82834 AUTHOR ACCEPTED MANUSCRIPT FINAL PUBLICATION INFORMATION The Energy Transition of the Transition Economies : An Empirical Analysis The definitive version of the text was subsequently published in Energy Economics, 40(November 2013), 2013-09-16 Published by Elsevier THE FINAL PUBLISHED VERSION OF THIS ARTICLE IS AVAILABLE ON THE PUBLISHER’S PLATFORM This Author Accepted Manuscript is copyrighted by the World Bank and published by Elsevier. It is posted here by agreement between them. Changes resulting from the publishing process—such as editing, corrections, structural formatting, and other quality control mechanisms—may not be reflected in this version of the text. You may download, copy, and distribute this Author Accepted Manuscript for noncommercial purposes. Your license is limited by the following restrictions: (1) You may use this Author Accepted Manuscript for noncommercial purposes only under a CC BY-NC-ND 3.0 Unported license http://creativecommons.org/licenses/by-nc-nd/3.0/. (2) The integrity of the work and identification of the author, copyright owner, and publisher must be preserved in any copy. (3) You must attribute this Author Accepted Manuscript in the following format: This is an Author Accepted Manuscript of an Article by Zhang, Fan The Energy Transition of the Transition Economies : An Empirical Analysis © World Bank, published in the Energy Economics40(November 2013) 2013-09-16 http:// creativecommons.org/licenses/by-nc-nd/3.0/ © 2013 The World Bank The Energy Transition of the Transition Economies: An empirical analysis 1 Fan Zhang 2 Abstract The aggregate manufacturing energy intensity of 28 countries in Eastern Europe and Central Asia had declined by 35 percent during 1998-2008. This study reveals strong evidence of convergence: less efficient countries improved more rapidly and the cross-country variance in energy productivity narrowed over time. An index decomposition analysis indicates that energy intensities declined largely because of more efficient energy use rather than shifts from energy intensive to less intensive manufacturing activities. Income growth and energy price increases were the main drivers of the convergence. They dominated the impact of trade, which led to specialization in energy intensive industries. Key Words: Manufacturing Energy Intensity, Energy Productivity Convergence, Index Decomposition Analysis, Trade, Eastern Europe and Central Asia JEL Classification: O13, O47, P28, Q40 1 I thank Uwe Deichmann and Indermit Gill for useful comments and suggestions. I thank Sofia Chiarucci and Po Ying Wong for excellent research assistance. The findings, interpretations, and conclusions expressed in this paper are entirely those of the author. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. 2 World Bank, 1818 H ST NW, Washington DC. 20433. fzhang1@worldbank.org 1 I. Introduction The Eastern Europe and Central Asia (ECA) region had one of the least efficient manufacturing sectors in the world because of the region’s historical emphasis on heavy industries and artificially low energy prices during the Soviet era. 3 Since the transition period started two decades ago, the sector’s energy productivity (value added per unit of energy input) has consistently improved –by almost 35 percent between 1998 and 2008. As a result, although manufacturing value added increased by more than 50 percent over the period, energy consumption had been stabilized at around 250 million tons of oil equivalent (mtoe). How did the region achieve this impressive decoupling of economic growth from energy use? This paper seeks to answer the question by empirically examining the trend and determinants of manufacturing energy use for 28 countries in ECA. 4 Using detailed sector and country-level data, we conduct three types of analysis. First, we compare energy productivity performance over time and across countries. We ask the following questions: Is the trend of declining energy intensity consistent across the region? Do lagging countries catch up with technological leaders? Second, we decompose the manufacturing sector into 11 subsectors and investigate whether the reduction in energy intensity is more attributable to changes in the mix of production activities (the structural effect), or to the fundamental improvement in the use of energy (the efficiency effect). Finally, we identify the economic factors that have driven changes in energy intensity, energy efficiency and the structural composition of industrial activities. The analysis reveals strong evidence of cross-country convergence: not only the inter-country variation in energy intensity has reduced; but there is a tendency of more rapid improvement in countries that had high energy intensity initially. Beyond this general trend, we find some Central Asian countries are notable exceptions where energy intensity had remained largely unchanged. Interestingly, intrinsic efficiency improvement turns out to have contributed to most of the reduction in manufacturing energy intensity, whereas structural change had had a less influence, and had led to higher energy intensity in some countries. Country-level analysis indicates that rising per capita income and higher energy prices are strong predictors of more efficient energy use, both through the channel of efficiency rather than structural mix. A one percent increase in per capita GDP is on average associated with a 1.4 percentage point reduction in manufacturing energy intensity. A 10 percent increase in the price 3 Manufacturing consists of all industrial activity outside of mining and construction. The detailed composition of manufacturing industry is described in section III. 4 Following the definition of World Bank regions, Eastern Europe and Central Asia includes Albania, Armenia, Azerbaijan, Belarus, Bosnia & Herzegovina, Bulgaria, Croatia, the Czech Republic, Estonia, Georgia, FYR Macedonia, Hungary, Kazakhstan, Kosovo, the Kyrgyz Republic, Latvia, Lithuania, Moldova, Montenegro, Poland, Romania, the Russian Federation, the Slovak Republic, Slovenia, Tajikistan, Turkey, Turkmenistan, Ukraine and Uzbekistan. Kosovo and Montenegro are not included in the analysis because their energy consumption data are not available. 