Explaining the Energy Consumption Portfolio in a Cross-Section of Countries: Are the BRICs Different?

Board of Governors of the Federal Reserve System International Finance Discussion Papers IFDP 1015 February 2011 Explaining the Energy Consumption Portfolio in a Cross-section of Countries: Are the BRICs Different? David M. Arseneau NOTE: International Finance Discussion Papers are preliminary materials circulated to stimulate discussion and critical comment. References in publications to International Finance Discussion Papers (other than an acknowledgment that the writer has had access to unpublished material) should be cleared with the author or authors. Recent IFDPs are available on the Web at www.federalreserve.gov/pubs/ifdp/. This paper can be downloaded without charge from Social Science Research Network electronic library at http://www.sssrn.com.

Explaining the Energy Consumption Portfolio in a Cross-section of Countries: Are the BRICs Di⁄erent? (cid:3) David M. Arseneau y Federal Reserve Board First Draft: October 2010 This Draft: February 2011 Abstract This paper uses disaggregated data from a broad cross-section of countries to empirically assess di⁄erences in energy consumption pro(cid:133)les across countries. We (cid:133)nd empirical support for the energy ladder hypothesis, which contends that as an economy develops it transits away from a heavier reliance on traditionalfuelsourcestowardsanincreaseintheuseofmoderncommercialenergy sources. We also (cid:133)nd empirical support for the hypothesis that structural transformation(cid:151)theideathatasaneconomymatures, ittransformsawayfrom agriculture-basedactivityintoindustrialactivityand,(cid:133)nally,fullymaturesinto a service-oriented economy(cid:151)is an important driver for the distribution of enduse energy consumption. However, even when these two hypotheses are taken into account, we continue to (cid:133)nd evidence suggesting that the patterns of energy consumption in the BRIC economies are importantly di⁄erent from those of other economies. JEL Classi(cid:133)cation: Q41; Q43 Keywords: Energy and development; Energy ladder hypothesis; Structural transformation This paper has bene(cid:133)ted from helpful comments from Neil Ericsson. (cid:3) The views expressed in this paper are those of the author and do not represent those of the y Board of Governors of the Federal Reserve System or other members of its sta⁄. E-mail address: david.m.arseneau@frb.gov. 1

1 Introduction Oneofthede(cid:133)ningcharacteristicsofglobalenergymarketsoverthepastdecadeisthe rapid growth of energy consumption in the emerging market economies. The 2010 Annual BP Statistical Review of World Energy shows that over the past ten years the average annual growth rate of global total (cid:133)nal energy consumption was just under 1 percent. Overthisperiod, energyconsumptioninOECDeconomiesdeclinedslightly. In contrast, the emerging market economies experienced a collective growth rate of roughly 2 percent, making it clear that the developing world has been the primary engine for global energy consumption growth. Moreover, much of this growth was concentrated in just four countries(cid:151)the so-called BRIC economies of Brazil, Russia, India, and China. These four economies accounted for approximately half of the growth in emerging markets taken as a whole over the past decade. This growth di⁄erential has potentially important implications for global energy markets going forward. Existing research suggests that the dynamics of energy consumption in emerging market economies are importantly di⁄erent fromthe developed world.1 If the growth di⁄erentials observed over the past ten years persist, the resulting shift in the distribution of global consumption could give rise to a markedly di⁄erent energy landscape; one that is much more heavily weighted toward developments in the emerging markets. In light of this, understanding the behavior of energy consumption in the emerging markets(cid:150)and in the BRICs in particular(cid:150)is an increasingly pressing priority for energy economists. Existing literature has made some strides in this direction, however it very much remains an open area of research. This paper moves in this direction by using disaggregated micro-level data to examine energy consumption patterns in a wide cross-section of countries. We constructadatasetdetailingenergyusagein35di⁄erentcountrieswhich, takentogether, comprise roughly 80 percent of global total (cid:133)nal energy consumption. These data are then used to empirically assess two alternative theoretical explanations for why energy consumption portfolios di⁄er across countries. In order to do this we examine the data from two separate dimensions. The (cid:133)rst is what we refer to as the fuel intensity pro(cid:133)le, which describes the fraction of energy consumption, either at the aggregate level or disaggregated at the sectorial- or industry-level, derivedfromagivensourcefuel. Here, weareinterestedinidentifying characteristics that make a country more (or less) reliant on a speci(cid:133)c fuel source for energy generation. The so-called (cid:147)energy ladder hypothesis(cid:148)o⁄ers a theoretical guide around which we organize our empirical investigation. This hypothesis contends that as the level of economic development in a country rises, substitution takes place away from traditional biomass, including wood and agricultural and animal waste, as a primary fuel 1See, for example, Gately and Huntington (2002) and Dargay, Gately and Huntington (2007), document notable di⁄erences in oil and/or energy consumption dynamics across di⁄erent subsets of countries. 2

source and into more modern, cheaper, and cleaner (less polluting) energy sources such as natural gas, oil and petroleum products, and electricity.2 This transition along the energy ladder occurs not only in residential usage, but also in industrial, commercial, and agricultural usage as technologies and physical infrastructure for energy generation using these fuels become more widespread.3 To test this hypothesis we exploit the systematic variance between fuel intensity pro(cid:133)les and the level of economic development. In particular, the fuel intensity pro(cid:133)le should vary in such a way that higher income countries tend to rely more heavily on higher quality, cleaner fuels. In fact, this is exactly what we (cid:133)nd in the data(cid:150)bothintheaggregatedataaswellasthedisaggregateddataatboththesectorand industry-level. Thus, the (cid:133)rst main result of this paper is that there is strong empirical support for the energy ladder hypothesis as a determinate of a countries(cid:146) fuel intensity portfolio. The second dimension we explore is a countries(cid:146)end-use consumption pro(cid:133)le, whichdescribesthefractionoftotalenergyconsumedinagivensectoroftheeconomy or, at a more disaggregated level, in a given industry within a sector. Along this dimension, the goal is to identify characteristics that lead to a country to consume a higher (or lower) fraction of total energy in one particular sector of the economy relative to other countries. Our empirical investigation here is guided by the so-called (cid:147)structural transformation hypothesis(cid:148), which keys o⁄ the widely accepted view that an economies(cid:146) industrial structure changes endogenously as it undergoes the process of economic development.4 Economic activity in underdeveloped countries tends to be focused mainly in agriculture. However, as a country grows agricultural activity gives way to industry as a country begins to develop. At later stages of development, once industrialization is complete industrial activity tends to decline as the process of development transforms the economy toward more service-oriented activity. This shift in the composition of the economy implied by the process of structural transformation has implications for patterns of end-use energy consumption.5 We test these implications at both the sector- and industry-level and we (cid:133)nd that, in general, the data are supportive of the structural transformation hypothesis. Thus, the second main result of the paper is that the process of structural transformation is an important determinant of a countries(cid:146)end-use consumption pro(cid:133)le. 2Hosier and Dowd (1987), Leach (1992), Barnes and Floor (1996), Heltberg (2004), and Hosier (2004) all examine the energy ladder hypothesis using micro data on residential usage. 3Gr(cid:252)bler (2004), Bashmakov (2007), Marcotullio, and Schulz (2007) all provide descriptive evidence of how the energy mix changes with economic development. Burke (2010a,b) explicitly tests this hypothesis in two contributions concentrating on the total energy mix and on the electricity mix, respectively. 4The link between economic development and structural change owes to Kuznets (1971). 5Judson, Schmalensee, and Stoker (1999), Medlock and Soligo (2001), and Sch(cid:228)fer (2005) all examine implicaitons of structural change for energy demand from an empirical standpoint. See Arbex and Perobelli (2010) and Stefanski (2010) for some recent theoretical contributions. 3

Backed with these two empirically-relevant theoretical explanations for why and how energy consumption pro(cid:133)les might di⁄er across countries, we next ask the question: Are the BRICs di⁄erent? In short, we (cid:133)nd that they are indeed notably di⁄erent along a number of dimensions. This is an important (cid:133)nding both from the perspective of energy economists trying to understand ongoing market developments as well as from the perspective of policy-makers who ultimately need to deal with the consequences of these developments. As noted above, the BRICs have been a signi(cid:133)cant engine of growth for global energy consumption and are likely to remain so in the future. Accordingly, these economies in particular will play an increasingly important role in shaping the energy landscape of the future. The results of this paper highlight the need for future research to shed more light on energy consumption dynamics(cid:150)both at long-run as well as at cyclical frequencies(cid:150)in the emerging markets, in general, and the BRICs, in particular. A key aspect of this research will inevitably involve delving further into the data at an even more disaggregated level, suggesting that continuing to improve the depth, scope, quality, and ease of dissemination of energy usage statistics should be a top priority. Regarding related literature, one paper in particular deserves further discussion. Using a panel dataset, Burke (2010b) also (cid:133)nds evidence in favor of the energy ladder hypothesis. Along this dimension, we reach a broadly similar conclusion here, thus our (cid:133)ndings can be viewed as complimentary to Burke (2010b). Nevertheless, there are a number of important di⁄erences across the two papers. For example, the two papers reach similar conclusions despite the use of di⁄erent data. While the country coverage in our data is smaller and there is no time series dimension, we exploit data at a more disaggregated level than does Burke (2010b). Data di⁄erences notwithstanding, the key point of di⁄erentiation between the two papers is the focus here on behavior of the BRIC economies as outliers. Theremainderofthepaperisorganizedasfollows. Thenextsectiondiscussesthe data and presents the empirical methodology used to assess the validity of the energy ladder hypothesis to describe cross country di⁄erences in the fuel intensity pro(cid:133)le and the structural transformation hypotheses to explain cross country di⁄erences in the end-use consumption pro(cid:133)le. The main results are presented in Section 3. Section 4 investigates whether or not the energy consumption pro(cid:133)les of the BRIC economies are signi(cid:133)cantly di⁄erent from that of other countries beyond what can be explained bythecorehypothesesoutlinedinsection2. Finally, section5o⁄erssomeconcluding comments as well as some suggested areas for further research. 2 Data and Empirical Methodology Thedatausedintheanalysisconsistofthe2007annualenergyconsumptionportfolios of 35 di⁄erent countries, listed in Table 1, from various geographic regions and levels of economic development. We use the 2007 data because it is the most recently 4

