FDI Flows, Exchange Rates and Crisis: Determinants of Inward Foreign Direct Investment of the Euro Members

(1)

FDI Flows, Exchange Rates and Crisis:

Determinants of Inward Foreign Direct

Investment of the Euro Members

Master Thesis

University of Groningen

Faculty of Economics and Business

University of Göttingen

Faculty of Economic Sciences

Student: Alexander Vey Student Number: S2804301 E-‐Mail: a.vey@student.rug.nl Supervisor (1st_{):
Sorin
Krammer}

Supervisor (2nd_{):
Christian
Bruns}

(2)

Abstract

This paper utilizes a panel data set of north-north FDI flows to gauge the effect of exchange rate movements and the occurrence of financial crises in source countries on inward FDI among euro members. In order to shed some light on these issues and to explain cross-country differences in the magnitude of FDI flows a gravity model is estimated. The analysis suggests that macroeconomic variables accounting for motives of horizontal FDI are important to explain differences in inward FDI between the host countries. Moreover, it is found that exchange rate movements have a significant and robust impact on inward foreign direct investment in the currency union.

Keywords: Foreign Direct Investment, Determinants, Gravity Model

(3)

List of Abbreviations

EU – European Union

FDI – Foreign Direct Investment FE – Fixed Effects

GDP – Gross Domestic Product IMF – International Monetary Fund M&A – Mergers and Acquisitions MNE – Multinational Enterprise

OECD – Organisation of Economic Co-operation and Development OLS – Ordinary Least Squares

(5)

List of Tables

Table 1 – Summary Statistics p. 21

Table 2 – Pooled OLS regressions p. 24

Table 3 – Pooled OLS regressions with fixed effects p. 27

Table 4 – FDI Flow (Robustness Check Dependent Variable) Annex B2

Table 5 – Stock of FDI (Robustness Check Dependent Variable) Annex B3

Table 6 – Robustness Check Independent Variable Annex B4

(6)

1. Introduction

Currently, several European countries are still dealing with the aftermath of the latest financial crisis starting in 2007, which had been triggered by the default of US-subprime loans. Starting with the bankruptcy of Lehman Brothers, the global financial system has seen a crisis that has been extraordinarily dangerous for individuals, banks and entire states. Compared to most former financial crises, which had been relatively regionally constrained, the financial crisis in 2007-09 was extremely contagious and led to developed countries tumbling into recessions all over the globe. In Europe, this crisis was promptly succeeded by a sovereign debt crisis. The occurrence of those two financial crises has put a lot of stress on the Euro currency members. Due to the commitment to a mutual currency, countries could not make use of individual monetary policy measures in order to relief some pressure on their economies. In fact, some countries were hit really hard because European monetary authorities had not been well prepared to deal with such a huge and contagious crisis. The growing inter-linkages in the world banking system, as well as the abandoning of individual monetary policy measures among euro member countries makes it necessary to derive a deeper understanding of how individuals, firms and financial markets act in troublesome times.

(7)

currency union, and how can one explain differences in the magnitude of foreign direct investment flows? To answer these questions, a gravity-approach with focus on macro-economic variables is employed. Specifically, some light is shed on the questions i) what host country-specific factors might explain differences in the magnitude of FDI flows across the Euro members ii) how foreign direct investment in the currency union is affected by the commitment to a mutual currency in terms of exchange rate movements and iii) how financial crises among developed countries affect foreign direct investment in the currency union. It is found that exchange rate movements have a highly significant impact in explaining FDI flows from OECD countries into the monetary union. Moreover, the underlying analysis suggests that the currency union is an attractive destination for horizontal FDI.

(8)

2. Literature Review

This chapter shall briefly outline the theory that is used to explain FDI flows, introduce possible determinants of foreign direct investment derived from this theory, and provide a set of testable hypothesis. While reviewing the literature, special attention is paid to the relationship between financial crisis, exchange rates and foreign direct investment.

The term foreign direct investment generally describes the economic cross-border engagement of private firms. The main difference between FDI and portfolio investments (PI) is that the former is generally associated with a long-lasting interest of the investor as well as some element of control. Thus, to be classified as FDI by international standards, at least 10 per cent of voting shares have to be acquired.1 Foreign direct investment can happen in different ways: As building factories from scratch (greenfield investment), acquisition of already existing physical property (brownfield investment), mergers between two firms or simply through acquisition of voting shares (Gilroy & Lukas, 2005).

2.1 Vertical and Horizontal FDI

The theoretical justification of the country-specific factors that may influence foreign direct investment in the currency union is derived from the theory of vertical and horizontal FDI. Note, that there are several competing theories trying to explain the occurrence of foreign direct investment between two countries. The most important ones among these other theories explaining FDI are probably the OLI-Framework2 of Dunning (1979, 1980), which incorporates several insights from earlier theories, and the knowledge-capital theory. Among all theories explaining FDI, there is no theory that is strictly superior over another - rather they focus on different aspects of foreign direct investment and should be seen as complements rather than substitutes (Faeth, 2009). Due to the macro-economic setting of this paper, the theory of vertical and horizontal FDI is preferred over the OLI-Framework.

1_{If e.g. less than 10 per cent of the shares of a foreign firm would be acquired, the transaction would be}

classified as PI.

2_{The OLI-Framework states that a firm needs a ownership, location or internalization advantage in order to}

(9)

2.1.1 Horizontal FDI

Generally, the term horizontal FDI refers to multi-national enterprises (MNEs) that enter a foreign country in order to produce the same product or service as in their home country with the aim to serve the foreign market via local production. Horizontal FDI is mainly driven by a MNE’s desire to gain access to markets and is also referred to as market-seeking in the literature (Busse & Hefeker, 2007). Underlying reasons for this are to be near customers, realize returns to scale or bypass barriers to trade such as tariffs and transport costs. The former is in so far important as nearness to customers helps firms to adapt their products to the foreign market. In a world with no barriers to trade, as well as low or even no transportation cost, a MNE would concentrate its production and choose to serve the foreign market via exporting instead of local production (Fukao & Wei, 2008). Therefore, horizontal FDI is often assumed to act as a substitute for exporting and licensing (Beugelsdijk, Smeets & Zwinkels, 2008). The proximity-concentration hypothesis states that MNEs face a trade-off between minimizing the distance to customers and realizing economics of scale (Faeth, 2009). It is found that important determinants of horizontal FDI are trade barriers such as tariffs (Busse & Hefeker, 2007) and transport cost (Bevan & Estrin, 2004) as well as market size (Busse & Hefeker, 2007; Janicki & Wunnava, 2004; Buckley, Clegg, Cross, Liu, Voss & Zheng, 2007) and wages of the host economy (Botric & Skuflic, 2006; Janicki &Wunnava, 2004). A big market, as well as high wages signal attractive business opportunities. Walsh and Yu (2010) argue that a high currency valuation might act as an entry barrier for horizontal FDI. Assuming economics of scale and sufficiently high trade costs, the theory of horizontal FDI is able to explain multinational activity between similar countries (Markusen & Venables, 2000; Markusen & Maskus, 2001). Given the relatively homogenous group of source and host countries that is reviewed in this paper, it is reasonable to assume that a significant proportion of inward FDI in Europe is due to a horizontal motive.