2 of electricity for industry is on average associated with a 2.2 percentage point reduction in manufacturing energy intensity, ceteris paribus. Furthermore, energy intensity slightly increases with the increase of manufacturing exports and imports, suggesting that trade induces specialization on more energy-intensive industries. In spite of many existing cross-country studies on energy intensity comparison and determinants (for example, Howarth et al. 1991; Uanander et al, 1991; Mulder and de Groot 2007), relatively few have looked at the trends in developing countries. Fankhauser and Cornillie (2004), Markandya et al. (2005) and EBRD (2011) are some exceptions that examine economy-wide energy intensity change in the ECA region. Our analysis differs from their studies in that we focus on the manufacturing sector, the largest energy user in the region and a sector that perhaps also offers the largest opportunities for energy savings. We also cover more countries and years and look at changes during the second decade of the transition. The rest of the paper proceeds as follows. Section II provides background information about the manufacturing sector in ECA. Section III describes the data. Section IV presents empirical methods and results. Section V discusses the conclusions and policy implications. II. Background – Manufacturing in Eastern Europe and Central Asia During the Soviet era, manufacturing was the cornerstone of centrally planned economies of ECA. Leaders emphasized rapid industrialization, particularly of heavy industry, to foster economic growth. This heavy industry was also vastly inefficient. To facilitate industrial production and social welfare, central planners kept energy prices well below cost-recovery levels. Of all non-OECD subsidies to fossil fuels (including electricity) in 1991, totaling US$270-US$330 billion, roughly two-thirds were in the former Soviet Union and Eastern Europe (Myers and Kent, 2001). Subsidies distorted price and led to extraordinary profligacy in the use of energy. In 1990, the former Soviet Union’s energy intensity-the ratio of primary energy consumption to GDP-was 70 percent higher than in the United States and 2.5 times that of Western Europe. The gap was especially evident in the industrial sector, where energy use per unit of output was three times higher than in the United States and 3.5 times higher than in the Western Europe (U.S. Congress, 1993). The early years of transition brought a sharp contraction in manufacturing output following the disruption of traditional trade and financial links and an end to centrally planned production (Havrylyshyn et al, 1998). Between 1991 and 1993, manufacturing value added fell on average by 30 percent. In parallel, manufacturing energy consumption dropped on average by 27 percent across all countries. Most transition countries continued the comprehensive reform programs, including the liberalization of energy prices. Price supports for fossil fuels in Russia and Eastern Europe were greatly reduced. Bringing laws into agreement with the European Union’s (EU) further accelerated the process of eliminating remaining subsidies. Even in Russia, gasoline and diesel prices for industry (which received two-thirds of energy subsidies in 1994) have increased 3 to world market levels, although gas prices still remain below border price levels (World Bank, 2008). The market-friendly reforms had a far-reaching impact on the transformation of the energy economies. When the sharp initial decline gave way to gradual economic recovery, energy consumption continued to decline and then stabilized at around 250 mtoe per year. Meanwhile, manufacturing output had more than doubled between 2002 and 2008, as shown in Figures 1-a and 1-b. This has occurred because the average manufacturing energy intensity declined by 35 percent from 0.4 kilogram of oil equivalent (koe) per dollar (in purchasing power parity (PPP) terms) in 1998 to 0.26 koe per dollar (PPP) in 2008. 5 If energy intensity had remained at its 2002 level, energy demand would have been 776 mtoe higher, the equivalent of 10 years’ total energy consumption in Ukraine today. While manufacturing energy intensity at the regional level had sharply declined, subregions show disparate behavior. Figure 2-a presents energy intensity trend for six subregions/countries. Central Asia is notably lagging behind in terms of the reduction in energy intensity. In fact, Kazakhstan, the Kyrgyz Republic, and Tajikistan had all experienced an increase in manufacturing energy intensity since 2000 as shown in Figure 2-b. In the rest of the paper, we explore the driving factors of manufacturing energy intensity to explain why countries were not doing equally well. III. Data Data used in this paper come from three sources. The energy consumption data are obtained from the International Energy Agency (IEA) World Energy Statistics and Balances database (2012). The database reports final energy consumption of manufacturing in 28 ECA countries. The manufacturing sector is further broken out by 11 subsectors at the two-digit level of the international standard industrial classification Revision 4 (ISIC Rev. 4). The aggregate value added of manufacturing is reported in constant 2000 US dollars in the World Bank World Development Indicators (WDI) (2012). To compare energy intensity levels across countries, we apply PPP exchange rates to convert all values into constant 2005 international dollars. The theoretically ideal PPP exchange rate to use at the sector level would be based on the output prices of the sector, rather than on aggregate expenditure prices. However, due to the limited availability of producer price-based PPPs, we follow the standard practice of using expenditure PPPs for cross-country convergence analysis (such as in Dollar and Wolff, 1993 and Mulder and de Groot, 2007). 5 Manufacturing energy intensity is measured by energy use per unit of manufacturing value added. Energy use is defined as final energy consumption excluding energy losses in the generation, transmission and distribution of purchased electric power. Value added presents the net economic output, or gross economic output less the value of purchased inputs. It is derived by subtracting the cost of materials, supplies, fuel and purchased electricity from the value of final outputs. 4 For decomposition analysis, we obtain value added of manufacturing at the subsector level from United Nations Industrial Development Organization (UNIDO) Yearbook of Industrial Statistics (INDSTAT4) (2011). INDSTAT4 reports manufacturing value added at the three-digit level of ISIC Rev.3 for selected ECA countries. We match IEA (2012) with INDSTAT4 (2011) to calculate subsector energy intensity at the two-digit level of ISIC Rev.4. INDSTAT4 reports value added in current prices, denominated in US dollars. We use the GDP deflator of each country to convert values of different years into 2000 prices and then to constant international dollars using 2005 PPP rates. Per capital GDP (PPP based), the capital-labor ratio of manufacturing and the share of exports and imports in total manufacturing value added are all obtained from WDI (2012). Energy prices are obtained from Energy Regulators Regional Association’s tariff database and Eurostat. IV. Empirical Strategy and Results In this section, we present three types of empirical analysis to identify the sources and mechanisms of energy intensity change in the manufacturing sector of ECA. First, we conduct a cross-country convergence analysis to test whether all countries experienced improvement in energy productivity and whether backward countries were catching up with