International Specialization, Sector Structure and the Evolution of Manufacturing Energy Intensity in OECD Countries

Abstract We present new evidence that changes in sector structure explain a considerable and increasing part of Manufacturing energy intensity trends across 19 OECD countries. We show that cross-country convergence of Manufacturing energy intensity levels is caused by efficiency improvements in lagging countries, while undermined by increasing international differences in sector structure. Particularly, we find that efficiency-driven catching-up processes only began to dominate the diverging impact of structural changes after 1995, reversing gradual crosscountry divergence of Manufacturing energy intensity levels into rapid convergence. Subsequently, we link sector structure dynamics to changing global production patterns under influence of international trade and specialization. We conclude that increasing trade and market integration helped reducing energy productivity gaps across countries, despite the contribution of increasing specialization to growing cross-country variation in sector structure. These trends are mainly driven by energy-intensive sectors, while various countries specialize in sectors for which they do not have a comparative energy productivity advantage.


INTRODUCTION
Changes in aggregate energy intensity result from energy efficiency improvements at the sector and firm level, as well as from changes in the structure of the economy.Efficiency effects are thought to be driven by energy-saving technological change, whereas the impact of a changing production structure (i.e.sector composition) results from the fact that some production processes inherently require more energy inputs than others, relative to capital and labor inputs.In the field of energy studies a popular line of research has been to separate the efficiency effect from the structure effect by means of index number decomposition (or shift-share) analysis (see Ang and Zhang (2000) and Liu and Ang (2007) for surveys).Despite differences as regards the use of data, methodologies, scope and sector detail in the numerous decomposition studies, it is safe to conclude from this literature that in the last three decades of the 20th century trends in aggregate energy intensity have been influenced more by energy efficiency change than by changes in the production structure.
(Section 4).Third, we test for the assumption that countries with relatively high initial Manufacturing energy intensity levels tend to catch-up to more advanced countries (Section 5).Fourth, we examine how and to what extent these catch-up processes are influenced by changes in the structure of economies.To this aim we combine index number decomposition analysis with so-called βconvergence and σ-convergence analyses inspired by the empirical macroeconomic growth literature.Fifth, we identify the degree of specialization across countries and its evolution over time, both in terms of energy consumption and value added (Section 6).To this aim we measure the geographical concentration of Manufacturing energy use over time, and to what extent countries specialize in sectors for which they have a comparative advantage in terms of energy intensity.Finally, we link the observed patterns of specialization and concentration in Manufacturing to the role that structural changes play in driving the aggregate evolution of cross-country differences in Manufacturing energy intensity We make use of a new dataset that was introduced by Mulder and De Groot (2012), who used the data to present an economy-wide analysis of energy intensity trends across OECD countries.In this paper we single out the Manufacturing sector, to provide a more in-depth analysis.Unlike Mulder and De Groot (2012), we assess how energy intensity trends are influenced by changing global production patterns under influence of international trade and specialization, and we identify in detail which Manufacturing sub-sectors drive the observed trends.Clearly, given data availability, there is a trade-off between the number of sectors and the number of countries that one can include in the analysis.Cross-country decomposition and convergence analyses in the field of energy studies typically include up to 15 sectors, and mostly six or seven sectors (see, for example, Aldy 2006, Duro and Padilla 2011, Jakob et al. 2012, Markandya et al. 2006, Miketa and Mulder 2005, Liddle 2009, Romero-Avila 2008, Sun 2002, Unander 2007).In contrast, our analysis includes 25 Manufacturing sectors for 19 OECD countries, covering the period 1980-2005.This is a major advantage because aggregate energy intensity trends may obscure considerable differences across industries (see, for example, Huntington 2010, Jorgenson 1984, Mulder and De Groot 2007, Mulder and De Groot 2012) and the inclusion of a limited sector detail may underestimate the importance of structure effects in a decomposition analysis.
In our analysis we find that, notwithstanding considerable sector heterogeneity, in most countries the rate of energy-intensity decline in Manufacturing accelerated considerably after 1995, especially in the USA and Eastern Europe.Structural changes indeed explain a considerable and increasing part of the observed trends in Manufacturing energy intensity.Cross-country convergence of Manufacturing energy intensity levels is entirely driven by efficiency improvements in lagging countries, while undermined by increasing international differences in sector structure.More specifically, we find that efficiency-driven catching-up processes only began to dominate the diverging impact of structural changes on aggregate energy intensity patterns by the end the 1990s, reversing gradual cross-country divergence of Manufacturing energy intensity levels into robust and rapid convergence.In agreement with this clear break in the trend, we show that increasing trade and market integration helped reducing energy productivity gaps across countries, despite the fact that increasing specialization contributed to growing cross-country variation in sector structure.Energyintensive sectors play a major role in driving these trends, while various countries increasingly specialize in sectors for which they do not have a comparative advantage in terms of energy intensity performance.
The paper proceeds as follows.In Section two we describe our data in more detail.In Sections three to six we present the various steps of our analysis, as described before.Section seven concludes.