(10)

2.1.2 Vertical FDI

Efficiency seeking is the main motivation of the vertically integrating firm. The vertical model of FDI assumes that differences in international factor prices are important for the location choice of a MNE following this strategy of direct investment. According to Baldwin (2006) the “great unbundlings” shaped the landscape of firms substantially. The first unbundling was driven by a drastic decrease in transportation costs and describes the separation of customers and producers. The second unbundling, starting in the 1960s, was driven by rapidly falling communication and coordination costs and refers to the separation of production stages. This means, that there was no need for performing manufacturing stages near to each other anymore (Baldwin, 2006). In order to exploit international differences in factor abundance and prices, MNEs split up their value chains and became “truly global” (Gereffi & Fernandez-Stark, 2011). The term vertical FDI is therefore mostly associated with the offshoring of low-value added activities of a firm’s value chain to low-cost countries (Timmer, Los, Stehrer & de Vries, 2013). In theoretical models of FDI, a multinational often engages in headquarter activities in countries that are abundant in physical and human capital, whereas production activities are carried out in low-skilled labour-abundant countries (Markusen, 1997). Production processes are fragmented and located in countries that provide the most favourable environment (Buckley et. al., 2007).

For these reasons, the location determinants of vertical FDI are likely to differ from those of horizontal FDI. Important determinants of vertical FDI are differences in labour costs (Bevan & Estrin, 2004; Vijayakumar, Sridharan & Rao, 2010), openness (Vijayakumar et. al, 2010), infrastructure (Vijayakumar et. al, 2010), resource availability (Buckley et. al., 2007), workforce and transport costs (Fukuao & Wei, 2008; Busse & Hefeker, 2007). Due to the focus on rather developed countries, it seems reasonable to consider also factors like human-capital and research capabilities as driving forces of vertical FDI, even though the offshoring of high-value added activities is not seen often.3 The finding of Markusen and Maskus (2001) that similarities rather than differences between countries account for most cross-border firm activity leads to the believe that the majority of inward FDI in the currency union is likely to be not of vertical nature.

Hypothesis (H2): Foreign direct investment by non-euro OECD countries in the currency union is not substantially driven by motives of vertical FDI.

3_{The offshoring of high-value added activities technically also belongs to vertical FDI even though it is more}

(11)

2.2 Exchange Rate Movements

For a long time economists like Mundell (1968) argued that exchange rate movements should not affect firms’ decision to invest in foreign countries. This reasoning was based on the assumption of perfect capital markets. The assumption of perfect capital markets states that capital is fully mobile around the globe or in other words: All investors and firms would have equal access to all capital markets. This in term would lead to the equalization of risk-adjusted expected returns of all international assets – and it was assumed that returns on assets rather than their prices were the only thing that should matter for the acquisition of foreign assets. Since assets usually yield returns in their denominated currency, exchange rate movements should have no impact on FDI flows.

However, Froot and Stein (1991), being among the first who challenged the assumption of perfect capital markets, argued that with informational asymmetries MNEs are not able to finance investments solely by external funds. This implies that exchange rate movements can change the relative wealth position between domestic and foreign investors. Thus, a depreciation of the host country currency can provide an advantage for foreign firms over domestic firms when it comes to the bidding for assets in the potential host country. At the heart of the discussion is the question why foreign firms should value an asset more than domestic firms. Froot and Stein (1991) provide a first transmission channel why this could be the case. Another link between FDI and exchange rate movements arises from imperfections in good markets. Bloningen (1997) argued that if firms do not have equal access to all good markets and therefore differ in their opportunities to generate revenues from firm-specific assets, it might be the case that their individual valuation of a firm-specific asset differs. Even though both firms bid in the same currency for a firm-specific asset, it is possible to due good market frictions that they generate returns in different currencies on the same asset. This might be the case if a certain technology is acquired. This contradicts the formerly held belief that foreign direct investment is somewhat similar to the acquisition of bonds that yield returns only in their denominated currency. So, exchange rate movements can change the reservation bid of foreign firms for domestic assets. For example, Buch and Kleinert (2006) find support for the goods market friction hypothesis.

(12)

periods of crisis. Urata and Kiyota (2004) also find that a depreciation of a currency promotes FDI inflows. They argue that an appreciation of the home country currency lowers the production cost in the foreign country as well as the prices foreign investors have to pay for host country assets. A positive impact of a currency devaluation on FDI flows is also found by e.g. Udomkerdmongkol, Morrisey and Görg (2009).

Hypothesis 3 (H3): A devaluation of the Euro increases FDI inflows in the Euro member countries.

2.3 Financial Crisis

In this paper a systemic financial crisis is characterized by two criteria: Firstly, significant signs of financial distress in the banking system indicated by bank runs, losses in the banking system and/or bank liquidations. Secondly, significant banking policy intervention measures in response to significant losses in the banking system. By using this definition Laeven and Valencia (2012) is followed.

2.3.1 FDI Flows During Financial Crisis

(13)

with up-stream activities can posses a crucial function within the value chain of a multinational enterprise. So, the production of an intermediate product that plays a major role in the processing of a MNEs final good could be assigned to a foreign subsidiary, making it harder to replace this subsidiary. Hence, subsidiaries that are closely linked to the main business activities of an enterprise are more resistant to short-term changes (Hill & Jongwanich, 2009; Alfaro & Chen, 2010). Hill and Jongwanich (2009), as well as Lipsey (2001) find that in response to crisis, affiliates of foreign firms may engage in market-switching to withstand troublesome times. Additionally, foreign affiliates can rely on already existing networks and better access to international markets, compared to local firms, making them more robust against financial and economic distress (Alfaro & Chen, 2010). Beginning with Krugman (2000), there is also a substantial strand of literature dealing with the phenomena of fire-sale FDI4 (Aguiar & Gopinath, 2005; Acharya & Shin, 2009).