efficient ones. Second, using an index decomposition method, we investigate the extent to which the change in manufacturing energy intensity is induced by structural shifts vs. efficiency improvements. Finally we explore the determinants of energy intensity, the structural mix and the efficiency levels based on regression analyses. A. Energy Intensity Convergence Convergence analysis is typically used in the context of economic growth and goes back to Barro and Sala-i-Martin (1992). There are different definitions of convergence in the literature. The reduction in cross-sectional variance in productivity is called σ-convergence. The tendency of less productive countries to grow faster is called β-convergence. To measure σ-convergence in the context of the energy intensity analysis, we look at the trend of cross-country variance over time. The mean and standard deviation of manufacturing energy intensity for an unbalanced panel of 28 countries in ECA during 1998-2008 are shown in Figure 3. To avoid selection bias, we also present the same statistics for a balanced panel of 23 countries. 6 For both samples of countries, the average energy intensity had declined steadily on average at a rate of 3.2 percent per year. The group of countries as a whole was converging towards more efficient economies (EU-15) with the gap in manufacturing energy intensity between ECA and EU-15 average more than halved over the sample period. The standard 6 Value added data are not available for Albania during 2004-2006, Romania in 2002, Russia during 1998-2001, Serbia during 1998-1999 and Turkmenistan during 2005- 2008. 5 deviation of energy intensity had also reduced, indicating that the gap between the less efficient and the technology leaders was narrowed over time. To test the existence of β-convergence, we conduct regressions of the rate of energy intensity change, , on initial levels of energy intensity, , for the 25 countries where data on energy intensities are available for both the initial (1998) and the end (2008) years. The regression equation and results are presented below. The coefficient on the initial levels is negative and statistically significant at 10-percent level when all countries are considered. Excluding Central Asian countries, the coefficient of the initial level almost doubles in absolute term and is statistically significant at the 5 percent level. The R2 also increases. The results indicate that countries with higher energy intensity at outset were catching up to more productive countries, especially for those outside Central Asia. However, the simple regression only has limited ability to explain cross-country difference in energy intensity. Even without accounting for Central Asia, the regression accounts for only 30 percent of the cross-country variance in the rate of energy intensity change. In the following, we further explore the underlying process of energy-intensity change across countries. B. Energy Intensity Decomposition Manufacturing energy intensity can drop for two main reasons: manufacturing shifts toward different products that require less energy to make (the structural effect), or increased efficiency of manufacturing processes so they require less energy to produce the same products (the efficiency effect). Specifically, assuming there are N subsectors within manufacturing, the manufacturing energy intensity can be written as: where Et and Yt are the overall energy consumption and value added of manufacturing in year t. eit and yit are energy consumption and value added of subsector i in year t. The aggregate energy intensity (It) is thus expressed in terms of sectoral energy efficiency (Cit=eit/yit) and production structure (Sit) which is measured by the sector share of the manufacturing value added (Sit=yit/Yit). To attribute changes in energy intensity between changes in efficiency and structural composition, various index decomposition methods have been used to isolate the impact of one variable from the other. For example, in multiplicative decomposition, the overall change in 6 energy intensity between the end year (t=T) and the base year (t=0) can be described as the multiplication of the structural (Dstr) and efficiency (DEff) indexes: where Dint is the intensity index, and IT and I0 are energy intensities at the end and the base year. One can use Laspeyres index to calculate the impact of structural or efficiency change by holding the efficiency or structural level constant at the base year value and use base year value as the fixed weight. The Laspeyres structural and efficiency indexes are given by: By reversing the role of base year and end year, one obtains the Passches indexes: There is a large literature in index decomposition analysis. See, for example, Ang and Zhang (2000) for the underlying concept and a literature review. Ang (2004) provides a thorough review of the application of index decomposition analysis in energy studies and recommends guidelines for methods selection. Liu and Ang (2007) present a comprehensive survey of energy decomposition analysis for the industrial sector during 1976-2005. 7 In this paper, we adopt the Fisher (1921) ideal index approach. The Fisher ideal index is the geometric mean of the Laspeyres and Paasche numbers: Fisher ideal index provides a perfect decomposition of aggregate energy intensity into structural and efficiency components, which is a very attractive property for interpreting the relative importance of industry composition and efficiency effects. 8 Liu and Ang (2003) first propose the use of Fisher ideal index in industrial energy demand studies. Ang, Liu and Chung (2004) then extend the two-factor Fisher index decomposition approach to multi-factors. On empirical analysis, Boyd and Roop (2004) use Fisher ideal index to examine the sources of energy intensity change in the context of U.S. manufacturing sector. Metcalf (2008) applies Fisher ideal index to analyze the trends of economy-wide energy intensity in the United States. 9 EBRD (2011) adopts the same decomposition method to analyze energy intensity change in Eastern Europe and the Caucasus. 7 Liu and Ang (2007) find that for most countries during the period of 1976 to 2005, trends in aggregate energy intensity for industry have been influenced more by efficiency effect than by structural change. 8 See Boyd and Roop (2004) for a discussion on the superlative theoretical properties of a Fisher ideal index. 9 Metcalf (2008) further advanced the technique by using intensity, structural and efficiency indexes as the basis of regression analysis to identify the causal determinants of efficiency and activity mix. Regression analysis of energy indexes in Section IV.C in this paper is conducted along the same lines. 