National levels of corruption and foreign direct investment

Journal of Comparative Economics xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Journal of Comparative Economics journal homepage: www.elsevier.com/locate/jce

National levels of corruption and foreign direct investment Josef C. Bradaa,b, , Zdenek Drabekb, Jose A. Mendeza, M. Fabricio Perezc ⁎

a

Arizona State University, Tempe, AZ, USA CERGE- Economic Institute of the Czech Academy of Sciences and Charles University, Prague, Czech Republic c Wilfrid Laurier University, Waterloo, Canada b

ARTICLE INFO

ABSTRACT

JEL Classification: F21 F23 D21 D22 K42

We model the relationship between bilateral foreign direct investment (FDI) and the level of corruption in multinational firms’ (MCNs’) home and host countries. We construct and test a model of bilateral FDI between countries that differ in their levels of corruption. FDI is affected negatively both by the level of corruption in the host country and by differences in home- and host-country corruption. Our model emphasizes that MNCs develop skills for dealing with homecountry corruption, and these skills become a competitive advantage in similarly corrupt host countries. We test the model using data on bilateral FDI stocks among a large number of home and host countries, using a variety of specifications and estimation strategies to provide robustness. Our results show that the effects of host-country corruption and of differences in corruption levels between home and host countries are statistically and economically significant.

Keywords: Corruption Foreign direct investment Competitive advantage Multinational firms

1. Introduction In 1998, El Salvador privatized its telecom sector through the sale of the national telephone company to France Telecom. After a five-year period of erratic profits and some losses, France Telecom sold a controlling interest in its Salvadoran affiliate to a Mexican company, which quickly succeeded in improving the telecom's financial performance. What can explain the success of the Mexican owner in improving the affiliate's performance so markedly? A Salvadoran telecom manager told one of the authors of this paper, “… the Mexicans understood how to get things done in El Salvador, and they were uninhibited and ruthless in comparison to the French.” This vignette reflects a broader phenomenon, the tendency of multinational corporations (MNCs) from relatively corrupt countries to become successful investors in host countries that have similarly high levels of corruption.1 While the ability of MNCs to undertake successful foreign direct investment (FDI) can be explained by many factors, in this paper we explore one specific channel that reflects the insight of the Salvadoran manager that the skills for dealing with corruption are a competitive advantage of MNCs investing in corrupt economies. If the MNC's home country has weak market and legal institutions and thus high levels of corruption, the MNC will develop skills and governance structures that enable it to cope with such an environment. Conversely, firms from countries with good market institutions will develop different competitive skills and Corresponding author. E-mail address: [email protected] (J.C. Brada). 1 For example, Cuervo-Cazurra (2006) and Cuervo-Cazurra and Genc (2008) find that MNCs from developing countries invest more frequently in “least-developed” countries where corruption is high. Perhaps the most often cited example of this phenomenon is China, whose outward foreign direct investment is heavily slanted toward host countries that have weak market institutions and high levels of state involvement in the economy (Buckley et al., 2007; Kolstad & Wiig, 2012). Chinese firms, by virtue of the institutions they encounter in their home country, develop skills in dealing with “complex patron-client relationships and personal and institutional favors in relatively opaque and difficult business environments’’ (Yeung & Liu, 2008). ⁎

https://doi.org/10.1016/j.jce.2018.10.005 Received 8 September 2017; Received in revised form 26 August 2018; Accepted 29 October 2018 0147-5967/ © 2018 Association for Comparative Economic Studies. Published by Elsevier Inc. All rights reserved.

Please cite this article as: Brada, J., Journal of Comparative Economics, https://doi.org/10.1016/j.jce.2018.10.005

Journal of Comparative Economics xxx (xxxx) xxx–xxx

J.C. Brada et al.

organizational structures that reflect their home-country environment. Once these skills and organizational features are developed in the home country, its MNCs are also able to exploit them abroad in host countries with similar levels of corruption and institutions. This idea stems from the seminal work of John Dunning (1998), who argued that MNCs invest in host countries that offer the economic and institutional environment that allows the MNCs’ competitive advantages to be efficiently utilized and that these competitive advantages reflect the MNCs’ home-country environment. Our model of the effects of corruption on MNCs' investment decisions retains the widely accepted view that host-country corruption influences MNCs’ foreign investment decisions by creating economic environments that are either favorable or unfavorable to rational economic activity. Rather than viewing corruption only as an environmental factor that either facilitates or hinders firms’ activities, we construct a model where a home country's institutions have an important impact on the skills and organization of homebased MNCs and where these skills, in turn, influence MNCs’ choices of host countries in which to invest as well as the amount that they invest. We test the model using data on bilateral FDI stocks between 43 home countries and 151 host countries for the period 2005–2009. Our results confirm that both host-country corruption levels and differences in corruption levels between home and host countries influence the likelihood and size of bilateral FDI stocks. Our most conservative estimates show that average FDI stocks will be 19% lower, ceteris paribus, for every one-unit decline in the host country's corruption perception index (CPI), and bilateral FDI stocks will be 4% smaller for each additional CPI unit difference between the home and host country. In the next section of this paper, we present a brief literature survey to place our contribution in the context of the research on MNCs’ FDI decisions. We also discuss the economic intuition for our model and then formally develop it and draw out its implications. Section 3 is devoted to the empirical test of the model, including data description, methodology and empirical results. Robustness checks are reported in Section 4, and Section 5 concludes. 2. A general model of corruption and MNCs’ competitive advantage 2.1. Literature review and economic intuition of the model There is an extensive literature on the effects of corruption on FDI. One stream of work (e.g., Al-Sadig (2009); Drabek and Payne (2002); Hines, (1995); Smarzynska and Wei (2002); Wei (2000)) focuses on the effect of host-country corruption on inward FDI. With some exceptions, studies of this type find that host-country corruption had a negative effect on inward FDI because MNCs investing in corrupt host countries encounter predation both by the government and by private agents, a lack of protection for their property, including intellectual property, and costly contracting and transactions with other economic agents. Such an environment imposes costs on firms that reduce their productivity and profits. While accepting the findings of a negative influence of host-country corruption on FDI, a second stream of research emphasizes that differences in home- and host-country corruption levels also influence bilateral FDI. Habib and Zurawicki (2002) find that skills in managing corruption in the MNC's home country give it a competitive advantage in corrupt host countries although such an advantage is diminished or can become a competitive disadvantage in less corrupt host countries. Wu (2006) also finds that corruption differences serve to reduce bilateral FDI and that the effect of such a difference is asymmetrical, meaning that whether the difference between home and host corruption measures is positive or negative determines the size of the effect on bilateral FDI, as also suggested by Habib and Zurawicki. Wu emphasizes skills in bribing officials as the key way in which skills learned in a corrupt home country can help MNCs succeed in corrupt host countries. Holburn and Zelner (2010) take a wider view of the skills that MNCs acquire in their home countries, stressing the ability to cope with opportunistic government officials and dealing with pressures for redistribution. Using data for FDI in the power generating sector, they verify that investors who face such environments in their home countries are less resistant to investing in countries with similar environments.2 Cezar and Escobar (2016) also take a broad view of home- and host-country environments, constructing an index that uses several measures of institutional quality of which corruption is only one. They find that greater home-host country differences in this index reduce both the likelihood of bilateral FDI taking place and of the amount of FDI that does take place.3 One novelty of our paper relative to the previous literature is that we use a larger sample of countries and, in particular, that we include non-OECD developing or middle-income counties in our sample of home countries, providing a larger range of home country corruption levels. Our sample consists of 43 home countries and 151 host countries. This larger sample and the inclusion of a broader range of home countries improves the precision of our estimates. By comparison, Habib and Zurawicki (2002) have a sample of only seven home countries, Germany, Italy, Japan, Korea, Spain, UK and the USA, so home-host corruption differences in their sample tend to be in favor of the home country. Wu (2006) uses 24 OECD countries and 52 host countries, although 23 of the host countries are other OECD countries weighing the sample in favor of FDI among relatively high-income and low-corruption countries. Cezar and Escobar (2016) use a larger sample yet, but it, too, uses only the 31 OECD countries as home countries and 125 host countries. A second contribution of this paper to the empirics of the corruption-FDI nexus is that we employ a number of different estimation 2 In general, the management literature presents a richer picture of the skills and strategies that enable MNCs to cope with host-country environments, See, for example Oliver and Holzinger (2008). 3 Brada, Drabek, and Perez (2012) also conclude that home-host country differences in corruption reduce both the likelihood and the volume of bilateral FDI, although the result comes from a model that posits non-linearities in the effects of home- and host-country corruption, and they do not include a corruption difference variable in their specification.

2

Journal of Comparative Economics xxx (xxxx) xxx–xxx

J.C. Brada et al.

techniques to account for issues such as non-linearity, selection bias and the fact that there are many country pairs that do not undertake bilateral FDI. The panel nature of our data enables us to test for non-linearity, to include home country and time fixed effects and to account for a number of statistical problems that plague studies of bilateral FDI. Previous studies have used a variety of specifications to explain bilateral FDI, some based broadly on the gravity equation, others more ad hoc. In general, the number of explanatory variables in these studies has been limited, leaving open the possibility of missing explanatory variables. In this paper, we use the work of Blonigen and Piger (2014), who identify the most likely correlates for a gravity specification of FDI stocks. In our base model, we employ their most likely 11 covariates, and we also estimate a specification with additional covariates. This procedure allows us to minimize the likelihood of omitted variable bias and also allows us to test the findings of these earlier studies. Moreover, our specification enables us to provide dollar estimates of the effect of host-country corruption and of home- and hostcountry differences in corruption. On the theoretical side, we set out an explicit model of how cultural distance influences FDI. Much of the work reviewed above provides largely intuitive and verbal explanations of why cultural distance reduces FDI flows. While formal models may miss the richness of cultural and institutional effects of differences between home and host countries on FDI, formal models can sharpen our understanding of the mechanisms through which corruption and FDI interact. Specifically, we model the effect of both host-county corruption and the corruption distance between home and host countries.4 We solve the model for bilateral FDI in a multi-country setting where countries are identical except in their levels of corruption. Using this model, we then make predictions about the pattern of bilateral FDI that we test in our empirical work. In our model, we propose an explanation for greater FDI between countries with similar levels of corruption or institutional quality that is based on the work of John Dunning (1998), who argued that MNCs develop competitive competencies and governance structures that reflect specific aspects of their home-country environments such as resource endowments, factor prices, levels of development and intensity of domestic competition. Host-countries that offer similar environments, which complement the MNC's core competitive advantages, will be attractive candidates for FDI. We propose that MNCs also develop competencies in dealing with their home country's level of corruption or institutional quality. Firms in corrupt countries will learn how to deal with an environment characterized by predation, uncertainties in the security of transactions and ambiguous property rights. Conversely, they are less likely to develop firm-specific advantages such as proprietary technologies, brand names, etc. because these are largely unprotected in their home countries, and thus they will find it hard to compete in countries where corruption is low and domestic firms do have such firm-specific advantages. Skills in coping with corruption and weak home-country institutions should be transferable to other countries where the firm decides to invest, but the value or transferability of the skill will depend on the level of corruption in the host country. If both home and host country have similar levels of corruption, the skills learned in the home country can provide valuable competitive advantages in the host country, but if the host country has a very different corruption level, then the skills learned in the home country will yield small or no competitive advantages abroad. Similarly, firms in home countries with low levels of corruption will have less experience and ability in dealing with a corrupt environment in host countries.5 Thus, there are two ways in which corruption influences MNCs’ foreign investment decisions. One is what we call the corruption environment effect (CEE), which reflects the effects of host-country corruption levels on the productivity or economic efficiency that firms can achieve given the institutional quality of the host country. The second effect is what we call the skillmatching effect (SME), which is the effect on MNCs’ FDI decisions of similarities between home- and host-country levels of corruption. In this paper, we propose a model that incorporates these two ways in which corruption influences MNCs’ investment decisions. The model predicts that MNCs’ foreign investment will be greater to host countries with lower levels of corruption, and that MNCs’ foreign investment will be greater the more similar are home and host countries in their levels of corruption. Moreover, differences in corruption levels will influence both the likelihood of investment between any two countries taking place and the volume of FDI that takes place. 2.2. Model of corruption and foreign direct investment To focus on the role played by differences in cross-country corruption levels, our model makes simplifying assumptions that make home and host country corruption levels the only country-level determinants of MNCs’ investment decisions. Specifically, we assume that all countries are identical in terms of size, available production technologies, consumer preferences, endowments, etc. The ten stylized countries in our model are assumed to differ only in the level of corruption that prevails throughout their economy. Each country falls within one of ten corruption categories where a value of 1 signifies that a nation is most corrupt and 10 signifies the one that is least corrupt.6 Additionally, we treat the skills and experience in coping with corruption as a public good so that their transfer 4