All these arguments supplement the view that foreign direct investment is a flow of long-term capital and that long-term returns are what is important for MNEs (Athukorala, 2003). Especially, when looking at green-field FDI projects, which may take a while until they are up and running, it becomes clear that MNEs engaging in foreign direct investment must possess a longer time horizon than portfolio investors. Implementing FDI is time consuming (Bevan & Estrin, 2004). Hence, FDI flows appear to be more stable or resilient in times of financial crisis compared to other private capital flows.

2.3.2 Financial Crisis and Real Economy

This section is intended to shed some light on how a financial crisis might affect the real economy. It is important to outline the effects which financial crisis might have in the host economies of direct investment in order to make an attempt to distinguish those effects from the effects of financial crisis in the source economy itself. Blot, Bayon, Lemoine and Levasseur (2009) outline four transmission channels of how financial shocks translate into the real economy. Two of them seem to be particularly interesting regarding the foreign investment decision of firms: The capital cost channel and the uncertainty channel.

For the capital cost channel or interest rate channel to work, one has to assume some kind of price and wage stickiness. If so, a rise in interest rates would lead to an increased cost of capital, thereby suppressing investments of firms, which results in less aggregated demand

4_{During the Asian financial crisis a substantial amount of short-term capital left the region and major sell-offs of}

(14)

and ultimately decreases output. Even though financial shocks during the latest crisis mainly affected the short-term interest rate, those changes are able to affect the long-term interest rate, which is assumed to be of primary interest for households and firms, by changing the term-structure of interest (Blot et. al, 2009). Closely related to this channel are so-called credit crunches. A credit crunch refers to the drastic decrease of private credit provided by the financial sector. Even though interest rates are low, the reluctance of banks to finance investment suppresses investments of firms and virtually raises the cost of capital since firms are denied access (Tong & Wei 2009). This is also supported by Dell’Ariccia, Detragiache and Rajan (2007) who find that the performance of sectors that are more reliant on external financing perform worse during times of financial turmoil.

The second important channel through which financial crisis might effect the behaviour of firms is the uncertainty channel. Beyond its direct financial impact, the occurrence of financial crisis also raises the uncertainty of firms about future growth perspectives and government actions. In each period, firms, but also households, need to make a decision between investment/consumption and saving. In times of a risky environment, households tend to save more, leading to firms delaying planned investments. Furceri and Mourougane (2009) find that financial crises have a significant negative effect on potential output.

So, economic literature suggests that foreign direct investment exhibits some sort of resilience against financial distress for which reason it is assumed that the occurrence of financial crisis should have no significant effect on FDI flows.

Hypothesis (H4): The occurrence of financial crises in source countries exhibits no effect on foreign direct investment in potential host countries.

2.4 Theoretical Framework

(15)

Stein, 2003). However, in economics, the masses of objects (countries) are reflected by their gross domestic products (GDP). In economics, GDP can be interpreted as the mass respectively weight of a country in the global economy. The simple gravity equation looks like the following:

𝐹𝐷𝐼_!" = 𝑐 𝐺𝐷𝑃! ∗ 𝐺𝐷𝑃!

𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒_!" (𝐈)

Where 𝐹𝐷𝐼_!" is the flow of foreign direct investment from country i to country j in year t, 𝐺𝐷𝑃! is the GDP of source country i in year t, 𝐺𝐷𝑃! is the GDP of the host economy j in year t and 𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒!" represents the distance between country i and j. C represents a gravity

constant. Note that time indices are suppressed in equation (I).

This equation implies that expected FDI flows between two countries should be positively affected by their respective GDPs and negatively related to the distance between them. The striking feature of this approach in economics is that the GDP of an economy, as well as distance exhibit a significant and robust impact across a wide range of empirical models explaining bilateral trade and FDI flows. As far as the economic literature on FDI is concerned, this baseline equation, in combination with a wide range of other possible variables, has been used to answer the question where capital is going (Faeth, 2009).5

5_{Generally, economists assume that foreign direct investment is beneficial for economic growth and}

(16)

3 Data and Empirical Methodology

This chapter is meant to specify the final sample, introduce the dependent as well independent variables used in the empirical model and outline the reasoning for the selection of econometric methods.

3.1 Data Sources and Specification of Final Sample

The initial data set contains data for 19 OECD member countries that posses a currency other than the Euro and 12 European countries that belong to the monetary union. The initial panel data set covers a period of 19 years, starting in 1995 and ending in 2013. The data is mainly derived from EUROSTAT databases, the World Bank World Development Indicator Database, and the OECD Library. A detailed attribution of variables to distinct sources can be found in annex A1 and A2. In the underlying data set, non-euro OECD countries represent the source countries of foreign direct investment whereas the euro-currency members are the recipients of those capital flows. OECD countries were chosen as source and host countries because the most recent crisis had a notable impact on all of them, which makes it a particularly interesting set to conduct analyses on. Additionally, restricting the attention to OECD countries minimizes in a first attempt the issues of zero and non-reported flows that are general concerns of gravity based models. This is important because it is found that a great amount of those are likely to cause biased results in those kinds of models (Bellak et. al., 2010).

Due to a considerable number of missing observations of the dependent variable for the years before 1999 and after 2012 those years are excluded from the initial sample, which reduces the covered period to 14 years. 6 Since the Euro as mutual currency was formally introduced in 1999, this year is believed to represent a proper starting point for the underlying analysis. By ensuring that every year is adequately represented in the final sample, consistency of insights derived from the empirical estimation will be strengthened. For the same underlying reasoning, Greece is dropped from set of host countries because more than 50 per cent of all observations for the dependent variable were reported as missing for those 14 years.

6_{More than 50 percent of all observations for the bilateral FDI flows from source to host country are reported as}

(17)

Moreover, it shall be noted that it is tempting to also remove Luxembourg from the set of host countries because observations for the dependent variable of Luxembourg exhibit significantly larger values compared to all other host countries in the sample. Before the transformation of the dependent variable, Luxembourg is the source of several outliers. These large values might be explained by the existence of several holding companies located in Luxembourg (Lipsey, 2006). The most striking outlier is Luxembourg’s reported FDI flow from the United States in 2013, which amounts to 250 million Euro. This is almost as much as the total inflows of all host countries in the previous year combined. But since years after 2012 are excluded anyway and the remaining outliers after transformation do not appear too severe anymore Luxembourg is not excluded from the final sample. Instead, for the sake of sample size, it is controlled for being Luxembourg by introducing a country dummy later on to which special attention will be paid. After reducing the initial sample size, the final data set contains information on 19 source countries, 11 host countries and covers a timespan of 14 years. A list of the source and host countries can be found in annex A3.

