Abstract
In this chapter we go on to test the hypotheses about FDI in the CEECs empirically. We employed panel data models to the CEE members and candidate countries of the EU during the period 1992–2015. The countries covered were Albania, Bosnia and Herzegovina, Bulgaria, Croatia, the Czech Republic, Estonia, Georgia, Hungary, Latvia, Lithuania, Moldova, Montenegro, Poland, Romania, Serbia, Slovakia, Slovenia, TFYR Macedonia, and Ukraine.
The first empirical part of the chapter describes the factors of the empirical model. In the second part, its robustness is checked, and in the third results are discussed. In the fourth section, we made an attempt to identify the main differences of FDI’s impact on different sub-regions and time periods.
The previous chapter is aimed at the formulation of a theoretical model that we will empirically test in this chapter. We also present here the estimation strategy and the explanation of empirical results. Thus, the main objective of this chapter is to estimate the factors that affected FDI location in CEE.
4.1 Factors
The location of FDI is closely related to a country’s comparative advantage, which in turn affects a multinational’s expected profits in a host country.
The components of host country location motives could be broadly classified into two types: first, there are traditional factors, which mainly consist of market potential, production costs and macroeconomic stability; second, there exists an integration effect, which implies access to neighbouring markets. In the previous chapter, we developed an empirical model that combines traditional FDI determinants and integration factors that are expected to play an important role in attracting foreign investors.
In the empirical literature, there is no consensus on the factors of FDI that need to be included in an empirical model. In addition, determinants of FDI may vary across characteristics of industry, production factor intensity, and the nature and source of investment. However, some variables such as market size, labour costs, are represented as the traditional factors, are generally incorporated in the empirical models (see Sect. 2.3). The preferences for other less prominent determinants may vary from one empirical model to another (see Appendix B, Table 2.6). Thus, the criteria for a variable choice include ease of data availability, sound theoretical justifications, and the variable’s robustness in the empirical FDI literature. All additional variables we include progressively into our empirical model.
In the previous chapter we obtained an equation that can be estimated empirically:
where β are the estimated coefficients, sjT represents the host country size, ljT is the labour cost in the location country, ∑sk is the neighbouring market size, tjT stays for internal trade barriers and/or international trade openness.
4.1.1 Dependent Variable
Our dependent variable, fjT, is the aggregate annual FDI inflows in a country j. Because of data limitation, we do not differentiate between either the countries of origin or the location sectors or the types of FDI.
We take FDI inflows rather than accumulated FDI inward stocks because stocks are almost time invariant. That is, as long as FDI stocks are relatively large, annual changes to these stocks are likely to be negligible. Moreover, significant stocks result from past investment decisions. They, however, do not depict a present attractiveness of a country, causing an empirical model to have difficulty identifying the determinants of the dependent variable. In addition, the calculation of FDI stocks is often not homogeneous across countries (Globerman and Shapiro 2002).
We also use FDI inflows instead of net FDI flows. Since we do not observe the country of origin, we are not able to distinguish whether, for example, a positive change in net FDI flows is driven by increased inflows or reduced outflows. This also makes it difficult to select the determinants, because the inward and outward flows depend on different conditions. Moreover, the results are difficult to interpret.
In our database, we can distinguish between missing data and a flow of zero. If data is indeed missing in the original dataset, we drop missing observations from the dataset. The usage of logarithms, however, generates problems with zero and negative FDI inflows. In the literature, there are three main approaches for dealing with negative and zero values of FDI inflows: either all negative or zero values of FDI are deleted from the analysis; or they are replaced by a very small positive number; or two-stage estimation (Heckman model (Heckman 1979)) is used (it is complicated procedure, needs additional assumptions and is more effective in the case of the large number of zero flows (stocks)—for example, trade flows) (Bos and van de Laar 2004).
For negative and zero observations, we make a transformation widely used in empirical literature: ln (|FDIjt| + 1). Thus, all zero flows will be zero. There are only 5 negative cases among 440 observations; we suppose, thus, that they will not significantly change our estimation results.
4.1.2 Market-Seeking Factors
Market-related factors are generally considered the most significant determining factors of FDI. There are several reasons behind this. Firms are always seeking new market opportunities for their products. Once the Iron Curtain fell, it led to the reintegration into the world economy of a relatively large market made up of 370 million inhabitants.
Firms are also attracted by current demand and a relatively low competition in the market. Thus, an investment location in CEE could be a strategic decision for multinationals looking to maintain or increase their international strategic position. In addition, the big market size allows companies to achieve economies of scale and to reach optimum scale. It also leaves room for new factories and avoids a fall of prices when total industrial productive capacity goes up.
The importance of the local market size, sjT, can explain the high FDI in Poland, Romania, Ukraine and Serbia (see Figs. 2.6 and 2.9). Among all studied CEE countries, they attracted the biggest amount of investments. However, all large countries except Poland, despite their potential, attracted fewer investments because of their “stop and go transition” (Fig. 4.1). This confirms our assumption that the market size is not the single factor of FDI.
In a survey-based study, Lankes and Venables (1996) found evidence that a majority of firms that invested in transitional economies were looking for new market opportunities. The empirical evidence of the importance of market-related factors is also extensive: Altomonte (1998), Clausing and Dorobantu (2005), Carstensen and Toubal (2004). Meyer (1998) as well as Brenton (1999), Kinoshita and Campos (2003) and Faeth (2009) found that market size is the primary determinant for foreign direct investment in the CEE region.
In contrast, Holland and Pain (1998) and Asiedu (2002) declaired growth and market size as insignificant determinants of FDI flow. Indeed, market size has different implications for FDI inflows in accordance with its motive. For instance, it might be crucially important for FDI stimulated by horizontal motives, while it might offer little incentive for vertical FDI. Furthermore, a high rate of growth in the host country’s market identifies a good development for the future, which suggests that a high growth rate in the host country would promote FDI inflows.
In the previous literature, several proxies were employed to measure market factors. Among them we can find total GDP, GDP per capita and population. The results of the main papers that explicitly consider market factors are presented in Appendix B, Table 2.6.
In our book, we employ GDP, which represents the host country’s economic conditions and the potential demand for their output. It is an important element in FDI decision making. The figures are drawn from UNCTAD (2016c) databases. The expected relationship between the size of the market and FDI is positive.
4.1.3 Resource-Seeking Factors
Labour and natural resources are the production factors which have been suggested as determinants of FDI into Central and Eastern Europe in a number of empirical studies.
The availability of low-cost skilled labour, ljT, is also one of the prime attractions for a multinational, enabling them to take advantage of lower production costs. The main idea behind is very simple: firms move different stages of production process to countries with lower costs. We use the nominal wage rate as a proxy for labour cost. Our data for average wages come from UNECE databases and from ILO (2016) and ILOSTAT (2016) Databases.
We would generally expect a negative sign of the coefficient (e.g. countries with lower labour costs would attract more FDI).
However, the expected effect of low-cost labour availability cannot be established apriori. CEE might have attracted investment due to its cheap labour but once the decision was taken to locate it there, finding the cheapest possible labour within a region already characterised by low wages might not be so important. The existent differences in labour cost terms between the studied countries are not significant enough to have a strong influence on the location choice. Czech Republic, Hungary, Croatia, Slovakia and Slovenia have the highest annual compensation per employee but they received the biggest amounts of investment, while the Western Balkans and Moldova and Ukraine, which registered cheaper labour costs, received less FDI (Fig. 4.2).
CEE still has low labour costs, compared to most of the countries in the Western Europe, however, they are higher than in many Asian countries that are more preferable for the location of the resource-seeking investments.
The previous results are also rather inconclusive. Thus, Lankes and Venables (1996), Clausing and Dorobantu (2005) find no statistically significant evidence of labour costs as a determinant for FDI. Holland and Pain (1998) presented evidence of the importance of relative labour costs for the location decision within Eastern Europe and Lansbury et al. (1996) found evidence of the importance of the relatively lower wages in Eastern Europe for re-orientation of investment from other low-wage regions in Southern Europe to Eastern Europe. When also taking into account the productivity of labour, as suggested by Lankes and Venables (1996), the evidence of the importance of labour as a determinant of FDI into CEE is more extensive. Carstensen and Toubal (2004) and Bevan and Estrin (2000) found significant evidence of the productivity-adjusted labour costs’ impact on FDI. As showed Bevan and Estrin (2000), not only an increase in productivity-adjusted labour deter FDI, but also the rate of growth of productivity-adjusted cost can be negatively correlated with the growth rate of FDI.
Kinoshita and Campos (2003) argue that the relationship between labour costs and inward FDI could be also positive because higher labour costs reflect a higher skill level. A more educated labour force can learn and adopt new technology faster and is generally more productive. Indeed, the influence of wage costs on FDI decisions varies among industries, depending on their factor combinations (labour or capital intensive) and investment motives (domestic or export market oriented) (Agarwal 1997). Some industries, such as computing, require high-skilled labour, which is often associated with high wages. Therefore the effect of labour costs on FDI might vary with the type of industry. This implies that the high-skilled labour associated with high level of wages might attract FDI in some industries instead of deterring it.
4.1.4 Integration Factors
It is natural that the sudden opening of the CEE market and its free access to the EU market,∑sk, attracted the interest of foreign investors. By locating itself in a host country with access to the EU market, an investing multinational firm gains not only an access to the local market of its host economy but importantly unlimited access to the market of 500 million consumers. Indeed, compared to other developing countries in Latin America and Asia and due to the fact that the trade barriers almost disappeared as a result of the FTAs and AAs, the dimension of the CEECs might be seen as insignificant; but their geographical and cultural proximity might have attracted a considerable number of multinationals. Immediately after opening up, the FDI inflows were low, but as the liberalisation of the market advanced the FDI inflows increased.
To measure the market potential of integration with the EU we estimate the coefficients of three variables. They are all calculated as the sum of the GDPs of all available for free-trade markets, depending on the agreements signed. They all correspond to the market potential defined in Sect. 3.4: the FTAs, AAs, and membership in the EU (see Figs. 4.3 and 4.4). The market size of the neighbouring markets has never been used in the empirical literature (see Appendix B). To estimate the effect of the integration dummy variables were commonly used. The exception are some theoretical studies (e.g. Altomonte 2007) that employed the alternative measure of market accessibility initially proposed by Harris (1954). He measured GDPs with weights equal to the inverse of the distance between the host and each partner. By doing so Harris (1954) included possible transportation costs that arise between distant FDI locations. Unfortunately, we can employ only aggregate data that do not identify the source of FDI.
We also employ three independent variables to capture the integration effect instead of just one because it enables different agreements to have different degrees of impact as well as some degree of variation in term of statistical significance, in case that the regional integration agreements have non-linear impact on FDI.
