Assessing the impact of bilateral trade policy measures on bilateral FDI using the structural gravity modelling

Since the bilateral impact of trade policy variables on bilateral FDI stocks is the main interest here, this article follows the strand of the literature on gravity modelling controlling for multilateral resistances (Anderson and van Wincoop, 2003). Therefore, in order to control for country-level time-variant variables like GDP, country-time fixed effects are used. Therefore, host-time, and home-time fixed effects are used in addition to country-pair fixed effects controlling for any time-invariant relationship between the two countries such as distance, colonial relationship, language and sharing borders. In this FDI model, with these fixed effects, the only correlating variation between the FDI stocks and trade policy measures would be what is left between i and j that changes over time.
Bilateral FDI stocks cover many zero values. In fact, 27% of the sample of estimations includes zero FDI stocks. Changing the FDI stocks to logarithms will eliminate 27% of the sample causing biased estimations. In order to have unbiased estimation results, the PPML estimator is used following Silva and Tenreyro. The PPML estimation procedure including multidimensional fixed effects is run using the Stata package implemented by Correia et al., (2019). Robust standard errors are calculated to control for the heteroscedasticity in the error term.

Stata codes ran for the estimation results of the manuscript are written in the Stata do.do file for the data_estimation.dta data file. These files are compressed in the following shared .rar file.

References:
• Anderson, J. E., & van Wincoop, E. (2003). Gravity with Gravitas: A Solution to the Border Puzzle. American Economic Review, 93(1), 170-192.
• Silva, J. S., & Tenreyro, S. (2006). The log of gravity. The Review of Economics and statistics, 88(4), 641-658.
• Correia, S., Guimarães, P., & Zylkin, T. (2019). PPMLHDFE: Fast poisson estimation with high-dimensional fixed effects. arXiv preprint arXiv:1903.01690.