Abatement Expenditures, Technology Choice, and Environmental Performance in Mexico (PDF)

Journal of Development Economics 133 (2018) 264–274 Contents lists available at ScienceDirect Journal of Development Economics...

Journal of Development Economics 133 (2018) 264–274 Contents lists available at ScienceDirect Journal of Development Economics journal homepage: www.elsevier.com/locate/devec Abatement expenditures, technology choice, and environmental performance: Evidence from ﬁrm responses to import competition in Mexico Emilio Gutiérrez a , Kensuke Teshima b,* a ITAM, Department of Economics, Mexico b ITAM-CIE, Mexico A R T I C L E I N F O A B S T R A C T JEL code: Abatement expenditures are not the only available tool for ﬁrms to decrease emissions. Technology choice can F18 also indirectly aﬀect environmental performance. We assess the impact of import competition on plants’ envi- O33 ronmental outcomes. In particular, exploiting a unique combination of Mexican plant-level and satellite imagery Q56 data, we measure the eﬀect of tariﬀ changes due to free-trade agreements on three main outcomes: plants’ fuel use, plants’ abatement expenditures, and measures of air pollution around plants’ location. Our ﬁndings show Keywords: that import competition induced plants in Mexico to increase energy eﬃciency, reduce emissions, and in turn Environment reduce direct investment in environmental protection. Our ﬁndings suggest that the general technology upgrading Technological change eﬀect of any policy could be an important determinant of environmental performance in developing countries Remote sensing data and that this eﬀect may not be captured in abatement data. Plant-level response to trade 1. Introduction higher, and lower environmental quality can be both attributed to inferior technology and low regulatory capacity (Greenstone and Jack Reductions in trade barriers between developed and developing (2015)). Bloom et al. (2013) shows that gains from better manage- nations can aﬀect the environment in a variety of ways: through ment practice is substantial for Indian textile ﬁrms. If ﬁrms in develop- changes in the location of production (Grossman and Krueger (1992), ing countries are then further away from the technological frontier or Copeland and Taylor (1994), and Antweiler et al. (2001))1 and best management practice, the potential impact of trade on technolog- consumption across countries (Davis and Kahn (2010)), or through ical choice can play a larger role in these contexts than in developed resource reallocation from low-productivity to high-productivity plants nations. A broad understanding of the impacts of trade on environ- within and across nations (Holladay (2016) and Yokoo (2009)). mental outcomes seems also particularly relevant in developing coun- For the U.S., recent literature has shown that improvements in tries, as high pollution levels impose larger health and productivity environmental outcomes can also be driven by a change in technol- costs.2 ogy (Levinson (2009)), and that this change can occur within plants The trade literature has accumulated evidence in the context of (Shapiro and Walker (2017)). Speciﬁcally in the U.S. context, changes developing countries that the impact of trade on technology choice in regulatory eﬀort are the main drivers of the change in technol- is substantial.3 In particular, in the same Mexican context, Teshima ogy choice. Nonetheless, the channels through which trade can aﬀect (2010) ﬁnds that a decline in Mexican trade barriers induced Mexican environmental outcomes, and the consequences of those changes in ﬁrms to upgrade their technology. In light of these ﬁndings, we explore developing countries may diﬀer substantially from those in developed whether in the same context, tariﬀ reductions had an indirect eﬀect on nations. Pollution concentrations in developing countries are remarkably * Corresponding author. Avenida Camino a Santa Teresa 930, México D.F. 10700, Mexico. E-mail addresses: [email protected] (E. Gutiérrez), [email protected] (K. Teshima). 1 Copeland and Taylor (2003) summarize nicely the literature in this line of research. 2 For example, see Arceo et al. (2016) and Hanna and Oliva (2015). 3 See, for example, Verhoogen (2008) and Bustos (2011). Tanaka (2017) analyzes the impact of trade on management practice in Myanmar. https://doi.org/10.1016/j.jdeveco.2017.11.004 Received 17 April 2017; Received in revised form 24 November 2017; Accepted 26 November 2017 Available online 2 December 2017 0304-3878/© 2017 Elsevier B.V. All rights reserved. E. Gutiérrez and K. Teshima Journal of Development Economics 133 (2018) 264–274 environmental outcomes precisely through their impact on technology matter concentrations around ﬁrms’ location as a result of import com- choice.4 petition. Being aware of the diﬀerent challenges for the control of and Our results show that the impact of trade barriers on technology the consequences of pollution concentrations in developing countries, choice and, as a result, environmental outcomes are an important mech- economists and policy makers have implemented policies aimed at anisms for a broader understanding of the impact of trade on the envi- improving regulatory capacity (Duﬂo et al. (2013) and Foster and ronment in developing countries’ settings. Moreover, they suggest than Gutierrez (2016)), and fostering the adoption of cleaner technologies even when detailed data at the plant level are available, caution should by ﬁrms (Ryan (2015)).5 be taken when trying to measure the eﬀects of any policy on environ- As for the evaluation of these policies, measuring the impact of mental performance. In our setting, one would have been tempted to a decrease in trade barriers on environmental performance is a very conclude that import competition aﬀects the environment negatively if diﬃcult task, as it requires concrete and reliable measures of envi- only abatement expenditures had been available. Since trade has eﬀects ronmental performance at the ﬁrm level, which are relatively hard to not only on the incentives to pollute but also on the adoption of diﬀer- obtain, particularly in developing countries. While looking at interme- ent technologies (which may already be more eﬃcient and less pollut- diate outcomes such as energy eﬃciency (Ryan (2015)) or abatement ing), abatement expenditures may decrease even as emissions decrease. expenditures (Wang (2002)) may be informative, the ﬁnal mapping of It is then necessary to obtain data both on plants’ abatement expendi- these measures into total emissions may not be trivial (Conrad and Mor- tures and on environmental performance in general, in order to better rison (1989)). Furthermore, the environmental impact of any policy understand the relationship between output price (driven in this case by that aﬀects technology choice can be particularly hard to gauge when trade openness), technology adoption, and the aggregate eﬀect of both only intermediate measures of environmental performance are avail- on pollution emissions. This message generally applies to any evalua- able. For instance, if the adoption of new technologies decreases emis- tion of the determinants of environmental performance. sions, it may also decrease abatement expenditures (particularly if these Our paper is related to several strands of literature. As in other top- are devoted to end-of-pipe abatement strategies). This point has been ics in international trade, trade economists have been increasingly ana- long recognized, at least implicitly, but ﬁnding evidence is hard because lyzing ﬁrm/plant level data to study the relationship between trade researchers would need data on all of the abatement eﬀort, technology and the environment. Holladay (2016) theorizes that exporters pol- and environmental outcomes and an exogenous shock to ﬁrm behavior lute less per unit of output than non-exporters in the same indus- to disentangle the causality. try, and ﬁnds supporting empirical evidence for US plants. Forslid In this paper, we show that in the Mexican context, abatement et al. (2015) advance this line of research further ﬁnding both the- expenditures and technological change can respond diﬀerently to a oretically and empirically that this is because exporters invest more change in trade tariﬀs, exploiting a unique combination of Mexican in abatement. In a developing country