HOW DOES ENERGY EFFICIENCY AFFECT THE GROWTH IN THE REAL SECTOR? EVIDENCE FROM UKRAINIAN REGIONS Borys Dodonov Senior Economist Bureau of Economic and Social Technologies 3A Desyatynna Str., 01025, Kyiv, Ukraine Tel. +380 99 501 8320, email. bdodonov@gmail.com June, 2013 Abstract This paper analyzes the link between energy efficiency and growth in the real sector. The impact of energy efficiency on industrial growth is studied using a four-year panel dataset containing information on 26 regions and 10 manufacturing industries. In order to avoid endogeneity bias I employ difference-in difference approach described in Rajan and Zingales (1998). I find statistically significant positive effect of energy efficiency on growth, the magnitude of which depends on energy intensity of the industry. The results are robust to different model specifications. KEYWORDS: Energy Efficiency, Energy Intensity, Industrial Growth JEL CLASSIFICATIONS: 013, 014, 047, Q43 2013 The proposal for the 32rd United States Association for Energy Economics meeting, 27 31 of July,
1 Introduction After the drastic energy price increase over the last decade the energy efficiency became a precondition for industrial competitiveness in the modern world. Unfortunately, the empirical evidence on the link between energy efficiency and growth in the real sector is still rather limited. This study aims at partially filling this gap for manufacturing industries. Some industries are more energy intensive than the others due to their technological characteristics. Therefore, the share of energy inputs in total costs varies substantially across manufacturing industries. For example, energy expenditures are much larger for the steel and cement industries than, e.g., for the food or machinery. If the firm is more energy efficient than its counterpart it has lower total costs and larger opportunities for capital expenditures. Thus, it is expected to grow faster. So energy efficiency should disproportionatly help energy intensive industries for their growth. Therefore, I expect that energy efficiency positively affects industrial growth and the effect increases with energy intensity of the industry. To test this hypothesis and avoid endogeneity bias I employ difference-in difference approach described in Rajan and Zingales (1998). The impact of energy efficiency on industrial growth is studied using a four-year panel dataset containing information on 26 regions and 10 manufacturing industries. I find statistically significant positive effect of energy efficiency on growth, the magnitude of which depends on energy intensity of the industry. The results are robust to different model specifications. 1
2 Methodology 2.1 Estimation strategy I employ the difference in difference approach proposed by Rajan and Zingales (1998) who studied the effect of financial development on industrial growth. This approach allows avoiding the typical for growth regressions omitted variable and model specification biases and determining the impact of the independent variable on growth based on its exogenous variation. 1 Following their approach I rank the industry by energy intensity (using the data for the average intensity in the EU that is exogenous to Ukraine) and then study the growth of industries depending on their energy efficiency in the region and energy intensity. The regional and industry characteristics are captured by dummy variables for each region and industry. The primary variable of interest is the interaction between industry s energy efficiency in each region of Ukraine and its energy intensity in the EU. Although energy efficiency is likely to be endogenous due to omitted variable bias, its interaction with the average intensity in the EU is exogenous. The only additional explanatory variables included that vary both with industry and region are the industry s share in manufacturing and capital expenditures in 2007. Thus, the model we need to estimate is then growth ij = constant + β 1...k Region Dummies + β k+1...m Industry Dummies + β m+1 (Industry s j s share of manufacturing in region i in 2007) + β m+2 (Efficiency of industry j in region i) (Intensity of industry j in the EU) + u ij, (1) where subscripts i and j denote region and industry respectively. growth it is the growth 1 See Angrist and Krueger (1999) and Bertrand et al. (2004) for detailed difference in-difference method review. 2
rate of industry s real value added or real capital expenditures, Efficiency is the energy efficiency of the industry j in the particular region i of Ukraine and Intensity is the index of energy intensity of industry j in the EU. Energy efficiency at industry level is calculated according to International Energy Agency (IEA) method of final energy consumption decomposition. The average energy consumption of the same sub sectors in the EU is proposed to use as the energy efficiency benchmark. Most EU countries are already more energy efficient than Ukraine would be in 2030 according its Energy Strategy to 2030. Thus, current EU energy consumption should be a good benchmark for an efficient energy consumption in Ukraine. 2 3 Data In my analysis I use final energy consumption data in 10 manufacturing industries defined by International Standard Industry Classification Code (ISIC) 3.1 food and tobacco (15-16), textile and leather (17-19), wood (20), paper and printing (21-22), chemical industry (24), non metallic minerals (26), primary metals (27), machinery (28-32), transport equipment (34 35) and others (sum of plastic and rubber (25), medical equipment (33) and furniture (36)). 3 I defined the energy intensity (EI) for each industry as final energy consumption (FEC) per Euro2000 of value added (VA) adjusted for purchasing power parity EI j = F EC j V A j. The estimation employs annual data from 2007 to 2010 for 27 Ukrainian regions and the European Union. The data on energy consumption, value added and capital expenditures for each region and industry are obtained from State Statistical Service of Ukraine. 2 See Dodonov et al. (2012) for details. 3 This choice is dictated by the availability of energy intensity data for the manufacturing industries in the EU. 3
