Sample and Descriptive Statistics
Many existing studies on FDI flows are based on country-level datasets, which say little about the strategies of firms and about the choice of ownership structures in particular (e.g., Bevan & Estrin, 2004; Globerman & Shapiro, 2003; Henisz, 2000; Kaufmann & Wei, 1999; Merlevede & Schoors, 2009; for criticism of this approach, see Wu et al., 2009). In contrast, and in line with Hines (1995) and Javorcik and Wei (2009), we use matched information on foreign parent firms and their foreign affiliates in 16 CEE countries (Bosnia and Herzegovina, Bulgaria, Croatia, the Czech Republic, Estonia, Hungary, Latvia, Lithuania, Montenegro, Poland, Romania, Russia, Serbia, Slovakia, Slovenia, and Ukraine) for the period 2003–2011. The dataset has been drawn from ORBIS,2 which has been widely used (e.g., Driffield, Mickiewicz, & Temouri, 2013; Lumineau & Malhotra, 2011; Temouri, Driffield, & Higon, 2008) and includes an unbalanced panel of, on average, 184,640 firms in CEE countries over 2003–2011. Only 6.6 percent of all firms have foreign investment and the remaining vast majority of the firms in the sample have no foreign investment.
As our panel is unbalanced, the number of firms in the sample period 2003–2011 differs: each firm may be in the panel for the entire sample period or only a few years, depending on its date of incorporation, its potential exit from the market, as well as its reporting requirements (e.g., very small firms are usually exempt from filing detailed firm information).
Thus, in terms of the sampling strategy, we started off with the entire population of firms in the 16 CEE countries as reported in ORBIS, which subsequently was reduced to firms for which we had information on the key variables in our analysis. We only included firms that showed ownership information (local or foreign), sales, employment, intangible and tangible fixed assets and material inputs, and other key variables that are crucial for the estimation of total factor productivity (TFP) in the subsequent analysis.3 Moreover, as we have the identity of the foreign partner in ORBIS, we were able to match the financial information of the foreign parent firm to the affiliate in the transition countries. This results in a sample of 4,137 affiliates, for which we have information on both affiliate and foreign parent firms from 43 countries of origin.
The percentage of firms with foreign presence differs across countries, and is higher for countries where the institutional environment is stronger. This is illustrated in Figure 1, where the horizontal axis corresponds to the corruption index (based on the International Country Risk Guide data, but with the sign reversed, so higher values represent more extensive corruption), and the vertical axis represents the percentage of firms in a given country that have some foreign ownership presence. In the lower right-hand corner, we find Russia and Ukraine, with high perceived corruption and less than 5 percent of firms in our sample having any foreign presence. In contrast, Estonia and Hungary are the two countries with the lowest perceived corruption and about 15 percent of firms with foreign share.
In turn, for firms with foreign investment, the average percentage of equity held by the foreign partner is very high, ranging from 55 percent in Montenegro to 88 percent in the Czech Republic closely followed by Poland (86 percent), Slovakia (86 percent), Estonia (80 percent), Latvia (80 percent), and Hungary (79 percent). Some 60 percent of foreign affiliates have at least 90 percent of their equity held by the foreign parent, while 48 percent are wholly owned. This is in line with the findings of Mani et al. (2007), who report high equity shares retained by parent firms for their sample of Japanese investments in 38 countries. The average percentage held by a local partner is lower where the institutional environment of the host country is stronger. This is illustrated in Figure 2. In the upper right-hand corner are Moldova, Montenegro, Russia, and Ukraine, being the countries with the highest perceived corruption and the highest mean local ownership. In the lower left-hand corner are Estonia, Slovenia, and Hungary, with the lowest perceived corruption; in these countries, the mean local partner share in equity is around 10 percent.
We construct our dependent variable in the following way. First we identify CEE companies where the foreign owner is the largest shareholder. Next we look for the largest local shareholder, if any. The share of locally held equity becomes our dependent variable, with mean country values illustrated in Figure 2. There are only a handful of cases where the second largest shareholder is also a foreign owner and, as this does not affect our results, we retain these observations.