DATA
As mentioned before, we make use of a new dataset developed by Mulder and De Groot (2012), which is based on the EU KLEMS database (O'Mahony and Timmer 2009). 1 The dataset offers industry-level measures of both value added and energy inputs, together with supplementary input and productivity data series, derived from a consistent framework of national accounts and supply-and-use tables across countries.As a result, our energy and economic input series and value added series are mutually consistent and avoid the usual matching and aggregation problems caused by different sector definitions across countries and across sources.We measure energy intensity by the ratio of intermediate energy input to gross value added.
Value added data originate from National Accounts and have been converted to constant 1997 US$, using a new and comprehensive dataset of industry-specific Purchasing Power Parities (PPPs).Hence, the underlying data in our analysis do not any longer rely on aggregate countryspecific PPPs, as is often the case in cross-country studies (see for example Miketa andMulder 2005, Mulder andDe Groot 2007).In the EU KLEMS project, the PPP series were constructed by double deflation of gross output and intermediate inputs within a consistent input-output framework.Price series for output and total intermediate inputs at the industry level are taken directly from the National Accounts.The EU KLEMS database has largely been constructed on the basis of data from national statistical institutes (NSIs) and processed according to harmonized procedures.These procedures were developed to ensure international comparability of the basic data and include the use of similar price concepts for inputs and outputs (O'Mahony and Timmer 2009).
Energy data consist of expenditure based intermediate inputs that encompass all energy mining products, oil refining products and electricity and gas products.2Using detailed supply-anduse tables, energy expenditures at the sector-level have been deflated in the EU KLEMS project by the relative price index of each fuel (energy carrier).Finally, because these intermediate energy input series are provided in volume indices, Mulder and De Groot (2012) have matched these with final energy consumption data provided by the International Energy Agency (IEA), to obtain energy consumption in kilo tonnes of oil equivalent (ktoe).The latter allows us to not only assess energy intensity growth rates but also energy intensity levels across countries and subsectors.
Our analysis covers 25 Manufacturing sectors (10 main sectors and 15 subsectors) and includes 16 EU member countries, the USA, Japan and South Korea.To ensure comparability of data across countries, our analysis covers the period 1980-2005 and distinguishes the period 1980-1995 (14 countries) from the period 1995-2005 (19 countries).Table 1 provides an overview of country-and time coverage, including the clustering of countries in various regional aggregates.

ENERGY INTENSITY TRENDS
We start our analysis by presenting in Table 2 key stylized facts about the role of Manufacturing in the economy, in the first and last year of our analysis (1980 and 2005).The first two columns in Table 2 clearly show a trend in developed countries towards a smaller and less energy  The remaining part of Table 2 shows within the Manufacturing sector a clear overall shift away from energy intensive sectors (Paper, Chemicals, Non-Metallic Minerals and Basic Metals)except for the EU8 region.As regards the latter, underlying data reveal that this is due to a combination of i) an increasing value added share of the Chemical sector, ii) an increasing energy share of the Paper sector, and iii) a relatively slow decrease of the Basic Metals industry in the EU as compared to the USA and the OECD11 sample.More detailed information can be found in the Data Annex to this paper, Table A1 (available online). 3 Next, we present in Table 3 the average annual growth rate of aggregate Manufacturing energy intensity, per country and for two different time periods (1980-2005 and 1995-2005).Table 3 leads to various important observations.First, over a longer period of time in most countries Manufacturing energy intensity levels are decreasing on average.Second, energy intensity growth rates differ substantially across countries, and also include exceptions-most notably Italy, Portugal and Spain, that all exhibit increasing energy intensity levels after 1995.Third, in general, Manufacturing energy intensity levels decreased relatively slow before 1995 and relatively fast afterwards, especially in the USA (6.4%) and the EU4 region (5.2%).As can be seen from the right-hand side 3. See www.petermulder.net/publications.Information in the Data Annex reveals various interesting trends in the less energy-intensive sectors.First, the relative share of the Machinery sector is increasing in the USA and OECD11 sample, but decreasing among Western European economies (EU8).Second, the share of the Transport Equipment sector in aggregate Manufacturing value added is decreasing in the USA but increasing in the EU8 group, while its energy consumption share is increasing across all countries.Moreover, it can be seen that overall the Textiles sector considerably lost economic weight.Finally, the Food sector presents a mixed picture, with a decreasing value added share and an increasing energy share.
Copyright © 2015 by the IAEE.All rights reserved.
of Table 3, these results imply that Manufacturing energy intensity levels are gradually converging within the OECD.Particularly the USA and Eastern European countries (EU4) have been catchingup to the relatively low energy intensity levels in Western European economies (EU8). 4At the same time, cross-country differences in energy intensity levels remain substantial with, in 2005, Finland and Slovakia having an aggregate Manufacturing energy intensity level almost twice that of the OECD11 average, and more than two times higher than Sweden.
Because of space limitations we cannot discuss in detail all underlying country-specific sector patterns that drive these aggregate trends.We refer to the Data Annex to this paper (available online), where we list energy intensity growth rates of individual Manufacturing sectors per country, differentiated for the periods 1980-2005, 1980-1995 and 1995-2005.Here, we limit ourselves to explain the most outstanding results presented in Table 3.The strong decline in Manufacturing energy intensity performance in Finland after 1995 is mainly due to improved performance in Machinery.The slowdown of Manufacturing energy intensity decrease in Denmark after 1995 is caused by a relatively poor performance in the sectors Textiles, Basic Metals, Transport Equipment and Machinery.The relatively large decrease in Manufacturing energy intensity level in Sweden results mainly from improved performance of the sectors Machinery, Chemicals and Non-Metallic Minerals.Finally, the substantial increase in Spanish Manufacturing energy intensity level is mainly due to increasing energy intensity levels in the sectors Food, Non-Metallic Minerals, Basic Metals, Transport Equipment and Machinery.
We conclude this section by summarizing in Table 4 sector-specific trends for the same selection of (groups of) countries as shown before, measured in terms of annualized energy intensity growth rates at subsector level.The Table shows that across countries, in general, energy intensity has decreased in nearly all Manufacturing sectors, except for the Tobacco industry.By far the largest energy intensity reductions have been realized in the sectors Office/Accounting/Computing Machinery and Electrical Engineering.In addition, across regions energy intensity has also reduced relatively fast in Chemicals.The same is true for the Pulp and Paper industry, except for the EU8 region (as noted before; cf.Table 2).Furthermore, Table 4 shows that the catching-up process by the USA appears to be driven by a relatively strong energy intensity decline in the sectors Motor vehicles, Rubber and Plastics, Basic Metals and Textiles.The catching-up process in the EU4 also seems to find its cause in the sectors Motor vehicles and Rubber and Plastics, as well as in the Non-Metallic Minerals and Leather and Footwear industries.We discuss and decompose these catchingup processes in more detail in Section 5.