available. Taken together these 35 countries constitute 80 percent of global total (cid:133)nal energy consumption. In what follows, let n be an integer that indexes country, where n [1;35]: All data are obtained from the Energy Balances of OECD and 2 Non-OECD Countries published by the International Energy Administration (IEA). Thesedataarepresentedalongtwoprimarydimensionsforeachofthen countries in the sample. The (cid:133)rst dimension is energy usage by primary fuel source. Let the integer k index primary fuel source, where f [1;6] indicating energy generated 2 from: Combustibles, renewable energy sources, and waste (f = 1); coal and peat (f = 2); crude oil and petroleum products (f = 3); natural gas (f = 4); geothermal, hydroelectric, and/or nuclear energy (f = 5); and electricity (f = 6). The data are also presented along a second dimension of end-use consumption broken out by sector as well as by industry within a given sector. In terms of notation, let the integer s index sector, where s [1;4] indicating energy consumed 2 in the: Industrial sector (s = 1); transportation sector (s = 2); residential and commercial sector (s = 3); and agricultural sector (s = 4). Moving down one level of aggregation,lettheintegeriindexindustrywithinsectors: Intherawdatapresented by the IEA the upper limit of the index i is conditional on the sector of interest. For example, thedatafortheindustrialsectorcanbedisaggregatedintothirteenseparate industries. Similarly, there are six industries within the transportation sector and threewithintheresidentialandcommercialsectorexcludingagriculture, forestry, and (cid:133)shing, which we have chosen to break out as a separate category. When all is said and done, at the most disaggregated level the dataset consists of a (23 6) matrix for every country in the sample, totaling 4;830 individual data (cid:2) points across the entire sample. These data are su¢ ciently detailed to describe, for example, energy derived from coal and peat that is consumed in the iron and steel industry expressed as a fraction of aggregate energy consumption for country n. In the interest of simplicity, as well as for the ease of presentation, we aggregate the industry-level data into just two industries per sector, so that i [1;2] regardless 2 of s: For the industrial sector, we group industries into those that are more energy intensive and those that are less energy intensive based on classi(cid:133)cations presented by the U.S. Department of Energy (DOE).6 The transportation sector is grouped into road transportation and non-road transportation7. Finally, both residential and commercial energy usage are broken out as separate industries. The agricultural sector is not disaggregated further. The resulting condensed dataset is an (6 6) (cid:2) matrix of data for each country in the sample, consisting of 1;260 individual data points. 6The following industries are classi(cid:133)ed as (cid:147)more energy intenstive(cid:148): Iron and steel, chemical and petrochemical, non-ferrous metals, non-metallic minerals, and paper pulp and printing. The remainder, transportation equipment, machinery, mining and quarrying, food and tobbacco, wood and wood products, construction, and textile and leather, are classi(cid:133)ed as (cid:147)less energy intensive(cid:148). 7Road transportation consists of both private and commercial transportation. Non-road transportation consists of domestic aviation, rail, pipeline transport, and domestic navigation. 5

Our goal in the analysis is to explain cross-country di⁄erences in energy consumption portfolios broken out along the two dimensions of fuel source and end-use consumption. Before we provide formal de(cid:133)nitions of the metrics that we will use to empirically describe these two dimensions, some additional notation is useful. At the lowest level of aggregation, let c denote consumption for country n of n;f;s;i fuel f in industry i of sector s. At the other extreme, let C denote aggregate n;;; energy consumption for country n across all fuels and end-use s(cid:1)e(cid:1)c(cid:1)tors, where C n;;; is de(cid:133)ned as: (cid:1)(cid:1)(cid:1) C = c n;;; n;f;s;i (cid:1)(cid:1)(cid:1) f s i XXX Thus, ournotationhasaconsumptionaggregatedenotedbyanuppercaseC : The n;;; subscript n; ; ; reveals that the aggregate is for a given country, n; while t(cid:1)h(cid:1)e(cid:1) (lack (cid:1) (cid:1) (cid:1) of a) dots ( ) reveals the level of aggregation. Generally speaking, a dot in place of a (cid:1) given subscript n; f; s; or i means that we have aggregated over that dimension, so more dots in the subscript implies a higher level of aggregation. For example: C n;;; is aggregate consumption summed over all fuels, f; sectors, s; and industries, i; C (cid:1)(cid:1)(cid:1) n;f;; is consumption by fuel f aggregated across all sectors, s, and industries, i; C i(cid:1)s(cid:1) n;;s; consumption by sector s aggregated across all fuels, f, and industries, i; C (cid:1) (cid:1) is n;f;s; consumption by fuel f in sector s aggregated across all industries; i; and so forth.(cid:1) With this notation in mind, we turn now to a formal de(cid:133)nition of the variables of interest and a description of the empirical models that will be used to explain them. 2.1 Fuel Intensity Portfolio The empirical metric used to summarize the energy portfolio along the fuel source dimension is fuel intensity. We aim to explain the cross-country variation in fuel intensity at three di⁄erent levels of aggregation. Aggregate fuel intensity is simply a measure of the share of aggregate energy consumption accounted for by fuel f aggregated across all sectors and industries for country n. A formal de(cid:133)nition is as follows: c n;f;s;i C AFI = n;f; (cid:1) ; (cid:1) = X s X i C c n;;; (cid:1)(cid:1)(cid:1) n;f;s;i f s i XXX where: AFI denotes aggregate fuel intensity; C denotes aggregate energy conn;f;; sumption accounted for by fuel f across all secto(cid:1)r(cid:1)s and industries; and C is n;;; aggregate energy consumption across all fuels, sectors, and industries. (cid:1)(cid:1)(cid:1) Disaggregating one level gives sector-level fuel intensity, which measures the share of energy consumption in sector s accounted for by fuel f, formally de(cid:133)ned as: 6

c n;f;s;i C SFI = n;f;s; (cid:1) = X i C c n;;s; (cid:1) (cid:1) n;f;s;i f i XX where: SFI denotes sector-level fuel intensity; C denotes energy consumption n;f;s; in sector s accounted for by fuel f across all industrie(cid:1)s, i; and C is energy conn;;s; sumption within sector s across all fuels and industries. (cid:1) (cid:1) Finally, the lowest level of aggregation gives industry-level fuel intensity, which measures the share of energy consumption in industry i of sector s accounted for by fuel f: A formal de(cid:133)nition follows: c c n;f;s;i n;f;s;i IFI = = C c n;;s;i (cid:1) n;f;s;i f X where: IFI denotes industry-level fuel intensity; c denotes energy consumption n;f;s;i in industry i of sector s accounted for by fuel f; and C is energy consumption n;;s;i within industry i of sector s across all fuels. (cid:1) Note that the three indices, AFI; SFI; and IFI; are normalized di⁄erently. The aggregate index is created by normalizing with total energy consumption. It can address the intensity of coal usage in aggregate energy consumption, for example. The sectorial(cid:150)level index is created by normalizing by total energy consumption within the sector. It measures the intensity of oil usage within the industrial energy consumption, for example. Finally, the industry-speci(cid:133)c index is created by normalizing by total energy consumption within an industry speci(cid:133)c to a given sector. It addresses the use of renewables and was in the non-energy intensive industrial sector, for example 2.1.1 Empirical Model The goal is to explain the portfolio of fuel intensity at each of three levels of aggregation for a given country. At the aggregate level, our analysis aims at explaining, for example, why India is more reliant on combustibles, renewables, and waste for energy generation than is either Brazil or Germany. At lower levels of aggregation, the point of our analysis is to identify country characteristics that can help to explain the di⁄erence between the fuel intensity portfolios in two di⁄erent countries at the sector level(cid:151)why Mexico uses more energy generated from oil and petroleum products and less energy generated from coal and peat than does the U.S. Going one step further, we would also like to explain cross-country di⁄erences at the industry level within a given sector. There are two primary hypotheses for structural factors that might be important in determining the fuel intensity pro(cid:133)le for a given country, regardless of the level 7

