The currency union effect on trade and the foreign direct investment channel José de Sousa∗

Julie Lochard†

June 2009‡

Abstract

Ten years ago, European countries took the historic decision to create a single currency. One major motivation was to reinforce economic integration by fostering international trade and investments. Since then, many studies have attempted to evaluate the impact of the Economic and Monetary Union (EMU) on trade or foreign direct investment (FDI). In this paper, we argue that these two issues are intertwined and that there is an interrelationship between the trade and FDI effects of EMU. Using data for 21 OECD countries on the period 1982-2005, we find that part of the euro impact on trade is driven by additional FDI. Our findings appear to be robust to endogeneity problems, as well as sample and specification changes. Keywords: EMU, Trade, FDI JEL classification: F15, F21, F33 Total word counts: 8112 ∗

University of Rennes 2, CREM University of Rennes 1 and CES University of Paris 1 Panthéon-

Sorbonne. Email: [email protected]. † Erudite, University of Paris 12 Val de Marne. Email: [email protected]. ‡ We are grateful to Matthieu Crozet, Eric Fisher, Carl Gaigné, Ann Harrison, Roman Horvath, Beata Javorcik-Smarzynska, Miklos Koren, John Romalis, Farid Toubal, Alessandra Tucci, Maurizio Zanardi and seminar participants at the Universities of Prague, Paris 13 (RIEF 2004), Nottingham (ETSG 2004), Ljubljana (EIIE 2005), Paris 1 and INRA-ESR for extremely helpful comments and suggestions. We also thank Françoise Legallo (INSEE) for providing us with data on bilateral trade, and Daniel Mirza for labour market data and helpful suggestions.

1

1

Introduction

Ten years ago, ten European countries took the historic decision to create the Economic and Monetary Union (EMU). One major motivation was to reinforce economic integration by fostering trade and investment relationships. Since then, many studies document a positive impact of the euro on trade (e.g. Bun and Klaassen, 2007; Flam and Nordstrom, 2003; Micco et al., 2003)1 or on foreign direct investments (FDI) (Coeurdacier et al. 2008; de Sousa and Lochard, 2006, 2008; Petroulas, 2007; Schiavo, 2007)2 . Other studies estimate the euro impact on both trade and FDI but do not take into account their interrelationship (e.g. Flam and Nordström, 2007). In this paper, we aim to account for the interrelationship between the trade and FDI effects of EMU. In particular, we assess whether some trade effects related to the launch of the euro may occur through an increase in FDI. The trade and FDI effects of the euro may be intertwined due to their strong linkages. Theoretically, the relationship between trade and foreign investment decisions is ambiguous, depending on the motive of FDI. If the motive is the duplication of a stage of production abroad (horizontal motive), FDI tend to substitute for exports to serve the foreign market. In contrast, if the motive is the exploitation of international factor-price differences (vertical motive), FDI lead to a geographical fragmentation of the production process and tend to complement trade. Empirically, data confirm the existence of horizontal and vertical motives of FDI according to the product-level considered (Blonigen, 2001). However, the distinction between horizontal and vertical motives is much starker than what is observed in practice. At least three other factors may create complementary linkages between FDI and trade, whatever the FDI motive. First, the presence of foreign affiliates in the host country may stimulate demand for products that originate in the parent country. Second, we observe export-platform FDI (Hanson et al., 2001; Ekholm et al., 2003) in which the foreign affiliate’s output is sold in third markets rather than in the parent or host markets. Finally, even in the case of horizontal FDI, trade can be expanded if 1

For instance, Micco et al. (2003) find that the euro has fostered bilateral trade between member

countries by 4% to 10%. Baldwin et al. (2008) review the literature on the euro effect on trade and report an average effect of 5%. 2 Beyond simple stylized facts, de Sousa and Lochard (2006) and Petroulas (2007) are, as far as we know, among the firsts to document a positive impact of euro on FDI in the early years of the EMU. Based on a multi-country theoretical framework and a larger time span (1992-2005), de Sousa and Lochard (2008) show first that the euro has raised bilateral FDI by around 30% in average and second that this positive effect has been smoothly increasing over time. Coeurdacier et al. (2008) confirm a positive euro effect on cross-border mergers and acquisitions (M&A) in the manufacturing sector.

