Do voluntary environmental programs reduce emissions? EMAS in the German manufacturing sector

G Model

ARTICLE IN PRESS

ENEECO-104558; No. of Pages 12

Energy Economics xxx (xxxx) xxx

Contents lists available at ScienceDirect

Energy Economics journal homepage: www.elsevier.com/locate/eneeco

Do voluntary environmental programs reduce emissions? EMAS in the German manufacturing sector Roland Kube a,∗ , Kathrine von Graevenitz b , Andreas Löschel a,b,c,d , Philipp Massier b a

University of Münster, Münster, Germany ZEW – Leibniz Centre for European Economic Research, Mannheim, Germany Fraunhofer Center for Economics of Materials (CEM), Halle, Germany d Research Institute for Global Value Chains, University of International Business and Economics, Beijing, China b c

a r t i c l e

i n f o

Article history: Received 15 January 2019 Received in revised form 16 October 2019 Accepted 21 October 2019 Available online xxx Keywords: Voluntary environmental programs Firm-level energy behavior Matching difference-in-differences

a b s t r a c t Voluntary environmental management programs for firms are a popular instrument of environmental policy. However, the effectiveness of such programs is ambiguous based on the existing literature. We evaluate the Eco-Management and Audit Scheme (EMAS) introduced in 1995 by the European Union. EMAS is a premium certification of continuous pro-environmental efforts above regulatory minimum standards. It is more demanding than other voluntary programs as it requires publication of annual reports by firms on their environmental performance and targets. We use official firm-level production census data on the German manufacturing sector. Our research design combines the Coarsened Exact Matching approach with a Difference-in-Differences estimation. We find weak evidence of reductions in CO2 intensity by about 9 percent for firms that were certified in the early years of the program. For firms certified in later years we find no statistically significant evidence of emission reductions. Moreover, we find no statistically significant evidence of higher investments in environmental and climate protection or an increased use of renewable energy sources for certified firms. © 2019 Elsevier B.V. All rights reserved.

1. Introduction Meeting the national climate policy targets defined in the Paris Agreement requires a more efficient use of energy. Energy efficiency improvements are the single most important way to reduce global greenhouse gas emissions growth, as they counteract rising energy demand due to income and population growth (IEA, 2017). Increasing energy costs resulting from environmental regulation, such as fuel taxes, standards or emissions trading, incentivize firms to use energy more efficiently. However, empirical studies reveal that management quality and practices also play a crucial part in increasing industrial energy efficiency and decreasing CO2 emissions (Bloom et al., 2010; Martin et al., 2012; Boyd and Curtis, 2014; Löschel et al., 2017). In the literature, untapped potential for cost-efficient energy savings is often denoted as an “energyefficiency gap” (IEA, 2015; Jaffe and Stavins, 1994; Allcott and Greenstone, 2012). Common explanations for the existence of this phenomenon are imperfect information, transaction costs, and

∗ Corresponding author at: University of Münster, Am Stadtgraben 9, 48143 Münster, Germany. E-mail address: [email protected] (R. Kube).

uncertainty about future energy costs and regulation (Gerarden et al., 2017). Voluntary environmental management programs for firms are a popular instrument of environmental policy as an alternative to more stringent regulations. The basic idea is to provide a credible certification of firms’ pro-environmental efforts and overcompliance with regulatory minimum standards. The underlying assumption is that firms benefit from an environmentally friendly image, for instance, by tapping into consumer willingness to pay for environmentally friendly production. In addition, it is assumed that the process of establishing an environmental management system reduces information problems regarding the types and costs of resource-saving measures (Bloom et al., 2010). Thus, such management programs should lessen remaining inefficiencies in resource usage, and spur pro-environmental behavior at low enforcement costs (Barla, 2007). In this paper, we thus examine whether a specific voluntary program really does effectively reduce energy use and hence CO2 emissions by firms. The Eco-Management and Audit Scheme (EMAS) for plants is a prime example of a voluntary certification scheme. The program addresses the complete environmental footprint of a plant, i.e. local air, water, noise and land pollution as well as energy use and the resulting greenhouse gas emissions. Introduced by the European

https://doi.org/10.1016/j.eneco.2019.104558 0140-9883/© 2019 Elsevier B.V. All rights reserved.

Please cite this article as: R. Kube, K. von Graevenitz, A. Löschel et al., Do voluntary environmental programs reduce emissions? EMAS in the German manufacturing sector, Energy Economics, https://doi.org/10.1016/j.eneco.2019.104558

G Model ENEECO-104558; No. of Pages 12 2

ARTICLE IN PRESS R. Kube, K. von Graevenitz, A. Löschel et al. / Energy Economics xxx (xxxx) xxx

Union (EU) in 1995, EMAS is integrated into the legal system of the EU and its member states, which provides EMAS certification with grater credibility than a private sector initiative. To obtain certification, a plant needs to undergo an initial assessment of its environmental impact and develop an environmental policy and action plan for improvements. While other voluntary programs might present similar demands, they do not oblige plants to publish annual reports on their environmental performance, which is a requirement under the more stringent EMAS program. In these reports, plants have to define their own targets for improvements and explicitly state strategies and measures for achieving these targets. The program, however, does not prescribe the ambitiousness of these targets, which might lead to certification despite a mere business-as-usual effort. Nonetheless, the commitment and a systematic review of the plant’s resource usage may ultimately still translate into improvements. Despite a growing impact evaluation literature, the effectiveness of such voluntary programs in terms of environmental impact remain ambiguous. The wide heterogeneity in program design, pollutants addressed and distinct regional and institutional settings, make it difficult to generalize results. Well-studied programs include the U.S. EPA’s 33/50 (e.g. Arora and Cason, 1996; CarriónFlores et al., 2013; Gamper-Rabindran, 2006, Khanna and Damon, 1999, Vidovic and Khanna, 2007) and the private sector ISO14001 norm (e.g. Arimura et al., 2008; Barla, 2007; Nakamura et al., 2001). For the Strategic Goals program among firms in the U.S. metal industry, Brouhle et al. (2009) report little, if any, additional emissions reductions. Dasgupta et al. (2000) suggest that subsidies for environmental management in Mexico were beneficial for compliance, whereas Blackman et al. (2010) find no lasting impact of the Mexican Clean Industry Program on environmental performance. These studies mostly investigate the effect on toxic releases, as well as local air and water pollution, and find mixed results. The evidence on the effectiveness of programs with a particular focus on greenhouse gas emissions so far also suggests little to no emissions reductions. For the U.S. programs Climate Wise and 1605(b), Pizer et al. (2011) find no substantial effects on firm emissions and energy costs respectively. Kim and Lyon, 2011 report that 1605(b) program participants even increased their CO2 intensity while reporting reductions. A study on the Canadian Greenhouse Gas program VCR asserts no difference in reported emissions at the end of the program period (Brouhle and Ramirez Harrington, 2010). For Europe, the empirical evidence on the effectiveness of voluntary programs is rather scarce. This is surprising, given that the EU is considered a major industrial economy with relatively stringent environmental policy (Bracke et al., 2008). The EMAS program has been studied from different angles, yet, to the best of our knowledge, there is no study evaluating participation with data on environmental performance, such as energy usage or emissions. Bracke et al. (2008) investigate decisive financial characteristics of European firms for EMAS, based on Bureau-van-Dijk firm-level data. Several studies evaluate EMAS through firm surveys with selfreported evidence, e.g., regarding product and process innovations (Frondel et al., 2004; Rennings et al., 2006) or manager perceptions of program effectiveness (Iraldo et al., 2009), and generally find a positive influence on environmental innovations and performance. With this paper we add to the literature by using administrative data, including firms’ actual environmental performance, to consider the effectiveness of voluntary programs. While EMAS targets a variety of environmental indicators, we investigate CO2 and energy intensity as the main outcome variables, as well as the use of renewable energy and investments in environmental protection. Our analysis focuses on the German manufacturing sector,

which is directly responsible for at least 20 percent of national GHG emissions (Federal Environment Agency, 2016a). We link the EMAS Register with the official firm-level production census dataset AFiD (Amtliche Firmendaten für Deutschland – Official firm data for Germany) provided by the Research Data Centre of the Federal Statistical Office and Statistical Offices of the Länder. Our firm panel covers the years 1995–2014 and contains a wide set of economic indicators, energy consumption by energy source and environmental protection investment activity. This highly reliable dataset is confidential and has been utilized for previous studies on the effects of energy and climate policy (e.g. Löschel et al., 2019; Lutz, 2016). In order to assess the EMAS program, it is necessary to understand why a plant or firm may expect a positive net benefit from participation and what kind of firms choose to enter the program. Our summary statistics show that the covariate distribution across participation status differs substantially. EMAS firms are larger, more energy and emissions intensive and have larger revenues and exports than Non-EMAS firms. Especially in its early years, the program attracted firms with high policy relevance, i.e. relatively large energy users and CO2 emitters. The share of multinational corporations with multiple plants is larger among EMAS participants. Often, these firms have their own electricity generation capacities and a higher propensity to invest in environmental and climate protection. To account for observable differences in characteristics of EMAS and Non-EMAS firms, we pre-process the data via matching (Ho et al., 2007). Pre-processing the data in this way makes the analysis less sensitive to functional form assumptions, as we reduce heterogeneity in covariate space. We apply the Coarsened Exact Matching approach as proposed by Iacus et al. (2011, 2012), who show that its statistical properties are preferable to other matching methods, such as Propensity Score Matching (Rosenbaum and Rubin, 1983; Dehejia and Wahba, 2002) or Nearest-Neighbor Matching. We then estimate the Difference-in-Differences of outcomes. This approach has become well established in the program evaluation literature for addressing possible selection bias (e.g. Blackman et al., 2010; Fowlie et al., 2012, and Löschel et al., 2019). Overall, our point estimates suggest, that EMAS was effective in reducing CO2 and energy intensity at least in the earlier part of the period analyzed. However, these effects are imprecisely estimated. EMAS participants may also reduce other environmental impacts addressed by the program, such as local air, water and land pollution, which we do not observe in our dataset. We find no statistically significant evidence of increased investments in environmental and climate protection by EMAS firms, though the point estimate is positive. We further find no significant impact on the use of renewable energy sources. A conservative interpretation of our findings is that the effectiveness of EMAS – above and beyond other (unobservable) certification schemes such as ISO 14,001 – may be limited. Given the lack of precision in our estimates, further research is needed to substantiate the effects. The analysis continues with a description of the EMAS program and insights from existing research on voluntary measures. In Section 3, we introduce the data and explain our empirical approach. In Section 4, we present and discuss the main results. Section 5 provides robustness checks and Section 6 discusses the findings.

