Revisiting the Income Elasticity of Energy Consumption: A Heterogeneous, Common Factor, Dynamic OECD & non-OECD Country Panel Analysis.

• Author: Liddle, Brantley

1. INTRODUCTION

   Since the OPEC oil embargo in 1973, studies have emphasized the role of energy prices in shaping energy use and factor allocation. Similarly, in the applied energy economic literature, there has long been interest in estimating/understanding the macro-energy-GDP elasticity--the percentage change in energy consumption associated with a 1% change in GDP. Previous work on the income elasticity of energy consumption has found a lack of leapfrogging (i.e., economic growth has not become less energy intensive in developing/industrializing countries), despite obvious technology transfer (current developing countries employ technology more advanced than that used in OECD countries circa 1960-1970). Also, estimating the relationship between economic development and energy demand and determining whether that relationship changes as levels of development change have been popular questions in energy economics (e.g., Judson et al. 1999; Medlock and Soligo 2001; and van Benthem and Romani 2009). Understanding more about the energy demand-GDP relationship and its dynamics are important for several reasons. Knowing the income elasticity of energy consumption can help in assessing the feasibility/stringency of intensity-based targets (e.g., energy or carbon emissions over GDP); and the elasticity is utilized in energy forecasting and as an input to larger energy systems or integrated assessment models (IAM) that are used to examine climate change options.

   Indeed, the macro energy elasticity of GDP is useful in projecting energy consumption for a given economic growth rate, and several countries, as part of the Paris Agreement on climate change, have committed to reduce their emissions intensity (i.e., the ratio of carbon emissions to GDP). But if the macro energy (or carbon) elasticity of GDP is less than unity, then energy/emissions intensity will fall in a business-as-usual economic growth scenario. Among the countries to set intensity-based targets are China, India, Malaysia, and U.S. Several other countries have set goals to reduce emissions off a business-as-usual (growth) scenario, including Indonesia, Thailand, and Republic of Korea. (Indonesia has a goal to lower its macro energy elasticity of GDP, too.) In addition, the APEC economies have an aspirational goal of lowering APEC aggregate energy intensity by 45% from 2005 levels by 2035; and the ASEAN countries have a goal of lowering energy intensity by 20% from 2005 levels by 2020 and 30% by 2025. These national actions demonstrate that energy intensity and energy demand responses are a critical element in climate change strategies, even though a host of other issues must also be addressed in mitigating greenhouse gas emissions.

   The current paper contributes to the literature on aggregate energy demand estimation by assembling an unusually wide panel dataset that covers not only energy consumption and economic growth but also end-use energy prices for 37 OECD and 41 non-OECD countries. Moreover, the approach also employs estimation methods that address nonstationarity, heterogeneity, and cross-sectional dependence. (A detailed description of the data and methods follows in Section 3.) We seek to determine (i) whether (and if so, by how much) energy intensity will fall in business-as-usual economic growth scenarios, and (ii) whether energy forecasts/IAMs need to allow for energy elasticities that change with economic growth.

   A major weakness in many previous empirical studies of international energy demand is the lack of energy price data that reflects local domestic conditions. Global oil prices are available for an extended period, but they exclude the prices of other important fuels as well as electric power. In addition, price controls and subsidies for domestic fuel production may distort world price conditions for any specific fuel. Occasionally, there may exist published domestic energy price data for select countries over limited horizons, but reliance upon this information alone may severely restrict the scope of the analysis. These problems may potentially introduce serious biases in the estimates of income elasticities and other responses because energy prices are not represented well.

   A substantial reason for revisiting the income elasticity was our effort to splice various energy price series together to form a more complete set of energy prices for 78 countries covering an extended number of years. This important data development effort--a contribution in its own right--'provides a useful perspective on the potential advantages of including energy price information for a full range of countries.