3.2 Variable Specification

This section is meant to specify the variables used in the empirical estimation in the next chapter. This includes potential transformations of variables, short definitions as well as a brief reasoning for including these variables.

3.2.1 Dependent Variable

As dependent variable yearly bilateral FDI flows are chosen, which are reported in million Euros per annum. Since bilateral FDI flows are flow measures, which means that they are measured between two points in time, it is assumed that FDI flows exhibit some kind of timing properties. Unlike the Stock of FDI, single FDI flows do not contain any (or at least significantly less) information about previously executed direct investments. It is believed that this feature makes them especially suitable to investigate time-specific determinants of foreign direct investment over short periods of time and to gauge the effect of recent events such as the occurrence of financial crisis on FDI.

(18)

stock of FDI is also found to be significant in explaining cross-country differences in the magnitude of FDI flows by e.g. Kiyota and Urata (2004), which implies that the stock of FDI contains information about (future) FDI flows. A large stock of FDI might serve as (public) signal for MNEs that indicates an attractive destination. Especially, in less developed countries where the acquisition of information is relatively costly and the environment is uncertain MNEs might use the stock of FDI to choose between a set of countries. Compared to the flow measure the stock of FDI possess the minor drawback that stocks are valued at historic cost which may cause changes due to changes in valuation (Bellak. et. al., 2008). As one will see later, the estimated effect of crisis on FDI is in fact sensitive to the decision to use FDI flows instead of stocks. In chapter 4.3, the results of the model with bilateral FDI stocks will be presented.

Proposition I:

The stock of FDI is more appropriate to evaluate country-specific determinants and cross-country differences whereas the flow of FDI is more suitable to investigate the time-specific properties of FDI.7

Due to the occurrence of zero and negative values of the dependent variable it, is not possible to simply transform it into natural logarithmic form (Schmidt & Boll, 2010). One possible way to deal with this issue is to exclude all observations that cannot be transformed into logarithmic form as some authors suggest (e.g. Bellak et. al. 2008). However, it is believed that those (negative) observations may contain important information and simply deleting them could distort the results (Marián & Vilma, 2011). In order to deal with this issue the proposed transformation of Levy-Yeati et. al. (2003) is adopted:

𝐿𝐹𝐷𝐼 = 𝑠𝑖𝑔𝑛 𝐹𝐷𝐼 ∗ log (1 + 𝐹𝐷𝐼 )

Even though this transformation may cause distortions in interpreting small values of the dependent variable, larger values can still be interpreted as elasticities. By using the log of FDI, e.g. Takagi and Shi (2011) and Frenkel et. al. (2004) are followed who also use the log of bilateral (real) FDI flows as dependent variable.

7_{Albeit in economic literature FDI flows are also commonly used to evaluate country-specific determinants of}

(19)

3.2.2 Independent Variables

Since one idea of this paper is to investigate host-country determinants that according to the theory may explain cross-country differences, two sets of host country-specific variables are introduced - two variables for each hypothesis concerned with location factors of FDI (H1 & H2). Additionally, a variable that measures the impact of exchange rate movements (H3), a set of variables controlling for the impact of financial crisis in the host economy, and a crisis dummy (H4) is added to the baseline equation (I).

Distance

Including distance in a model explaining FDI flows is rooted in the idea that distance is an important control for transaction costs and cultural difference (Frenkel et. al. 2004). A greater distance may act as an obstacle to FDI since a greater distance increases the cost of communication, the cost of information, and hinders face-to-face communication (Buch, Kleinert & Toubal, 2003; Portes & Rey, 2000). Even though a sample of developed countries is investigated, cultural differences might still be present, which increase with distance. Furthermore, distance may act as a measure for transport costs. Therefore, the coefficient of

dist is believed to be negative. In order to make the effect more feasible to interpret, distance

is transformed into logarithmic form. The distance measures are taken from the CEPII GeoDist database and are reported in km. The CEPII measure of distance does not only account for the distance between two countries but also takes the distribution of cities within source and host country into account. It is believed that this measure reflects distance between two countries more accurately than just taking distances between e.g. capitals or borders.

Gross Domestic Product

The underlying reasoning for using variables controlling for the size of source as well as host country markets is, that larger economies are expected to not only exhibit a greater outflow of capital, but also receive more investments (Vijayakumar et. al., 2010). Market size also serves as an indicator for a country’s product demand and supply capacity (Bevan & Estrin, 2004). A large market may also appear more attractive to MNEs because it signals business opportunities. Market sizes are assumed to be decisive in determining the magnitude of bilateral capital flows.8 GDP or GDP per Capita are the most common proxies for market size

8_{It should be critically mentioned that GDP can proxy for a wide range of other country-specific factors such as}

(20)

in economic models. In this paper, GDP is used as a proxy for market size and is defined as real GDP in million US $ in year t. Real GDP is used to account for changes in the level of prices in the source, respectively, host country. In order to make the effect more feasible to interpret and to account for large differences in GDP across countries, the GDP variables are transformed into logarithmic form. The coefficients for the level of host as well as source GDP are expected to be positive.

So, after transforming the independent variable as well as the gravity variables the empirically testable model of equation (I) becomes:

𝐿𝐹𝐷𝐼_!"# = 𝛽_!ln 𝑑𝑖𝑠𝑡_!" + 𝛽_!ln 𝐺𝐷𝑃_!" + 𝛽_!ln 𝐺𝐷𝑃_!" + 𝜀 _!"# (𝐈𝐈)

Thereby, 𝜀 represents the error term. Due to transformation into logarithmic form the coefficients can be interpreted as elasticities. Note that transforming the equation in logarithmic form causes the gravity constant c to drop out (Walsh & Yu, 2010; Janicki & Wunnava, 2004)

Horizontal FDI - Wages & Tariffs (H1)

Two host-country determinants are derived from the theory of horizontal FDI. Wages and tariffs are meant to reflect the impact of horizontal FDI (H1). The variable wage is defined as the average annual wage of host country j in year t. High wages are assumed to signal attractive business opportunities for MNEs since the purchasing power of the population is greater (Busse & Hefeker, 2007). Wages adjusted for purchasing power parity are used to capture this effect across countries most accurately. Specifically, wages are measured in 2013 constant prices at 2013 USD purchasing power parity. Additionally, the variable wage is transformed into logarithmic form. Thus, the coefficient of wage can be interpreted as elasticity and is expected to be positive, since high wages are assumed to be attractive for the horizontally investing firm.

