Total Factor Productivity, Foreign Direct Investment, and Entry Barriers in the Chinese Automotive Industry

ABSTRACT

We explore three questions on foreign direct investment (FDI): (1) What are the differences in entry barriers for foreign, public, and private investors? (2) What are the effects of past productivity levels on future foreign direct investment (FDI) decisions? (3) What is the effect of equity structure on future total factor productivity (TFP) levels? The empirical results based on a monopolistic competition model and using a firm-level data set from the Chinese automobile industry suggest that foreign investors face higher entry barriers and react stronger to past TFP levels. FDI is also found to improve future TFP more than other forms of investment. Finally, World Trade Organization (WTO) accession is found to reduce entry barriers for foreign and domestic private investors while increasing entry barriers for public investors.

KEY WORDS:

• auto-manufacturing industry
• China
• entry barriers
• foreign direct investment
• total factor productivity

Notes

¹. Standard Industrial Code (SIC) 3711 includes firms “primarily engaged in manufacturing or assembling complete passenger automobiles, trucks, commercial cars and buses, and special purpose motor vehicles which are for highway use” (United States Department of Labor Occupational Safety and Health Administration, 2014, available at https://www.osha.gov/pls/imis/sic_manual.display?id=840&;tab=description).

². The surveys report two additional categories of investors (with comparatively small weights), which are those from Macau and Hong Kong, and the collective ownership variable belonging to towns and villages. Given the difficulty fitting these two categories into either domestic or foreign investors or public investors, we decided to leave them out of our empirical analysis. As a result, the sums in Figure 1 do not add up to 100% percent.

³. The number of observations included in the regression analysis is lower due to the use of a dynamic specification and variable lags as instruments in the GMM estimation. Furthermore, the difference between the number of observations for the aggregate TFP measure and those for subgroups arises as the TFP averages for the latter refer to cases with positive observations.

⁴. The estimation is done in Stata 12.1 using Levinsohn and Petrin (2003) methodology.

⁵. Here and in later specifications, we add one to the level of Investment given the presence of zero levels of foreign, public, and (domestic) private stock of equity ownership.

⁶. In the literature, ownership types are usually classified by either the majority equity shareholders or a benchmark threshold level of ownership. Since we have more than two types of ownership and we are interested in exploring the relationship among entry costs, TFP, and equity structure, we do not divide firms according to an ex post benchmark level of equity ownership and include them all in a single equation. The commonly used dummy variable approach in the literature requires the choice of a cutoff point for each type of investor, such as 10 percent. In that case, it is difficult to come up with a classification system to designate a firm that receives, for example, 10 percent foreign investment, 10 percent private investment, and 80 percent public investment. Moreover, the dummy variables will capture time-variant shifts in ownership structure only if these are above or below the cutoff points. For example, in the case of a 10 percent threshold rule, if a firm increases its foreign equity share from 11 percent to 12 percent, this won’t be reflected in the dummy variable measurement. If we use a continuous variable at levels and include them in a single equation, then the issue will be the correct identification as all three types of investment can increase or decrease by the same amount at the same time. Furthermore, the specification of entry cost variable poses a serious challenge as it is practically impossible in a reduced form setting to estimate three separate entry costs for each type of investors all at the same time. Therefore, in all our estimations, we use the level of equity investment by the three types of investors to identify a firm’s having access to foreign, public, and private equity capital and then run three separate regressions for each.

⁷. Although there is sufficient variation in this variable, which ranges between sixty-two and seventy with a standard deviation of 3.19, it is cross-section invariant. However, our TFP measure already accounts for the effects of macroeconomic changes. Furthermore, the ICRG measure is intended to capture the reaction of foreign, public, and private investors to a common time-variant shock to institutional structure.

⁸. The numbers of observations in each regression analysis are equal across three groups as they are based on the same firms with a level equity ownership being equal to or greater than zero.

⁹. Different from the standard gravity model, Eaton and Tamura (1994) assume nonhomogeneous relation between the dependent and independent variables. Particularly, for dependent variable V, the logarithm of a_v+V_it, where a_v is an intercept variable, is “linear homogenous in the logarithm of independent variables” (Eaton and Tamura 1994, 490). As their dependent variables are censored below zero, a positive value of a_v implies a minimum threshold value (entry cost) before positive values of V are observed.

¹⁰. Estimating the TFP part of Equation (16) alone would cause several problems, including (1) an identification issue as we need to identify α₂ and γ₀ simultaneously, using the variation in investment; (2) an endogeneity problem regarding the realized, as opposed to the optimal, level of investments, as well as the TFP effect.