context, Barrows and Ollivier plant-level and satellite imagery data that together allow us to con- (2014) further analyze the impact of export market access on ﬁrm- struct three main outcomes: plants’ fuel use, plants’ abatement expen- level product choice and its consequences on ﬁrm-level substance use ditures measured as investment in eﬃcient energy and environmental by Indian ﬁrms.9 Rodrigue and Soumonni (2014) analyze the rela- protection, and measures of air pollution around plants’ location. As tionship between exports and environmental innovation, using direct an exogenous factor to ﬁrms, we analyze import competition induced investments in environmental protection as an outcome. Most of the by free trade agreements.6 Constructing ﬁrm-speciﬁc and time-varying papers analyze exports. An exception is Cherniwchan (2017), who ana- measures of output tariﬀs faced by Mexican ﬁrms during the 2000–2003 lyzes the impact of NAFTA on the emission of U.S. plants. Apart from period, we explore whether tariﬀ changes aﬀect energy eﬃciency, focusing on the impact of trade on environmental outcomes for the abatement expenditures and pollution concentrations around plants’ Mexican context, our paper diﬀers from Cherniwchan (2017) by look- locations, controlling for industry-speciﬁc and state-speciﬁc time vari- ing at the interaction between self-reported information on environ- ation in the outcomes of interest.7 We ﬁnd that import competition is mental performance and independently collected measured of pollution related to an increase in plants’ energy eﬃciency, but a decrease in concentrations form satellite imagery.10 In addition, an important mes- abatement expenditures.8 The overall eﬀect of import competition on sage of our paper is that improvement in energy eﬃciency and thus environmental quality is then impossible to infer from these two coun- environmental quality due to trade may not be found in abatement teracting forces, particularly in a setting in which reliable measures data. of emissions at the ﬁrm level are unavailable. We provide further evi- Apart from exploring a diﬀerent channel through which trade can dence on the impact of import competition on pollution concentrations aﬀect developing countries’ ﬁrms’ environmental performance, while by exploiting particulate matter concentration measures obtained from focusing on a very speciﬁc policy change apparently unrelated to envi- satellite-imagery, ﬁnding small but positive reductions in particulate ronmental regulations, our results are informative to the literature that measures the eﬀect of diﬀerent policies on environmental out- comes (Duﬂo et al. (2013) and Foster and Gutierrez (2016)). Our paper 4 Throughout the text, we use the term “technology choice” rather broadly. As we also speaks to the literature trying to test for a link between envi- will explain in better detail below, the data at hand makes it impossible to identify the ronmental regulation and technological choice (Porter (1996), Jaﬀe precise technological changes that drive the changes in environmental performance in et al. (2002) and Acemoglu et al. (2012)) and between input prices this context. The robustness section presents evidence suggesting that the main driver and innovation (Popp (2002)). An important message from this paper of the increases in energy eﬃciency is very likely a change in cost-cutting practices and improvements in technical eﬃciency. 5 Policies that try to encourage households to adopt cleaner technologies have also been implemented. See for example Davis et al. (2014). 6 9 Lipscomb (2008) also extends Melitz’s heterogeneous-ﬁrm model to analyze how We show in the data section that substantial fractions of the tariﬀ reduction during this period is driven by phaseout of tariﬀs due to NAFTA. This is attractive for our purpose environmental regulation aﬀects the production decisions of multi-product plants, and as Kowalczyk and Davis (1998) argue that Mexican tariﬀ reductions due to NAFTA were how the reallocation of resources resulting from these decisions aﬀects industry-level driven by U.S. interests, and not those of Mexican ﬁrms. We present further econometric environmental outcomes in India. evidence on this point in Section 5. 10 The classic papers concerning the eﬀect of trade on the environment are Grossman 7 Our empirical strategy is similar to that in Teshima (2010), which ﬁnds that import and Krueger (1992), Copeland and Taylor (1994, 1995). Antweiler et al. (2001) disen- competition has an eﬀect on ﬁrms’ technology choice, but we provide independent and tangle the eﬀects of trade into scale, composition and technology eﬀects. Our paper high- more checks on the assumption of exogeneity of tariﬀs in our context and sample. lights a particular channel through which trade could aﬀect the environment through its 8 We are not the ﬁrst to interpret a reduction in output tariﬀs as an increase in com- eﬀect on technology, although there are other channels. Analyzing them and decompos- petition. See, for example, Holmes and Schmitz (2010) and De Loecker and Goldberg ing the whole eﬀects through which trade could aﬀect the environment is beyond the (2014). scope of this paper. 265 E. Gutiérrez and K. Teshima Journal of Development Economics 133 (2018) 264–274 is that the fact that ﬁrms might lower their emissions of via cost- [National Council of Science and Technology]. saving process improvements, either through technology upgrading or The survey contains information on several aspects of innovative management practice, but without investment in abatement, should be activities of manufacturing plants: expenditures, human resources and considered when designing environmental policy for developing coun- collaboration between ﬁrms and research institutions. It includes infor- tries. mation on expenditures for each type of R&D: product R&D and pro- Finally, our paper is also related to studies using satellite image cess R&D. We use the 2002 and 2004 surveys. Each survey elicits data. The use of data obtained from satellite imagery has become information for the previous two years. This allows us to construct a widespread practice in empirical papers in economics (Donaldson an unbalanced panel from 2000 to 2003. In addition to the standard and Storeygard, 2016). In particular, the use of Aerosol Optical Depth technology-related variables, the survey asks how much plants spend on (AOD) as a measure of particulate matter concentrations is not unique socio-economic activities. Speciﬁcally, the survey asks how much plants to this paper.11 The usefulness of AOD measures of particulate matter spend on (1) care and control of the environment (cuidado y control del concentrations relies on their availability in contexts where no ground medio ambiente), which includes prevention, detection and improve- measures of pollution concentrations exist, and on the independent ment of contamination of land, water, and air,12 (2) rational production nature of the data collection process, which does not respond to politi- and use of energy (Producción y uso racional de la energía),13 and (3) cal pressures. All Bombardini and Li (2016), Gutierrez (2015) and Chen health except pollution reduction (salud (excluyendo contaminación)). et al. (2013) document a very close relationship between ground mea- We use (1) and (2) as abatement expenditures for the main analysis and sures and estimates of pollution concentrations and AOD in contexts in (3) in the robustness check section. We call (1) environmental invest- which both are available. ment, (2) energy investment and (3) health investment.14 To our knowledge, apart from this paper, only Bombardini and Li There are ESIDET surveys for the various sectors. The survey for (2016) measure the impact of trade shocks, export market access in the manufacturing sector addresses plants with more than 50 employ- their case, on environmental quality using AOD as dependent variable. ees. The survey uses the Economic Census of 1999 to draw a sample. Apart from the diﬀerence in the particular channel of trade in which Among the plants in the Economic Census of 1999, the plants with more their and our study focus respectively, the main innovation in our study than 500 employees are included in the sample with certainty.15 Plants is then to compare the results using remote sensing images with those with at most 