The Location Choice of Foreign Direct Investment Within Turkey: An Empirical Analysis

Abstract

This paper investigates the location choice of foreign direct investment (FDI) in Turkey over the period 1996–2003. Using FDI data at the provincial level and negative binomial model, the empirical evidence confirms that agglomeration economies and information costs are the most important determinants of FDI location in Turkey. Specifically, both foreign and domestic agglomerations and in particular urbanization economies strongly affect the location decision of foreign investors. The results also suggest that foreign investors are attracted to provinces with fast growing market, more educated labour force, high density of road network, greater public investment, milder climate, larger area and better quality of life. On the other hand, the empirical findings show that wage, productivity, labour availability, unionization, sea and air transportation, free zones and instability have no significant impacts on the location decision.

Acknowledgements

I am very grateful to Niv Farago for checking my English and anonymous referees for their comments on an earlier version of the manuscript. Any errors are, of course, my own.

Notes

Some empirical studies have examined location decisions of foreign firms choosing Turkey as an investment location (Tatoğlu & Glaister, 1998; Coşkun, 2001; Erdal & Tatoğlu, 2002). But, those studies, which deal with FDI location in Turkey, have approached FDI only at aggregate and national levels.

Because CLM limits observations to the actual choices made by investors (Ó'hUallacháin & Reid, 1997, p. 409).

Due to the scope of the paper, FDI inflows to Turkey from 2004 to present are excluded from the paper.

For comprehensive details of count models and the regression-based test of the Poisson restriction, see Cameron and Trivedi (1998) and Greene (2003).

We also tried to use additional/alternative variables such as urbanization (instead of population density), distance from Istanbul (instead of principal city effect) and ethnicity (instead of political instability), which are not reported in Tables 1–3. However, these variables were either insignificant or highly correlated with each other and/or some other variables, leading to serious multicollinearity problems. Therefore, we decided to remove these variables from the models.

Beaverstock et al. (1999) classified the major cities of the world at three levels (alpha, beta and gamma world cities) in terms of the world city status by adopting Sassen's approach. In their list, Istanbul has been placed as a Gamma world city (third level), which means that Istanbul is a global service centre in attracting major multinational corporations in advertising, banking and legal services.

The state of emergency regions, called OHAL in Turkish, created by the Turkish government for the security reason as temporary regional governorship, cover 14 provinces where the clashes took place. These 14 provinces are Adiyaman, Ağri, Batman, Bingöl, Bitlis, Diyarbakır, Elazığ, Hakkari, Mardin, Muş, Siirt, Tunceli, Şırnak and Van (Figure 2).

For example, domestic agglomeration and labour quality are not significant, and the quality of life displays wrong sign.

The removal of agglomeration variables in model 6 decreases the fit of the model as indicated by the log likelihood. The log likelihood starts at −169.5 in the initial model and continues in each successive column, −176.1, −176.6, −175.9 and −177.9, respectively, and ultimately reaching −180.9. This result implies that agglomeration economies have significant effects on FDI location.

Not surprisingly, the inclusion of population density variable affects the statistical significance of the Istanbul variable, and thus this variable becomes statistically significant only at the 10% level.

Adding any of the agglomeration variables, except the service, makes the road variable statistically insignificant across models.

The inclusion of population density makes the climate variable statistically insignificant.

By including the domestic agglomeration into the models, the land variable becomes statistically insignificant.