7 To construct the structural and the efficiency indexes, we divide the manufacturing sector into 11 subsectors based on ISIC Rev. 4. These subsectors are iron and steel, chemical and petrochemical, non-ferrous metals, non-metallic minerals, transport equipment, machinery, food and tobacco, paper and pulp, wood and wood products, and textile and leather. Continuous time series value added data are only available at the subsector level for 13 countries mostly during 2001-2007. 10 This data availability dictates the countries selected and the period of investigation for the decomposition analysis. Figure 4 presents the decomposition results for the group of 13 countries as a whole between 2001 and 2007. Table 1 presents the decomposition results at the country level. As shown in Figure 4 and Table 1, efficiency improvements contributed to most of the reduction in energy intensity, while structural shifts had actually offset some of the efficiency gains in a handful of countries. For example, in Bulgaria, aggregate energy intensity of the manufacturing sector in 2007 was of 61 percent of its intensity level in 1998. The structural index was 105 percent of its level in 1998 while the efficiency index was 57 percent of its 1998 level. In other words, had the energy efficiency stayed the same, total energy intensity would have increased by 5 percent between 1998 and 2007. In Russia, the 32 percent reduction in energy intensity can be attributed almost entirely to improvements in energy efficiency during 2001-2007. Several factors need to be underlined when interpreting the above results. First, the decomposition analysis (at the two-digit level of ISIC) may mask some important structural shifts within an industrial subsector. For example, within the paper and pulp sector, pulp is much more energy intensive than paper; within the steel sector, shifts from oxygen steel to electric steel will reduce the intensity of energy use. Intra sector shifts from basic to specialized manufacturing are not analyzed here, but are certainly relevant. Second, although value added data allow the measurement of efficiency and activity levels for heterogeneous industries, value added is also influenced by a range of pricing effects unrelated to changes in the level of physical production. For example, prices of steel and chemicals have surged during the worldwide commodity boom during the 2000s. The steel and chemical sectors’ considerable decline in energy intensity may in part reflect the price development. 11 In a recent study, Ang and Xu (2013) extend the conventional monetary-based index decomposition analysis to incorporate physical activity indicators in industrial energy studies. Finally, it should be noted that structural change in the manufacturing industry is different from structural change at the aggregate, economy-wide level. Under market forces, all transition economies faced a drastic structural adjustment, which led to the downsizing of the heavy industry. Several studies confirm the general patterns of structural transformation from industry 10 Observations on manufacturing value added in Turkey are dropped to avoid potential price irregularities during the economic crisis in the early 2000s. Including Turkey data in the analysis does not qualitatively change the results. 11 We also measure energy intensity in terms of energy consumption per volume of product physical outputs for the steel and cement industry. The observed trend of increasing energy efficiency is robust to the alternative indicator. 8 to service in the region (Fankhauser and Cornillie, 2004 and EBRD, 2010). EBRD (2011), however, also find that improvements in intrinsic efficiency have dominated structural changes across most of the subregions in Eastern Europe and the Caucasus to contribute to an overall decline in the economy-wide energy intensity. C Determinants of Manufacturing Energy Intensity What have driven changes in manufacturing energy intensity, as well as energy efficiency and the composition of manufacturing activities? A number of factors may have played a role. These include: Income - As income grows, consumer demand for manufactured goods expand. As countries increase capacity utilization 12 and/or build new capital stock – equipment and factories- to meet increasing demand, efficiency rises. By fostering learning and innovation, income growth could also encourage companies to move up the value chain toward more value- added and less energy-intensive products. Energy prices – When energy is relatively cheap, there is little incentive for industry to reduce energy use or shift to less energy-intensive sectors. Capital intensity of manufacturing - Depending on whether capital and energy are substitutes or complements, there is a negative or positive relationship between capital and energy intensities. Trade – On one hand, trade could facilitate information and technology spillover that help improve energy efficiency; trade could also expose enterprises to a higher level of competition and create pressure to cut costs and save energy (EBRD, 2010). On the other hand, trade could contribute to cross-country divergence in energy intensity by stimulating international specialization (Grossman and Helpman, 1991). Finally, government energy policies and public awareness about energy efficiency can be important in promoting the rational use of energy. To understand which factors are more important in determining the patterns of ECA’s manufacturing energy intensity change, we estimate the following equation: where is manufacturing energy intensity in country i at year t. Git is per capita GDP measured in 2005 PPP based international dollars. Pit , industrial electricity prices measured in 2000 US dollars, is used as a proxy for industrial energy prices. 13 Rit is the capital-labor ratio indicating the capital intensity of the manufacturing activity. 14 Eit is the percent of the sum of 12 Bernstein et al. (2003) find that there are economies of scale in energy use and as industry capacity is fully utilized, it minimizes the amount of energy required to produce output. 13 We also include industrial gas prices in the regression, which do not qualitatively affect the estimation results but reduce the sample size. Ideally, one would also control for the prices of other major industrial fuels, i.e. coal, oil etc. However, cross-country time-series data on these prices are scarce. Given that electricity prices for industries are cost-reflective and are directly linked to the costs of input fuels in most ECA countries, we consider industrial electricity prices a reasonable proxy for the wider industrial energy prices. 