Foreign production and the environment: does the type of FDI matter?

ABSTRACT

We examine the relationship between foreign ownership and the environmental performance of firms. We make a distinction between export-oriented and horizontal multinationals as they have different motivations and firm characteristics. Horizontal foreign direct investment substitutes for exports when trade costs are high, while export-oriented vertical multinationals geographically separate the stages of production primarily to exploit production cost differences across countries. Theoretically, it is not clear which type of foreign direct investment is dirtier. In this paper, we use microdata from Chile and find that export-oriented foreign firms have lower emission intensity than horizontal affiliates and domestic firms.

KEYWORDS:

• Emission intensity
• export status
• foreign ownership

JEL CLASSIFICATION:

• F18
• Q56.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1. This is by no means an exhaustive list. See (Cole, Elliott, and Zhang 2017) for a detailed survey.

2. Either way, these firms are likely to be motivated by cost differences. In reality, only a small number of firms are strictly vertical or horizontal, while the majority of multinationals engage in complex integration strategies (Yeaple 2003; Feinberg and Keane 2006).

3. This dataset is considered of high quality and has been used widely in the trade and productivity estimation literature. See, for example, Pavcnic (2002), Levinsohn and Petrin (2003), Oberfield (2013), and Ackerberg, Caves, and Frazer (2015) among many others.

4. Industries are based on International Standard Industrial Classifications (ISIC) Rev.2. Chile is divided into 15 geographic regions.

5. Some empirical studies employing the same data set as we do use the number of blue-collar and white-collar workers (or the wage bill for these) as free variables, and add fuels as an additional proxy. However, there are a lot of zeros for these variables, which result in reduced sample size when the natural logarithm of the variables are taken.

6. These cases are likely to be a result of data misreport. However, we have fewer than 300 such cases, which corresponds to 0.5% of total observations.

7. When we take the ratio of white-collar to blue-collar workers to calculate skill intensity, we add 1 to the denominator since a significant number of plants report zero for the number of blue- or white-collar workers.

8. Our results are robust to different ownership thresholds. In particular, we get qualitatively similar results if we consider only the firms for which the foreign ownership share is at least 50% as foreign.

9. Source: United Nations Conference on Trade and Development (UNCTAD).

10. The aggregate data do not have details about the activities of affiliates, such as export status. Thus, we cannot pinpoint their motivations. Our microdata, on the other hand, have details about firms’ export status and foreign ownership but do not have any information regarding the source countries.

11. We initially considered adding the total number of workers as a control for plant size. However, it is highly correlated with other variables, especially with TFP, and creates multicollinearity. Although our main results do not change, the standard errors get inflated significantly.

12. We report under-identification and weak identification test statistics in instrumental variable regressions throughout the paper. Our regressions are just identified, so we do not report an overidentification statistic. We report p-values for the under-identification test. We suggest readers to use 10 as a rough critical value for the weak identification test (Baum, Schaffer, and Stillman 2007).

13. We thank the anonymous referee for suggesting these.

14. The under-identification and weak identification test results cast some doubt about the validity of this instrument. The null hypothesis that the regression equation is under-identified cannot be rejected at 1% significance level. Also, the test statistic for weak identification is 3.11, which is less than the rule of thumb value of 10.