500 employees are sampled with probability depending that can be indirectly inferred from survey measures of environmen- on whether they have employees (a) between 50 and 100, (b) 101 and tal performance which, may lead, as we show, to misleading conclu- 250 and (c) 251 and 500. sions. In particular, we use them to explore how the interaction between changes in direct investment in abatement eﬀorts and in fuel use from 2.1.2. EIA the part of ﬁrms sum up to the total change in air quality as a result of In order to obtain energy-related expenditures and sales, and thus trade shocks. energy eﬃciency, we draw the Encuesta Industrial Anual (EIA) [Annual This paper is organized as follows. The next section describes the Industrial Survey]. The EIA is a longitudinal plant level dataset in 205 new combination of datasets, and presents descriptive statistics of of the 305 6-digit industries in manufacturing. The EIA is also compiled plant-level variables as well as the air pollution measures. Section 3 by INEGI. EIA sampling design selects largest producers from Economic describes our econometric strategy. Section 4 presents the key results Census and continues adding to sample until the target share, 85 per- of the eﬀects of competition on plants’ energy eﬃciency, environmen- cent, of the covered domestic sales at each industry is reached.16 tal investment, and the pollution level at the plants’ locations. Section 5 presents a series of robustness checks. The ﬁnal section concludes. 2.1.3. SIEM For information on the output and input categories of the ﬁrms to 2. Data calculate output and input tariﬀs at the plant level, we use the Sis- tema de Información Empresarial Mexicano (SIEM) [Mexican Company 2.1. Plant-level data Information System] compiled by Mexico’s Secretaría de Economía [Ministry of Economy]. It is a directory of ﬁrms in Mexico to facilitate We combine three types of plant-level data for the analysis. The business contacts between ﬁrms in Mexico and foreign ﬁrms. SIEM lists ﬁrst is a specialized survey on innovative activities from which we ﬁrms’ inputs and outputs at the 6-digit or 8-digit trade-classiﬁcation draw abatement expenditures, measured as investment in environmen- level regardless of whether the ﬁrms export or import. It does not have tal technology. The second is a standard plant-level survey from which information on the volumes of each output or input, or on whether the we draw information on energy eﬃciency, measured as expenditure on plants export or import. The SIEM starts in 1997, but detailed infor- fuel and electricity divided by total sales. The third is a registry of plants mation about ﬁrms’ inputs and outputs are available only from 2001. that includes information on the trade-classiﬁcation category of plants’ Firms are legally obliged to report; therefore in principle the SIEM can outputs and inputs from which we construct measures of plant-level be regarded as a census of ﬁrms in the formal economy. The SIEM has tariﬀ changes. 2.1.1. ESIDET The source for the information on the environmental and energy 12 An example of this type of investment is investment for development and installation investment is the Encuesta Sobre Investigación y Desarrollo de Tecnología for measuring, preventing and controlling pollutants. 13 (ESIDET) [Survey on Research and Development of Technology]. This An example of this type of investment is investment for generating and distribut- ing electricity within plants. Importantly for our purpose, expenses on energy eﬃcient is a conﬁdential survey carried out by the Instituto Nacional de Estadís- equipment are not included. tica, Geografía (INEGI) [National Institute of Statistics and Geography] 14 Both environmental and energy investment may be noisy for our purposes. Environ- of Mexico for the Consejo Nacional de Ciencia y Tecnología (CONACYT) mental investment may contain eﬀort towards reducing water or land pollution, while we focus on air pollution. Energy investment may include investment on production of energy. We ﬁnd that the results are similar though sometimes less precise when we analyze environment investment and energy investment separately. Importantly, these investment refers only to technology-related investment, thus excludes advertisement expenditures, for example. 11 15 Plants for Tobacco, Ship-building, Airplane, and Electronic components are included See for instance Gutierrez (2015) and Foster et al. (2009) for papers using these data for the Mexican context, and Jayachandran (2009), Chen et al. (2013) and Bombardini with certainty regardless of the size. and Li (2016) for studies in other contexts than the Mexican. 16 Further details on EIA can be found in Appendix II in Verhoogen (2008). 266 E. Gutiérrez and K. Teshima Journal of Development Economics 133 (2018) 264–274 been linked by INEGI personnel to the EIA and ESIDET using informa- geographic locations. The estimated AOD daily value for each zip-code tion on ﬁrm name, state, municipality, street address, and industry. was averaged for each month in the sample. The yearly average is the mean of all monthly averages. 2.2. Descriptive statistics of plant-level variables 2.4. Descriptive statistics of the pollution measure Appendix Table B.1 presents summary statistics of environmental and energy investment for the ESIDET-EIA-SIEM panel. Consistent with Appendix Fig. B.1 shows a map with the calculated AOD level for the trade literature on exporting ﬁrms, exporters are larger in terms the year 2000 in all Mexican zip codes for which we have precise geo- of employment. Exporters not only spend more on fuel and electric- graphic coordinates and for which AOD measures are available. The ity but also have a higher share of these expenditures on total sales, lighter dots represent zip codes with lower AOD levels. Although, as though this may be reﬂecting the industry composition of exporters stated, diﬀerences in AOD levels across regions can be due to geo- and non-exporters. Exporters are more likely to be engaging in environ- graphic and climatic conditions unrelated to concentrations of partic- mental and energy investment and have higher expenditure. However, ulate matter, AOD measures do appear higher around metropolitan only 6% of these exporters report a positive amount of environmental areas and along the Gulf Coast (possibly due to the importance of the investment. This ratio is 4% for all the plants and 2% for non-exporters. oil industry in this region). Appendix Fig. B.2, in contrast, maps the changes in our AOD measure within zip codes between 2000 and 2003. 2.3. Satellite imagery data Darker dots represent the zip codes that experienced higher increases in AOD during this period. Clear geographic patterns on the increase or In order to assess the overall impact of the changes in tariﬀs on reduction of our AOD measures during the period are not evident. plants’ environmental performance, we constructed a zip-code level Appendix Table B.2 shows statistics for both AOD levels in 2000 and dataset, which assigns, along with measures of weighted tariﬀ changes changes in AOD between 2000 and 2003 for all 378 zip codes matched in each zip-code, measures of pollution concentrations in the atmo- with our ﬁrm-level dataset. The mean AOD level in our sample in the sphere around them. For this, we obtained daily measures of Aerosol year 2000 was 0.39, ranging from 0.02 to 0.98 and with a standard Optical Depth (AOD) at a 5 km spatial resolution for cloud-free images deviation of 0.22. As our regression estimates diﬀerence out variations for the entire land area of Mexico over the 2000–2003 time period. in AOD levels across zip codes with the zip code ﬁxed eﬀects, the rel- A higher value of AOD means less transparency (lower air quality). evant variation exploited in this paper corresponds to changes in this The data were obtained from the Moderate Resolution Imaging Spec- variable. Between 2000 and 2003, for all zip-codes in our sample, the troradiometer (MODIS onboard the Terra Satellite), of NASA’s God- change in AOD (on average close to zero) ranges from −0.32 to 0.38 dard Space Flight Center Earth Sciences Distributed Active Archive with a standard deviation of 0.09 (more than 20% of the average AOD Center (DAAC). Aerosols are liquid and solid particles suspended in level in 2000). To put this variation in context, it is perhaps useful