Foreign direct investment, financial development and the global financial crisis

Transcription

1 Foreign direct investment, financial development and the global financial crisis Rodolphe Desbordes Shang-Jin Wei Abstract Little is known about the empirical impact of a well-functioning and sophisticated financial system on outward and inward foreign direct investment (FDI). The purpose of this paper is to fill this void in the existing cross-country literature by using a novel database on bilateral real greenfield manufacturing FDI and two distinct but complementary econometric approaches. We exploit differences in firm- or sector-specific dependence on external finance and the occurrence of the banking crises in developed countries. Our main results are that source countries financial development (FD) tends to contribute strongly to the promotion of FDI, especially in sectors typically dependent on external finance, whereas destination countries FD matter much less and may, in sectors not typically dependent on external finance, even have a negative impact on FDI. The latter finding is likely due to both the ability of MNEs to use their internal capital markets to finance part or all their activities and the increase in domestic competition which typically comes along with higher FD. Our detailed study of the effects of the banking crises in developed countries provides additional insights. We find some empirical evidence that firms located in developed countries which did not formally experience a banking crisis also seem to have faced tighter credit conditions during the global financial crisis than in previous years, hindering their ability to invest abroad. Finally, the negative effects of the global financial crisis on the availability of external finance in developed countries appear to have progressively worsened over the years. Keywords: Banking crisis; Financial constraints; Financial Development; Foreign Direct Investment. JEL: F23; O16. Corresponding author. University of Strathclyde. Address: Department of Economics, Sir William Duncan Building, University of Strathclyde, 130 Rottenrow, Glasgow G4 0GE, Scotland, United Kingdom. Telephone/Fax number: / rodolphe.desbordes@strath.ac.uk Columbia Business School, NBER and CEPR. Corresponding author at: Graduate School of Business, Columbia University, Uris Hall, 3022 Broadway, Room 619, New York, NY 10027, United States. Telephone/Fax number: / shangjin.wei@columbia.edu

2 1 Introduction Depending on the perspective adopted, the empirical literature on foreign direct investment (FDI) and financial development (FD) may seem rich or scarce. While many cross-country studies have been devoted to investigating the interaction of FDI and FD in the context of economic growth (e.g. Alfaro, Chanda, Kalemli-Ozcan, and Sayek (2004)), much less attention has been paid to the impact of FD, in source or destination countries, on FDI. This paper attempts to fill this void in the literature by making four empirical contributions to current knowledge of the effects of a well-functioning and sophisticated financial system on outward and inward FDI. To do so, we use a novel database on sector-specific bilateral real greenfield manufacturing FDI over the recent period, with a wide coverage of source and destination countries and sectors. 1 As we will shortly explain, having information on the sectors in which FDI takes place is a crucial component of our empirical analysis. Our two first contributions consist of examining the influence of source and destination countries FD on outward and inward FDI. First, we investigate, in line with the recent literature on firm heterogeneity and credit constraints, whether source countries FD foster outward FDI. If FD plays such a role in determining outward FDI, this should be particularly apparent for FDI in sectors which tend to rely heavily on external finance. Second, we look for evidence that the overall influence of destination countries FD on inward FDI depends on the relative strengths of two FD-related but opposite effects. More specifically, some theoretical works have suggested that the impact of destination countries FD on inward FDI is ambiguous because the positive effect resulting from greater local access to external finance must be weighted against the negative impact of stronger domestic competition induced by FD. Hence, the relationship between destination countries FD and inward FDI may be positive or negative, depending on whether the access to external finance effect or the competition effect dominates. This relationship is likely to be sector-specific. Firms operating in sectors relatively more in need of external finance than other sectors, i.e. firms in externally financially dependent (ED) sectors, should put relatively more weight on the access to external finance than firms in non-ed sectors which may be able to rely solely on their internal capital markets. 1 For this paper, FDI is defined as a firm expanding its activities outside the territorial boundaries of its home country, whichever the way this expansion is financed. When we talk about financial FDI, we refer to cross-border FDI flows recorded in balance of payments statistics. These data only capture the proportion of the funding of existing and new foreign affiliates coming from related parties outside the host country, such as an equity investment or a loan by the parent company. Finally, the focus of this paper, greenfield FDI, makes reference to the decision by a firm to create a new affiliate in a foreign country, instead of investing in an already existing foreign affiliate or buying an existing domestic firm. The last entry mode would correspond to Mergers&Acquisitions (M&A) FDI. 1