2

foreign affiliates require imported inputs from the parent country (Head and Ries, 2001). Overall, these factors, combined with vertical motives, may explain why complementary effects between FDI and trade prevail empirically at high levels of data aggregation (Swenson, 2004).3 In this case, the effect of the euro on FDI and trade could reinforce one another. As far as we know, the influence of the interrelationship between trade and FDI on the euro effect has received little attention.4 Nevertheless, this interrelation may provide a complementary and partial explanation of the positive euro effect on trade (see for instance Baldwin, 2006: 27). To explore the interrelationship between trade and FDI, we use a simple empirical framework: a gravity equation. This is the traditional tool for explaining bilateral trade and its theoretical foundations date from Anderson (1979). Bilateral trade is related to country specific factors, such as total income and price indexes, as well as bilateral factors, such as geographic distance, economic and monetary integration agreements (including the euro) and FDI relationships. Using aggregate bilateral data, we find complementarity between FDI and trade. We also show that part of the euro effect on trade appears to be indirect, coming from an increase in FDI following the creation of the single currency. We interpret this result as a ‘FDI channel’. We also tackle potential simultaneity between FDI and trade by employing a two stage least squares estimation procedure. Our finding about the FDI channel appears to be robust to endogeneity problems but also to a variety of sensitivity tests related to sample and specification changes.

The remainder of the paper is organized as follows. In the next section, we present the empirical model and describe the data set. In section 3, we report our main results and provide some sensitivity tests related to the sample and specification. In section 4, we deal with endogeneity problems: we first account for the sequential nature of the FDI channel and then we use an instrumental variables estimator for the FDI variable. In the final section, we draw some conclusions. 3

Aizenman and Noy (2006) emphasize also strong linkages between FDI and trade with positive feed-

back effects. 4 Brouwer et al. (2008) use a simulation-based technique to assess the implications, in terms of trade and FDI, of a potential EMU enlargement. For this purpose they estimate the impact of the euro on both trade and FDI. They find a complementarity between trade and foreign investments. However, they do not account for a potential simultaneity problem between both variables.

3

2

Methodology and data

2.1

The empirical model

A general theoretical gravity equation can be represented by the following specification: Mij = GCi Cj Bij ,

(1)

where the value of imports of country i from country j (Mij ) is related to the importing country characteristics Ci , to the exporting country characteristics Cj , to the bilateral factor Bij , and to the gravitational constant G which does not vary across countries.5 What country characteristics (Ci ) and (Cj ) stand for? This is a first estimation issue. From the seminal work of Anderson and van Wincoop (2003), we know that total incomes (Yi and Yj ) and implicit price indexes (Pi and Pj ) represent the main importing and exporting country characteristics. Price indexes (Pi ) and (Pj ), named “multilateral resistance” indexes, account for the fact that “the more resistant to trade with all others a region is, the more it is pushed to trade with a given bilateral partner” (Anderson and van Wincoop, 2003). A fairly simple and efficient solution to control for these unobserved country terms is to introduce country specific dummies (αi and αj ) in the estimated equation (Feenstra, 2004).6 A second estimation issue concerns the bilateral factor (Bij ). We follow other authors in hypothesizing that (Bij ) is a log-linear function of some observables: emu × FDIφ , × beu Bij = distδij × volatγij × arta ij ij × cij ij 5

(2)

In the Anderson and van Wincoop (2003)’s theoretical derivation of the gravity model, G represents

the world GDP. 6 Our data vary across time and this solution is not totally satisfactory when a gravity equation is estimated on a such panel data set. Indeed, using country fixed effects amounts to consider that price terms are time-independent. However, other solutions are not yet totally satisfactory as well (see e.g. Head et al. 2008). For instance, applications suggest that the Anderson and van Wincoop’s non-linear estimates are not robust to alternate ways of calculating ‘internal’ distance. This method presents also computational difficulties. The Baier and Bergstrand (2009) bonus vetus OLS approach is stimulating but, as suggested by Head et al. (2008), it is not sure whether this approach generalizes to panel data. The ‘tetrad’ approach of Head et al. (2008) is also appealing but the results are sensitive to the choice of the reference countries. One simple computational solution would be to introduce country dummies interacted with time dummies. However, this would lead to a disproportionate loss of degrees of freedom and is computationally burdensome.

4

• (distij ) is the bilateral distance and is a proxy for trade costs, such as transport costs, which traditionally dampens trade. We expect δ < 0. • (volatij ) is the volatility of bilateral exchange rate. According to the literature, it is expected to increase trade costs and therefore to reduce trade. We expect γ < 0. The following dummy variables capture economic and monetary integration effects. They are expected to reduce trade barriers and therefore to influence trade positively. We expect a > 0, b > 0 and c > 0. • (arta ) captures the regional economic integration effect related to the signature of regional ij trade agreements. In a more formal way, the dummy variable (rta) equals 1 if i and j are members of the same regional trade agreement and 0 otherwise (see Table 5 in appendix for details). • (beu ij ) captures the effect of the European economic integration process. Formally, the dummy variable (eu) equals 1 if i and j are members of the European Union and 0 otherwise. • (cemu ) measures the effect of the EMU beyond the exchange rate stabilization and ecoij nomic integration process. In a more formal way, (emu) equals 1 if the two countries belong to the EMU and 0 otherwise. The final term of Bij captures the strong linkages between FDI and trade (see above). The FDIij variable is computed as the stock of bilateral FDI between country i and country j. We expect φ to be positive if complementary effects between trade and FDI prevail at our aggregate level of estimation. This bilateral factor modeling limits the potential upward bias of the currency union effect. In fact, its magnitude relies on the comparison of trade between a pair of currency union members and a similar pair using different currencies. Consequently, if we do not control properly for trade characteristics and arrangements among the latter pair, the currency union effect might be biased upwards (see Baldwin, 2006). Benefiting from the multiplicative nature of the gravity equation, we estimate a common stochastic and log-linear form of (1). Theory and assumptions about Ci , Cj and Bij imply that: ln(Mijt ) = gt + β1 ln(Yit ) + β2 ln(Yjt ) + δ ln(dist)ij + γ ln(volat)ijt + β3 (rta)ijt + β4 (eu)ijt + β5 (emu)ijt + φ ln(fdi)ijt + ναi + µαj + λt + ijt ,