DECOMPOSING ENERGY INTENSITY LEVELS
In this section we use index number decomposition (or shift-share) analysis to decompose changes in aggregate energy intensity levels into a so-called structure effect and an efficiency effect.The structure effect measures the change in aggregate Manufacturing energy intensity due to the changing composition of subsectors within Manufacturing.The efficiency effect, in contrast, measures changes due to efficiency improvements within each Manufacturing subsector at a constant subsector structure.In the field of energy studies index number decomposition methodology has been widely used to decompose aggregate changes in energy use, energy intensity, or emission intensity (see Ang andZhang 2000 andLiu andAng 2007 for reviews).We use the so-called log    mean Divisia index method (LMDI I) as introduced by Ang and Liu ( 2001), which is defined as the geometric average of the Laspeyres and Paasche indices and, in our (two-factor) case, equivalent to the Fisher ideal index method (Ang, 2004, De Boer, 2009, Boyd and Roop, 2004). 5We refer to the Annex for a definition of the decomposition method.
Figure 1 shows for different (groups of) countries the impact of the structural effect and the efficiency effect on aggregate Manufacturing energy intensity trends over time.The left-hand side of Figure presents the evolution of aggregate Manufacturing energy intensity insofar driven by efficiency improvements within each Manufacturing subsector at a constant subsector structure.It clearly shows that once we correct for the impact of structural changes, aggregate Manufacturing energy intensity features a consistent decline over time.As noted before, this decline is especially strong in the USA after 1995.
The right-hand side of Figure 1 shows aggregate Manufacturing energy intensity trends insofar driven by changes in the subsector structure.It reveals that, in contrast to the efficiency effect, these structural changes caused aggregate Manufacturing energy intensity to increase in the period before 1995, especially in the EU8 region.After 1995, however, this trend reversed, with structural changes increasingly contributing to decreasing aggregate energy intensity levels, especially in the EU4 region.In addition, Figure 1 suggests that cross-country differences in aggregate Manufacturing energy intensity are increasingly driven by differences in sector structure, rather than differences in energy efficiency.We return to this issue in the next section.In the remaining part of this section we look in more detail at the underlying trends.
To do so, we present in Table 5 for each country and two time periods (1980-2005 and 1995-2005) the average annualized energy intensity growth rates insofar driven by, respectively, efficiency improvements within Manufacturing subsectors (EFF) and structural changes (STR) in 5.The choice for this approach is primarily motivated by its ability to satisfy the factor-reversal test, i.e. it provides perfect decomposition results without a residual (interaction) term, which frequently is relatively large and by definition difficult to interpret.Moreover, this approach can handle zero values effectively, the results are invariant to scaling and it satisfies the time-reversal test, i.e. estimated values between period 0 and T and period T and 0 are equal (in absolute terms).For the aforementioned reasons the LMDI and Fisher ideal index methods have emerged as the preferred methods in energy decomposition analysis (Ang 2004).Recent applications include Metcalf (2008) and Huntington (2010).Copyright © 2015 by the IAEE.All rights reserved.
the composition of the Manufacturing sector-which together add up to the aggregate growth rates (TOT) presented in Table 3.The results in Table 5 show that within-sector efficiency improvements lead to negative Manufacturing energy intensity growth rates, except for South Korea as well as Italy, Portugal and Spain after 1995.In contrast, measured over the period 1980-2005, structural changes have caused aggregate Manufacturing energy intensity levels to grow and decline, depending on the country.After 1995, however, in nearly all countries (Austria and the Netherlands being the exceptions) structural changes have encouraged reductions in aggregate Manufacturing energy intensity.Finally, it can be concluded that structural changes play an increasingly important role in explaining Manufacturing energy intensity changes: especially after 1995, in a range of countries the structure effect is stronger than the efficiency effect.Because of differences in sector detail, time periods and decomposition methods, our results are not straightforwardly comparable to related country-specific studies in the literature.However, recent research on energy intensity of the U.S. economy (see, for example, IEA 2004, Lescaroux 2008 andHuntington 2010) does allow for such a comparison.It appears that overall our results correspond well with the findings of these studies. 6Also, our results regarding the role 6.Energy intensity change according to our data: -1.97% for the period 1980-1995, -3.72% for the period 1980-2005 and -6.35% for the period 1995-2005.IEA (2004): -2.7% for the period 1973-1998;Lescaroux (2008): -2.2% for the period 1974-1998 (41.9% decline over 24 years); Huntington (2010): -5.75% for the period 1997-2006.Our results largely reconcile with these findings if we consider the different time periods: the decline in energy intensity accelerated after the first oil price shock of 1973 and slowed down since the mid-1980s with the fall in energy prices.
Copyright © 2015 by the IAEE.All rights reserved.of structural change in explaining these reductions are in line with what other studies have reported.According to our data, about 18% to 22% of the reduction in U.S. aggregate energy intensity is due to changes in the sectoral composition of the U.S. economy.Using a similar two-digit decomposition approach Metcalfe (2008) and Lescaroux (2008) find an 18% and 17% contribution of structural change in the periods 1974-1997 and 1974-1998, respectively. 7Using a 65-sector structure Huntington (2010) finds that structural changes explain about 39% of U.S. manufacturing energy intensity reductions between 1997 and 2006.Given similarity in decomposition methods used, the difference between the latter finding and our results is to be attributed to differences in sector detail and data used (a lower degree of disaggregation obscures structural shifts, which then show up as efficiency improvements).
In order to examine the role of individual Manufacturing sectors in driving the results presented above, we identified per individual Manufacturing sector the percentage contribution of the total efficiency effect and the total structure effect to the growth rate of aggregate Manufacturing energy intensity.Again, due to space limitations we cannot present the detailed figures here-we refer to Table A2 in the Data Annex.In summary, the results show that across countries, in general, the overall efficiency effect is primarily driven by efficiency improvements in the energy intensive sectors Chemicals, Non-Metallic Minerals (before 1995), Basic Metals (except EU8 after 1995), and Pulp and Paper (except EU8 before 1995).In the Eastern European region (EU4) developments are somewhat different: the contribution of sectoral efficiency improvements after 1995 is, in addition to Chemicals, particularly strong in Non-Metallic Minerals as well as in Food and Beverages and Motor Vehicles, but on the other hand exceptionally weak in Pulp and Paper and even absent in Basic Metals.In addition, efficiency improvements in the Food and Beverage make a considerable contribution to the overall efficiency effect in the USA, while in the EU8 region increasing efficiency in the Rubber and Plastics industry considerable contribute to the overall efficiency effect.
With regard to the role of structural changes, the sub-sector data show that across countries, in general, the overall structure effect is primarily driven by a structural shift away from Basic Metals and, except for the EU8 region, Chemicals.In the EU4 region this effect is further enhanced by a decreasing relative size of Non-Metallic Minerals and, to a lesser extent, Food and Beverages.In addition, in the EU8 region a considerable increase in the relative size of the Rubber and Plastics industry has weakened the overall effect of structural changes on lowering energy intensity levels.
Finally, on the subject of the distinct role of the energy-intensive sectors, detailed country figures reveal that efficiency improvements in the Chemicals sector have been especially strong in Austria, Germany, Japan (before 1995), Sweden and the UK; in the Non-Metallic Minerals energy efficiency growth have been especially large in the Czech Republic, Slovakia and Poland; the Basic Metals industry experienced its largest efficiency gains in France and Slovakia; efficiency increases in the Pulp and Paper have especially high in the Czech Republic, Finland (after 1995) and Hungary.In addition, the structural shift away from Basic Metals has been most notable in Belgium, Japan, Spain and the UK, while the relative size of this sector has largely increased in France.The structural shift away from Chemicals has been especially strong in France, South Korea, the USA and the Eastern European countries, while the opposite is true in Belgium, Denmark, Germany, the UK and especially the Netherlands-the latter countries increasingly specialize in Chemicals.France, Belgium and Denmark are mainly responsible the considerable increase in the relative size of the Rubber and Plastics industry in the EU8 region.We refer to the Data Annex for more detailed information on country-specific decomposition results.
In short, thus far our analysis showed that notwithstanding considerable cross-country heterogeneity, Manufacturing energy intensity levels decreased relatively slow before 1995 and relatively fast afterwards, that changes in sector structure explain a considerable and increasing part of Manufacturing energy intensity trends across 19 OECD countries, that Manufacturing energy intensity levels tend to gradually converge across these countries, but that remaining cross-country differences in energy intensity levels are increasingly driven by differences in sector structure, rather than differences in energy efficiency.Together, these conclusions make us to further analyze the evolution of cross-country differences in energy intensity levels overtime.This is the topic of the next section.