3 Our last two contributions provide an alternative way to understand the role played by source and destination countries FD in promoting FDI, by exploring how outward and inward FDI reacted to the temporary disruptions in access to external finance that the banking crises in developed countries caused. For our third contribution, we examine whether the credit crunch occurring in those countries has hindered the ability of resident firms to invest abroad. As for source countries FD, if access to external finance truly matters for firms to engage in FDI, the effects of a tightening of credit conditions should have been more strongly felt by firms operating mainly in ED sectors. Finally, our last contribution is to analyse the sector-specific response of multinational enterprises (MNEs) to a temporary rise in credit constraints in crisis-affected destination developed countries. In addition to offering further evidence on the existence and relative strengths of the access to external finance effect and the competition effect, such an analysis can also help to test whether MNEs were able to take advantage of their internal capital markets to substitute internal borrowing for external borrowing in this period of crisis. To explore the links between FDI and a well-functioning and sophisticated financial system, we use two distinct but complementary empirical approaches. We first investigate the impacts on FDI of source and destination countries FD, as well as those of the banking crises in developed countries, by applying a method originally introduced by Rajan and Zingales (1998). Their approach relies on identification of the effects of FD through within-country pair, between-industry differences and the assumption that some sectors are structurally more ED than others. To our knowledge, this method has never been employed in the context of cross-country FDI. However, while this method dramatically reduces potential simultaneity and omitted variable biases, it only provides information on the effects of a sophisticated and well-functioning financial system on the volume of FDI in ED sectors relative to the volume of FDI in non-ed sectors. Hence, in a second stage, we directly test for the effects of destination countries FD and the banking crises on the absolute volume of FDI in ED and non-ed sectors, giving us the opportunity to investigate the existence and relative strengths of the competition effect and the access to external finance effect. We will show that any potential omitted variable bias is mild and does not affect our interpretations. Finally, to gain a better understanding of how FD and banking crises shape FDI, we also estimate two-part models where the probability of FDI to occur and the volume of FDI given that FDI takes place are allowed to be two different processes. In our second approach, we provide a much more robust and dynamic analysis of how MNEs 2

4 react to temporary disruptions in access to external finance. We use the occurrence of the banking crises in developed countries as a natural experiment to provide identification of the effects of distressed financial systems on outward and inward FDI within a difference-in-differencein differences framework, where we exploit both the time dimension of our data and firm- or sectorspecific differences in ED. Variations in the sample of source countries will offer additional insights regarding the credit market conditions in developed countries which did not formally experience a banking crisis during the global financial crisis and on the potential substitution of parent borrowing for external borrowing in crisis-affected destination developed countries. To preview our results, we find, holding other factors constant, that source countries FD fosters FDI in all sectors, but disproportionately more in ED sectors in both relative and absolute terms. Destination countries FD increases the relative volume of FDI in ED sectors, but tends to have a negative impact on the absolute volume of FDI in non-ed sectors and a positive, but not statistically significant effect, on the absolute volume of FDI in ED sectors. We interpret this last finding as evidence that two opposite FD-related effects, the access to external finance effect and the competition effect, are at play in the relationship between destination countries FD and FDI. As a corollary, this also suggests an overall ability of MNEs, especially in non-ed sectors, to use their internal capital markets to overcome poor access to external finance in their destination countries. Regarding the effects of the banking crises in developed countries, the banking crises reduced relative and absolute volumes of financially vulnerable outward FDI, 2 in comparison to the outward FDI performance of non-crisis affected countries and holding other factors constant. On the other hand, the absolute volume of non-financially vulnerable outward FDI does not appear to have been depressed in crisis-affected countries, presumably due to investing MNEs relative lack of reliance on external finance and a possible substitution effect between domestic and foreign investment. This substitution effect may have occurred as investing abroad became relative more attractive than investing at home where growth prospects were low and uncertain. The relative stability of non-financially vulnerable outward FDI does not mean that these FDI flows were not affected by the financial crisis. Rather, they dropped, on average, as much as those from the other main source countries, the non-crisis affected developed countries. 2 We use the expression financially vulnerable FDI to refer either to FDI done by firms owned by parent firms observed to operate mainly in ED sectors or to FDI in ED sectors, depending on whether we have exploited parent-specific or sectorspecific differences in ED. Likewise, we use the expression non-financially vulnerable FDI to refer either to FDI done by firms owned by parent firms observed to operate mainly in non-ed sectors or to FDI in non-ed sectors, depending on whether we have exploited parent-specific or sector-specific differences in ED. As we will show, our qualitative results are not sensitive to the measure of ED used. 3