This leads us, in our empirical work, to provide estimates of the relative importance of these two effects. For evidence that skills in dealing with institutional weaknesses developed by MNCs in their home country carryover to host countries, see Henisz (2016), Oliver and Holzinger (2008), Westermann-Behaylo, Rehbein, and Fort (2015). The closely related idea that countries’ comparative advantage may be shaped by domestic institutions is well developed in the literature. See, for example, Belloc (2006) and Levchenko (2007). Moreover, Athanasouli and Goujard, (2015) and Mironov (2015) demonstrate the advantages that managers used to corrupt environments have over managers from less corrupt countries when an MNC's foreign affiliate is located in a country characterized by high levels of corruption. MNCs may also learn skills for coping with corruption by doing business in corrupt host countries rather than in their home country, but we view such learning to be of secondary importance. 6 The categorization is motivated by Transparency International's Corruption Perceptions Index that ranks countries’ corruption on a scale of 1 to 10, with 10 being least corrupt. This is the scale used in the empirical part of this paper. 5

3

Journal of Comparative Economics xxx (xxxx) xxx–xxx

J.C. Brada et al.

and employment in another country does not diminish the amount available for use by the MNC elsewhere.7 Each country has a group of MNCs that are identified with that nation such as American multinationals or French multinationals, etc. Additionally, consumers in each country view the output of their national group of MNCs as similar, but not identical, to the outputs of the other nine foreign multinational groups operating in their country. That is, in each nation, the preferences of consumers are assumed to be as in Armington (1969) in that the product produced domestically by the domestic firms and of foreign affiliates of the MNCs from the other nine nations fall within the same broad product group, say manufacturing, but consumers treat each product as a different variety or product segment. Thus, each nation has a unique corruption rating and product variety. As a result, in what follows, we use a nation's corruption rating to identify both the nation and its product variety. Specifically, subscripts i and j refer to nation i and nation j respectively or to product varieties i and j where i denotes home countries and j host countries. The context will clarify the appropriate usage. We now turn to the formal model, beginning with demand side conditions in each host country. Assume that consumers in host country j spend a fraction βij of their total budget (Yj) on product variety i within the MNC product group. Also assume that the sum of ij < 1. If we ignore products outside the broad MNC product group and adopt Cobb-Douglas prethese shares is less than one: ferences, we can then write the utility function for consumers in country j as 10

Uj =

i=1

Xij ij

where Xij is the amount of variety i consumed by consumers in host country j. This specification yields simple product demand functions so that the amount of each variety demanded, Dij is given by:

Dij =

1 Pij

ij Yj

i = 1, …, 10

(1)

where Pij is the price of variety i in host country j. In the ensuing analysis we assume that the βij are identical across the varieties and the same for all ten countries.8 The production side of the model assumes that all firms within this “manufacturing” product segment populated by the ten groups of multinationals share the same Cobb-Douglas production technology where the output of variety i by a MNC from country i in host country j is given by:

Qij = Aj K ij k Lij l

i = 1, …, 10

where αk and αl are, respectively, the capital and labor factor shares, Kij and Lij are the capital and labor inputs used by MNCs from country i in country j, and Aj is total factor productivity. This production function yields a tractable conditional factor demand function for capital. Letting wj and rj represent, respectively, the wage and rental rate for capital in country j, the solution to the standard cost minimization problem yields the following demand for capital by MNCs from country i in country j:

Kij = Assuming

Kij =

k

wj

k

rj

l

+

l

l/( l +

k)

Qij

1/( l + k )

Aj

= 1, simplifies this expression to:

wj

k

rj

l

l

Qi Aj

(2)

These expressions, known as conditional input demand equations, indicate each investing firm's optimal or desired capital stock in country j. We now turn to the treatment of corruption and its impact on investment. We can proceed in one of two ways. One is to treat corruption as a tax on the sale of a final good. This would be appropriate in those cases where corruption is facilitated by the physical presence of each unit so that the additional costs per unit are in turn transferred to consumers in terms of higher prices. An alternative approach is to treat corruption as causing a reduction in total factor productivity (A). Here, productivity declines as producers divert time, energy, and resources to addressing the demands placed on them by a corrupt environment and, as a result, they produce less output from a given level of inputs. Both approaches lead to identical results in our framework, and thus, without loss of generality, we treat the effects of corruption akin to the imposition of a tax on the final good.9 In the analysis that follows, we assume that two broad influences impact the magnitude of the corruption tax that a multinational from country i faces when operating its affiliate in country j. This tax, denoted as τij, will reflect the skill matching effect and the corruption environment effect. 7

This assumption simplifies the model, but it is not critical. Note that if there are costs to transferring such coping skills internationally, then the negative effect of differences in home-host corruption on FDI would be greater. Such costs could be modeled as are iceberg trading costs in models of international trade, although, in this case, it is the amount of capital transferred that would be “melted” by cultural distance. 8 These assumptions greatly simplify the exposition of the model, but they are not essential for the qualitative results that follow later. 9 In this formulation we follow Wei (2000), who also views corruption as imposing a tax on the foreign investor and estimates the tax-equivalent of a host country's level of corruption. 4

Journal of Comparative Economics xxx (xxxx) xxx–xxx

J.C. Brada et al.

The skill-matching effect depends on two factors. First, it depends directly on |Ci Cj|, the absolute value of the difference in corruption levels between the home country, the location where the multinational gained its experience in addressing corruption, and the host nation.10 The more similar the home and host countries’ corruption ratings, the lower is the value of |Ci Cj| and hence the lower the implicit tax rate on foreign MNCs; the higher the value of |Ci Cj| the higher the tax. Second, in terms of the skill-matching effect, we also consider the breadth or span of applicability of the skills learned in one setting to a different setting. The corruption-tax level may be low if the host and source nation have similar corruption level, but how close must the host country's corruption level be to the source nation's level? To capture the transferability of the skills learned in the home-country setting to the host country, we introduce the parameter δ. If the skills learned coping with the home-country level of corruption are specific to a given level of corruption, then δ has a low value, say 1, meaning that skills learned in a nation with corruption level 3 may not be very applicable in a nation with a corruption classification only one level higher (corruption level 4) or lower (corruption level 2). If the skills are easily transferable and relevant across a broader range of corruption environments, so that δ has a higher value, say 5, then skills learned in a country with one level of corruption will, to a lesser extent, also be applicable in host nations with a corruption classification that is somewhat different from that of the home country. In the extreme, skills learned in a home country could be so highly applicable in any host nation that the corruption tax becomes irrelevant, in which case δ = 10. To capture the importance of the span of applicability of skills gained coping with corruption, we scale |Ci Cj| by (10 – δ) where δ is assumed to range from 1 to 10. Putting these two elements together, the skill-matching (SME) component of the corruption tax is:

SMEij = 1 +

1

(10

) Ci

Cj

(3) 11

where λ is a scalar used to parameterize the corruption tax rate. The corruption environment (CEE) component is straightforward to model. All else the same, a host nation with a higher level of corruption will impose greater costs on firms than one with less corruption.12 Thus,

1

CEEij =

(10

Cj )

(4)

where ϕ is a scalar that determines the maximum tax rate. For instance, if when Ci = 1. Combining the SME and CEE effects, we have ij

=1+

ij

=1+

1

1

(10

Cj ) +

(10

Cj )

1

(10

) Ci

Cj

if i

j

= 20 , the maximum tax rate is to 45%, which occurs

(5)

if i = j

(6)

Eq. 6 reflects the fact that the skill matching effect does not apply to the home country MNC group when it invests in the home country economy, but the corruption environment effect (CEE) does apply to those firms as well. Also, note that when the least corrupt country is also the host country, then 10,10 = 1. The final set of assumptions pertains to the market conditions under which the various firms operate. We abstract from general equilibrium effects so that the home and multinational firms are price takers in the host countries’ product and factor markets. This means that all firms take factor prices, the wage rate w for labor and the rental rate r for capital, as set exogenously. It also means that all firms are price takers in their respective product market and that they set price equal to marginal cost. In light of the earlier assumptions regarding the similarity of the technology, all firms have the same unit costs. Using duality results and defining c as marginal cost, the competitive pricing conditions (i.e., price equals unit cost) for the host and other MNC groups may be expressed as:

Pij =

wj l

l

rj k

k

1 Aj

ij

=c

ij

(7)

for the output of the multinational firms. Foreign firms will encounter a higher implicit tax due to the SME effect, which will raise their costs and thus the price of their products in the host country, and this will reduce host-country demand for their product. As we noted earlier, corruption could also be modeled as reducing the efficiency of multinationals, which would be captured by a reduction in A. This also would lead to a higher price, which confirms our earlier comment regarding the equivalence of the two approaches to modeling the effects of corruption. We are now able to derive an expression that indicates how the desired level of investment in a host country is impacted by its level of corruption. We begin by substituting Eq. 7, the price equation, into the set of product demand equations identified as Eq. 1. Doing so allows us to determine the values of the Qij’s or industry outputs for the home firms and each of the foreign firms. These values are in turn substituted into the conditional factor demands given by Eq. 2 to obtain, K ij* , the desired stocks of capital of homecountry firms and multinational firms as: 10

Using the absolute value rather than the difference assumes symmetry for this effect. We test for symmetry in our empirical work. Since the tax is multiplicative, if either δ = 10 or Ci = Cj, then SME = 1 meaning there is no SME effect. 12 So as not to complicate the model unnecessarily, we ignore the effect of home country corruption on FDI stocks, although we include it in our empirical tests. 11

5

Journal of Comparative Economics xxx (xxxx) xxx–xxx

J.C. Brada et al.