14 We also control for the capital investment growth ratio of the manufacturing sector. High turn-over rate of capital stocks could mean more efficient facilities. It turns out there is no statistically significant relationship between the 9 exports and imports in manufacturing value added and is included as a measure of trade intensity. ki is an unobserved country fixed effect, which includes country specific characteristics that are fixed over time, such as culture, climate zone 15 and government regulation and so on. We also include year fixed effects, Tt (t = 1999-2008), to control for common cyclical components such as a common technology shock. uit is an idiosyncratic error term. β0, β1, β2, β3, and β4 are parameters to be estimated. Summary statistics of variables are presented in Table 2. If the country-specific effects ki are correlated with other explanatory variables which in turn are orthogonal to the error term uit, then OLS and random effects estimation would be biased but a within transformation will yield consistent estimators. As discussed in the previous section, energy intensity changes can be disaggregated into structural and efficiency components. Similar to Metcalf (2008), we estimate the determinants of structural and efficiency changes using index numbers of the 13 countries presented in Table 1, based on the same model specification of equation (7). Columns (1) , (2) and (3) of Table 3 report results from estimating determinants of manufacturing energy intensity identified in equation (7) via OLS, fixed effects and random effects models. The parameter estimates are very sensitive to the choice of model. The Hausman test decisively rejects the null hypothesis that the unobserved variable ki is uncorrelated with the other independent variables at the 1 percent level. In this case, only fixed effects specification reported in column (2) yields consistent estimators, and is the preferred specification. Controlling for country and year fixed effects, the parameter estimates suggest that all independent variables reported in column (2) of Table 3 are strong predictors of a country’s manufacturing energy intensity. A one percent increase in per capita GDP is on average associated with a 1.4 percentage point decrease in energy intensity. A 10 percent increase in industrial electricity price is on average associated with a 2.2 percentage point decrease in energy intensity. Both results are statistically significant at the 1 percent level. The estimates of capital-labor ratio show that manufacturing energy intensity first increases then decreases with the increase of capital intensity with the turning point slightly below the sample mean. This inverted-U relationship suggests that diminishing returns to capital may partially explain energy-productivity convergence across countries. 16 growth of capital stock and energy intensity. Including this variable does not qualitatively change the estimation results of other parameters but reduces the number of observations. 15 In some manufacturing sectors, portion of the energy consumption is influenced by weather conditions. The influence on energy consumption is different form industry to industry. Given the lack of time-series data on country-level weather conditions, and that heating and cooling degree days tend to be stable within country, the analysis does not specifically control for weather conditions in analyzing the energy consumption performance within the manufacturing sector. 16 This is consistent with findings of Metcalf (2008) based on the US data. However, to be noted, there has been no consensus about the sign of the elasticities of substitution between energy and capital. Some authors suggest the need to disaggregate capital inputs into working capital and physical capital, and find that energy is generally 10 Another interesting finding emerged from the results is that energy intensity increases with the increase of the share of exports and imports in total manufacturing, lending support for the hypothesis that trade induces specialization on more energy intensive industries. This result is not entirely surprising. As demonstrated in the model in Krugman (1987), comparative advantage leads to specialization in the tradable-goods sectors. Because the region has abundant energy resources, relatively low energy prices and more lenient emissions rules, many of the region’s countries are net carbon exporters. In fact, Ukraine, Russia, and Kazakhstan are the first, third, and fourth most carbon- (and energy-) intensive exporters in the world (Davis and Caldera, 2010). EU enlargement also offers access to a larger market and opportunities for specialization. Within the EU, some new member states including Bulgaria, Estonia, Hungary, Latvia and Romania have the most specialized industrial structures. For example, Hungary is highly specialized in refined petroleum products, while Bulgaria shows high specialization in mining and quarrying (European Commission, 2011). This specialization effect dominates other, potentially positive effects of trade on energy productivity, such as technological spillover and increased competition. One concern with the above analysis is that the causal effect between economic growth and energy intensity is likely to be two-way. Energy efficient technologies can improve productivity (Cleveland et al, 2000); reductions in energy demand could also free up capital required for energy requirement and cut expenses. Both effects can spur economic growth. It is also likely that industrial energy prices are affected by the amount of energy consumed in the industry - that is energy price may be endogenously determined by energy intensity. Furthermore, the above analysis assumes an instant response of energy intensity to income and energy price change within a year. More realistically, energy using behavior may respond to price changes with some lags. To address these concerns, we estimate a dynamic panel model where the current level of energy intensity depends on the level of energy intensity in the last period as described in equation (8). We then use the difference generalized methods of moments (GMM) technique outlined in Arellano and Bond (1991) to obtain consistent estimators. To account for the potential endogeneity of income and energy prices, we use the log of per capital income and the log of electricity prices lagged 2 to 4 periods as instruments in the GMM model. The estimation results are presented in column (4) of Table 3. Although some estimated coefficients changed in magnitude 17, none changed in sign and these results are similar to those obtained by treating income and price as exogenous. The specification also easily passes the overidentification test. complement to physical capital and a substitute for working capital (Field and Grebenstein, 1980). Solow (1987) demonstrates that either relationship is possible when using aggregated capital data. 17 Price elasticity is no longer statistically significant probably due to the reduced sample size resulting from the use of lagged variables as instruments. 