DECOMPOSING ENERGY INTENSITY DIFFERENCES
In this section we analyze and decompose the evolution of cross-country energy intensity differences over time.Cross-country divergence or convergence of energy intensity levels can be understood both in terms of levels and growth rates, which-building on the empirical macroeconomic growth literature-translates into a distinction between so-called σ-convergence and β-convergence (e.g., Barro 1991, Barro andSala-i-Martin 1992).The former refers to a decreasing crosscountry variance of productivity or intensity levels, while the latter refers to a tendency of countries with relatively high (low) initial intensity (productivity) levels to grow relatively fast, building upon the proposition that growth rates tend to decline as countries approach their steady state. 8 First, we conduct a β-convergence analysis, testing for the assumption that countries with relatively high initial energy intensity levels tend to catch-up to more advanced countries.We do so by regressing for each sector i the growth rate of energy productivity I* between period 0 and T, on its initial level. 9A statistically significant negative relation between the initial energy productivity level (at period 0) and its subsequent growth rate (measured by the coefficient β) is then taken as evidence of convergence.Subsequently, we use the estimated values of β, to derive the speed k at which the energy-intensity levels are converging to their steady-state level (Barro and Sala-i-Martin, 1992;Islam, 1995). 10We refer to the Annex for details.
In addition, we calculate to what extent aggregate β-convergence is to be explained from, respectively, convergence in the underlying sector structure and convergence of efficiency improvements within individual Manufacturing sectors.To this aim we use a modified version of the sectoral β-decomposition approach developed by Wong (2006).This sectoral β-decomposition approach is based on the identity: β ≡ ∑β + ∑β , where β eff and β str are the coefficient estimates obtained eff str i i from regressing for each sector i the weighted growth rate of, respectively, its energy productivity and value added share between period 0 and T on the initial level of aggregate (economy-wide) energy productivity.In these regressions the weights are derived from the log mean Divisia index 8.Obviously, both concepts of convergence are closely related.β-Convergence is a necessary condition for σ-convergence: decreasing variation in energy intensity levels across countries implies that countries with a relatively high initial energy intensity level tend to improve their performance relatively fast.However, as has been argued by Quah (1993), βconvergence is not a sufficient condition for σ-convergence, because of potential regression towards the mean.For this reason we use both methodologies to assess trends of convergence and/or divergence.9. We use energy productivity (I*) instead of energy intensity (I), in analogy to β-convergence analyses in the mainstream economic literature, which predominantly focus on convergence of labour productivity and total factor productivity.
10. Using the estimated values of β, the speed k at which the energy-intensity levels are converging to their steady-state level is derived according to: k = -[(1/T)log(β + 1)], with T denoting the length of the time interval under consideration.
Copyright © 2015 by the IAEE.All rights reserved.decomposition method (LMDI), like in the previous section; we refer to the Annex for details.The results of this analysis are presented in Table 6.
Table 6 provides evidence that, across all samples, countries with relatively high initial aggregate Manufacturing energy intensity levels indeed tend to catch-up to more advanced countries (TOT).Moreover, we find that this process is entirely driven by a relatively fast improvement of lagging energy efficiency levels within lagging individual Manufacturing sectors (EFF).In contrast, the underlying sector structure shows clear evidence of divergence (STR), thus weakening the overall catching-up process.Measured over the whole sample period , in the OECD 11 and EU8 samples, respectively about 32% and 18% of energy efficiency catch-up was undone by a diverging sector structure.11 After 1995, these percentages increase, especially in the larger samples of OECD19 and EU16, where structural changes undermined most of the catch-up in energy efficiency.Our results show that the inclusion of Eastern European countries in the sample (cf.EU16 and OECD19 samples), substantially increases the role of structural changes in attenuating the catching-up process across countries.Furthermore, our results show that the speed of β-convergence is largest in the relatively homogeneous EU8 sample and lowest in the relatively heterogeneous OECD19 sample.This means that, notwithstanding the significant catch-up of the USA and Eastern Europe to Western Europe (see Section 3), cross-country differences in energy intensity levels tend to decrease relatively fast among neighbouring countries close to the frontier within Western Europe.
As noted before, β-convergence is a necessary but not a sufficient condition for decreasing cross-country differences in levels of energy productivity or intensity-also known as σ-convergence.Hence, we continue by assessing to what extent the observed catching-up processes led to decreasing variance in energy intensity levels among countries.To this aim we present in Figure 2 for different samples of countries the evolution of cross-country differences in energy intensity levels, measured as the cross-country standard deviation of the logarithm of energy intensity.The left-hand side of the Figure is based on actual energy intensity levels; the right-hand side is based on hypothetical energy intensity levels that correct actual levels for cross-country differences in sector structure.More precisely, the so-called 'common structure' energy intensity level defines what a country's intensity level would have been if it had the same Manufacturing sector structure as all other countries. 12 The left-hand side of Figure 3 reveals a remarkable strong and robust break in the trend at the end of the 1990s: before that period cross-country differences in energy intensity levels increase over time, where after this period energy intensity levels rapidly converge across countries.The right-hand side of Figure 2 shows a less pronounced picture: if we assume that all countries have the same Manufacturing sector structure, we find a long-term tendency of energy intensity levels to converge across countries, even though this trend is again particularly strong since the end of the 1990s.Together, these findings confirm that during the period 1980-2000 a diverging Manufacturing sector structure (see also the right-hand side of Figure 1) has more than offset the efficiency-based catching-up processes (β-convergence), leading to cross-country divergence of actual aggregate Manufacturing energy intensity levels (σ-divergence).Since the end of the 1990s, however, this trend clearly reversed, with efficiency improvements in lagging countries (catch-up) caus-11.Viz.47/147 (OECD11) and 21/121 (EU8).12.It is calculated as the aggregate product of actual energy intensity levels within each manufacturing subsector and a common sector structure.ing a rapid convergence of Manufacturing energy intensity levels (σ-convergence), despite a diverging sector structure.