Desired FDI as % of Country 10 K*

120% 100% 80%

Host CC=1 Home Ci=1 Host CC=2 Home Ci=2

60%

Host CC=4 Home Ci=4 40%

Home Ci=6 Host CC=6 Home Ci=8 Host CC=8

20%

Home Ci=10 Host CC=10 0% 1

2

3

4

5

6

7

8

9

10

Cj (host country corrup!on)

Fig. 1. Desired Capital Stock K ij*. The Figure shows the desired capital stock K ij* in host country j from home country i, as a function of countries’ corruption classifications. K ij* is measured as a percentage of the capital stock desired by home-country firms in the least corrupt country (Ci = 10 ). Note: δ = 2, λ = 180.

K ij* =

1 ij

Yj

k

rj

l

l

/

l k

k

+

k l

l

=c

1 (8)

ij

It is evident that the K ij* are functions of corruption levels across countries as captured by the τij term. Higher levels of τij, whether due to higher levels of corruption and thus greater CEE effects or to greater SME effects resulting for greater differences in home-host corruption levels, reduce the amount of foreign investment that MNCs desire to undertake. Note that since all MNCs in a given country are identical, their FDI decisions will also be the same, and thus aggregate FDI stocks reflect firm-level FDI decisions. Consequently, our model can be tested using aggregate rather than firm-level FDI stocks. Several conclusions flow from Eq. 8. If there is no corruption in any of the countries, τij = 1, and the desired capital stock for both home- and host-nation multinationals in each country is the same given our assumptions of identical endowments, tastes and technology. This means that each country's MNCs would invest in all other countries and that they would have the same size of production facility and charge the same price in all countries, including in their home country. With inter-country differences in corruption levels, the CEE effect will result in smaller desired capital stocks for both foreign and home-country firms in more corrupt countries. The skill matching effect implies that the optimal capital stock, K ij* , desired in host country j is highest for its own country multinationals since there is no skill matching tax on domestic firms (compare Eqs. 5 and 6). MNCs from other countries will have smaller desired capital stocks, with MNCs from countries that differ the most from the host country in their corruption level seeking the smallest investments. Fig. 1 shows the influence of the corruption environment and skill-matching effects on desired levels of investment as predicted by a numerical solution of our model. The Figure shows the optimal capital stock K ij* directed to host country j from MNCs from various home countries i. In each country, the highest levels of desired capital come from home-country MNCs (as depicted by the peaks in each line) because home-country MNCs are not subject the SME tax. They are, however, subject to the CEE tax, and so MNCs from more corrupt countries will have smaller K ij* in their home country than do MNCs whose home is in less corrupt ones. The SME effect is shown by the “inverse V” shape of each home country's MNCs’ K ij* , which reflects the growing cost of the SME effect, and thus the lower K ij* , as the corruption differences between the home and host countries increase. To sum up, our model predicts that, ceteris paribus, differences in the corruption level between home and host countries reduce the desired volume of bilateral FDI. Moreover, countries with either very high or very low levels of corruption may also expect to receive or undertake fewer investment projects in total. Finally, MNCs from less corrupt countries will undertake larger FDI projects than do MNCs from more corrupt ones. Cezar and Escobar (2016) construct a model of FDI that arrives at similar conclusions regarding both the size and likelihood of FDI from one country to another. Their and our models do, however, have substantive differences. First, both models focus on differences between home- and host-country levels of corruption, but our analysis through the introduction of the corruption environment effect (CEE) also considers the level of corruption in the receiving country, independent of differences in home and host corruption levels (see the discussion above related to Equations 4 through 6). Second, in Cezar and Escobar's framework, the differences in crosscountry corruption levels are treated as fixed costs (‘adaptation costs’ ) borne by the firm when establishing a facility in the host country. These adaptation costs in turn bear directly on the firm's decision to invest abroad. The more similar the two nations are in terms of institutions, the lower these adaptation costs and the more likely it is that a firm will decide to build a facility in a given host 6

Journal of Comparative Economics xxx (xxxx) xxx–xxx

J.C. Brada et al.

country.13 While such a fixed cost approach may capture the effects of paying bribes to enter a host country, we believe that this is only one aspect of the effects of corruption and that other aspects of corruption continue to burden firms once they enter a country. Thus, in our model we allow for the influence of bribery or corruption on the firm's costs on an ongoing basis. Finally, our modeling approach also incorporates more recent concepts developed by Johanson and Vahlne (1977, 2009), work that also motivates Cezar and Escobar (2016). For instance, Johanson and Vahlne (1977) introduced the notion of psychic distance, which they defined as the “factors that make it difficult to understand foreign environments.” These factors encompass, among other things, differences in language, laws and rules. In an update of their work, Johanson and Vahlene (2009), note that more “general knowledge of foreign environments” can also have the effect of reducing psychic distance more broadly and, as a result, it makes it more likely that a firm will decide to invest abroad in a foreign environment that does not necessarily match closely the conditions in the home market. Our notion of the breath of applicability of the knowledge gained, reflected in the second term of the SME, is consistent with this notion of “general knowledge”. The skills learned by a Mexican MNC in Mexico, for example, are more broadly applicable in a group of countries that are similar. This is captured by a larger value for our applicability parameter, which translates directly into a lower cost of operating in the foreign environment. 3. Testing the model In this section, we test the predictions of our model regarding corruption and MNCs’ investment decisions. Our tests are based on bilateral foreign direct investment stocks obtained from the UNCTAD FDI database for the period 2005–2009. We use FDI stocks to be consistent with our theoretical model since FDI stock data are an indicator of size of MNC affiliates in host countries, which is what our theoretical model addresses. Moreover, we use stock data in order to be able to take advantage of the findings of Blonigen and Piger (2014) regarding the appropriate specification for the gravity equation used to estimate bilateral FDI stocks. There are, of course, broader arguments for using stock data rather than flow data. One argument is that stock data is less volatile than is flow data and it presents fewer zero and negative observations (Cezar and Escobar, 2016). Stock data also better capture the size of a foreign firm's investments in the host country because some of the stock may be financed by borrowing from host-country banks and investors, something that flow data would miss. At least in theory, if not always in practice, stock data should also capture the effects of exchange rate changes, forgiveness of affiliate debt by parent companies, expropriation, etc., that is not captured by FDI flows (Fujita, 2008).14 3.1. The benchmark gravity model The model developed in Section 2 assumes that all countries are identical except for their levels of corruption. In our empirical work, we embed the theoretical insights of that model into a broader empirical model of bilateral FDI based on the gravity equation with additional covariates that capture important drivers of FDI. There is little consensus in the literature about the appropriate set of variables that should be included in a model of FDI. Blonigen and Piger (2014) argue that the specification of the model is itself a “variable” and they utilize a Bayesian approach to select the set of variables that should be used in a gravity equation explaining FDI. Our benchmark model follows their results for estimation of FDI models based on stock levels.15 Of the 52 variables analyzed in their study, only 7 have inclusion probabilities of more than 50% in a model of FDI. We include the 11 variables with inclusion probability above 30%. These include: (1) home-country real GDP, (2) host-country real GDP, (3) distance, (4) similarity of host- and home-country real GDPs, (5) squared home and host GDP differences, (6) host skill level, (7) interaction of GDP differences with education differences, (8) common official language, (9) past colonial relationship, (10) existence of a regional trade agreement,16 (11) existence of a bilateral investment treaty (BIT).17 13 Cezar and Escobar (2016) also assume that adaptation costs (and thus the fixed cost of the facility in the host nation) are inversely related to the firm's productivity, transferred from the home country. The greater the number of firms having a productivity level exceeding a threshold required for its investment to be profitable in the host country, the greater the volume of FDI from home to host country. This assumption of differences in firm productivity makes their approach similar to that of Brada et al. (2012), who also construct a model of FDI where both corruption levels and corruption differences play a role. In the Brada et al. model, firms in each home country differ in their capabilities for dealing with corrupt environments, and, if their capabilities match those of firms in countries with different levels of corruption, then they will undertake FDI to that country. 14 We acknowledge the sometimes-deep division in the literature on the appropriateness of using FDI stocks or FDI flows as the dependent variable in studies such as this. While both stock and flow data have their shortcomings, as Wilkins (2012) concludes, “… FDI stock continues to be an excellent indicator of MNE activities and one that can be used as public policies are formulated.” Nevertheless, given the existing disagreements over stocks versus flows, we ask the reader to note that we have obtained qualitatively similar results using FDI flows as the dependent variable. These results are available from the authors upon request. 15 As we explain in our empirical section we do not estimate the model in log form (i.e., using log(FDI) as the dependent variable) since that would drop all country years’ observations with zero FDI stock and could also lead to biased estimators as shown in Silva and Tenreyro (2006). We deal with possible skewness in the FDI stocks by using a Poisson pseudo-maximum-likelihood (PPML) that includes zero FDI stocks. 16 Blonigen and Piger (2014) use service sector agreements, defined as economic integration agreements in services between countries. Due to data availability we use regional trade agreements instead, as these also capture trade openness between countries. This choice of variable is unlikely to affect our results since our corruption variables should share the same level of correlation with either of these two proxies. 17 It may seem surprising that bilateral trade flows are not among the explanatory factors suggested by Blonigen and Piger (2014). This is likely due to two problems that the use of bilateral trade flows as an explanatory variable for bilateral FDI create. One problem is conceptual: bilateral FDI

7

Journal of Comparative Economics xxx (xxxx) xxx–xxx

J.C. Brada et al.

