What Drives Carbon Emissions in German Manufacturing: Scale, Technique or Composition?

Rottner, Elisa; von Graevenitz, Kathrine Article — Published Version What Drives Carbon Emissions in German Manufacturing: Scale, Technique or Composition? Environmental and Resource Economics Provided in Cooperation with: Springer Nature Suggested Citation: Rottner, Elisa; von Graevenitz, Kathrine (2024) : What Drives Carbon Emissions in German Manufacturing: Scale, Technique or Composition?, Environmental and Resource Economics, ISSN 1573-1502, Springer Netherlands, Dordrecht, Vol. 87, Iss. 9, pp. 2521-2542, https://doi.org/10.1007/s10640-024-00894-7 This Version is available at: https://hdl.handle.net/10419/315258 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by/4.0/ Vol.:(0123456789) Environmental and Resource Economics (2024) 87:2521–2542 https://doi.org/10.1007/s10640-024-00894-7 1 3 What Drives Carbon Emissions inGerman Manufacturing: Scale, Technique orComposition? ElisaRottner1· KathrinevonGraevenitz2 Accepted: 20 June 2024 / Published online: 16 July 2024 © The Author(s) 2024 Abstract Emissions of local pollutants from industry have declined across many developed countries over the last decades. For carbon emissions such reductions have yet to materialize. Using German administrative micro-data at the product level, we apply the workhorse model for decomposing emission changes into scale, composition (changes in the mix of goods produced) and technology (emission factors of production) effects. We find that the production composition in German manufacturing shifted towards less CO2 intensive goods, while emission intensities of production increased. We show that data aggregation matters. Both effects are substantially underestimated when using data at the sector level as compared to the product level. A complementary plant level decomposition reveals that emission intensities of production increased both within plant, and due to a reallocation of production to more emission intensive plants. Keywords Carbon emissions· Climate policy· Statistical decomposition· Manufacturing We are grateful to two excellent referees who helped us improve the paper substantially. We thank Sylwia Bialek, Claire Brunel, Andreas Gerster, Arti Grover, Beat Hintermann, Arik Levinson, Dominik Schober, Bodo Sturm and Ulrich Wagner for suggestions and insightful comments. We also thank seminar participants at the ZEW–Leibniz-Centre for European Economic Research, the 9th Mannheim Conference on Energy and the Environment, the European Association of Environmental and Resource Economists, the annual meeting of the Association of Germanspeaking Economists (VfS), the AURÖ junior workshop and the World Bank brown bag series Greening Private and Financial Sector. We gratefully acknowledge the Research Data Centre (FDZ) of the Federal Statistical Office and the Statistical Offices of the German Länder for granting us access to the AFiD data and for the use of their research facilities, in particular Denise Henker, Stefan Seitz, Kerstin Stockmayer and Diane Zabel for their advice and technical support. We thank the Federal Ministry of Education and Research (BMBF) for the financial support through the Project TRACE (Grant number 01LA1815A). The views expressed in this paper are those of the authors and do not necessarily represent those of the institutions mentioned above. * Kathrine von Graevenitz [email protected] Elisa Rottner [email protected] 1 ZEW - Leibniz Centre forEuropean Economic Research andUniversity ofBasel, P.O. Box10 34 43, 68034Mannheim, Germany 2 ZEW - Leibniz Centre forEuropean Economic Research andUniversity ofMannheim, P.O. Box10 34 43, 68034Mannheim, Germany 2522 E.Rottner, K.von Graevenitz 1 3 JEL Classification D22· L60· Q41· Q48 1 Introduction Environmental policy to improve air quality has been remarkably effective. Local pollutant emissions have been regulated in many countries since the 1970s and 1980s. In contrast, we have yet to make substantial progress on the reduction of CO2 emissions. Regulation of CO2 emissions is much more recent, and not as widespread even among developed economies. For an individual country, the direct benefits of reducing carbon emissions are less clear than the benefits of reducing local pollutants, since damages depend on emissions globally and vary across countries. Perhaps in consequence of less stringent regulation, end-of-pipe technologies for CO2 removal are not yet commercially viable, and emissions remain closely linked to activity levels increasing the perceived costs of regulation. The industrial sector accounts for the lion’s share of global greenhouse gas emissions, together with the power sector (IPCC 2014). In this sense, carbon emissions are similar to the case of local pollutants. Electricity generation has become cleaner over time through expansion of renewable electricity generation from wind and solar installations. For industry, a reduction in greenhouse gas emissions, and, most prominently, CO2 has not materialized, even in countries where climate policies are implemented and are comparatively strict from a global point of view: In Germany, carbon emissions from manufacturing have increased in recent years and were about 32 million tonnes higher in 2017 than in 2003 (von Graevenitz and Rottner 2023). The lack of progress in absolute terms, even in such an industrialized economy with climate policies in place, is problematic—especially if countries like Germany are to serve as a blueprint for decarbonization without deindustrialization in other countries. What drives the development of CO2 emissions in the German manufacturing sector? Answering this question is crucial to determine appropriate countermeasures. Are emission increases solely grounded in a rising production scale? Do we have to worry about the outsourcing of dirty products and carbon leakage (production composition)? Are there technological improvements that might be exported to other countries and facilitate emission reductions there as well (production technique)? Decomposition methods are a useful tool to disentangle these emission drivers and have