5

(3)

where time subscripts stress the points that we are estimating equation (3) on panel data and some variables are time-dependent. The dependent variable is the value of bilateral imports of country i from country j at time t. The constant g = ln G. β3 = ln(a), β4 = ln(b) and β5 = ln(c) are coefficients to be estimated. The coefficients, ν = ln(Pi ) and µ = ln(Pj ), are supposed to estimate the multilateral resistance terms. Our specification also includes a vector of time dummies (λt ), which controls for the general evolution of trade and g across time periods. ijt is an error term reflecting measurement error in trade. This equation is linear in parameters and can be estimated by ordinary least squares (OLS) estimator. In this equation, the coefficient on the EMU dummy (β5 ) measures the ‘net’ effect of EMU on trade, i.e. accounting for interrelationships between FDI and trade. To reveal a FDI channel, we compare this ‘net’ effect to a ‘gross’ effect of EMU obtained by regressing equation (3) without the FDI variable. A significant positive difference between the gross and net effect of EMU is interpreted as evidence for the FDI channel hypothesis. We further check the result of this comparison by addressing a potential problem of simultaneity between trade and FDI (see section 4).

2.2

Data

We estimate equation (3) at the aggregate level on a sample of 21 OECD countries: 14 European union members (Austria, Belgium-Luxembourg, Germany, Denmark, Spain, Finland, France, Great-Britain, Greece, Ireland, Italy, the Netherlands, Portugal, Sweden) and 7 non-European union members (Australia, Canada, Japan, Korea, Norway, Switzerland, the United States) on the time period 1982-2005. Bilateral imports (Mijt ) come from OECD, as well as outward FDI stocks (fdiijt ) which are extracted from the International Direct Investment Statistics Yearbook.7 FDI data are based for the most part on balance of payments statistics published by Central Banks and Statistical Offices.8 Total income (Y ) is proxied by GDP coming from the World Bank (World Development Indicators). Distance between country i and country j (distij ) is computed as the great circle 7

We would have preferred using more disaggregated data but there are no bilateral FDI data at the

sectoral or firm level for a large sample of countries and recent years. 8 Besides OECD, FDI data are also collected by the Statistical Office of the European Communities (Eurostat), UNCTAD, World Bank and the IMF. One of the major advantages of the OECD database is that flows and stocks of FDI are registered by country of origin and destination since 1980. Eurostat collects data via common OECD-EUROSTAT questionnaires and harmonizes national data. However, Eurostat covers fewer countries and years than the OECD database. The two series are moreover highly correlated.

6

distance between the major cities of the two countries. The exchange rate volatility variable (volatijt ) is computed as the standard deviation of the first difference in the log of the monthly nominal exchange rate in the preceding and current year. Volatility is set equal to zero for countries belonging to the EMU. The appendix gives more details about sources and computation of the variables (see Table 5 for details and Table 6 for descriptive statistics).

3

Results

Baseline results We pool the data and use the OLS estimator to run pair-regressions.9 We first estimate equation (3) by dropping the FDI variable to measure the gross impact of EMU on trade. Then, we run (3), including the FDI variable, to account for the ‘FDI channel’ and estimate the net impact of EMU. Our results are reported in Table 1. On the overall sample (column 1), all coefficients are economically and statistically significant (p 0.5

(1)

0.21a (0.05)

(2) C. Original EU 6 countries

(1)

0.06

0.20a

(0.04)

(0.01)

0.29a (0.07)

(2) D. DM bloc countries

(1)

0.20a

0.11a

(0.07)

(0.01)

0.30a (0.08)

E. Nordic countries

(2)

0.21a

0.11a

(0.07)

(0.01)

(1)

0.21a (0.05)

F. Poorer countries

(2)

0.09b

0.15a

(0.05)

(0.01)

(1)

0.13a (0.03)

(2) G. Richer countries

(1)

0.06b

0.17a

(0.03)

(0.01)

0.20a (0.04)

(2)

0.12a

0.15a

(0.04)

(0.01)

Notes: Dependent variable: log of bilateral imports of country i from country j. Heteroscedastic-consistent (White-robust) standard errors in parentheses. a and b denote resp. the significance at the 1% and 5% level. (1) corresponds to the estimation of equation (3) without FDI and (2) reports estimation of equation (3). Estimations are carried with the OLS estimator including country and year dummies. To save space, all other coefficients are not reported but are available upon request.