Countries’ GDPs capture market size. Larger countries attract greater FDI stocks not only because the demand for the MNC's products is greater but also because larger markets offer scale and scope economies and because MNCs find larger host countries more profitable if there are fixed costs of undertaking FDI. Furthermore, larger economies are more likely to have a greater variety of specialized factors of production and resources attractive to foreign investors. These effects are captured by home- and host-countries’ GDPs, squared GDP differences, (GDPi GDPj )2 , and the similarity of host and home real GDP, which is defined as:

GDPij Similarity = [GDPi/(GDPi + GDPj )] × [GDPj /(GDPi + GDPj )] where GDPi and GDPj are the real GDPs of the home and host countries respectively. The distance between the capitals of home and host countries measures transportation and trade costs between countries. The effect of distance is ambiguous because FDI overcomes high transportation costs for low-value bulky goods or for non-tradables and thus distance promotes FDI, but proximity also has a positive effect on FDI because it implies similar tastes and consumption patterns, promoting FDI to serve host-country needs. Host country factor endowments (Skillj) are also an important motive for FDI (Helpman, 1984; Markusen and Maskus, 2002). We use the United Nations Human Development Index (HDI) to capture skill endowments.18 The HDI is a comprehensive measure of development that aggregates country-level achievements in education, life expectancy and income levels. We also use alternative measures of skills in our extended model below. In addition, in order to account for nonlinear effects, interaction terms are appropriate, and we include the interaction of GDP differences with education (skill) differences (Interaction1). The benchmark model also includes variables that capture cultural distance factors such as common official language and past colonial relationship, defined as dummy variables that take the value of 1 if countries have a common language or a past colonial relationship respectively. Finally, we also include variables capturing established relations between countries such as regional trade agreements and bilateral investment protection agreements. Additionally, in order to capture potential time effects and home-country effects, we include time and home-country fixed effects in our linear specifications. 3.2. Data Our sample of countries, reported in Table 1, covers bilateral FDI stocks for 43 home countries and 151 host countries.19 The explanatory variables described above come from the World Bank's World Development Indicators. As in many of the related studies of FDI mentioned above, our home and host country corruption measure is the Transparency International Corruption Perceptions Index (CPI).20 The bilateral investment stocks, measured in billions of US$, are taken from the UNCTAD FDI database, and we employ data for 2005–2009. The home countries encompass a broad variety of country sizes, locations, levels of development and corruption levels. The host countries make up a large fraction of all countries. The bilateral distances and other cultural distance variables are obtained from the French Research Center in International Economics’ (CEPII) Geodist data set. Descriptive statistics of the FDI stocks in our sample are presented in Table 2. Positive FDI stocks represent 76.44% of our sample, and 23.56% are zero. There is significant time variation in FDI stocks. The size of positive stocks peaks in 2009, with the average bilateral investment stock of 7 billion US dollars (US$), although the spread, as is to be expected, is quite large. In Table 3 we present descriptive statistics of the corruption index by year and host or home country. There is a small difference between the average level of corruption of home and host countries. Home countries have a lower average corruption level (10-CPI) but, as Table 3 shows, both home- and host-country samples encompass a broad range of corruption levels, and a more detailed examination of the data shows that differences in corruption levels between pairs of home and host countries vary over the sample period. 3.3. Estimation In order to estimate the model and test its implications, we use both a non-linear specification to reflect the theoretical model and linear specifications to account for issues that frequently arise in studies of FDI. We also estimate the model using Poisson pseudomaximum-likelihood (PPML) estimation to deal with possible skewness in the dependent variable. 3.3.1. Non-linear specification According to our theoretical model, the relation between the desired capital stock of an MNC and the SME and CEE effects is given (footnote continued) is likely to “cause” bilateral trade because, for example, sales and purchases between parent firms and their foreign affiliates can increase when an MNC invests in the host country, but it is also well known in the literature that trade “causes” FDI because entry into foreign markets via exports is often a prelude to entering those markets through FDI. This would cause an obvious causality problem with our estimates. The second problem is an econometric one: if bilateral trade is well explained by the gravity variables (incomes, distance, population, etc.), then bilateral trade flows are effectively a linear combination of these explanatory variables (with some error term added). Thus, adding bilateral trade flows as an explanatory variable introduces a variable that is close to a linear combination of the explanatory variables being used in the gravity regression for FDI, which leads to serious collinearity problems. 18 The Human Development Index is published by the United Nations Development Programme in their Human Development Reports. 19 Data sets used to study MNCs decisions at the firm level often incorporate only a few countries, resulting in low variation in corruption levels. See, for example, Javorcik and Wei (2009). 20 Cuervo-Cazurra (2006) and Wei (2000) find that substituting other measures of corruption for the CPI does not change their conclusions regarding the effect of corruption on FDI. We use other measures of corruption in Section 4. 8

Journal of Comparative Economics xxx (xxxx) xxx–xxx

J.C. Brada et al.

Table 1 List of Home and Host Countries. Home countries (43): Australia

Croatia

Iceland

Malaysia

Spain

Austria Belarus Belgium Brazil Bulgaria Canada Chile China

Cyprus Czech Republic Denmark Finland France Germany Greece Hungary

Ireland Israel Italy Japan Korea, Rep. of * Latvia Lithuania Macedonia *

Netherlands New Zealand * Norway Oman * Poland Portugal Romania Russian Federation

Sweden Switzerland Thailand Turkey United Kingdom United States

Host countries (151): Afghanistan *

Croatia

Ireland

Namibia

Spain

Albania Algeria * Angola Argentina Armenia Australia Austria Bahrain * Bangladesh * Barbados * Belarus Belgium Belize * Benin * Bolivia Bosnia and H. * Botswana Brazil Brunei D. * Bulgaria Burkina Faso Cambodia * Cameroon Canada Chile China Colombia Congo* Congo Dem. Rep* Costa Rica

Cyprus Czech Republic Cote d'Ivoire Denmark Dominican Rep. Ecuador Egypt El Salvador Equat. Guinea* Estonia Ethiopia* Fiji* Finland France Gabon* Georgia Germany Ghana* Greece Guatemala Guinea Guyana* Haiti* Honduras Hong Kong* Hungary Iceland India* Indonesia* Iran*

Israel Italy Jamaica* Japan Jordan Kazakhstan Kenya* Kuwait* Kyrgyzstan* Lao PDR* Latvia Lebanon* Liberia Libya* Lithuania Luxembourg Macedonia, * Madagascar* Malawi* Malaysia Maldives* Mali Malta* Mauritania Mauritius Mexico Moldova Mongolia Morocco Mozambique

Nepal* Netherlands New Zealand* Nicaragua Niger* Nigeria Norway Oman* Pakistan Panama Papua New Guinea* Paraguay Peru Philippines Poland Portugal Qatar* Romania Russian Federation Rwanda* Samoa* Saudi Arabia* Senegal* Seychelles* Sierra Leone* Singapore* Slovakia Slovenia Solomon Islands* South Africa

Sri Lanka Sudan Suriname* Swaziland Sweden Switzerland Syrian Arab Republic* Tajikistan Thailand Togo Trinidad and Tobago* Tunisia* Turkey Uganda Ukraine United Arab Emirates* United Kingdom Tanzania, TFYR United States Uruguay Uzbekistan* Venezuela Vietnam Yemen* Zambia Zimbabwe*

(*) Countries not included in the extended model.

Table 2 FDI Descriptive Statistics. FDI in US million dollars. Positive FDI Stocks Year

Observations

Average

Std. Dev.

Min

Max

2005 2006 2007 2008 2009

1,598 1,656 1,809 1,839 2,001

4.88 5.47 6.25 6.65 7.08

21.09 23.37 27.08 27.57 29.09

0.00 0.00 0.00 0.00 0.00

351.51 406.36 426.36 448.41 497.47

FDI Sample Composition Zero FDI stock Positive FDI stock Total

Observations

Percentage

Average

2,744 8,903 11,647

23.56 76.44 100

0 6.12 4.68

9

Journal of Comparative Economics xxx (xxxx) xxx–xxx

J.C. Brada et al.

Table 3 Corruption Statistics. Transparency International Corruption Perception Index (CPI). Home Country Corruption (10-CPI) Year Observations

Mean

Std. Dev.

Min

Max

2005 2006 2007 2008 2009 All

3.73 3.70 3.67 3.72 3.84 3.73

2.40 2.37 2.20 2.03 2.14 2.20

0.30 0.40 0.60 0.70 0.60 0.30

7.40 7.90 7.90 8.00 7.80 8.00

Host Country Corruption (10-CPI) Year Observations

Mean

Std. Dev.

Min

Max

2005 2006 2007 2008 2009 All

5.81 5.79 5.85 5.81 5.80 5.80

2.22 2.21 2.16 2.13 2.17 2.17

0.30 0.40 0.60 0.70 0.60 0.30

8.30 8.10 8.30 8.50 8.70 8.70

38 39 39 42 43 40

143 136 141 142 144 141

by Equation 8, above, as:

K ij* = c where:

ij

1 ij

= SMEij + CEEij , and

SMEij = 1 +

CEEij =

1

1

(10

(10

) Ci

Cj

Cj ).

We reparametrize our model to obtain identification. For the CEE effects, as explained in Section 2, the parameter ∅ measures the 1 effects of the corruption environment. We define the parameter CEE = to capture these effects. For the SME, the parameter 1 captures the effects of skill matching on the desired stock of capital, and δ captures the level of transferability of these skills across countries. It is impossible to disentangle these two effects because a direct measure that captures the level of transferability is not 1 ) .21 available. Thus, our empirical model combines both effects in the parameter SME = ( )(10 Our final specification is as follows:

FDIij = c

1+

CEE (10

1 Cj ) +

SME

Ci

Cj

+ Xij +

+

ij

(9)

where, FDIij, the dependent variable, is the observed positive FDI stock of home country i in host country j measured in billions of US $, and Xij are control variables described above. Our data is a panel, but we omit time subscripts to reduce notation.22 We expect that the coefficients c, γCEE and γSME will be positive and significant, confirming negative CEE and SME effects on capital stocks. We estimate our model using non-linear least squares with robust standard errors, and report the results in Table 4, column 1. The estimated coefficients c, γCEE and γSME are positive and significant. This confirms the prediction of our model that the corruption environment of the host country (CEE) and differences in corruption levels of home and host countries (SME) both affect FDI negatively. The parameter estimates for the control variables are also significant and in accord with theoretical expectations, and we further analyze these coefficients when we discuss the linear specification below. A disadvantage of the non-linear model is that the magnitude of the effect of CEE and SME on specific FDI stocks cannot be directly observed from the estimated parameters. Marginal effects have to be calculated for every host-country corruption level and for every home-host difference in corruption to obtain these economic effects, and we report the results of this exercise in Section 3.4. 3.3.2. Linear specification The advantage of the non-linear specification is that it links the theoretical results with the available data directly. However, it has 21 Our estimated coefficient can be also interpreted as the pure SME effect if = 9, which represents the lower bound for the SME effect. The SME ). effects assuming other values of δ can be calculated by dividing γSME by (10 22 Our estimation is based on pooled panel regressions. Below, we use the panel nature of our data to control for possible omitted variable bias due to home-country and year fixed effects.

10

Journal of Comparative Economics xxx (xxxx) xxx–xxx

J.C. Brada et al.