(21)

be positive because the theory of vertical FDI states that higher trade barrier makes it relatively more profitable to enter foreign markets directly.

Vertical FDI - Labour Cost & Knowledge-Seeking (H2)

Since vertical FDI is driven by efficiency seeking, it is plausible to assume that labour costs are an important determinant of this type of FDI. Therefore, the variable ulc is added to the model. ulc is defined as the relative unit labour cost in country j in year t. The coefficient of

ulc is assumed to be negative since a higher cost of labour is assumed to act as an obstacle for

the vertically integrating firm because it directly decreases profitability (Bevan & Estrin, 2004). On the other hand the behaviour of MNEs could be driven by the availability of educated workforce and firm-specific knowledge in the host country. It is reasonable to assume that skilled labour is more expensive than unskilled labour and so it is important to control for technology-seeking motive of the multinational enterprise. Otherwise, the unit labour cost might not be able to capture the effect accurately and might even appear with the wrong sign (Bellak et. al., 2008). Thus, a variable measuring the high-technology exports of country j in year t as per cent of total exports of country j in year t is added to the model. This measure is preferred over e.g. years of education, patents, or number of researchers as it is assumed to reflect more accurately the actual capabilities of the host economy. Years of education do not vary much across the set of host countries whereas the latter two exhibit a strong correlation with GDP. It is assumed that countries with a high percentage of technology exports appear more attractive for the knowledge-seeking MNE and hence a positive coefficient of the variable hightech_ex is expected.

Exchange Rate (H3)

In order to capture the effect of exchange rate movements on foreign direct investment, the bilateral exchange rate is used. The bilateral exchange rate is preferred over the real exchange rate because it is believed that the former reflects changes in bilateral wealth positions more accurately.9 In order to account for the change in prices, the exchange rate is multiplied by the ratio of price levels. So, the bilateral exchange rate fx is defined as the average of monthly observations for the exchange rate in year t in terms of foreign currency times the ratio of price levels. Thereby, I follow the definition used by Schmidt and Boll (2010) and Campa (1993). This definition implies that a rise in the variable fx is associated with a depreciation of the euro and hence a positive coefficient is expected. In a similar manner the bilateral

9_{The real exchange rate is the domestic exchange rate weighted against a basket of major currencies and}

(22)

exchange rate is defined in Takagi and Shi (2011). However, they include monthly observations of the exchange rate for the year t – 1 in their definition, but lack an explanation for doing so. I do not adopt this definition because one could argue that, even though past movements of the exchange rate might influence the investment decision, only the current exchange rate matters when it comes to the actual investment. Furthermore, the actual levels of the exchange rate may also serve as proxy for past expectations of investors.

Crisis Dummy (H4) and Control Variables for Crisis

To capture the effects of financial crises in the source countries, a dummy variable for the occurrence of a financial crisis is added to the model. It is a binary dummy variable that takes on the value of 1 if a financial crisis is reported in the source country. So, it is accounted for fact that some countries were hit earlier by the latest financial crisis than other countries. Thereby, I follow the approach of Furceri and Zdzienicka (2011), Furceri and Mourougane (2009), Dell’Ariccia et al. (2008) and similar to the one of Lee and Min (2011). The values for the dummy variables are taken from the International Monetary Fund (IMF) crisis database (Leaven & Valencia, 2012). It is worth to note that according to the IMF crisis database, all host countries except Finland were subject to financial distress in the years 2008 – 2012. However, it will be abstained from adding a second crisis dummy capturing the effects in host countries because such a dummy is considerably correlated with the source country crisis dummy. Due to the findings that FDI flows are less volatile during troublesome times, compared to portfolio investment, the crisis dummy is expected to be insignificant.

Since not only a sub-set of source countries10, but almost all host countries were subject to the latest financial crisis, two additional variables are added, which might influence a MNE in its investment decision. The first variable is privcredit that is defined as the private credit in the host country over host country GDP in year t. It is divided by GDP to account for the fact that the total amount of credit is bigger in larger economies. This variable is meant to reflect the

true cost of capital (Klein, Peek & Rosengren, 2000). The obvious measure for the cost of

capital would be the interest rate, but the latest financial crisis showed that even though interest rates were low, banks were reluctant to provide credit to the private sector. Thus, the interest rate does not accurately reflect the cost of capital since enterprises had no access to it. One should note that interest rates and the amount of credit is somewhat related, but a

10_{According to IMF crisis data base the following source countries experienced a financial crisis in the period}

under coverage: Czech Republic, Denmark, Hungary, Iceland, Japan, Switzerland, Sweden, Turkey, United

(23)

decrease in interest rates does not necessarily lead to more credit. To overcome this issue, credit to the private sector is used. The coefficient of privcredit is expected to be positive since more credit is associated with more access to finance.

The second important variable being added is govdebt. It is defined as the debt of the central government in million US $ of country j in year t as percentage of country j’s GDP in year t measured in million of US $. Since Europe was not only subject to a banking crisis, but also struggled with increasing sovereign debt, it seems important to include a measure for the debt burden of the host country in the empirical estimation. The underlying reasoning for including this variable, is to model the expectations of foreign firms. As FDI is associated with a long-term interest of the investor, rapidly increasing government debt in the potential host country must lead investors to anticipate that the government will make attempts to increase its future income. Popular government measures to do so are the increase of taxes, cutting tax benefits, the raise of tariffs and the cutting of subsidies – all measures which may have an direct impact on future profitability of direct investment activities. For this reason, the coefficient of

govdebt is expected to be negative.

Common Language & Trade Union Dummy

In addition to the above-mentioned variables, country-pair specific dummy variables controlling for common language and mutual EU membership are added. The variable

comlang_off is added because it is found that common language has a significant impact in

(24)

3.3 Applied Econometric Methods

In order to estimate the empirical model, it was first checked whether the Hausman-test supports the use of the random-effects (RE) model because the RE model would allow for the inclusion of time-invariant variables. Additionally, RE procedure would allow for better control of heteroscedasticity and auto-correlation within the panel data set compared to the pooled OLS model. The Hausman-test compares the coefficient estimates of the FE model with the estimates of the RE model. The FE model assumes that there is a systematic difference in intercepts of each individual (country-pair) whereas the RE model assumes that individuals differ in their error terms. Estimates from both models are consistent, if there is no correlation between the error term of regression and the explanatory variables. If both models yield consistent estimators, then both models should converge in large samples to the same true coefficient estimates. Even if the explanatory variables and the error term of regression are correlated, the FE model still yields consistent estimators whereas the RE estimators would converge to some other values. Thus, coefficient estimates of both models would differ (Adkins & Hill, 2011; Hill, Griffiths & Lim, 2011).