to the air, and AOD can be described as the extinction of beam power mention that Foster et al. (2009), exploiting variation in AOD and infant caused by the presence of these particles in the atmosphere. AOD mea- mortality within municipalities, found that the elasticity of infant mor- sures obtained form satellite imagery are particularly useful in contexts tality with respect to AOD in Mexico is approximately 4. where no ground measures of pollution concentrations exist, and are likely to be more reliable than alternative sources due to the indepen- 2.5. Tariﬀ data dent nature of the data collection process. For the Mexican context, these AOD measures have already made it possible to evaluate pollution We construct tariﬀ data using (1) Mexican import statistics pub- abatement policies (Foster and Gutierrez (2016)), and their potential lished in trade statistics yearbooks and (2) tariﬀ information from relationship with health outcomes (Gutierrez (2015, 2010) and Foster the tariﬀ law of Mexico and from the documents of the free trade et al. (2009)).17 agreements between Mexico and other countries. The ﬁrst subsection The strong relationship between AOD and other measures of partic- describes the method to calculate plant-level tariﬀs. The second subsec- ulate matter concentrations in the atmosphere is well documented in tion describes the summary statistics for the tariﬀ data. the existing literature. For example, Bombardini and Li (2016), Gutier- rez (2015) and Chen et al. (2013) document a very close relationship 2.5.1. Construction of plant-level tariﬀ measures between ground measures and estimates of pollution concentrations Because of free trade agreements, tariﬀs for one product diﬀer and AOD in contexts in which both are available.18 Kumar et al. (2007) depending on the country of origin. We ﬁrst aggregate the country-good show that linear regression estimates suggest that a 10 percent change speciﬁc tariﬀs to good-level tariﬀs by taking the weighted average with in AOD explains a 0.52 percent change in their ground measure of par- the initial volume of imports used as weights. Importsg jct is imports of ticulate matter (PM2.5,) with an R-squared of 0.71. However, while good g in industry j from country c at time t. Tarif fg jct is tariﬀ of good AOD is a good predictor of general levels of suspended particles in g in industry j from country c at time t. the atmosphere, it is worth mentioning that it does not allow to make ∑ any distinction between pollutants, and that comparisons across regions Tarif fgjt = 𝛼c Tarif fgjct (1) c with diﬀerent climate and geographic conditions are hard to make. Our analysis will then focus on changes in AOD levels within zip codes. Importsgjc2000 where 𝛼c = ∑. As the information at the plant level is available yearly in our analy- c Importsgjc2000 Next, using this good-level tariﬀ data Tarif fg jt , we take the simple sis, we constructed a measure of the average yearly AOD level for each average of the tariﬀs of each plant’s outputs to construct the output zip code in our data set.19 Using GIS, the observed measures of AOD tariﬀs at the plant level.20 from the satellite images were overlapped with each zip-code’s exact ∑ g ∈Gi Tarif fgjt Output Tarif figt = (2) Ni 17 See Jayachandran (2009), Chen et al. (2013) and Bombardini and Li (2016) for stud- ies in other contexts than the Mexican. 18 See also Chu et al. (2003) and Gupta et al. (2006). 19 We ﬁnd qualitatively similar eﬀects when we use 95 percentile of yearly AOD and 20 when we calculate the average level of AOD of each month and use the highest value We have to use the simple average because SIEM data does not allow one to obtain among monthly AOD for a given year. the information on the volumes of each product by plant. 267 E. Gutiérrez and K. Teshima Journal of Development Economics 133 (2018) 264–274 where Gi is the set of products that plant i produces, and Ni is the 3.2. Zip-code level analysis number of products of plant i produces, respectively. Similarly, we take the simple average of the tariﬀs of each plant’s As stated, in order to assess the aggregate eﬀect that the changes inputs in the initial period to construct the input tariﬀs at the plant in plant-level outcomes translate into changes in pollution emissions, level. Note that we always use the outputs and inputs information from we present a set of results relating the changes in tariﬀs to changes in year 2001 to compute the output and input tariﬀs for each year.21 Thus environmental performance by directly looking at measures of pollution all the variation of the tariﬀ of a good is coming from the changes in concentrations around plants’ location. If a measure of environmental the tariﬀ of the good but not from the changes in the volume of the performance at the plant level were available, we would run the same imports of the good. This is to avoid bias due to the changes in output speciﬁcation as in the previous sub-section, using this measure as our mix or in input mix in response to the tariﬀ reduction. outcome variable. However, AOD measures pollution concentrations in When we calculate the weighted average tariﬀs for imports from all the atmosphere at the zip-code level, and more than one plant can be the countries as well as from four groups of sets of countries: NAFTA, located in the same zip-code. We then assume that the pollution concen- EU, countries to which most favored nations (MFN) tariﬀs are applied, trations in each zip-code are a weighted average of the pollution emis- and other countries that are not in NAFTA or in EU and that have a free sions by each plant in that zip-code. We calculated a weighted average trade agreement with Mexico, we see that the tariﬀ changes are largely of the tariﬀ variable in the main regression equation for each zip-code, coming from tariﬀ changes scheduled, late NAFTA liberalization and using the total number of employees reported by each plant divided by the free trade agreement with EU. In terms of plant-level tariﬀs, average the total number of employees in each zip code (the sum of the employ- output tariﬀs decreased from 7.7% in 2000 to 4.1% in 2003. Appendix ees of all plants in the SIEM database in each zip code) as the weight Table B.3 presents summary statistics for tariﬀs. It shows that the tariﬀ for each of the plant-level observations, and run regressions, with each changes are largely coming from tariﬀ changes scheduled in free trade of these variables as regressors, at the zip-code level. Speciﬁcally, we agreements, evidencing that changes are exogenous. run the following regression: AODzjt = 𝛽1 Output Tarif fzt + 𝜆z + 𝜇jt + 𝛾 Xmt + 𝜖zjt (4) 3. Speciﬁcation where z denotes zip-code. 𝜆z captures the idiosyncratic eﬀect of each 3.1. Plant-level analysis zip-code. 𝜇jt is a dummy variable indicating whether the zipcode has any plants in industry j. 𝛾 Xmt include municipality-level weather- The baseline econometric model is the following: related variables, more speciﬁcally temperature and dew, and their polynomials, taken from one of the authors’ earlier work (Gutierrez Yijt = 𝛽1 Output Tarif fit + 𝜆i + 𝜇jt + 𝜖ijt (3) (2015)). where i, j, and t index plants, industries, and years, respectively; Yijt 4. Results denotes the dependent variable: (Inverse) energy eﬃciency measured as the share of expenditures on fuel or/and electricity over total sales, 4.1. Results: plant-level measures abatement expenditures measured as the sum of environmental invest- ment and energy investment; Output Tariﬀ it is output tariﬀs at the plant 4.1.1. Results: energy use level constructed in the manner described in the tariﬀ data section; 𝜆i Table 1 presents the regression results for diﬀerent measures of is a plant ﬁxed eﬀect; 𝜇jt is an industry-year ﬁxed eﬀect; 𝜖ijt is an error energy use on the output tariﬀ. Columns (1) and (2) use the sum of elec- term.22 tricity and fuel expenditures over total sales as the dependent variable. The coeﬃcient of interest in these regressions is 𝛽1. 