of disaggregation of the data. First, resource endowment is likely to be important. All else equal, countries that are rich in coal reserves, such as the U.S., are likely to use coal more intensely to meet domestic energy demand at all levels of aggregation relative to a country where coal is relatively scarce. A similar case can be made for oil; recent experience in Saudi Arabia, where the use of crude oil for electricity generation is increasingly frequent, stands out as a case in point. A less dramatic, but equally relevant, example is the extensive use of natural gas in Russia. In the most simple terms, exploiting domestically abundant energy resources is desirable for both economic as well as political reasons and we would expect a countries(cid:146)fuel intensity pro(cid:133)le to re(cid:135)ect this. The second hypothesis for the determinates of a given countries(cid:146)fuel intensity pro(cid:133)lerelatestothelevelofeconomicdevelopment. Existingresearchhasdrawnlinks between economic development and the development of energy infrastructure. This is commonly referred to in the literature as the (cid:147)energy ladder(cid:148)whereby economic development leads tomaturationinthe technologyavailable forenergyprovision. As a country develops it cycles from relatively ine¢ cient fuels, such as combustibles, to more e¢ cient fuels such as coal and, eventually, matures to the current technological frontier in energy provision, exploiting re(cid:133)ned fuels derived from petroleum as well as natural gas and electricity. We test these two candidate hypothesis to explain cross-country di⁄erences in fuel intensity pro(cid:133)les using the following regression framework FI = (cid:12)f +(cid:12)fENDOW +(cid:12)fRGDP +(cid:12)fREGION +" (1) 0 1 n;f 2 n 3 n n where: FI is a fuel intensity measure de(cid:133)ned at one of the three levels of aggregation (that is, in our empirical analysis FI is given by one of the three variables AFI; SFI; or IFI de(cid:133)ned in the previous section depending on the level of disaggregation desired) for fuel f in country n; ENDOW is the share of global proved reserves n;f for fuel f held by country n, which is intended to capture resource abundance for that particular fuel; RGDP is (log) real per capita GDP for country n, which is n a direct measure the level of economic development; (cid:133)nally, REGION is a vector n of dummy variables, each of which takes on a value of one if country n is classi(cid:133)ed as a European, Developed Asian, Latin American, Emerging Asian, or Emerging Other economy, respectively, and takes on a value of zero otherwise. (Accordingly, the estimated coe¢ cients on the regional dummies are interpreted as the regional e⁄ect relative to North America.) The speci(cid:133)c regions are chosen based on existing literature which has shown that these country groupings are relevant for explaining cross-country di⁄erences in oil consumption. The dummies are intended to control for all other unobserved factors within a given region that may help to determine the fuel intensity pro(cid:133)le. Finally, the error term is assumed to be independent and identically distributed, " N(0;(cid:27)2);. The equation is estimated using simple n (cid:24) n ordinary least squares (OLS). 8

Within this regression framework we test the following two hypotheses: HEndowment : (cid:12)f > 0 0 1 and (cid:12)f < 0 for f = 1; 2 HEnergyLadder : 2 f g 0 (cid:12)f > 0 for f = 3; 4; 5; 6 2 f g The (cid:133)rst tests the statistical validity of the endowment hypothesis. If the hypothesis is valid we would expect that the aggregate fuel intensity of fuel f in country n is increasing in the resource endowment of that fuel, thus the coe¢ cient estimate for (cid:12) should be positive and signi(cid:133)cantly di⁄erent from zero. 2 The second tests the validity of the energy ladder hypothesis. Here, we would expect the aggregate fuel intensity of lower quality fuels such as combustibles, renewables, and waste (f = 1) and coal and peat (f = 2) to decrease as a country becomes more developed and makes its way (cid:147)up the energy ladder(cid:148)as it adapts more e¢ cient, cleaner technologies for energy generation. Hence, for these fuels we would expect the coe¢ cient estimate for (cid:12) to be negative and signi(cid:133)cantly di⁄erent from zero. 2 In contrast, for the higher quality fuels such as oil (f = 3); natural gas (f = 4), geothermal, hydoelectric, and nuclear (f = 5), and electricity (f = 6) we expect that aggregate fuel intensity should increase with the level of development. We would expect the coe¢ cient estimate for (cid:12) to be positive and signi(cid:133)cantly di⁄erent from 2 zero for these fuels. 2.2 End-use Portfolio The second dimension of the energy portfolio that we would like to explain is the cross-country variation in end-use consumption. We summarize this aspect of the energy portfolio with the empirical metric, energy usage de(cid:133)ned at two levels of disaggregation. Sectorial energy usage measures of the share of aggregate energy consumption accounted for by sector s aggregated across all fuels and industries for country n. A formal de(cid:133)nition is as follows: c n;f;s;i C n;;s; f i SEU = (cid:1) (cid:1) = XX C c n;;; (cid:1)(cid:1)(cid:1) n;f;s;i f s i XXX where: SEU denotes sectorial energy usage; C denotes aggregate energy conn;;s; sumption accounted for by sector s across all fuels(cid:1), f(cid:1) , and industries, i. Similarly, moving down one level of aggregation, industry-level energy usage measures the share of aggregate energy consumption accounted for by industry i aggregated across all fuels, f, for country n: We formalize this as 9

c n;f;s;i C n;;s;i f IEU = (cid:1) = X C c n;;; (cid:1)(cid:1)(cid:1) n;f;s;i f s i XXX where: IEUn denotes industrial-level energy usage; C denotes energy consumps;i n;;s;i tion accounted for by industry i within sector s, aggrega(cid:1)ted across all fuels, f. 2.2.1 Empirical Model With regard to end-use consumption, our analysis aims to explain, for example, why consumption in the industrial sector comprises a larger fraction of total energy consumed in Argentina (41.1 percent) as opposed to Hong Kong (28.6 percent). At a higher level of disaggregation, road transport (consisting of both passenger and commercial transport activity) comprises 33.1 percent of aggregate energy consumption in Spain, but only 21.7 percent in Canada. What can explain the di⁄erence? In short, as with fuel intensity above, the point of the analysis here is to identify characteristics that can help to explain cross-country di⁄erences in end-use consumption portfolios at both the sectorial and the industry level. We examine three hypotheses. The (cid:133)rst two relate to sector size and the energy e¢ ciency of the sector in question, respectively. All else equal, as the economic size of a given sector increases we might expect energy consumption within that sector to grow as a fraction of total energy consumption. On the other hand, as the energy e¢ ciency of a given sector increases we might expect energy consumption within that sector to decline as a fraction of total energy consumption. Beyond size and e¢ ciency, we also explore the structural transformation hypothesis. There is a well-known, established literature dating to Kuznets (1971) which contends that a countries(cid:146)industrial structure changes endogenously as it undergoes the process of economic development. Initially, for countries at low levels of development, agricultural production constitutes the largest share of economic activity. However, as an economy begins to develop industrialization causes the share of industry in total output to rise as economic activity moves away from agriculture and into heavy industry. Later phases of development tend to be characterized by a decline in manufacturing activity as industrialization eventually gives way to a transformation toward a more service-oriented economy. Transformation of the industrial structure, of course, has implications for energy usage. For countries at low levels of economic development the structural transformation hypothesis suggests that end-use consumption pro(cid:133)les should be weighted toward greater energy usage in the residential and agricultural sectors and relatively low weights on industry. As a country develops and undergoes the process of industrialization, industries(cid:146)share of total energy usage should rise at the expense of agriculture and residential usage. Finally, at high levels of development, after in- 10

dustrialization has occurred and the transformation toward a more service oriented economy underway, the share of residential and commercial usage should rise at the expense of industry. Thus, there are two empirical implications of the structural transformation hypothesis for energy usage that can be tested, both of which exploit the compositional shift of economic activity implied by the process of structural transformation. The (cid:133)rst keys o⁄ the change in industries(cid:146)share of total energy usage, which according to the structural transformation hypothesis should be increasing with income for relatively low levels of economic development(cid:151)re(cid:135)ecting the e⁄ect of industrialization on energy usage(cid:151)and then decreasing for su¢ ciently high levels of development(cid:151) re(cid:135)ectingdeindustrializationastheeconomytransformsintoservice-orientedactivity. Thesecondkeyso⁄thechangeinresidentialandcommercialusage. Accordingtothe structural transformation hypothesis, residential usage should be declining with income at low levels of development and then increasing, along with commercial usage, at su¢ ciently high levels of development. Wetestthethreecandidatehypothesistoexplaincross-countrydi⁄erencesinenduse energy consumption pro(cid:133)les using the following general regression framework EUn = (cid:12)s +(cid:12)sSIZE +(cid:12)sEFFICIENCY (2) s;i 0 1 n;s 2 n;s +(cid:12)sRGDP +(cid:12)sRGDP2 +(cid:12)sREGION +" 3 n 4 n 5 n n where: EUk is the end-use consumption measure de(cid:133)ned at one of the two levels of n aggregation (either SEUn; or IEUn as de(cid:133)ned in the previous section depending on s s;i the level of disaggregation desired) for fuel s in country n; SIZE is the value added n;s (expressed in percentage terms) the sector s in total output for country n; as in the the previous subsection; EFFICIENCY is the total energy consumed in sector s, n;s measured in units of thousands of tones of oil equivalent, expressed per U.S. dollar of real GDP; RGDP is (log) real per capita GDP for country n which, for reasons n discussed below, enters quadratically into the regression framework to capture the non-linear response of the sectorial and industry shares to income at di⁄erent stages of a structural transformation; (cid:133)nally, as above we include the vector of regional dummies, REGION ; to control for other unobserved factors. The error term is n assumed to be iid and normally distributed, " N(0;(cid:27)2): In order to address n (cid:24) n possible endogeneity between our metric for end-use consumption and the proxy for sectorial energy e¢ ciency, the equation is estimated using two stage least squares (2SLS) using aggregate energy e¢ ciency as an instrument for energy e¢ ciency at the sectorial level. Within this regression framework, we examine whether or not sector size is an important determinate of the end-use energy consumption pro(cid:133)le by testing the following hypothesis. HSize : (cid:12)s > 0 0 1 We expect that the share of total energy consumption in sector s is increasing in the 11