Table 6: β-convergence and Its Decomposition
To further understand this break in the energy productivity trend, we provide in Table 7 a breakdown of the σ-convergence trends.The left-hand side of the Table presents the numbers underlying Figure 2, showing the degree of cross-country variation in actual and common structure energy intensity levels, again measured as the standard deviation of the logarithm of energy intensity.A systematic comparison of the two data series clearly reveals that over time an increasing part of cross-country differences in Manufacturing energy intensity levels is to be attributed to cross-country variation in sector structure (i.e. a structural effect).More precisely, in 2005 crosscountry variation in the Manufacturing sector structure explain about 20% of cross-country differences in Manufacturing energy intensity levels.The right-hand side of Table 7 summarizes these trends in percentage change of cross-country variation in energy intensity levels for the various time periods under consideration.The results shows again that across the various groups of coun- tries, changes in the sector structure have encouraged cross-country divergence of Manufacturing energy intensity levels, especially in the period 1980-1995.This finding confirms that cross-country differences in aggregate Manufacturing energy intensity indeed are increasingly driven by differences in sector structure, while trends in energy efficiency performance within Manufacturing subsectors encourage convergence of aggregate Manufacturing energy intensity levels-as already suggested by the decomposition results presented in the previous section (see also Figure 1).
To asses which Manufacturing sectors drive the aggregate trends discussed above, we calculated per Manufacturing sector the speed of β-convergence (k), in terms of both the efficiency and structure effect. 13The results shows that the overall catching-up process in the Manufacturing sector is largely driven by catching-up processes in energy-intensive sectors-most notably Non-Metallic Minerals, Basic Metals and Rubber and Plastics-as well as in the Food and Beverages industry.After 1995, the sector Printing and Publishing also plays a leading role.Interestingly, the main source of the overall negative aggregate structural effect is also to be found in the energyintensive industry.More precisely, in particular structural changes in the sectors Non-Metallic Minerals, Basic Metals and Pulp and Paper contribute to the overall diverging sector structure that weakens energy efficiency catching-up in the aggregate Manufacturing sector.Finally, the sectors Pulp and Paper as well as Chemicals stand out for their large and contrasting contribution to aggregate convergence patterns after 1995.Across the various samples The Paper and Pulp industry is characterised by a strong pattern of divergence, both in terms of sector structure and energy efficiency levels.Remarkably, except for the EU8 region, the Chemical sector combines strong divergence in energy efficiency levels with strong convergence in sector structure.
We conclude this section by presenting in Table 8 sector dynamics that drive the aggregate trends of σ-convergence.From the Table it can be seen that the aforementioned trend-break in Manufacturing is largely caused by strong cross-country convergence of energy intensity levels in the sectors Food and Basic Metals, and further enhanced by convergence in the sectors Pulp & Paper, Textiles (in the EU region) and Transport Equipment (in the OECD 19 and EU 16 sample).Furthermore, the Table shows that overall within the Manufacturing sector σ-convergence is particularly strong in the sectors Textiles and Leather (especially in the EU8 region), Wood and Cork, Basic Metals and, before 1995, also Non-Metallic Minerals.In contrast, in the Machinery industry differences in energy intensity levels across countries increased considerably, especially within the EU8 region.Albeit much weaker, during the period 1995-2005 we also find evidence of σ-divergence in the Chemical and Non-Specified Industries, especially among the EU16 and OECD19 sample.Underlying data reveal that as a result of these and other trends, in 2005 cross-country variation in energy intensity levels is relatively high in the subsectors Wood and Non-Specified Industry and relatively low in the subsectors Food and Textiles.Except for the Paper & Pulp industry, cross-country differences in the energy-intensive subsectors are relatively low.
In short, in this section we have provided evidence for cross-country convergence of Manufacturing energy intensity levels, and we have shown that it is caused by efficiency improvements in lagging countries, while undermined by increasing international differences in sector structure.As such, we have confirmed and reinforced the key conclusion of section 4 that Manufacturing cross-country differences in energy intensity levels are increasingly driven by differences in sector structure, rather than differences in energy efficiency.Also, in section 4 we found that changes in sector structure explain a considerable and increasing part of Manufacturing energy intensity trends across our sample of OECD countries.In this section, we found that gradual crosscountry divergence of Manufacturing energy intensity levels reversed into rapid convergence only after 1995, because only after 1995 the efficiency-driven catching-up processes began to dominate the diverging impact of structural changes.Together, these findings asks for a further inquiry into the nature and causes of structural changes across OECD economies.To this aim, in the next section we link the observed sector structure dynamics to changing global production patterns under influence of international trade and specialization.