We include factors which may affect firms' ownership structure. First, for every host firm, we have the share of intangible assets in total assets (similar to Qian & Strahan, 2007, who use the ratio of tangible to total assets instead; and, for example, Barth & Kasznik, 1999). This proxies for a firm's knowledge capabilities, and includes formation expenses, research expenses, goodwill, development expenses, and all other expenses with a long-term effect. As argued above, this is assumed to increase the need for closer monitoring and the difficulty in defining resource contribution of local and foreign partners via contractual obligations, and therefore we relate this variable to hypothesis 2. We also control for host firm size based on assets, following Pan (1996), as larger size implies a higher absolute investment risk for the parent firm.
Next, we include host firm TFP derived as the residual of the production function using the Levinsohn and Petrin (2003) semi-parametric approach, which is an econometric technique to address endogeneity in inputs.4 Furthermore, we control for market share of a host firm, as product market competition may be a good substitute for incentivizing the stakeholders (Loredo & Suarez, 2000; Randøy & Jenssen, 2004). Consistent with this, we use a logarithm of market share in sales in a given sector (Cooper, 1993) and expect it to be associated with higher share of local ownership. As we have matched host firms with their foreign parent, we can also control for the share of intangible assets in the parent firm. Similar to what we use for the host firm, we utilize the share of intangible assets in total assets. We also take into account the size of the parent firm.
As in Judge et al. (2008), all our firm-level variables are lagged one year to alleviate simultaneity bias. We also include full sets of random host country-year effects and industry and country of origin dummies to control for industry-, country-, and time-specific factors that may affect a firm's ownership structure.
Empirical Model and Estimation
The essential problem here is to model the ownership structure of foreign affiliates allowing for the fact that most firms in a given location do not attract foreign investment. In order to do this, we employ the Wooldridge (1995) estimator. Suppose in a given year the foreign ownership in the i-th host firm operating in the j-th sector is denoted by a variable Fijc* determined as follows:
For a given year t, t = 2003, … , 2011, we use this ownership information Fijc* to construct the following binary foreign entry variable Fijc, indicating whether the i-th host firm operating in the j-th sector in country c has been successful in attracting foreign investment:
Equation (1) thus provides an underlying structural model for the determination of the probability of foreign investment in a host firm. This is similar in spirit to the analysis of Agarwal and Ramaswami (1992) for example. The X vector is the set of one-period lagged explanatory variables explaining this probability, namely: TFP; firm size; intangible to total assets ratio; and the volume of cash available to the firm. Given the multi-level data at our disposal, we also allow for sector-specific (γj) and country-specific (γc) effects that capture common unobserved shocks at the relevant level. The remaining errors are included in the independently and identically distributed error term ε. Estimation of equation (1) for each year t in the sample allows us to determine the inverse Mill's ratios λit for t = 2003, … , 2011.
After selecting the firms with some foreign ownership (F = 1), we estimate a second model to determine the level of ownership of the local partner Dijct in the i-th host firm with foreign ownership in sector j, country c and year t as follows:
where Z is the set of host firm characteristics, incorporating a subset of X from (1). H captures the foreign MNE characteristics, while C refers to the country-level characteristics, namely, measures of the institutional environment. In equation (2), industry-specific fixed effects are denoted by δj and year-specific fixed effects by δt. Note that the λit term is the so-called inverse-Mills ratio obtained from estimating equation (1) above, and used to control for potential selectivity bias. The remaining errors are included in the independently and identically distributed error term ν.
Following Amiti and Wakelin (2003), we argue that characteristics of the host firms (lagged values of firm size – medium and large, TFP as well as cash flow) play a crucial role in the probability of foreign investment. Note, however, that the cash variable is only included in equation (1) as cash availability may be driving investment decisions; thus it serves as an exclusion restriction for equation (2).