The energy efficiency of manufacturing industries at regional level in Ukraine were taken from Dodonov et al. (2012) and transformed into a normalized intensity index. The energy intensities of EU manufacturing industries were retrieved from ODYSSEE database. The database contains energy and economic data on the 27 EU member countries plus Croatia and Norway. The data on value added in the ODYSSEE database are presented in euros as of 2000. The value added data for Ukraine were converted into euros as of 2000 on a purchasing power parity basis in order to ensure correct comparison of energy consumption efficiency. To this end, exchange rate data, adjusted for the purchasing power parity (from the World Bank 4 and OECD 5 ) and the Eurozone GDP deflator (from the European Central Bank 6 ), were used. Summary statististics are reported in Table 1. 4 Results Table 2 reports the estimates of the basic specification (1). The coefficient estimate for the interaction term is positive and statistically significant at conventional 5% level. 7 Thus, we might conclude that energy efficiency positively affects the real sector growth and the effect grows for industries that are energy intensive due to the technological characteristics. In order to get the sense of its magnitude consider two biggest regions in Ukraine (Donentsk and Dnipropetrovsk) and two biggest industries in terms of value added (primary metals and food). The primary metals is the most energy intensive industry in the EU with an energy intensity index equal to one. The correspondent figure for food industry is only 0.13. The energy efficiency of primary metals and food industry are 47% and 53% from the EU level in Donetsk and 36% and 40% in Dnipropetrovsk region respectively. The coefficient estimate then predict that primary metals will grow 4 http://data.worldbank.org/country/ukraine 5 http://stats.oecd.org/index.aspx?datasetcode=cpl 6 http://sdw.ecb.europa.eu/browse.do?node=2120780 7 The robust to heteroskedasticity standard errors are reported in the parenthesis. 4
5.2% faster in Donetsk than in Dnipropetrosk region but food industry will grow only 0.8% faster despite bigger difference in energy efficiency. The difference in growth rates of capital expenditures attributed to differences in energy efficiency is substantially larger than for growth rates of value added (18.6% and 2.8%). Table 3 contains the estimates of the model on the annual growth data. The coefficient estimates for interaction term are statistically significant at conventional level of significance, they are larger in absolute values but standard errors are larger as well. 5 Conclusion The empirical evidence on the link between energy efficiency and growth in the real sector is still rather limited. This study aims at partially filling this gap for manufacturing industries. Some industries are more energy intensive than the others due to their technological characteristics. Thus, industries using more energy resources per unit of value added due to technological characteristics of the industry (e.g. steel) would gain more from energy energy efficiency than less energy intensive industries (e.g. food or machinery). In order to test this hypothesis and avoid typical for growth regressions endogeneity bias I employ difference-in difference approach described in Rajan and Zingales (1998). I find statistically significant positive effect of energy efficiency on growth of real value added and capital expenditures, the magnitude of which depends on energy intensity of the industry. The results are robust to different model specifications. References Angrist, Joshua and Alan Krueger, Empirical Strategies in Labor Economics, in Orley Ashenfelter and David E. Card, eds., Handbook of Labor Economics, Vol. 3, New York: Elsevier, 1999, pp. 1277 1366. 5
Bertrand, Marianne, Esther Duflo, and Sendhil Mullainathan, How Much Should We Trust Differences-in-Differences Estimates?, Quarterly Journal of Economics, 2004, 119 (1), 249 275. Dodonov, Borys, Valeriy Gladkiy, Yuriy Gladkyy, and Ganna Tsarenko, Energy Efficiency Rankings of the Regions of Ukraine, Donetsk, Ukraine: System Capital Management, 2012. Rajan, Raghuram G. and Luigi Zingales, Financial Dependence and Growth, American Economic Review, 1998, 88 (3), 559 586. 6
Table 1: Descriptive Statistics, 2007 2010 Variable Obs. Mean Std. Min Max Real Value Added Growth Growth 615.1400.4166 2.4695 2.3585 Real Value Capital Expenditure 482.0050.9002-3.4158 3.6156 Energy efficiency of the Industry 828.4404.2807.02 1.24 Manufacturing Energy Intensities in the EU 10.2231.2336.03 1.16 Industry j s share of manufacturing 828.1173.1292.0004.6137 in region k in 2007 Efficiency*Intensity EU 828.0854.0978.0007.71 7
Table 2: Energy Efficiency and Manufacturing Growth at Regional Level, average growth rates The dependent variables are the average growth rate of real value added in column 1 and the average growth rate of real capital expenditures in column 2 (logarithmic change) for the period 2007 2010 for each ISIC manufacturing industry in each region. (1) (2) Industry j s share of manufacturing 0.1799 0.5681 in region k in 2007 (0.137) (0.364) Efficiency*Intensity 0.4594** 1.6503** (0.227) (0.705) Region fixed effects YES YES Industry fixed effects YES YES R2 Adj 0.212 0.129 Number of observations 219 170 Notes:, and denotes statistical significance at the 1% level, 5% level and 10% level respectively, s.e. corrected for heteroscedasticity are presented in parenthesis. 8
Table 3: Energy Efficiency and Manufacturing Growth at Regional Level, annual growth rates The dependent variables are the annual growth rate of real value added in column 1 and the annual growth rate of real capital expenditures in column 2 (logarithmic change) for the period 2007 2010 for each ISIC manufacturing industry in each region. (1) (2) Industry j s share of manufacturing 0.9198*** 0.6596* in region k (0.183) (0.353) Efficiency*Intensity 0.8660*** 1.0457* (0.308) (0.617) Region fixed effects YES YES Industry fixed effects YES YES R2 Adj 0.083 0.013 Number of observations 615 482 Notes:, and denotes statistical significance at the 1% level, 5% level and 10% level respectively, s.e. corrected for heteroscedasticity are presented in parenthesis. 9