𝛽1 corresponds Column (1) shows a positive and statistically signiﬁcant coeﬃcient for to the changes in the dependent variables in response to a one percent our measure of output tariﬀs, suggesting that an increase in competition point change in the output tariﬀ, which captures (the inverse of) the increases energy eﬃciency in general. A one percentage point decrease eﬀect of competition. The plant ﬁxed eﬀects capture all observed or in the output tariﬀ implies that energy-related expenditures over sales unobserved time-invariant heterogeneity across plants. The industry- fall by about 0.05 percentage points.23 Column (2) shows that the result year ﬁxed eﬀects capture all observed or unobserved shocks at the is robust to the inclusion of state-year ﬁxed eﬀects, suggesting that the industry level. Thus, the coeﬃcient of interest is identiﬁed on the results are not driven by changes in geographic conditions or state-level basis of within-plant changes in the three types of tariﬀs and within- policies. plant changes in the dependent variables controlling for industry-level The next four columns show the results of the same speciﬁcation, idiosyncratic shocks. The identiﬁcation assumption of this econometric disaggregating the dependent variable into electricity over sales and model is that no unobservable factors are correlated with the output fuel over sales. Column (3) of Table 1 shows that there is a signiﬁ- tariﬀs after controlling for time-invariant plant-level heterogeneity and cant and positive eﬀect of output tariﬀs on electricity use over sales, industry-level idiosyncratic shocks. suggesting that the increase in competition driven by the change in tar- Note that a positive value of the coeﬃcient means that output tariﬀ iﬀs increases electricity eﬃciency. A one percentage point decrease in reduction aﬀects the dependent variable negatively. A priori, there is the output tariﬀ implies that electricity expenditures over sales fall by no clear theoretical prediction on whether the coeﬃcients should be about 0.02 percentage points.24 This result is again robust to the inclu- positive or negative. In some speciﬁcations, we also control for state- sion of state-year ﬁxed eﬀects (Column (4)). Using fuel eﬃciency as year ﬁxed eﬀects to control for any shocks at the region level. the dependent variable, Columns (5) and (6) show that the coeﬃcients of the output tariﬀs are similar in sign and quantitatively larger than those in the previous two columns. However, these coeﬃcients are not signiﬁcantly diﬀerent from zero. Given this, it is diﬃcult to conclude 21 This is not ideal as the plant-level data set starts in 2000. However, as product-level which of the two types of expenditure is most aﬀected by the change in information is available at the SIEM only since 2001. The output and input composition does not change much at least between 2001 and 2002. output tariﬀs. 22 In the robustness check section, we show that our results are robust to inclusion of Overall, we ﬁnd in this analysis that the increase in import com- other tariﬀs: input tariﬀs, which are Mexican tariﬀs imposed on plants’ imported interme- petition induced by output tariﬀ reductions leads to an increase in the diate inputs, and US tariﬀs, which are US tariﬀs imposed on plants exports of outputs to US. Furthermore, we present results with robust standard errors. Results clustering stan- 23 dard errors at the industry level do not change the signiﬁcance level of our coeﬃcients The mean of energy-related expenditures over total sales is 2 percent. of interest. The results are available upon request. 24 The mean of electricity expenditure over total sales is 1 percent. 268 E. Gutiérrez and K. Teshima Journal of Development Economics 133 (2018) 264–274 Table 1 Regressions of the intensity of electricity and fuel over sales on output tariﬀs, ESIDET-EIA-SIEM panel 2000–2003a. (1) (2) (3) (4) (5) (6) Dependent Variable Sum Sum Electricity Electricity Fuel Fuel Output Tariﬀ 0.0481** 0.0551** 0.0262* 0.0223* 0.0262 0.0258 (0.0235) (0.0274) (0.0149) (0.0133) (0.0344) (0.0315) Plant ﬁxed eﬀects Yes Yes Yes Yes Yes Yes Industry-year eﬀects Yes Yes Yes Yes Yes Yes State-year eﬀects No Yes No Yes No Yes Observations 1776 1776 1776 1776 1776 1776 R2 0.17 0.22 0.18 0.22 0.19 0.23 a Notes: The table reports coeﬃcients on the output tariﬀs from plant-level regressions of the intensity of expenditures on electricity and fuel over total sales on the output tariﬀs, plant ﬁxed eﬀects, industry-year ﬁxed eﬀects and in some cases state- year ﬁxed eﬀects. Plant-level output tariﬀ for a plant is the simple average of the product-level tariﬀs of the products that the plants produce. Robust standard errors in parentheses. Signiﬁcance: * 10 percent, ** 5 percent, *** 1 percent. Table 2 Regressions of the environmental and energy investment on output tariﬀs, ESIDET-EIA-SIEM panel 2000–2003a. (1) (2) (3) (4) (5) (6) Dependent Variable Sum of Environmental and Energy Investment Intensity Dummy Log Output Tariﬀ 0.0023* 0.0026* 0.0072** 0.0074** 0.059*** 0.066*** (0.0014) (0.0014) (0.0028) (0.0029) (0.021) (0.021) Plant ﬁxed eﬀects Yes Yes Yes Yes Yes Yes Industry-year eﬀects Yes Yes Yes Yes Yes Yes Region-year eﬀects No Yes No Yes No Yes Observations 1776 1776 1776 1776 1776 1776 R2 0.23 0.27 0.24 0.27 0.22 0.27 a Notes: The table reports coeﬃcients on the output tariﬀs from plant-level regressions of the intensity, dummy and the log of the sum of environmental and energy investment on the output tariﬀs, plant ﬁxed eﬀects and industry-year ﬁxed eﬀects. Plant-level output tariﬀ for a plant is the simple average of the product-level tariﬀs of the products that the plants produce. Robust standard errors in parentheses. Signiﬁcance: * 10 percent, ** 5 percent, *** 1 percent. energy eﬃciency of aﬀected plants. Our interpretation is that this is environmental investment at the plant level. The overall eﬀect of the due to the improvement in plants’ general technology, which previous tariﬀ change on environmental performance is therefore uncertain. empirical studies have shown to be a result of the same tariﬀ changes. In order to shed some light into the aggregate eﬀect of changes in Speciﬁcally, Teshima (2010) ﬁnds that, for the same plants and over tariﬀs on environmental performance, Table 3 presents the results for the same time period, increased competition (measured by the same our zip code-level regressions, with our measure of pollution concen- changes in output tariﬀs) increases total R&D and process R&D. These trations (AOD) as the dependent variable. Columns (1) only includes increases might have been accompanied by the adoption of new tech- zip-code and industry-year eﬀects and zip-code level controls such as nologies which brought on savings in electricity and/or fuel expendi- total sales from the plants in our sample. Column (5) present the results tures. We return to this point in Section 5.2. including zip-code, industry-year and state-year ﬁxed eﬀects. Because, in our setting, air quality should improve if the positive eﬀect of tech- 4.1.2. Results: environmental and energy investment nology adoption is higher than the potential negative eﬀect driven by Next, we report the results of regressions that use measures of envi- the decrease in investment in preventing emissions, Columns (2) and ronmental and energy investment as dependent variables. We use three (6) include an interaction term between initial AOD levels and out- types of investment measures: investment intensity, measured as invest- put tariﬀs. The extent to which the scope for technology adoption to ment over total sales, the log of investment, and an investment dummy. reduce emissions is larger for initially more polluting technologies, we Table 2 shows the results. Columns (1) and (2) suggest that a one expect the coeﬃcient associated with this interaction to be positive. To percentage point decrease in output tariﬀs implies a decrease in envi- explore this directly, we perform two types of analysis. First, Columns ronmental and energy investment over sales by about 0.002 percentage (3) and (7) include an interaction term between the output tariﬀs and points. Columns (3) and (4) suggest that the same decrease in output the initial average energy intensity of plants in the zip code, constructed tariﬀs leads to a 0.7 percentage points decrease in the likelihood of as total-sales-weighted average of energy use intensity of the plants in investing in energy and the environment. Columns (5) and (6) suggest the zip-code. Second, Columns (4) and (8) include an interaction term that the same one percentage point decrease in output tariﬀ leads to between the output tariﬀs and a dummy variable indicating whether a a 5–6 percent decrease in the amount spent on such types of invest- zip code has at least one ﬁrm in a relatively more polluting industry.25 ment. Columns (2), (4) and (6) show the results after also controlling As Column (1) shows, the coeﬃcient of the output tariﬀ implies for state-year eﬀects, suggesting that none of these results are driven that tariﬀ reductions decrease pollution concentrations around plants’ by state-speciﬁc economic ﬂuctuations. Overall, the increase in import locations. An increase of 1% in tariﬀs is associated with an increase in competition induced by output tariﬀ reductions led to a decrease in the AOD of 0.0017 points (around 0.4 percentage points). The eﬀect of the environmental and energy investment of the aﬀected plants. 