The effects of exports on facility environmental performance: Evidence from a matching approach

abstract

This paper employs matching techniques to investigate the effects of facility export status on environmental performance. Using facility-level criteria air emission data in the US manufacturing industry, we find the industry-specific effects of export status on emission intensity, measured by emissions per value of sale. In some industries, there is consistent and robust evidence supporting the superior environmental performance of exporters relative to non-exporters in terms of emission intensity for all criteria air pollutants tracked. In other industries, we find weak evidence that exporters appear to have a higher emission intensity than non-exporters. This industrial heterogeneity in the effects of exporting on the environment is closely related to industrial characteristics including pollution abatement capital expenditure, trade costs, capital intensity and others.

KEYWORDS:

• Criteria air emissions
• exports
• propensity score matching, industrial heterogeneity

JEL Classifications:

• F18
• Q56

Acknowledgments

We gratefully acknowledge support from GianCarlo Moschini for providing access to the National Establishment Time Series database. We also thank Jonathan Colmer, Robert Elliott, Ralf Martin, and conference participants in 2013 AAEA meeting at Washington, DC, the United States, and 2014 WCERE at Istanbul, Turkey.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. The PSM technique has been extensively used to identify the causal effects of exports on firm size and productivity growth (Girma, Greenaway, and Kneller 2003 2004; Loecker 2007; Wagner 2002). In addition, List et al. (2003) employ this technique to identify the effects of environmental regulation on manufacturing plant birth.

2. The scale effect measures the increase in emissions due to the scale-up of economy. The technique effect refers to lower pollution as a result of the improvement in pollution abatement technologies. The composition effect explains the mixed results of changing shares of dirty good on pollution.

3. Holladay (2016) employs the same data set on plants’ economic characteristics as we use in this paper. He matches this data with plants’ toxic releases, while we match the data with plants’ criteria air emissions.

4. See Becker and Ichino (2002) for a detailed discussion.

5. For facility-level pollution data, we only have their reports for years 2002, 2005, and 2008. For the remaining facility economic characteristics, and county and industry characteristics, the data cover the entire period between 2000 and 2008.

6. Some major caveats for this NEI data are summarized in the Appendix; for example, duplicate emission data in 2005.

7. According to the EPA technical document, emission data for filterable and condensable components of particulate matter are incomplete through sample years, and, hence, are not suggested to use in any aggregate level.

8. The NETS data have been utilized to study issues related to job creations and destructions, business relocation, and business ownership (Kolko and Neumark 2008 2010; Neumark, Wall, and Zhang 2011). Neumark, Wall, and Zhang (2011) provide a detailed description of the NETS and an assessment of the quality of the NETS database along many dimensions. One dimension related to our study is the estimated data versus actual data regarding employment size. In our study, this problem is not critical, because about 90% of employment data have indicators suggesting actual data.

9. According to IHS Global Services, US seaborne trade with the rest of the world accounts for 78.05% by volume (millions of metric tons) in 2008.

10. The ‘Haversine’ formula calculates the great-circle distance between two points, that is, the shortest distance over the earth's surface.

11. As defined in Greenstone (2002), an industry is designated as a dirty emitter of a pollutant if it accounts for at least 7% of industrial sector emissions. See Table A2 (Online Supplemental data) of the Annual Industrial Sector Pollutant Release by Industry in Greenstone (2002) for details.

12. The fraction of observations with annual emissions less than 0.001 tons is as follows: 6.6% for SO₂, 0.46% for CO, 0.16% for O₃, and 1.17% for TSPs.

13. It is possible one treated unit in the east coast might be accidently paired with another control unit in the west coast. These types of matched pairs could bias the estimated ATET results, due to the confounding geographic location-specific unobservable, such as state-level environmental regulations, natural geographic advantage of shipping products abroad, or county-level pollutant nonattainment status. To remove the unobservable, we further restrict the matched exporters and non-exporters from the same US state and the same year, or the same county. All these matching results along with alternative matching estimators including strata and kernel matching are presented in Tables A1--A4 in the Online Appendix. The price for this restricted matching procedure is a reduced sample size and number of matched pairs, and, hence, less statistically significant estimators.

14. For robustness checks, we conduct the OLS regressions, using alternative measures of TETs at the matched-pairs level from different matching estimators and matching restrictions. Main conclusions are robust to these alternative plant-level TET estimates. These OLS results are available upon request.

Additional information

Funding

National Natural Science Fund of China [grant number 71603191].