10

In the trade literature, the largest euro effect is found for the most central and tightly integrated countries (Micco et al., 2003 and Baldwin, 2006). Is the FDI channel result robust to the drop of such countries? To address this question we run our estimation excluding alternatively: (1) the EU 6 original members, which are considered as the most tightly integrated countries (row C) (i.e. Belgium, France, Germany, the Netherlands, Italy, Luxembourg), (2) the “Deutsche Mark bloc countries” (row D) (i.e. Austria, Belgium, Luxembourg, the Netherlands, Denmark, France) and (3) the Nordic countries (row E) (i.e. Denmark, Finland, Norway, Sweden). The FDI channel result appears quite robust to these sample modifications. When we introduce the FDI variable, the EMU coefficient is again reduced (and sometimes less significant), indicating that part of the effect of the monetary union on trade might go through an increase in FDI. Finally, we check if the FDI channel is robust to the drop of the poorest or richest countries of the sample. We estimate our model without countries for which GDP per capita is below 15,000 US$ on average on the period 1982-2005 (row F) (Greece, Korea, Portugal and Spain) and then without countries for which GDP per capita is above 25,000 US$ (row G) (Denmark, Finland, Germany, Japan, Norway, Switzerland, Sweden, USA). Again, the FDI channel appears to be robust to these sample modifications.

4

Endogeneity issues

The estimation of equation (3) may be affected by endogeneity problems in the relationship between FDI and trade. A positive correlation between FDI and trade might simply reflect an omitted variable bias due to third unobserved factors (see e.g. Head and Ries, 2001) or a simultaneity bias. For instance, Neary (2007, 2009) suggests that the relationship may go from trade to FDI. In particular, he demonstrated that trade liberalization can encourage Mergers and Acquisitions (M&A) by fostering competition.14 To deal with the endogeneity issue, we use two techniques. First, we account for the sequential nature of the FDI channel and lag the FDI and the EMU variables. Second, we use an instrumental variables estimator for the FDI variable.

4.1

Sequential decisions of FDI and trade

One simple approach to addressing endogeneity is to lag the FDI variable. Indeed, we may think that FDI and trade are not contemporaneous in the sense that present trade is influenced by past investments. Consequently, we estimate equation (3) and lag the FDI variable by one year 14

Aizenman and Noy (2006) also emphasize bi-directional linkages between FDI and trade.

11

to remove contemporaneous shocks influencing simultaneously FDI and trade. This estimation is reported in column (2) of Table (3). Then, we compare this estimation with estimates of equation (3) in which the FDI variable has not been included and the sample restricted to observations for which one-year lagged FDI data are available (column 1). From this comparison, we observe that the EMU coefficient has again been reduced by half. The “net” euro effect, after introducing the FDI channel, is about 11%, consistent with the estimated coefficient of Table 1 (column 3).15

Table 3: Trade, FDI and monetary union: lagged variables Dependent variable Model ln Yit ln Yjt ln Distanceij Exchange rate volatilityijt rtaijt euijt emuijt

OLS (1) 0.64a (0.07) 0.50a (0.06) -0.71a (0.02) -0.04a (0.01) 0.89a (0.05) 0.98a (0.06) 0.20a (0.04)

Log of bilateral OLS (2) 0.49a (0.07) 0.40a (0.05) -0.59a (0.02) -0.04a (0.01) 0.86a (0.05) 0.94a (0.05) 0.10a (0.03)

imports of country i from country OLS OLS OLS (3) (4) (5) 0.64a 0.49a 0.65a (0.07) (0.07) (0.07) 0.50a 0.40a 0.46a (0.06) (0.05) (0.06) a a -0.71 -0.59 -0.71a (0.02) (0.02) (0.02) a a -0.04 -0.04 -0.05a (0.01) (0.01) (0.01) a a 0.89 0.86 0.88a (0.05) (0.05) (0.05) 0.98a 0.94a 0.98a (0.06) (0.05) (0.06)

0.21a (0.04)

emuij(t−1)

0.19a (0.04)

The euro effect\

22.11a (4.32)

0.15a (0.01) 10.79a (3.65)