economicsizeofthesectorasmeasuredbyvalueaddedinGDP,sothatthecoe¢ cient estimate for (cid:12)s should be positive and signi(cid:133)cantly di⁄erent from zero. 1 Next, we test the validity of the hypothesis that increased energy e¢ ciency in sector s leads to a decrease in that sectors share of aggregate energy consumption. HEfficiency : (cid:12)s < 0 0 2 If sectorial-level e¢ ciency is an important determinate for the end-use energy consumption pro(cid:133)le we would expect the coe¢ cient estimate for (cid:12)s to be negative and 2 signi(cid:133)cantly di⁄erent from zero. Finally, we test the validity of the structural transformation hypothesis as follows. (cid:12)s > 0; (cid:12)s < 0 for s = 1 HTransform : 3 4 0 (cid:12)s < 0; (cid:12)s > 0 for s = 3 3 4 Asdiscussedabove,thehypothesispredictsthatindustries(cid:146)shareoftotalenergyusage will have an inverse U-shaped relationship with the level of income, which should be captured by the quadratic income term with (cid:12)s=1 > 0 and (cid:12)s=1 < 0: In contrast, 3 4 commercial and residential usage should have a U-shaped relationship with the level of income, falling for low levels of development and then growing at a su¢ ciently high level of development, which should be captured by the quadratic income term with (cid:12)s=3 < 0 and (cid:12)s=3 > 0. For the (cid:133)nal two sectors, we expect transportation(cid:146)s share 3 4 to increase with income, so that (cid:12)s=2 > 0, and agricultural(cid:146)s share to decrease, so 3 that (cid:12)s=4 < 0, but do not necessarily have reason to think that either should enter 3 into the regression in a non-linear way. 3 Main Results The main results are presented below in the following two subsections. The (cid:133)rst examines cross-country di⁄erences in fuel intensity pro(cid:133)les while the second examines di⁄erences in end-use consumption. 3.1 Fuel Intensity Pro(cid:133)le Table 2 presents summary statistics for the share of total energy usage broken out by source. The table shows that the dominate energy source comes from crude oil and petroleum products, which alone accounts for about half of all energy consumed globally. Electricity accounts for about 20 percent of global energy consumption, followed by natural gas at roughly 15 percent. The remaining share is comprised of combustibles, renewables, and waste as well as coal and peat, which account under 15 percent of global energy consumption. The remaining fraction comes from geothermal,hydroelectric,ornuclearpower,whichtakentogetheraccountforatrivial fraction of global usage. 12

Comparing the developed economies to the emerging markets economies hints at some key di⁄erences when energy usage pro(cid:133)les are broken out by primary source. The data show that, relative to emerging market economies, developed economies tend to rely more heavily on petroleum products, natural gas, and electricity as primary sources of energy. In contrast, developing economies tend to rely more heavily on coal and peat as well as combustibles, renewables, and waste. Thus, even a cursory glance at the data suggests that there may be some systematic di⁄erence in the energy usage portfolio between the two sets of countries. A more formal assessment can be found in Table 3, which presents the regression results for Equation (1). The table shows a set of results for each fuel, with one set corresponding to the regression without the regional dummies ((cid:133)rst column of numbers) and the second set corresponding to the regression with the dummies (second column). Concentrating on the (cid:133)rst column of numbers for each fuel, we see that there is strong support for the energy ladder hypothesis across nearly all the fuels. For (cid:133)ve of the six, the estimated coe¢ cient has the predicted sign and is signi(cid:133)cantly di⁄erent from zero at the 95 percent con(cid:133)dence level.8 As real per capita GDP rises, countries shift their aggregate fuel intensity portfolios away from lower quality, more polluting fuels such as combustibles, renewables, and waste as well as coal and peat and into higher quality, cleaner fuels such as re(cid:133)ned petroleum products, natural gas, andelectricity. Moreover, inlookingat the quantitativemagnitudeof thecoe¢ cients and the precision with which they are estimated, the evidence in favor of the energy ladder hypothesis is clearly strongest at the two extremes of the ladder. There is a large, highly signi(cid:133)cant negative correlation between income and the lower end of the quality ladder(cid:151)the usage of combustibles, renewables, and waste(cid:151)while the opposite is true at the higher end of the ladder as re(cid:135)ected in electricity usage. The coe¢ cients for the intermediate fuels tend to be smaller in magnitude and, although many are statistically signi(cid:133)cant, taken as a whole they tend to be more imprecisely estimated. In contrast to the energy ladder hypothesis, there is only mild support for the endowment hypothesis. Although the estimated coe¢ cients have the correct sign for all three fuels for which we have an empirical proxy for endowment available, only in the case of natural gas do we (cid:133)nd that proved reserves are signi(cid:133)cantly correlated with the share of natural gas in total energy usage. Natural gas, more so that coal or oil and petroleum products, may be particularly susceptible to the endowment hypothesis given the relatively large (even for the energy industry) capital expenses associated with international trade in natural gas either via pipeline or in lique(cid:133)ed form. Moving to the regressions with the regional dummies, we (cid:133)nd that support for the 8The lone exception is for geothermal, hydoelectric, and nuclear, but that likely re(cid:135)ects the fact thatthiscategoryonlyaccountsforaminisculefractionofenergyinmostcountriesandiscompletely absent in many others. 13

energyladderhypothesisislargelyrobusttocontrollingforunobservedregion-speci(cid:133)c characteristics For three of the six fuels (combustibles, renewables, and waste, oil and petroleum products, and electricity) the estimated coe¢ cient on the real GDP per capita remains of the correct sign and continues to be statistically signi(cid:133)cant at high con(cid:133)dence levels. Evidence in favor of the endowment hypothesis is marginally strongerduetotheintroductionoftheregionaldummiesowinglargelytonaturalgas. With regard to the country dummies themselves, two things stand out. First, the Asian economies(cid:151)both developed and developing(cid:151)tend to rely heavily on coal for energy generation relative to other countries in the sample. Importantly, this is true even when controlling for resource endowment. Second, the emerging other category, which includes Israel, Russia, and Saudi Arabia, stands out in its low reliance on renewables, combustibles, and waste relative to other countries. and its high reliance on geothermal, hydroelectric, and nuclear power and strong electricity usage. Extending the analysis to disaggregated data at the sectorial and industry level reveal that much of the support for the energy ladder hypothesis stems from the industrial as well as the residential and commercial sectors. Fuel intensity in the transportation and agricultural sectors does not, in general, (cid:133)t well into our hypothesized determinates. For the sake of brevity, I do not present the full set of disaggregated regression results and instead simply highlight some of the interesting insights.9 Support for the energy ladder hypothesis comes primarily from the industrial as well astheresidential andcommercial sectorand, muchliketheaggregatedata, tends to be strongest at the two extreme ends of the energy ladder. Speci(cid:133)cally, usage of combustibles, renewables, andwastefallssigni(cid:133)cantlywithincomeinbothresidential as well as commercial usage and also in non-energy intensive industries. At the other extreme of the energy ladder, electricity usage rises signi(cid:133)cantly in both residential and commercial usage as well as in both energy-intense and non-intense industrial usage. The evidence is somewhat more mixed for the intermediate fuels in these sectors. Coal usage falls with income amongst energy-intensive industries. Natural gas usage rises with income in non-energy intensive industrial usage as well as in both residential and commercial usage, although the results for natural gas are not robust to the inclusion of regional dummies. Industrial usage of oil and petroleum products is interesting because it falls with income for energy-intense industries, but rises with income for energy non-intense industries, suggesting that there is fuel switching within industries usage itself. Finally, there is very little, if any, evidence for the energy ladder hypothesis in the transport sector while oil and petroleum product usage declines with income in the agricultural sector. With regard to the endowment hypothesis, the disaggregated data reveal that support comes primarily from coal usage in commercial and agricultural activity, oil and petroleum product usage in non-energy intensive industries, as well as from 9Thefullsetofresultswouldrequireasetoftablesdescribingresultsfrom60di⁄erentregressions, whichisistoocumbersometoincludeinthepaper. However,theresultsareavailableuponrequest. 14

residential and non-road transport natural gas usage. 3.2 End-use Pro(cid:133)le Table 4 presents summary statistics for the share of total energy usage by sector. For the sample as a whole, industrial usage accounts for the largest share of global energy consumptionat37percent,whiletransportationandresidentialandcommercialusage each account for roughly 30 percent. Agricultural energy usage accounts for the remaining 2.5 percent. This carries over to the industry-level with each sector as well. Thus,incontrasttoaggregatefuelintensity, acursoryglanceatthedatareveals very little di⁄erence in the sectorial distribution of energy usage between developed and emerging market countries. Regression results for Equation 2 are presented in Table 5. Again, we present two setsofresultsforeachend-usesector,onewithregionaldummies((cid:133)rstcolumnforeach sector) and one without regional dummies (second column). Generally speaking, the resultsarerobustacrossbothspeci(cid:133)cations. Thereislittleevidencethateithersector sizeorsector-speci(cid:133)ce¢ ciencyisanimportantdeterminateofenergyusage. Thereis, however, support for the structural transformation hypothesis. For industrial usage the coe¢ cient on (logged) real GDP is positive and signi(cid:133)cant while the coe¢ cient on the log real GDP squared is negative and signi(cid:133)cant. This indicates that the share of industrial energy usage starts out at a low level for relatively undeveloped economies. As these economies grow, the industrial share of energy usage increases re(cid:135)ecting the process of industrialization which is a key component of economic development. However, once a country reaches a certain level of development, deindustrialization occurs as the economy transforms into more service-oriented activity; hence, industry shareoftotalenergyconsumptionbeginstofallonceaneconomyhasreachedacertain level of development. In our estimates, this peak occurs at a real per capita level of roughly $10,500, about the level of development of Brazil. This inverse U-shape for the share of industrial energy usage is very much in line with the structural transformation hypothesis. For residential and commercial energy usage, we see the opposite pattern. The coe¢ cienton(logged)realGDPisnegativeandsigni(cid:133)cantwhilethecoe¢ cientonthe logrealGDPsquaredispositiveandsigni(cid:133)cant.Thisisalsoinlinewiththestructural transformation hypothesis in the sense that at low levels of economic development residentialusagecarriesalargefractionoftotalusage,butthisdeclinesasaneconomy grows and industrialization occurs. Eventually, the economy hits a point at which the emergence of the service sector causes the share of commercial usage to increase. In addition, the share of residential usage increases as the demand for energy-intense consumer durables begins to pick up at su¢ ciently high income levels. The net e⁄ect gives rise to a U-shaped pattern for the share of residential and commercial usage taken as a whole. According to the regression results reported in [Table 5], the turning point at which residential and commercial usage stops declining and begins 15