The first proxy, FTAMAjT, measures the size of the FTA market. It is the sum of the GDPs of all partner countries in the European FTAs (CEFTA, CEFTA 2006, and BAFTA).
The signing of the Association Agreement provides access to the EU market. That is why the AAMAjT variable measures the size of the EU’s GDP. During this time horizon the EU has absorbed new countries, that is why the size of the associated market has also increased in a number of participants (see Fig. 4.3).
Consequently, membership of the EU provides an access not only to all members of the EU but also to all countries that have trade and association agreements with the EU. This proxy sums up the GDPs of 57 countries (however, the amount varies depending on the agreements in force). Such huge markets as Turkey (FTA since 1995), Norway (FTA since 1994), Switzerland (FTA since 2002) and many other large markets worldwide (see Table 3.4) are among them. Thus, the membership in the EU provides for all member-countries an access to a greater amount of markets on preferable conditions.
Thus, market access variables include all possible relevant markets that are open for free trade according to the agreements signed. Such a framework is consistent with the emergence of integration strategies, namely the export-platforms, which includes the market potential of the associated countries. For the purposes of the book, a main working hypothesis can thus be derived: the greater the degree of trade integration, the larger the increase in each market’s potential and hence the higher the profits obtainable by production relocation through FDI.
There is also some empirical evidence that contiguity and proximity to the EU were important factors in observed trade and investment decisions. Benacek et al. (2000) suggested that national and regional market access was the primary factor that influenced potential investors, with market potential as another dominant factor (Benacek et al. 2000).
Trade costs are a very important determinant of FDI, and former studies take into account a variety of components of such costs including transport costs, distance, and trade policy barriers (see Table 2.6, Appendix B). To measure the degree of openness of an economy we employ the ratio of international trade to GDP. In the literature, this ratio is often used as the measure of openness of a country and is also often interpreted as a measure of trade restrictions. An alternative measure of the openness is tariff levels and revenues of duties on imports (see Table 2.6), but these data are scarce. Moreover, tariffs vary across industries. The tariffs also are different for different trade partners according to the bilateral and multilateral trade agreements signed. Finally, low tariffs do not always indicate a more open economy due to the presence of non-tariff barriers which are notoriously difficult to measure. An important element of the trade cost is the time it takes to ship products between plants. Proximity is a good variable to measure these costs. However, we do not observe sources of FDI, thus, we are not able to add these distance weights in our model.
The relation between trade openness and FDI may differ by the type of investment: FDI can substitute trade (in the case of tariff jumping), or stimulate exports (in the case of export-platforms). Holland and Pain (1998) found that trade openness should be not an important determinant for horizontal FDI, in contrast, export-oriented vertical FDI would be greatly affected export volume with positive re-action (Holland and Pain 1998). In the context of CEECs, the literature reveals a positive correlation between FDI and trade openness (see Appendix B). The countries that are more liberal in their trade approach tend to export more and this might attract foreign investors, especially ones which are export driven (Bevan and Estrin 2000). Resmini (2000) suggested that export-driven investors were mainly attracted by low labour costs (which give the possibility of reducing costs) and by the degree of openness. Kinoshita and Campos (2003) showed that trade openness is an important attractor for the less developed of the transition countries (CIS), but less relevant for the more developed ones. Most findings indicate that investors prefer countries with liberal trade regimes, located in regions with national free-trade arrangements. Therefore, we expect that openness should have a positive influence on FDI.
4.1.5 Efficiency-Seeking Factors
We include supplementary efficiency-seeking factors in our model. They are GDP growth rate, Gross Capital formation and agglomeration. These factors are used to measure macroeconomic stability and the economic effectiveness of a host country.
As anticipated, market growth (GDP growth rate) positively influences FDI. A country which has a stable macroeconomic condition with high and sustained GDP growth rates will receive more FDI inflows than a more volatile economy. Higher GDP growth indicates a potentially larger market and more promising prospects. It also implies better infrastructure, provides greater incentive for inward FDI and positively influences the business climate for inward FDI. Moreover, rapid growth may also give rise to the presence of economic rents that will encourage inward FDI (Globerman and Shapiro 1999).
We also include in our model Gross fixed capital formation (GCF). According to the World Bank definition, it includes land improvements (fences, ditches, drains, and so on); plants, machinery, and equipment purchases; and the construction of roads, railways, and the like, including schools, offices, hospitals, private residential dwellings, and commercial and industrial buildings (The World Bank 2016). Apparently, the increase in Gross capital formation results in the improvements of infrastructure which further attracts higher FDI inflows. However, the relationship between FDI and Capital Formation is not clear observable (Krkoska 2001). For example, privatisation attracts FDI into a country; however, there can be no improvements in the infrastructure and investment climate. Thus, the possible effect of GCF can be not observed in countries in transition (Krkoska 2001).
To measure the agglomeration effect, we include in our model lagged for one period FDI inflows. The use of this variable implies the presence of dynamic effects in the model, i.e., AR(1) process in the model. There are a few motives for including this proxy in our model. Firstly, new foreign investors follow the past investment decisions of former investors about where to invest. Secondly, FDI inflows include among others reinvested earnings, thus present inflows are bound with former investment inflows. Thus, Kinoshita and Campos (2003) confirmed that the agglomeration plays a crucial role in attracting FDI in a country.
To proxy the institutional effects of the integration we include EU dummy variables in our regression model. The recent accession of the CEECs into the EU has been suggested to be a major driver of the transformation of institutions within the region. The institutional type of explanation states that countries facing the prospect of accession to the European Union had a strong incentive to establish well-functioning democracies and legal institutions. The new member states had to introduce all institutions of the acquis communautaire, the joint institutions of the European Union. There was strong monitoring of the progress of accession candidates in introducing the required institutions.
European Union membership also opens the possibility of a country’s adopting the euro, which further harmonises the country’s macroeconomic policies with those of the rest of Europe. Both effects serve to reduce investors’ perceived level of country risk within CEE. By using the accession announcements made by the European Council as a measure of the accession progress of the candidate countries, Clausing and Dorobantu (2005) and Bevan and Estrin (2000) find evidence supporting the hypothesis that the EU accession process has had an impact on FDI into accession countries. The EU’s commitment to accept qualified candidate countries and the recent enlargement of the EU are suggested as major sources for reducing the perceived level of risk when investing in accession countries.
Thus, we estimate the effect of membership in CEFTA or BAFTA, the signing of the Association Agreement, entry into force of the Association Agreement, application for membership, granting of candidate status, the start of negotiations, the signing of the Accession Treaty with the EU, membership in the EU, and Euro area membership (all stages are described in Sect. 2.3). The value 1 indicates that a country was approved for a particular stage of the integration before the 1st of July in a particular year, and 0 displays that the stage has not been reached yet.
To capture the effect from announcements we also employ individual proxies because of the non-linear effect of the integration. This non-linear effect can arise when multinationals and other potential investors are not very sensitive to the progress in the integration or when progress in the EU accession from a specific stage to the next contains the use of FDI-friendly instruments, such as direct subsidies and corporate tax exemption and, hence, offsets the positive effect of EU enlargement (Iwasaki and Suganuma 2009).
We expect to observe a positive effect on FDI inflows because integration into the EU results in the improvement of institutions and investment climate.
4.1.6 Completed Model
The final overview of our empirical equation is:
Where β are coefficients, GDPjT and WAGEjT are traditional factors of FDI, FTAMAjT, AAMAjT, EUMAjT represent the markets size of the markets to which countries get access after signing FTAs, AAs and Accession Treaty correspondingly, and OPENjT is the overall openness of the economy. FDIjT − 1, GCFjT and GDPGRjT are additional macroeconomic factors. The description of the proxies, their sources and measures are summarised in Table 4.1.
The dataset is an unbalanced panel of 19 acceding countries over the 1992–2015 period. FDI in Central and Eastern Europe was practically non-existent before 1992 and data are only available through 2015.
All values are expressed in current US dollars. Our dependent variable is the natural logarithm of the FDI flow in millions of U.S. dollars to country j at time T. Some independent variables are also measured in logarithmic values (see Table 4.1). This has two advantages: such reduces the likelihood of outliers; the coefficients can be directly interpreted as elasticities and semi-elasticities.
We additionally calculate the effect of lagged for 1 year variables because there is a clear theoretical reason to expect that the effect of an explanatory variable influences the investment decisions only with a one-period lag.
There are several other explanatory variables that could be added to the model specification. Measures of institutional factors may be also important. However, we do not have enough data to measure the effects of institutions. That is why we estimate this effect by including EU accession dummies. In doing so we rely on Berglöf and Roland (1997) who highlighted the role the European Union played as the institutional anchor for transition economies from Central and Eastern Europe. Roland and Verdier (2003) showed how the prospect of admission to the European Union served as a coordination device to introduce the rule of law in CEECs (Berglof and Bolton 2002).
4.2 Robustness Check
4.2.1 Descriptive Statistics
Before proceeding the estimation of the panel data analysis, we provide descriptive statistics analysis. The outcomes of the analysis are summarized in Table 4.12. It gives the descriptive statistics both for the dependent and independent variables, namely number of the observations, means, standard deviations for each variable, as well as minimum and maximum values and rage of the variables, their skew, kurtosis and standard error.
Among descriptive statistics of explanatory variables, several points should be pointed out. The CEE markets are mostly small economies. The highest value of GDP (545795,70 million dollars) was obtained by Poland in 2014. However, Poland only had the 25th largest GDP in the world in 2016; other countries are even smaller (International Monetary Fund 2016).
The average wage in CEECs is equal to $543,19 which is much lower than in many developed European countries (see Fig. 4.5).
Gross average monthly wages by subregions and year, US dollars. Source: Author’s calculations based on the data from UNECE. Note: Some values are not available and added as zeroes
FDI inflows in CEE less integration effect. Source: Author’s calculations. Note: Based of the 2SLS FE estimations
Estimated effect of integration. Source: Author’s calculations. Note: Based of the 2SLS FE estimations
The lowest values were recorded in the early periods of transition: in Georgia in 1994 and 1995 with $8,5 and $9,6 average monthly wages; in 1992 average wages in Ukraine was $10,15. They were so low because in those most of the salaries were paid in coupons (National Statistic Office of Georgia; State Statistic Service of Ukraine). This payment pattern was common to all former members of the USSR.
Figures 4.9 and 4.10 in Appendix B present coefficients of pair correlation. All pair-correlation coefficients have acceptable values. Coefficients of correlation greater than 0,50 appear only between factors that represent the accession stages. Dummies FTAjT, AaforjT and EUjT are highly correlated (at 100%) with corresponding market access variables, FTAMAjT, AAMAjT, and EUMAjT. That is why we estimate these variables in different specifications to eliminate the problem of multicollinearity.