Adj. R-sq # of observations Country dummies Year dummies

0.91 5595 yes yes

0.92 5595 yes yes

OLS (6) 0.50a (0.07) 0.37a (0.06) -0.59a (0.02) -0.05a (0.01) 0.86a (0.05) 0.94a (0.05)

0.12a (0.03)

23.25a (4.46)

0.15a (0.01) 12.69a (3.77)

21.45a (4.45)

0.12a (0.03) 0.15a (0.01) 12.68a (3.74)

0.91 5595 yes yes

0.92 5595 yes yes

0.91 5511 yes yes

0.92 5511 yes yes

emuij(t−2) ln FDI outstockij(t−1)

j

Notes: Dependent variable: log of bilateral imports of country i from country j. Heteroscedastic-consistent (White-robust) standard errors in parentheses with a denoting the significance at the 1% level. \ The euro effect is computed as [exp(emuijt )-1]×100 and its standard errors are computed using the Delta method. See text for details. 15

The results are quite similar with longer lags.

12

Furthermore, we may also argue that the decision to create a currency union is not contemporaneous to trade. A plausible sequence is that the creation of a currency union first gives incentives to invest abroad and then, once the investment is operational, trade begins. Actualy, unless the foreign investment is realized through the acquisition of a firm already involved in international trade, this sequence takes time. To deal with this sequentiality, we lag the EMU variable by one or two years (columns 3 to 6). In column (3), we modify equation (3) in three ways: we lag the EMU dummy by one-year, we exclude the FDI variable and use the sample restricted to observations for which one-year lagged FDI data are available. Then, in column (4), we add the one-year lagged FDI variable. The results remain broadly unchanged. In columns (5) and (6), we replace the one-year lagged EMU with a two-year lagged EMU dummy. We still find a positive euro effect and a reduction of its magnitude when introducing the FDI variable. However, as recognized by Frankel (1997: 132), introducing lagged variables is not totally satisfactory to addressing endogeneity, since “precedence does not ensure causality”. In particular, firms may have invested abroad before the currency union creation because they have anticipated further trade following the euro creation. Hence, we provide additional estimates using an instrumental variables estimator.

4.2

Instrumental variables estimator

The instrumental variables (IV) method is a common way to deal with the problem of endogeneity (either due to omitted variables, measurement error or simultaneity).16 The most complicated task is to find appropriate instruments for the FDI variable. The instruments must satisfy two requirements: (1) they must be correlated with the endogenous variable (i.e. FDI) but (2) uncorrelated with the error term. Labor market characteristics are natural instrumental variables candidates (see Clausing, 2000).17 They certainly satisfy the first requirement. Javorcik and Spatareanu (2005) find strong evidence that foreign investors care about labor market regulations 16

The joint determination of FDI and trade might also be described with a simultaneous equations

model (SEM). However, this model basically comes to the same thing since the leading method for estimating SEMs is the instrumental variables approach. But crafting the SEMs is more challenging. It requires to find out exogenous instruments for bilateral FDI, to identify the trade equation, and for bilateral trade, to identify the FDI equation. The latter requirement is all the most difficult. 17 Clausing (2000) uses average employee compensation as an instrumental variable. Frankel (1997) proposes a dummy variable indicating the existence of a bilateral tax treaty as an instrument for the FDI variable. However, Blonigen and Davies (2004) and Hallward-Dreimeier (2003) find little evidence that bilateral international tax treaties affect FDI.

13

in the host country. As a proxy for such regulations, we use the employment protection level in the host country [legislation (host)] (see Table 5 in appendix for a description of this variable and data sources). Employment protection may dissuade inward (vertical) FDI, as it represents potential costs for investors. Typically, more strict labor regulations in the host country are associated with lower FDI (Javorcik and Spatareanu, 2005). Moreover, we are confident that such regulations satisfy the second requirement and are uncorrelated with the error term. Actually, they are generally driven by policy objectives and not directly related to bilateral trade flows. As indicated by the first stage regression, reported in column (1) of Table 8 (see appendix), the host country’s employment protection level is a significant determinant of bilateral FDI, with an expected negative estimate. The partial F-test is about 10, which seems reasonable (Staiger and Stock, 1997). The result of the second stage least squares estimation is reported in column (2) of Table (4). For this IV model, we also report in column (1) the OLS results obtained by estimating equation (3) without the FDI variable. This estimation has been done on the sample restricted to the availability of the exogenous instrument in order to ease the comparison of the gross and net effect of EMU. The IV FDI estimate depicted in column (2) is statistically significant. It also appears to be larger than the OLS FDI estimate reported in Table (1) (column 3). This suggests that the OLS estimator tends to underestimate the trade-creating effect of FDI and thus to underate the FDI channel.18 As a result, with the IV estimator, the magnitude of the EMU coefficient is considerably reduced compared to the OLS EMU estimate of column (1). Moreover, it is not statistically significant. This may be due to the fact that the IV method tends to increase the variance of the estimator and the correlation between regressors, inducing larger standard errors (see Woolridge, 2003: 479). Since we cannot be totally sure that our instrument is uncorrelated with any unobservable factors affecting trade, we add a second instrument to undertake a simple overidentification test. This second instrument is still related to labor market characteristics. We use a variable measuring the union density in the parent country [union density (parent)] (see Table 5 in appendix). This variable is also likely to be correlated with FDI. An increase in union density in the parent country may encourage firms to (re)locate activities abroad. Again, we argue that the second requirement is satisfied since such a variable is not directly related to bilateral trade flows. Indeed, union density appears to be largely cultural and related to legal constraints (Cahuc and Zylberberg, 2004).19 The first stage regression, reported in column (2) of Table 8 (see appendix) 18