to rise is roughly $14,000, about the level of Mexico or Argentina. Incontrastwithindustrialusageandresidentialandcommercialusage,neitherthe transportationnortheagriculturalsector(cid:133)tneatlyintothestructuraltransformation hypothesis. For both the estimated coe¢ cients on the level of income is positive while the squared term is negative, but both are insigni(cid:133)cant. While the size of the agricultural sector helps to explain cross-country variation in the agriculture share of totalenergyusage, wedidnothavemuchsuccessinexplainingcross-countryvariation in energy usage with the transportation sector. Table 6 shows regression results for the industry-level data. Support for the structural transformation hypothesis is not robust at the disaggregate level. For the residential and commercial sector we see that the nonlinear relationship at the sectorial level is driven by residential usage. In contrast, commercial usage, like agricultural usage, appears to be driven by sector size. Finally, we have a bit more success in explaining transportation usage at the disaggregated level. In particular, for road transport we see the share is rising in income presumably re(cid:135)ecting increased automobile purchases at higher income levels. Non-road transport is signi(cid:133)cantly correlated with e¢ ciency, indicating that the share tends to be higher in countries where transport usage is relatively ine¢ cient. 4 Are the BRICs Di⁄erent? Much of the impetus for the shift toward emerging market economies and away from developed economies as the primary driver of global energy consumption growth has come from the so-called BRIC economies of Brazil, Russia, India, and China. Not surprisingly, these economies have garnered a lot of attention from energy market participants in particular, as well as (cid:133)nancial market participants more generally as well as policy-makers interested in understanding developments in commodity markets. GiventhattheBRICeconomiesareplayingalargerandlargerroleinglobal energy consumption, it seems natural to ask whether there is something inherently di⁄erent about the consumption patterns in these countries in particular. Methodologically, weanswerthisquestionbysimplyintroducingdummyvariables into the regression equations 1 and 2 both for the BRICs as a whole (i.e., a single indicator variable that takes on the value of one if the country is a BRIC member and is zero otherwise) and then for each of the BRICs individually. If energy usage in the BRICs is di⁄erent in some way not already addressed by the hypotheses laid out in the previous section, then the dummies will capture this di⁄erence. The aggregate BRIC dummy is intended to capture systematic di⁄erences in the BRIC economies as a whole, while the individual dummies are intended to capture country-speci(cid:133)c di⁄erences. We are interestedinansweringtwoquestions. First, howdoes the inclusionof the BRIC dummies in(cid:135)uence our conclusions regarding our hypothesized determinates of the energy consumption portfolio? Second, given that we control for these hypoth- 16

esized determinates, do the BRICs themselves, either taken together as a group or individually, have systematically di⁄erent consumption portfolios from other countries? Results are reported below in two subsections. 4.1 Fuel Intensity Referring back to Table 1, the fuel intensity pro(cid:133)les of the BRIC economies stand out in two respects. First, they tend to rely more heavily on combustibles, renewables, and waste as well as coal for energy generation relative to other economies. Taken together these two fuel sources constitute nearly 35 percent of total energy consumption, whereas comparable number for the developed economies and non- BRIC emerging market countries are 9 and 15 percent, respectively. Second, they tend to rely less heavily on oil and petroleum products, which constitute 31 percent of the fuel intensity pro(cid:133)le in the BRIC economies as opposed to 48 and 52 percent, respectively, in the developed and non-BRIC emerging markets. Thus, a preliminary look at the data suggests that the BRICs may indeed be di⁄erent with respect to the fuel intensity pro(cid:133)le. Table 7 presents regression results from Equation 1 estimated with the separate BRIC dummies, which are directly comparable to what was reported above in Table 2. The table reveals that the high share of combustibles, renewables, and waste in the BRIC economies is largely driven by Brazil and Russia. The disaggregated data showthatforBrazilthehighshareofcombustibles,renewables,andwastecomesfrom non-energy intensive industries as well as both road an non-road transportation. For Russia, thehighsharestemsprimarilyfromcommercialusage. Thestrongcoalusage is driven by China, which uses coal more intensely than the other countries in all four sectors. Importantly, this is true even after controlling for China(cid:146)s relatively large endowment of coal. On the other hand, the relatively low share of oil and petroleum products in the fuel intensity pro(cid:133)le of the BRIC economies appears to be largely due to India and China. In summary, even after controlling for some hypothesized determinatesofthefuelintensitypro(cid:133)le,theBRICeconomiesstillseemtobedi⁄erent formothercountriesinthesensethattheyhaveanover-relianceonlowerqualityfuels and an under-reliance on oil an petroleum products relative to other countries. With regard to the main conclusions regarding the determinates of the fuel intensity pro(cid:133)le the inclusion of the BRIC dummies appear to have little impact. Even after allowing for a country-speci(cid:133)c e⁄ect for each of the BRIC economies, we continues to see strong support of the energy ladder hypothesis, principally at the two extremes of the energy ladder. For the intermediate fuels, the evidence remains mixed. For oil and petroleum products, support for the energy ladder hypothesis is not robust to the inclusion of the BRIC dummies due to the low usage in India and China. Instead, controlling for each of these two countries separately strengthens empirical support of the endowment hypothesis. Disaggregating data to the sectorial and industry level o⁄ers little in the way of 17

new insights. The results are essentially unchanged relative to those discussed in the previous section. 4.2 End-use Consumption Table4showsthatalthoughtheredonotappeartobeanynotabledi⁄erencesbetween developed and developing countries with respect to end-use consumption, there do appear to be big di⁄erences in the BRIC economies. In particular, the BRICs stand outasdi⁄erentinnearlyeverysectorandalsoinindustrieswithinagiven. Theytend tohavealargershareofindustrialenergyusage(cid:151)nearly10percenthigherthaneither developed economies or the non-BRIC emerging market economies(cid:151)and this extends down to both energy-intensive and non-energy-intensive industries. The BRICs also have a higher percentage of energy use in agricultural activity(cid:151)nearly double that of either developed or non-BRIC emerging market economies. In contrast, transportationdoesnotplayaslargearoleintheBRICsasitdoesinothereconomies. When we look at the disaggregated industry data we can see that this is primarily due to low energy usage in road transport. Finally, residential energy usage carries a larger share in BRIC energy consumption relative to the rest of the world, while commercial energy usage plays a smaller share. Regression results from Equation 2 estimated with separate BRIC dummies are presented in Table 8 and are directly comparable to results presented in Table 5. At thesectorial level, itturns outthat oncewecontrol forthehypothesizeddeterminates oftheend-useconsumptionpro(cid:133)le, industrialenergyconsumptionintheBRICsisnot signi(cid:133)cantly di⁄erent from other countries. Thus, contrary to the impression created by the unconditional data in Table 4, it appears that there is nothing di⁄erent about industrial energy usage in the BRICs per se; instead, they simply tend to have higher shares of industrial usage primarily because these economies are undergoing a period of rapid industrialization. This sectorial-level result does not necessary apply when the data are disaggregated down to the industry level. Table 9 shows that China, in particular, is importantly di⁄erent in that it has a very high share of energy-intense industrial energy usage. The results in Table 8 also showthat the BRICs really don(cid:146)t stand out in terms of agricultural usage. But, at the sectorial-level what appears to set energy consumption apart in the BRICs is transportation, where energy usage is considerably lower relative to other countries, as well as residential and commercial usage, where the opposite is true primarily in China and India. A look at the disaggregated data in Table 9 reveals that the low transportation usage is due to road transport in India and China as well as with non-road transportation industries in Russia. With regard to the main conclusions regarding the determinates of the end-use consumption pro(cid:133)le, the inclusion of the BRIC dummies appear to have little impact. We continue to (cid:133)nd broad support for the structural transformation hypothesis at both the sectorial as well as the industry-level. 18