11 Table 4 reports estimates from three separate regressions of equation (7) using energy intensity, structure and efficiency indexes as the dependent variables. The income and price elasticities of the intensity and efficiency indexes are both negative and statistically significant at the 1 percent level. The income and price elasticity of the efficiency index are higher than those for the overall intensity index in absolute term. In contrast, the structural index is less responsive to income and price changes. The income and price elasticities of structural index are in the opposite sign and are statistically insignificant. The results are consistent with those from the previous section. They show that income growth and price reform help reduce energy intensity. This is achieved mostly through intrinsic efficiency improvements, whereas structural change in the mix of industrial activities had a less influence on the level of aggregate energy intensity and may have partially offset efficiency gains. A comparison of the coefficient estimates on trade intensity helps distinguish the impact of trade on efficiency and structural change. While a higher share of trade is negatively correlated with the efficiency index (showing improvement in energy efficiency), it is strongly positively correlated with the structural index with a p-value less than 0.001. The results provide further evidence that the specialization effect of trade dominates other positive effects of trade on energy efficiency, and may have acted to offset some of the gains from the efficiency improvement. V. Conclusions and Policy Implications Industry was the cornerstone of the centrally planned economies in ECA. Today, it still plays an important role, measured by both its economic contribution and its environmental impact. Industry is the largest energy user, accounting for 28 percent of total energy use, and the second- largest carbon emitter, responsible for 34 percent of total carbon dioxide (CO2) emissions in the region in 2009 (Deichmann and Zhang, 2013). This paper investigates the sources and mechanisms of energy intensity change in the manufacturing sector of ECA during 1998-2008. There had been a remarkable reduction in manufacturing energy intensity over the period. The aggregate energy intensity dropped by 35 percent by 2008. Meanwhile, there is evidence of both σ- and β-convergence: the gap in energy intensity between leaders and backward countries was narrowed over time; the reduction in energy intensity had been more rapid in countries that had higher energy intensity initially. Notable exceptions are Kazakhstan, the Kyrgyz Republic and Tajikistan, where energy intensity had increased since 2000. To shed light on the sources of industrial energy use change, we decompose aggregate manufacturing energy intensity into efficiency and structural components using Fisher ideal index numbers. The decomposition analysis suggests that the bulk of the reduction in energy intensity has occurred because of improvements in energy efficiency as opposed to shifts from energy intensive to less intensive manufacturing activities in the 13 countries where disaggregated value added data are available. 12 We then conduct regression analysis to identify factors driving the change in aggregate energy intensity, energy efficiency and structural composition. The results show that both income growth and rising energy prices helped lower overall manufacturing energy intensity in ECA countries. A one percent increase in per capita GDP is on average associated with a 1.4 percentage point reduction in energy intensity; a 10 percent increase in the price of electricity for industry is on average associated with a 2.2 percentage point reduction in energy intensity, ceteris paribus. In addition, the efficiency index is much more sensitive to income and price changes than the structural index, suggesting that both income and price drive intensity change largely through the efficiency rather than the structural channel. Finally, energy intensity increases slightly with the increase in manufacturing trade, implying that trade induces specialization in more energy intensive industries. Overall, the analysis suggests that economic development has played an especially important role in improving energy efficiency. Before the global financial crisis, GDP grew on average by 7 percent annually in ECA countries. This economic growth led to an increase in the demand for manufacturing commodities such as iron and steel. To satisfy increased demand for commodities, countries either increased capacity utilization, or added new, more efficient capacity, thereby reducing the share of smaller, less efficient production units. 18 In contrast, in Central Asia, where production levels had stalled, manufacturers had failed to upgrade to more efficient technologies. Rising energy prices also explains the convergence in energy intensity. Industrial electricity prices on average increased by 12 percent during 1998-2008 in ECA. Higher energy prices cause energy intensity to fall by encouraging the adoption of energy-saving technologies—especially in sectors where energy constitutes a significant portion of their total costs. These efforts include the introduction of more energy-efficient machines, production processes or materials 19. Most industrial sectors consist of a relatively small number of large players with similar manufacturing processes. It is therefore the part of the economy best organized to become more energy efficient. Although industry in ECA has become much more efficient over the past decade, the potential for further reducing energy use and associated CO2 emissions remains high. Many ECA countries still use several times more energy per unit of output than their Western partners, as shown in Figure 3. The region as a whole produces 5 percent of manufacturing value added in the world but consumes 9 percent of energy. If all countries adopt best-practice technologies, the energy savings potential is estimated to be 50 percent of current energy use in the energy-intensive sectors (iron and steel, cement, and pulp and paper) and 10-20 percent across the board (Deichmann and Zhang, 2013). Given the old age of many of the region’s 18 One widely reported example is Romania. Benefitting from strong growth in demand, industrialists in Romania rapidly introduced modern technologies into outmoded or newly built facilities; increased the production of commodities; and invested profits into further upgrading and expansion of industrial production. 19 For example, producing clinker from raw material is the main energy-consuming process in a cement factory. Using other synthetic material to substitute for clinker can significantly reduce energy use in cement production. 