been widely used for local pollutants in the environmental community (Brunel 2017; Levinson 2009, 2015; Najjar and Cherniwchan 2021; Shapiro and Walker 2018). Due to limited data availability, past research has conducted Levinson-style statistical decompositions at the sectoral level. Using fine-grained data at the product level distinguishing between more than 4600 9-digit products, we decompose the development of carbon emissions in German manufacturing. We show that data aggregation matters: Generally, we find that the German manufacturing sector has been shifting towards a cleaner product composition between 2005 and 2017. The clean up due to compositional change is 9.4% in our 9-digit baseline decomposition, but only 6.8% in the 3-digit decomposition (71%). Conversely, emission intensities of German manufacturing (production technique) have increased between 2005 and 2017. At the 9-digit level, this increase amounts to 4.5%, but only 2.3% at the 3-digit level (51%). The increase in production scale and the positive technique effect combined are the reasons why carbon emissions in German industry have not decreased. 2523 What Drives Carbon Emissions inGerman Manufacturing: Scale,… 1 3 The detailed micro-data available from the German Manufacturing Census allow us to combine the Levinson-style decomposition with a plant level decomposition in the style of Foster etal. (2008). In doing so we follow Barrows and Ollivier (2018). Conducting decomposition analyses using different units of observation (product versus plant), we can exploit the respective merits of each method and obtain a fuller picture of the channels through which emission intensities of production have increased. Our results show that selection (plant entry and exit) had negligible effects on the emission intensity of German industry. Emission intensities of production increased within plant. Re-allocation of production between plants has also led to an increase in emission intensity. The cross effect is large and negative indicating that emission intensity of plants with initially high intensities has grown relatively less, dampening the net impact on emission intensity of the manufacturing sector. With our paper, we contribute to the literature using decomposition tools as a means to understand the drivers of emission developments (e.g. Shapiro and Walker 2018; Levinson 2009, 2015, 2021; Brunel 2017; Najjar and Cherniwchan 2021). First, we contribute by characterizing the aggregation bias. Prior work by Lyubich etal. (2018) show that emission intensities in the US vary substantially within 375 narrowly defined NAICS industries. The German manufacturing sector displays similar heterogeneity (von Graevenitz and Rottner 2023). Part of this heterogeneity is due to different emission intensity of products within industries, i.e., the within industry composition. In decompositions of emissions conducted at the industry level, technique effects are calculated as within sector changes in emission intensity. Changing composition of products within an industry gives rise to aggregation bias in the technique and composition effect. The direction of bias depends on the sign of the “true” technique and composition effects. With negative composition effects and positive technique effects as in our case, both effects are underestimated. When both effects are negative, the technique effect would tend to be overestimated in absolute terms: Emission reductions attributed to the technique effect in reality stem from within sector compositional shifts. Whenever the true composition effect is negative, the bias works towards predicting a technique effect that is too clean. Either, it is positive but too small, or negative and too large. These findings suggest that the clean-up attributed to the technique effect in prior analyses at a less granular level is likely overstated. Second, we contribute to the decomposition literature by providing evidence on carbon emissions. Several papers examine the drivers of local pollutant emissions (e.g. Shapiro and Walker 2018; Levinson 2009, 2015; Brunel 2017; Najjar and Cherniwchan 2021). These papers generally document large and negative technique effects suggesting that technology adoption has been a main driver of local pollutant emission reductions. A much smaller literature decomposes carbon emissions (Barrows and Ollivier 2018; Levinson 2021; Brunel and Levinson 2022), but these studies have focused on the US and India where climate policy has been comparatively lax. We provide evidence from a developed and export-oriented economy in which carbon emissions have been regulated since the early 2000s. Lastly, we contribute to the literature evaluating the effect of environmental policies on the manufacturing sector. A growing literature studies the relationship between climate policies and energy demand as well as firm-performance in Germany (e.g. Flues and Lutz 2015, Lutz, 2016; Gerster and Lamp, 2020; Hintermann etal., 2020; Lehr etal., 2020; von Graevenitz and Rottner, 2022). This literature has focused on identifying causal effects of specific policies such as the European emissions trading scheme or the exemption from the German renewable energy surcharge. Exploiting quasi-natural experiments, this body of work mostly quantifies effects on emissions of large and energy intensive manufacturing 2524 E.Rottner, K.von Graevenitz 1 3 plants. Our analysis provides context for these individual policy evaluations. We are the first to apply the approaches by Levinson (2015) and Foster etal. (2008) to German emissions.1 2 Data We conduct our analysis using the official plant level micro-data from the federal statistical offices of the Bund and the Länder. All manufacturing plants in Germany with more than 20 employees must report in the surveys. We use data from 2005 to 2017. Our analysis requires information on each plant’s emissions and total output, as well as aggregate emissions and output. Moreover, we need information on the output share and emission intensity of each subsector and plant. The German Manufacturing Census does not contain any information on carbon emissions. We