THE ROLE OF SPECIALIZATION AND GEOGRAPHIC CONCENTRATION
In this section we analyze the role of international specialization and geographical concentration of Manufacturing production in explaining energy intensity trends, through its impact on changes in production patterns across OECD countries.We do so by using two commonly-used indicators to measure the (spatial) structure of economies.The first indicator is the Krugman specialization index, which measures the extent to which a country's production pattern differs from those of a comparison group of countries (Krugman 1991).The second indicator is the spatial Hirschman-Herfindahl Index, which captures the degree of concentration of a particular industry in a particular geographic market (Hirschman 1964, McCann 2001).We measure the degree of specialization and concentration both in terms of value added and energy consumption.
Letting S i,j be the share of industry i in country's j Manufacturing output (energy consumption) and S i * that share in the comparison group, the Krugman index in country j is defined as (1) i It runs from zero if the country and group produce the same goods in the same proportions, to two if they produce only different goods.Because of its geographical coverage and consistent data availability, the comparison group is defined as the aggregate of the OECD11 countries.The Hirschman-Herfindahl (HH) Index for industry i is defined as Copyright © 2015 by the IAEE.All rights reserved.with Q ij the share of country j in the output (energy use) of industry i in the entire geographical market, and Q j * the share of country j in the total output (energy use) of the entire geographical market.A high value of the HH Index indicates concentration of industry i in a limited numbers of countries.Table 9 summarizes for various country samples the development of the Krugman and HH-indicators over time, expressed in terms of an average percentage change, in two time periods.The Table provides compelling evidence that, measured in value added, both the degree of specialization and geographical concentration have greatly increased in the post-1995 period.This development is in sharp contrast with the pre-1995 period, which, in general, was characterized by a clear trend of decreasing specialization and geographical concentration of industries.In addition, Table 9 shows that the degree of specialization also has greatly increased in terms of energy use after 1995, which means that the economic structure of energy consumption increasingly differs across countries.In contrast, Table 9 shows that geographical concentration of energy use has decreased considerably in the period after 1995.Together, these findings are clearly in agreement with the previously presented evidence that lagging countries catch-up in terms of energy efficiency performance, while increasing international differences in sector structure undermine this convergence process.
In Figure 3 we present additional evidence with regard to the geographical concentration of industries.Figure 3A shows that energy extensive sectors are more concentrated than energyintensive sectors (ratio >1).Also, it leads to the conclusion that within the group of EU8 countries, energy intensive industries increasingly concentrate, relative to energy extensive industries.A closer look at the data reveals that since 1995 indeed sectors like Chemicals, Non-Metallic Minerals, Basic Metals have been subject to a relatively high increase in geographical concentration (see Table 10).Figure 3B provides weak evidence that sectors with low energy efficiency growth rates concentrate more than countries with high energy efficiency growth rates.In other words, increasingly high energy efficiency growth rates become geographically less concentrated.This accords well with our finding that across sectors lagging countries increasingly catch-up to the frontier, implying a pattern of convergence i.e. less spatial variation.Finally, Figure 3B also shows that this process is most advanced within the group of major EU countries (EU8), which is also an area with a relatively high degree of market integration.In the previous section we documented a clear structural break in the evolution of crosscountry differences in Manufacturing energy intensity levels: by the end of the 1990s gradual crosscountry divergence of Manufacturing energy intensity levels reversed into robust and rapid convergence.To further understand the role of international specialization in explaining this remarkable pattern, we continue our analysis by analyzing cross-country differences in specialization over time.Similar to the σ-convergence analysis of energy intensity levels presented in section 5, we calculated over time the cross-country standard deviation of the logarithm of the country-specific Krugman specialization indices.The result of this exercise is presented in Figure 4.
The Figure shows that within various samples, the differences in degree of specialization across countries have increased substantially since the first half of the 1990s.In other words, countries increasingly differ in terms of their deviation from the production pattern of the EU/ OECD group as a whole.Figure 4 thus provides additional evidence of increased heterogeneity in production processes across countries.This patterns suggests the existence of increasing fragmentation of production processes across countries, a ´la Krugman (1995), under influence of increasing cross-country market integration, facilitated by increasing trade.Fascinatingly, the remarkable strong trend-break (increase) in cross-country variation of specialization patterns by the first half of the 1990s (as shown in Figure 4) clearly is-with a 3-year lag-in parallel with the strong trend- break (decrease) in cross-country variation of Manufacturing energy intensity levels in the post-1995 period (as shown in Figure 2).
In the previous section we argued that the structural break in cross-country variation of energy intensity levels by the end of the 1990s is caused by efficiency-driven catching-up processes that began to dominate the diverging impact of structural changes.Together with the evidence on the parallel structural break in cross-country variation of specialization patterns (as shown in Figure 4) this suggests that increasing trade and market integration contributed to cross-country convergence of energy intensity levels by accelerating knowledge diffusion and/or equalization of factor prices, for example via high-tech imports or increasing international competition.Moreover, it suggests that since the second half of the 1990s this mechanism has (by far) outweighed the countervailing power of specialization (increased heterogeneity) to increase divergence of energy intensity levels via its contribution to diverging production structures.This idea also finds support in Mulder and De Groot (2012), who show that aggregate convergence patterns are almost exclusively caused by convergence of within-sector energy productivity levels (i.e. a 'technology effect'), and not by convergence of the sectoral composition of economies.Certainly, our results make that this hypothesis deserves to be explicitly and carefully tested in future work.
To further identify the link between the patterns of specialization described above and changing patterns of Manufacturing energy intensity we develop a new index that builds upon the Krugman specialization index.Recall that the Krugman index measures the extent to which a country's production pattern differs from those of a comparison group of countries.We modify this index such that we measure the extent to which the energy intensity performance embodied in a country's production pattern differs from the comparison group of countries, according to: with S i,j the share of industry i in country's j Manufacturing output and S i * that share in the group.
Similarly, I i,j is the energy intensity of industry i in country j and I i is that energy intensity in the group.The group is defined as the OECD11 sample, given its geographical coverage and consistent data availability during the entire sample period 1980-2005.The index is zero if the country and group have the same sector structure and energy intensity level within each sector; it is negative if the country specializes in sectors that are less energy intensive than the group average, and it is positive if the country specializes in sectors that are more energy intensive than the group average.Based on this 'energy specialization index' Q, Figure 5 presents the evolution of specialization patterns over time, measuring the extent to which countries specialize in sectors in which their energy intensity performance is relatively good (<0) or bad (>0), as compared to the OECD11 average (0).The Figure leads to a couple of observations.First, production patterns in Eastern European countries and Finland used to be highly energy intensive, but are now rapidly converging to the OECD11 average-in line with results presented in previous sections.In contrast, the energy intensity level of the production pattern in Austria, Belgium, Denmark, Spain and the United Kingdom used to be (considerably) lower than the OECD11 average, but these countries are increasingly specializing in sectors for which they do not (any longer) have an energy intensity advantage.The same appears to be true for Japan, and after 1995 also for South Korea.Finally, the USA, France, Germany, Italy, The Netherlands, Portugal and Sweden do not feature a clear relation between specialization patterns and a comparative (dis)advantage in energy intensity performance, although especially in Italy and Portugal the production patterns is characterized by a fairly substantial and constant relatively low level of energy intensity.
To asses which sectoral developments drive these aggregate trends we calculated for each Manufacturing sector its percentage contribution to the total change in the value of the specialization index Q.Again, because of space limitations we limit ourselves to highlight the main results, while all country-specific details can be found in the Data Annex (Table A4).First, the strong decrease in the relative energy intensity of the production process in Finland is almost entirely caused by a strong increase in energy efficiency in the sector Printing and Publishing (see also the Data Annex).Next, in the Eastern European countries, the convergence to the OECD11 average is driven by increasing relative energy efficiency performance in a variety of sectors, most notably Basic Metals and Machinery, as well (except for Poland) the Textile and Leathers industry.Furthermore, the data do not reveal a common specialization pattern underlying the specialization in relative energy intensive sectors in Austria, Belgium, Denmark, Spain and the United Kingdom.The same is true for Japan, and also, by definition, for the group of countries that do not feature a clear relation between specialization patterns and a comparative (dis)advantage in energy intensity performancei.e. the USA, France, Germany, Italy, The Netherlands, Portugal and Sweden.It is to be noted, however, that France and Germany feature a relative high degree of sectoral dynamics as compared to other Western European countries, whereas the small improvement in the relative position of Sweden is almost completely driven by energy efficiency improvements in its relative large Motor vehicles industry.