The Hausman-test does not support the use of random-effects model and hence would favour the fixed-effects model (Vijayakumar et. al., 2010). Test results are displayed in annex C2. However, the FE model is not used for the estimation procedure because i) one is not able to estimate the effect of time-invariant variables and ii) the FE model assumes that there is a systematic difference in intercepts for each country pair. Issue i) poses in so far a dilemma as e.g. distance is assumed to be decisive for explaining bilateral FDI flows. Omission of the distance variable distorts the gravity approach and causes the coefficients of the GDP variables to turn negative and insignificant. Thus, leading to different estimates of the FE and RE model and therefore to the rejection of the 𝐻! of the Hausman-test. Issue ii) is problematic

because intercepts in the FE model would reflect host as well as source country characteristics at the same time and hence making a comparison of those difficult, if meaningful at all. Therefore, a pooled OLS regression for the underlying panel data set is chosen.

(25)

confirmation of the underlying regression model can be derived from the RE model (Portes & Rey, 2000). Moreover, it is argued that the assumption of the RE model which states that the observations for each country pair differ in the error terms rather than intercepts is more reasonable to assume in the underlying case. The Breusch and Pagan Lagrangian multiplier test for random effects would suggest to choose the RE model, if one would had to choose between a RE and a pooled OLS model.11

The general procedure is structured as follows: First, each set of variables, dedicated to answer the hypotheses, which are stated in the second chapter of this paper, is added separately to the base line equation in order to verify its empirical significance. For each of those sets the joint significance is tested. Due to relatively small increases in the R-squared statistics, it seems necessary to conduct Chow-tests for significant model difference. This test utilizes the R-squared statistics of the restricted model (the baseline model) and the unrestricted model to state whether there is a significant difference between these models. If this would not be the case, the Chow-test would recommend the use of the unrestricted model with fewer variables. After the empirical significance for each set of variables is tested, the full model is estimated.

11_{One has to be careful in interpreting this test because it is only really meaningful in combination with the}

(26)

4 Empirical Results

This chapter is dedicated to the description of data, the model specification and estimation itself. Moreover, a set of robustness checks are performed and the results are interpreted. Additionally, limitations are discussed.

4.1 Data Description

Table 1 – Summary Statistics

(1) (2) (3) (4) (5)

VARIABLES N mean sd min max

lfdi 2,484 1.909 4.587 -‐10.61 11.35 lgdp_source 2,926 26.67 1.590 22.94 30.28 lgdp_host 2,926 26.86 1.275 23.87 28.92 ldist 2,926 8.045 1.148 5.374 9.883 lwage 2,926 10.59 0.218 10.02 10.92 tariff 2,926 1.920 0.571 1.020 3.130 ulc 2,926 98.73 7.029 78.31 120.2 hightech_ex 2,926 15.75 9.323 3.413 47.84 fx 2,750 0.361 0.411 0.000564 2.236 govdebt 2,888 60.97 28.36 3.671 126.2 privcredit 2,907 120.1 40.68 51.38 224.0 crisis 2,926 0.147 0.354 0 1 comlang_off 2,926 0.0861 0.281 0 1 EU 2,926 0.259 0.438 0 1

Even though some observations for the dependent variable are missing, it is important to not exclude them in the final estimation because otherwise the statistical software cannot calculate confidence intervals, respectively standard errors, properly. Worthy of mention is that for five variables – lfdi, fx, crisis, comlang_off and EU – the standard deviation is greater than their mean value. Some observations of the variable lfdi take on negative values. If a host country reported a negative bilateral FDI flow, then a dis-investment occurred. In such a scenario investment flows reverse and the bilateral stock of FDI shrinks. It is assumed that the negative sign of the variable lfdi carries important information with regard to the occurrence of financial crisis. For fx this observation is rooted in rather small values for three currencies.12 The dummy variables possess greater standard deviations than means because only few source countries in the sample are actually members of the EU or share a common language or only rarely experienced years of financial distress. In principal, the model can still be estimated via ordinary least squares, even though some of those explanatory variables

(27)

run the risk of lacking explanatory power because the true effect could not be captured (Hill, Griffiths & Lim, 2011). The table of correlations can be found in annex B1. Due to the careful selection of variables, no striking evidence for multi-collinearity is found. Additionally, the sample was tested for variance inflation and the reported values for all variables are well below 10 which is considered the threshold for confidently assuming the absence of linear combinations among the explanatory variables. Annex C2 reports the mean variance inflation.

4.2 Model Specification and Estimation

The most general form of the empirical model tested is the following:

𝑙𝑓𝑑𝑖_!"# = 𝜃_!+ 𝜃_!𝑆𝑂𝑈𝑅𝐶𝐸_!" + 𝜃_!𝐻𝑂𝑆𝑇_!"+ 𝜃_!𝑃𝐴𝐼𝑅_!"#+ 𝜃_!𝐶𝑂𝑁𝑇𝑅𝑂𝐿 + 𝜌_! + 𝜎_!+ 𝜏_! + 𝜖_!"# (𝐈𝐈𝐈)

Where 𝑙𝑓𝑑𝑖!"# represents the log of bilateral FDI flows from source country i to host country j

at time t. 𝑆𝑂𝑈𝑅𝐶𝐸!" is a vector of source country characteristics which vary over time (e.g.

GDP), 𝐻𝑂𝑆𝑇_!" is a vector of source country variables which vary over time (e.g. labor costs),

𝑃𝐴𝐼𝑅_!" is vector of country-pair specific variables that may or may not vary over time (e.g.

distance). 𝐶𝑂𝑁𝑇𝑅𝑂𝐿 is a vector containing control variables (inflation, tertiary education of workforce, unemployment), 𝜌_! represents unobserved source country characteristics, 𝜎_! are unobserved host country characteristics, 𝜏_! accounts for unobserved effects in time and 𝜖_!"# is the error term of the regression. Specifically, the unobserved effects are I –1 respectively J -1 country dummies and T -1 time dummies. Thereby, Bellak et. al (2008), is followed who recommend including variables for unobserved effects in gravity equation models with three indices. The time dummies are meant to capture business cycle effects and therefore account for common cyclical effects influencing the dependent variable (Levy-Yeati et. al., 2003). At first, the model is estimated without control variables and unobserved effects.