Table 4 Corruption environment and the skill-matching effects and FDI. We report estimation results for the CEE and SME FDIij = c

1 + CEE (10

1 Cj) + SME Ci

Cj

+ Xij +

+

ij .

on

FDI.

The

The OLS (2) model is FDIij =

non-linear + Xij +

model

CEE (10

in

Cj ) +

column SME |Ci

(1) Cj| +

is ij

as

follows:

and the PPML

model (3) is: FDIij = exp( + Xij + CEE (10 Cj ) + SME |Ci Cj|) ij . All control variables Xij are described in the text. Ci and Cj are the home and host country CPI indexes. FDI is measured in billion US$. t-statistics are in parenthesis. Standard errors are robust to heteroskedasticity. *, **, *** indicate significance at the 10%, 5%, and 1% levels. Variables

(1) Non-linear

γCEE

0.36*** (12.38) 0.42*** (3.18) 6.06*** (38.64) 5.54*** (32.88) −0.84*** (−15.45) 38.29*** (16.56) −0.22***

γSME GDPi GDPj Distanceij GDPijSimilarity

(GDPi

GDPj )2

Skillj Interaction1 Common Lang. Common Colony Bilateral Invest.Agg. Regional Trade Agg. Observations Adjusted R-squared

(2) OLS

(−17.58) −12.50*** (−6.47) −0.37*** (−14.65) 10.55*** (−14.39) 5.09*** (−5.31) −2.47*** (−6.14) 0.36*** −3.52

(3) PPML

6.26*** (5.45) 7.42*** (11.04) −0.81*** (−13.89) 30.85*** (13.14) −0.41***

−1.04*** (−8.11) −1.29*** (−13.53) 6.23*** (5.45) 7.23*** (10.87) −0.88*** (−14.00) 32.30*** (13.61) −0.39***

0.15*** (7.77) 0.22*** (87.66) −0.13*** (−70.11) 4.07*** (48.38) −0.00***

−0.23*** (−51.15) −0.04*** (−8.11) 0.14*** (7.42) 0.23*** (91.80) −0.13*** (−68.86) 4.20*** (49.44) −0.00***

(−8.30) −0.60 (−0.55) −0.47*** (−6.15) 12.09*** (8.91) 3.67** (1.97) −4.29*** (−15.33) 1.09** (2.39) 11,739 0.30

(−8.15) −16.18*** (−8.64) −0.47*** (−6.03) 11.15*** (8.16) 3.69** (2.00) −2.62*** (−10.18) −0.37 (−0.75) 11,647 0.31

(−13.76) 4.80*** (59.78) −0.01*** (−16.42) 0.81*** (68.11) 0.31*** (19.44) −0.70*** (−39.17) 0.31*** (19.76) 11,739 0.74

(−10.92) 0.66*** (6.55) −0.01*** (−18.06) 0.70*** (58.56) 0.27*** (16.47) −0.36*** (−18.65) 0.32*** (20.20) 11,647 0.75

limitations in dealing with econometric issues often observed in empirical work on FDI. For example, a common issue is the possibility of omitted variable bias and potential time trends. Incorporating fixed effects that control for these problems in a non-linear model is challenging due to convergence problems. However, they can be easily incorporated in a linear model. We estimate a linear specification of the model as follows:

FDIij =

+ Xij +

CEE (10

Cj ) +

SME

Ci

Cj +

(10)

ij

This differs from the non-linear model in that the corruption effects CEE and SME affect FDI linearly and in that we also control for year and home country fixed effects.23 The disadvantage of the linear model is that the estimates cannot be linked directly to the parameters of the theoretical model. Nevertheless, the implications of a negative relationship between CEE and SME and FDI stocks can still be tested using the linear specification by examining the significance and sign of the coefficients of γCEE and γSME. If our theoretical model is correct, we expect negative and significant signs for γCEE and γSME. Results using the OLS model including time and home country fixed effects are reported in columns 2 and 3 of Table 4. The parameter estimates for the benchmark gravity model are significant, in accord with theoretical expectations and very similar in magnitude to the parameters of the non-linear specification. Moreover, the addition of the corruption variables does not change the parameter estimates much, which suggests that the corruption variables do not serve as proxies for some of the traditional variables. The corruption environment effect (CEE) coefficients are negative and significant. This negative coefficient means that more corrupt countries have smaller FDI stocks as our theory predicts. The coefficient for the skill matching effect (SME) is also negative, meaning that the greater the difference in corruption levels between the home and host countries, the smaller bilateral FDI will be.24 The traditional gravity equation variables, GDPs and distance are of the expected sign and significant as are the additional covariates proposed by Blonigen and Piger (2014). The coefficient for host country factor endowment (Skillj) varies across specification as does the coefficient for the existence of a regional trade agreement involving the home and host counties. These are the only variables whose effect on FDI varies by specification. A result that may be surprising is that the coefficient for the existence of a 23

Notice that the total effect of host-country corruption on FDI still is non-linear in this linear specification. Recall that a positive coefficient for the CEE and SME effects in the non-linear specification carry the same implication as negative coefficients in the linear specification. 24

11

Journal of Comparative Economics xxx (xxxx) xxx–xxx

J.C. Brada et al.

bilateral investment treaty (BIT) is negative because the objective of such treaties is to increase bilateral FDI. However, most studies that find some positive effect for BITs use FDI from developed, capital-exporting countries to less developed capital-importing countries. In the case of our sample, the effect of BITs is measured using FDI from wealthy capital-exporting countries and the reverse flows from poorer countries to rich ones, and it also includes FDI among high-income capital-exporting countries and among capitalimporting countries. Jang (2011) shows that, for FDI among high-income countries, BITs do indeed have a negative effect on FDI and thus that positive effects for BITs are to be found largely in rich-to-poor FDI, which is only part of our sample. 3.3.3. PPML specification Bilateral FDI stocks exhibit skewness. A common practice to address this problem is to use a log-level regression, which involves using log(FDI) as a dependent variable. However, a log-level regression has an important drawback, because there are numerous country pairs for which the FDI stock is zero. A common practice is to retain these observations in the sample by adding a small constant to each FDI stock before taking logarithms, or alternatively, to eliminate these observations from the sample. Silva and Tenreyro (2006) show that both solutions severely bias estimation, and they propose using a Poisson pseudo-maximum-likelihood (PPML) estimate as an unbiased alternative. In the PPML approach, our gravity model with corruption can be expressed as follows:

FDIij = exp ( + Xij +

CEE (10

Cj ) +

Ci

SME

Cj )

ij

25

Here the coefficients can be interpreted as elasticities. Our model also includes year and home country dummies to control for fixed effects. Estimation results under the PPML are reported in columns 4 and 5 of Table 4. Results are qualitatively similar to the non-linear and linear specifications in terms of signs and significance levels. The CEE and SME effects remain significant and negative; however, the magnitude of the effect is larger for CEE. 3.4. The economic significance of corruption In this section we evaluate the economic significance of the skill-matching and corruption environment effects. We start with the non-linear model as given by Eq. 9. In order to understand the economic effects of CEE and SME on FDI, we need to calculate marginal effects. Given the non-linear nature of the model these effects are not constant as they would be in a linear model. Specifically, to measure the pure CEE, we start by assuming that there are no SME effects ( SME = 0 ), and thus the marginal effects from Eq. 9 can be calculated as follows:

FDIij Ci

=

c

CEE

1 1 + CEE (10

2

Cj )

(11)

Similarly, we calculate marginal effects for SME by assuming no CEE effects (

FDIij (Ci Cj )

=

c

SME

c

SME

(c (c

1 SME 1 + SME Ci

1 SME 1 + SME Ci

), 2

Cj

) , if 2

Cj

not defined, if Ci

if Ci Ci

CEE

= 0 ), as follows:

Cj > 0 Cj < 0

Cj = 0

(12)

Thus, for a home country with corruption index of 5 (Cj = 5) investing in a host country with corruption level 6 (Ci = 6), an increase in the host country's corruption index will have a negative effect on FDI. However, if the same home country is investing in a host country with corruption level 3 (Ci = 3), an increase in the host-country corruption index will increase FDI. In Table 5 we report the marginal effects calculated using Eqs. 11 and 12. The CEE effects change depending on the host-country corruption level. The largest effects are observed for countries with low corruption levels (high Cj= host country corruption perception index). For example, for a very honest country (10 Cj = 1), a unit increase in the level of corruption (i.e. 10 Cj = 2 ) results in a reduction of FDI stocks of US$6,318 million. The impact for highly corrupt countries is smaller. For example, for a corrupt country with 10 Cj = 8 a unit increase in corruption to 10 Cj = 9, will result in a reduction of US$767 million in FDI stocks. Since the average host country level of corruption (10 Cj ) in our sample if 5.80, then, for an average country, a unit increase in corruption, from 6 to 7 for example, will result in a reduction of US$1,158 million in its bilateral FDI stock. Since the average positive FDI stock in our sample is US$6,120 million, the average CEE effect will be a reduction of 18.9% in the FDI stock. The SME marginal effects differ depending on the difference in corruption levels between home and host countries. Thus, in Table 5, we present estimation results for all possible corruption differences (i.e., 0 to 9). For a home-host country pair with a difference in corruption of 1 CPI unit, an increase in this difference by one unit can reduce the bilateral FDI stock by as much as US $672 million. The SME marginal effects decrease as the difference in countries’ CPI increases, with an expected reduction on FDI stock of US$59 million for a home-host country pair with 9 CPI units of difference. The average corruption difference in our sample is 2.80, so for a country pair with average corruption differences, an increase of one unit of corruption implies a reduction of US$400 million, which represents a 6.53% reduction in FDI for the average home-host country pair. 1 ) captures the effects of skill matching, 1 , and also the level of transferability of these The regression coefficient SME = ( )(10 25

For example, the elasticity for the coefficient β is given by (exp( )

1) × 100 . 12

Journal of Comparative Economics xxx (xxxx) xxx–xxx

J.C. Brada et al.