calculate plant level emissions by combining information on manufacturing plants’ consumption of 14 different fuels and electricity with appropriate emission factors retrieved from the German Environmental Agency (Umweltbundesamt 2008, 2020a, b). Emission factors are national but time-varying. The emission factor for electricity reflects the German electricity mix as well as transmission losses.2 Aggregate emissions are calculated by summing up plant level emissions. In the analysis, we abstract from analysing process emissions due to data limitations.3 Figure1 shows the development of aggregate carbon emissions in the German manufacturing sector between 2003 and 2017. In 2017, carbon emissions were roughly 32 mio. tonnes higher than in 2003. The German Manufacturing Census contains information about products at the 9-digit level. The 9-digit level is very fine-grained: In our base year 2005, we distinguish between 4,672 different 9-digit products. For comparison, there are only 1,356 products at the 6-digit level; 221 different sectors at the 4-digit level, and 91 sectors at the 3-digit level. Figure2 shows an example of the breakdown of a 2-digit sector to the 9-digit product level for illustrative purposes. Aggregate output is calculated by summing gross output of each product produced by individual manufacturing plants.4 We deflate gross output using producer price indices (base year 2015) from the Federal Statistical Office (DeStatis 2018).5 1 Petrick (2013); Kube and Petrick (2019) analyse CO2 emissions at the firm level using the Logarithmic Mean Divisia Index. They do not take the effects of product mix into account. 2 There are slight regional differences in emissions performance across German power plants. For example, coal fired power plants from the Rhineland emitted 113 t CO2/TJ in 2016 whereas a coal fired power plant in middle Germany emitted 104 t CO2/TJ in the same year due to differences in heat rates (Umweltbundesamt 2019a). Germany constitutes one electricity market and in general there is no way to determine which region has delivered electricity to individual plants. 3 According to the EEA (2015, 2016, 2017) process emissions made up about 10-16% of overall German emissions regulated under the EU Emissions Trading Scheme in the years 2013 to 2015. Within manufacturing, the sectors in which process emissions occur tend to be the same sectors that also have larger emissions from fuel combustion, i.e. pulp and paper, coke and petroleum, chemicals, other non-metallic mineral production and metal production. 4 Note that by using gross output as a measure for manufacturing activity, we cannot rule out the possibility that results are driven by the manufacturing sector outsourcing/starting to produce intermediate inputs that were produced/imported beforehand. Information on value added is only available at the firm level for a stratified sample of firms. 5 Where available, product level gross output is deflated using price indices at the 9-digit product level. In total, roughly 80% of gross output are deflated at the 9-digit level, 13% at the 6-digit level and the remaining 7% at the 4-digit level. 2525 What Drives Carbon Emissions inGerman Manufacturing: Scale,… 1 3 Energy use and by extension CO2 emissions are available at the plant level. We calculate emission intensities at the product level by allocating plant emissions to the different products based on output shares. The procedure is described in more detail in Additional file1. Note that by using output shares to allocate plant level emissions onto the products, we implicitly assume that all products within a plant are produced with the same emission Fig. 1 The development of carbon emissions from fuel combustion in the German manufacturing sector. Notes: Source: DOI 10.21242/43531.2017.00.03.1.1.0. Own calculations Fig. 2 Industries, sectors and products. Notes: Source: Goods catalogue of production statistics (DeStatis 2011) 2526 E.Rottner, K.von Graevenitz 1 3 intensity.6 We conduct the Levinson-style decomposition analysis at the 3-, 4-, 6and 9-digit sector level to quantify aggregation bias. 3 Statistical decomposition methods Decomposition tools are frequently used to disentangle the sources of emission changes. A wide variety of different approaches exist. We focus here on the decomposition into scale, composition and technique effects as in Levinson (2015), which is the "workhorse approach" in environmental economics. With our exceptionally granular data, we are able to analyse composition effects in more detail than past research and to assess the extent of aggregation bias in quantifying composition and technique effects using more aggregate data. We complement this product level decomposition of aggregate emissions with a plant level decomposition of emission intensity in the manufacturing sector inspired by Foster etal. (2008). The plant level decomposition does not allow us to analyse the role of product composition on emissions development in detail, but speaks to the effect of production allocation across plants and selection (entry and exit) among manufacturing plants in driving emission intensities.7 The combination of both decomposition approaches allows us to obtain a comprehensive picture of the drivers of emissions. 3.1 Product level decomposition ofaggregate emissions The statistical decomposition of aggregate emissions following Levinson (2015) can be carried out at different levels of sectoral disaggregation. It is grounded in a representation of total emissions, Pt , of, in this case, CO2 in the German manufacturing sector at time t as the sum of emissions from s different subsectors in manufacturing. In each subsector, emissions are determined by the product of output produced, vst , and the emission intensity of that subsector, zst . Hence, total emissions from manufacturing depend on aggregate output Vt of the manufacturing sector as a whole, the share of each subsector from aggregate output 𝜃st and the emission factors of production in each subsector. In vector notation, this is: where 𝛉𝐭 and 𝐳𝐭 are s×1 vectors containing the market shares and emission intensities of each of the s different industries. Total differentiation and division by emissions yields an expression for emission changes (with time subscripts dropped for notational convenience): (1) P t= ∑ s pst = ∑ s vstzst =Vt ∑ s 𝜃stzst (2) Pt =V t 𝛉 � 𝐭 𝐳 𝐭 (3) dP P =dV V +d𝜃 𝜃 + dz z 6 This most likely leads to some measurement error, as Barrows and Ollivier (2018) found substantial variation in emission intensities within multi-product firms across product lines. To assess robustness, we also conduct the product level analysis based on single-product plants. 