2. LITERATURE REVIEW

   There is now an extensive literature on energy demand studies that include evaluations of specific fuels and their substitution with other energy forms. Dahl and Roman (2004) survey the literature on many fuels through the early 2000s and conclude that most energy forms are both price and income inelastic in the long run. Although income responses are higher, similar responses apply for gasoline consumption (Dahl, 2014). Similar conclusions are also reached by Labandeira et al. (2017), who provide a meta-analysis combining estimated responses from many studies. Studies included in these analyses are primarily from the richer OECD members. Preliminary conclusions offered by Huntington et al. (2019) suggest that price and income responses within the major industrializing economies outside the OECD do not appear to be dramatically different from those of OECD economies, although the studies for these rapidly growing economies are much sparser.

   One of the major challenges in reaching conclusions from the previous literature is that researchers apply different methodologies to different countries and time periods as well as use different data sources. With some important exceptions like the early estimates by Pindyck (1979a, 1979b), few researchers have applied the same analytical framework to evaluate global energy demand in as many economies as possible.

   The current paper is most like Galli (1998), Medlock and Soligo (2001), Gately and Huntington (2002) (2), van Benthem and Romani (2009), and van Benthem (2015), in that we analyze data that have both time series and cross-sectional dimensions, consider both GDP and energy prices, and employ a dynamic model. Galli analyzed 10 developing Asian economies over 1973-1990 that included Korea and Taiwan. Medlock and Soligo compiled data from 28 countries, of which seven were non-OECD (Brazil and six Asian countries), over 1978-1995. The dataset van Benthem and Romani analyzed contained 17 developing countries (including Israel and Korea)--for which individual country, end-use prices were available--and spanned 1978-2003. The individual country, end-use price data that van Benthem (2015) used ran from 1978-2006 and included observations from 58 countries. It appears only the Galli dataset was balanced. Also, only Galli's data--which was sourced from Pesaran et al. 1997--is publically available.

   In addition to using a standard demand-type model in which energy consumption per capita is a function of GDP per capita and real energy price (all in natural logs), Medlock and Soligo (2001), van Benthem and Romani (2009), and van Benthem (2015) employed the partial adjustment mechanism of Koyck (1954); whereas, Galli (1998) estimated an error correction model. All papers used a homogeneous, fixed effects estimator; van Benthem and Romani (2009) and van Benthem (2015) included time effects, too; Galli also considered the weighted mean group procedure of Swamy (1971), and Medlock and Soligo (2001) employed the two-stage least squares approach of Balestra and Nerlove (1966) to address dynamic panel bias. To capture potential nonlinearities, all four papers added a GDP per capita squared term (van Benthem and Romani 2009 included price squared as well). In addition, van Benthem (2015) estimated a linear model across several income bands.

   All papers uncovered evidence of a nonlinear relationship between energy consumption and GDP, i.e., significant coefficients for both GDP and GDP squared (although, for Galli, those coefficients were insignificant when the mean group procedure was used). Hence, their results suggested that the income elasticity of energy changed with GDP. However, the shapes of the GDP-energy relationship were not always the same. Galli (1998) and Medlock and Soligo (2001) estimated inverted U-shaped relationships (i.e., the GDP term was positive while the GDP squared term was negative), as did van Benthem (2015) for income in the $10,000-$40,000 range. In contrast, van Benthem and Romani (2009) estimated U-shaped relationships (i.e., a negative GDP coefficient but a positive GDP squared one), as did van Benthem (2015) for GDP per capita less than $10,000 (where the linear GDP term was insignificant). The van Benthem and Romani result of an increasing income elasticity appears to have been caused by observations from income levels less than $5,000 since a subsequent regression based on income levels between $5,000-$10,000 produced an inverted-U shaped relationship.

   For a linear model, Galli estimated long run income and price elasticities of 1.18 and -0.32, respectively. Also, country-specific income elasticities were typically above unity for the developing Asian economies that Galli analyzed; yet, the forecasts implied that nearly all the countries considered would have an elasticity below unity today (given the higher income levels used in the forecasts and the negative GDP squared coefficient in Galli's model). In their linear model, van Benthem and Romani (2009), also considering developing countries, found substantially different GDP and price elasticities of 0.64 and -0.55, respectively (although, that regression did not include time effects, which are demonstrated to be significant). Another regression that was based on a $5,000-$10,000 income band and that...