Table 5 Economic Implications of SME and CEE. We report estimates of FDI stocks motivated by the corruption environment effect (CEE) and the skill matching effect (SME) as predicted by our non-lineal model in Table 4, column 1. For the CEE we report the average reductions in FDI stocks motivated by a unit increase in corruption (i.e. a unit increase in CPI). Results are reported for host countries with different levels of CPI. For the SME we report the average reductions in FDI stocks motivated by a unit increase in the difference in the level of corruption of home and host countries. Results are reported for home-home countries pairs with different levels of corruption differences. All numbers are in million US dollars. CEE Host Corruption 10

SME

Cj

1 2 3 4 5 6 7 8 9 10

FDI reduction

Corruption Differences |Cj

−6,318 −3,935 −2,684 −1,947 −1,476 −1,158 −932 −767 −642 −545

0 1 2 3 4 5 6 7 8 9

Ci |

FDI reduction 0 −672 −400 −265 −189 −141 −109 −87 −71 −59

skills across countries, δ. Thus, the estimated SME effects in Table 5 can be interpreted as the pure SME effects if = 9. This can be interpreted as the lower limit of the pure skill matching effect. For comparison, we also analyze the economic significance of corruption under the PPML specification. The coefficients on the PPML specification can be interpreted as elasticities. The magnitude of the CEE and SME effects in the PPML regression are very similar to the average effects in the non-linear model reported above. The reported PPML coefficients imply than an increase in the host country corruption level (CEE) decreases FDI stock by 20.6%.26 Similarly, a unit increase in the differences between home and host corruption levels, decreases FDI stock by 3.92%. To summarize, our most conservative estimates predict an average reduction of 19% (18.9%) in FDI stocks caused by an increase of one unit of corruption in the host country; and a 4% (3.92%) decrease in FDI for an increase in corruption differences between the home and host country. 4. Robustness checks 4.1. Omitted variables If important variables are omitted from our regression and are correlated with corruption levels, our estimation results will be biased. Our benchmark model has the advantage of including a set of variables capturing the most relevant determinants of FDI with data that is available for a large set of countries. However, it may be that it fails to include other variables proposed in the literature for which data are not available for the large set of countries analyzed in this study. In order to make our empirical exercise robust to the exclusion of such variables, we propose an extended model that includes additional determinants of FDI. These additional explanatory variables come at the cost of a smaller sample of countries. Data availability for our extended model reduces the sample to 39 home countries and 94 host countries (see Table 1) with 4,254 time-country observations. In addition to all variables included in the benchmark gravity model, the extended model includes the following variables: (1) home-country real GDP per capita (GDPperCap), (2) absolute difference between home- and host-country GDP per capita, (3) differences between home- and host-country skill levels, (4) host-country trade cost, (5) host-country political conditions, (6) interaction between home- and host-country political conditions, (7) home-country income inequality (Gini coefficient), (8) host-country Gini coefficient, (9) interaction between the home- country Gini coefficient and host-country political conditions (9) home-country level of ethno-fractionalization (10) interaction between host-country ethno-fractionalization and host-country political conditions and (11) home-country corruption level, (10 Ci ) . The inclusion of measures of GDP per capita is motivated by models that analyze the trade-off between serving host countries via exports or FDI and suggest that FDI is more likely to occur between countries with similar per capita income levels (Fajgelbaum et al., 2015). Also, several studies (Cuervo-Cazurra, 2008; Javorcik and Wei, 2009) use GDP per capita differences to capture differences in countries’ factor endowments instead of the HDI index we use in the benchmark model. Higher trade costs in the host country should encourage FDI since foreign firms will prefer to serve the market through affiliates rather than through foreign trade. We measure trade costs using Cost to export + Cost to import (both measured in deflated US$ per container) using data from the Doing Business project of the World Bank. The extended model also includes measures of countries’ institutional environments (other than corruption) and income 26

The effect can be calculated as (e

1) × 100 where β is the corresponding PPML coefficient. 13

Journal of Comparative Economics xxx (xxxx) xxx–xxx

J.C. Brada et al.

inequality. First, we follow Holburn and Zelner (2010) and include POLCON to capture policy risk or stability (Henisz, 2016). This variable takes values from 0 to 1, where 0 reflects the complete concentration of policy making authority.27 The idea is that MNCs invest less in host countries with greater policy risk, defined as the risk that a government will alter policies to expropriate an investing firm's profits or assets but that MNCs from risky environments will be less deterred from such investments. Second, Holburn and Zelner (2010) also argue that host-country policy risk may vary depending on the level of income inequality and the degree of fragmentation among resident ethnic groups, so we include the Gini coefficients from the World Bank's World Development Indicators to measure income inequality. Levels of ethno-fractionalization capture the probability that individuals in a country do not belong to the same ethnic group (Alberto et al. 2003).28 Thus, we include variables such as host- and home-countries’ POLCON (Pol Cond), Gini index and ethno-fractionalization index (Ethno) and interactions between home- and host-country political conditions and between the home-country Gini coefficient and host-country political conditions. However, due to the inclusion of home-country fixed effects, home-country POLCON and ethno-fractionalization are dropped from the regression. Finally, our theoretical model does not directly predict the effect of home country corruption on optimal capital stock, since the skill matching effect only measures differences in host- and home-country corruption levels. In order to disentangle skill-matching effects from home-country corruption effects, we include home-country corruption (10 Ci ) as a control variable. As in previous regressions we also control for year and home country fixed effects. We report estimation results in Table 6 for both the OLS and PPML specifications.29 All results are robust to this extended version of the model. In fact, the SME effect in the PPML model is even larger than in the case of the benchmark gravity model. We are also able to partially confirm the effects of the variables proposed in previous studies. For example, part of the Holburn, and Zelner (2010) results obtained for FDI in the power generating sector can be extended to all FDI. First, firms from home countries with weaker institutional constrains are less sensitive to host country policy risk. In our regressions, the positive coefficient of host political conditions Pol Condj implies that host countries with better political conditions attract more FDI, and the negative coefficient on the interaction Pol Condi × Pol Condj implies that this effect is smaller for home countries with poor political conditions. Similarly, these effects are smaller for home countries with greater redistributive pressures measured by the home country Gini coefficient, as the negative coefficient of Ginii × Pol Condj shows. However, unlike Holburn and Zelner (2010), we find a positive or non-significant effect for ethnic fractionalization. 4.2. Sample selection bias Next, we deal with the possibility of sample selection bias because our sample includes a number of observations with zero FDI stock. Suppose that the propensity of a home-country firm to invest in a given host country is determined by the variables in Eq. 10. If this propensity reaches a given threshold value, we will observe a positive bilateral FDI stock. However, if this propensity is low, we will observe zero FDI stocks. In order to deal with this potential problem, we use the Heckman (1979) selection model. We model the FDI decision as a two-stage process. In the first stage, the investor selects the host countries in which to invest. Then, in the second stage, she determines the amount to be invested. The first stage of the Heckman model includes a Probit regression, where the dependent variable is a dummy variable equal to 1 if country j receives FDI from country i and 0 otherwise and the independent variables are gravity variables and other determinants of the FDI decision. In the second stage, we estimate the effect of corruption on the volume of bilateral FDI stocks by controlling for sample selection. Specifically, we use a specification similar to Eq. 10 although we add a selectivity regressor obtained from the first stage, yielding the following specification:

FDIij =

+ Xij +

CEE (10

Cj ) +

SME

Ci

Cj +

Mills Millsij

+

ij

(14)

where FDIij, the dependent variable, is the observed positive FDI stock from home country i to host country j measured in millions of US$. We include the variable Millsij, the selectivity regressor that controls for possible sample selection in our data. The selectivity regressor corresponds to the Inverse Mills ratio of the fitted values of the first Probit stage. Estimation results for Eq. 14 are reported in the third column of Table 6. Once again, the coefficients for SME and CEE are negative and significant. Comparing the results with the OLS results (column 1, Table 6) reveals that the CEE is smaller and the SME larger in magnitude in the Heckman model. 4.3. Alternative measures of corruption Next, we analyze the effects of an alternative measure of corruption. We use two alternative corruption measures. The first alternative corruption measure is the political risk index from the International Country Risk Guide Methodology (ICRG) of the PRS group. This index is especially relevant for our study because it measures corruption within the political system, which is an important threat to foreign investors because it distorts the economic and financial environment, reduces the efficiency of government and business, and creates inherent instability in the political process.30 The index ranges from 0 to 6, where 6 represents countries with the lowest level of corruption, so we multiple the index by −1 in order to make it comparable with our CEE effect. Thus, a higher ICRG corruption index will represent a more corrupt country in our regressions. 27

Holburn and Zelner (2010) use 1-POLCON to measure political risk. The data were obtained from Romain Wacziarg's webpage: http://www.anderson.ucla.edu/faculty_pages/romain.wacziarg/papersum.html 29 The solution for the non-linear specification failed to converge, possibly due to the large number of variables in the specification. 30 From PRS Group Inc. definition of the variables in the ICRG rating system. Howell, (2011). 28

14

Journal of Comparative Economics xxx (xxxx) xxx–xxx

J.C. Brada et al.

Table 6 Extended Model and Selection Effects. We report estimation results of the CEE and SME on FDI. The OLS (2) and PPML (2) as in Table 4. The Heckman selection model is estimated based on FDIij = + Xij + CEE (10 Cj ) + SME |Ci Cj| + Mills Millsij + ij , Millsij, is the selectivity regressor that controls for possible sample selection. All control variables Xij are described in the text. FDI is measured in billion USS$. t-statistics are in parenthesis. Standard errors are robust to heteroskedasticity. *, **, *** indicate significance at the at the 10%, 5%, and 1% levels. Variables

OLS

PPML

Heckman

γCEE

−1.74*** (−7.51) −1.10*** (−3.49) 11.11** (2.01) 7.05*** (5.74) −1.11*** (−10.16) 43.53*** (8.97) −0.45***

−0.23*** (−34.77) −0.06*** (−8.48) 0.21*** (2.77) 0.21*** (34.31) −0.23*** (−39.90) 3.80*** (30.05) −0.00***

−1.43*** (−2.86) −1.25** (−2.39) 15.36** (2.11) 10.49*** (11.57) −2.43*** (−7.41) 70.39*** (7.82) −0.57***

(−3.93) −55.36 (−0.81) −0.89** (−2.46) 17.83*** (5.55) 0.42 (0.18) −2.37*** (−4.94) −2.55*** (−2.65) −0.01 (−0.15) −28.65 (−0.77) −44.17 (−0.65) 0.36*** (5.54) −11.43 (−1.40)

(−2.90) 0.77 (0.17) −0.01*** (−4.85) 0.66*** (31.05) 0.33*** (11.22) −0.38*** (−12.90) −0.21*** (−5.14) −0.01*** (−3.41) 2.97*** (5.04) −2.04 (−0.46) 0.08*** (16.06) 1.36*** (3.80)

(−6.92) −203.01 (−0.82) −0.81*** (−3.51) 18.31*** (5.60) 7.52** (2.29) 1.18 (0.75) −0.35 (−0.15) −0.10 (−0.40) −15.84 (−0.29) −216.54 (−0.87) 0.40 (1.60) –

γSME GDPi GDPj Distanceij GDPijSimilarity

GDPj )2

(GDPi Skill j

Interaction1 Common Lang. Common Colony. Bilateral Inv. Agg. Regional Trade Agg. GDPperCapi Dif GDPperCapij Dif Skilij Trade Costj Political Condj

Variables

OLS

PPML

Heckman

Pol Condi × Pol Condj

−1.67 (−0.27) 0.49 (1.29) 0.11*** (3.18) 0.27 (1.03) 7.75*** (4.67) 0.88 (0.09) −0.77 (−0.72)

−0.88** (−2.22) 0.02* (1.71) 0.05*** (24.59) −0.09*** (−7.35) 0.09** (2.08) 0.50 (1.63) −0.06 (−1.44)

4,254 0.34

4,254 0.73

1.09 (0.05) 4.95 (0.29) 0.01 (0.02) 0.20* (1.68) −0.20 (−0.27) 17.37*** (4.34) −31.75 (−1.48) 37.14*** (8.69) 4254 0.34

Ginii Ginij Ginii × Pol Condj ETFj ETFi × Pol Condj

(10

Ci )

Millsij Observations Adjusted R-squared

Second, we use the Control of Corruption measure from the Worldwide Governance Indicators (WGI) project of the World Bank. This index captures perceptions of the extent to which public power is exercised for private gain, including both petty and grand forms of corruption, as well as “capture” of the state by elites and private interests (Kaufmann et al., 2011). The index ranges from 15

Journal of Comparative Economics xxx (xxxx) xxx–xxx

J.C. Brada et al.