7 We thank an anonymous referee for this suggestion. 2527 What Drives Carbon Emissions inGerman Manufacturing: Scale,… 1 3 The first term of the equation is the scale effect. The scale effect is given by the change in aggregate output and summarises how emissions would have evolved if only the production volume had changed. The second term, the composition effect, describes how emissions would have evolved if the sectoral composition of manufacturing had followed its historical path while keeping scale and emission intensities fixed. The third term is the technique effect and explains how emissions would have evolved if emission intensities had followed their historical paths while production scale and composition were fixed.8 The decomposition is straightforward on a conceptual level. The accounting identity in Eq.3) is used to directly calculate the scale effect (keeping composition and technique constant) and either the technique or the composition effect (keeping scale and composition or scale and technique constant). Scale, composition and technique effect add up to the actually observed emission changes as demonstrated in Eq. (3). Based on this identity, the remaining component is determined as the residual once scale and composition or scale and technique have been subtracted from the actual observed emissions. If, given scale and composition, emissions would have been higher than they actually were, the technique effect is negative. Conversely, if, given scale and composition, emissions would have been lower than they actually were, the technique effect is positive. Note that estimating scale, technique and composition in this way attributes all interactions that might arise between the three effects to the effect estimated as the residual. Different estimates of the technique effect depending on whether it is estimated directly or indirectly as a residual are not an issue of robustness of the decomposition. Rather such differences arise when interaction effects are large. Levinson (2009, 2015) lists several potential types of interactions, e.g., larger industries having increasing returns to scale to pollution abatement or shrinking industries closing down the dirtiest plants first. It is not obvious which channel these interactions should be attributed to. In many studies, the choice of whether to calculate technique or composition directly is motivated by data availability. Our data allow us to calculate both composition and technique effect directly and as a residual, thereby allowing us to assess the importance of interaction effects. For the sake of interpretation, the technique effect can also be depicted in a Laspeyreslike index. In this case, the predicted emissions with actual emission intensity developments but base year production scale and composition are divided by base year emissions. This index is equal to one when emission intensities remain unchanged as compared to the base year. Falling or rising emission intensities in contrast lead to the technique index taking on values smaller or larger than one. The index for the technique effect is given by the following equation:9 8 Our decomposition is based on the workhorse approach which relies on revenues. Recent work by Rodrigue etal. (2022) on SO2 emissions in China has shown that this decomposition may be biased when markups change over the period under study. De Loecker and Eeckhout (2021) estimate markups for several countries including Germany to examine their development over time. In Germany markups remained approximately constant throughout our study period. While De Loecker and Eeckhout (2021) do not discuss markup heterogeneity within Germany, their finding of constant markups suggests that the bias is limited. 9 Similarly, the composition effect in index form can be written as: C L= ∑ s𝜃st zs0 ∑ s 𝜃 s 0z s 0 2528 E.Rottner, K.von Graevenitz 1 3 where 0 indicates the base year to which emission intensity changes are compared.10 Reformulating the estimated technique effect in this way clarifies some of the properties of the calculated effect. Specifically, the aggregate index consists in changes in individual subsectors s that are weighted by their relative importance in the base year 0. For our application to emission intensities, this means that the estimated technique effect is a weighted average of emission intensity changes in individual subsectors s where the weights are given by the subsectors’ shares of total manufacturing emissions in the base year 𝜇s0 .11 In each subsector, our measure of the technique effect does not constitute a technique improvement within plant. The change in the emission intensity also depends on allocation of production across plants. Equation (7) shows this: The first term in the last expression is the share of plant j in aggregate production of subsector s. The second term is the plant’s emission intensity. Clearly, the emission intensity of different plants is weighted by their output share, which means that emission intensity improvements (i.e., the technique effect in the analysis) can be achieved both by emission intensity improvements within plants, and by a reallocation of production toward less emission intensive plants. To clearly disentangle the role of within plant changes versus across plant reallocation, we complement the sector level decomposition with a plant level decomposition of the technique effect. 