4.2. Zip code level results on the pollution measure 25 We use the World Bank’s Industrial Pollution Projection System (IPPS), which includes emission intensities by four-digit SIC code. We manually matched SIC code to the The results on energy eﬃciency and environmental investment seem Mexican industry classiﬁcation. We used the ﬁrst principal component of emission inten- to go in opposite directions. While tariﬀ reductions (increased compe- sity of CO, SO2 , and NO2 , and deﬁne relatively more polluting industries as industries tition) imply higher energy eﬃciency, they are also related to lower whose value is higher than the median. 269 E. Gutiérrez and K. Teshima Journal of Development Economics 133 (2018) 264–274 Table 3 Regressions of the AOD measure on output tariﬀs:2000-2003a. Dependent Variable (1) (2) (3) (4) (5) (6) (7) (8) AOD Output Tariﬀ 0.002* −0.009*** −0.006*** −0.007** 0.001 −0.008*** −0.007*** −0.007** (0.001) (0.001) (0.001) (0.002) (0.001) (0.002) (0.002) (0.003) Output Tariﬀ 0.021*** 0.016*** *AOD2000 (0.005) (0.005) Output Tariﬀ 0.25*** 0.19*** *Energy Intensity 2000 (0.06) (0.05) Output Tariﬀ 0.027*** 0.022*** *Dummy (Pollution Intensity more than (0.006) (0.005) Median 2000) Zip-code ﬁxed eﬀects Yes Yes Yes Yes Yes Yes Yes Yes Zip-code Controls Yes Yes Yes Yes Yes Yes Yes Yes Industry-year eﬀects Yes Yes Yes Yes Yes Yes Yes Yes Region-year eﬀects No No No No Yes Yes Yes Yes Observations 1512 1512 1512 1512 1512 1512 1512 1512 R2 0.96 0.96 0.97 0.97 0.96 0.97 0.97 0.97 a Notes: The table reports coeﬃcients on the output tariﬀs and its interaction term with the initial level of AOD from zip-code level regressions of the AOD measure on the output tariﬀs, the interaction term, zip-code ﬁxed eﬀects, industry-year ﬁxed eﬀects and state-year eﬀects. Zip-code-level output tariﬀ is the weighted average of the plant-level tariﬀs of the products that the plants in the zip-code produce. Energy intensity 2000 is constructed by taking the zip-code level total-sales-weighted average of energy intensity (energy expenditure divided by total sales) of the plants in the zip-code. A variable, Pollution Intensity more than Median 2000, is a dummy variable indicating whether a zip-code has at least one ﬁrm in a relatively more polluting industry. Robust standard errors in parentheses. Signiﬁcance: * 10 percent, ** 5 percent, *** 1 percent. output tariﬀ is not robust to the inclusion of state-year eﬀects; its coef- particular given that our measure of energy eﬃciency is the quotient ﬁcient becomes insigniﬁcant in Column (5). However, the eﬀect of the between energy expenditures and total sales, both price-dependent. We output tariﬀ for zipcodes with higher initial AOD, with higher energy start by discussing this potential source of bias. Unfortunately, the level intensity, or with plants in more polluting industries, is positive and sta- of disaggregation in available producer price indices for the Mexican tistically signiﬁcant, with and without state-year eﬀects (Columns (2), context is lower than the level of disaggregation in our industry-level (3), (4), (6), (7) and (8)). ﬁxed eﬀects. Given that we already control for industry-year eﬀects in The results together show that the overall impact of tariﬀ changes on all the analyses, using existing price indices cannot address this poten- pollution emissions is driven by changes in energy eﬃciency, and not tial concern. by changes in environmental investment. This suggests that, through However, during our study period, gas and other oil derivatives’ competition, trade liberalization can have distinct eﬀects on plant-level prices were constant within states, as PEMEX, a government-owned environmental performance and that, even when direct abatement mea- ﬁrm, was the only existing one in the industry. While some diﬀer- sures are available, empirical studies should be careful when interpret- ences in prices across states existed (aimed at avoiding cross-border ing results. Regulation is usually based on capping emissions, and trade shopping), within each state, oil and gas prices were ﬁxed each year has eﬀects not only on the incentives to pollute but also on the adoption by PEMEX. Electricity prices were determined diﬀerently as there was of diﬀerent technologies. If this is the case, when adopting new, more more than one provider at the national level. However, the exist- eﬃcient and less polluting technologies, abatement eﬀorts (which are ing electricity providers in the country had monopoly power at the possibly made only to comply with environmental regulation, and not state level, and prices were constant across industries within each to improve overall eﬃciency) may decrease. We believe our results to state. The inclusion of state-year ﬁxed eﬀects should then control suggest the need to obtain data both on plant’s direct investment in for changes in our outcome variable driven by changes in energy pollution abatement and on environmental performance when trying to prices. empirically test for the relationship between trade openness, technol- Changes in output prices, however, may indeed vary diﬀerentially ogy adoption, and the aggregate eﬀect of both on pollution emissions. across industries during our study period. The direction of the bias induced by this potential confounding factor depends then on the cor- 5. Robustness relation between changes in tariﬀs and changes in output prices.26 However, this correlation is very likely positive: output prices must 5.1. Endogeneity concerns decrease when output tariﬀs are reduced.27 Recalling that our mea- sure of energy eﬃciency is the quotient between energy expendi- Given the non-experimental nature of the variation in tariﬀs that tures and sales, in face of a drop in output prices, all else constant, we exploit in the empirical analysis, our identiﬁcation strategy may fail we would observe an increase in the dependent variable, as only to estimate the causal impact of tariﬀ changes on the outcomes ana- the denominator will be lower. Our results show that when tariﬀs lyzed if there are omitted variables correlated with both our explana- decrease, energy expenditures over sales decrease. We then argue that tory and dependent variables, or in the presence of reverse causality. In the potential bias induced by the potential correlation between changes this section, we discuss the potential sources of bias, and present evi- in tariﬀs and output prices is, if anything, biasing our results towards dence suggesting that they are not likely an important concern for the zero. interpretation of our ﬁndings. For this purpose, it is perhaps useful to recall that our preferred estimates control for diﬀerential time trends by industry (industry-year ﬁxed eﬀects), and diﬀerential trends by region 26 The Industrial Organization literature, for example, De Loecker and Warzynski (2012) (state-year ﬁxed eﬀects). The set of potentially omitted variables is then and De Loecker et al. (2016) study how changes in input tariﬀs aﬀect mark-ups indepen- restricted to those that, as our measure of tariﬀs, vary diﬀerentially over dently of quantity. One implication of this literature is that caution should be taken when time and across industries in the same region. interpreting the results when researchers do not have information on quantity and price separately. 27 Columns (1) and (2) in Table 5 show that the output tariﬀ reductions decreased 5.1.1. Prices the values of total sales and domestic sales of plants. This is consistent with a positive Obvious candidates to satisfy this last requirement are prices, in correlation between output tariﬀs and prices. 