5 While we initially found that the relative and absolute volume of inward FDI in ED sectors tended to be lower in crisis-affected destination developed countries, further investigations reveal that these effects of the banking crises in developed countries captured the fact that key source partner countries, i.e. the other developed countries, had also been affected by the global financial crisis, reducing their capacity to invest abroad in their developed partners. Finally, as for inward FDI in ED sectors, inward FDI in non-ed sectors did not seem to be relatively lower than in other countries, providing further evidence that MNEs can substitute, at least temporarily, internal borrowing for external borrowing through their internal capital markets when they are confronted to poor credit conditions in their destination countries. Overall, our results indicate that source countries FD tends to strongly foster FDI, and disproportionately more so in ED sectors, whereas destination countries FD matter much less and may even, in non-ed sectors, have a negative impact on FDI. The rest of the paper proceeds as follows. In section 2, we motivate in detail the novelty of our four contributions by reviewing what has been accomplished in the existing literature. In section 3, we describe our two distinct econometric approaches, the data used and the estimation method. In section 4, we present our empirical results. Finally, we conclude in section 5. 2 Related literature In this section, we survey the existing literature on FDI, FD and banking crises, and we put forward the empirical contributions that we intend to make. Hints about the impact of FD can be found in the international trade literature, where both theoretical and empirical work have shown that FD can be a source of comparative advantage. The intuition behind this result is that firms in some sectors are more dependent on external finance than others, due for instance to high upfront costs, long operating and cash conversion cycles, a short harvest period or the requirement for continuing investment (Rajan and Zingales, 1998). The reduction of financial market frictions that occur with FD increases the availability and affordability of external finance, allowing the expansion of ED sectors. In this context, if a high FD country decides to engage in international trade with another country only distinguishable by lower FD, standard Ricardian or Heckscher-Ohlin trade models predict that the pattern of trade will be determined by relative FD levels (Kletzer and Bardhan, 1987; Beck, 2002, 2003; Matsuyama, 2005; Ju and Wei, 2011). The high FD country has a comparative advantage in goods produced by the ED sector and therefore specialises 4

6 in the production of these goods. As a result, the low FD country will specialise in the production of goods produced in the non-ed sector. The impact of FD on exports can also be analysed in a heterogeneous-firm trade model with credit constraints (Beck, 2003; Manova, forthcoming). Firms may have to rely on external finance to cover part of the fixed cost incurred by exporters. The productivity cut-off for exporting will decrease for all firms as FD increases, but this impact will be larger for firms in ED sectors, where the fixed cost of exporting is mostly externally-funded. Like Beck (2002, 2003), Manova (forthcoming) empirically finds that FD promotes exports, especially in ED sectors, suggesting that FD is indeed a source of comparative advantage. 3 Understanding the mechanisms underlying the impact of FD on international trade is important because they can be naturally applied when considering the influence of FD in the source country on FDI. Helpman, Melitz, and Yeaple (2004) show how heterogeneity in firm productivity leads to a continuum of entry modes in the foreign market, where only the most productive firms can recover the additional fixed cost of engaging in FDI. Profitable firms below the productivity cut-off for FDI export or produce only for the domestic market. Incorporating credit constraints and sector-specific ED would lead to predictions similar to those obtained for exports in Manova (forthcoming). As FD increases, FDI should also increases, and this effect should be more pronounced in ED sectors. To our knowledge, no test of this hypothesis has previously been done in FDI literature. Given the paucity of empirical evidence, our first contribution to the FDI literature is to investigate whether source countries FD promotes greenfield FDI, especially in ED sectors. Destination countries FD may also matter. The introduction of the U.S. Foreign Direct Investment Program (FDIP) in provides early evidence of the importance of destination countries FD. The purpose of the FDIP was to limit the impact of outward FDI on the U.S. balance of payments by restricting the amount of U.S. investment abroad that could be financed through capital transfers and reinvested earnings, in the absence of offsetting borrowing abroad. 5 In 1969, whereas the volume of U.S. investment abroad did not appear to have fallen relative to the pre-fdip period, the use of longterm foreign borrowing quadrupled, accounting for 45% of the funding sources (Wiley, 1970). Hence, thanks to their destination countries FD, U.S. MNEs were able to circumvent the financing constraints imposed by their source country and to carry on expanding abroad uninterrupted. The extensive use 3 She also comments, albeit in a footnote, that she found that FD in the importing country also matters but to a lower extent than FD in the exporting country. 4 The FDIP was terminated in See also Beenstock (1982) on the impact of exchange control on outward FDI from the United Kingdom. 5

7 of foreign borrowing by U.S. foreign affiliates has continued unabated in the subsequent decades. 6 However, large variations in the external borrowing to affiliate assets ratio exist across countries. Desai, Foley, and Hines (2004) and Aggarwal and Kyaw (2008) show that, among other determinants, 7 the use of local debt by U.S. foreign affiliates is strongly determined by host countries FD. Affiliates located in low FD countries face a higher relative cost of external finance, resulting in a substitution of parent borrowing for external borrowing. This does not necessarily mean that FD in destination countries is irrelevant for MNEs. Desai, Foley, and Hines (2004) report that substitution of parent for external debt is incomplete suggesting that, despite the inherent financial advantage that U.S. foreign affiliates have over local domestic firms as a result of their access to internal capital markets, poor local access to external finance can constrain their expansion. Indeed, Desai, Foley, and Hines (2006) find that the fall in local borrowing rates and in capital account restrictions induced by capital control liberalisations results in an accelerated growth of the sales, assets and capital expenditures of the U.S. foreign affiliates located in the liberalising countries. More recently, Antras, Desai, and Foley (2009) develop a theoretical model in which capital market imperfections in a given destination country decisively influence the entry mode of MNEs, the capital structure of their foreign affiliates, and the scale of their activities. Empirically, while they find results consistent with a positive effect of FD on the scale of multinational activity, they are not able to find a direct statistically significant impact of FD on the volume of sales of U.S. foreign affiliates. This elusive impact of FD on inward FDI is prevalent in the few papers that have tested its role as a FDI determinant in a cross-country setting. Hausmann and Fernàndez-Arias (2000) do not find that higher FD is associated with a higher financial FDI to GDP ratio. Albuquerque, Loayza, and Servén (2005) add nuance to this conclusion by highlighting that FD matters, but only in developing countries. The use by these two papers of balance of payments FDI data as proxy for the activities of MNEs may explain these results, given that home and host sources of funding can be substitutes, which could in itself lead to a negative impact of FD on balance of payments FDI flows; foreign financing substituting for domestic financing would not figure in those flows. Alternatively, the fact that the coefficients on FD variables are not statistically different from zero can be interpreted as the outcome of two conflicting forces: an access to external finance effect and a competition effect. The theoretical model of Ju and Wei (2010) makes clear this second effect. A key insight of their theory 6 Calculations by Feldstein (1995) and Lehmann, Sayek, and Kang (2004) suggest that the share of the assets owned abroad by U.S. MNEs which was financed locally by debt fluctuated between 35-45% during the period. 7 Other determinants of external borrowing are creditor rights, political risk, currency risk, corporate taxes or local management incentives. 6