3.2 The aggregation bias instatistical decompositions The level of sectoral disaggregation has direct implications for the calculation of the composition and technique effect. Suppose, e.g., that no sector data are available at all, but only data on the aggregate manufacturing sector. In that scenario, it would not be possible to identify any composition effect. All changes in emissions would be attributed to either scale or technique effect. Specifically, changes in emission intensities caused by production composition changes would be attributed to the technique effect. More detailed sector level data makes it possible to separate a composition effect from the technique effect. If (4) T L= ∑ s z st v s0 ∑ z s0 v s0 =� s zst z s 0 ∗ zs0vs0 ∑s z s0 v s0 ≡� s zst z s 0 ∗𝜇s 0 (5) zst = pst vst = ∑ jpjst vst = ∑ jzjst ∗vjst vst =� j vsjt vst ∗z jst 10 In our application, 2005 constitutes the base year. 11 Alternatively, Levinson (2015) proposes to use a Paasche-like measure where the weights are given by current shares: Differences between Laspeyres and Paasche indices capture one interaction attributed to the term estimated as a residual, namely whether or not the manufacturing sector shifts towards or away from sectors in which pollution intensities decline the most. The comparison does not capture all possible interactions between scale, composition and technique effect. We report comparisons between Laspeyres and Paasche technique effects in Additional file1. Generally, we find that sectors for which the difference between Laspeyres and Paasche indices is big also display a big difference between estimating the composition or the technique effect as a residual. This suggests that a large share of interaction effects are between composition and technique rather than due to scale. T P= ∑ s z st v st ∑ s z s 0v st 2535 What Drives Carbon Emissions inGerman Manufacturing: Scale,… 1 3 switches. The importance of product switches for the observed emission intensity improvements is relatively small. This is grounded in the fact that only a relatively small share of single product plants switched their products between 2005 and 2017 (overall roughly 6%, from year to year less than 1%). Still, on average, plants seem to have switched toward less emission intensive products (i.e., they became less emission intensive after a product switch). This switching behaviour might be one driving force for the clean composition effect identified in the product level decomposition. 4.4 Carbon leakage andachange intheproduct composition The clear trend towards a less emission intensive production composition that we find at the 9-digit product level could indicate carbon leakage. If especially energy intensive sectors outsource the production of the most emission intensive products, this would lead to a less carbon intensive domestic production composition. Suggestive evidence for the occurrence of carbon leakage and for the loss of competitiveness of German energy intensive sectors could be found if we observed a simultaneous shift in German exports toward less emission intensive sectors and in German imports toward more emission intensive sectors. We decompose German trade flows following Eq. (3), akin to the approach taken in Levinson (2009). The decomposition is conducted at the 3-digit sector level using imports and Table 2 Allocation and selection Source: Own calculations. The table shows the results of the plant level decomposition of emission intensities calculated at the 4-digit sector level. Sectors are aggregated using output weights. The first row shows results in long differences between 2005 and 2017. The three remaining rows report averages of year-on-year changes for distinct periods Time period Total change Within plant Between plants Cross effect Entry Exit 2005–2017 4.2 19.5 10.5 -25.4 -0.7 -0.3 2005–2009 1.3 3.6 5.4 -7.6 0.4 0.4 2010–2013 -0.0 2.9 1.0 -3.9 -0.3 -0.1 2014–2017 -0.3 0.5 2.0 -3.2 0.8 0.4 Table 3 Product entry effects within single product plants Source: Own calculations. The table shows the results of a product level decomposition of emission intensities calculated at the plant level. Plants are aggregated using output weights. The first row shows results in long differences between 2005 and 2017. The second row reports averages of year-onyear changes between 2005 and 2017 Plants Long differences Total change Product switch Technique Single product Yes −11.99 −1.48 −10.52 Single product No −8.47 −0.67 −7.80 2536 E.Rottner, K.von Graevenitz 1 3 exports from Eurostat (2023). Base year, i.e., 2005 emission intensities at the 3-digit level are computed from the German Manufacturing Census.20 Table 4 shows Laspeyres and Paasche indices of the composition effect for German manufacturing imports and exports, respectively. Clearly, the composition shift toward less emission intensive sectors is not limited to overall German production, but also extends to both German imports and German exports. Imports have even shifted faster toward cleaner sectors than domestic production. This suggests that carbon leakage and stringent German climate policies may not be at the roots of the German production shift toward less emission intensive goods. Table 4 Laspeyres and Paasche indices for the composition effect of German trade flows Source: Own calculations. The table shows the Laspeyres and Paasche indices for the composition effect calculated for the decomposition of German exports (left) and German imports (right). Comparing the two shows, that the Paasche index is consistently smaller than the Laspeyres index indicating that trade flows in industries with faster falling/more slowly growing carbon intensities grew at a faster rate Year Laspeyres Paasche Laspeyres Paasche Exports Exports Imports Imports 2005 1 1 1 1 2006 1.002387 1.001079 1.008533 1.003687 2007 0.9569227 0.9535437 1.009588 1.003167 2008 0.950685 0.9394742 0.9987149 0.9887199 2009 0.9790129 0.9720478 0.8985757 0.8895001 2010 0.9624456 0.9484019 0.9516796 0.9256257 2011 0.9259918 0.9122524 0.9341809 0.910885 2012 0.9189055 0.9039791 0.9045228 0.8831315 2013 0.9114583 0.8901056 0.901652 0.8795794 2014 0.907735 0.8805468 0.9000167 0.8723223 2015 0.8880442 0.8578597 0.8838269 0.8516514 2016 0.8824099 0.8520127 0.8584881 0.819602 2017 0.8790151 0.8361229 0.8638813 0.8129885 