Correlation matrix. Source: Author’s calculations. Note: Signif. codes: 0 ‘***’ 0,001 ‘**’ 0,01 ‘*’ 0,05 ‘.’ 0,1 ‘ ’ 1
Pair correlations. Source: Author’s calculation based on UNCTAD (2016c). Note: Red line is trend line
For a regression analysis, there is an important assumption in a classical regression model, that is, the sequence must be stable. If the sequence is unstable, the test will be invalid or the regression will be false. In practice, the sequence stability test is usually completed by Dickley-Fuller (ADF) test and Phillips-Perron Unit Root Test (PP-test) (Dickey and Fuller 1979; Phillips and Perron 1988). Both unit root tests were conducted on all variables to check whether they are stationary at same level or not. The results of the unit root testing procedure are presented in Table 4.13.
All the variables were found to be stationary.
Having identified and checked the determinants of FDI, the next step is to outline the model to empirically test the level of influence of the aforementioned variables on FDI.
4.2.2 Choice of the Model
The basic linear panel models used in econometrics can be described through suitable restrictions of the following general model:
where j denotes a country, T denotes time, β0 is an absolute term, β is a vector of the factor coefficients, xjT (x1, jT, x2, jT, …, xp, jT) is a vector of explanatory variables, νjT represents the vector of the error components with:
The error term has two separate components, one of which, uj, is specific to a country and does not change over time. We regard uj as a proxy of the combined effect on yjT of all unobserved variables that are constant over time (Wooldridge 2010).
The error εjT, is often called the idiosyncratic error or time-varying error, because it denotes the remaining disturbance, and varies with individuals and time and affect yjT. The individual component, εjT, may either be independent from the regressors or correlated. It is assumed to be normally distributed with mean of zero and finite variance (Wooldridge 2010).
In our book, we analyse and present consistent estimators for three basic models of panel data: the pooling model, the fixed effects or least squares dummy variables (LSDV) model, and the random effects or error components model. They differ mainly in the assumption of the intercept and error term.
In Appendix F, in Tables 4.14–4.18, we present ten model specifications.
The baseline specifications, namely (1) and (2), only include traditional determinants of FDI inflows (market size and production costs) but exclude the determinants of the export-platform FDI. The only difference between the first and the second one is that we use the factors lagged for one period. In specifications (3) and (4), we additionally include the variables that measure market access and openness. Further, we progressively add variables to our baseline model until Adjusted R-squared does not significantly change. In specifications (5) and (6) we add GCF; we add GDP growth rate in specifications (7) and (8).
In the last two specifications, (9) and (10), we replace the market access variables with integration dummies.
Choosing between present and lagged for 1-year values, we refer to the relative better results of the present values. On the whole, specifications (7) and (9) provide the best estimations in terms of R-squared, and expected signs.
In Table 4.2, we represent some of the results from the panel data analysis. First, we consider the OLS estimator results for both specifications. Almost all variables enter the regression with expected signs in addition to their statistical significance. The adjusted R-squared for the specifications (7) and (9) indicates that the variables included explain approximately 76,35% and 77,14% of the variation in FDI inflows in a country.
However, the estimations of the OLS estimator would give inconsistent results because of the correlation between the dependent variable and error term. Random and fixed-effects models remove the inconsistency because they include the individual effect of the countries (Wooldridge 2010).
Generally, in the panel data analysis, the fixed effects (FE) model assumes that each country differs in its intercept term, whereas the random effects (RE) model assumes that each country differs in its error term (Wooldridge 2010).
We used F-test or Chow and Wald test to compare FE and OLS estimators. The null hypothesis is that all the constants are the same (homogeneous), and, therefore, the Common constant method (OLS) is applicable (Wooldridge (2013), pp. 245–248). We decline the null hypothesis (see results in Table 4.3) and conclude that the fixed effect model is more preferable than the pooling model.
We carried out the Breusch Pagan (1980) Lagrange multiplier (LM) test to choose between pooled and random effects estimators (Greene 2003, pp. 223–225). With the large chi-squared (see Table 4.3) too much of the variance is explained by the additional explanatory variables. That is why we reject the null hypothesis in favour of the random group effect model.
Taking into the consideration the results of the tests, OLS model would generate inconsistent estimates, because the correlation between the dependent variable and error term biases the estimations of β0 and β (Wooldridge 2013, p. 468). That is why we do not further consider it.
We need to apply tests to check whether fixed or random effects should be included in the model. Turning to the choice of deciding between fixed and random effects, the random effects model is preferred when there is no a significant correlation between the unobserved sector-specific random effects and the regressors. But if there is such a correlation, the random effects model would be inconsistently estimated and the fixed effects model would be the model of choice (Clark et al. 2010).
Therefore, standard restriction tests should be carried out so that an appropriate statistical model can be chosen.
We performed the specification test to choose between the fixed and random effect model (Wooldridge 2013, pp. 495–496). According to Ahn and Moon (2002), the Hausman statistic is viewed as a distance measure between the fixed effects and the random effects estimators. The null hypothesis of Hausman test is that time-invariant error uj is not correlated with any explanatory variable in all time periods. If the null hypothesis is rejected, then the fixed effect model should be used. Whereas, if the null hypothesis is accepted, the random effect model should be used (Wooldridge 2013, pp. 495–496).
The result of the test clearly rejects the null hypothesis assumption (Table 4.3). Under alternative hypothesis, the fixed effects estimator is still consistent, but the random effects is inconsistent. We conclude that endogeneity is a problem for the random effects estimator and we should use the fixed effects estimator.
In principle, by including country fixed effects, we are controlling for the average differences across countries in any observable or unobservable predictors. This greatly reduces the threat of omitted variable bias.
The results of all conducted tests are reported in Table 4.3.
4.2.3 Robustness Tests
The estimates in the previous section may not account for several important patterns in the model residuals. In the following discussion, we identify some of these problems and consider how they might be addressed.
According to the Gauss-Markov theorem (Greene 2003, pp. 10–17), the estimator is BLUE (Best Linear Unbiased Estimator) when the expectation of errors is zero, given any values of the independent variables:
they are uncorrelated:
and have equal variances:
The errors do not need to be normal, nor do they need to independent and identically distributed.
In this part, we test the pooling and fixed effect models of the specifications (7) and (9) that are the baseline models in our research. The issues we consider are non-linearity, cross panel heteroskedasticity and sectoral correlation, multicollinearity, and endogeneity.
4.2.3.1 Non-linearity and Influential Data
Violations of linearity or additivity are extremely serious: if a linear model includes data which is nonlinearly or nonadditively related, the predictions are likely to be seriously in error.
We tested our data for normal distribution, unusualness of independent variables and for outliers of the independent variables.
First, we constructed the Q-Q plot (see Appendix G, Figs. 4.11 and 4.12). The Q-Q plot, (quantile-quantile plot), is a graphical tool to assess whether the data is normally distributed. It plots two sets of quantiles against one another. If both sets of quantiles came from the same distribution, they form a symmetrical shape around the mean. From the plot we see that our sample data is not skewed. We can identify light tails; thus, we can conclude that the residuals are normally distributed.
QQ-plot for specification (7) with FE. Source: Author’s calculations
QQ-plot for specification (9) with FE. Source: Author’s calculations
However, the Q-Q plot is only a visual check and somewhat subjective, but it allows for the identification of the presence of general problems and whether our assumptions are plausible (Atkinson 1987; Fox and Weisberg 2011).
In multiple regression models, nonlinearity or nonadditivity may also be revealed by systematic patterns in the plots of the residuals versus individual independent variables (see Appendix G, Fig. 4.14). Added variable plots help to project multidimensional data in the two-dimensional world. They identify the presence of outliers that determine the slope (Fox and Weisberg 2011).
The added-variable plot (partial-regression leverage plot) depicts the relationship between dependent and one independent variable, adjusting for the effects of other independent variables (Fox and Weisberg 2011).
High leverage observations are shown in added variable plots as points horizontally distant from the rest of the data. Figure 4.14 suggests a problem in determining the coefficient of GDPGRjT, GDPjT, and FDIjT − 1 because of the points on the bottom of the plot.
Leverage plots also help to identify the “unusualness” of independent variables (see Appendix G, Fig. 4.13). An observation that has an extreme value on a predictor variable—i.e. it is far from the independent variable’s mean—has leverage on (i.e. the potential to influence) the regression line (Fox and Weisberg 2011; Cook and Weisberg 1999).
Leverage plots. Source: Author’s calculations
Added-variable plots. Source: Author’s calculations
High leverage does not necessarily mean that it influences the regression coefficients. In our case GDPGRjT, GDPjT, and FDIjT − 1 have high leverages and yet follow straight in line with the pattern of the rest of the data.
Our plots reveal also some “abnormal” observations on WAGEjT. We have identified a “doubtful” observation that reveals very low wages in Ukraine and Georgia in the early 1990s. This observation contains valuable information about the process under investigation. We conclude that our outlier does not indicate incorrectly measured or recorded data. In this situation, it is not legitimate to simply drop it.
Moreover, Cook’s distance confirms that all observations lie within the acceptable range (see Appendix G, Fig. 4.15).
A threshold level of the Cook’s distance is calculated as 4/N or 4/(N−k−1), where N is the number of observations and k the number of explanatory variables (Cook and Weisberg 1999; Fox and Weisberg 2011). The latter formula should yield a threshold around 0,09. An observation with a Cook’s distance larger than three times the mean might be an outlier. In our case the values do not exceed value of 0,20 (see Appendix G, Fig. 4.15).
Diagnostics plots. Source: Author’s calculations
4.2.3.2 Heteroskedasticity and Autocorrelation
The third condition of the Gauss-Markov theorem is that the variance of the error term is homoskedastic, the error v has the same variance given any values of the explanatory variables. If the error terms do not have constant variance, they are said to be heteroskedastic.
In other words:
If this assumption fails, then the model exhibits heteroskedasticity. This means that the variance in the error term, v, conditional on the explanatory variables, is the same for all combinations of outcomes of the explanatory variables. If Var(u|x) is not constant, OLS is no longer BLUE. That is why the violation of this assumption makes the estimators of βj biased. Since the OLS standard errors are directly based on these variances, they are no longer valid for constructing confidence intervals and t statistics. Similarly, F statistics are no longer F distributed, and the LM statistic no longer has an asymptotic chi-square distribution (Wooldridge 2013, pp. 268–271).
It is important to mention that heteroskedasticity does not cause bias or inconsistency in the OLS estimators of the βj, whereas the omitting of an important variable would have this effect.