CONCLUSIONS
We analyzed for 19 OECD countries how and to what extent aggregate Manufacturing energy intensity trends during the period 1980-2005 have been influenced by changes in the production structure (i.e.sector composition).Using a new dataset and a combination of index number decomposition analysis and convergence analysis techniques, we decomposed the development of both levels and differences across countries in Manufacturing energy intensity over time.Our results provide new evidence that structural changes explain a considerable and increasing part of Manufacturing energy intensity dynamics across countries.In addition, we discovered a clear structural break in the evolution of cross-country differences in Manufacturing energy intensity: by the end of the 1990s, a historic and acknowledged pattern of gradual increasing cross-country variation in energy intensity levels reversed into a robust pattern of rapid cross-country convergence.We found that convergence of Manufacturing energy intensity levels is entirely driven by efficiency improvements in lagging countries, while undermined by increasing international differences in sector structure.We showed that by the end of the 1990s, efficiency-driven catching-up processes began to dominate the diverging impact of structural changes on aggregate energy intensity patterns, causing a trend-break toward rapidly decreasing cross-country differences in Manufacturing energy intensity.
Subsequently, we related these findings to changing global production patterns under influence of international trade and specialization.We provided evidence that in more recent years, among OECD countries specialization and geographical concentration of Manufacturing production have greatly increased, whereas geographical concentration of energy use has considerably decreased.This is clearly in agreement with our findings regarding the development and driving forces of cross-country convergence in Manufacturing energy intensity levels.Indeed, if, as we found, lagging countries increasingly catch-up in terms of energy efficiency performance, one would indeed expect decreasing geographical concentration of energy use.Also, increasing specialization and geographical concentration of Manufacturing production obviously leads to increasing international differences in sector structure, and this development, as we showed, undermines cross-country convergence.With regard to the latter, in the paper we documented a trend-break in cross-country variation of specialization patterns that is strikingly parallel to the previously mentioned trend-break in cross-country variation of Manufacturing energy intensity levels.In sum, our evidence suggests that since the second half of the 1990s increasing trade and market integration helped reducing energy productivity gaps across countries.
The broader literature identified various factors that may help explain this result, including the role that increasing trade and market integration play in facilitating knowledge diffusion and equalization of factor prices, via high-tech imports or increasing international competition.In addition, it may be that increasing political and institutional integration in the European Union also plays a role, for example via cross-country harmonization of environmental standards.Also, it is likely that energy-intensity trends have been influenced by the increasing international oil price, which tripled in the period between 1999 and 2005.Certainly, these suggestions deserve to be carefully tested for in future work.Our results show that since about 15 years the forces that stimulate cross-country convergence of Manufacturing energy intensity levels visibly outweigh the contribution of increased specialization to diverging cross-country energy intensity levels via increasing international differences in sector structure.
Finally, we found that within countries changes in the sector structure increasingly contribute to decreasing energy intensity levels, while some countries increasingly specialize in sectors for which they do not have a clear energy productivity advantage.While the former may be explained by relocation of energy intensive industries to non-OECD countries, this is contradicted by the latter observation.Resolving this puzzle requires more insight into changing relative factor prices and patterns of technology diffusion under influence of increasing trade and market integration.
In this context, an important issue to consider in future research is the new situation about gas prices.Over the last years (since 2008/9), natural gas prices rose in Europe and Asia, but fell in North America, where rising US natural gas output pushed gas prices to record discounts against both crude oil and international gas prices.This may slow down the relatively rapid decrease in US Manufacturing energy intensity that we observed in this paper for the period 1995-2005.Also, increasing internationally diverging energy prices may further contribute to international specialization, with energy-intensive sectors increasingly concentrating in North America.If so, this would reinforce the observed trend that structural changes explain a considerable and increasing part of aggregate Manufacturing energy dynamics across countries.Future research should assess whether this will undermine the current trend towards international convergence of Manufacturing energy intensity levels.