20 In the decomposition of exports, we implicitly assume that exports are produced with the same emission intensity as overall production. While this is not necessarily true (see, e.g., Forslid etal. 2018 for evidence that exporters tend to be less emission intensive), note that the decomposition exploits variation relative to the base year. There will only be bias if the “export premium” in terms of emission intensity differs systematically across 3-digit sectors. In the decomposition of imports, for the lack of international emissions data at the 3-digit sector level, we also rely on German emission intensities, not the emission intensities of production in the origin country. We trust that Germany serves as a reasonable approximation to the relative emission intensity of different sectors, even if emission intensities from different origin countries might differ. Alternatively, as in Levinson (2009), the exercise can be viewed as analysing the emissions displaced, i.e., the emissions that would have occurred in Germany absent imports. 2537 What Drives Carbon Emissions inGerman Manufacturing: Scale,… 1 3 4.5 Sectoral heterogeneity inthedevelopment ofemissions andemission intensities We aggregate sector and product specific developments using weights for their emission share in the base year 2005 (as shown in Eq. (5)). This means that the product and sector specific decomposition results are heavily driven by developments in the very energy and emission intensive sectors metal production, chemicals, coke and petroleum, other nonmetallic mineral products and pulp and paper.21,22 In this section, we show disaggregated results by 2-digit sector. This reveals substantial heterogeneity in emission drivers across sectors. It also shows that aggregation and selection bias matter to a different extent in different sectors. Figure4 contrasts the Laspeyres indices for the composition and the technique effect at the 2-digit sector level for the year 2017 relative to 2005. Both sign and magnitude of technique and composition effects vary substantially.23 Table5 provides a list of brief sector descriptions with the sector codes. Despite the positive technique effect in 2017 for the manufacturing sector overall, in the same year, negative technique effects prevail when it comes to the 2-digit sectors (i.e., most sectors lie below the horizontal line in the figure). Several sectors experienced continuous improvements in terms of their emission intensity (among others, NACE 15: manufacture of leather and related products, NACE 18: printing and reproduction of recorded media, and NACE 26: manufacture of computer, electronic and optical products). The more energy intensive sectors (highlighted by the black boxes in the figure), however, are the ones that really matter for manufacturing’s emissions, and they tend to display positive, i.e., emission increasing, technique effects.24 Why have especially those sectors that matter in terms of manufacturing’s emissions not managed to improve their emission intensity of production? A potential reason for energy intensive sectors generally experiencing more of an increase in emission intensities could lie in those sectors benefiting to a higher degree from exemption and compensation schemes that were introduced alongside climate policies to prevent a loss in competitiveness and leakage effects. In the observation period, large and intensive electricity users could, e.g., get exempt from paying the Renewable Energy Surcharge which accounted for roughly a third of electricity prices. Manufacturing plants above a certain electricity procurement threshold could also benefit from paying reduced electricity network tariffs, and since 2013 a subset receive electricity cost compensation due to higher electricity prices owing to the EU ETS. Also, through the leakage list of the EU ETS, energy intensive and trade exposed sectors are eligible for a higher share of free allowances or electricity price 21 The development of energy consumption in German manufacturing, split across 2-digit sectors, is shown in Figure7 in the Additional file1. 22 A similar argument applies to the plant level decomposition which is aggregated based on output shares. Heterogeneity in terms of output shares is of a similar magnitude, but only chemicals belong to the top five sectors in both categories. The other four top five sectors in terms of output shares are food, machinery, motor vehicles, and metal products. The food sector holds a 6th place in terms of energy intensity. 23 The same can be seen from Tables9 and 10 containing the Laspeyres indices for technique and composition effects in the Additional file1. 24 The outliers in Figure4 (e.g. NACE 33 (Repairs) and NACE 26 (Computers and electronics) and NACE (Tobacco)) are small in terms of emissions. They do not contribute substantially to the main results. The development of emission intensities in computing is partially driven by technological development and resulting price decreases affecting deflated output. Repairs are generally an additional product with limited importance for manufacturing plants producing machinery. In consequence activity displays substantial variation over time. 2538 E.Rottner, K.von Graevenitz 1 3 compensation in the study period.25 Aside from directly facing different prices, the energy intensive sectors tend to use a different energy mix. Specifically, they rely a lot more on onsite electricity generation, and as such have been less exposed to the regulation-induced electricity price increases over our observation period. Consequently, they might have been subject to a lower shadow price on emissions and therefore have faced lower incentives to decrease their emission intensity. Composition effects also vary across sectors with sectors like NACE 13: textiles, or NACE 20: chemicals shifting towards less carbon intensive goods. Sectors like NACE 16: manufacture of wood and wood products, and NACE 26: manufacture of computer, electronic and optical products shift toward more emission intensive goods. In some sectors, the production composition remains virtually unchanged over time