Heteroskedasticity may occur with the quantitative change of independent variables. In this case we assume that there are modelling errors and some important variables not included in the model. Measurement errors can also cause heteroskedasticity. Some respondents might provide more accurate responses than others. Heteroskedasticity can also appear if there are subpopulation differences or other interaction effects. This problem also can be eliminated by incorporating such differences into the model (Wooldridge 2013, pp. 268–271).
The plots on the Figs. 4.16 and 4.17 in Appendix H show that there is a possible heterogeneity across countries. Thus, the including of the country-specific effect (FE) may eliminate this problem. The heterogeneity across years seems to not be a problem in our analysis.
Plot group means and confidence intervals, by countries. Source: Author’s calculations
Plot group means and confidence intervals, by years. Source: Author’s calculations
There are a great number of tests for heteroskedasticity. Some of them are able to detect heteroskedasticity, however, they do not directly test the assumption that the variance of the error does not depend upon the independent variables (Wooldridge 2013, p. 268). Apart these tests we also employ tests which detect the kind of heteroskedasticity.
Firstly, we carried out Score Test for Non-Constant Error Variance in the pool and fixed effect model for both specifications. This test checks at the hypothesis of constant error variance against the alternative that the error variance changes with the level of the response. This test is often called the Breusch-Pagan test; it was independently suggested with some extension by Cook and Weisberg (1999) (Wooldridge 2013, pp. 275–278; Greene 2003, pp. 222–225). The test rejected the null hypothesis of constant variance or homoskedastic standard errors and indicated the presence of heteroskedastic standard errors in both specifications (see Table 4.4).
We conduct two additional test to identify the non-homoskedastic errors. The Goldfeld–Quandt test compares the variances of two submodels divided by a specified breakpoint and rejects null hypothesis if the variances differ, and, hence, the test is sometimes called a two-group test. If the null hypothesis is rejected there is ground to suppose that the standard deviation of errors is proportional to some variable (Greene 2003, pp. 223–224). In our case Goldfeld-Quandt test against heteroskedasticity rejected the null hypothesis and showed that the variances differ into subgroups (Table 4.4).
The Harrison–McCabe test is similar to the Goldfeld-Quand test. Its statistic is the fraction of the residual sum of squares that relates to the fraction of the data before the breakpoint. Under null hypothesis the test statistic should be close to the size of this fraction, e.g. in the default case close to 0,5 (Harrison and McCabe 1979). In our case, the null hypothesis is rejected because the statistic is too small (Table 4.4).
Both tests confirmed the presence of heteroskedasticity.
Unfortunately, the Goldfeld–Quandt and the Harrison–McCabe tests are not very robust to specification errors. They detect non-homoskedastic errors but cannot distinguish between heteroskedastic error structure and an underlying specification problem such as an incorrect functional form or an omitted variable. Jerry G. Thursby proposed a modification of the Goldfeld–Quandt test using a variation of the Ramsey RESET test in order to provide some measure of robustness (Thursby 1982).
The Ramsey Regression Equation Specification Error Test (RESET) test is a general specification test for the linear regression model. It tests whether non-linear combinations of the explanatory variables have any power in explaining the response variable. More specifically, it tests whether non-linear combinations of the fitted values help explain the response variable (Wooldridge 2013, pp. 306–307). To implement RESET, we must decide how many functions of the fitted values to include in an expanded regression. There is no right answer to this question, but the squared and cubed terms have proven to be useful in most applications (Wooldridge 2013, pp. 306–307). In our test, we use quadratic influence of the fitted response.
RESET test shows the correctness of the structure of our model both for specification (7) and (9) for OLS and FE estimators.
Some tests reveal the presence of cross-panel heteroskedasticity in our models, which can significantly influence standard errors and therefore affect hypothesis testing.
Apart from groupwise heteroskedasticity, panel data might also suffer from serial correlation. Serial correlation occurs in time-series studies when the errors associated with a given time period carry over into future time periods. The errors should fulfil the next condition:
The consequences of autocorrelation are similar to heteroskedasticity, but the problems caused by the latter are usually more severe. Autocorrelation of errors violates the ordinary least squares (OLS) assumption that the error terms are uncorrelated, meaning that the Gauss Markov theorem does not apply, and that OLS estimators are no longer the Best Linear Unbiased Estimators (BLUE) (Wooldridge 2013, pp. 353–354).
Models with lagged values of dependent variables as factors should be carefully tested for the presence of autocorrelation because the autocorrelation of idiosyncratic errors may lead to biased and inconsistent estimates.
We apply several tests for residual autocorrelation. Lagrange multiplier test developed by Baltagi and Li (1995) (Greene 2003, pp. 270–271) and a “corrected” versions of the standard LM test developed by Bera et al. (2001) are used to test for first-order serial correlation in residuals (Croissant and Millo 2008). The results of our tests (see Table 4.5) reveal the presence of serial correlation in OLS models, but taking into account the country-specific factors (FE) eliminates this problem.
Secondly, we applied the Breusch-Godfrey/Wooldridge test. It is a Lagrange multiplier test on the residuals, which should be serially uncorrelated under the null hypothesis. This test allows for lagged dependent variables (Wooldridge 2013, p. 422). The Breusch-Godfrey/Wooldridge test for serial correlation also does not reveal the presence of autocorrelation in FE models.
A Wooldridge test for autocorrelation in in fixed-effects panel models confirms the null hypothesis of no first-order serial correlation (Table 4.5). FE coefficient estimates are consistent, unbiased and asymptotically efficient (Wooldridge 2010, pp. 310–312).
A popular test for serial correlation is the Durbin-Watson statistic. The DW statistic lies in the 0–4 range, with a value near two indicating no first-order serial correlation. Positive serial correlation is associated with DW values below 2 and negative serial correlation with DW values above 2 (Greene 2003, p. 270; Wooldridge 2013, pp. 418–419). This test also confirms the robustness of the results (Table 4.5).
The tests revealed the presence of heteroskedasticity in OLS and FE models, and autocorrelation in OLS models. We can easily eliminate the problem of autocorrelation by including country specific effects (FE). In order to account for heteroskedasticity within panels we can employ White-corrected standard errors (see Appendix I).
4.2.3.3 Multicollinearity
Multicollinearity is a high (but not perfect) correlation between independent variables. Correlation among regressors is quite common in time-series data and even extreme multicollinearity (so long as it is not perfect) does not affect the stability of the model. OLS estimates are still unbiased and BLUE. Nevertheless, the greater is the multicollinearity, the greater are the standard errors because it affects calculations regarding individual predictors (Greene 2003, pp. 56–59). To reduce the negative effects of the multicollinerity, each explanatory variable was progressively added (Appendix F, Tables 4.14–4.18).
We have already mentioned multicollinearity issues in Sect. 4.2.1. The potential multicorrelation problems include correlation between integration variables. The matter of fact, they represent the gradual process in which some stages appear only once the agreements were achieved on the previous stages.
The only problem that can arise is only in pairs of dummies and market access variables. The pair correlation coefficients are 100%. That is why we consider them in different specifications (specification (7) and (8); (9) and (10)).
Formal tests for multicollinearity (Variance Inflation Factor (VIF = 1/1−R2)) is not clearly applicable to panel data (Greene 2003, pp. 56–59). The result for the proxies are reported in Table 4.6 and confirm that none of the variables are highly correlated, as none of VIF’s are excessively high based on rule of VIF > 4 for the existence of problematic multicollinearity. The correlation coefficients between the explanatory variables are also not alarming high (R2 > 0,80 or R2 > 0,90), as can be seen from the correlation matrix presented in Appendix E, Fig. 4.9.
4.2.3.4 Endogeneity
One of the most important assumptions of the regression models is that factors are exogenous. The violation of this condition leads to a significant deterioration of the estimates. Endogeneity, or two-ways causality, occurs when some of the explanatory variables are wholly or partly influenced by the dependant variable. This may cause problems not only in inference but also for estimations as the independent variable is potentially correlated with the variation in the dependent variable that is relegated to the error term.
Endogeneity is often described as having three sources: omitted variables, measurement error, and simultaneity. All the sources of the endogeneity may appear simultaneously, they can compensate or strengthen one another.
In our model, the relation between FDI and GDP attracts special attention. The correlation between FDI and GDP could arise from an endogenous determination of GDP, that is, GDP itself may be influenced by wages, GCF or integration. In these circumstances there would exist a correlation between GDP and the country-specific error term, which would bias the estimated coefficients.
Moreover, as a matter of fact, high market potential attracts more FDI. Foreign investors add capital stock in the host country, what in the issue stimulates the economic growth. Moreover, domestic investment financed by FDI is included in the definition of GDP. These and other potential endogeneity problems with GDP are non-trivial, and have prompted several studies to move the GDP level to the left-hand side, estimating FDI as a share of GDP.
Therefore, in order to gauge the magnitude of the potential endogeneity problems, we employ the Granger causality tests. “Granger causality” is a term for a specific notion of causality in time-series analysis. Its main idea of this test is to evaluate two models with and without doubtful factors and compare them. Evaluations are made in both directions with lagged values of dependent variables (see Table 4.7) (Greene 2003, p. 592).
The results are shown in Table 4.7 and suggest that there is causality between FDI and GDP. This suggests that endogeneity may be a significant problem in our sample.
As far as I know, the Granger causality test in panel data is not generally used because it does not account for country differences. Thus, we estimate equations adding county dummies and compare models by ANOVA (see Table 4.8).
Furthermore, we have made previous references to the extensive literature that has established a robust positive relationship between FDI on the left-hand side and market size.
In this section, we conducted the basic tests for non-linearity, heteroskedasticity and autoregression, multicollinearity and endogeneity. The baseline models showed no evidence for unmanageable multicollinearity and autocorrelation, but confirmed the presence of heteroskedasticity and endogeneity.
The endogeneity and unobserved heterogeneity often refer to the same thing. Endogeneity comes, for example, from omitted variables; unobserved heterogeneity is caused by the same problem. Thus, the elimination of endogeneity will automatically eliminate heteroskedasticity (Wooldridge 2012).
In next section we will try to account for these problems by running fixed-effect instrumental variables estimator.
4.2.4 Instrumental Variables (IV) Estimators
In our book, we use dynamic panel data methods to examine the determinants of FDI in the Central and Eastern European countries. There are several dynamic model estimators for panel data, such as 2SLS and GMM, that have been developed in the econometric literature to solve the problem of endogeneity. For our estimations we employ 2SLS and GMM with fixed effects.
A common problem for this type of analysis is coming up with a set of instruments that have to be correlated with the endogenous regressors but must not be correlated with the time-varying error term:
and
The choice of instrumental variables in the literature is arbitrary and there is no consensus on a set of variables that are widely used (Wooldridge 2010, pp. 83–84).