Note however that obtaining rigorously the direction of the bias in the coefficients between OLS and

IV regressions is a complicated issue (Wooldridge, 2003: 507). 19 For instance, in France or Spain, the relatively low rate of union density can be explained by the fact

14

uses the set of both instruments. As expected, the regression indicates economically and statistically negative significant effects of the host country’s legislation and the parent country’s union density on bilateral FDI. The first stage F-statistic of joint significance is still larger than ten. The result of the second stage least squares estimation is reported in column (4) of Table (4). The corresponding OLS results obtained by estimating equation (3) without the FDI variable on the restricted sample are presented in column (3). The Chi2 statistic of the Sargan test of overidentifying restrictions presented in column (4) indicates that we cannot reject the null hypothesis that the excluded instruments are valid instruments, i.e. uncorrelated with the error term. Here again, FDI and trade are positively related and the EMU coefficient is reduced when introducing the instrumented FDI variable. However, the statistical significance of both the EMU and the FDI estimates have been reduced.20 Again, this seems to be due to the larger standard errors obtained with the IV estimator compared to the OLS estimator (see above). Nevertheless, comparing the EMU and FDI OLS estimates of Table 1 (column 3) with the IV estimates of Table 4 (columns 2 and 4), we may speculate that the simultaneity between FDI and trade underestimates the FDI effect on trade and thus underestimates the FDI channel. Thus, we can argue that the IV estimates do not reject the FDI channel hypothesis.

5

Concluding remarks

In this paper, we attempted to estimate the euro effect on trade by taking into account interrelationships between trade and FDI effects. Our results reveal that the euro effects on trade and FDI reinforce one another. Part of the euro effect on trade appears to be indirect, going through a rise in foreign direct investments. Furthermore, this conclusion appears to hold for a variety of sensitivity tests related to sample and specification changes. Our finding appears also to be robust to potential endogeneity problems due to omitted variables or simultaneity between trade and FDI. In empirical papers, EMU member countries are also found to trade more with non-member countries (see e.g. Micco et al., 2003). The FDI channel could clear up the reason for this trade that collective agreements do not have the right to discriminate between unionized and non-unionized workers. In contrast, the right to discriminate between the two classes of workers in Australia or in the United Kingdom favor union membership (Cahuc and Zylberberg, 2004: 371). 20 The probability associated with the Student test is 6.9% for the FDI coefficient and 8% for the EMU coefficient.

15

Table 4: Trade, FDI and monetary union: IV estimates Model ln Yit ln Yjt ln Distanceij Exchange rate volatilityijt ftaijt euijt emuijt ln FDI outstockijt Adj. R-sq # of observations Country dummies Year dummies Sargan test [p-value]

Dependent variable: ln of bilateral imports OLS IV OLS IV (1) (2) (3) (4) 0.65a 0.35b 0.69a 0.58a (0.07) (0.16) (0.08) (0.09) a a a 0.65 0.42 0.62 0.49a (0.06) (0.13) (0.08) (0.10) a a a -0.76 -0.52 -0.76 -0.66a (0.02) (0.12) (0.02) (0.06) a a a -0.04 -0.04 -0.04 -0.04a (0.01) (0.01) (0.01) (0.01) a a a 0.78 0.75 0.80 0.78a (0.05) (0.05) (0.06) (0.05) 0.84a 0.78a 0.87a 0.85a (0.06) (0.06) (0.06) (0.05) 0.23a 0.04 0.18a 0.10c (0.04) (0.10) (0.05) (0.06) 0.30b 0.13c (0.15) (0.07) 0.90 5413 yes yes

0.90 5413 yes yes

0.90 4505 yes yes

0.91 4505 yes yes 0.01 [0.97]

Notes: Dependent variable: log of bilateral imports of country i from country j. Heteroscedastic-consistent (White-robust) standard errors in parentheses with a , b and c denoting respectively the significance at the 1%, 5% and 10% level. The constant, country and time dummies are not reported. Column (2): FDI instrumented by legislation (host); Column (4): FDI instrumented by legislation (host) and union density (parent). See text for more details.