with regards to the emission intensity (e.g., NACE 22: manufacture of rubber and plastic products). Aggregation bias is also present in the sector-specific decompositions. Moving from the 3-digit decomposition to the 9-digit decomposition, it is clear that dispersion increases in the scatter plot. This is consistent with underestimation due to aggregation Table 5 Overview over two-digit sectors in manufacturing Sector Code Sector name 10 Manufacture of food products 11 Manufacture of beverages 12 Manufacture of tobacco products 13 Manufacture of textiles 14 Manufacture of wearing apparel 15 Manufacture of leather and related products 16 Manufacture of wood and of products of wood and cork, except furniture 17 Manufacture of pulp, paper and paper products 18 Printing and reproduction of recorded media 19 Manufacture of coke and refined petroleum products 20 Manufacture of chemical products 21 Manufacture of basic pharmaceutical products and pharmaceutical preparations 22 Manufacture of rubber and plastic products 23 Manufacture of other non-metallic mineral products 24 Metal production and processing 25 Manufacture of fabricated metal products 26 Manufacture of computer, electronic and optical products 27 Manufacture of electrical equipment 28 Machine manufacturing 29 Manufacture of motor vehicles 30 Other transport equipment 31 Manufacture of furniture 32 Other manufacturing 33 Repair and installation of machinery and equipment 25 An overview of existing exemptions in Germany for energy and electricity taxes and levies with the relevant thresholds can be found in a publication by the Umweltbundesamt (2019b). 2539 What Drives Carbon Emissions inGerman Manufacturing: Scale,… 1 3 bias. We also see that the pattern is not entirely consistent. Some sectors shift around considerably depending at the aggregation level. For example, sector 14: Clothing goes from having a moderate positive composition effect and a small but negative technique effect in the 3-digit decomposition to having a small positive composition effect and a large positive technique effect in the 9-digit decomposition. There are two reasons for these large shifts: First, composition might have changed substantially within the clothing industry, with a shift toward less emission intensive products within 3-digit industries. Second, there might be many product entries and exits such that the decompositions for the various aggregations differ in terms of their plant-coverage. Despite these shifts, a general pattern towards more dispersion at higher levels of disaggregation is visible. 5 Conclusion We have yet to make substantial progress on the reduction of CO2 emissions. Despite the introduction of several climate policies in Germany in the period between 2005 and 2017 carbon emissions continued to increase. To understand drivers of this emissions Fig. 4 Laspeyres indices for composition and technique effect in 2017 across sectors. Source: DOI 10.212 42/43531.2017.00.03.1.1.0. Own calculations. Notes: The figures display scatter plots of Laspeyres indices for technique and composition aggregated at the 2-digit sector level. The six most energy intensive sectors are marked by black boxes, whereas the remaining sectors are marked with circles. Values above one indicate that the sector uses more emissions intensive techniques in 2017 as compared to 2005 (Y-axis) and/or produces more emissions intensive products in 2017 as compared to 2005 (X-axis) 2540 E.Rottner, K.von Graevenitz 1 3 development, we apply a workhorse decomposition method from the environmental economics literature to extraordinarily fine-grained data on 9-digit products. We show that the increase in carbon emissions from the manufacturing sector can to a large extent be attributed to an increase in production scale. We observe a cleanup of German manufacturing in the order of 9% reduction of emissions compared to what they would have been had composition and technique remained unchanged since 2005. This clean-up is due to a change in product composition. From 2011 onwards, the German manufacturing sector has shifted towards greener products. In contrast, emission intensities of production have mostly increased as compared to 2005. We show that data aggregation matters in decomposition analyses. Both the composition and the technique effect are substantially underestimated if conducting the analysis at the more standard 3-digit sector level, rather than using data on 9-digit products: The clean-up resulting from changing product composition is smaller (71% of the effect calculated at the 9-digit level) and the increase in emission intensity of production technique is also attenuated (51% of the effect calculated at the 9-digit level), though the conclusions still hold. The changing composition of production raises concern whether these patterns are due to outsourcing of more carbon intensive products and could be indicative of carbon leakage. Our preliminary analyses decomposing imports and exports do not support this conclusion, but more research is required to rule out such concerns. Our measure of the technique effect depends on within plant emission intensity as well as production allocation across plants. We complement the product level decomposition with a plant level decomposition to better understand the roles of selection and allocation in driving the increase in emission intensities. This second decomposition shows that emission intensity within plants generally increased over the period. The between plants effect is also positive indicating that allocation of production shifted towards more emission intensive plants. A large and negative cross effect indicates, that plants that were initially highly emission intensive improved their emission performance more than less emission intensive plants. Effects of net entry on aggregate emission intensity are negligible. The net effect remains an increase in emission intensity in the manufacturing sector over the period studied. Supplementary Information The online version contains supplementary material available at https:// doi. org/ 10. 