2. The EMAS program The Eco Management and Audit Scheme (EMAS) was developed by the European Commission (Council Regulation No 1836/93) and

Please cite this article as: R. Kube, K. von Graevenitz, A. Löschel et al., Do voluntary environmental programs reduce emissions? EMAS in the German manufacturing sector, Energy Economics, https://doi.org/10.1016/j.eneco.2019.104558

G Model ENEECO-104558; No. of Pages 12

ARTICLE IN PRESS R. Kube, K. von Graevenitz, A. Löschel et al. / Energy Economics xxx (xxxx) xxx

implemented in Germany in 1995.1 EMAS is a public voluntary regulation that is integrated into the legal system of the EU and its member states, in contrast to other private sector initiatives like the ISO 14,001 norm. To obtain certification under EMAS, plants need to undergo an initial inspection by an external auditor with a systematic review of the plant’s environmental impact regarding resource usage and emissions. On this basis, the plant has to set up an environmental management system, essentially a management tool that is based on the Plan–Do–Check–Act cycle with the following steps: Step 1 (Plan): defining an environmental policy with targets and projects, Step 2 (Do): operational implementation and appointment of an environmental officer in charge and involvement of employees, Step 3 (Check): management review of progress, and Step 4 (Act): taking corrective action where necessary. This subset of requirements is similar to the more popular ISO 14,001 certification2 and was harmonized under the EMAS II revision (EG Nr. 761/2001). Whereas ISO 14,001 is often considered a necessity in the market place, EMAS is generally perceived as the “premium” certification and a stronger signal of actual environmental responsibility (Neugebauer, 2012; Bracke et al., 2008; Arimura et al., 2011). This is due to the latter’s additional requirements: EMAS mandates annual reports on a firm’s current environmental performance, and the definition of own targets for improvements, as well as particular strategies to achieve these goals. Reports have to be accessible to the public. Compliance with self-set targets is mandatory for continuous certification. The plant enters the EMAS Register when the initial statement is positively evaluated for plausibility by external auditors. Each year, the environmental reports are to be validated and the plant is to be inspected every three years. In phone interviews with several auditors as of February 2018, we gained a better understanding of the certification routine and best practice examples. EMAS firms are encouraged to increase their efforts if auditors find that targets can be reached without making more than a business-as-usual effort. It was pointed out, however, that the impact of the program is the result of the commitment to allocate extra resources to exploring means of improvement and setting timelines, rather than of achieving the targets. Further, it was stated that EMAS firms do not immediately lose their certification when failing to achieve a target, as long as the firm can plausibly explain the reason to the auditor, e.g. due to an increase in production or unexpected higher costs. Potentially, such shortcomings can be compensated for by further efforts in other parts of the environmental strategy. However, the EMAS certification is invalidated if a violation of environmental laws is detected. However, almost all firms leaving the program did so owing to the costs of staying in the program, which are relatively larger for small and medium enterprises. With the latest EMAS III revision (EG Nr. 1221/2009), that came into effect in January 2010, the reporting of environmental performance was harmonized with key performance indicators, such as

1 Gesetz zur Ausführung der Verordnung (EG) Nr. 1221/2009 des Europäischen Parlaments und des Rates vom 25. November 2009 über die freiwillige Teilnahme von Organisationen an einem Gemeinschaftssystem für Umweltmanagement und Umweltbetriebsprüfung und zur Aufhebung der Verordnung (EG) Nr. 761/2001, sowie der Beschlüsse der Kommission 2001/681EG und 2006/193/EG. The EMAS certification system is administered by the Umweltgutachterausschuss, a subsidiary of the Federal Ministry of the Environment. EMAS auditors also need to be accredited by the DAU GmbH (Deutsche Akkreditierungs- und Zulassungsgesellschaft für Umweltgutachter mbH). 2 Because of the relatively larger effort and costs, the number of EMAS certified plants in Germany as of 2017 (about 1,213, EMAS Register, 2019) is smaller than ISO14001 certified plants (about 8,000, Federal Environment Agency, 2017). According to a survey by the Federal Environment Agency (2013), about half of EMAS participants are also ISO14001 certified.

3

energy use or greenhouse gas emissions, and the program was no longer restricted to EU member states. 2.1. Reasons for participation In order to assess the EMAS program, it is necessary to understand why firms expect a positive net benefit from participation, and what kind of firms are attracted to the voluntary program. In a survey by the Federal Environment Agency (2013), German EMAS firms that operate in the manufacturing sector state initial costs of D 20,000 at the median, ranging up to D 75,000 for larger firms. Estimated variable costs are less than D 10,000 on average and only exceed D 20,000 for 20 percent of respondents.3 In contrast, average estimated cost savings are more than D 10,000 per year, also increasing with firm size, and mostly attributed to improved resource efficiency, especially regarding energy usage. The literature suggests that firms mainly join EMAS due to internal firm motivation or due to regulatory incentives. Neugebauer (2012) reports that the main reasons for participation given in interviews with managers of German EMAS firms were reputation, precautions regarding regulatory inspections or stakeholder demands. In an OECD survey covering German EMAS firms that operate in the manufacturing sector, Frondel et al. (2004) report ensuring regulatory compliance as the main reason for adoption, partly owing to the principle of strict liability in Germany. Other important motives are firm reputation, incident prevention and cost-savings due to improved resource efficiency. Bracke et al. (2008) investigate the role of financial characteristics for EMAS participation among European firms and find that the program mainly attracts large firms with a sound financial structure and above-average labor costs. This is in line with Frondel et al. (2004) and the summary statistics of our sample (see Section 3).4 In a survey of 140 EMAS firms that operate in the manufacturing sector, Kube and Capellen (2017) questioned respondents on their main reasons for participating. The most frequent answers related to corporate identity and a credible certification of proenvironmental efforts, optimized documentation, expectations from business customers, regulatory compliance, workplace safety and government incentives. As in Frondel et al. (2004), the initiative for participation often came from internal stakeholders, mainly corporate management. EMAS participation is currently incentivized by regulation. In fact, German participants face less documentation effort and fewer inspections, e.g., regarding wastewater and recycling (WHG, BImSchG, KrWG). Since 2012, energy-intensive firms that apply for an exemption from the renewable energy surcharge on electricity prices (EEG) also have to provide EMAS, or an alternative certification of their energy management system. Similar rules were introduced regarding the reduced electricity tax rate in 2013 (EnergieStG), and EMAS participants are exempt from energy audits that have been mandatory for large firms since 2015. In the next Section, we present our data and discuss our empirical strategy for evaluating the impact of EMAS on manufacturing firm environmental performance.

3 Empirical investigations of stock market reactions to other voluntary environmental actions by firms do not show a clearly positive picture (Fisher-Vanden and Thorburn, 2011; Ziegler et al., 2011), which is why shareholder wealth maximization may not necessarily be a prime reason for EMAS participation. 4 Even in the U.S., Videras and Alberini (2000) find that large firms are generally more likely to join different EPA voluntary programs (Green Lights, WasteWi$e and 33/50), regardless of program requirements and pollutants covered, presumably due to receiving more attention from consumers and regulators.

Please cite this article as: R. Kube, K. von Graevenitz, A. Löschel et al., Do voluntary environmental programs reduce emissions? EMAS in the German manufacturing sector, Energy Economics, https://doi.org/10.1016/j.eneco.2019.104558

G Model ENEECO-104558; No. of Pages 12

ARTICLE IN PRESS R. Kube, K. von Graevenitz, A. Löschel et al. / Energy Economics xxx (xxxx) xxx