(28)

𝑙𝑓𝑑𝑖_!" = 𝛽_!+ 𝛽_! 𝑙𝑔𝑑𝑝_! + 𝛽_!𝑙𝑔𝑑𝑝_! + 𝛽_!𝑙𝑑𝑖𝑠𝑡_!" + 𝛽_!𝑙𝑤𝑎𝑔𝑒_!+ 𝛽_!𝑡𝑎𝑟𝑖𝑓𝑓_!+ 𝛽_!𝑢𝑙𝑐_!

+ 𝛽!ℎ𝑖𝑔ℎ𝑡𝑒𝑐ℎ_𝑒𝑥! + 𝛽!𝑓𝑥!" + 𝛽!𝑔𝑜𝑣𝑑𝑒𝑏𝑡!+ 𝛽!"𝑝𝑟𝑖𝑣𝑐𝑟𝑒𝑑𝑖𝑡! + 𝐷!𝑐𝑟𝑖𝑠𝑖𝑠!

+ 𝐷!𝑐𝑜𝑚𝑙𝑎𝑛𝑔_𝑜𝑓𝑓!"+ 𝐷!𝐸𝑈!" + 𝜃!𝐶𝑂𝑁𝑇𝑅𝑂𝐿 + 𝜌!+ 𝜎!+ 𝜏!

+ 𝜖_!"# (𝐈𝐕)

Where 𝑙𝑓𝑑𝑖_!" is the log of net FDI flows form source country i to host country j in year t denominated in million Euro, 𝑙𝑔𝑑𝑝_! is the log of real GDP of source country i in year t in US$, 𝑙𝑔𝑑𝑝! is the log of real GDP of host country j in year t in US$, 𝑙𝑑𝑖𝑠𝑡!" is the log of

distance between source country i and host country j in kilometres, 𝑙𝑤𝑎𝑔𝑒! is the log of

average yearly wage of host country j in year t adjusted for purchasing power parity, 𝑡𝑎𝑟𝑖𝑓𝑓_! is the weighted-average applied tariff rate for all products of host country j in year t in percentage points, 𝑢𝑙𝑐_! represents the relative unit labor cost of host country j in year t, ℎ𝑖𝑔ℎ𝑡𝑒𝑐ℎ_𝑒𝑥_! is the share of high-tech exports as share of total manufactured exports of host country j in year t, 𝑓𝑥_!" is the annual average of the bilateral exchange rate of the currencies of source country i and host country j in year t adjusted for prices changes, 𝑔𝑜𝑣𝑑𝑒𝑏𝑡! is the

debt of the central government as percentage of GDP of host country j in year t, 𝑝𝑟𝑖𝑣𝑐𝑟𝑒𝑑𝑖𝑡!

is the amount of credit granted to the private sector by the financial sector as percentage of GDP in host country j in year t, 𝑐𝑟𝑖𝑠𝑖𝑠_! is a dummy variable indicating financial crisis in source country i in year t, 𝑐𝑜𝑚𝑙𝑎𝑛𝑔_𝑜𝑓𝑓_!" is a dummy variable indicating a mutual language between source country i and host country j and 𝐸𝑈_!" is a dummy variable indicating whether source country i and host country j both belong to the EU. Note that for clarity purposes, sub-indices indicating the time are left out in specification (IV), except in the case of the time dummies.

Table 2 reports the regression results for the baseline model (column 1), results for the different sets of explanatory variables which are added to the baseline model (columns 2 – 6) and the full model (column 7). For each set of variables, a F-test for joint significance is performed. In each case, the added variables are jointly significant, even though some appear to be individually insignificant at first glance. Since the F-tests do not support the exclusion of variables, they are all kept in the full model.

(29)

whereas the unrestricted models are models (2) – (7). If the null hypothesis of this test cannot be rejected, good economic practice would suggest to choose the restricted model over the unrestricted version, but in each case the calculate F-statistics exceed the critical values. This leads to the belief that every model is significantly different from the baseline model and is hence of use.

There is strong evidence against homoscedasticity in the sample. This insight is gained through reviewing the residuals plots of the regressions from the individual variables against the dependent variable as well as performing the Breusch-Pagan-test for heteroscedasticity. Since there are multiple sources of heteroscedasticity and no consistent pattern of heteroscedasticity, simply applying weights or splitting the sample does not resolve the issue (Buckley et. al., 2007). To assess whether autocorrelation is an issue the Wooldridge-test for autocorrelation is performed. The results are displayed in annex C2. Therefore, the model (7) in table 2 is estimated with standard errors that are consistent, even in the presence of heteroskedasticity and auto-correlation.

Table 2 – Pooled OLS Regressions

(1) (2) (3) (4) (5) (6) (7) VARIABLES lfdi lfdi lfdi lfdi lfdi lfdi lfdi

lgdp_host 0.128* 0.184** 0.192*** 0.139* 0.257*** 0.139* 0.510*** (0.0716) (0.0722) (0.0730) (0.0746) (0.0802) (0.0716) (0.0917) lgdp_source 0.565*** 0.576*** 0.570*** 0.471*** 0.597*** 0.553*** 0.501*** (0.0595) (0.0595) (0.0593) (0.0678) (0.0618) (0.0605) (0.0621) ldist -‐0.644*** -‐0.610*** -‐0.657*** -‐0.604*** -‐0.684*** -‐0.676*** -‐0.614*** (0.0795) (0.0796) (0.0793) (0.0862) (0.0841) (0.101) (0.102) lwage 1.945*** 2.416*** (0.434) (0.437) tariff 0.655*** 0.774*** (0.173) (0.193) ulc -‐0.0605*** -‐0.0849*** (0.0151) (0.0167) hightech_ex -‐0.00788 -‐0.0582*** (0.0107) (0.0134) fx 0.944*** 0.823*** (0.250) (0.255) govdebt -‐0.0131*** -‐0.0143*** (0.00362) (0.00424) privcredit -‐0.000641 0.00542* (0.00245) (0.00293) crisis -‐0.520* -‐0.570* (0.273) (0.327) comlang_off 0.930*** 0.916** (0.335) (0.429) EU -‐0.272 -‐0.0437 (0.252) (0.265) Constant -‐11.53*** -‐35.43*** -‐7.122*** -‐9.981*** -‐14.60*** -‐11.27*** -‐38.07*** (2.488) (5.631) (2.704) (2.618) (2.659) (2.508) (5.669) Observations 2,484 2,484 2,484 2,361 2,484 2,484 2,361 R-‐squared 0.049 0.060 0.056 0.052 0.056 0.053 0.088

(30)