5 Conclusion This paper used a dataset detailing energy usage in a broad cross-section of countries to explain country-to-country di⁄erences in energy consumption portfolios along two separate dimensions: the fuel intensity pro(cid:133)le and the end-use consumption pro(cid:133)le. Speci(cid:133)cally, we tested two hypotheses regarding determinates of the di⁄erences in consumption portfolios across countries. The energy ladder hypothesis implies that as the level of economic development increases energy consumption will transit from lower quality, cheaper fuels such as biomass (wood and animal and plant waste) to higher quality fuels such as natural gas and petroleum products. The structural transformationhypothesisimpliesthatasthelevelofeconomicdevelopmentincreases the bulk of end-use energy demand will shift away from agricultural usage toward industrial usage as an economy undergoes a structural transformation. Once the transformationhas occurredhigherlevels of economicdevelopment will pushthe bulk of end-use energy demand out of industrial usage and into residential and commercial usage as the economy becomes more service-oriented. We found statistical evidence to support both of these hypotheses. In addition, the paper also showed that even when these determinants of the energy consumption portfolio are taken into account, the energy consumption portfolios of the BRIC economies are still notably di⁄erent from those of other countries. The BRICs tend to rely more heavily on lower quality fuel sources(cid:151)combustibles, renewables, and waste, as well as coal and peat(cid:151)and, in terms of end-use consumption, tend to underconsume energy in the transportation sector relative to other countries. In addition, we found that China consumes a large fraction of total energy in energy-intense industry(cid:151)even more than what can be explained by the structural transformation hypothesis. The policy implications of this paper are relatively straight-forward. From the perspectiveof energyanalysts andpolicy-makers, theempirical results presentedhere suggest that understanding global energy market developments probably requires a more intense focus on developments at the country- and industry-speci(cid:133)c level. In this sense, this paper is very much in line with the broad conclusions of Stefanski (2009) and Arbex and Perobelli (2010), which emphasize that microeconomic foundations are important for understanding global energy developments. Future empirical work should concentrate on examining how far the systematic di⁄erences in energy consumption portfolios can go in explaining di⁄erences in the dynamics of energy consumption over the business cycle. Arseneau (2010) is a paper that moves in this direction. Such an explanation seems promising in explaining why countryspeci(cid:133)c heterogeneity is typically so important to control for when estimating price and income elasticity parameters for energy demand. 19

References [1] Arbex, M. and F. S. Perobelli, (2010) (cid:147)Solow meets Leontief: Economic Growth and Energy Consumption(cid:148), Energy Economics, 32, 43-53. [2] Arseneau, D.M., (2010) (cid:147)An International Perspective on Oil Consumption over the Business Cycle(cid:148), Federal Reserve Board mimeo. [3] Barnes, D. and Floor, W., (1996), (cid:147)Rural Energy and the Poor in Developing Countries: A Challenge for Economic Development(cid:148), Annual Review of Energy and the Environment, 21, 497-530. [4] Bashmakov, I, (2007), (cid:147)Three Laws of Energy Transmission(cid:148), Energy Policy,35(7), 3583-3594. [5] Burke, P., (2010a), (cid:147)Income, Resources, and Electricity Mix(cid:148), Energy Economics, 32(3), 616-177. [6] Burke, P., (2010b), (cid:147)Energy Ladders of Supply and Demand(cid:148), Australian National University mimeo. [7] Dargay, J. M., D. Gately, and H. G. Huntington (2007), (cid:147)Price and Income Responsiveness of World Oil Demand, by Product(cid:148), Energy Modeling Forum Working Paper EMF OP 61, Stanford University [8] Gately, D. and H. Huntington, (2002), (cid:147)The Asymmetric E⁄ects of Changes in Prices and Income on Energy and Oil Demand(cid:148), The Energy Journal, 23(1), 19-55. [9] Gr(cid:252)bler, A., (2004), (cid:147)Transitions in Energy Use(cid:148), Encyclopedia of Energy, 6, 163-177. [10] Heltberg, R., (2004), (cid:147)Fuel Switching: Evidence from Eight Developing Countries(cid:148), Energy Economics, 26(5), 869-887. [11] Hosier, R., (2004), (cid:147)Energy Ladder in Developing Countries(cid:148), Encyclopedia of Energy, 2, 423-435. [12] Hosier, R. and J. Dowd, (1987), (cid:147)Household Fuel Choice in Zimbabwe: An Empirical Test of the Energy Ladder Hypothesis(cid:148), Resources and Energy, 9(4), 347-361. [13] International Energy Agency, Energy Balances of Non-OECD Countries, (2009), Paris, France. [14] International Energy Agency, Energy Balances of OECD Countries, (2009), Paris, France. 20

[15] Judson, R.A., R. Schmalensee, and T.M. Stoker, (1999), (cid:147)Economic Development and the Structure of Demand for Commercial Energy(cid:148), The Energy Journal, 20(2), 29-57. [16] Kuznets, S. (1971), Economic Growth of Nations, Cambridge: Harvard University Press, 1971. [17] Leach, G. (1992), (cid:147)The Energy Transition(cid:148), Energy Policy, 20(2), 116-123. [18] Marcotullio, P., and N. Schulz, (2007), (cid:147)Comparison of Energy Transitions in the United States and Developing and Industrializing Economies(cid:148), World Development, 35(10), 1650-1683. [19] Medlock, K. G., and R. Soligo, (2001), (cid:147)Economic Development and End-use Energy Demand(cid:148), The Energy Journal, 22(2), 77- 105. [20] Sch(cid:228)fer, A. (2005), (cid:147)Structural Change in Energy Use(cid:148), Energy Policy, 33(4), 429-437. [21] Stefanski,Radek,(2010),(cid:147)StructuralTransformationandtheOilPrice(cid:148),Mimeo, University of Oxford. 21

Table 1. Countries in sample, by region Developed Economies Emerging Market Economies Europe North America Developed Asia Latin America Emerging Asia Emerging Other BRICs Austria Canada Australia Argentina Hong Kong Israel Brazil Belgium Mexico Japan Chile Indonesia Saudi Arabia China Finland US South Korea Colombia Malaysia India France Venezuela Philippines Russia Germany Singapore Ireland Thailand Italy Netherlands Portugal Sweden Spain Switzerland UK

Table 2. Fuel intensity profile Coal and Peat Crude Oil and Petroleum Products (f = 1) (f = 2) Mean Median St. Dev. Min. Max. N Mean Median St. Dev. Min. Max. N World 0.047 0.026 0.062 0 0.330 34 0.479 0.478 0.118 0.232 0.763 34 Developed Economies 0.029 0.022 0.020 0.008 0.090 19 0.488 0.478 0.085 0.317 0.639 19 BRIC Economies 0.131 0.080 0.138 0.035 0.330 4 0.308 0.276 0.097 0.232 0.446 4 Non-BRIC Emerging Markets 0.048 0.029 0.052 0 0.144 11 0.523 0.517 0.127 0.332 0.763 11 Natural Gas Nuclear, Geothermal, and Hydro. (f = 3) (f = 4) Mean Median St. Dev. Min. Max. N Mean Median St. Dev. Min. Max. N World 0.144 0.128 0.107 0 0.363 30 0.003 0 0.010 0 0.058 30 Developed Economies 0.177 0.169 0.093 0.016 0.340 19 0.002 0.001 0.002 0 0.009 19 BRIC Economies 0.111 0.052 0.130 0.036 0.305 4 0.001 0.001 0.002 0 0.003 4 Non-BRIC Emerging Markets 0.102 0.084 0.112 0 0.363 11 0.005 0 0.017 0 0.058 11 Combust., Renew., and Waste Electricity and Heat (f = 5) (f = 6) Mean Median St. Dev. Min. Max. N Mean Median St. Dev. Min. Max. N World 0.091 0.045 0.105 0 0.409 30 0.228 0.217 0.089 0.072 0.447 30 Developed Economies 0.058 0.045 0.051 0.007 0.169 19 0.246 0.227 0.076 0.147 0.447 19 BRIC Economies 0.213 0.218 0.172 0.006 0.409 4 0.236 0.202 0.127 0.124 0.416 4 Non-BRIC Emerging Markets 0.102 0.037 0.119 0 0.351 11 0.197 0.171 0.094 0.072 0.439 11

Table 3. Cross-country differences in aggregate fuel intensity profiles (AFI) Combustables, Oil and Geothermal, Hydroelectrical, Electricity and Renewables and Waste Coal and Peat Petroleum Products Natural Gas and Nuclear Power Heat Generation (f = 1) (f = 2) (f = 3) (f = 4) (f = 5) (f = 6) Constant 1.05 1.22 0.42 0.27 0.004 -0.18 -0.26 -0.02 -0.01 -0.02 -0.3 -0.43 (7.12) (5.78) (3.85) (1.86) (0.02) (-0.48) (-1.23) (-0.07) (-0.44) (-0.52) (-1.81) (-1.70) Europe . 0.04 . 0.04 . -0.09 . -0.004 . 0.001 . 0.05 . (0.92) . (1.29) . (-1.05) . (-0.06) . (0.11) . (1.06) Developed Asia . -0.01 . 0.07 . -0.02 . -0.06 . 0.001 . 0.06 . (-0.24) . (1.85) . (-0.21) . (-0.75) . (0.06) . (0.85) Latin America . -0.02 . 0.02 . 0.01 . 0.01 . 0.001 . 0.03 . (-0.33) . (0.39) . (0.11) . (0.84) . (0.11) . (0.47) Emerggingg Asia . -0.01 . 0.09 . -0.01 . -0.1 . 0.002 . 0.08 . (-0.23) . (2.38) . (-0.14) . (-1.34) . (0.21) . (1.32) Emerging Other . -0.09 . -0.004 . 0.01 . -0.14 . 0.02 . 0.12 . (-1.66) . (-0.09) . (0.11) . (-1.56) . (2.55) . (1.85) Economic Development -0.22 -0.26 -0.09 -0.06 0.11 0.16 0.09 0.05 0.003 0.004 0.12 0.14 (6.53) (-5.64) (-3.48) (2.01) (1.94) (1.92) (1.92) (0.68) (0.57) (0.56) (3.20) (2.50) Resource Endowment . . 0.23 0.32 0.6 -0.17 0.83 1.25 . . . . . . (1.63) (2.22) (0.62) (0.13) (2.03) (2.49) . . . . R2 0.56 0.67 0.33 0.57 0.11 0.18 0.18 0.37 0.01 0.29 0.24 0.35 ˆˆˆˆˆˆ222222 0.005 0.004 0.003 0.002 0.013 0.014 0.010 0.009 0.000 0.000 0.006 0.006 Nobs 35 35 35 35 35 35 35 35 35 35 35 35