4

3. Data and methodology 3.1. Data Various datasets are combined for our analysis. First, from the official EMAS Register lists we collect all EMAS certified plants as of March 2016 in various manufacturing sectors (see Appendix A in Supplementary material). We obtain the respective firms’ Trade Register numbers from the Hoppenstedt Firm Information Database and VAT numbers from company websites in order to append the EMAS participation status to our main dataset: the German production census data AFiD (Amtliche Firmendaten für Deutschland – Official firm data for Germany) provided by the Research Data Centre of the Federal Statistical Office and Statistical Offices of the Länder. This highly reliable dataset is confidential and can thus only be accessed by approved researchers for scientific purposes. It has been utilized in previous studies on the effects of energy and climate policy (e.g. Löschel et al., 2019; Lutz, 2016; Petrick and Wagner, 2014). Our firm panel covers the years 1995–2014 and has the following modular structure: AFiD-Panel Industrial Units (DOI: The 10.21242/42111.2014.00.01.1.1.0) is based upon the German Production Census, the Monthly Report on Plant Operation and the Investment Census. Participation is mandatory for all plants within manufacturing firms that have at least 20 employees. From the plant-level panel, we obtain economic variables such as the number of Employees (our proxy of firm size), Revenue (in D 1000) and Gross Output (in D 1000). We define Export Ratio as the value of exports divided by the value of total revenue. Furthermore, a dummy (Multi-plant firm) identifies firms consisting of more than one plant. Appended at the plant-level, the AFiD-Module Use of Energy (DOI: 10.21242/43531.2014.00.03.1.1.0) provides consumption data (in kWh) for electricity and 14 different fuel types, e.g. natural gas, coal and oil products. Thus, we obtain the total Energy Consumption (in MWh) and define Energy Intensity as the total Energy Consumption (MWh) per Gross Output (in D 1000). As in Löschel et al. (2019), the fuel-specific data enables us to estimate direct CO2 Emissions of the firm (see Appendix B in Supplementary material) via emissions factors (Energiebilanzen: Energiebilanz der Bundesrepublik, 1995; Federal Environment Agency, 2008; Federal Environment Agency 2016a, 2016b). We define CO2 Intensity as respective CO2 Emissions (in tons) per Gross Output (in D 1000). The module also provides the usage of renewable energy, e.g., solar, wind and hydro power (in MWh), and we generate a dummy (Renewables) to indicate firms with any positive values. We further use a dummy (Fossil) to identify firms that generate their own electricity from fossil fuels or have fuel-switching options (i.e. firms using fossil fuels apart from natural gas and oil for heating purposes, but also coal, lignite or process gas and others). Available since 2003, the AFiD-Module Environmental Protection Investments (DOI: 10.21242/32511.2015.00.03.1.1.0 and 10.21242/32511.2005.00.03.1.1.0) comprises the value of investments (in D 1000) in protecting the environment (Air, Water, Waste, Noise, Nature & Landscape, Soil, and Noise) and climate (Energy Efficiency, Greenhouse Gas Mitigation, Renewable Energy). We also generate a dummy to identify firms that have invested in protecting the environment and climate (Eco-Investments). From the EMAS Register as of March 2016, we obtain a sample of 423 plants within 271 firms with at least 20 employees operating within selected industries of the German manufacturing sector that were certified between 1995 and 2013 (see Appendix A in Supplementary material). Of these, we could identify 208 firms within the AFiD data. We aggregate the plant data to the firm level to account for potential spillover effects and specific internal accounting methods. Hence, a firm is coded as newly EMAS certified in the

year that a firm’s first plant enters the EMAS Register. Furthermore, a firm is coded as EMAS certified, even if just one of its plants is certified.5 All monetary variables are deflated to 2010 levels for each 2digit level industry via the index of producer prices for industrial products.6 We reduce the dataset to sectors with our EMAS participants and exclude outliers, i.e. observations in which values of CO2 intensity or energy intensity are above the respective 99th percentile. 3.2. Descriptive statistics In Table 1, we show the number of newly certified EMAS firms which could be identified in our AFiD panel dataset over the years. The program quickly attracted firms initially, but certification rates slowed down since 2000. Presumably due to government incentives for large energy-consuming firms (as discussed in the previous Section), rates have increased again after 2010. The majority of EMAS firms operate in the following industries: fabricated metals, chemicals, machinery, rubber and plastic, and automotive. In Table 2, we display the summary statistics for the years 2000 and 2010 by certification status. A key insight is that, especially in its first years, the program was able to attract policy-relevant firms, i.e. large, energy and emission-intensive firms with more revenue and exports than the average firm. Participants are often multinational firms with multiple plants, which already invest in protecting the environment and climate. They are also more likely to have their own electricity-generation capacities and/or use fossil fuels (such as coal, lignite or liquified natural gas). Next, we describe our empirical strategy. 3.3. Empirical strategy Formally, we are interested in the Average Treatment Effect on the Treated (ATT), which we infer by using Non-EMAS firms as counterfactual observations for EMAS participants. However, as firms voluntarily adopt EMAS, we have to deal with the problem of self-selection (Heckman, 1979). In previous evaluations of voluntary programs, the favored identification strategy is an instrumental variable (IV) approach, such as in Anton et al. (2004); Arimura et al. (2008); Arora and Cason (1996); Barla (2007); Bui and Kapon (2012); Carrión-Flores et al. (2013); Khanna and Damon (1999); Nishitani (2011), or Vidovic and Khanna (2007). Popular instruments are governmental encouragement to participate in previous environmental programs, or the frequency of inspections and of fines. However, the extent to which such instruments are truly exogenous varies from case to case. In our study, variables that could serve as instruments are not available. Prior certification in other programs such as ISO14001, could certainly provide explanatory power for EMAS participation, but it would be difficult to argue that there were no direct effects from prior programs on the outcome variables, i.e. CO2 and energy intensity. There is also no official ISO14001 Register for Germany.

5 We merge the EMAS Register into the census data based on a firm identifier, although this makes it impossible to conduct the analysis at the plant level. However, the majority of the firms in our sample are single-plant firms and we carry out the analysis for these firms only in our main specification. 6 The price indices data is available on the website of the Federal Statistical Office: https://www-genesis.destatis.de/genesis/online (Producer Price Index 61241-0003). From 2009 onwards, the industry classification of the dataset has been based on NACE rev. 2. In earlier years, the dataset was based on NACE rev 1.1, which we manually transfer to NACE rev. 2 via four-digit industry codes and the official reclassification guide of the German statistical offices.

Please cite this article as: R. Kube, K. von Graevenitz, A. Löschel et al., Do voluntary environmental programs reduce emissions? EMAS in the German manufacturing sector, Energy Economics, https://doi.org/10.1016/j.eneco.2019.104558

G Model

ARTICLE IN PRESS

ENEECO-104558; No. of Pages 12

R. Kube, K. von Graevenitz, A. Löschel et al. / Energy Economics xxx (xxxx) xxx

5

Table 1 Number of firms by EMAS certification status across 2-digit industries in selected years. Non-EMAS firms

EMAS firms

Industry (NACE classification) / Year

2000

2005

2010

2013

2000

2005

2010

2013

17) Paper and paper products 20) Chemicals and chemical products 21) Pharmaceutical products 22) Rubber and plastic products 23) Other nonmetallic mineral products 24) Basic metals 25) Fabricated metal products 28) Machinery and equipment n.e.c. 29) Motor vehicles and trailers Total

785 924 227 2,634 1,812 866 6,151 5,214 968 19,581

740 988 228 2,561 1,446 820 5,889 4,915 982 18,569

707 994 221 2,591 1,330 822 6,299 4,955 970 18,889

679 1053 237 2,698 1,363 818 6,669 5,154 956 19,627

4 19 5 10 3 6 23 15 8 93

5 31 5 13 3 6 34 21 12 130

6 35 6 19 5 9 48 28 16 172

10 41 5 25 5 12 57 33 20 208

Note: The table shows the number of firms in our AFiD dataset across 2-digit industries in selected years. We split the figures according to firm EMAS certification status in that year. The columns “Non-EMAS firms” show the number of firms in our AFiD dataset that are not EMAS certified in the respective years. The columns “EMAS firms” show the number of firms that are EMAS certified in the respective years and identified in our AFiD dataset. We focus on certain 2-digit industries in our analysis (see Appendix A in Supplementary material). (Source: Research Data Centre of the Statistical Offices Germany: AFiD-Panel Industrial Units, 2019, AFiD Module Use of Energy, AFiD Module Environmental Protection Investments, survey years 1995-2014, own calculations). Table 2 Summary Statistics for firms by EMAS certification status in the years 2000 and 2010. Year 2000

Year 2010

EMAS firms

Non-EMAS firms

EMAS firms

Non-EMAS firms

Covariate

Mean

Median

Mean

Median

Mean

Median

Mean

Median

Energy Consumption (MWh)

40,803 [55,253] 0.46 [0.53] 24,544 [37,533] 0.23 [0.23] 8,436 [17,161] 141,160 [249,843] 0.30 [0.25] – – 0.10 [0.30] – – 0.25 [0.43] 93

12,285

6,045 [18,845] 0.32 [0.64] 3,563 [18,166] 0.13 [0.21] 1,602 [3,640] 24,779 [80,748] 0.19 [0.23] – – 0.03 [0.18] – – 0.09 [0.30] 19,581

949

38,313 [56,000] 0.49 [0.56] 24,462 [50,062] 0.22 [0.22] 5868 [10,638] 112,566 [192,498] 0.32 [0.25] 0.10 [0.31] 0.17 [0.38] 0.41 [0.49] 0.22 [0.41] 172

13,822

7,112 [20,835] 0.35 [0.61] 3,981 [21,692] 0.14 [0.21] 1,547 [3,411] 27,418 [101,024] 0.24 [0.26] 0.05 [0.22] 0.08 [0.28] 0.12 [0.33] 0.09 [0.29] 18,889

1,278

Energy Intensity (MWh/D 1,000) CO2 Emissions (t) CO2 Intensity (t/D 1,000) Employees (Firm size) Revenue (D 1,000) Export Ratio (percent) Renewables (dummy) Fossil (dummy) Eco-Investments (dummy) Multi-plant firm (dummy) Nr. of observations

0.24 8,583 0.13 3,400 59,313 0.26 – 0.00 – 0.00

0.14 436 0.07 671 7,423 0.09 – 0.00 – 0.00

0.27 6,113 0.15 2,404 40,326 0.28 0.00 0.00 0.00 0.00

0.18 538 0.08 694 7,775 0.16 0.00 0.00 0.00 0.00

Note: The table shows the summary statistics (Mean, Median and Standard deviation) of firms in selected years, separately for firms’ current EMAS certification status in the respective year. Standard deviation is shown in parentheses. The covariates Renewables and Eco-investments have only been available since 2003 (Source: Research Data Centre of the Statistical Offices Germany: AFiD-Panel Industrial Units, 2019, AFiD Module Use of Energy, AFiD Module Environmental Protection Investments, survey years 1995-2014, own calculations).