8 is that the return to physical capital and the interest on financial savings are not, in general, identical. The wedge between the two tends to be larger if a country s financial system is weaker, reflected in either higher financial intermediation costs or poorer corporate governance. Therefore, a country may have a high return to physical capital, but either high financial intermediation costs or poor corporate governance can deter domestic savers from providing more funding to local firms, severely limiting their expansion. On the other hand, foreign investors, who by their nature are less financially constrained, can invest in profitable local projects, enjoying high returns to investment. It follows that higher levels of FD will allow more domestic firms to emerge and expand, lowering the return to physical capital and making the country less attractive to FDI. 8 Hence, by both improving access to external finance and increasing market competition, FD may simultaneously deter and attract FDI, resulting in little impact on aggregate FDI. This could explain why, with the exception of the research done using firm-level data on already established U.S. foreign affiliates, there is surprisingly little empirical evidence that destination countries FD influences FDI. 9 Hence, our second contribution to the literature is to investigate whether the FD-related access to external finance effect or the FDrelated competition effect dominates in the relationship between greenfield FDI from a large number of source countries and destination countries FD. This relationship is likely to be sector-specific as MNEs operating in ED sectors may put relatively more weight on access to external finance than MNEs operating in non-ed sectors. Indeed, a corollary of the competition effect is that the latter can rely on their internal capital markets to obtain the funds they need to conduct their operations in destination countries with poor financial conditions. The well-functioning of financial systems is regularly disrupted by banking crises. Financial distress in the banking system generally results in a credit crunch, including in countries with high FD pre-crisis. Looking at the impacts of banking crises on FDI may thus be an alternative way to assess the role of source and destination countries FD in promoting FDI. On the source country side, the literature on the effects of banking crises on FDI is extremely scarce. Nevertheless, the investigation by Klein, Peek, and Rosengren (2002) of the impact of the Japanese banking crisis in the nineties on Japanese FDI to the United States suggests that source countries FD matters to explain FDI. This study finds that the FDI activity of Japanese firms was inversely correlated with the deterioration of the financial health of their main bank, as measured by 8 Qualitatively similar conclusions can be found in Chor, Foley, and Manova (2008). 9 The positive impact found by Desai, Foley, and Hines (2006) of capital account liberalisations on capital expenditures by U.S. foreign affiliates would suggest that the external access to finance effect dominated the competition effects possibly thanks to U.S. MNEs superior firm-specific advantages. 7