1007/ s1064002400894-7. Funding Open Access funding enabled and organized by Projekt DEAL. Declarations Competing interests The authors have no competing interests to declare that are relevant to the content of this article. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. 2541 What Drives Carbon Emissions inGerman Manufacturing: Scale,… 1 3 References Barrows G, Ollivier H (2018) Cleaner firms or cleaner products? How product mix shapes emission intensity from manufacturing. J Environ Econ Manag 88:134–158 Brunel C (2017) Pollution offshoring and emission reductions in EU and US manufacturing. Environ Resour Econ 68:621–641 Brunel C, Levinson A (2022) Globalization and greenhouse gas emissions: is the United States a pollution haven?, Edward Elgar Publishing De Loecker J, Eeckhout J (2021) Global Market Power. Working Paper DeStatis (2011) Güterverzeichnis für Produktionsstatistiken (GP 2009). Further information: https:// www. dest a tis. de/ DE/ Metho den/ Klass ifika tionen/ GueterWir ts chaft sklas sifik ation en/ klass ifik a tiongp19. html. DeStatis (2018) Erzeugerpreisindex gewerblicher Produkte. Further information: https:// www. desta tis. de/ DE/ Themen/ Wirts chaft/ Preise/ Erzeu gerpr eisin dexgewer blicheProdu kteinhalt. html Ederington J, Levinson A, Minier J (2004) Trade liberalization and pollution havens. B.E. J Econ Anal Policy 4(2) EEA (2015). Application of the EU Emissions Trading Directive - Analysis of national responses under Article 21 of the EU ETS Directive in 2014. EEA Technical Report No. 3 EEA (2016) Application of the EU Emissions Trading Directive—analysis of national responses under Article 21 of the EU ETS Directive in 2015. EEA Report No. 6 EEA (2017) Application of the EU Emissions Trading Directive—analysis of national responses under Article 21 of the EU ETS Directive in 2016. EEA Report No. 4 Eurostat (2023). EU trade 1988-2022 by CPA 2008 (DS-057009). Available at https:// ec. europa. eu/ euros tat/ datab rowser/ (accessed on 25 May 2023) Flues F, Lutz BJ (2015) The effect of electricity taxation on the German manufacturing sector: a regression discontinuity approach. ZEW discussion papers No. 15-013 Forslid R, Okubo T, Ulltveit-Moe KH (2018) Why are firms that export cleaner? International trade, abatement, and environmental emissions. J Environ Econ Manag 91:166–183 Foster L, Haltiwanger J, Syverson C (2008) Reallocation, firm turnover, and efficiency: selection on productivity or profitability? Am Econ Rev 98(1):394–425 Gerster A, Lamp S (2020) Energy tax exemptions and industrial production. Working Paper Hintermann B, Zarkovic M, diMaria C, Wagner U (2020) The effect of climate policy on productivity and cost pass-through in the German manufacturing sector. CRC TR224 Discussion Paper No. 249 IPCC (2014) Climate change 2014: mitigation of climate change. Contribution of working group III to the fifth assessment report of the intergovernmental panel on climate change Kube R, Petrick S (2019) Cleaner production or structural change? A breakdown of CO2 emissions reductions in the German manufacturing sector using firm-level data. Unpublished working paper Lehr J, Gerster A, Pieper S, Wagner U (2020) The impact of carbon trading on industry: evidence from German manufacturing firms. mimeo Levinson A (2009) Technology, international trade, and pollution from US manufacturing. Am Econ Rev 99(5):2177–2192 Levinson A (2015) A direct estimate of the technique effect: changes in the pollution intensity of US manufacturing, 1900–2008. J Assoc Environ Resour Econ 2(1):43–56 Levinson A (2021). Energy intensity: Deindustrialization, composition, prices, and policies in U.S. states. Resour Energy Econ 65 Lutz BJ (2016) Emissions trading and productivity: firm-level evidence from German manufacturing. ZEW Discussion Paper, 16-067 Lyubich E, Shapiro JS, Walker R (2018) Regulating mismeasured pollution: implications of firm heterogeneity for environmental policy. AEA Papers and Proceedings 108:136–142 Najjar N, Cherniwchan J (2021) Environmental regulations and the clean-up of manufacturing: plantlevel evidence. Rev Eon Stat 103(3):476–491 Petrick S (2013) Carbon efficiency, technology, and the role of innovation patterns: evidence from German plant-level microdata. Kiel Working Papers 1833 Rodrigue J, Sheng D, Tan Y (2022) The curious case of the missing Chinese emissions. J Assoc Environ Resour Econ 9(4):755–805 Shapiro JS, Walker R (2018) Why Is pollution from US manufacturing declining? The roles of environmental regulation, productivity, and trade. Am Econ Rev 108(12):3814–3854 Umweltbundesamt (2008) Bestimmung spezifischer Treibhausgas-Emissionsfaktoren für Fernwärme. Climate Change. 08/08 2542 E.Rottner, K.von Graevenitz 1 3 Umweltbundesamt (2019a) Aktualisierung der Emissionsfaktoren für Großfeuerungsanlagen. Texte, 140/2019, further information (only available in German): https:// www. umwel tbund esamt. de/ publi katio nen/ aktua lisie rungemiss ionsf aktor engross feuer ungsa nlagen Umweltbundesamt (2019b) Reform und Harmonisierung der unternehmensbezogenen Ausnahmeregelungen im Energiebereich. Texte, 23/2019, further information (only available in German): https:// www. umwel tbund esamt. de/ publi katio nen/ reformharmo nisie rungderunter nehme nsbez ogenen Umweltbundesamt (2020a) Carbon Dioxide Emissions for the German Atmospheric Emission Reporting 1900–2018 Umweltbundesamt (2020b) Entwicklung der spezifischen Kohlendioxid-Emissionen des deutschen Strommix in den Jahren 1990–2019. Climate Change, 13/2020, further information (only available in German): https:// www. umwel tbund esamt. de/ publi katio nen/ entwi cklungderspezi fisch enkohle ndiox id-2 von Graevenitz K, Rottner E (2022) Do manufacturing plants respond to exogenous changes in electricity prices? Evidence from administrative micro-data. ZEW Discussion Paper, 22-038 von Graevenitz K, Rottner E (2023) Energy Use Patterns in German Manufacturing from 2003 to 2017. Jahrbücher für Nationalökonomie und Statistik 243(3–4):319–354 von Graevenitz K, Rottner E, Richter P (2023) Is Germany becoming the European Pollution Haven? ZEW Discussion Paper, 23–069 Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.