FDI performance: a stochastic frontier analysis of location and variance determinants

Abstract

The opening up process of the eastern European countries was marked by greater integration of FDI with their western neighbouring countries. Using the single-step ML approach to stochastic frontier analysis, the location and variance determinants of FDI are estimated using the knowledge capital (KK) model framework. The findings, based on a panel of bilateral FDI stocks from 10 western to 10 eastern European countries over the 1996–2007 period, suggest FDI is determined by both horizontal and vertical motives while the process of liberalization and infrastructural developments significantly reduces the variance of FDI. In using a stochastic frontier specification of the KK model, the efficiency of FDI performance is identified relative to maximum levels. The bilateral efficiency scores suggest a mixed performance, indicating scope to improve the efficiency of FDI.

Keywords:

• knowledge capital model
• stochastic frontier analysis
• FDI performance

JEL Classification:

• C33
• F21
• P27

Notes

¹ Investment projects often involve a large initial outlay of capital – in purchasing a factory unit, for example – and may incur smaller expenditures thereafter, implying FDI data can be ‘lumpy’ in its range between large values one year and low or zero values the next. Missing values suggest no FDI takes place while disinvestment (negative values) can occur if a company divests its subsidiary of assets for more productive use elsewhere. Note that FDI stock data help iron out the irregular patterns of FDI flows.

² National firms can also arise if the country is large and is skilled labour abundant; if trade costs are low and countries are similar in size and in relative endowments; or if there are high investment barriers in the foreign country.

³ SFA has also been applied to other contexts. For example, Mosheim and Knox Lovell (2009) analyse dairy farm performance; Park and Davis (2011) evaluate human resource practices in food retailing; while Obeng (2013) estimates technical efficiency for public transit systems.

⁴ Data limitations (missing and zero values for FDI) restrict the sample size to 10 western and 10 eastern European countries; these countries represent the main parents and hosts of FDI into the region. In a conventional setting, missing bias is usually accounted for in one of three ways. First and most frequent, missing observations are simply discarded. Second, missing observations are treated as zeros (e.g. see Carr et al., 2001, who apply a Tobit procedure to the expanded set of observations). Finally, and more recently, the Heckman two-stage selection procedure has been employed as a correction method for sample selection (Razin et al., 2004). The Tobit and Heckman procedures, however, are not compatible with SFA estimation, hence the first option is used and the missing data are dropped. Moreover, to obtain a measure of FDI performance using SFA, positive values for FDI are required. At any rate, missing data for FDI stocks are much fewer than for FDI flows, accounting for just one-fifth of observations in the data set.

⁵ To prevent loss of information, period averages substitute for missing data before 1999 and secondary school enrolment ratios substitute for missing German data.

⁶ The use of index variables compiled from survey data to rank a country’s degree of trade openness is not unusual in the existing empirical literature (e.g. see Brainard, 1997; Carr et al., 2001; Blonigen et al., 2003). Indeed, Edwards (1998) points out that in capturing different aspects of trade policy a composite index can lead to efficiency gains when compared with separately introducing into an equation the various aspects of trade policy as independent variables. Using an index of distortions in international trade from the Heritage Foundation as one of nine indexes of openness in a growth model, Edwards (1998) finds no discernible differences in the results.

⁷ Outward FDI between the 10 western and the 10 eastern European countries over the period 1996–2007 is characterized by 241 missing values, 10 zero values and 7 negative values, thus reducing the potential number of FDI observations in the panel data set from 1200 to 942.

⁸ Dropping the normality assumption from the null hypothesis does not alter this result; the F-statistic is 25.68 and the ${\chi }^2$ (iid) is 25.05.

⁹ To an extent, these effects are also picked up by the host country cost indexes, although not significantly.