9 Moody s downgrades. 10 Furthermore, Alfaro and Chen (2012) show that the performance of foreign affiliates during the global crisis period was negatively influenced by the incidence of a financial crisis in their parents countries. This indicates that foreign affiliates are sensitive to financial conditions in their parents countries. Overall, little remains known about the impact of a banking crisis on outward FDI. In this context, our third contribution to the literature is to investigate how a banking crisis influences outward greenfield FDI from major source countries, taking into account sector-specific ED. On the destination country side, studies have stressed that domestic and foreign firms do not face the same financial constraints during a financial crisis. Contrary to local firms, foreign affiliates are not limited to local sources of finance, thanks to their international financial linkages with their parent company or related foreign affiliates in other countries. These linkages form a multinational financial system, in which foreign affiliates have access to internal capital markets offering an alternative source of funds, e.g. loans or equity infusions from their parents, when credit constraints in their host countries are particularly stringent. 11 In addition to FDI being relative stable during financial crises (Lipsey (2001), Tong and Wei (2010)), the financial advantage that foreign affiliates enjoy over local firms has the paradoxical implication that financial crises can coincide with increased inward FDI relative to pre-crisis levels, as foreign firms are able to take advantage of crisis-induced opportunities denied to credit-constrained domestic firms. 12 For instance, in agreement with Krugman (2000) on fire-sale FDI, Aguiar and Gopinath (2005) show that M&A FDI significantly increased in the midst of the Asian crisis, with a preference for liquidity-constrained, but high growth prospects, target firms. Hawkins and Macaluso (1977) find that the fixed capital spending of U.S. foreign affiliates rises relative to that of local firms during periods of tight credit conditions in several developed countries. They argue that MNEs profit from lower domestic competition to enhance their own competitive position in the host country. Such an argument is in agreement with the results of Desai, Foley, and Forbes (2008). They show that U.S. foreign affiliates, including those mainly serving the local market, increase their capital expenditures significantly following a large depreciation in their host 10 In the field of international trade, Chor and Manova (2012) shows that during the recent financial crisis in developed countries (identified as the period in the paper), countries with tighter credit availability, as measured by interbank rates, exported less to the United States than countries where credit conditions were more favourable. This effect was particularly pronounced in ED sectors. 11 See Shapiro (2006) for a comprehensive discussion of the MNEs cost of capital and the value of the multinational financial system. Oxelheim, Randøy, and Stonehill (2001) discuss the reactive and proactive strategies that MNEs can adopt to minimise their cost of capital and maximise their availability of capital relative to their competitors. 12 Aghion, Bacchetta, and Banerjee (2004) describe a theoretical model in which there is a stabilising effect of unrestricted FDI during slumps. In addition to FDI not depending on the creditworthiness of domestic investors, foreign investors may be attracted by the depressed price of the country-specific factor (e.g. skilled labour or real estate). 8

10 country, whereas the fixed investment of local firms falls on average. They attribute this divergence of responses to the inability of local firms to overcome the financial constraints that result from the currency crisis, whereas foreign affiliates can rely on internal capital markets to secure the additional funding required to exploit the benefits of a sharp depreciation, in terms both of a weaker currency, and, presumably, weaker domestic competition. 13 Overall, referring back to our discussion on the ambiguous impact of destination countries FD on FDI, it seems that when a crisis temporarily depresses FD, the competition effect dominates the external access to finance effect. However, while the research surveyed has improved our understanding of the role played by MNEs internal capital markets during times of crisis, it leaves many questions unanswered. Existing research mainly looks at the response of M&A FDI or existing foreign affiliates to financial crises in emerging countries with relatively underdeveloped financial markets during periods when parent companies were unlikely to be themselves credit constrained. By contrast, it is not clear how greenfield FDI would react to a banking crisis in a major developed destination country, with the additional possibility that the source country also experiences a banking crisis. MNEs may find it hard to set-up a new foreign affiliate in the midst of a financial crisis, despite a fall in input prices and potentially weaker domestic competition, due to the absence of pre-existing trade and credit linkages with local suppliers, customers or banks. The attractiveness of the crisis-affected destination country may be further reduced if the industrial landscape is dominated by resilient large domestic firms, if vertical linkages are impaired by customers or suppliers normally relying on external finance to conduct their day-to-day business or if the destination of the sales is mostly the local market. These possibilities are more likely to arise in developed countries, rich and with high FD, than in developing countries. Finally, a simultaneous banking crisis in MNEs source countries can disrupt the good functioning of their internal capital markets. On the basis of these considerations, our fourth contribution to the literature is to investigate the responsiveness of inward greenfield FDI to banking crises in developed countries, distinguishing between sector-specific ED and whether the source country also experiences a banking crisis. In the next section, we describe the two econometric approaches that we employ to realise these four contributions, as well as the data and estimation method used. 13 See also, among other works, Blalock, Gertler, and Levine (2008), Kalemli-Ozcan, Kamil, and Villegas-Sanchez (2010), Kolasa, Rubaszek, and Taglioni (2010) or Alfaro and Chen (2012) on the superior performance of foreign firms over domestic firms during financial or twin (currency and banking) crises, partly thanks to their exclusive access to non-local sources of finance. 9

11 3 Econometric approaches, data and estimation method 3.1 Econometric approaches We empirically investigate the influence of a sophisticated and well-functional financial system on inward and outward greenfield FDI in two different, but complementary ways. In a first stage, we mainly focus on the role of source and destination countries FD in promoting greenfield FDI by following the now standard Rajan and Zingales (1998) s approach. 14 More precisely, we use the following log-linear model: ln[e(f DI ijs x ij, αs)] = α ij + α s + β 1 ln(initial FD) i X ED s + β 2 ln(initial FD) j X ED s +γ 1 Banking crisis_f i X ED s + γ 2 Banking crisis_f j X ED s δ p Control variables i X ED s + δ p Control variables j X ED s p=1 p=3 (1) where F DI ijs is a measure of cumulated FDI flows between source country i and destination country j in sector s over a given period, x ij is a vector of regressors, the αs are fixed effects, initial FD is a measure of pre-sample period financial development, ED s is a binary variable indicating whether the sector depends relatively more on external financing than other sectors and the Banking crisis_f variables correspond to the proportion of time spent by a given country into a banking crisis. There are two main advantages to the Rajan and Zingales (1998) s approach. First, the use of presample FD levels and the inclusion of country-specific, country-pair-specific and industry-specific effects 15 help to reduce the risks of simultaneity bias, 16 omitted variable bias or model misspecification. Second, the identification of the impact of a well-functioning and sophisticated financial system on FDI is achieved by focusing on a specific channel, the greater need for MNEs in ED sectors to 14 See, among others, Beck (2002), Kroszner, Laeven, and Klingebiel (2007), Manova, Wei, and Zhang (2011) or Chor and Manova (2012) for an application of this approach in various contexts. 15 Note that including country-pair specific effects is equivalent to a specification including, in addition to country-pairspecific effects, source and destination country-specific effects. See Christensen (2011). 16 Among other factors, the entry of foreign banks can be explained by them following their domestic clients abroad (see Clarke, Cull, Peria, and Sanchez (2003) for a good survey of the literature). If their entry fosters destination countries FD, by increasing for instance the efficiency of the domestic banking system, reverse causality between non-financial FDI and FD can occur. Such a mechanism may explain why Harrison, Love, and McMillan (2004) find that higher financial FDI is associated with an ease of domestic firms financing constraints. 10