In our case, the instrument choice is rather obvious. We use population as an alternative proxy for market size. It is highly correlated with GDP, and has lower correlation with FDI. Moreover, it does not cause much difference in the estimation results. We also dummy for years as an instrumental variable.
Thus, we have two exogenous instruments that are not included in the model: natural logarithm value of population, and a factor of year. Lagged for one period value of FDI inflows, wages, market access and openness proxies, as well as GCF, and GDP growth are used as endogenous instrumental variables.
In Table 4.11, we reported results from the 2SLS fixed effect model and GMM fixed effects model.
The overall validity of instruments is checked by Sargan test. Sargan test statistics (1988) is generally used to detect the correlation between instrumental variables and error term. Weak correlation between instrumental variables and endogenous variables lead to the problem of identification of equations (Kleiber and Zeileis 2008; Wooldridge 2010, pp. 122–124). As Sargan test in Table 4.9 indicates, the null hypothesis that the instruments are valid instruments is not rejected for both models. In other words, the instruments are not correlated with the error term and the choice of instruments is appropriate.
In addition, Sargan and Basmann tests evaluate whether the equation is misspecified and whether one or more of the excluded exogenous variables should in fact be included in the structural equation (structural model is heteroskedastic). Both tests assume that the errors are independent and identically distributed random variables. The estimations are invalid when this condition is not fulfilled (Kleiber and Zeileis 2008). This seems not to be true. Sargan statistic is valid at 0,1% level, though we do reject the hypothesis that our instruments are invalid or our model is incorrectly specified.
The results of above mentioned test show that the elimination of endogeneity also accounted the problem of heterogeneity as was assumed in Sect. 4.2.3 Heterogeneity and Autocorrelation.
Wu-Hausman F and Durbin-Wu-Hausman tests are commonly used to estimate the quality of instruments. These tests compare the evaluation results before and applying the instruments. They determine whether the instruments are endogenous. The null hypothesis of the Durbin and Wu–Hausman tests is that the variable under consideration can be treated as exogenous. If the endogenous regressors are in fact exogenous, then the OLS estimator is more efficient (Wooldridge 2010, pp. 118–122). As Wu-Hausman F Test and Durbin-Wu-Hausman test in Table 4.9 indicate that the exogeneity of the variable FDI is clearly rejected. Both estimates of the error variance are consistent.
To test for weak instruments we implement F test of the first stage regression proposed Cragg-Donald (1993) and Kleibergen and Paap (2002) LM test to provide further model evaluation. Stock and Yogo (2005) completed this test by calculating critical values to measure relative bias of 2SLS; they also provided critical values for worst-case rejection rates (Wooldridge 2010, pp. 92–94). Overall, Crag-Donald F statistic is just below the Stock and Yogo 1% critical value (see Table 4.9). Hence, the instruments are strong and lead to relatively small biases in 2SLS with FE.
Generalised Method of Moments (GMM) is another estimator for ARDL models. This estimator was developed by Arellano and Bond (1991) to get asymptotically efficient estimators. This estimator is designed for datasets with many panels and periods, and it requires that there be no autocorrelation in the idiosyncratic errors (see Table 4.11).
To estimate the validity of instruments in our GMM model we employ the Hansen’s J test which is analogous to Sagran and Basmann tests. Hansen test is used to check the correlation between an endogenous variable and time-varying error term. The null hypothesis is that instrumental variables are exogenous. The alternative hypothesis is that some of the instrumental variables are not exogenous or are correlated with the time error term (Greene 2003, pp. 154–155). Table 4.10 shows that J test rejects the hypothesis of the overidentifying restriction in both specifications.
The results reported in Table 4.11 are robust to heteroskedasticity and autocorrelation (HAC).
4.3 Empirical Results
In this section, we analyse the empirical results obtained from the previous section. The results (Table 4.11) show that regression model with FDI as a dependent variable fits well with independent variables, since value of adjusted R-Squared is high (0,77). The R-square (0,79) shows the proportion of variance in FDI inflows which can be predicted by independent variables.
The empirical results also support our hypotheses, as although some are not significant, all the explanatory variables have shown signs as expected. The estimations of 2SLS and GMM are similar, but some necessary points should be mentioned.
4.3.1 Traditional Determinants of FDI
According to the results presented in Table 4.11, GDP is found to be a fundamental factor of attraction for multinationals, even accounting for endogeneity in the model. The coefficient indicates that a 1% increase in GDP is associated with a proportional increase from 0,69% to 1,13% of FDI inflow (2SLS and GMM estimations of the specification (7) and (9)).
The coefficient of GDP accurately reflects theoretical expectations (Table 4.1). Flows are expected to be greater in larger economies with well-built markets. As investing in a given countries implies large fixed costs, multinationals are most likely interested in capturing a greater share of the market when expanding to the CEE countries.
Unfortunately, we do not clearly observe the motives of FDI in CEE: whether FDI in the region is market-seeking (tariff-jumping motive prevail) or resource-seeking (off-shoring and export-platform motives also exist) (see Dunning 1973). Still, the analysis of the sector structure can shed additional light upon this question. Most FDI has been domestic-market seeking in traditional services such as finance, tourism and trading, or in industries such as electricity, water or telecommunications. This largely reflects the non-tradable nature of these sectors. They are non-storable and transportable and hence need to be produced when and where they are consumed.
Section 2.2 shows that in the largest host countries of the CEE region, the industry composition of inward FDI gradually shifted from manufacturing in the 1990s towards services in the early 2000s, and within services, from privatised network industries in earlier years towards business services (UNCTAD 2004). In the Czech Republic, Hungary and Poland, services already became dominant in terms of FDI in the late 1990s. Generally, in the early 2000s the countries of CEE were characterised by substantial FDI penetration in infrastructure services (e.g. banking, telecommunications, water, electricity) (UNCTAD 2004).
Thus, significant FDI in the service attributed market-seeking FDI inflows in Poland, Hungary, the Czech Republic, Romania and Ukraine in the 2000s (see Sects. 2.2.3 and Sect. 2.2.4). Further, some evidence of a negative relationship between FDI and tariff rates confirms the presence of a tariff-jumping motive (see Table 4.11).
These results suggest also that countries with a smaller market size may have problems in attracting foreign investors, all other factors being equal. Countries such as Albania, Macedonia, and Moldova may become less desirable for foreign investors seeking high purchasing-power and demand, but trade agreements with the EU (preferential or association agreements) may affect market size, one of the key determinants of FDI. As trade blocs and regional links grow, the significance of national markets thus diminishes.
The importance of market size was also supported by a survey analysis undertaken by Altzinger and Bellak (1999). They questioned more than 150 Austrian firms investing in the CEECs, and found that market potential is the most important factor. According to Savary (1997), 22 surveyed French industrial firms that invested in the CEECs also mentioned market size as being among the most important factors. Thus, they found Poland more attractive than most South-East European countries. However, market size was found to be slightly less important than factor costs in other smaller countries of the region (Savary 1997).
In our model, the value of the labour cost coefficient implies that a decrease in average monthly wages by 1% would generally lead to an up to 0,15% increase in the magnitude of FDI flow.
Thus, high levels of FDI were expected to flow into the CEECs, mainly due to low wages and open access to EU markets. As the output from resource-seeking FDI is typically intended for export, and intermediate goods are imported from outside the host economy, the reduction of trade barriers will ease international trade and, thus, will increase the level of FDI in low-wage countries. The presence of export-platforms also can be further confirmed by sector analysis. Besides market-seeking FDI in services, the CEE countries attracted FDI in the automobile industry which is typically export-oriented (see Sect. 2.2). There are plenty of examples of resource-seeking FDI in the region. As mentioned previously, these include; Toyota and VW/Skoda in the Czech Republic; Suzuki and Audi in Hungary; Fiat, GM/Opel, Daewoo, VW in Poland; PSA/Peugeot and VW in Slovakia; and Renault in Slovenia (see UNCTAD 2003, p. 61).
According to our estimations (see Fig. 4.5), average wages in the new EU member states are still on average 60% lower than the EU-15 average. The results prove the further potential for installation of export platforms in the integrating countries. We can also conclude that the countries with the lowest average wages in the region such as the countries in the Western Balkans, Georgia, Moldova and Ukraine that have just begun integrating with the EU will attract more FDI from both outside and inside of the EU. We can even observe in the future, the reallocation of production not only from the EU-15 but also from new EU members.
However, we should also point out that the coefficient of labour cost is insignificant. The issue of whether labour costs affect the decision to invest in the transition economies is an important one, and the subject of some debate due to the inconclusive results (See Appendix F)—it appears both with positive and negative signs. The explanation was given by Rolfe and Woodward (2004) who pointed out that once the decision to locate in a low wage area—as Eastern Europe—has been taken, “finding the cheapest possible labour within an already low wage region may or may not be an important consideration” (Rolfe and Woodward 2004). Elsewhere, labour cost advantage was viewed as less important than market access (Janssens and Konings 1996; Savary 1997; Altzinger and Bellak 1999).
4.3.2 Macroeconomic Variables
The pattern of FDI may be different among countries with similar market size and production costs. The location choices of multinationals between countries are increasingly related to advantages arising from other factors that influence the supply capacities of host countries, such as scale economies (particularly in the manufacturing sector) and clustering (agglomeration economies), as well as institutional and policy variables. The investors address this as “business climate”, which comprises many individual factors.
The positive and statistically significant effect (at up to 0,1% level) of Gross capital formation (or gross domestic investment) in relation to FDI, indicates that the improvement of infrastructure positively affects FDI inflow in the CEE countries. From the results, it is clear that physical infrastructure (ports, roads, power, telecommunications), knowledge infrastructure (universities, technology parks, etc.) and business infrastructure (EPZs, clusters, etc.) increase the attractiveness of a country.
Agglomeration also exhibits a high degree of statistical significance, and has a large and positive impact on the location of FDI. This also confirms the importance of infrastructure and firm concentration. Firms benefit from locating close to other firms, to markets and factors of production, and from the availability of specialised skills, developed production factor markets, and well-built supply chains.
In Sect. 3.1, we mentioned NEG studies (Fujita et al. 1999; Puga and Venables 1997; etc.) that identified the importance of agglomeration for industrial structures and international trade. Head and Mayer (2004) studied the factors that influence the location of Japanese firms within Europe. They found that that firms prefer to locate “where the markets are.” (Head and Mayer 2004).