ACKNOWLEDGMENTS
I would like to thank Henri L. F. de Groot and three anonymous referees of this journal for valuable comments on earlier versions of this article.

ANNEX-BETA CONVERGENCE AND ITS DECOMPOSITION
This Annex describes the decomposition and convergence methods used in Sections 4 and 5.In section 4 we use the log mean Divisia index method (LMDI I) to decompose changes in aggregate energy intensity levels into a so-called structure effect and an efficiency effect, as follows.Aggregate energy intensity I is defined as the ratio of energy (E) to output (Y) according to where w i is the weighting function defined as w = L(V ,V ), with V = ∑I S and L the logarithmic In section 5 we conduct a β-convergence analysis, testing for the assumption that countries with relatively high initial energy intensity levels tend to catch-up to more advanced countries.We do so by regressing for each sector i the growth rate of energy productivity I* between period 0 and t, on its initial level, generating an estimate of β, according to: DI jt * = β ln (I j * 0 ) + l j + e j t (A.4) with l j representing unspecified country-specific (fixed) effects.These fixed effects are unobservable individual 'country-effects' capturing country-specific tangible and intangible factors that may impact energy intensity levels-implicitly assuming that energy intensity levels converge to multiple steady-states, which are conditional on country-specific characteristics.A statistically significant negative estimate of β is taken as evidence of convergence.Following Islam (1995) we calculate DI * in equation (A.4) as an average growth rate using five-year time intervals, in order to reduce the influence of business-cycle fluctuations and serial correlation on the error term.Subsequently, we use the estimated values of β, to derive the speed k at which the energy-intensity levels are converging to their steady-state level, according to k = -[(1/T)log(β + 1)] , with T denoting the length of the time interval under consideration (Barro and Sala-i-Martin, 1992;Islam, 1995).In addition, we calculate to what extent aggregate β-convergence (according to equation A.4) is to be explained from, respectively, convergence in the underlying sector structure and convergence of efficiency improvements within individual Manufacturing sectors.To this aim we use a modified version of the sectoral β-decomposition approach developed by Wong (2006).This sectoral βdecomposition approach is based on the identity:

E E Y I = =
In equation (A.5) β eff is the coefficient estimate obtained from regressing for each sector i the weighted growth rate of its energy productivity between period 0 and t on the initial level of aggregate (economy-wide) energy producticity according to: w DI * = β ln (I * ) + l + e (A.6) i i eff j,0 j j t and β str is the coefficient estimate obtained from regressing for each sector i the weighted growth rate of its value added share between period 0 and t on the initial level of aggregate (economywide) energy intensity according to: w DS = β ln (I * ) + l + e .(A.7) Copyright © 2015 by the IAEE.All rights reserved.
In equations (A.6) and (A.7) w i is the decomposition weighting function (see Appendix); again, j denotes the cross-country dimension, e jt the standard error and l j the unspecified country-specific (fixed) effects.Wong (2006) derives w i from the traditional General Parametric Divisia method (PDM).In related work Duro et al (2010) and Duro and Padilla (2011) develop a similar approach, based on a Theil inequality index.In contrast, our approach is based on the log mean Divisia index method (LMDI), like in Section 4.

Figure 1 :
Figure 1: Aggregate Manufacturing Energy Intensity insofar Driven by Within-Subsector Efficiency Improvements (left-hand side) or Changes in the Subsector Structure (right-hand side) measures the relation between initial energy productivity level and its subsequent growth rate, thus indicating β-convergence.k measures the speed at which the energy-intensity levels are converging to their steady-state level.Significance levels: * 1%, ** 5%, ***10%.

Figure 2 :
Figure 2: Manufacturing σ-Convergence Analysis, Measured as Standard Deviation of Log(energy intensity)

Figure 3 :
Figure 3: Geographical Concentration of Industries

Figure 4 :
Figure 4: σ-Convergence Analysis of Krugman Specialization Index, Measured as Standard Deviation of Log(Krugman index)

Figure 5 :
Figure 5: Index of Specialization, Measuring the Extent to which Countries Specialize in Sectors in which Their Energy Intensity Performance is Relatively Good (<0) or Bad (>0), as Compared to the OECD11 Average (0) i Y i with I i the energy intensity of sub-sector i and S i is the share of sub-sector i in total value added.In its additive form LDMI I calculates the efficiency effect (DI eff ) between period 0 and T as positive numbers a and b given by L(a, b) = (a -b)/ln(a/b).

Table 2 : Value Added and Energy Shares of Manufacturing and Its Subsectors
a Economy = Agriculture + Manufacturing + Services + Construction + Transport b Pulp & Paper, Chemicals, Non-Metallic Minerals, Basic Metals c Food, Textiles, Wood, Machinery, Transport Equipment, Non-Specified Industry intensive Manufacturing sector: between 1980 and 2005 the relative weight of the Manufacturing sector in the aggregate economy has dropped, both in terms of value added and energy.The decline in energy share has been relatively strong in the USA ( -33%), while the value added share has fallen relatively rapidly among Western European (EU8) economies ( -24%).

Table 7 : Decomposition of σ-convergence Trends
Degree of cross-country variation* % Change of cross-country variation * Measured as standard deviation of log(energy intensity)

Table 9 : Change in Degree of Specialization and Concentration, in Terms of Output and Energy Use
a Based on Krugman Specialization index, average across countries.b Based on Hirschmann-Herfindhal index, average across sectors.