12 have access to external finance to engage in FDI, relative to MNEs in non-ed sectors. In doing so, we strongly limit the range of alternative interpretations regarding a link between FDI and FD or FDI and banking crises. Nevertheless, in order to ensure that our interaction terms truly isolate the impact of external finance availability on FDI in ED sectors, we also control for other country characteristics, potentially correlated with FD and which may have a differential impact on FDI in ED sectors, by interacting them with the ED dummy. If the initial level of FD rightly reflects the ease of access to external finance in normal times, initial FD in source and/or destination countries should have a disproportionate influence on cumulated bilateral outward and inward FDI in sectors heavily reliant on external finance for their growth, holding other factors constant. We should find β 1 > 0 and β 2 > 0. On the other hand, if a country experienced a banking crisis during the period under investigation, resulting in a temporary tightening of credit constraints, we would expect outward and inward FDI in ED sectors in this crisis-affected country to be disproportionately affected, relative to other sectors. We should find γ 1 < 0 and γ 2 < 0. These coefficients need to be interpreted in relative terms because, using Rajan and Zingales (1998) s approach, we cannot estimate the impacts of FD or banking crises on the absolute volume of FDI in non-ed sectors, and by extension, on the absolute volume of FDI in ED sectors. In other words, using for illustrative purpose the simplified econometric model ln(e(f DI is x i, α i )) = α i + θ 1 F D i + θ 2 ED s + β 1 F D i ED s or E(F DI is x) = exp(α i + θ 1 F D i + θ 2 ED s + β 1 F D i ED s ), we can estimate β 1 but not θ 1, due to the presence of country-specific fixed effects α i. Therefore we cannot calculate, for example, the sector-specific elasticity of the expected value of FDI with respect to FD. On the other hand, we can focus on the impact of FD on the ratio of the expected value of FDI in ED sectors to the expected value of FDI in non-ed sectors: R = exp(α i+θ 1 F D i +θ 2 +β 1 F D i ) exp(α i +θ 1 F D i ) = exp(θ 2 + β 1 F D i ). If, for instance, FD increases by 1 unit, R becomes R +1 = exp(θ 2 +β 1 F D i )exp(β1) and therefore the multiplicative change in this ratio caused by a one-unit change in FD is equal to exp(β 1 ). β 1 can be seen as a difference-in-differences estimator and, in equation (1), the coefficients on the interaction terms can thus be interpreted as indicating how FD or a banking crisis disproportionately influences the relative volume of FDI in ED sectors. However, estimating a sector-specific impact of a sophisticated and well-functioning financial system on the absolute volume of FDI is also of interest, especially if we wish to assess the existence of a sector-specific interplay between a competition effect and an access to external finance effect in destination countries. A clear manifestation of these two effects would be θ 1 < 0 and θ 1 + β 1 > 0 11

13 where the competition effect dominates the access to external finance effect for FDI in non-ed sectors and vice-versa for FDI in ED sectors. We will therefore also estimate a variant of equation (1) where the country-pair specific effects are replaced by observed variables, allowing us to estimate the overall effects of FD and banking crises on non-ed and ED sectors. While doing so increases the risk of an omitted variable bias, we will show that any potential bias is mild and does not impact on inference. Finally, to gain a better understanding how FD and banking crises shape FDI, we will also estimate two-part models in which the probability of observing FDI and the volume of FDI occurring when positive FDI is observed are separately modeled. In this way, the impact of a given FDI determinant is not constrained to be the same for both parts of the FDI decision process. In equation (1), it is necessary to control for the occurrence of the banking crises in developed countries in order to avoid omitted variable bias as these countries combine high FDI activity and high FD. However equation (1) is inherently a static model, depicting long-term relationships. It is ill-suited to understand how MNEs react dynamically to a temporary tightening in credit constraints, such as those induced by a banking crisis in countries with a normally sophisticated financial system. Furthermore, it does not allow us to control either for sector-country-pair-specific effects or for the contemporaneous values of other FDI determinants at the time of the banking crisis. Finally, there is always the risk that FD may proxy for an omitted variable. Hence in a second stage, we investigate the link between access to external finance and FDI by examining how financially and non-financially vulnerable FDI flows responded to the deteriorating economic and credit conditions engendered by the banking crises in developed countries. The occurrence of these crises provides us with a useful natural experiment. The affected countries are relatively homogeneous, invested or attracted a large amount of FDI across sectors before the crisis, and are numerous. Homogeneity reduces the risk that any identified effects of a banking crisis pick up the effect of an omitted time-varying factor. Large FDI flows offer a good chance to identify the effects of a banking crisis. Finally, a large number of affected countries increases the power of statistical tests and strengthens the external validity our findings. Here, we take advantage of the various dimensions of our data (time, sector, firm) and use the following log-linear model: 12