For these reasons, our empirical strategy is a Difference-inDifferences regression. In the program evaluation literature, the -Difference-in-Differences approach has become well established and is used in studies such as Blackman et al. (2010); Fowlie et al. (2012), or Löschel et al. (2019). We use matching in order to preprocess the data (Ho et al., 2007), so as to reduce sensitivity to functional form assumptions. We implement the Coarsened Exact Matching (CEM) approach proposed by Iacus et al. (2011, 2012), and model participation as a one-time decision. First, the CEM algorithm stratifies matching covariates – binary or continuous into bins, and each participant is matched to all observations that are in exactly the same bins for all matching covariates. The nonparametric nature of the procedure accounts for higher moments of covariates and reduces model dependence. By definition, it yields strong balance in covariates and a lower root mean square error than existing approaches, such as Propensity Score Matching or Nearest-Neighbor Matching (Iacus, 2011, 2012). CEM has been applied successfully in previous studies, e.g., on the effectiveness of

energy audits (Schleich and Fleiter, 2017) or spillover effects from municipal green-building procurement (Simcoe and Toffel, 2014). 3.4. Matching covariates We match firms that are newly certified with EMAS in a given year to firms that are not certified in the same year. The matching algorithm is based on a vector of lagged firm-specific characteristics from pre-certification periods, of which some proxy the firm’s “greenness”. The average duration from the decision to participate until certification takes place, is 15 months, and less than 24 months for almost all respondents in a government survey (Federal Environment Agency, 2013).7 Thus, we use covariate values from

7 Alternatively, we could use the covariate average over more previous years, but this would reduce our sample size critically and also not account for the most recent dynamics of the firms.

Please cite this article as: R. Kube, K. von Graevenitz, A. Löschel et al., Do voluntary environmental programs reduce emissions? EMAS in the German manufacturing sector, Energy Economics, https://doi.org/10.1016/j.eneco.2019.104558

G Model ENEECO-104558; No. of Pages 12 6

ARTICLE IN PRESS R. Kube, K. von Graevenitz, A. Löschel et al. / Energy Economics xxx (xxxx) xxx

Fig. 1. Dynamic matching structure for firms certified with EMAS in arbitrarily chosen period t = 4. Note: The figure depicts the dynamic participation structure and our matching approach. The blue boxes indicate the increasing number of firms that enter the EMAS Register over time, starting with the first adopters in the arbitrary period t = 1. The white box indicates all remaining firms that are not EMAS certified in the respective years, of which some enter EMAS in a later period (indicated by additional blue boxes in future periods, e.g. “Certified in t = 2 in period t = 2) and some firms never enter EMAS during our sample period (which corresponds to the white box in period t = 5). Our approach shows arbitrary firms that enter the EMAS Register in period t = 4 (indicated by the blue box with the green frame in period t = 4): we assume that these firms decided to enter EMAS in period t = 2, according to the EMAS survey of the Federal Environment Agency (2013).Thus, in the two years prior to certification, the firms make an effort to enter the EMAS Register (indicated by the red “certification process” sign), which is why we only consider previous periods as the unaffected baseline years. Therefore, we match these new EMAS firms to those that never participate in EMAS (indicated e.g. by the red box in period t = 4) according to their covariate values in periods t = 1 (and previous periods not shown here for the sake of brevity), indicated by the black box in period t = 1 (and previous periods not shown here for brevity). The matched firms are kept as matches, not only in the certification period t = 4, but also in all other periods of our sample period, and we repeat this matching procedure for the other certification years. We then use regression analysis to compare the change in outcomes over time across the matched firms. (Source: own depiction based on Brand and Xie (2007)).

three years prior to certification that are most likely exogenous to the participation decision. Our approach is related to Brand and Xie (2007) and depicted in Fig. 1. If a firm is certified in an arbitrary year t = 4, we assume it decided to participate in year t = 22 and match according to covariate values from year t = 1. We select the matching covariates that have emerged as most relevant in the previously discussed literature and our own prestudy survey (Kube and Capellen, 2017). First, we include the log of Energy Consumption and the CO2 Intensity as determinants of the environmental impact and costs associated with resource usage.8 An EEG dummy is equal to one if the firm was eligible for exemption from the renewable energy surcharge three years prior to certification.9 This way we control for the possibility that eligibility may have encouraged EMAS certification and created further incentives for energy-saving behavior, or in the case of the EEG exemption, the converse (Gerster, 2017).

8 Although this implies a double-counting of intermediate goods, we prefer gross output to value added, since it more accurately reflects disembodied technological change and does not assume separability between intermediate goods and value added. Finally, the dataset only provides value added data for a subset of firms with more than 500 employees, which would drastically reduce the number of observations for our purpose. 9 Since 2008, energy-intensive firms (energy consumption >10 GWh/a) that are eligible for an exemption must have an energy management system certification like EMAS, ISO14001, ISO50001, or a comparable certification by an external auditor (EEG 2009). Since 2012, only EMAS or ISO50001 are accepted certifications (EEG 2012). However, among the firms that were EMAS certified between 2008 and 2013, only nine are on the list of firms exempt from the surcharge between 2010 and 2013. Similarly, since 2013 energy-intensive manufacturing firms applying for a reduced electricity tax rate also require EMAS or a comparable energy management system certification such as ISO50001. However, all new EMAS participants from 2013 onwards in our dataset exceed the threshold electricity consumption.

Financial capabilities for installing and maintaining a certified environmental management system are proxied via Revenue. We also include the share of revenue from exports (Export Ratio), given that Richter and Schiersch (2017) show that a higher export intensity is associated with significantly lower emissions intensity for German manufacturing firms. A global market scope may also be associated with additional pressure from international stakeholders to manage environmental impacts and certify sustainable production (Khanna and Anton, 2002). Environmental management systems may also be more attractive for firms with their own electricity generation and with an option for fuel-switching decisions, which we indicate with a dummy (Fossil). From 2003 onwards, we proxy awareness of lowcarbon alternatives and pro-environmental attitudes of the firm via dummies that indicate the usage of renewable energy (Renewables) and investments in environmental and climate protection (Eco-Investments). In order to account for industry heterogeneity, we match firms exactly in the same 2-digit industry.10 3.5. Outcome estimation Based on the matched sample, we assess the program’s impact on firms’ environmental performance, by comparing the changes in outcomes across participant status. While EMAS addresses various

10 Researchers face a trade-off between the quality of the match in terms of industry and in terms of other observable characteristics, as matching exactly on 3-digit NACE codes would reduce the number of potential controls substantially in some cases. We follow the literature based on AFiD by matching exactly on 2-digit NACE codes to account for industry heterogeneity (cf. Löschel et al., 2018; Lutz, 2016; Petrick and Wagner, 2014).

Please cite this article as: R. Kube, K. von Graevenitz, A. Löschel et al., Do voluntary environmental programs reduce emissions? EMAS in the German manufacturing sector, Energy Economics, https://doi.org/10.1016/j.eneco.2019.104558

G Model ENEECO-104558; No. of Pages 12

ARTICLE IN PRESS R. Kube, K. von Graevenitz, A. Löschel et al. / Energy Economics xxx (xxxx) xxx

environmental externalities, we focus on CO2 and energy intensity, and further assess the usage of renewable energy and investments in environmental and climate protection. Formally, we estimate the Difference-in-Differences of outcomes (Heckman et al., 1997, 1998) via the following regression of our outcome variable Yit (e.g. log CO2 intensity) of firm i in year t: Yit = ˛i + t + ˇ1 · EMAS i · Preit + ˇ2 · EMAS i · Post it + εit ,

(1)

where ˛i is a firm fixed-effect, t are year dummies. EMAS i is the treatment dummy (and equal to one for all EMAS firms). EMAS i · Preit is equal to one for EMAS firms in one and two years prior to certification. EMAS i · Post it identifies EMAS firms in the certification year and all further years. The error term is εit .We cluster standard errors at the level of the matches, as suggested by Abadie and Spiess (2019). The coefficients of interest are ˇ1 and ˇ2 , to be interpreted as the percentage change of the outcome variable, relative to the baseline level from up to three years prior to certification. In our baseline analysis, we exclude multi-plant firms, because EMAS certification is at the plant level. As a result, treatment intensity at the firm level may vary substantially for multi-plant firms; the EMAS dummy takes the value 1, even if just one of the firm’s plants is certified. This one plant may contribute negligibly to the overall emissions and energy use of the firm, making it hard to justify observed changes in these variables as a result of EMAS. As we are not able to match the EMAS registry at the plant level, we cannot determine the energy use or emissions of the certified plants – except in those cases where the firm only has one plant. One of our robustness checks in Section 5 is to run the CEM regressions, including multi-plant firms. Due to a structural change in the energy statistics in 2003, we split the sample and estimate the models, using two sub-samples: years 1995–2002 and years 2003-2014. Rehdanz, Petrick and Wagner (2011) describe the changes in statistical reporting of the energy use variables over the two periods which make it unadvisable to pool the data across periods as the variable may not be comparable over time. Accordingly, we only estimate the effect for firms that were certified 1998–2002 and 2006-2013.11 Another challenge to our analysis is the fact that the EMAS Register, 2019 only lists currently certified plants, i.e., we cannot observe firms that left the program before 2016. We were able to obtain a list of certified EMAS firms from 2001 from the EMAS study of Rennings et al. (2006). A comparison revealed that out of the 574 manufacturing firms certified by 2001, only 174 firms were still certified in 2016.12 We cannot observe the particular year in which former participants left the EMAS program. However, as we include a large number of Non-EMAS firms as control units also in the CEM-matched sample, it is unlikely that our findings are driven by dropout firms. The problem caused by dropouts is likely to be most severe in the early sub-period until 2002. In a phone interview with the German EMAS Register Helpdesk (March 2016), we were assured that dropout numbers are particularly low among manufacturing firms.13 Nonetheless, it should be kept in mind that our

11 Matching firms on lagged covariates forces us to drop those that joined between 1995 and 1997, losing 55 participant observations in the first sub-period. Likewise we must drop 21 firms that joined between 2003 and 2005. We also omit EMAS firms certified in 2014, since we cannot estimate post-certification effects for these participants. Moreover, we lose observations due to firms that enter the AFiD dataset shortly before EMAS certification. 12 The comparison of firms in EMAS across the years was impeded by the fact that industry codes and classification changed twice between 2001 and 2016. Moreover, the total figures of certified firms deviate from our EMAS sample, as we did not include firms from all manufacturing industries, in contrast to the 2001 sample (see Appendix A in Supplementary material). 13 In phone interviews with EMAS auditors (as of February 2018), we were assured that firms mainly decided to leave EMAS due to its running costs, not because of new requirements following the revisions in 2001 and 2010.