13 industrial plants, upgrading to energy-efficient technology is likely to be cost effective if undertaken as part of the natural cycle of plant replacement and offers an excellent opportunity for achieving energy savings and associated emission reductions in the near to medium term. Empirical results in this paper provide several policy implications for further improving manufacturing energy efficiency in the region: First, one obvious strategy to improve energy productivity is to target energy efficiency efforts on industries with high energy intensity, high total energy use, and high economic growth. Industries undergoing rapid growth are likely to be less capital-restricted and more receptive to efforts to improve their competitive edge through increased management of energy costs. Second, despite recent progress in energy price reforms, subsidies still persist in many ECA countries, especially in energy exporting countries. For example, Uzbekistan and Turkmenistan respectively spent 28 percent and 23 percent of GDP on energy consumption subsidies in 2011, the highest in the world. The effectiveness of continued energy price reforms in the region will largely determine its potential to save energy. Finally, the lack of energy-use data is one of the stumbling blocks to developing rational energy efficiency strategies and policies. Important data gaps remain, especially in Central Asian countries. For example, more than 80 percent of the industrial energy use is reported as “non- specified” in Kazakhstan, Kyrgyz, Turkmenistan and Uzbekistan. Collection of reliable, timely, comparable and detailed data that go well beyond those currently available should be a priority in these countries to provide the basis for rational policy making. 14 References Ang, B.W. 2004. Decomposition Analysis for Policymaking in Energy: Which is the Preferred Method? Energy Policy 32, 1131-1139. Ang, B.W. and F.Q.Zhang. 2000. A Survey of Index Decomposition Analysis in Energy and Environmental Studies. Energy 25: 1149-1176. Ang, B.W., Liu F.L., Chung, H.S. 2004. A Generalized Fisher Index Approach to Energy Decomposition Analysis. Energy Economics 26: 757-763. Ang, B.W., Xu, X.Y., 2013. Tracking Industrial Energy Efficiency Trends using Index Decomposition Analysis. Energy Economics doi: 10.1016/j.eneco.2013.05.014 Arellano, M. and S. R. Bond. 1991. Some Specification Tests for Panel Data: Monte Carlo Evidence and an Application to Employment Equations. Review of Economic Studies 58:277-298. Barro, R. J. and X. Sala-I-Marin. 1992. Convergence. Journal of Political Economy 100(2): 223- 251. Bernstein, M., Fonkych, K., Loeb. S. and Loughran, D. 2003. State-level Changes in Energy Intensity and their National Implications. Rand Publishing. Santa Monica, CA. Boyd G.A. and J. M. Roop. 2004. A Note on the Fisher Ideal Index Decomposition for Structural Change in Energy Intensity. The Energy Journal 25(1): 87-101. Cleveland, C. J. Kaufmann, R. K. and Stern D. I. 2000. Aggregation and the Role of Energy in the Economy. Ecological Economics 32: 301-318. Deichmann, U. and F. Zhang. 2013 Growing Green: the Economic Benefits of Climate Action. World Bank Publication. Washington DC. 2013. Dollar, D. and E. Wolff. 1993. Competitiveness, Convergence and International Specialization. MIT Press. EBRD (European Bank of Reconstruction and Development). 2010. Transition Report 2010: Recovery and Reform. Report of the Office of the Chief Economist, EBRD Publications, London. EBRD. 2011. The Low Carbon Transition. Special Report on Climate Change. EBRD Publications. London. European Commission. 2011. EU Industrial Structure: Trends and Performance. Fankhauser, S. and Cornillie, J. 2004. The Energy Intensity of Transition Countries. Energy Economics. V(26): 283-295. 15 Field, B. and Grebenstein, C. 1980. "Capital- Energy Substitution in U.S. Manufacturing." Review of Economics and Statistics, V.62: 207-12. Fisher, I. 1921. The Best Form of Index Number. Quarterly Publications of the American Statistical Association 17 (133): 533-537 Grossman, G. M. and Helpman E. 1991. Innovation and Growth in the Global Economy, Cambridge, MA. MIT Press. Havrylyshyn, O., I. Izvorski, and R. Rooden. 1998. “Recovery and Growth in Transition Economies 1990–97: A Stylised Regression Analysis.” Working Paper WP/98/141, International Monetary Fund, Washington, DC. Howarth, R.B., L. Schipper, P.A. Duerr and S. Strom. 1991. Manufacturing Energy Use in 8 OECD Countries. Energy Economics 13: 135-142. IEA (2012), "World energy statistics", IEA World Energy Statistics and Balances (database). Krugman, P. 1987. The Narrow Moving Bank, the Dutch Disease, and the Competitive Consequences of Mrs. Thatcher. Journal of Development Economics, 27(1-2): 41-55. Liu, F.L., Ang, B.W. 2003. Eight Methods for Decomposing Aggregate Energy Intensity of Industry. Applied Energy 76: 15-23. Liu, N., Ang, B.W. 2007. Factor Shaping Aggregate Energy Intensity Trend for Industry: Energy Intensity versus Product Mix. Energy Economics 29: 609-635. Metcalf, E. G. 2008. An Empirical Analysis of Energy Intensity and Its Determinants at the State Level. The Energy Journal. 29 (3): 1-26. Markandya, A., Pedroso-Galinato, S., Streimikiene, D., 2005. Energy Intensity in transition economies: Is there convergence towards the EU average? Energy Economics. 28 (1): 121-145. Mulder, P. and H.L.,de Groot. 2007. Sectoral Energy- and Labor-Productivity Convergence. In L. Bretschger and S. Smulders (eds.), Sustainable Resource Use and Economic Dynamics, 165-190. Myers, N. and Kent. J. 2001. Perverse subsidies: how tax dollars can undercut the environment and the economy. Island Press. Solow, J.L. 1987. The Capital-Energy Complementary Debate Revisited. American Economic Review. V.77(40): 605-614. Uanander, F., S. Karbuz, L. Schipper, M. Khrushch and M. Ting. 1999. Manufacturing Energy Use in OECD Countries: Decomposition of Long Term Trends. Energy Policy 27:769-778. 16 U.S. Congress, office of Technology Assessment, 1993. Energy Efficiency Technologies for Central and Eastern Europe, OTA-E-562. Washington, DC. World Bank. 2008. “Energy Efficiency in Russia: Untapped Reserves.” Study report. World Bank Group, Washington, DC. 17 Figure 1 Decoupling of Energy Consumption from Manufacturing Output a. Eastern Europe and Central Asia (excluding Russia and Serbia) 1998-2008 250 700 650 200 600 150 550 500 100 450 400 50 350 0 300 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 Energy Consumption (Million tons of oil equivalence) (left axis) Manufacturing Value Added (2005 PPP billion dollars)(right axis) b. Eastern Europe and Central Asia (including Russia and Serbia) 2002-2008 450 1050 350 950 250 850 150 750 50 650 2002 2003 2004 2005 2006 2007 2008 Energy Consumption (Million tons of oil equivalence) (left axis) Manufacturing Value Added (2005 PPP billion dollars)(right axis) Note: Manufacturing value added data are not available for Russia and Serbia before 2002. PPP = purchasing power parity 18 Figure 2 Manufacturing Energy Intensity in Eastern Europe and Central Asia, by subregion/country, 1998–2008 a. By Sub-region Central Asia .8 Energy Intensity (Kgoe/PPP$) .6 Neighborhood .4 Russia EU10 .2 Accession Turkey 0 1998 2000 2002 2004 2006 2008 Note: kgoe = kilograms of oil equivalent. PPP = purchasing power parity. EU-10 countries include Bulgaria, the Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Romania, the Slovak Republic, and Slovenia. EU accession countries in this analysis include Albania, Bosnia and Herzegovina, Croatia, the Former Yugoslav Republic of Macedonia, and Serbia. Neighborhood states include Armenia, Azerbaijan, Belarus, Georgia, Moldova, and Ukraine. Central Asian nations include Kazakhstan, the Kyrgyz Republic, Tajikistan, Turkmenistan, and Uzbekistan. b. By country in Central Asia 2 Energy Intensity (Kgoe/PPP$) 1.5 Uzbekistan 1 Kazakhstan Kyrgyz Republic .5 Tajikistan 0 Turkmenistan 1998 2000 2002 2004 2006 2008 19 Figure 3 σ-convergence in Manufacturing Energy Intensity in ECA, 1998-2008 .6 Energy Intensity (kgoe/PPP$) .5 .4 .3 .2 .1 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 ECA mean (unbalanced panel) ECA Std.Dev.