14 ln[e(f DI fpijst x ijt, αs)] = β 1 Banking crisis dummy it + β 2 Banking crisis dummy jt + +β 3 Banking crisis it X ED p + β 4 Banking crisis jt X ED p 19 + δ p Control variables + α ijs + α st + α t X DEV ij p=1 (2) where F DI fpijst is a measure of the FDI done by firm f owned by parent firm p located in source country i in sector s of destination country j at time t, x ijt is a vector of regressors, the αs are fixed effects, ED p is a parent-specific ED dummy and the Banking crisis variables indicate whether the source or destination country was experiencing a banking crisis in a given year. After having estimated this model, we will examine in greater details the dynamic effects of the banking crises in developed countries by creating a series of dummies D that we will interact with the ED variable. These dummies, which will allow us to track the impacts of the banking crisis on FDI over time, correspond to D 1ct = 1 if t = T and zero otherwise; D 2ct = 1 if t = T + 1 and zero otherwise; D 3ct = 1 if t = T + 2 and zero otherwise; D 4ct = 1 if t = T + 3 and zero otherwise; T denotes the starting year of the banking crisis in country c. These four dummies, and their interaction terms with the ED variables, are created for both source and destination countries. Note that we do not control for FD in equation (2). As previously mentioned, this variable may be endogenous and its presence would interfere with the direct relationships that we wish to uncover between banking crises and FDI. The greenfield FDI database that we use does not provide any firm-specific data, besides identification numbers for firms and their parent companies. 17 We could simply aggregate the FDI data at the sector-country-pair level since a property of the Poisson regression model that we will use to estimate equations (1) and (2) is that use of either aggregated data or unit-level data yields the same maximum likelihood estimates, for a given set of covariate values (Rabe-Hesketh and Skrondal, 2012). However, in an attempt to exploit the firm-level dimension of our dataset, we construct a parent-level indicator of ED, based on the average ED status of the sectors in which the firms affiliated to a common parent company engaged in FDI over the entire period of the sample. This approach is motivated by the fact that observing a FDI project in an ED sector does not necessarily mean that the parent of the firm undertaking this project only invests in ED sectors. Indeed, in our data, 50% of parents only invested 17 The firm and the parent company can be the same entities. 13

15 in ED sectors, 36% only invested in non-ed sectors, but 14% were multi-sectoral as they invested in a mix of ED and non-ed sectors. The impact of the credit constraints induced by a banking crisis on the ability of firms to engage in FDI may depend on the extent to which their parent company is involved in ED sectors (0-100%). Given data constraints, this parent ED indicator is measured with error, since it is based on the observed greenfield foreign activities of the parent companies over the period. For instance, firms related to a common parent company may have invested abroad in ED sectors during the sample period, despite the parent company mainly operating in non-ed sectors in its home country. Alternatively, a multi-sectoral parent company may have decided to invest only in non-ed sectors during the financial crisis. Hence, the parent-specific ED indicator may not accurately reflect the ED of a given parent company. However, despite these drawbacks, this parent-level ED indicator may still provide a better picture of the financial vulnerability of firms than that achieved through use of a sector-level ED dummy, which does not take into account that some firms investing in ED sectors abroad belong to parent companies which are not necessarily heavy users of external finance. Making this distinction may be particularly important if firms investing abroad in different sectors tend to be large, since their observations could have a large influence on the estimates. In addition, it is possible that the relevant measure of ED depends on whether the crisis-affected country is the source or destination of a given foreign project. On the source country side, the correct measure of ED may be at the parent-level. If a multi-sectoral parent firm usually relies in part on the source country s financial system to finance its day-to-day activities at home and, potentially, abroad, a banking crisis will reduce its ability to invest abroad in any sector, including non-ed sectors. On the destination country side, the correct measure of ED may be at the sector-level. MNEs, whether multisectoral or operating purely in financially vulnerable sectors, may decide to invest in the ED sectors of a given destination country only if local external finance is available to support their activities in those sectors, which is unlikely to be the case during a banking crisis. We will start by using ED p for both source and destination countries, as reported in equation (2), and then compare these results with those where we employ ED s in order to establish the most relevant measure of ED on each side of bilateral FDI. We will show that our qualitative results are robust to the measure of ED used. Coefficients β 1 and β 2 are of interest, since they indicate how the absolute volume of nonfinancially vulnerable FDI 18 reacted to the banking crisis in source and destination crisis- 18 As previously mentioned, we use the expression non-financially vulnerable FDI to refer either to FDI done by firms owned by parent firms observed to operate mainly in non-ed sectors or to FDI in non-ed sectors, depending on whether 14