7

results reflect the effect of remaining EMAS-certified rather than the certification decision as such. In Table 3, we display the number of observations available for the regressions, i.e., observations with lagged covariate values, separately for newly certified EMAS firms and Non-EMAS firms in the respective sub-periods. In the first sub-period, there are 42 newly certified EMAS firms with lagged covariate values. Of these, CEM could match 38 to 5011 Non-EMAS firms. In the second sub-period, there are 50 newly certified EMAS firms with lagged covariate values. Of these, 47 could be matched to 5831 Non-EMAS firms. Owing to the large number of Non-EMAS firms used as control units, we are confident that EMAS dropouts do not hamper our analysis. In the last two columns, the same figures are shown for the subsample of a robustness check, in which we do not exclude multi-plant firms (see Section 5). Given these data limitations and the missing information on certification according to ISO14001, our findings should conservatively be interpreted as the net effect of adopting EMAS beyond any such pro-environmental efforts of the firms that we cannot observe. 4. Main results and discussion 4.1. Common trends Inferring the average treatment effect on the treated (ATT) via Difference-in-Differences “. . . permits selection to be based on potential program outcomes and allows for selection on unobservables.” (Heckman et al., 1997). This is valid under the assumption that outcome trends would be equal for both groups in the absence of participation in the program. We assess the common trend assumption in two ways. First, we provide graphical evidence on the trends of log CO2 intensity (Figure C1) and log energy intensity (Figure C2) and how they differ up to five years before and after each certification year, between Non-EMAS firms and newly EMAS certified firms from each certification year. Overall, the trends suggest a parallel structures during the years prior to certification. Second, we run regression analyses where we test for significant differences in outcomes prior to the (assumed) year in which firms decide to participate. The results in Table C1 of the appendix show no significant differences for our main outcomes CO2 intensity and energy intensity in the main specification. However, the test shows a higher propensity of EMAS firms to invest in environmental and climate protection in years prior to the assumed participation decision. In consequence, the regression results for this outcome in Tables 4–6 should be interpreted with care. The remaining results suggest that it is advisable to use the matched sample that excludes multi-plant firms. 4.2. Regression results We now turn to the regressions of our main outcomes within the preferred CEM-matched sample of single-plant firms only (Table 4). For brevity, we only report the EMAS coefficients, as well as the respective standard errors (in parentheses), the 95 percent confidence intervals (in squared brackets) and the p-values (indicated with stars) of each regression. The first two outcome columns in Table 4 show estimates of the average change of outcomes across participation status before (Pre x EMAS) or after (Post x EMAS) certification, relative to the baseline level (all periods up to three years prior to the certification year). The regression results are shown in separate columns for each subperiod. In the first row, the Post x EMAS coefficient measures the average difference between EMAS and Non-EMAS firms regarding the change in log CO2 intensity towards post-certification years

Please cite this article as: R. Kube, K. von Graevenitz, A. Löschel et al., Do voluntary environmental programs reduce emissions? EMAS in the German manufacturing sector, Energy Economics, https://doi.org/10.1016/j.eneco.2019.104558

G Model ENEECO-104558; No. of Pages 12

ARTICLE IN PRESS R. Kube, K. von Graevenitz, A. Löschel et al. / Energy Economics xxx (xxxx) xxx

8

Table 3 Number of observations with lagged covariate values per sub-period in the full sample and CEM matched sample, over all industries, both excluding and including multi-plant firms. Excluding multi-plant firms

Including multi-plant firms

Sub-period

1995-2002

2003-2014

1995-2002

2003-2014

Newly certified EMAS firms Matched: Non-EMAS firms Matched:

42 38 87,629 5,011

50 47 137,265 5,831

52 50 96,832 6,364

58 54 151,933 6,724

Note: The table shows the number of newly certified EMAS firms (identified in our AFiD dataset) and Non-EMAS firms. We only retain firms for which we could include covariates with at least a three-year lag for the purpose of matching. The first two columns show the number of firms when we exclude multi-plant firms, shown separately by sub-period. The last two columns contain the number of firms when we do not exclude multi-plant firms, separately by sub-period. The upper row shows the total number of newly certified EMAS firms and below that we display how many of these firms could be matched via the CEM approach. The last two rows show the total number of available Non-EMAS firms and the number of Non-EMAS firms that were matched to EMAS firms via the CEM approach (Source: Research Data Centre of the Statistical Offices Germany: AFiD-Panel Industrial Units, 2019, AFiD Module Use of Energy, AFiD Module Environmental Protection Investments, survey years 1995–2014, own calculations).

Table 4 Main outcomes for the CEM-matched sample - excluding multi-plant firms. Dep. Var.

Log CO2 Intensity

Sub-period / Margin

1995-2002

2003-2014

1995-2002

Post x EMAS

−0.094* (0.056) [-0.202 ; 0.015] −0.031 (0.038) [-0.106 ; 0.045]

−0.043 (0.063) [-0.163 ; 0.077] −0.058 (0.049) [-0.151 ; 0.036]

−0.061 (0.062) [-0.181 ; 0.058] −0.030 (0.044) [-0.117 ; 0.055]

X X 29,534 <0.0001 0.0202

x x 44,209 <0.0001 0.0134

X X 29,534 <0.0001 0.0157

Pre x EMAS

F-Test for Pre & Post Year dummies Firm fixed effects N Prob > F / Chi2 R2 within

Log Energy Intensity

Renewables (2003-14)

Eco-Investments (2003-14)

2003-2014

Extensive

Intensive

Extensive

Intensive

−0.045 (0.061) [-0.160 ; 0.071] −0.060 (0.054) [-0.161 ; 0.041]

−0.232 (1.007)

32.273 (38.247) [-43 ; 107] −115.740 (102.359) [-317 ; 85]

0.682 (0.420)

40.071 (71.345) [-100 ; 180] 77.741 (77.145) [-74 ; 229]

X X 44,211 <0.0001 0.0274

x x 4659 <0.0001

x x 1903 <0.0001 0.0329

x x 15,225 <0.0001

−1.186 (1.025)

0.119 (0.467)

x x 3296 0.1224 0.0604

Note: Dependent variables are the change in log CO2 intensity and log energy intensity, as well as the usage of renewables (Extensive: OLS regression of the usage dummy, Intensive: OLS regression of the change in consumption in MWh for which we include only firms that use renewable energy), and the investments in environmental and climate protection (Extensive: OLS regression of the usage dummy, Intensive: OLS regression of the change in investments in D 1000, for which we include only investing firms). The latter two outcome variables are only available from 2003 onwards. Standard errors are clustered at the match level and shown in round parentheses. The squared brackets show the 95 percent confidence interval of the coefficients. P-values: *: p < 0.1, **: p < 0.05, ***: p < 0.01 (Source: Research Data Centre of the Statistical Offices Germany: AFiD-Panel Industrial Units, 2019, AFiD Module Use of Energy, AFiD Module Environmental Protection Investments, survey years 1995–2014, own calculations).

Table 5 Robustness check within the CEM-matched sample – including multi-plant firms. Dep. Var.

Log CO2 Intensity

Renewables (2003-14)

Eco-Investments (2003-14)

Sub-period / Margin Post x EMAS

1995-2002 −0.094* (0.052) [-0.196 ; 0.007] −0.044 (0.038) [-0.119 ; 0.030]

2003-2014 −0.093 (0.059) [-0.205 ; 0.019] −0.094** (0.044) [-0.205 ; 0.019]

1995-2002 −0.105* (0.054) [-0.210 ; -0.000] −0.054 (0.041) [-0.135 ; 0.026]

2003-2014 −0.070 (0.051) [-0.169 ; 0.029] −0.078* (0.045) [-0.167 ; 0.010]

Extensive 0.245 (1.026)

Intensive 3.842 (42.317) [-79 ; 87] −47.062 (79.147) [-202 ; 108]

Extensive 0.568 (0.371)

x x 35,883 <0.0001 0.0181

x x 49,515 <0.0001 0.0131

x x 35,884 <0.0001 0.0109

x x 49,517 <0.0001 0.0246

X X 5131 <0.0001

x x 2127 <0.0001 0.0259

x x 17,544 <0.0001

Pre x EMAS

F-Test for Pre & Post Year dummies Firm fixed effects N Prob > F / Chi2 R2 within

Log Energy Intensity

−0.598 (0.954)

0.093 (0.419)

Intensive 98.073 (92.193) [-83 ; 279 114.136 (76.628) [-36 ; 264] x x 3967 0.0958 0.0258

Note: Dependent variables are the change in log CO2 intensity and log energy intensity, as well as the usage of renewables (Extensive: OLS regression of the usage dummy, Intensive: OLS regression of the change in consumption in MWh for which we include only firms that use renewable energy), and the investments in environmental and climate protection (Extensive: OLS regression of the usage dummy, Intensive: OLS regression of the change in investments in D 1000, for which we include only investing firms). The latter two outcome variables are only available from 2003 onwards. Standard errors are clustered at the match level and shown in parentheses. The squared brackets show the 95 percent confidence interval of the coefficients. P-values: *: p < 0.1, **: p < 0.05, ***: p < 0.01 (Source: Research Data Centre of the Statistical Offices Germany: AFiD-Panel Industrial Units, 2019, AfiD Module Use of Energy, AfiD Module Environmental Protection Investments, survey years 1995–2014, own calculations).

relative to the baseline years. Likewise, in the row below, the coefficient Pre x EMAS indicates the average difference between EMAS and Non-EMAS firms regarding the change towards the two precertification years, relative to baseline years. This accounts for the effort shortly after deciding to join EMAS. The row below shows the F-Test statistics for the joint significance of the Pre x EMAS and Post x EMAS coefficients.