Table 4. End-use consumption profile Industry Transportation (S = 1) (S = 2) Mean Median St. Dev. Min. Max. N Mean Median St. Dev. Min. Max. N World 0.371 0.390 0.097 0.198 0.545 35 0.290 0.294 0.093 0.104 0.522 35 Developed Economies 0.364 0.355 0.087 0.216 0.513 19 0.294 0.294 0.075 0.173 0.443 19 BRIC Economies 0.454 0.441 0.071 0.390 0.545 16 0.184 0.163 0.096 0.104 0.307 16 Emerging Market Economies 0.356 0.375 0.112 0.198 0.508 4 0.318 0.317 0.099 0.168 0.522 4 Residential and Commercial Agriculture (S = 3) (S = 4) Mean Median St. Dev. Min. Max. N Mean Median St. Dev. Min. Max. N World 0.314 0.306 0.097 0.143 0.534 35 0.025 0.024 0.021 0.000 0.099 35 Developed Economies 0.317 0.316 0.073 0.194 0.475 19 0.025 0.024 0.020 0.006 0.099 19 BRIC Economies 0.320 0.328 0.122 0.166 0.461 16 0.041 0.042 0.010 0.028 0.052 16 Emerging Market Economies 0.306 0.267 0.128 0.143 0.534 4 0.020 0.007 0.024 0.000 0.059 4

Table 5. Cross-country differences in end-use energy consumption, by sector (SEU) Sector Industrial Transportation Residential and Commercial Agriculture (S = 1) (S = 2) (S = 3) (S = 4) Constant -3.83 -6.61 -2.52 -0.55 7.70 8.61 -0.65 -0.43 (-1.71) (-2.64) (-1.13) (-0.19) (4.00) (3.92) (-1.48) (-0.79) Europe . 0.04 . -0.11 . 0.06 . 0.004 . (0.79) . (-1.77) . (1.34) . (0.37) DDeevveellooppeedd AAssiiaa . 00.1111 . --00.1111 . 00.000088 . --00.0011 . (1.65) . (-1.47) . (0.14) . (-0.64) Latin America . 0.004 . -0.02 . 0.03 . 0.001 . (0.06) . (-0.27) . (0.49) . (0.07) Emerging Asia . 0.05 . -0.07 . 0.04 . -0.02 . (0.77) . (-0.98) . (0.69) . (-1.24) Emerging Other . -0.14 . -0.03 . 0.19 . -0.09 . (-1.96) . (-0.42) . (2.98) . (-0.63) Economic Development 2.11 3.45 1.32 0.36 -3.62 -4.07 0.29 0.19 (1.91) (2.81) (1.21) (0.25) (-3.87) (-3.79) (1.43) (0.74) Economic Development (Squared) -0.26 -0.43 -0.15 -0.03 0.44 0.49 -0.03 -0.02 (-1.95) (-2.86) (-1.15) (-0.18) (3.89) (3.79) (1.34) (-0.67) Sector Size 0.06 0.06 -0.48 -0.84 0.17 0.32 0.31 0.37 (0.32) (0.34) (-0.52) (-0.85) (0.51) (0.98) (2.19) (2.41) Efficiency 0.64 1.28 0.14 0.02 -1.08 -1.68 0.32 0.37 (0.61) (1.31) (0.13) (0.02) (-1.24) (-2.06) (0.15) (0.15) R2 0.24 0.52 0.11 0.25 0.41 0.61 0.21 0.36 ˆˆˆˆˆˆˆ2222222 00.000088 00.000066 00.000099 00.000099 00.000066 00.000055 00.000011 00.000011 Nobs 35 35 35 35 35 35 35 35

Table 6. Cross-country differences in end-use energy consumption, by industry within sector (IEU) Industrial Sector Transportation Sector Residential and Commercial Sector Energy Intensive Non-energy Intensive Industries Industries Road Industries Non-road Industries Residential Commercial (S = 1; I = 1) (S = 1; I = 2) (S = 2; I = 1) (S = 2; I = 2) (S = 3; I = 1) (S = 3; I = 2) Constant -2.31 -1.84 -1.10 -0.64 -4.50 -5.15 -0.64 0.50 7.42 8.32 1.39 0.26 (-0.93) (-0.62) (-0.95) (-0.44) (-1.53) (-1.36) (-0.89) (0.55) (3.91) (3.64) (0.87) (0.16) Europe . 0.06 . 0.02 . -0.11 . -0.03 . 0.06 . 0.00 . (1.04) . (0.70) . (-1.46) . (-1.71) . (1.29) . (-0.07) Developed Asia . 0.09 . 0.04 . -0.13 . -0.02 . -0.02 . 0.05 . (1.12) . (1.12) . (-1.29) . (-0.75) . (-0.27) . (1.24) Latin America . 0.00 . 0.00 . -0.01 . -0.01 . 0.00 . 0.07 . (-0.02) . (-0.02) . (-0.11) . (-0.57) . (-0.07) . (1.56) Emerging Asia . -0.06 . -0.03 . 0.05 . -0.05 . -0.02 . 0.15 . (-0.85) . (-0.91) . (0.53) . (-2.25) . (-0.41) . (3.65) Emerging Other . -0.09 . -0.03 . -0.01 . -0.01 . 0.11 . 0.08 . (-1.16) . (-0.84) . (-0.14) . (-0.51) . (1.74) . (1.71) Economic Development 1.20 1.06 0.62 0.43 2.34 2.57 0.30 -0.23 -3.33 -3.71 -0.78 -0.38 (0.98) (0.73) (1.09) (0.62) (1.64) (1.39) (0.85) (-0.52) (-3.61) (-3.33) (-1.01) (-0.49) Economic Development (Squared) -0.15 -0.14 -0.08 -0.06 -0.28 -0.29 -0.04 0.03 0.38 0.42 0.11 0.07 ((--00.9977)) ((--00.7799)) ((--11.1133)) ((--00.7722)) ((--11.6600)) ((--11.2299)) ((--00.8833)) ((00.5500)) ((33.4466)) ((33.1144)) ((11.1166)) ((00.7799)) Sector Size -0.03 0.04 -0.08 -0.06 -0.95 -0.02 0.29 0.32 -0.19 -0.02 0.78 0.62 (-0.16) (0.19) (-0.89) (-0.66) (-0.80) (-1.49) (0.97) (1.06) (-0.59) (-0.06) (2.81) (2.62) Efficiency 1.12 1.23 -0.05 -0.01 -0.66 -0.76 0.81 0.75 0.02 -0.53 -1.17 -0.94 (0.96) (1.06) (-0.09) (-0.02) (-0.47) (-0.53) (2.35) (2.14) (0.03) (-0.63) (-1.63) (-1.58) R2 0.07 0.33 0.08 0.29 0.13 0.3 0.19 0.37 0.52 0.57 0.52 0.57 ˆˆ22 0.010 0.008 0.002 0.002 0.015 0.014 0.001 0.001 0.006 0.005 0.004 0.003 Nobs 35 35 35 35 35 35 35 35 35 35 35 35

Table 7. Cross-country differences in fuel intensity profiles and the BRIC economies (AFI) Combustables, Oil and Geothermal, Hydroelectrical, Electricity and Renewables and Waste Coal and Peat Petroleum Products Natural Gas and Nuclear Power Heat Generation (f = 1) (f = 2) (f = 3) (f = 4) (f = 5) (f = 6) Constant 0.90 1.09 0.26 0.19 0.32 0.19 -0.17 -0.004 -0.01 -0.01 -0.43 -0.55 (5.62) (5.10) (2.91) (1.85) (1.26) (0.54) (-0.64) (-0.01) (-0.38) (-0.38) (-2.30) (-2.02) Europe . 0.04 . 0.01 . -0.06 . 0.04 . 0.001 . 0.05 . (0.96) . (0.51) . (-0.91) . (0.63) . (0.13) . (1.06) Developed Asia . -0.01 . 0.05 . -0.002 . -0.02 . 0.001 . 0.05 . (-0.29) . (2.05) . (-0.03) . (-0.29) . (0.07) . (0.87) Latin America . -0.04 . -0.003 . -0.02 . 0.06 . 0 . 0.03 . (-0.75) . (-0.13) . (0.23) . (0.81) . (0.04) . (0.54) Emerging Asia . -0.01 . 0.05 . 0.003 . -0.07 . 0 . 0.08 . (-0.10) . (2.02) . (0.04) . (-0.89) . (0.04) . (1.42) Emerging Other . -0.07 . -0.02 . 0.12 . -0.13 . 0.028 . 0.05 . ((-1.33)) . ((-0.67)) . ((1.23)) . ((-1.52)) . ((3.64)) . ((0.74)) Brazil 0.13 0.16 -0.02 0.004 -0.05 -0.03 -0.06 -0.13 0 0 0.01 0.03 (2.05) (2.35) (-0.60) (0.13) (-0.44) (0.27) (-0.58) (-1.26) (-0.09) (0.10) (0.13) (0.39) Russia 0.16 0.12 0.03 0.01 -0.16 -0.15 -0.05 -0.01 0 0 0.04 0.02 (2.17) (1.74) (0.83) (0.38) (-1.37) (-1.26) (-0.39) (-0.12) (0.03) (0.21) (0.44) (0.23) India -0.11 -0.04 -0.01 0.03 -0.35 -0.43 -0.46 -0.69 0 -0.03 0.22 0.22 ((--11.6666)) ((--00.5522)) ((--00.2211)) ((00.6677)) ((--33.1100)) ((--33.4411)) ((--11.1111)) ((--11.5599)) ((--00.2211)) ((--22.7700)) ((22.9944)) ((22.3355)) China -0.04 -0.06 0.26 0.23 -0.24 -0.23 -0.1 -0.06 0 0 0.09 0.07 (0.58) (-0.88) (6.35) (6.48) (-2.18) (-2.07) (-0.86) (-0.57) (0.22) (0.46) (1.18) (0.84) Economic Development -0.19 -0.23 -0.05 -0.04 0.04 0.07 0.07 0.03 0.003 0.003 0.15 0.17 (5.17) (-4.93) (-2.50) (-1.70) (0.64) (0.93) (1.18) (0.42) (0.49) (0.41) (3.53) (2.76) Resource Endowment . . 0.02 0.03 1.66 0.71 2.61 4.10 . . . . . . (0.15) (0.24) (1.80) (0.63) (1.54) (2.19) . . . . R2 0.70 0.77 0.75 0.86 0.41 0.51 0.24 0.47 0.01 0.46 0.43 0.48 ˆˆˆˆˆˆˆ2222222 0.005 0.005 0.001 0.001 0.012 0.010 0.011 0.011 0.000 0.000 0.007 0.008 Nobs 35 35 35 35 35 35 35 35 35 35 35 35