In addition, the presence of other foreign firms in the region reduces the risk of TNCs by demonstration effect, wherein, multinationals tend to place more trust in locations where other multinationals are present. Thus, Kinoshita and Campos (2003) found that the existence of agglomeration economies has a positive and statistically significant effect on FDI, especially in non-CIS countries. It appears because a country’s environment is not well explored. That is why foreign investors may view the investment decisions of others as a good signal for favourable conditions, and invest there too to reduce uncertainty (Kinoshita and Campos 2003).
The reasons for the importance of agglomeration and gross domestic investment for location decisions are not difficult to find. Foreign firms want to place their subsidiaries close to centres of corporate, political and financial decision-making, high levels of income, access to technology and, especially, innovative activities, universities, institutions and modern infrastructure (including easy access to international air transport) and quality of life.
The coefficient for GDP growth is positive and highly significant in our specifications. This proxy enters both specifications estimated by IV with coefficient up to 0,06% and 0,1% level of significance. The results show that the strong economic growth of new integrating countries and their favourable growth prospects are very attractive for market-seeking investors. Indeed, the real GDP growth rate in CEE is more than double the EU-15 average (UNCTAD 2016d).
There are several reasons why foreign investors might prefer faster-growing markets. Firstly, higher growth rates signal higher rates of return on investment. Thus, investment in countries with rapid economic development promises growing profits. Secondly, GDP growth rate also measures the increasing market size (Blonigen et al. 2007; Vernon 1966; Carstensen and Toubal 2004). Therefore, investors are attracted by better production opportunities and increasing demand.
4.3.3 Integration Effect
We also want to analyse how FDI inflows are affected when countries start building integration unions. For this purpose, we analyse trade openness and access to foreign markets.
The negative sign of the trade openness confirms the presence of tariff-jumping FDI motives. Indeed, the existence of tariff and non-tariff barriers could affect location choice with respect to servicing markets. The results show that more open economies are attractive for export strategies; whereas the presence of trade barriers stimulates more FDI as a supply opportunity. Thus, some host countries and member-countries of regional trade blocs can intentionally use tariffs, quotas, and local standards to encourage direct investment. However, a proxy for openness is significant only for 2SLS estimation of specification (9).
The variable FTAMAjt appears with negative and insignificant coefficient (specification (7)). Indeed, membership of CEFTA and BAFTA did not significantly improve the market size of the economies (see Fig. 4.5). Moreover, members of CEFTA or BAFTA got access only to the countries of a similar economy and demand structure (see Sect. 2.3.3). Thus, foreign firms did not get the advantage of installing an export-platform within the European FTAs.
These results are in line with the estimations of specification (9). The FTA dummy variable has a negative sign. One explanation for the negative sign is that both FTAs (the Baltic Free Trade Area (BAFTA) and the Central European Free Trade Agreement (CEFTA) were created to favour the development of intra-regional trade and competition in these countries. This is why FTAs are an important determinant of economic trade, since they institute free trade among a number of nations.
The further results show that integration agreements with the EU have significant and positive effects on attraction of FDI. AAMAjt, and EUMAjt variables appear with 0,011%, and 0,012% coefficients in 2SLS; and 0,012% and 0,010% in GMM, correspondingly.
The Association Agreements form a free-trade area with the EU, remove trade barriers, and allow free access to the EU market. Thus, in the early 1990s, when the AAs with the CEE countries were signed, the countries were afforded access to a market with a population of 380 million, and $7711 billion GDP. Current AAs provide access to a market of already 506 million people and $17868 billion GDP (UNCTAD 2016d). Foreign investors cannot but use the opportunity to supply such a huge and significant market.
Access to the EU market has the stronger effect according to the t-value. Thereby, the reduction of internal trade costs associated with regional integration affected volumes and patterns of FDI both into and within the integrated area. This made the CEECs more attractive to foreign investors from the US and Japan that wanted to bypass the EU’s external trade barriers so as to gain access to the whole European market. Moreover, reduced trade barriers stimulated a shift of EU production towards countries with relatively low production costs (negative sign of WAGE supports this statement). A decrease in tariff barriers pushes FDI in the CEECs which confirms off-shoring and export-platform motives. However, we do not observe the sources of FDI, and that is why we are not able to conclude whether inflows from inside or outside of the region dominate in CEE.
According to the t-value, the effect of market access after becoming an EU member-country is weaker than the effect after signing the AAs. The reason behind is that CEE enjoyed free access to the EU market for many years before their accession to the EU. Companies that wanted to benefit from it had invested in these countries long before they became full EU members. Moreover, newly accessible markets are usually distant lands. As previous literature suggests, proximity plays an important role both for FDI and trade flows (Altononte and Guagliano 2003; Carstensen and Toubal 2004). Taking labour cost differences into account also, CEE countries are most unlikely to become export-platforms for Asian, African, Latin American and Oceania countries. Therefore, significant changes in FDI inflow around or following accession to the EU are not to be expected.
These results have important implications for understanding the determinants of FDI. European integration expands the determinants of FDI from the level of the country to the level of the whole region. As a result, the complete set of FDI determinants includes both country and neighbourhood characteristics.
Another aspect of European integration is the ongoing transformation of the countries in CEE to market economies (institutional transformation). Thus, the signing of the AAs indicates a favourable investment climate, as it guarantees investors the country’s adherence to certain economic standards, as well as the adoption of regulations designed to harmonise its business and legal environment with that of the EU. However, the effect of AAs dummies is not significant. We can conclude that in the case of the AAs, the effect of market access is stronger than the effect of institutional reforms.
Nevertheless, we found the effects of EU accession from the application for membership onwards are positive and statistically significant (specification (9)). The processes determining EU membership are based on an evaluation of progress in transition, which is a determining factor of FDI. Integrating countries became more attractive to foreign investors as their economic systems and regulatory frameworks became more similar to those of the Union, and as the dynamic effects of the association agreements began to manifest themselves. The admission into the Union raises the likelihood of free trade with the EU member states and the adoption of EU law.
The accession countries also developed the institutional framework to administer and properly channel the variety of funds available from European Community sources for assisting economic development. In their search for international competitiveness under EU membership, some accession countries are also lowering their corporate taxes. This combination of factors, combined with a favourable business climate, highly skilled workforce and free access to the rest of the EU market, made the accession countries attractive locations for FDI. This applied particularly to efficiency-seeking FDI.
Thus, we also observe the confirmation of our hypothesis—the greater the degree of trade integration, the larger the increase in each market’s potential, hence the higher the obtainable profits are by relocation production through FDI. While the European Agreements have implied an opening up of the markets only for trade, capital and factor movements, the EU accession has involved an economic and political change within the candidate countries. This has an impact on inward FDI into CEE.
On average, CEECs are predicted to receive about 0,58% more FDI after they apply for membership, 0,51% after they are granted candidate status and 0,76% after membership negotiations with the EU start. The effect is even more apparent if we advert to specification (10) with lagged variable values (See Table 4.22). Four variables appear with a significant coefficient. We may conclude that the effects resulting from the EU announcements are stronger with time.
In constructing a linear chart on coefficients of specification (8), where the nine dummy variables are designed to measure the effect of the EU announcements, we proved that this effect increases over time. The significant result of the variables supports the hypothesis that even deeper EU integration of the CEE countries contributes to a greater increase in FDI inflows (see Fig. 4.6). The empirical results reinforce the assumptions provided in Sect. 2.4.
From Fig. 4.6 we can also conclude that the EU accession negotiation process and the FDI inflows are not a simple monotonic relationship. Hence, we assume that the positive effect of the integration diminishes as a country becomes a full member of the EU than at the accession negotiation stage mostly because membership in the EU does not imply further economy transformations.
Thus, the countries that have successfully implemented transition policies join the EU relatively quickly, which further accelerates FDI and generates more growth and development. In contrast, countries that are less successful in implementing transition policies gain less FDI because they are less attractive for investors in this context, and also, because in the near future they will not become full members of the EU and be able to enjoy the advantages of membership.
In spite of the fact that we used other proxies to estimate the effect of EU integration, we obtained results in line with the previous findings. Bevan and Estrin (2004) concluded that EU announcements about potential accession have significant independent effects on FDI flows to transition countries.
Clausing and Dorobantu (2005) showed that future EU members receive more foreign direct investment. Results indicate that the Copenhagen announcement is associated with positive and statistically significant effects on foreign direct investment, and after the release of Agenda 2000, both first and second wave countries experienced continued simulative effects regarding foreign direct investment.
4.3.4 Future FDI Flows
We now turn to the question of whether FDI inflows in the CEECs are close to their “normal” levels, or whether they are overestimated or underestimated by investors in some countries. For this purpose, we compare fitted and observed values of inward FDI.
We use the estimated by 2SLS coefficients of the specification (7) to calculate the expected FDI levels for the period 1992–2015. Although these countries have seen differing progress in terms of integration and stabilisation, they follow the same transition and integration strategies.
Appendix H reports the expected and actual annual FDI inflows. The expected levels of inflow fit well with the observed values of inward FDI in CEE, with only minor differences between them. However, as can be seen from Fig. 4.18 in Appendix K, the early periods of transition are characterised by shortfalls in FDI inflow. In the mid-1990s FDI inflows reached the corresponding amounts. Thus, both good macroeconomic performance and integration have contributed to the growth of FDI in the transition economies.
We can also identify a period of economic boom, 2005–2009, when the observed values surpassed the estimated values. Indeed, this period is characterised by excessive investor activity in the CEECs (see Sect. 2.2.3).
Further, we mark out a dramatic drop in FDI. This supports our conclusions in Sect. 2.2.4, that the FDI inflows in CEE overestimated the consequences of the economic crisis. During this period, considerable differences between estimated and real values are observed in Lithuania, Latvia, and Slovenia. The empirical results confirm our observations that these countries experienced the strongest reduction in FDI inflow among all CEE countries what amounted decline by 2–4 times.
We can conclude that our model performs well. Our findings suggest that a large part of the FDI inflows to these countries can be attributed to macroeconomic performance and integration progress, including the positive effects of increased market access. Moreover, since we employ FE estimators, we also account for the effect of each country’s economic characteristics on FDI inflow. In this regard, the starting conditions of transformation are also covered by our estimations.
Therefore, from our findings, FDI inflows in CEE can be forecasted and evaluated based on these countries’ macroeconomic characteristics and the trade agreements signed.
In this part, we also want to estimate the amount of FDI adjusted to the integration effect. Figure 4.7 accounts for the effect of the AAs and EU membership on FDI inflows. Since FDI inflows are volatile, the effect of EU integration is not good observable.
That is why we also constructed the difference plot to show the additional FDI inflows generated under influence of EU integration (Fig. 4.8). Our estimations show that integration with the EU has brought an additional $215,52 billion to the region since 1992.
On the whole, the effect of improved access to neighbouring markets had a continuous and significant effects upon countries’ attractiveness for multinationals.