16 affected developed countries. However, they should be interpreted carefully because they are catch-all coefficients which are not tightly linked to a finance-specific channel. For instance, they can capture, in addition to any access to external finance effect, a substitution between domestic and foreign investment effect on the source country side and a competition effect, or the negative impact of economic uncertainty, on the destination country side. This ambiguity explains our initial focus on β 3 and β 4, which indicate whether financially vulnerable FDI flows are disproportionately affected by a banking crisis as compared to other FDI flows. If crisis-induced credit constraints matter for FDI, their impacts should be particularly manifest for financially vulnerable FDI flows. The latter rely heavily on external finance, relative to non-financially vulnerable FDI flows which are not heavy users of external finance. Finding β 3 < 0 and β 4 < 0 would suggest that MNES tend to rely on both source and destination countries banking sector to fund their FDI. To isolate the effects of the banking crises on financially vulnerable FDI of crisisaffected countries, we exploit both the time and the combined firm-sectoral dimension of our data by employing what amounts to a difference-in-difference-in differences identification strategy, where the availability of two control groups allows us to control for two kinds of potentially confounding trends. The first potential confounding trend may be that financially vulnerable FDI fell in all countries during the period, irrespective of whether a given country was experiencing a banking crisis or not. To control for changes in FDI in ED sectors across all countries, we can use as control group (outward and inward) financially vulnerable FDI of countries which did not experience a banking crisis during the period. This involves including variables accounting for systematic differences in FDI across sector-country pairs (fixed effects α ijs, time-varying control variables, and time effects specific to the source or destination group of developed countries α t X DEV ij, e.g. a shift towards M&A FDI ) and interactions between industry-specific and time-specific effects (α st ). Note that α st also control for changes in non-financially vulnerable FDI across all countries. The second potential confounding trend occurs if all FDI flows within the crisis-affected country changed in the same way, irrespective of whether FDI was financially vulnerable or not. To control for changes common to all FDI in the crisis-affected country, we can use as control group outward we have used in our interactions terms a Parent ED dummy or a Sector ED dummy. Likewise, we use the expression financially vulnerable FDI to refer either to FDI done by firms owned by parent firms observed to operate mainly in ED sectors or to FDI in ED sectors, depending on whether we have used in our interactions terms a Parent ED dummy or a Sector ED dummy. 19 Note again that including these sector-country pair effects is equivalent to including all pairwise interactions between sector-specific effects, source country-specific effects, destination country-specific effects and country-pair specific effects. 15

17 and inward non-financially vulnerable FDI in crisis-affected countries. This involves including the banking crisis variables (Banking crisis dummy it and Banking crisis dummy jt ) and their interactions with the parent-specific or sector-specific ED dummy. 20 If the coefficients β 3 and β 4 are statistically different from zero and negative, after having controlled for factors common to all sectors in the crisisaffected countries and global factors affecting non-financially vulnerable and financially vulnerable FDI worldwide, we can conclude with some certainty that, in crisis-affected countries and holding other factors constant, tighter credit constraints resulted in lower outward and inward financially vulnerable FDI. This result could be then interpreted as implying that FD, in the sense of offering easy access to external finance in normal times, contributes to the promotion of FDI, at least on the source country side. On the destination country side, the overall sector-specific effect of FD the volume of inward FDI would still depend on the relative strengths of the competition effect and the access to external finance effect. Comparisons of the effects of the banking crises on inward FDI in non-ed and ED sectors can help to assess the role played by destination countries FD in the attraction of FDI. We will conclude this section by varying the sample of source countries used, allowing us to gain deeper insights into the effects of the global financial crisis on FDI. This will allow us to check whether resident firms in source developed countries classified as non-crisis affected countries have avoided any rise in credit constraints during the global financial crisis, which might otherwise have limited their outward FDI. Investigating this possibility is important to avoid wrongly attributing the effects of higher credit constraints in the source country to the effects of a banking crisis in a destination developed country. Finally, we will be able to look for some indirect evidence on the substitution of parent borrowing for external borrowing during the banking crises in some destination developed countries. 20 Two comments need to be made when we use ED p as the measure of ED. First, when we use ED p as the measure of ED, there is no longer a well-defined control group in crisis-affected countries. Given that ED p is not a binary variable, the degree to which firms are likely to have suffered from crisis-induced credit constraints (the treatment intensity) becomes parent-specific. Second, when we use ED p, we also need to include interactions between this variable and time-specific effects, in order to account for all pairwise interactions between the constitutive terms of the interaction terms. 16