In the first sub-period, the CEM point estimates suggest a substantial reduction in CO2 intensity of 9.4 percent in postcertification years relative to the baseline years. The coefficient is significant at the ten percent level. The 95 percent confidence interval illustrates that the effect size is very uncertain and could also be very low or entirely absent. In the pre-certification period, the regression does not show evidence of a significant effect. Moreover, the F-Test statistic suggests

Please cite this article as: R. Kube, K. von Graevenitz, A. Löschel et al., Do voluntary environmental programs reduce emissions? EMAS in the German manufacturing sector, Energy Economics, https://doi.org/10.1016/j.eneco.2019.104558

G Model

ARTICLE IN PRESS

ENEECO-104558; No. of Pages 12

R. Kube, K. von Graevenitz, A. Löschel et al. / Energy Economics xxx (xxxx) xxx

9

Table 6 Robustness check using OLS on the full sample – excluding multi-plant firms. Dep. Var.

Log CO2 Intensity

Log Energy Intensity

Renewables (2003-14)

Eco-Investments (2003-14)

Sub-period / Margin

1995-2002

2003-2014

1995-2002

2003-2014

Extensive

Intensive

Extensive

Intensive

Post x EMAS

−0.099* (0.051) [-0.212; 0.014] −0.035 (0.038) [-0.122 ; 0.053]

−0.058 (0.053) [-0.161 ; 0.0442] −0.068* (0.041) [-0.149 ; 0.014]

−0.070 (0.056) [-0.191 ; 0.051] −0.015 (0.041) [-0.110 ; 0.079]

−0.057 (0.049) [-0.153 ; 0.040] −0.080* (0.043) [-0.170 ; 0.009]

0.499 (1.053)

−224.066 (201.337) [-619 ; 171] −8.558 (168.923) [-340 ; 323]

0.550 (0.433)

13.266 (50.821) [-100 ; 127] 91.247* (50.453) [-20 ; 203]

x x 70,958 <0.0001 0.0103

x x 176,434 <0.0001 0.0125

x x 70,965 <0.0001 0.0088

x x 176,451 <0.0001 0.0244

x x 17,781 <0.0001

x x 8438 <0.0001 0.0269

x x 68,844 .

Pre x EMAS

F-Test for Pre & Post Year dummies Firm fixed effects N Prob > F / Chi2 R2 within

0.115 (0.984)

0.658 (0.453)

X X 18,354 0.0027 0.0024

(convergence not achieved) Note: Dependent variables are the change in log CO2 intensity and log energy intensity, as well as the usage of renewables (Extensive: Logit regression of the usage dummy, Intensive: OLS regression of the change in consumption in MWh for which we include only firms that use renewable energy), and the investments in environmental and climate protection (Extensive: Logit regression of the usage dummy, Intensive: OLS regression of the change in investments in D 1000, for which we include only investing firms). The latter two outcome variables are only available from 2003 onwards. Standard errors are clustered at the match level and shown in parentheses. The squared brackets show the 95 percent confidence interval of the coefficients. P-values: *: p < 0.1, **: p < 0.05, ***: p < 0.01 (Source: Research Data Centre of the Statistical Offices Germany: AFiD-Panel Industrial Units, 2019, AFiD Module Use of Energy, AFiD Module Environmental Protection Investments, survey years 1995–2014, own calculations).

no joint significance of the Pre x EMAS and Post x EMAS coefficients. This raises questions about the evidence of the Post x EMAS effect. The effect of EMAS participation on CO2 intensity is consistent across periods, but we lack the statistical power to precisely identify effects in the later years. For later adopters, both the Pre x EMAS and Post x EMAS coefficients are insignificant, but would imply a reduction by four and six percent, respectively. In other words, our findings suggest a marginally significant, but sizable impact of the program only for those EMAS firms that entered the program shortly after its introduction, when the interest in the program was greatest in terms of new certifications (as visible in Figure A1 for our sample). Next, we assess whether participation had an effect on the energy intensity of production, i.e. energy consumption per unit of gross output. In the first sub-period, the Post x EMAS coefficient is a six percent reduction, but is no longer significant at the ten percent level. This suggests that the observed reduction of CO2 intensity is not the result of a more efficient use of energy, but of other reasons such as changes in the fuel mix. The effects on energy intensity for the second sub-period closely resemble the findings for CO2 intensity. As another channel for environmental improvements, we assess whether EMAS led to an increasing uptake and usage of renewable energy sources. We measure these two outcomes, which correspond to the extensive and intensive margin, in two ways. In the “Extensive” columns, we apply a logit model using the dummy variable to measure any change in the likelihood of using renewable energy at all. In the “Intensive” columns, we use OLS to measure the change in renewable energy usage in MWh (power and other renewable sources) for firms that use renewable energy in the respective years.14 Owing to data availability, we can only analyze the sub-period 2003-2014. In general, we find no evidence that EMAS certification led to an increased uptake of renewable energy technology. Both the Pre x EMAS and Post x EMAS coefficients are insignificant in the logit model. Furthermore, the regression coefficients do not suggest an increased consumption of renewable energy. Our last outcome variable refers to investments in environmental and climate protection. The variable is also only available from

2003 onwards, and comprises various measures to, for example, reduce air and water pollution or waste, as well as investments into energy efficiency. A recent survey of German manufacturing firms by Löschel et al. (2017) reports a significant correlation between energy management practices and investments in energy-saving technologies. We measure this outcome analogously to renewable energy consumption: a Logit model with the dummy variable and a separate regression of the investment value (where we include only firms that already show investment activity).15 Again, we find no evidence that EMAS participation leads to a statistically significant increase in such activity, though the point estimates are positive.

14 An alternative measure using the share of energy from renewables was problematic, due to many observations with a zero share.

15 As for renewables, measuring the investments relative to revenue would lead to many zero values, due to the lumpy nature of investments.

5. Robustness checks In order to assess the robustness of our findings, we complement our CEM-matched sample regression of single-plant firms with different samples. First, we run the regression of the CEM-matched sample again, but now include multi-plant firms. In the first sub-period, this yields 52 newly certified EMAS firms. Of these, 50 were matched via the CEM algorithm to 6364 Non-EMAS firms (see Table 3). In the second sub-period, there are 58 newly certified EMAS firms and 54 could be matched to 6724 Non-EMAS firms. Second, we run an OLS regression on the full sample excluding multi-plant firms. We do not use a matching algorithm. Instead, we  (lagged by three periods) as controls include the covariates X it−3 in the regression model: Yit = ˛i + t + ˇ1 · EMAS i · Preit + ˇ2 · EMAS i · Post it +  · X



it−3

+ s + εit ,

(2)

Note that instead of exact matching within the same 2-digit industry, we now include 2-digit industry-dummies s . Both robustness checks increase the number of observations, but at a price: the inclusion of multi-plant firms entails a less precise treatment identification. Estimation based on the full sample relies on the accuracy of the linear functional form specified in Eq. (2). Moreover, our analysis of pre-treatment trends revealed some evidence of divergent trends in the multi-plant and full sample (see

Please cite this article as: R. Kube, K. von Graevenitz, A. Löschel et al., Do voluntary environmental programs reduce emissions? EMAS in the German manufacturing sector, Energy Economics, https://doi.org/10.1016/j.eneco.2019.104558

G Model ENEECO-104558; No. of Pages 12 10

ARTICLE IN PRESS R. Kube, K. von Graevenitz, A. Löschel et al. / Energy Economics xxx (xxxx) xxx

Table C1), which suggests that the findings should be interpreted with caution. In general, the robustness checks are in line with the results of our preferred specification, for which we detect a slight impact of EMAS certification on CO2 intensity in the years following certification of the first sub-period. In both robustness checks, the Post x EMAS coefficient indicates a 9 percent reduction of CO2 intensity at a significance level of ten percent (Tables 5 and 6). Moreover, the two alternative specifications further suggest that energy intensity decreases by about 8 percent in the years immediately preceding certification in the second sub-period. The results of the analysis including multi-plant firms in the CEM approach are shown in Table 5. In the first sub-period, the estimates suggest that a post reduction of energy intensity was driving the robust result of a reduced Post CO2 intensity. In the second sub-period, the model further suggests a significant reduction of CO2 intensity only in years prior to certification, but not after certification. Finally, neither robustness check suggests any statistically significant impact on the usage of renewable energy or investments in environmental and climate protection.