270 E. Gutiérrez and K. Teshima Journal of Development Economics 133 (2018) 264–274 5.1.2. Other types of tariﬀs Another potential misspeciﬁcation issue is related to the timing of the Nonetheless, while the changes in output prices driven directly by plants’ response to tariﬀ changes. The extent to which tariﬀ changes are import tariﬀs are not likely to induce the results presented in this paper, expected by plants in our dataset, and that plants may react in advance changes in output prices driven by other forces may bias our estimates, to these expected changes, the relationship between environmental out- to the extent that these forces can be correlated with output tariﬀ comes and contemporaneous tariﬀ changes presented in this paper may changes. In order to explore if this is indeed the case, we run our main fail to measure the relationship we seek to analyze. speciﬁcation including a larger set of controls, which could arguably be In order to explore if plants responded to expected, not current tar- correlated with both output tariﬀs and prices. In particular, we run the iﬀ changes, and more generally, to indirectly test the validity of our same speciﬁcation, this time additionally including input tariﬀs, aver- empirical strategy, we would ideally show that there were no diﬀeren- age Mexican tariﬀs on products ﬁrms use as inputs, and the average US tial background trends in the outcomes of interest between plants who tariﬀs Mexican ﬁrms would face if they exported their products to the faced decreasing output tariﬀs and those who did not. Unfortunately, US, as controls. In particular, we run regressions of the following form: the characteristics of the dataset exploited do not allow us to fully per- form this test, as the information on abatement expenditures as well Yijt = 𝛽1 Output Tarif fit + 𝛽2 Input Tarif fit + 𝛽3 US Tarif fit + (𝛾 Xit ) as AOD is only available for the survey rounds included in the main + 𝜆i + 𝜇jt + 𝜖ijt (5) results. Nonetheless, we can eﬀectively test for diﬀerential pre-trends in energy eﬃciency and other potentially relevant characteristics as infor- where Input Tarif fit denotes the input tariﬀ on plant i at time t, which mation on energy expenditures and other potentially relevant plant- was constructed in the same way as the output tariﬀs in the previous characteristics are available from the EIA. In order then to test whether sections and US Tarif fit denotes US tariﬀs on goods produced by plant the changes in tariﬀs from 2000 to 2003 are correlated with changes in i at time t. Plant-level control variables include employment, export these outcomes before the changes in tariﬀs eﬀectively took place, we ratio, and total sales and capital. run a series of regressions of the following form: In this framework, 𝛽1 , still captures changes in the dependent vari- ables in response to changes in the output tariﬀ, and can be interpreted △Yij1997−2000 = 𝛽1 △ Output Tarif fij + 𝛽2 △ Input Tarf fij as (the inverse of) the eﬀect of import competition. 𝛽2 captures the + 𝛽2 △ US Tarif fij + 𝜇j + 𝜖ij (6) eﬀect of changes in the input tariﬀ, interpreted as (the inverse of) the eﬀect of increased access to imported intermediate products. Finally, 𝛽3 Table 4 shows the results. As opposed to our main results, the coeﬃ- measures the eﬀect of changes in the US tariﬀ that plants would face if cients of interest are small and insigniﬁcantly diﬀerent from zero, sug- they exported, and can be interpreted as (the inverse of) the eﬀect of gesting that the parallel trends assumption in our diﬀ-in-diﬀ setting export market access. holds. Appendix Table A.1 shows the results of these regressions for the two plant-level dependent variables analyzed in previous sections, i.e. 5.2. Mechanisms and interpretations energy eﬃciency, environmental and energy investment. The magni- tude of the coeﬃcients on the output tariﬀ stays roughly the same as in Throughout the text, we speculate that the reduction in tariﬀs the previous speciﬁcations and is still signiﬁcant. Therefore, the results reduced abatement expenditures and increased energy eﬃciency due to we have been putting forth do not appear to be driven by other changes the fact that tougher competition may have induced ﬁrms to upgrade occurring during the same time period, such as increased access to their technology and, as a side-result, decrease their emissions, reducing imported intermediate products and increased access to export mar- the need for directly investing in abatement eﬀorts. We do not measure kets. Appendix Table A.2 shows the results of these regression for the technology or emissions directly. As a result, we are aware that the AOD measure. Again, our results are robust to inclusion of the other exploration of the precise mechanisms through which the eﬀects found types of tariﬀs.28 can be explained is a very diﬃcult task, and that our interpretation is It is perhaps worth stressing that our main results (presented in the not the only potential explanation of our ﬁndings. Nonetheless, several previous section) do not include input and US tariﬀs as control vari- key facts make us relatively conﬁdent that this is the most likely expla- ables, since input tariﬀs are found to be correlated with plants’ charac- nation for the eﬀects found. teristics and thus are very likely to not be exogenous (Teshima (2010)). In order to shed better light on the potential mechanisms behind Furthermore, for the sample analyzed, the eﬀects of input tariﬀ and the eﬀects found, Table 5 shows the regression results when investment US tariﬀ reductions are not likely strong, as they would aﬀect only in process R&D, sales and health investments are used as dependent the subset of plants that are using imported intermediate products or variables. exporting. 5.2.1. Technology and scale eﬀects 5.1.3. Endogeneity of tariﬀs First, we ﬁnd an impact of tariﬀ changes on investment in process A second potential concern is that industry characteristics may be R&D, replicating the results of Teshima (2010) in our sample. Columns correlated with tariﬀ changes. Kowalczyk and Davis (1998), however, (3) and (4) of Table 5 show the results. Thus, we ﬁnd evidence consis- show that tariﬀ reductions due to NAFTA were driven by U.S. interests, tent with our hypothesis that output tariﬀ reductions induced plants to and not those of Mexican ﬁrms. In order to provide further evidence upgrade technology in their production process. that this is not an important concern in our setting, Teshima (2010) As shown in Columns (1) and (2) of Table 5, tariﬀ reductions are provides evidence of no relationship between plant characteristics in associated with a decrease in the value of sales. While we cannot iden- 2000 and the subsequent output tariﬀ reductions within an industry. tify whether this decrease is due to prices or quantities, the extent to which it may be driven by a decrease in production is an important con- cern for the interpretation of our ﬁndings, as another potential mecha- 28 A reader may wonder why input tariﬀ reductions have the same sign as output tar- nism behind the observed increase in energy eﬃciency may be related iﬀs. We believe that the results on input tariﬀ (reductions) are consistent with our gen- to a change in scale. eral technology upgrading hypothesis. Trade literature emphasizes that the reduction in Our main results, however, suggest that the increase in energy eﬃ- input tariﬀs allow domestic ﬁrms to buy more cheaply foreign intermediate products that ciency is driven by both a decrease in electricity and fuel expenditures embody foreign better technology, and thus to do more technology upgrading and inno- vation (see for example Goldberg et al. (2010). Furthermore, Amiti and Konings (2007) over sales. We believe that existing installed capacity is very unlikely ﬁnd both output tariﬀ reductions and input tariﬀ reductions increase plant-level TFP. Our to be aﬀected by tariﬀ changes in the short term, as it may require results are consistent with these ﬁndings. substantial capital investments. Fuel is likely used mainly to power the 271 E. Gutiérrez and K. Teshima Journal of Development Economics 133 (2018) 264–274 Table 4 Correlations between the changes in initial sales and employment and changes in tariﬀs, EIA-SIEM panel 1997–2003a. (1) (2) (3) (4) (5) (6) (7) △Log △ Log △Exporter △Log △Log △Log △Energy Total Sales Domestic Sales Dummy Exports Employment