16

increase. In fact, EMU members tend also to invest more in non-members countries since the creation of the euro (see de Sousa and Lochard, 2008). These additional FDI with non-member countries could give rise to trade flows. Further research is needed to investigate this effect.

References Aizenman, J., Noy, I., 2006. FDI and Trade - Two Way Linkages?, The Quarterly Review of Economics and Finance 46, 317-337. Anderson, J., 1979. A Theoretical Foundation for the Gravity Equation, American Economic Review 69(1), 106-116. Anderson, J., van Wincoop, E., 2003. Gravity With Gravitas: A Solution to the Border Puzzle. American Economic Review 93, 170-192. Baldwin, R., 2006. The euro’s trade effects. European Central Bank, Working Paper no. 594, March. Baldwin, R., Di Nino, V., Fontagné, L., De Santis, R., Taglioni, D., 2008. Study on the Impact of the Euro on Trade and Foreign Direct Investment. European Economy Economics Paper, 321. Blonigen, B., 2001. In Search of Substitution Between Foreign Production and Exports. Journal of International Economics 53, 81-104. Blonigen, B., Davies, R., 2004. The Effects of Bilateral Tax Treaties on U.S. FDI Activity. International Tax and Public Finance 11, 601-622. Brouwer, J., Paap, R., Viaene, J.-M., 2008. The Trade and FDI Effects of EMU Enlargement. Journal of International Money and Finance, 27(2), 188-208. Bun, M., Klaassen, F., 2007. The Euro Effect on Trade is not as Large as Commonly Thought. Oxford Bulletin of Economics and Statistics, 69, 473-496. Cahuc, P., Zylberberg, A., 2004. Labor Economics, MIT Press, Cambridge. Clausing, K., 2000. Does Multinational Activity Displace Trade? Economic Inquiry 38, 190-205.

17

Coeurdacier, N., de Santis, R., Aviat, A., 2008. Cross-border mergers and acquisitions: Financial and institutional forces. Economic Policy, forthcoming. De Sousa, J. and Lochard, J., 2006. Union monétaire et IDE. Quels sont les effets de l’euro ?, Revue économique 57 (3), 419-430. De Sousa, J., Lochard, J., 2008. Does the Single Currency Affect FDI? A Gravity-Like Approach. Mimeo. Ekholm, K, Forslid, R., Markusen, J., 2003. Export-Platform Foreign Direct Investment. NBER working paper, 9517. Feenstra, R., 2004. Advanced International Trade: Theory and Evidence. Princeton University Press, Princeton. Flam, H., Nordström, H., 2003, Trade Effects of the Euro: Aggregate and Sector Estimates, IIES Seminar Paper No. 746. Flam, H., Nordström, H., 2007, The Euro and Single Market impact on trade and FDI. Mimeograph. Frankel, J., 1997. Regional Trading Blocs in the World Economic System. Institute for International Economics, Washington DC. Hallward-Dreimeier, M., 2003. Do Bilateral Investment Treaties Attract FDI? Only a bit... and it might bite. World Bank Policy Research Working Paper Series 3121, Washington, World Bank. Hanson, G.H., Mataloni R.J., Slaughter, M.J., 2001. Expansion Strategies of U.S. Multinational Firms. In: Rodrik, D., Collins D. (Eds.), Brookings Trade Forum, pp. 245-294. Head, K., Ries, J., 2001. Overseas Investment and Firm Exports. Review of International Economics 9, 108-122. Head, K., Mayer, T., Ries, J., 2008. The erosion of colonial trade linkages after independence. Mimeo. Javorcik, B., Spatareanu, M., 2005. Do Foreign Investors Care about Labor Market Regulations?. Review of World Economics/Weltwirtschafliches Archiv 141, 375-403.

18

Micco, A., Stein, E., Ordonez, G., 2003. The Currency Union Effect on Trade: Early Evidence from EMU. Economic Policy 37, 315-356. Neary, P., 2007. Cross-border mergers as instruments of comparative advantage. Review of Economic Studies 74(4), 1229-1257. Neary, P., 2009. Trade costs and foreign direct investment. International Review of Economics and Finance 18(2), 207-218 Petroulas, P., 2007. The effect of the Euro on Foreign Direct Investment. European Economic Review 51, 1468-91. Rauch, J.E. and V. Trindade, 2002. Ethnic Chinese Networks in International Trade. Review of Economics and Statistics 88, 641-658. Santos Silva, J.M.C., Tenreyro, S., 2006. The Log of Gravity. Review of Economics and Statistics 84(1), 116-130. Schiavo, S., 2007. Common currencies and FDI flows. Oxford Economic Papers 59(3), 536-560. Staiger, D., Stock, J., 1997. Instrumental Variables Regression with Weak Instruments. Econometrica 65, 557-586. Swenson, D., 2004. Foreign Investment and the Mediation of Trade Flows. Review of International Economics 12(4), 609–629. Wooldridge, J., 2003, Introductory Econometrics. A Modern Approach, 2e, South-Western College Publishing, South-West.

