (1998) uses a panel data set of ten developing Asian economies and uses a quadratic function of log income to estimate total energy demand. She finds evidence of declining energy intensity as these Asian economies become wealthier (the energy intensity starts declining around $4,000 PPP), providing further evidence for the non-linear specification. Both these papers find relatively low price elasticity levels, and acknowledge that price data are elusive to collect, particularly for developing countries. Judson et al. (1999) use a panel of up to 123 countries to estimate energy demand in three sectors: households, transportation and industry. They adopt a flexible form for income, and use time and country fixed effects. They find that as income increases, the household sector's share of aggregate energy demand decreases, while transportation's share rises and industry follows an inverted U-shape. While their work is very comprehensive, the absence of prices is an important omission. This paper addresses some of these shortcomings. It adopts a non-linear form for income, using a larger data set of low and middle income countries only, using a new price series of domestic energy end-use prices, and allowing them to enter the estimation non-linearly as well.
There is another important challenge that undermines some of the recent literature: identifying the impact of economic growth on energy demand independently from the impact of technology. Estimating these two effects, technology and economic growth, separately is not easy if at all possible. Technological change (and preferences) is both a determinant of and is determined by economic growth. Additionally, technological change is determined by other factors as well, which are not directly related to economic growth. The identification problem can be divided into two separate issues: on one side how to control for technological progress implicit in economic development. On the other side, how to control for technological progress above and beyond the normal trends associated with economic development. The first issue is important, but not overall crucial. If technology is correlated with income growth, then the estimated income elasticity will simply include the associated technological change.
But the second issue presents a challenge, especially in models that rely on long time series for many countries, where explicit data on technology is not easily available. As highlighted by Griffin and Schulman (2005), most of the studies that look at energy demand and economic growth will face a potential bias, as it is very complex to explicitly control for technological change. Both time fixed effects (Griffin and Schulman, 2005) and asymmetric price effects (Gately and Huntington, 2002) have recently been interpreted as tools to control for instances when technology has improved energy efficiency significantly. Another complication is that time fixed effects include all unobserved variables that influence a country's energy demand, which include technological change but also e.g. demographic changes, changes in the structure of the economy, weather, etc.
Huntington (2006) and Adeyemi et al. (2008) discuss how to compare these different methods. Few papers in this literature have explicitly applied these tools to control for technological change and other unobserved variables, but rather rely only on income and prices to capture technological change implicitly (Galli, 1998; Medlock and Soligo, 2001). Adeyemi and Hunt (2007) test the relevance of these tools in the industrial sector in OECD countries and find evidence to support the asymmetric price specification. Our paper contributes to the methodological debate described above by empirically testing both time fixed effects and asymmetric prices in the context of developing countries.
3. DATA
The data (see Appendix 1 for more details) were provided by Rice University, based on the IEA Extended Energy Balances for non-OECD countries and various country-specific sources. The final data set contains annual data for the period 1978-2003 and includes three sectors: transport, industry and residential, commercial and agriculture. (1)
Total final consumption (TFC) data, expressed in kton of oil equivalent, are available for all countries and sectors. GDP at PPP, expressed in constant international US dollars, is taken from the World Bank's World Development Indicators. (2) Figure 1 plots these two variables, which illustrate the key relationship of the "energy ladder".
Following the literature on demand estimation for OECD countries, we prefer GDP at PPP to GDP at market exchange rates because it enables comparison across countries as it accounts for differences in purchasing power (Medlock and Soligo, 2001). Two energy price variables are available: the more commonly used international oil prices as well as a domestic end-use energy price series for each country and sector. Both are real price indices expressed in local currencies. Oil prices are taken from the EIA, (3) end-use prices from a combination of the IEA Energy Prices and Taxes (4) and national sources. Population data are taken from the United Nations Population Division. (5) The TFC data by sector contain a few discontinuities. Whenever this is the result of a clear accounting reason (such as a reclassification of sectors leading to a discontinuous allocation of energy consumption to sectors), we added a dummy variable for the sector, country and years to which the discontinuity pertains.
[FIGURE 1 OMITTED]
Table 1 provides an overview of the values of TFC per capita, real GDP at PPP per capita, the real oil price index and the weighted-average real domestic end-use price index in 1980 and 2002, as well as the compound average growth in this period. This table consists of country-aggregated TFC data and gives a first indication of the statistical relationship between growth in per capita energy consumption and per capita income. Figure 2 summarizes this in a scatter plot.
FEED