Table 2 shows that the gravity parameters across all specifications exhibit the expected signs. The GDP of the source country as well as distance are significant at all common levels for alpha whereas host country GDP in models (1), (4) and (6) is only barely significant. The coefficients for measuring the impact of wages and tariffs both exhibit the expected positive signs and are very significant, with wages having a relatively strong impact on the dependent variable. The coefficient for the unit labour cost is also very significant and appears with the expected negative sign contrarily to the coefficient of hightech_ex that has not the expected positive sign in model (4). Interestingly, the coefficient of hightech_ex turns significant in specification (7), but keeps the wrong sign. Furthermore, the coefficient of fx measuring the exchange rate effect is very significant and has the expected positive sign in specification (4) as well as (7). In specification (5) and (7), the coefficients of govdebt exhibit the expected negative signs and are significant. Privcredit appears with the wrong sign in specification (5) and is insignificant. Both issues seem to be resolved in the final specification. The crisis dummy is weakly significant and exhibits a negative impact on dependent variable. Specification (6) points towards the positive impact of sharing a mutual language on bilateral FDI. The EU dummy is insignificant in both specifications. It could be the case that the effect of the common trading block is partly picked up by the distance variable; for this reason the EU dummy might appear insignificant (Portes & Rey, 2000). The most striking observation from reviewing table 2 are the rise in the impact of the host country GDP in the final specification, the considerable impact of average annual wages on FDI, the unexpectedly negative sign for high-tech exports and lastly the significant impact of source country crisis on FDI. The former is likely explained by the addition of several host-country variables to the model that might exhibit some correlation with host country GDP.

(31)

currency union. This finding is well in line with literature since all three countries are known to be major source countries of foreign direct investment.

When it comes to the host country dummies, the majority of dummies appear to be insignificant. Even though many host country variables turn insignificant, it is still in so far satisfactory that one can confidently assume that the selected explanatory variables sufficiently capture differences between the host countries. Only the host country dummy variables for Germany, France and Netherlands exhibit a significant and positive impact. Interestingly, the dummy variable for Luxembourg appears to be insignificant. Also note that the coefficient of host country GDP is still significant but switches its sign. In case of the year dummies, only the dummy for the year 2000 appears to be significant with having a positive impact on FDI. Considering that FDI flows at first glance appear to be of cyclical nature, which is meant to be captured by the year dummies, this finding is somewhat surprising. So, it must be concluded that there are no common cyclical factors at work that exhibit an impact on FDI flows (Frenkel et. al., 2004). However, not surprising is the fact that the crisis dummy looses explanatory power when time dummies are included since its effect is picked up by the time dummies.

After examining the impact of all sets of fixed effects separately, specification (4) incorporates all fixed effects but is just presented for the sake of completeness. More meaningful is specification (5) in which only the significant dummies from regression (1) – (3) have been included.13 In this final specification, the dummies for South Korea as well as Germany turn insignificant. However, F-tests for joint significance suggest leaving them in the model. Besides, most variables of the model are significant and exhibit the expected sign yet the coefficient for hightech_ex is still negative but is notably robust across all specifications. It is not obvious why countries with a higher share of high-technology exports receive less FDI. However, one possible explanation could be that MNEs prefer to source technology intensive products via exporting from those countries rather than set up affiliates in those countries due to considerable sunk-costs in technology intensive industries. A higher-share of technology exports might also indicate a high competitiveness of firms in those countries that cannot be overcome by source country firms.

The results of specification (5) do not change even when a set of control variables is added to the model.14 No control variable exhibits a significant impact for which reason the regression

13_N_{amely, the dummies for Japan, South Korea, the United States, France, Germany, the Netherlands and the} year 2000.

(32)

outputs are not displayed in the annex. As mentioned before, the relatively low values of the goodness-of-fit measure R-squared does not necessarily indicate a bad specification of the model. Rather it has to do with the fact that OLS is generally not well suited to predict negative values of the dependent variable (Bellak et. al., 2008). This will become more evident in the next section in which a couple of robustness checks will be conducted.

Table 3 – pooled OLS regression with fixed effects

(1) (2) (3) (4) (5)

VARIABLES lfdi lfdi lfdi lfdi lfdi

lgdp_host 0.520*** -‐1.359* 0.529*** 2.874 0.235* (0.0918) (0.809) (0.0922) (4.049) (0.142) lgdp_source -‐0.432 0.519*** 0.510*** 0.175 0.304*** (0.498) (0.0605) (0.0610) (0.640) (0.0704) ldist -‐0.691** -‐0.607*** -‐0.601*** -‐0.642** -‐0.668*** (0.284) (0.0853) (0.0856) (0.299) (0.0874) lwage 2.573*** -‐3.260 2.661*** -‐5.426 2.592*** (0.486) (4.399) (0.454) (5.057) (0.448) tariff 0.441** 0.353 0.881*** 0.836 0.705*** (0.223) (0.263) (0.249) (0.691) (0.193) ulc -‐0.0777*** -‐0.0120 -‐0.0780*** 0.00476 -‐0.0651*** (0.0173) (0.0258) (0.0188) (0.0329) (0.0178) hightech_ex -‐0.0609*** -‐0.132*** -‐0.0646*** -‐0.139*** -‐0.0999*** (0.0137) (0.0425) (0.0140) (0.0448) (0.0176) fx 2.436*** 0.771*** 0.764*** 1.735** 1.152*** (0.773) (0.254) (0.256) (0.877) (0.289) govdebt -‐0.0134*** 0.00539 -‐0.0135*** 0.0209 -‐0.0115** (0.00422) (0.0103) (0.00436) (0.0135) (0.00524) privcredit 0.00678** -‐0.00151 0.00720** -‐0.00842 0.00499 (0.00303) (0.00716) (0.00320) (0.00969) (0.00338) crisis -‐0.946*** -‐0.524 -‐0.506 -‐0.923*** -‐0.755** (0.340) (0.329) (0.338) (0.349) (0.328) comlang_off 0.738 1.050** 0.967** 0.854* 1.244*** (0.501) (0.416) (0.414) (0.514) (0.422) Constant -‐15.28 64.63 -‐42.51*** -‐18.38 -‐28.86*** (14.24) (45.64) (6.118) (98.40) (6.543) Observations 2,361 2,361 2,361 2,361 2,361 R-‐squared 0.102 0.106 0.093 0.123 0.104

Source FE YES NO NO YES YES

Host FE NO YES NO YES YES

Year FE NO NO YES YES YES

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

4.3 Robustness Checks

FDI Flows, Exchange Rates and Crisis: Determinants of Inward Foreign Direct Investment of the Euro Members