19

Appendix Table 5: Data Imports Y Distance Exchange rate volatility

RTA

EU EMU

FDI Union density

Legislation

Bilateral imports data, in current dollars, come from OECD (International Trade by Commodity). Economic size is proxied by GDP coming from the World Development Indicators (World Bank). The bilateral distance is computed as the great circle distance between the major cities of the two countries. Bilateral exchange rate volatility is measured as the standard deviation of the first difference in the log of the monthly nominal exchange rate in the preceding and current year. Bilateral exchange rate come from International Financial Statistics (IMF). = 1 if countries i and j participate in a Regional Trade Agreement ate time t. Given our sample, we account for NAFTA (North American Free Trade Agreement), EFTA (European Free Trade Association) and EEA (European Economic Area) (EFTA-EU relations). = 1 if countries i and j are EU members at time t. = 1 if if countries i and j are EMU members at time t. In our sample, EMU comprises Austria, Belgium, Finland, France, Germany, Ireland, Italy, Luxembourg, the Netherlands, Portugal and Spain, since 1999 and Greece since 2001. Bilateral FDI data are taken from the International Direct Investment Statistics Yearbook (OECD). The share of workers unionized in the host country is expressed in percentage. Data come from OECD Labour Force Statistics database (available only until 2002 for most OECD countries). The employment protection legislation variable is an index constructed for 1990, 1998 and 2003 referring to both regulations concerning hiring (e.g. conditions for using fixed or temporary contracts, training requirements) and firing (e.g. redundancy procedures, severance payments). We reproduce the 1990 value until 1997, then the 1998 value until 2002 and finally the 2003 value until 2005. Data come from OECD Labour Force Statistics database.

20

Table 6: Descriptive statistics Variable ln(Imports) ln(GDP) ln(Distance) Exchange rate volatility RTA EU EMU ln(FDI Outstock) ln(Union density) ln(Legislation)

Nb. of obs 9720 10080 10080 10080 10080 10080 10080 5894 8320 9280

Mean 14.09 26.58 7.90 1.98 0.22 0.33 0.07 20.63 3.49 0.57

Standard deviation 1.75 1.33 1.14 1.28 0.42 0.47 0.26 2.59 0.59 0.73

Min 6.38 23.68 5.16 0 0 0 0 9.31 2.13 -1.56

Max 19.49 30.15 9.8 9.29 1 1 1 26.60 4.47 1.41

Table 7: Missing observations for the FDI outstock variable Year 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

Missing data ratio 0.80 0.79 0.68 0.66 0.58 0.48 0.43 0.44 0.42 0.36 0.35 0.28 0.20 0.27 0.22 0.23 0.14 0.13 0.15 0.18 0.26 0.30 0.66 0.56

country Australia Austria Belgium-Luxembourg Canada Denmark Germany Finland France Great Britain Greece Ireland Italy Japan the Netherlands Norway Portugal South Korea Spain Switzerland Sweden USA

Missing data ratio 0.53 0.25 0.60 0.09 0.63 0.09 0.45 0.28 0.19 0.75 0.62 0.27 0.26 0.26 0.41 0.63 0.40 0.60 0.52 0.53 0.01

Notes: Missing data ratio is calculated for each year or for each country as the number of missing observations divided by the total number of observations.

21

Table 8: First stage estimations Model ln Yit ln Yjt ln Distanceij Exchange rate volatilityijt ftaijt euijt emuijt ln Legislationjt ln Union densityit Adj. R-sq # of observations Country dummies Year dummies Test of excluded instruments\ [p-value]

Dependent variable: ln(FDI outstockijt ) OLS OLS (1) (2) 1.21a 1.19a (0.13) (0.14) a 0.85 1.12a (0.13) (0.15) -0.79a -0.80a (0.04) (0.04) -0.01 0.00 (0.02) (0.02) 0.15 0.14 (0.11) (0.12) 0.26b 0.23c (0.11) (0.12) 0.61a 0.59a (0.08) (0.09) a -0.49 -0.68a (0.16) (0.20) 0.77a (0.21) 0.82 5557 yes yes F(1,5486)=10.01 [0.001]

0.82 4649 yes yes F(2,4580)=11.44 [0.000]

Notes: Dependent variable: log of FDI bilateral outstock. Heteroscedasticconsistent (White-robust) standard errors in parentheses with a , b and c denoting respectively the significance at the 1%, 5% and 10% level. OLS estimations include country and year dummies. \ Excluded instruments: legislation in column (1); union density and legislation in column (2).

22