6. Conclusions In this paper, we investigate whether German manufacturing firms that participate in the voluntary EMAs program show significant changes in their CO2 intensity and energy intensity, relative to Non-EMAS firms. Furthermore, we investigate the impact of participation on the usage of renewable energy sources and investments in environmental and climate protection. The program has been implemented by the European Union, which is why EMAS may be a more credible certification than existing private sector initiatives. Furthermore, the list of requirements for EMAS is more demanding than for comparable voluntary programs. We use the CEM approach to create a matched sample based on lagged covariates, and estimate the Difference-in-Differences of outcomes via regression analysis. An assessment of common trends in CO2 and energy intensity shows no significant difference across participation status, which substantiates the validity of this approach. To check the robustness of our results, we also run OLS regressions with the set of matching covariates. We focus on changes in outcomes both before and after certification. In general, we find evidence of a substantial reduction of CO2 intensity by about 9 percent after EMAS certification. The coefficient is significant at the ten percent level, but confidence intervals are wide. Moreover, this result is only statistically significant for EMAS firms that were certified in the early years of the program, i.e. between 1995 and 2002. While the point estimates are consistently positive across periods, we lack the power to clearly pinpoint effect sizes in the later years where the point estimates are also smaller. It may be that EMAS firms focus their efforts on other externalities such as air, water and land pollution or waste, which we cannot observe in our dataset. Although the point estimates are positive, our findings provide no statistically significant evidence of higher investments in environmental and climate protection nor an increased use of renewable energy sources. Future work could involve pollution data to assess this issue. In addition, due to limitations in the data available to us, the effects measured here are effects over and above those of (unobservable) alternative certifications such as ISO14001. Many EMAS firms are large and energy-intensive firms with an energy management system already implemented. In such cases, an EMAS certification may not provide additional improvements or information (European Environment Agency, 2008). Almost half of EMAS participants are also certified with ISO14001, which in

itself demands an environmental management system (Federal Environment Agency, 2013). One further explanation of our findings could be that EMAS firms actually detect further options for improvements due to the program, but that these measures are just not profitable in conditions with low prices for energy and CO2 emissions. In a scenario of rising prices, a scheme like EMAS might be a stronger lever for environmental policy. Our findings are in line with the literature for other countries. The findings provide limited evidence for the effectiveness of voluntary measures in reducing greenhouse gas emissions. Although EMAS does not focus primarily on energy and resulting emissions, it is among its key performance indicators and a salient option for action. However, EMAS is only one of many certifications for firms’ pro-environmental efforts. Hence, further research on soft instruments is required to gain a more complete picture of the effectiveness of soft regulation. A meta-analysis based on existing research of individual programs may provide insights into which factors are important determinants of program effectiveness. Finally, future research could investigate measures in which firms participate by law, rather than by choice. Potentially, the effects of such policies are stronger as they may address firms with less knowledge about potential measures that ultimately translate into environmental improvements and cost savings. One example is the latest German climate policy measure which introduced energy audits for large firms in Germany in 2015 following the latest EU Energy Efficiency Directive (Federal Ministry for Economic Affairs and Energy, 2014, §§ 8 - 8d EDL-G, and Article 8 of the EU Energy Efficiency Directive (2012/27/EU)). As the empirical evidence for such mandatory audits is scarce (e.g. Fleiter et al., 2012a, b) and EMAS certification fulfills the requirements, our study of a voluntary audit program may provide insights into the potential impact of mandatory audits. If – as we argue above – EMAS firms are better informed about the environmental impact of the firm in advance of EMAS certification, the effects we measure are likely to be a lower bound on effects of a mandatory scheme. On the other hand, firms participating in a voluntary program may be more willing to act on resulting information, making our results more of an upper bound on effects on environmental performance.

Acknowledgements We would like to thank Dietrich Earnhart, Ulrich Wagner, Andreas Ziegler and two anonymous reviewers for their extremely helpful comments and the seminar audiences at the Centre for European Economic Research, the University of Münster, the University of Barcelona and the IAEE Groningen 2018. We gratefully acknowledge the Research Data Centre of the Statistical Offices Germany for granting access to the AFiD data, and in particular Stefan Seitz for his support with our estimation code. We thank Patrick Capellen for his invaluable research assistance. This work was supported by the Bundesministerium für Bildung und Forschung (BMBF) [Grant nr. 100303184 and grant nr. 03SFK4Z0, Verbundvorhaben ENavi - Energiewende-Navigationssystem zur Erfassung, Analyse und Simulation der systemischen Vernetzungen - Teilvorhaben F0, Kopernikus-Projekt der Energiewende].

Appendix A. Supplementary data Supplementary material related to this article can be found, in the online version, at doi:https://doi.org/10.1016/j.eneco.2019. 104558.

Please cite this article as: R. Kube, K. von Graevenitz, A. Löschel et al., Do voluntary environmental programs reduce emissions? EMAS in the German manufacturing sector, Energy Economics, https://doi.org/10.1016/j.eneco.2019.104558

G Model ENEECO-104558; No. of Pages 12

ARTICLE IN PRESS R. Kube, K. von Graevenitz, A. Löschel et al. / Energy Economics xxx (xxxx) xxx

References Abadie, A., Spiess, J., 2019. Robust Post-Matching Inference. MIT & Harvard Working Paper [20.08.2019] https://scholar.harvard.edu/spiess/publications/robustpost-matching-inference. AG Energiebilanzen: Energiebilanz der Bundesrepublik Deutschland 1995 -2014, https://ag-energiebilanzen.de/7-0-Bilanzen-1990-2016.html [10.01.2019]. Allcott, H., Greenstone, M., 2012. Is there an Energy Efficiency Gap? J. Econ. Perspect. 26 (1), 3–28. Anton, W., Deltas, G., Khanna, M., 2004. Incentives for environmental self-regulation and implications for environmental performance. J. Environ. Econ. Manage. 48, 632–654. Arimura, T., Hibiki, A., Katayama, H., 2008. Is a voluntary approach an effective environmental policy instrument? A case for environmental management systems. J. Environ. Econ. Manage. 55, 281–295. Arimura, T., Darnell, N., Katayama, H., 2011. Is ISO14001 a gateway to more advanced voluntary action? The case of green supply chain management. J. Environ. Econ. Manage. 61, 170–182. Arora, S., Cason, T., 1996. Why do firms volunteer to exceed environmental regulation? Understanding participation in EPA’s 33/50 program. Land Econ. 72 (4), 413–432. Barla, P., 2007. ISO 14001 certification and environmental performance in Quebec’s pulp and paper industry. J. Environ. Econ. Manage. 53, 291–306. ˜ C., 2010. Voluntary environmental Blackman, A., Lahiri, B., Pizer, W., Planter, M., Pina, regulation in developing countries: mexico’s Clean industry program. J. Environ. Econ. Manage. 60, 182–192. Bloom, N., Genakos, C., Martin, R., Sadun, R., 2010. Modern management: Good for the environment or just hot air? Econ. J. 120, 551–572. Boyd, G., Curtis, E., 2014. Evidence of an “Energy-Management Gap” in U.S. manufacturing: spillovers from firm management practices to energy efficiency. J. Environ. Econ. Manage. 68, 463–479. Bracke, R., Verbeke, T., Dejonckheere, V., 2008. What determines the decision to implement EMAS? A european firm level study. Environ. Resour. Econ. (Dordr) 41, 499–518. Brand, J., Xie, Y., 2007. Identification and estimation of causal effects with timevarying treatments and time-varying outcomes. Sociol. Methodol. 37, 393–434. Brouhle, K., Griffiths, C., Wolverton, A., 2009. Evaluating the role of EPA policy levers: an examination of a voluntary program and regulatory threat in the metal-finishing industry. J. Environ. Econ. Manage. 57, 166–181. Brouhle, K., Ramirez Harrington, D., 2010. GHG registries: participation and performance under the canadian voluntary climate challenge program. Environ. Resour. Econ. (Dordr) 47, 521–548. Bui, L., Kapon, S., 2012. The impact of voluntary programs on polluting behavior: evidence from pollution prevention programs and toxic releases. J. Environ. Econ. Manage. 64, 31–44. Carrión-Flores, C.E., Innes, R., Sam, A., 2013. Do voluntary pollution reduction programs (VPRs) spur or deter environmental innovation? Evidence from 33/50. J. Environ. Econ. Manage. 66, 444–459. Dasgupta, S., Hettige, H., Wheeler, D., 2000. What improves environmental compliance? Evidence from Mexican industry. J. Environ. Econ. Manage. 39, 39–66. Dehejia, R.H., Wahba, S., 2002. Propensity score matching methods for nonexperimental causal studies. Rev. Econ. Stat. 84 (1), 151–161. EMAS Register: http://www.emas-register.de/. [10.01.2019]. European Environment Agency, 2008. Environmental Agreements – Environmental Effectiveness. Summary – Case Study 3: Germany [10.01.2019] https://www. eea.europa.eu/publications/92-9167-052-9-sum/page006.html. Federal Environment Agency, 2013. EMAS In Germany - Evaluation 2012 [10.01.2019] https://www.umweltbundesamt.de/publikationen/emas-ingermany-evaluation-2012. Federal Environment Agency, 2008. Bestimmung Spezifischer TreibhausgasFür Fernwärme [10.01.2019] https://www. emissionsfaktoren umweltbundesamt.de/publikationen/bestimmung-spezifischer-treibhausgas. Federal Environment Agency, 2016a. Carbon Dioxide Emissions for the German Atmospheric Emission Reporting [10.01.2019] https://www. umweltbundesamt.de/themen/klima-energie/treibhausgas-emissionen. Federal Environment Agency, 2016b. Entwicklung Der Spezifischen Kohlendioxidemissionen Des Deutschen Strommix in Den Jahren 1990 Bis 2015 [10.01.2019] https://www.umweltbundesamt.de/publikationen/entwicklungder-spezifischen-kohlendioxid-2. Federal Environment Agency: https://www.umweltbundesamt.de/themen/ wirtschaft-konsum/wirtschaft-umwelt/umwelt-energiemanagement/iso14001-umweltmanagementsystemnorm#textpart-4, 2017, [10.01.2019]. Federal Ministry for Economic Affairs and Energy. National Action Plan on Energy Efficiency Online: https://www.bmwi.de/Redaktion/DE/Publikationen/Energie/ nationaler-aktionsplan-energieeffizienz-nape.html, 2014, [10.01.2019]. Fisher-Vanden, K., Thorburn, K., 2011. Voluntary corporate environmental initiatives and shareholder wealth. J. Environ. Econ. Manage. 62 (3), 430–445. Fleiter, T., Gruber, E., Eichhammer, W., Worrell, E., 2012a. The German energy audit program for firms—a cost-effective way to improve energy efficiency? Energy Effic. 5, 447–469. Fleiter, T., Schleich, J., Ravivanpong, R., 2012b. Adoption of energy-efficiency measures in SMEs—an empirical analysis based on energy audit data from Germany. Energy Policy 51, 863–875. Fowlie, M., Holland, S.P., Mansur, E.T., 2012. What do emissions markets deliver and to whom? Evidence from Southern California’s nox trading program. Am. Econ. Rev. 102 (2), 965–993.