4.4 Subtests
Although our results appear to be economically sound, there remains some possibility that the reported coefficients may be subject to bias given that the panel regression explores FDI determinants across CEE countries at different stages of transition and integration. In this section, we attempt various decompositions of our sample, namely by geographic region and time periods. This is to determine whether our aggregate results dissemble some important sub-trends. In addition, the estimates of this section will serve as extra tests to assess the robustness of the aggregate relationship between European integration and FDI in the smaller sub-samples.
As stated in the literature, the empirical results are most sensitive to the selected sample and selected time period (Heckman 1979). The existing sample exhibits large discrepancies in terms of economic and transition levels among the new EU members and candidate countries. Therefore, different groupings of countries are taken into account for the sensitivity analysis to provide an insight for potential determinants of FDI.
4.4.1 Decomposition by Region
We begin with a breakdown by sub-regions. It is important to control for differences across the main European regions because such differences may influence FDI inflows.
Firstly, we follow the procedure employed by Holland and Pain (1998) and test for common parameters using four country groups. Groups are formed on the basis of their inclusion in the accession process:
-
Eight Central European (CE) countries (the Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Slovakia, and Slovenia) that joined the EU in 2004;
-
South-East European (SEE) countries (Bulgaria, Romania, Croatia), the first two countries joined the EU in 2007. Croatia became the 28th member state of the EU on the 1st of July, 2013;
-
Three former members of the Soviet Union that signed the Accession Agreements in 2014, namely Georgia, the Republic of Moldova, and Ukraine (GMU);
-
The Western Balkans (WB) (Serbia, Bosnia and Herzegovina, Montenegro, Macedonia and Albania), the future members of the EU, that are at their different stages of integration with the EU (see Table 2.3).
The results of these tests also can be compared with the previous studies that estimated the effects of integration in smaller samples (see Table 2.1).
We re-estimate Eq. (4.2) allowing for separate slope parameters in each of the distinct country groups. The OLS and FE estimates of the baseline model (7) and (9) are shown in Appendix L. Both the coefficient values and their significance levels are not too different from the reference specification.
The coefficients on market size (GDP) are somewhat larger for GMU than in the reference model, and significant at 0,1%. That indicates that from three countries in the subregion, Ukraine, the largest country, attracted significantly more FDI mostly thanks to its market size. The strongest effect of GDP growth rate was also observed in CE. Indeed, Poland and the Czech Republic, the fastest growing economies in the region, were the leading recipients of FDI.
Wages appear with positive sign and 10% level of significance in CE. Probably, the reason for this is that FDI is attracted in sectors where high-skill labour is required. Indeed, the CE countries possess a higher level of knowledge in comparison to other countries of the region. This assumption is confirmed by the agglomeration effect. The agglomeration effect is present for the non-CE countries, but not in the CE countries. The greater importance of agglomeration in the non-CE countries is consistent with greater externalities in the manufacturing sector and positive externalities arising from specialized and low-cost labour.
Thus, the different FDI motives explain different patterns in non-CE and CE countries. In the non-CE countries that receive FDI mostly in the manufacturing sector, institutions and agglomeration are the main considerations for investors (Kinoshita and Campos 2003). Moreover, the significant evaluation for agglomeration shows investors are more likely to invest in markets that have been explored by others.
The effect of the integration is not uniform among CEE sub-regions. The FTAs have a negative influence for FDI inflows. In the GMU, the membership in FTA costs −0,05% decrease in FDI inflow; similarly in the WB it results in −0,06% decrease of FDI inflows. Signing of the Association Agreements is statistically significant in SEE and in CE. This effect appears because we estimate countries in different stages of integration: CE and SEE had more time to enjoy the positive contribution of EU integration towards FDI inflow, whereas the inclusion of the WB and GMU countries lasts only for a few years.
The role of EU membership in CE was the most evident and important among other regions. Results show that membership of the European Union brought CE countries a 0,015% rise in FDI inflow what is higher than average region value.
These findings are in line with Brenton et al. 1998) who made the distinction between first-round EU candidates (Poland, Hungary, Czech Republic), second-round candidates (Slovakia, Bulgaria, Romania) and CIS. They proved that the first-round EU candidates always take in the highest value of FDI from Europe and Finland, followed by the second-round and then by CIS.
Altomonte and Guagliano (2001) also discovered a greater capacity of CE in the attraction of FDI flow with respect to the Mediterranean countries due to the higher degree of integration achieved among the CE countries.
The coefficients of the macroeconomic factors in specification (9) are similar to the specifications (7). However, some important findings need to be mentioned.
We found potential candidate, candidate and accession status to have positive and statistically significant effects in the CE and SEE countries. However, the impact of the announcements was stronger in SEE. These findings are in line with findings of Clausing and Dorobantu (2005) that the release of the Agenda 2000 in July 1997 affected foreign direct investment in the first wave and second wave countries differently. The announcements had a large and statistically significant in the second wave countries, but positive and less significant effects in the first wave countries.
The difference in the strength of the effect lies in the intensity and speed of the reforms associated with European integration. The SEE countries were a relatively long time in AAs with the EU before they were recognised as candidate countries. Therefore, they implemented the most changes during that period. This is confirmed by the positive and significant effect of the AAs. On the contrary, the CE countries quickly adopted all necessary reforms and were immediately granted candidate status (Clausing and Dorobantu 2005).
4.4.2 Decomposition by Time Periods
Next, we turn to breaking down a full sample into separate time periods. We want to test whether European integration has a stronger or weaker effect on the inflow of FDI during different steps of the integration and transition process. Thus, we divide the investment experience of the CEE countries into four shorter periods: 1992–1998, 1999–2004, 2005–2008, 2009–2015, in accordance with the announcements and integration waves. The results of these experiments are summarised in Appendix M.
The first group includes 1992–1998 years. Policy changes in the countries of Central and Eastern Europe were very dramatic. During this time agglomeration appears to be the most important determinant of net FDI inflows. In early 1990s there were many uncertainties in the region, which is why FDI mostly presented itself in the reliable countries which opened up shortly before the collapse of the Soviet Union (Czechoslovakia, Poland and Hungary). As the result, most of the countries of the region became quite attractive for foreign investors because of large domestic markets with relatively high purchasing power. Stern (1997) also examined the patterns of FDI during the first 6 years of transition, and found that FDI was much stronger in countries with macroeconomic stability and stronger reforms (Stern 1997).
We do not find evidence for the presence of correlation between European integration and net FDI inflow during the 1990s. As the economic and political situation was not always stable and clear, the market-seeking motives dominated in the region during that period of time.
Focusing on the latter part of the 1990s, and early years of the new millennium, the motives of foreign investors have changed. The estimated coefficients for this sample are shown in Table 4.28.
The EU accession and acceding countries became more attractive to foreign investors as their economic systems and regulatory frameworks became more similar to those of the Union, and as the dynamic effects of the association agreements begin to manifest themselves. This integration stage increased FDI inflows by 0,64%.
In the third examined period (2005–2008), inflow in CEE grew substantially. 8+2 CEE countries that joined the EU adopted the EU law. The accession countries developed an institutional framework to administer and properly channel the variety of funds available from European Community sources for assisting economic development. Bulgaria, Romania and Croatia also undertook reforms related to judicial independence, accountability, fighting corruption, and tackling of organised crime. All these measures improved the business climate for investors.
Association and partnership agreements also shaped FDI-related policies in various countries such as Albania, Bosnia and Herzegovina, Serbia and Montenegro. EU reforms brought infrastructure investments and gave regulatory stability to the EU single market, but the economic and social costs of adjustment were also high.
The financial and credit crisis significantly impacted the volume of FDI inflows in the fourth period (2009–2015), because it added new uncertainties and risks to the world economy. During the economic and financial crisis, investment flows all over the world dropped due to the reduction of economic activity and loss of confidence in the existing economic and financial system, as a result, many investment plans were cancelled or postponed.
Over this period, we observe the extremely high importance of EU market access and EU-membership dummy variables in comparison to other variables. Neither GDP, nor agglomeration, nor wages interested foreign investors as attraction factors. This supports our assumption that EU membership played a leading hand. This is especially true for periods, when the overall economic decline is observed.
Decomposing the results by sub-regions and time-periods reveals very interesting and important results. This confirms that despite many similar trends in the whole sample there are some specific sub-sample traits that attract investors.
The decomposition results must be, however, considered with a great concern, because of the small sample sizes which affect the reliability of test statistics and limit the variation in dependent and independent variables, therefore preventing efficient estimation.