(unbalanced panel) ECA mean (balanced panel) ECA Std.Dev.(balanced panel) EU15 mean Note: Value added data are not available for Albania during 2004-2006, Romania in 2002, Russia during 1998-2001, Serbia during 1998-1999 and Turkmenistan during 2005- 2008. The balanced panel excludes observations of the five countries. The unbalanced panel includes all 28 countries. 20 Figure 4 Fisher Ideal Indexes for Manufacturing Energy Intensity, Structure and Energy Efficiency 1.2 1 .8 .6 .4 2001 2002 2003 2004 2005 2006 2007 year Intensity Index Structural Index Efficiency Index Note: The chart is based on index decomposition analysis of manufacturing energy intensity of the following countries during 2001-2007: Albania, Azerbaijan, Bulgaria, Czech Republic, Estonia, Georgia, Hungary, Latvia, Lithuania, Macedonia, Moldova, Poland, and Russia. 21 Table 1 Country-level energy intensity, structure and efficiency indexes Country Index 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 Albania Structure -- -- 1 1.051 1.505 0.969 1.038 0.416 0.463 0.442 0.866 Efficiency -- -- 1 0.927 0.595 0.707 0.485 1.198 0.830 0.734 0.375 Intensity -- -- 1 0.973 0.896 0.685 0.503 0.498 0.384 0.324 0.325 Azerbaijan Structure -- -- -- 1.000 1.055 0.930 0.977 1.124 1.090 1.271 1.135 Efficiency -- -- -- 1.000 0.711 0.605 0.468 0.412 0.384 0.180 0.200 Intensity -- -- -- 1.000 0.749 0.563 0.458 0.463 0.418 0.228 0.227 Bulgaria Structure 1 0.931 1.141 1.013 0.882 0.961 1.047 0.968 0.993 1.056 -- Efficiency 1 0.912 0.766 0.970 1.063 0.901 0.756 0.757 0.693 0.574 -- Intensity 1 0.849 0.874 0.982 0.937 0.866 0.792 0.733 0.689 0.606 -- Czech Republic Structure 1 0.931 0.958 0.957 0.917 0.917 0.945 0.930 0.893 0.666 -- Efficiency 1 0.909 0.887 0.873 0.813 0.706 0.647 0.683 0.602 0.978 -- Intensity 1 0.847 0.850 0.835 0.746 0.647 0.612 0.635 0.537 0.651 -- Estonia Structure -- -- 1 1.041 1.044 1.017 1.043 1.058 1.029 1.038 -- Efficiency -- -- 1 1.080 0.878 0.889 0.855 0.870 0.763 0.816 -- Intensity -- -- 1 1.124 0.917 0.905 0.891 0.921 0.785 0.848 -- Georgia Structure -- -- 1 1.182 1.243 0.955 1.254 1.142 1.569 1.736 1.487 Efficiency -- -- 1 0.491 0.556 0.713 0.439 0.390 0.297 0.242 0.292 Intensity -- -- 1 0.580 0.691 0.681 0.550 0.446 0.467 0.420 0.435 Hungary Structure 1 0.917 0.979 0.929 0.874 0.902 0.901 0.874 0.923 0.941 -- Efficiency 1 1.076 0.964 0.996 1.074 0.921 0.896 0.884 0.804 0.830 -- Intensity 1 0.987 0.944 0.925 0.939 0.831 0.807 0.772 0.741 0.781 -- Latvia Structure -- -- 1 1.000 0.975 1.045 1.128 1.013 1.088 1.065 -- Efficiency -- -- 1 1.001 0.923 0.804 0.759 0.856 0.735 0.691 -- Intensity -- -- 1 1.001 0.900 0.840 0.856 0.867 0.800 0.736 -- Lithuania Structure -- -- 1 1.159 1.116 1.083 1.052 1.159 1.314 1.501 1.525 Efficiency -- -- 1 0.790 0.826 0.730 0.642 0.579 0.538 0.420 0.429 Intensity -- -- 1 0.916 0.922 0.790 0.675 0.671 0.707 0.631 0.654 Macedonia Structure 1 1.009 1.124 1.049 0.746 0.710 1.135 0.720 0.704 -- -- Efficiency 1 0.745 0.743 0.758 1.454 1.351 0.868 1.261 1.212 -- -- Intensity 1 0.752 0.836 0.795 1.085 0.959 0.985 0.908 0.853 -- -- Moldova Structure -- -- -- 1 1.008 1.026 0.994 0.996 0.937 0.982 0.993 Efficiency -- -- -- 1 0.879 0.838 1.079 1.120 1.089 0.915 0.896 Intensity -- -- -- 1 0.886 0.859 1.073 1.115 1.020 0.898 0.890 Poland Structure -- -- 1 1.112 0.895 0.878 0.988 0.925 1.101 1.109 -- Efficiency -- -- 1 0.723 0.940 0.854 0.663 0.659 0.609 0.542 -- Intensity -- -- 1 0.803 0.841 0.749 0.655 0.610 0.671 0.601 -- Russia Structure -- -- -- 1 0.991 1.027 1.055 0.978 0.994 1.004 0.985 Efficiency -- -- -- 1 1.140 0.918 0.830 0.791 0.720 0.712 0.691 Intensity -- -- -- 1 1.129 0.943 0.875 0.774 0.716 0.715 0.680 22 Table 2 Summary Statistics of Variables Variable Obs. Mean Std. Dev. Min Max Energy Intensity (kgoe/ppp $) 116 0.26 0.28 0.05 1.74 Electricity price (US cents/kWh) 106 6.01 2.64 2.11 19.02 P.C. GDP (2005 PPP) 120 9490 4416 2123 18330 Capital-labor ratio (%) 120 4.79 4.26 0.22 36.41 Share of exports and imports in manufacturing (%) 116 3.38 1.37 0.86 6.95 23 Table 3 Determinants of Manufacturing Energy Intensity (1) (2) (3) (4) OLS Fixed Effects Random Effects GMM Ln (PCGDP) -0.0254 -1.388*** -0.025 -0.795** (0.123) (0.341) (0.120) (0.383) Ln (Industrial Electricity Price) -0.670*** -0.215*** -0.670*** -0.048 (0.143) (0.074) (0.145) (0.082) Ln(Capital Labor Ratio) -0.480** 0.398*** -0.480*** 0.178 (0.189) (0.091) (0.129) (0.131) Ln (Capita Labor Ratio)^2 0.072 -0.031** 0.072 -0.037 (0.062) (0.024) (0.049) (0.036) Ln(Share of Trade in Manufacturing) 0.149 0.636** 0.149 0.245* (0.122) (0.107) (0.123) (0.137) Ln(Energy Intensity)t-1 0.490*** (0.108) R2 0.546 0.145 0.546 Sargan test (p-value) 0.581 Obs. 102 102 102 70 Note: Columns (1), (2), and (3) report results from estimating determinants of manufacturing energy intensity identified in equation (7) via OLS, fixed effects and random effects models. Column (4) reports the difference-GMM results, using the log of per capita GDP (lnIit) and the log of electricity prices (lnPit) lagged 2 to 4 periods in levels as instruments. Year fixed effects are not reported. Standard errors adjusted for clustering on country are reported in parentheses. *** indicates significant at the 1 percent level. ** indicates significant at the 5 percent level. * indicates significant at the 10 percent level. Table 4 Determinants of Energy Intensity, Efficiency and Structure Indexes (1) (2) (3) Intensity Efficiency Structure Ln (PCGDP) -1.074*** -1.086*** 0.012 (0.202) (0.228) (0.219) Ln (Industrial Electricity Price) -0.351*** -0.361*** 0.010 (0.106) (0.121) (0.116) Ln(Capital Labor Ratio) -0.176 0.173 -0.349* (0.170) (0.193) (0.185) Ln (Capita Labor Ratio)^2 0.138 0.028 0.110 (0.064) (0.073) (0.070) Ln(Share of Trade in Manufacturing) 0.389*** -0.148 0.537*** (0.116) (0.119) (0.114) R2 0.076 0.128 0.001 Obs. 68 68 68 Note: Columns (1), (2), and (3) report results from estimating determinants of manufacturing energy intensity index, energy efficiency index and structural index identified in equation (7) based on a fixed effects model. Year fixed effects are not reported. Standard errors are reported in parentheses. *** indicates significant at the 1 percent level. ** indicates significant at the 5 percent level. * indicates significant at the 10 percent level. 24