Table 7 Additional Robustness test. We report PPML model coefficients for the model in Table 4, including several additional control variables and robustness test as described in the text. t-statistics are in parenthesis and standard errors are robust to heteroskedasticity. *, **, *** indicate significance at the 10%, 5%, and 1% levels. Variables

ICRG Corruption

WGI Corruption

Round Trip

Few host

Interactions

Anti-Bribery

γCEE

−0.25*** (−26.59) −0.04*** (−3.97) 0.19** (2.54) 0.18*** (30.44) −0.24*** (−40.09) 3.51*** (28.18) −0.00*

−0.55*** (−34.88) −0.09*** (−5.81) 0.23*** (3.08) 0.21*** (35.00) −0.24*** (−40.26) 3.78*** (30.06) −0.00***

−0.24*** (−34.64) −0.07*** (−9.77) 0.21*** (2.78) 0.21*** (34.00) −0.23*** (−38.27) 4.04*** (31.68) −0.00

−0.23*** (−34.37) −0.05*** (−8.09) 0.22*** (2.92) 0.21*** (34.07) −0.23*** (−39.85) 3.76*** (29.51) −0.00***

−0.20*** (−19.76) −0.17*** (−9.11) 0.20*** (2.66) 0.21*** (33.89) −0.23*** (−38.47) 3.67*** (28.82) −0.00***

−0.20*** (−26.86) −0.06*** (−8.17) 0.21*** (2.77) 0.21*** (34.71) −0.23*** (−39.49) 3.88*** (30.66) −0.00***

(−1.91) 2.42 (0.55) −0.01*** (−5.51) 0.68*** (32.63) 0.48*** (16.47) −0.58*** (−19.51) −0.19*** (−4.65) −0.01*** (−3.46) 0.07 (0.12) −2.77 (−0.63) 0.05*** (9.51) 1.43*** (3.92)

(−3.64) 2.40 (0.54) −0.01*** (−5.08) 0.66*** (31.00) 0.34*** (11.67) −0.36*** (−12.03) −0.27*** (−6.33) −0.01*** (−3.51) 2.03*** (3.45) −0.14 (−0.03) 0.07*** (16.16) 1.38*** (3.84)

(−1.40) −0.33 (−0.07) −0.01*** (−5.16) 0.67*** (31.44) 0.37*** (12.52) −0.27*** (−9.09) −0.13*** (−3.11) −0.01*** (−3.29) 3.09*** (5.19) −3.26 (−0.73) 0.08*** (17.41) 3.02*** (8.07)

(−2.82) 8.02* (1.74) −0.01*** (−4.62) 0.66*** (30.62) 0.31*** (10.25) −0.38*** (−12.78) −0.23*** (−5.56) −0.01 (−1.64) 2.42*** (4.06) 5.10 (1.11) 0.08*** (15.66) 1.39*** (3.87)

(−2.89) 3.05 (0.69) −0.02*** (−6.73) 0.65*** (30.76) 0.32*** (10.76) −0.42*** (−13.88) −0.18*** (−3.95) −0.01*** (−3.76) 2.58*** (4.36) 0.32 (0.07) 0.08*** (15.60) 1.49*** (4.17)

(−2.83) 0.48 (0.11) −0.01*** (−3.08) 0.65*** (30.77) 0.33*** (11.29) −0.39*** (−13.24) −0.19*** (−4.61) −0.01*** (−3.40) 3.46*** (5.86) −2.40 (−0.54) 0.08*** (15.67) 1.26*** (3.49)

γSME GDPi GDPj Distanceij GDPijSimilarity

GDPj )2

(GDPi Skill j

Interaction1 Common Lang. Common Colony. Bilateral Invest. Agg. Regional Trade Agg. GDPperCapi Dif GDPperCapij Dif Skilij Trade Costj Political Condj

Variables

ICRG Corruption

WGI Corruption

Round Trip

Few host

Interactions

Pol Condi × Pol Condj

−1.21*** (−2.96) 0.02** (1.96) 0.05*** (23.70) −0.09*** (−7.63) 0.27*** (6.20) 0.57* (1.80) −0.03 (−0.76)

−0.83** (−2.10) 0.01 (1.14) 0.05*** (24.01) −0.09*** (−7.50) 0.16*** (3.64) 0.55* (1.76) 0.12 (0.76)

−0.17 (−0.44) 0.02** (2.05) 0.05*** (23.45) −0.14*** (−11.78) 0.15*** (3.42) 0.49 (1.58) −0.06 (−1.54)

−0.35 (−0.83) 0.00 (0.41) 0.05*** (24.63) −0.09*** (−7.33) 0.11** (2.47) 0.21 (0.65) 0.04 (0.99)

−0.99** (−2.51) 0.02* (1.72) 0.05*** (24.55) −0.09*** (−7.67) 0.10** (2.35) 0.43 (1.37) −0.16*** (−3.67) 0.00

Ginii Ginij Ginii × Pol Condj Ethnoj Ethnoi × Pol Condj

(10

Ci )

(10

Cj ) × |Ci

Cj|

(10

Ci ) × |Ci

Cj|

(0.10) 0.04*** (10.53)

γCEE × Anti Bribery Observations Adjusted R-squared

4,254 0.73

4,254 0.73

4,234 0.73

4,218 0.73

4254 0.73

−0.96** (−2.42) 0.02* (1.65) 0.05*** (24.13) −0.08*** (−6.98) 0.10** (2.25) 0.47 (1.51) −0.06 (−1.46)

−0.07*** (−8.28) 4254 0.73

−2.50 to 2.50, where 2.50 represents countries with the lowest level of corruption, so we also multiply this index by −1 to in order to make it comparable to our CEE effect. We report results under these two alternative corruption measures in Table 7. To save space we only report results from the PPML 16

Journal of Comparative Economics xxx (xxxx) xxx–xxx

J.C. Brada et al.

model. Our main results are robust to these two alternative measures of corruption. The magnitude of the CEE effect is larger in comparison to the SME effect, but both effects are very similar to our main results reported in Table 4. 4.4. Data anomalies We also consider some characteristics of FDI flows that may skew our results. First, we analyze the effects of roundtripping FDI. In roundtripping, residents of a country move funds offshore to a foreign country where shell companies or local banks then invest the funds back into the investors’ home country in the form of inward FDI. Investors are motivated to undertake roundtripping for various reasons, including money laundering, tax evasion and the desire to have their investments in their own country be more secure from expropriation or to receive better treatment from their own country's government because their acquisition of home-country assets is seen as the investments of foreigners (Aykut et al. 2017). As a result, certain small countries such as Cyprus or Hong Kong have extremely high levels of outward FDI, much of it attributed to roundtripping by investors from a few partner countries. Such reported FDI distorts both partners’ measures of FDI. To check that our results are not biased by roundtripping, we eliminate observations for the following host-home country pairs, which are seen as the major round-tripping actors: China-Hong Kong, Russia-Cyprus, RussiaSingapore, India-Mauritius, Brazil-Luxemburg, Brazil-Netherlands, and Ukraine-Cyprus. Our results are reported in the third column of Table 7, and they are robust to this control for roundtripping and similar in magnitude to our benchmark model. Next, we analyze if our results are driven by home countries with very few linked host country pairs. In our sample, the average number of host countries for the 43 home countries is 61. We drop from our sample all home countries with FDI in less than 20 host countries and estimate our extended model. Once more, the results reported in Table 7 column 4 are similar to our benchmark model. Finally, we explore the possibility that the impact of the skill matching effect changes depending of the corruption level of home and host countries. We test this possibility by introducing interaction terms to our extended model. Specifically, we include (10 Cj ) × |Ci Cj| = CEE × SME and (10 Ci) × |Ci Cj| = Home country corruptioni × SME . Note that the coefficients of these interaction terms measure whether the SME effect changes depending on the level of corruption in the host country (first interaction) or the level of corruption of the home country (second interaction). Our results are reported in Table 7 column 5. The coefficients for the CEE and SME effects remain negative and significant. The effects of SME are significantly larger than the results obtained in previous specifications. Only the interaction term between home-country corruption and the SME effects is significant and positive. This implies that the reduction of FDI motivated by the SME effect is larger for home countries with low corruption levels. For example, for a country like Finland with a low corruption level of 1.1 (i.e. corruption perception index = 8.9), for each additional point of corruption distance, the volume of FDI stock in a host country will decrease by −12.18%, compared with a decrease of −5.25% for a home country like Spain with a corruption level of 3. 4.5. Effects of the OECD Anti- Bribery Convention The OECD Convention on Combating Bribery of Foreign Public Officials in International Business Transactions (hereafter “the Convention” ) seeks to reduce corruption in developing countries by encouraging sanctions against bribery in international business transactions, including, but not limited to, FDI, carried out by companies based in the countries that have adopted the Convention. The Convention requires adherents to criminalize offering or giving bribes, but not of soliciting or receiving bribes. The effects of the Convention, as well as of the earlier United States Foreign Corrupt Practices Act (hereafter “the Act” ), which applies only to United States firms, on FDI have been investigated in the literature, but the studies reach conflicting results. Wei (2000) and Smarzynska and Wei (2000) found that the Act had no effect on US MNCs investments in corrupt countries. In contrast, Hines (1995) found the Act reduced US MNC's investments in corrupt countries, and Cuervo-Cazurra (2008) found that investors from the United States and from other countries that implemented the Convention reduced their investments in corrupt countries. Graham and Stroup (2015), examined US enforcement actions under the Act in cases where US multinationals were found to be bribing foreigners while seeking to undertake FDI transactions, and they found that, following such enforcement actions by the US government, the sanctioned firms did not undertake further investments in the foreign country where they had engaged in bribery. Nevertheless, signing the Convention does not guarantee that the signatory states are effectively enforcing the Convention. It is necessary for them to enact relevant legislation and then to provide the resources for investigating and punishing bribery. Transparency International (2012) reports that many of the signatories have either not enacted appropriate legislation or have undertaken no or very few enforcement actions. Their report on the enforcement of the OECD Anti-Bribery Convention as of 2012 argues that only 7 of the 37 signatory countries are actively enforcing anti-bribery legislation. Those are the US, Germany, the UK, Italy, Switzerland, Norway and Denmark. This gap between de facto and de jure application of the Convention may explain some of the contradictory results regarding its effects on FDI reported by others. In order to control for the potential effects of the Convention on signatories’ FDI in corrupt countries, we create a dummy variable that takes the value of 1 if the home country is one of the seven countries that actively enforce the Convention. Then, we include in our model an interaction term consisting of the level of corruption of the host country with this dummy variable (γCEE × Anti Bribery). The logic behind this variable is that the more corrupt the host country, the more likely it is that bribery takes place in FDI transactions, so effective enforcement of anti-bribery legislation should lead to a larger decline in FDI from countries that are parties to the Convention. The results presented in the last column of Table 7 show that the CEE and SME coefficients remain significant and of the same order of magnitude as the ones reported in Table 4. The interaction term is also significant and negative, which supports the studies mentioned above that find the Convention reduces FDI to corrupt countries. In unreported results we used a dummy variable that takes the value of 1 for all home countries signing the Convention before 2009. In this case, too, our estimates for the SME and 17

Journal of Comparative Economics xxx (xxxx) xxx–xxx

J.C. Brada et al.