TFP Eﬃciency Output Tariﬀ Changes 0.0103 0.0088 −0.0012 −0.0157 −0.0100 −0.0015 0.0032 (0.0111) (0.0134) (0.0025) (0.0244) (0.0232) (0.0087) (0.0047) Input Tariﬀ Changes 0.0058 0.0133 −0.0112 −0.0123 −0.0122 −0.0110 −0.0011 (0.0088) (0.0145) (0.0108) (0.0241) (0.0223) (0.0321) (0.0043) US Tariﬀ Changes 0.0067 0.0059 0.0023 0.0234 −0.0136 0.0052 −0.0055 (0.0114) (0.0137) (0.0052) (0.00355) (0.0189) (0.0102) (0.0048) Industry dummies Yes Yes Yes Yes Yes Yes Yes Observations 625 625 625 416 625 625 625 R2 0.335 0.336 0.242 0.259 0.318 0.143 0.132 a Notes: The table reports coeﬃcients on the changes from 2000 to 2003 in output tariﬀs, input tariﬀs and U.S. tariﬀs from plant-level regressions of the changes in plant characteristics from 1997 to 2000 on the changes in these tariﬀs and industry eﬀects. TFP is estimated using Olley-Pakes method. Plant-level output tariﬀ for a plant is the simple averages of the product-level tariﬀs of the products that the plants produce. Similarly, the plant-level input tariﬀ for a plant is the simple averages of the product-level tariﬀs of the products that the plant uses as intermediate products. The plant-level U.S. tariﬀ for a plant is the simple averages of the U.S. tariﬀs for U.S imports from Mexico of the products that the plants produce. Robust standard errors in parentheses. Signiﬁcance: * 10 percent, ** 5 percent, *** 1 percent. Table 5 Regressions of other outcome variables on tariﬀs, ESIDET-EIA-SIEM panel 2000–2003a. (1) (2) (3) (4) (5) (6) Dependent Variable Log Total Log Domestic Process Process R&D Health Health Sales Sales R&D Intensity Dummy Investment Investment Intensity Dummy Output Tariﬀ 0.0125* 0.0164** −0.0224* −0.0107* −0.0022 −0.0065 (0.068) (0.0080) (0.0123) (0.0066) (0.0032) (0.0052) Input Tariﬀ −0.078 −0.0053 0.0135 0.0007 0.0035 0.0079 (0.0079) (0.0879) (0.0082) (0.0034) (0.0027) (0.0082) US Tariﬀ −0.0032 0.0015 −0.0152 −0.0027 −0.0033 −0.0100 (0.0031) (0.0027) (0.0111) (0.0024) (0.0029) (0.0101) Plant ﬁxed eﬀects Yes Yes Yes Yes Yes Yes Industry-year eﬀects Yes Yes Yes Yes Yes Yes Region-year eﬀects Yes Yes Yes Yes Yes Yes Observations 1776 1776 1776 1776 1776 1776 R2 0.06 0.04 0.16 0.20 0.14 0.13 a Notes: The table reports coeﬃcients on the output tariﬀs, the input tariﬀs, and the U.S. tariﬀs from plant-level regressions of various variables on these three tariﬀs, plant ﬁxed eﬀects, industry-year ﬁxed eﬀects and state-year ﬁxed eﬀects. Log Process R&Dit = log (Process R&Dit + 1), abd Log Health Investmentit = log (Health Investmentit + 1). Plant-level output tariﬀ for a plant is the simple average of the product-level tariﬀs of the products that the plants produce. Similarly, the plant-level input tariﬀ for a plant is the simple average of the product-level tariﬀs of the products that the plant uses as intermediate products. Similarly, the plant-level U.S. tariﬀ for a plant is the simple average of the product-level tariﬀs of the products that the plant would face if they export to the U.S. Robust standard errors in parentheses. Signiﬁcance: * 10 percent, ** 5 percent, *** 1 percent. existing machinery, which remains unchanged. Fuel use per unit of out- leave less room for such activities. If incentives for plants to spend on put is then very unlikely to decrease in face of a decrease in production abatement are mainly driven by altruism, and if overusing electricity in the context analyzed. Electricity, however, may be used to power can to some extent be some form of luxury (such as turning the lights machinery, but is also likely to vary with respect to total employees on when unnecessary) an increase in competition could reduce plants’ and other more ﬂexible production inputs. If our results were driven by proﬁts and, as a consequence, decrease abatement expenditures and a scale reduction, we would then expect the decrease in fuel per unit increase energy eﬃciency. Thus, the negative eﬀects of increased com- of output to decrease in a smaller magnitude than electricity per unit petition on environmental and energy investments found in this paper of output, which is not the case in the context of our study. We then could therefore be a result of reduction in CSR activities, not because of argue that the results presented in the main paper are unlikely driven substitutability between such investment and general technology. This by a change in scale. alternative hypothesis cannot fully explain why then the energy eﬃ- Furthermore, for each industry, the correlation between the ciency and pollution measures could improve. Nonetheless, we provide 2000–2003 changes in sales and changes in energy expenditure divided one more piece of evidence against this alternative explanation. We run by sales within industry is rather small. The mean of the correlation regressions estimating the eﬀects of the changes in the three types of coeﬃcient is 0.03. Less than 10% of the industries have a correlation tariﬀs on health investment. Our idea is that if environmental invest- coeﬃcient lower than −0.2. Thus, it appears to be the case that the ment decreased after an increase in competition because of the reduc- extent to which changes in sales mechanically drive our results on the tion of CSR activities, then we should also see the negative eﬀect of changes in energy eﬃciency is limited. competition on other types of social investment. Columns (5) and (6) of Table 5 show that there are no signiﬁcant eﬀects of any tariﬀs on health 5.2.2. Abatement expenditure as luxury goods investment and that the coeﬃcients on output tariﬀs have the opposite Both environmental and energy investment might be carried out sign to those found for environmental and energy investment. While we as a form of Corporate Social Responsibility (CSR) activities or lux- understand that this test does not fully address this potential issue, we ury expenditures from the part of ﬁrms. If so, competitive pressure may believe it is suggestive evidence that a decrease in luxurious expendi- 272 E. Gutiérrez and K. Teshima Journal of Development Economics 133 (2018) 264–274 tures is not the main driver of the decrease in abatement expenditures thank Arturo Blancas, Jorge Reyes, Gerald Leyva, Natalia Volkow, Abi- as a result of output tariﬀ reductions. gail Duran, Mauro Gascon, Adriana Ramirez, and Gabriel Romero of INEGI for assistance with establishment surveys. We also thank Alberto 5.2.3. Additional results Zarco of Secretaría de Economía for his help on the SIEM data. Both Finally, in the online appendix, we present evidence suggesting that authors received support from the Asociacion Mexicana de Cultura. We (a) an endogenous regulatory response is not likely driving our main are solely responsible for any errors. results, (b) exit due to tariﬀ reduction would not wholly explain our results on AOD, and (c) the results on AOD are not likely biased due to Appendix A. Supplementary data spatial spillover eﬀects. Supplementary data related to this article can be found at https:// 6. Conclusion doi.org/10.1016/j.jdeveco.2017.11.004. We have found evidence that the reduction of tariﬀs on the goods References produced by Mexican plants is associated with improved energy eﬃ- ciency, reduced air pollution near plants’ location, but also reduced spe- Acemoglu, Daron, Aghion, Philippe, Bursztyn, Leonardo, Hemous, David, 2012. The environment and directed technical change. Am. Econ. Rev. 102 (1), 131–166. ciﬁc environmental and energy investment. This suggests that import Amiti, Mary, Konings, Jozef, 2007. Trade liberalization, intermediate inputs, and competition induced by trade liberalization indirectly aﬀected the envi- productivity: evidence from Indonesia. Am. Econ. Rev. 97 (5), 1611–1638. ronment positively, through the plants’ incentive to change their energy Antweiler, Werner, Copeland, Brian R., Scott Taylor, M., 2001. Is free trade good for the environment? Am. Econ. Rev. 91 (4), 877–908. eﬃciency through general technological investment. This eﬀect would Arceo, Eva, Hanna, Rema, Oliva, Paulina, 2016. Does the eﬀect of pollution on infant have been diﬃcult to identify by solely analyzing the environment and mortality diﬀer between developing and developed countries? Evid. Mexico City. energy in

Abatement Expenditures, Technology Choice, and Environmental Performance in Mexico (PDF)

Document Details

Tags

Related

Summary

Full Transcript

Upgrade to continue