Table 8. Cross-country differences in end-use energy consumption and the BRIC economies, by sector (SEU) Sector Industrial Transportation Residential and Commercial Agriculture (S = 1) (S = 2) (S = 3) (S = 4) Constant -4.92 -7.63 0.03 1.11 6.48 7.83 -0.59 -0.32 (-1.79) (-2.75) (0.01) (0.37) (3.03) (3.52) (-1.12) (-0.47) Europe . 0.03 . -0.10 . 0.06 . 0.01 . (0.65) . (-1.90) . (1.63) . (0.54) Developed Asia . 0.09 . -0.14 . 0.04 . 0.01 .. ((11..4455)) .. ((--11..9944)) .. ((00..7766)) .. ((00..3322)) Latin America . -0.01 . -0.09 . 0.09 . 0.01 . (-0.20) . (-1.12) . (1.67) . (0.36) Emerging Asia . 0.04 . -0.10 . 0.07 . -0.01 . (0.63) . (-1.51) . (1.43) . (-0.14) Emerging Other . -0.19 . -0.02 . 0.22 . -0.01 . ((--22.5588)) . ((--00.2244)) . ((33.7722)) . ((00.2299)) Brazil 0.07 0.07 0.04 -0.04 -0.12 -0.05 0.02 0.02 (0.69) (0.74) (0.39) (-0.41) (-1.56) (-0.60) (0.83) (0.76) Russia 0.11 0.19 -0.14 -0.25 0.01 0.05 0.02 0.01 (0.94) (1.56) (-1.28) (-1.85) (0.09) (0.47) (0.90) (0.39) India -0.02 -0.04 -0.10 -0.17 0.13 0.21 -0.01 -0.003 ((-00.2233)) ((-00.4433)) ((-11.0077)) ((-11.6655)) ((11.6644)) ((22.8844)) ((-00.3300)) ((-00.1144)) China 0.15 0.15 -0.21 -0.29 0.05 0.14 0.01 0.01 (1.50) (1.55) (-2.30) (-2.74) (0.66) (1.74) (0.55) (0.29) Economic Development 2.58 3.86 0.03 -0.45 -2.92 -3.60 0.32 0.19 (1.93) (2.88) (0.03) (-0.31) (-2.81) (-3.35) (1.22) (0.57) Economic Development (Squared) -0.31 -0.47 0.00 0.06 0.35 0.43 -0.04 -0.02 ((-11.9955)) ((-22.9900)) ((00.0011)) ((00.3344)) ((22.8800)) ((33.3355)) ((-11.2266)) ((-00.6622)) Sector Size 0.09 0.18 0.33 0.36 -0.37 -0.50 -0.05 -0.04 (0.45) (1.02) (1.81) (1.85) (-2.39) (-3.52) (-1.32) (-0.97) Efficiency 0.48 0.65 0.10 -0.31 -0.76 -0.53 2.16 2.11 (0.42) (0.66) (0.09) (-0.28) (-0.86) (-0.66) (0.83) (0.72) R2 0.33 0.63 0.38 0.52 0.59 0.76 0.22 0.31 ˆˆˆˆˆˆˆˆ22222222 0.008 0.006 0.008 0.008 0.006 0.005 0.001 0.001 Nobs 35 35 35 35 35 35 35 35

Table 9. Cross-country differences in end-use energy consumption and the BRIC economies, by industry within sector (IEU) Industrial Sector Transportation Sector Residential and Commercial Sector Energy Intensive Non-energy Intensive Industries Industries Road Industries Non-road Industries Residential Commercial (S = 1; I = 1) (S = 1; I = 2) (S = 2; I = 1) (S = 2; I = 2) (S = 3; I = 1) (S = 3; I = 2) Constant -2.21 -0.48 -0.22 0.56 -3.10 -5.70 -0.04 1.16 5.05 6.68 2.34 -0.68 (-0.76) (-0.17) (-0.15) (0.34) (-1.01) (-1.75) (-0.04) (1.30) (2.22) (2.49) (1.02) (-0.35) Europe . 0.05 . 0.02 . -0.10 . -0.04 . 0.06 . -0.01 . (1.04) . (0.65) . (-1.69) . (-2.39) . (1.37) . (-0.32) Developed Asia . 0.08 . 0.04 . -0.16 . -0.02 . -0.01 . 0.06 . (1.13) . (1.08) . (-2.08) . (-0.98) . (-0.09) . (1.31) Latin America . 0.00 . -0.01 . -0.11 . 0.00 . 0.03 . 0.08 . (-0.001) . (-0.18) . (-1.39) . (-0.12) . (0.48) . (1.73) Emerging Asia . -0.09 . -0.04 . 0.02 . -0.05 . -0.01 . 0.18 . (-1.45) . (-1.13) . (0.27) . (-2.62) . (-0.17) . (4.37) Emerging Other . -0.14 . -0.04 . 0.05 . -0.05 . 0.12 . 0.07 . (-1.83) . (-0.90) . (0.59) . (-2.00) . (1.62) . (1.44) Brazil 0.10 0.08 0.06 0.05 -0.02 -0.07 0.00 -0.02 -0.13 -0.10 -0.03 0.05 (0.97) (0.83) (1.13) (0.83) (-0.14) (-0.63) (0.15) (-0.62) (-1.60) (-1.13) (-0.38) (0.71) Russia 0.05 -0.06 -0.05 -0.10 -0.09 -0.04 -0.01 -0.07 0.12 0.09 -0.05 0.16 (0.40) (-0.44) (-0.81) (-1.43) (-0.67) (-0.29) (-0.24) (-1.69) (1.21) (0.73) (-0.53) (1.89) India 0.06 0.05 0.00 0.00 -0.27 -0.33 0.09 0.07 0.08 0.12 0.05 0.09 ((00.5566)) ((00.5566)) ((00.004)) ((-00.001)) ((-2.3388)) ((-33.0033)) ((33.004)) ((2.42)) ((00.9933)) ((1.337)) ((00.5555)) ((1.337)) China 0.25 0.18 0.03 0.00 -0.29 -0.30 0.01 -0.02 -0.02 -0.01 0.01 0.15 (2.36) (1.72) (0.58) (-0.02) (-2.59) (-2.65) (0.33) (-0.76) (-0.21) (-0.07) (0.12) (2.25) Economic Development 1.09 0.34 0.19 -0.14 1.57 2.79 0.02 -0.55 -2.17 -2.92 -1.13 0.17 (0.77) (0.24) (0.27) (-0.17) (1.05) (1.77) (0.04) (-1.27) (-1.96) (-2.25) (-1.01) (0.18) Economic Development (Squared) -0.13 -0.05 -0.03 0.01 -0.19 -0.32 0.00 0.07 0.24 0.33 0.15 0.01 (-0.74) (-0.28) (-0.32) (0.08) (-1.02) (-1.72) (-0.02) (1.28) (1.84) (2.11) (1.10) (0.06) Sector Size 0.00 0.12 -0.07 -0.04 0.47 0.00 0.00 0.00 -0.13 -0.14 -0.20 -0.36 (0.01) (0.63) (-0.72) (-0.40) (2.12) (2.28) (-0.05) (0.03) (-0.83) (-0.80) (-1.20) (-2.93) Efficiency 0.63 0.40 -0.07 -0.18 -0.40 -0.73 0.49 0.50 0.59 -2.16 -11.25 -0.69 (0.53) (0.39) (-0.12) (-0.31) (-0.31) (-0.61) (1.35) (1.52) (0.05) (-0.19) (-0.99) (-0.08) R2 0.25 0.59 0.16 0.39 0.47 0.69 0.39 0.64 0.57 0.68 0.44 0.79 ˆˆ22 0.010 0.007 0.002 0.002 0.012 0.011 0.001 0.001 0.006 0.006 0.005 0.003 Nobs 35 35 35 35 35 35 35 35 35 35 35 35