CEE effects remain the same, but the interaction term for the effect of the Convention is not significant. This shows that signing the Convention is insufficient and that effective implementation is key, a point also raised by Cuervo-Cazurra (2008); Graham and Stroup (2015); Transparency International (2012) among others. 5. Conclusions In this paper we have developed and tested a theory of the influence of home- and host-country corruption on MNCs’ investment decisions. Our theory argues that corruption, or, more broadly, institutional quality has two effects on FDI decisions of MNCs. One effect works through the environment of the host country, which influences the costs to foreign MNCs of doing business in that environment. The second effect posits that corruption or institutional quality also shapes the abilities and skills of the firms to whom that environment is home. Thus, corruption determines the organization and skills of the firms in the home country, and these skills may be useful in a similar foreign environment. Our results show that host-country corruption has a negative effect on the volume and likelihood of receiving FDI and that these effects are, in economic terms, quite large. Our most conservative estimates show an average reduction of 19% in FDI stocks for one unit deterioration in the host-country corruption level. Corruption also influences the skills that a firm acquires in its home-country environment; firms from corrupt countries become adept at dealing with corruption and are thus able to transfer these skills to their affiliates in other corrupt countries, steering their FDI toward countries with similar levels of corruption. We estimate that average FDI stocks are reduced by at least 4% when the difference between home and host country corruption increases by one unit although the effect may be larger. Given the magnitude of these effects, it is clear that reducing corruption can be an important way for host countries to increase the FDI that they receive. Acknowledgement We thank the Journal's referees for their helpful comments and suggestions, which materially improved the paper. We are also grateful for comments received from participants during presentations of previous versions of this paper at the University of Perugia, Peking University, Southwest Jiatong University, the Asian Economic Community Forum and the Conference on Economic Chalenges in Enlarged Europe. Brada and Drabek acknowledge the financial support received from Czech Science Foundation Grant Number 1804630S. Perez acknowledges the financial support received from the Social Sciences and Humanaities Council of Canada's Grant Number 435-2017-0476 and and from the Arthur Wesley Downe Professorship of Finance. References Alberto, Alesina, Arnaud, Devleeschauwer, William, Easterly, Sergio, Kurlat, Romain, Wacziarg, 2003. Fractionalization. Journal of Economic Growth 8 (2), 155–194. Al-Sadig, Ali, 2009. The effects of corruption on FDI inflows. Cato J. 29 (2), 267–294. Armington, PaulS., 1969. A theory of demand for products distinguished by place of production. IMF Staff Pap. 16 (1), 159–178. Athanasouli, D., Goujard, Antoine, 2015. Corruption and management practices: firm level evidence. J. Comp. Econ. 43 (4), 1014–1034. Dilek, Aykut, Apurva, Sanghi, Gina, Kosmidou, 2017. What to Do When Foreign Direct Investment Is Not Direct or Foreign: FDI Round Tripping. In: World Bank Policy Research Working Paper 8046. Washington DC. Belloc, Marianna, 2006. Institutions and international trade: a reconsideration of comparative advantage. J. Econ. Surv. 20 (1), 3–27. Blonigen, BruceA., Piger, Jeremy, 2014. Determinants of foreign direct investment. Can. J. Econ. 47 (3), 775–812. Brada, JosefC., Drabek, Zdenek, Perez, M.Fabricio, 2012. The effect of home-country and host-country corruption on foreign direct investment. Rev. Dev. Econ. 16 (4), 640–663. Buckley, PeterJ., Clegg, JeremyL., Cross, AamR., Liu, X., Voss, Hinrich, Zheng, Ping, 2007. The determinants of Chinese outward foreign direct investment. J. Int. Bus. Stud. 38 (4), 499–518. Cezar, Rafael, Escobar, Octavio, 2016. Institutional distance and foreign direct investment. Rev. World Econ. 151 (4), 713–733. Cuervo-Cazurra, Alvaro, 2006. Who cares about corruption? J. Int. Bus. Stud. 37 (6), 807–822. Cuervo-Cazurra, Alvaro, 2008. The effectiveness of laws against bribery abroad. J. Int. Bus. Stud. 39 (4), 634–651. Cuervo-Cazurra, Alvaro, Genc, Mehmet, 2008. Transforming disadvantages into advantages: developing-country MNEs in the least developed countries. J. Int. Bus. Stud. 39 (6), 957–979. Drabek, Zdenek, Payne, Warren, 2002. The impact of transparency on foreign direct investment. J. Econ. Integr. 17 (4), 777–810. Dunning, JohnH., 1998. Location and the multinational enterprise: a neglected factor? J. Int. Bus. Stud. 29 (1), 45–66. Fajgelbaum, Pablo, Grossman, GeneM., Helpman, Elhanan, 2015. A Linder hypothesis for foreign direct investment. Rev. Econ. Stud. 82 (1), 83–121. Fujita, Masataka, 2008. A critical assessment of FDI data and policy implications. Transnatl. Corp. 17 (2), 107–123. Graham, Brad, Stroup, Caleb, 2015. Does anti-bribery enforcement deter foreign investment?. Appl. Econ. Lett. 23 (1), 63–67. Habib, Mohsin., Zurawicki, Leon, 2002. Corruption and foreign direct investment. J. Int. Bus. Stud. 33 (2), 291–307. Heckman, JamesJ., 1979. Sample selection bias as a specification error. Econometrica 47 (1), 153–161. Helpman, Elhanan, 1984. A simple theory of international trade with multinational corporations. J. Polit. Econ. 92 (3), 451–471. Henisz, WitoldJ., 2016. The dynamic capability of corporate diplomacy. Glob. Strateg. J. 6 (3), 183–196. Hines, James R. Jr. (1995). Forbidden payment: Foreign bribery and American business after 1977. NBER Working Paper No. 5266, Cambridge, MA. Holburn, GuyL.F., Zelner, BennetA., 2010. Political capabilities, policy risk, and international investment strategy: evidence from the global electric power generation industry. Strateg. Manag. J. 31 (12), 1290–1315. Howell, LlewellynD., 2011. International Country Risk Guide Methodology. PRS Group, East Syracuse, New York. Jang, YongJoon, 2011. The impact of bilateral free trade agreements on bilateral foreign direct investment among developed countries. World Econ. 34 (9), 1628–1651. Javorcik, BeataS., Wei, Shang-Jin, 2009. Corruption and cross-border investment in emerging markets: firm-level evidence. J. Internatl. Money Financ. 28 (4), 605–624. Johanson, Jan, Vahlne, Jan-Erik, 1977. The internationalization process of the firm: a model of knowledge development and increasing foreign market commitments. J. Int. Bus. Stud. 8 (1), 23–32. Johanson, Jan., Vahlne, Jan-Erik, 2009. The Uppsala internationalization process model revisited: from liability of foreignness to liability of outsidership. J. Int. Bus. Stud. 40 (9), 1411–1431. Kaufmann, Daniel, Kraay, Aart, Mastruzzi, Massimo, 2011. The worldwide governance indicators: methodology and analytical issues. Hague J. Rule Law 3 (2), 220–246.

18

Journal of Comparative Economics xxx (xxxx) xxx–xxx

J.C. Brada et al.

Kolstad, Ivar, Wiig, Arne, 2012. What determines Chinese outward FDI? J. World Bus. 47 (1), 26–34. Levchenko, AndreiA., 2007. Institutional quality and international trade. Rev. World Econ. 74 (3), 791–819. Markusen, JamesR., Maskus, KeithE., 2002. Discriminating among alternative theories of the multinational enterprise. Rev. Int. Econ. 10 (4), 694–707. Mironov, Maxim, 2015. Should one hire a corrupt CEO in a corrupt country? J. Financ. Econ. 117 (1), 29–42. Oliver, Christine., Holzinger, Ingo, 2008. The effectiveness of strategic political management: a dynamic capabilities framework. Acad. Manage. Rev. 33 (2), 496–520. Silva, J.M., Santos, C., Tenreyro, Silvana, 2006. The log of gravity. Rev. Econ. Stat. 88 (4), 641–658. Smarzynska, BeataK., Wei, Shang-Jin, 2002. Corruption and cross-border investment: Firm-level evidence. William Davidson Institute Working Paper No. 494. Ann Arbor, MI. Transparency International, 2012. Exporting Corruption? Country Enforcement of the OECD Anti-bribery Convention: Progress Report for 2012. Transparency International, Berlin Available at. https://www.transparency.org/whatwedo/publication/exporting_corruption_country_enforcement_of_the_oecd_anti_bribery_ conventio. Wei, Shang-Jin, 2000. How taxing is corruption on international investors? Rev. Econ. Stat. 82 (1), 1–11. Westermann-Behaylo, MichelleK., Rehbein, Kathleen, Fort, Timothy, 2015. Enhancing the concept of corporate diplomacy: encompassing political corporate social responsibility, international relations, and peace through commerce. Acad. Manag. Perspect. 29 (4), 387–404. Wu, Shih-Ying, 2006. Corruption and cross-border investment by multinational firms. J. Comp. Econ. 34 (4), 839–856. Yeung, HenryW., Liu, Weidong, 2008. Globalizing China: the rise of mainland firms in the global economy. Euroasian Geogr. Econ. 49 (1), 57–86.

19