Foreign Direct Investment, Pollution, and the Environmental Quality: A Model with Empirical Evidence From the Chinese Regions

Abstract

In this study, within the framework of the Environmental Kuznets Curve (EKC), we empirically investigate the effects of regional openness to foreign direct investment (FDI) and regional economic growth on pollution emission across the Chinese provinces. Our analysis shows FDI contributes to more serious pollution emission, where the effect of the former on the latter is realized through the former’s impacts on the input of natural resources or the industry mix, either of which is associated with the level of total factor productivity. Our analysis also shows that with the continuous growth of output and per capita output, pollution emission and pollution emission intensity would both first rise and then fall, which lends support to the EKC hypothesis.

KEYWORDS:

• foreign direct investment
• environment
• pollution emission
• Environmental Kuznets Curve
• economic growth

Notes

¹ FDI inflows and foreign trade in China are closely linked. Foreign Invested Enterprises (FIEs) generally account for over 50% of China’s exports and over 60% of China’s imports. See, for example, Whalley and Xin (2010) for a recent discussion of the relationship between China’s FDI inflows and its foreign trade.

² As Equation (7), our baseline theoretical model, is quite vacuous in the sense that it provides little information as to the functional form of the involved variables, we have to base our regression specification in (8) more or less on intuitive reasoning, with the quadratic terms included in the equation to take account of potential nonlinear relations.

³ In Equation (8), total regional levels of the variables are used instead of regional per capita (per worker) levels. This does not lead to problems associated with estimation biases because all of the variables are entered in logs and labor (in logs) is explicitly included (controlled for) in the regression equation.

⁴ The variable associated with natural resources is dropped from the regression specification for the reason of data unavailability.

⁵ To fix ideas, suppose for the moment the industry mix is held fixed.

⁶ This is, again, because output and the various inputs are controlled for in the regression specification. To fix ideas, we can suppose for the moment the input of natural resources is held fixed.

⁷ All three mechanisms may be related to the “income effect” discussed earlier.

⁸ The regions include provinces, ethnic minority autonomous regions, and provincial-level municipalities. Owing to missing data, three regions, Tibet, Chongqing, and Hainan are excluded from our sample.

⁹ The choice of this proxy variable is based on considerations regarding data availability and data consistency.

¹⁰ A Hausman test justifies this in the current case.

¹¹ Both the FE and FD methods have the desirable feature of explicitly controlling for the unobserved province effects. However, the choice between FE and FD hinges on the assumptions about the idiosyncratic errors. Generally, the FE estimator is more efficient if the errors are serially uncorrelated while the FD estimator is more when the errors follow a random walk. See, for example, Wooldridge (2001).

¹² Statistic significance mentioned here in this analysis is at the usual 5% level unless otherwise stated.