4.5 Appendix
4.5.1 Appendix D
4.5.2 Appendix E
4.5.3 Appendix F
4.5.4 Appendix G
4.5.5 Appendix H
4.5.6 Appendix I
4.5.7 Appendix J
4.5.8 Appendix K
Fitted FDI inflows vs. observed values based on 2SLS FE estimations. Source: Author’s calculations. Note: Blue line is for estimated values (2SLS FE), red one is for observed values of FDI inflows. The data for The former Yugoslav Republic of Macedonia 2002 and Georgia 1995–1996 are not available
4.5.9 Appendix L
4.5.10 Appendix M
References
Agarwal JP (1997) European integration and German FDI: implications for domestic investment and Central European economies. Natl Inst Econ Rev 160:100–111 http://www.freepatentsonline.com/article/National-Institute-Economic-Review/56908891.html
Ahn SC, Moon H (2002) Large-N and large-T properties of panel data estimators and the Hausman test. http://EconPapers.repec.org/RePEc:cpd:pd2002:a6-2
Altomonte C (1998) FDI in the CEEC’s and the theory of real options: an empirical assessment. https://ideas.repec.org/p/lic/licosd/7698.html
Altomonte C (2007) Regional economic integration and the location of multinational firms. Rev World Econ 143(2):277–305
Altomonte C, Guagliano C (2001) Competing locations? Market potential and FDI in Central and Eastern Europe vs the Mediterranean. https://ideas.repec.org/p/lic/licosd/10801.html
Altomonte C, Guagliano C (2003) Comparative study of FDI in Central and Eastern Europe and the Mediterranean. Econ Syst 27(2):223–246
Altzinger W, Bellak C (1999) Direct versus indirect FDI: impact on home country exports an employment—Empirical evidence for Austrian FDI in CEECs. Globalisation 1999:131–154
Arellano M, Bond S (1991) Some tests of specification for panel data. Monte Carlo evidence and an application to employment equations. In: The review of economic studies, vol 194, pp 277–297
Asiedu E (2002) On the determinants of foreign direct investment to developing countries: is Africa different? World Dev 30(1):107–119
Atkinson AC (1987) Plots, transformations, and regression: an introduction to graphical methods of diagnostic regression analysis. Clarendon Pr, Oxford. ISBN: 0198533713
Baltagi BH, Li Q (1995) Testing AR(1) against MA(1) disturbances in an error component model. J Econ 68(1):133–151. https://doi.org/10.1016/0304-4076(94)01646-H
Benacek V, Gronicki M, Holland D, Sass M (2000) The determinants and impact of foreign direct investment in Central and Eastern Europe: a comparison of survey and econometric evidence. Transl Corp 9(3):163–212 http://pure.iiasa.ac.at/5953/
Bera AK, Walter SE, Yoon MJ (2001) Tests for the error component model in the presence of local misspecification. J Econ 101(1):1–23
Berglof E, Bolton P (2002) The great divide and beyond: financial architecture in transition. J Econ Perspect 16(1):77–100
Berglöf E, Roland G (1997) The EU as an “outside anchor” for transition reforms. SITE working paper
Bevan AA, Estrin S (2000) The determinants of foreign direct investment in transition economies, 2638
Bevan AA, Estrin S (2004) The determinants of foreign direct investment into European transition economies. J Comp Econ 32(4):775–787
Blonigen BA, Davies RB, Waddell GR, Naughton HT (2007) FDI in space. Spatial autoregressive relationships in foreign direct investment. Eur Econ Rev 51(5):1303–1325
Bos J, van de Laar M (2004) Explaining foreign direct investment in Central and Eastern Europe: an extended gravity approach, 8
Brenton P (1999) Removing obstacles to deeper economic integration in the broad European trade area: Report of the CEPS working party on trade and investment in an enlarging Europe. CEPS, Brussels. ISBN: 9290792574
Brenton P, DiMauro F, Lücke M (1998) Economic integration and FDI: an empirical analysis of foreign investment in the EU and in Central and Eastern Europe. http://EconPapers.repec.org/RePEc:kie:kieliw:890
Carstensen K, Toubal F (2004) Foreign direct investment in Central and Eastern European countries: a dynamic panel analysis. J Comp Econ 32(1):3–22
Clark P, Crawford C, Steele F, Vignoles A (2010) The choice between fixed and random effects models: some considerations for educational research. CMPO, Bristol
Clausing K, Dorobantu C (2005) Re-entering Europe: does European Union candidacy boost foreign direct investment? Econ Transit 13(1):77–103 http://EconPapers.repec.org/RePEc:bla:etrans:v:13:y:2005:i:1:p:77-103
Cook RD, Weisberg S (1999) Applied regression including computing and graphics. Wiley, New York, NY. ISBN: 9780471317111. http://www.loc.gov/catdir/description/wiley037/99017200.html
Cragg JG, Donald SG (1993) Testing identifiability and specification in instrumental variable models. Econ Theory 9:222
Croissant Y, Millo G (2008) Panel data econometrics in R: the plm package. J Stat Softw 27(2)
Dickey DA, Fuller WA (1979) Distribution of the estimators for autoregressive time series with a unit root. J Am Stat Assoc 74(366):427
Dunning JH (1973) The determinants of international production. Oxf Econ Pap 25(3):289–336 http://EconPapers.repec.org/RePEc:oup:oxecpp:v:25:y:1973:i:3:p:289-336
Faeth I (2009) Determinants of foreign direct investment—A tale of nine theoretical models. J Econ Surv 23(1):165–196
Fox J, Weisberg S (2011) An R companion to applied regression. SAGE, Los Angeles. ISBN: 141297514X
Fujita, M, Venables, A, Krugman, PR (1999) The spatial economy. Cities, regions and international trade. Cambridge, Mass: MIT Press. Online verfügbar unter http://search.ebscohost.com/login.aspx?direct=true&scope=site&db=nlebk&db=nlabk&AN=9243.
Giles D (2011) Dummies for dummies. http://davegiles.blogspot.de/2011/03/dummies-for-dummies.html. Accessed Mar 2017
Globerman S, Shapiro DM (1999) The impact of government policies on foreign direct investment: the Canadian experience. J Int Bus Stud 30(3):513–532
Globerman S, Shapiro D (2002) Global foreign direct investment flows: the role of governance infrastructure. World Dev 30(11):1899–1919
Greene WH (2003) Econometric analysis. Prentice-Hall, Upper Saddle River, NJ. ISBN: 0-13-110849-2
Harris CD (1954) The market as a factor in the localization of industry in the United States
Harrison MJ, McCabe BPM (1979) A test for heteroscedasticity based on ordinary least squares residuals. J Am Stat Assoc 74(366):494
Head K, Mayer T (2004) Market potential and the location of Japanese investment in the European Union. Rev Econ Stat 86(4):959–972
Heckman JJ (1979) Sample selection bias as a specification error. Econometrica 47(1):153
Holland D, Pain N (1998) The determinants and impact of foreign direct investment in the transition
ILO (2016) ILO database of labour statistics: mean nominal monthly earnings of employees. http://ilo.org/global/statistics-and-databases/lang--en/index.htm
ILOSTAT (2016) Database: mean nominal monthly earnings of employees. http://www.ilo.org/ilostat/faces/oracle/webcenter/portalapp/pagehierarchy/Page137.jspx?_afrLoop=552230679810060&clean=true#!%40%40%3F_afrLoop%3D552230679810060%26clean%3Dtrue%26_adf.ctrl-state%3Dpak6ybvwc_9
International Monetary Fund (2016) World Economic Outlook Database. http://www.imf.org/external/pubs/ft/weo/2016/02/weodata/index.aspx. Accessed 25 Nov 2016
Iwasaki I, Suganuma K (2009) EU enlargement and foreign direct investment into transition economies revisited. Transl Corp 18(3):27–57
Janssens S, Konings J (1996) How do Western companies respond to the opening of Central and East European economies? Survey evidence from a small open economy-Belgium. Katholieke Universiteit, Leuven
Kinoshita Y, Campos N (2003) Why does FDI go where it goes? New Evidence from the transition economies. IMF Working Papers. International Monetary Fund (03/228). http://EconPapers.repec.org/RePEc:imf:imfwpa:03/228
Kleibergen F, Paap R (2002) Priors, posteriors and bayes factors for a Bayesian analysis of cointegration. J Econ 111(2):223–249
Kleiber C, Zeileis A (2008) Applied econometrics with R. Springer, New York. ISBN: 978-0-387-77316-2. https://doi.org/10.1007/978-0-387-77318-6
Krkoska L (2001) Foreign direct investment financing of capital formation in Central and Eastern Europe. http://www.ebrd.com/downloads/research/economics/workingpapers/wp0067.pdf
Lankes H-P, Venables AJ (1996) Foreign direct investment in economic transition: the changing pattern of investments. Econ Transit 4(2):331–347
Lansbury M, Pain N, Šmídková K (1996) Foreign direct investment in Central Europe since 1990: an econometric study. Natl Inst Econ Rev 156(1)
Meyer K (1998) Direct investment in economies in transition. University, Dissertation – London. Elgar, Cheltenham. ISBN: 978 1 85898 736 1
Phillips PCB, Perron P (1988) Testing for a unit root in time series regression. Biometrika 75(2):335–346
Puga D, Venables AJ (1997) Preferential trading arrangements and industrial location. J Int Econ 43(3–4):347–368
Resmini L (2000) The Determinants of foreign direct investment in the CEECs: new evidence from sectoral patterns. Econ Transit 8(3):665–689
Roland G, Verdier T (2003) Law enforcement and transition. Eur Econ Rev 47(4):669–685
Rolfe RJ, Woodward DP (2004) Attracting foreign investment through privatization: the Zambian experience. J Afr Bus 5(1):5–27
Savary J (1997) The French firms motivations for investing in Eastern Europe countries. A study of investors and potential investors. In: Witkowska J, Wysokinska Z, Zofia (eds) Motivations of foreign direct investors and their propensity to exports in the context of European integration process. University of Lódz, Lódz, pp 43–76
Stern N (1997) The transition in Eastern Europe and the former Soviet Union: some strategic lessons from the experience of 25 countries over six years
Stock JH, Yogo M (2005) Testing for weak instruments in linear IV regression. In: Identification and inference for econometric models : essays in honor of Thomas Rothenberg. Cambridge University Press, Cambridge, pp 80–120
The World Bank (2016) National accounts: output and expenditure. http://econ.worldbank.org/WBSITE/EXTERNAL/DATASTATISTICS/0,print:Y~isCURL:Y~contentMDK:20451590~menuPK:64133152~pagePK:64133150~piPK:64133175~theSitePK:239419,00.html
Thursby JG (1982) Misspecification, heteroscedasticity, and the chow and Goldfeld-Quandt tests. Rev Econ Stat 64(2):314
UNCTAD (2003) World investment report: FDI policies for development: national and international perspectives. United Nations, New York. ISBN: 92-1-112580-4
UNCTAD (2004) World investment report: the shift towards services. United Nations, New York. ISBN: 92-1-112644-4
UNCTAD (2016c) UNCTADstat: Foreign direct investment: inward and outward flows and stock, annual, 1970–2015 Table summary. http://unctadstat.unctad.org/wds/TableViewer/tableView.aspx?ReportId=96740
UNCTAD (2016d) UNCTADstat: Gross domestic product: total and per capita, current and constant (2005) prices, annual, 1970–2015. http://unctadstat.unctad.org/wds/TableViewer/tableView.aspx?ReportId=96
Vernon R (1966) International investment and international trade in the product cycle. Q J Econ 80(2):190–207
Wooldridge JM (2010) Econometric analysis of cross section and panel data. MIT Press, Cambridge, MA. ISBN: 9780262232586. http://search.ebscohost.com/login.aspx?direct=true&scope=site&db=nlebk&db=nlabk&AN=550663
Wooldridge JM (2012) Panel data models with heterogeneity and endogeneity: programme evaluation for policy analysis. https://www.ifs.org.uk/docs/wooldridge%20session%204.pdf
Wooldridge JM (2013) Introductory econometrics: a modern approach. South-Western Cengage Learning, Mason, OH. ISBN: 978-1-111-53104-1
Author information
Authors and Affiliations
Rights and permissions
Copyright information
© 2018 Springer International Publishing AG, part of Springer Nature
About this chapter
Cite this chapter
Makhavikova, H. (2018). Empirical Model. In: Determinants of FDI in Central and Eastern Europe. Contributions to Economics. Springer, Cham. https://doi.org/10.1007/978-3-319-91878-5_4
Download citation
DOI: https://doi.org/10.1007/978-3-319-91878-5_4
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-319-91877-8
Online ISBN: 978-3-319-91878-5
eBook Packages: Economics and FinanceEconomics and Finance (R0)