11

Frondel, M., Horbach, J., Rennings, K., Requate, T., 2004. Environmental Policy Tools and Firm-level Management Practices: Empirical Evidence for Germany. Economics Working Paper / Christian-albrechts-Universität Kiel. Department of Economics, No. 2004, 02. Working Paper available at https://www.econstor.eu/ handle/10419/21979. [10.01.2019]. Gamper-Rabindran, S., 2006. Did the EPA’s voluntary industrial toxics program reduce emissions? A GIS analysis of distributional impacts and by-media analysis of substitution. J. Environ. Econ. Manage. 52, 391–410. Gerarden, T., Newell, R.G., Stavins, R.N., 2017. Assessing the energy-efficiency gap. J. Econ. Lit. 55 (4), 1486–1525. Gerster, A., Working Paper available at https://ssrn.com/abstract=2603211 or http://dx.doi.org/10.2139/ssrn.2603211 [10.01.2019] 2017. Do Electricity Prices Matter? Plant-level Evidence From German Manufacturing. Heckman, J., 1979. Sample selection Bias as a specification error. Econometrica 47, 153–161. Heckman, J., Ichimura, H., Todd, P.E., 1997. Matching as an econometric evaluation estimator: evidence from evaluating a job training programme. Rev. Econ. Stud. 64 (4), 605–654. Heckman, J., Ichimura, H., Smith, J., Todd, P.E., 1998. Characterizing selection Bias Using experimental data. Econometrica 66, 1017–1098. Ho, D.E., Imai, K., King, G., 2007. Matching as nonparametric preprocessing for reducing model dependence in parametric causal inference. Political Anal. 15, 199–236. Iacus, S., King, G., Porro, G., 2011. Multivariate matching methods that are monotonic imbalance bounding. J. Am. Stat. Assoc. 106, 345–361. Iacus, S., King, G., Porro, G., 2012. Causal inference without balance checking: coarsened exact matching. Political Anal. 20 (1), 1–24. IEA, 2015. Capturing the Multiple Benefits of Energy Efficiency. International Energy Agency Publication [10.01.2019] https://www.iea.org/publications/ freepublications/publication/capturing-the-multiple-benefits-of-energyefficiency.html. IEA, 2017. Energy Efficiency. International Energy Agency Market Report Series [10.01.201] https://www.iea.org/efficiency. Iraldo, F., Testa, F., Frey, M., 2009. Is an environmental management system able to influence environmental and competitive performance? The case of the ecomanagement and audit scheme (EMAS) in the European Union. J. Clean. Prod. 17, 1444–1452. Jaffe, A.B., Stavins, R.N., 1994. The energy-efficiency gap: what does it mean. Energy Policy 22, 804–810. Khanna, M., Damon, L., 1999. EPA’s voluntary 33/50 program: impact on toxic releases and economic performance of firms. J. Environ. Econ. Manage. 37, 1–25. Khanna, M., Anton, W., 2002. Corporate environmental management: regulatory and market-based incentives. Land Econ. 78 (4), 539–558. Kim, E.-H., Lyon, T., 2011. Strategic environmental disclosure: evidence from the DOE’s voluntary greenhouse gas registry. J. Environ. Econ. Manage. 61, 311–326. Kube, R., Capellen, P., mimeo 2017. Why Do German Manufacturing Firms Participate in EMAS? Results of an Online Survey. Not Published but Available Upon Request. Lutz, B. J., Emissions Trading and Productivity: Firm-level Evidence From German Manufacturing ZEW Discussion Papers, No. 16-067. Working Paper available at https://ideas.repec.org/p/zbw/zewdip/16067.html, 2016, [10.01.2019]. Löschel, A., Lutz, B. J., Managi, S., The Impacts of the EU ETS on Efficiency and Economic Performance – an Empirical Analyses for German Manufacturing Firms. Resource and Energy Economics ISSN 0928-7655, https://doi.org/10.1016/j. reseneeco.2018.03.001, 2019, [10.01.2019]. Löschel, A., Lutz, B.J., Massier, P., 2017. Credit Constraints, Energy Management Practices, and Investments in Energy Saving Technologies: German Manufacturing in Close-Up (2017). ZEW - Centre for European Economic Research Discussion Paper, No. 17-072, Working Paper available at https://ssrn.com/ abstract=3091625 or http://dx.doi.org/10.2139/ssrn.3091625 [10.01.2019]. Martin, R., Muûls, M., de Preux, L., Wagner, U., 2012. Anatomy of a paradox: management practices, organizational structure and energy efficiency. J. Environ. Econ. Manage. 63, 208–223. Nakamura, S., Takahashi, T., Vertinsky, I., 2001. Why japanese firms choose to certify: a study of managerial responses to environmental issues. J. Environ. Econ. Manage. 42, 23–52. Neugebauer, F., 2012. EMAS and ISO 14001 in the German industry - complements or substitutes? J. Clean. Prod. 37, 249–256. Nishitani, K., 2011. An empirical analysis of the effects on firms’ economic performance of implementing environmental management systems. Environ. Resour. Econ. (Dordr) 48, 569–586. Petrick, S., Wagner, U., The Impact of Carbon Trading on Industry: Evidence From German Manufacturing Firms Working Paper available at http://ssrn.com/ abstract=2389800, 2014, [10.01.2019]. Pizer, W.A., Morgenstern, R., Shih, J.-S., 2011. The performance of industrial sector voluntary climate programs: climate Wise and 1605(b). Energy Policy 39, 7907–7916. Rennings, K., Ziegler, A., Ankele, K., Hoffmann, E., 2006. The influence of different characteristics of the EU environmental management and auditing scheme on technical environmental innovations and economic performance. Ecol. Econ. 57, 45–59. Research Data Centre of the Statistical Offices Germany: AFiD-Panel Industrial Units (DOI: 10.21242/42111.2014.00.01.1.1.0), AFiD-Module Use of Energy (DOI: 10.21242/43531.2014.00.03.1.1.0), AFiD Module Environmental Protection Investments (DOI: 10.21242/32511.2015.00.03.1.1.0 and 10.21242/32511.2005.00.03.1.1.0) for years 1995-2014, Further information: http://www.forschungsdatenzentrum.de/en/index.asp [10.01.2019].

Please cite this article as: R. Kube, K. von Graevenitz, A. Löschel et al., Do voluntary environmental programs reduce emissions? EMAS in the German manufacturing sector, Energy Economics, https://doi.org/10.1016/j.eneco.2019.104558

G Model ENEECO-104558; No. of Pages 12 12

ARTICLE IN PRESS R. Kube, K. von Graevenitz, A. Löschel et al. / Energy Economics xxx (xxxx) xxx

Richter, P.M., Schiersch, A., 2017. CO2 emission intensity and exporting: evidence from firm-level data. Eur. Econ. Rev. 98, 373–391. Rosenbaum, P., Rubin, D., 1983. The central role of the propensity score in observational studies for causal effects. Biometrika 70 (1), 41–55. Schleich, J., Fleiter, T., 2017. Effectiveness of energy audits in small business organizations. Resour. Energy Econ., http://dx.doi.org/10.1016/j.reseneeco.2017.08. 002 [10.01.2019]. Simcoe, T., Toffel, M.W., 2014. Government green procurement spillovers: evidence from municipal building policies in California. J. Environ. Econ. Manage. 68 (3), 411–434.

Videras, J., Alberini, A., 2000. The appeal of voluntary environmental programs: which firms participate and why? Contemp. Econ. Policy 18 (4), 449–461. Vidovic, M., Khanna, N., 2007. Can voluntary pollution prevention programs fulfill their promises? Further evidence from the EPA’s 33/50 Program. J. Environ. Econ. Manage. 53, 180–195. Ziegler, A., Busch, T., Hoffmann, V., 2011. Disclosed corporate responses to climate change and stock performance: an international empirical analysis. Energy Econ. 33, 1283–1294.

Please cite this article as: R. Kube, K. von Graevenitz, A. Löschel et al., Do voluntary environmental programs reduce emissions? EMAS in the German manufacturing sector, Energy Economics, https://doi.org/10.1016/j.eneco.2019.104558