An empirical study of the relationship between corruption and FDI: with sample selection error correction

Transcription

1 An empirical study of the relationship between corruption and FDI: with sample selection error correction Ying Zhou Economics Department University of Birmingham Edgbaston B15 2TT August 14, 2007 Abstract The aims of this paper will be threefold. First we will analysis the overall effect of non-discriminatory and discriminatory corruption on MNEs incentives to undertake FDI in a particular country. Then, we will try to determine the direction and magnitude of the effect of corruption on the level of FDI invested in the host country, after MNEs have decided to actually invest there. In particular, we show that the assumption of the independence between the MNEs s location and level of investment decisions that was made by the majority of the empirical studies in the literature might be incorrect. Therefore We take the location selection process of the MNEs into account when studying the relationship between corruption and the level of FDI in a host country by analysing these two decisions simultaneously using Heckman s model. 1 Introduction Corruption is not rare in international business operations, instead it is a frequent occurrence for international investors. A World Bank study (1999) revealed that more than 85 percent of polled multinational companies always or mostly encounter corruption while dealing with public sectors. Furthermore, according to TI s Bribe Payers Index (BPI) the subsidiaries of MNEs are the main suppliers of bribes all over the world (TI, 2006). International Organisations such as the OECD, World Bank, IMF and UN have 1

2 long acknowledged the fact that MNEs have played an important role in spreading corruption over the world through intense globalisation. 1 At the same time, it has been pointed out that MNEs are also part of the solution to control and fight corruption worldwide. 2 The grabbing-hand theory of corruption, supported by economists such as Shleifer and Vishny (1992,1993), Bliss and di Telia (1997) and Aidt (2003), claimed that corruption in an economy is like a grabbing hand that increases the costs of carrying out economic activities in the market. If we follow the reasoning of the grabbing-hand theory of corruption, it is seemingly uncontroversial to believe that corruption would increase the costs of foreign investors, just as it does to domestic investors, which would reduce the profitability of the investment projects and hence discourage FDI. This view is supported by surveys of international business. For instance, Kaufman (1997) found that the costs of investment in a relatively corrupt host country can be as much as 20 percent higher, compared with its uncorrupt counterpart. This show that corruption in the host country has been a major obstacle to international businesses and would discourage FDI from taking place. On the other hand, the helping-hand theory of corruption, supported by economists such as Lui (1985), Beck and Maher (1986) and Saha (2000), claimed that rather than obstacle for business, corruption could be an efficient lubrication for rigid economic regulation and red-tape. This is especially true for international business. By bribing the host government, the MNEs could get around the regulations or red-tape and potentially obtain a large amount of benefit from the host government in terms of profitable contracts, privileged access to markets or subsidies that cannot be obtained by exporting, which would act as an extra incentive for them to engage in FDI. For instance, as pointed out by Tanzi (1998), decisions to authorise major FDI project often provide MNEs with monopoly power in the host country, which would be extremely profitable for the investors. This provides a great incentive for MNEs to bribe host government officials, and in this case a corrupt host government would be preferred by the MNEs over an honest host government. The above discussion raises the interesting question of what the economic impact of corruption on MNEs FDI decisions could be. Despite more than a decades of empirical studies, the evidence on the sign of its effect is still unclear. One of the first empirical studies that investigates the possible correlation between FDI and corruption in a host country was made by Wheeler and Mody (1992). They studied U.S. manufacturing MNEs direct investment into 42 countries, between 1982 and Using panel data methods, it was shown that market size (measured by GDP per capita), a measure of agglomeration effects, such as current foreign investment, infrastructure and industrialization are significant determents of investment decisions, while corporate taxation and country specific factors, among which is corruption (called RISK in their study), are statistically insignificant. 1 See Tanzi (1998) and Svensson (2005). 2 For instance, the good practices to combat corruption suggested by the OECD, World Bank, UN and IMF normally involve encouragement for MNEs to resist bribe demands from host country officials, and severe penalties for MNEs that engage in corruption in the host countries. 2

3 Their results were contradicted by Wei s (2000a) study on the effects of corruption on international investors decisions. Using a cross-section of bilateral FDI stocks, from 12 source countries to 45 host countries, he found a significant negative effect of corruption on FDI investment. 3 He explains the differences between his and Wheeler and Mody s findings, as a result of different measurements. In Wei s study, instead of using aggregate measures with numbers of inter-related indicators, he adopted Business International s (BI) measure for corruption. Also, instead of using flows of FDI, Wei used stock data that might contribute to the difference in the result obtained. 4 By using average end of year bilateral investment stock data, for 14 source countries and 53 host countries, between 1994 and 1996, and corruption measures from the Global Competitive Report (GCP) and World Development Report (WDR), Wei (2000b) found that there is a significant negative impact of corruption on the level of FDI, even after controlling for the effect of net FDI incentives. 5 The same result regarding the effects of corruption on FDI was observed under various specifications and different measures of corruption. Drabek and Payne (2001) studied how a lack of transparency in a host country would affect FDI inflow, where transparency was measured by an index derived from The International Country Guild (ICG), with a higher number indicating higher transparency in the host country. 6 They pooled the aggregate bilateral FDI flow data from 52 countries between 1991 and 1995 together, and used Ordinary Least Square (OLS) and Two Stage Least Square (2SLS) estimations to show that lack of transparency is negatively correlated with the level of FDI inflows into a host country. A similar study was performed by Habib and Zurawick (2002). Again they pooled the aggregate bilateral FDI flows from 7 host countries into 89 host countries, between 1996 and 1998, and performed OLS estimations on the level of FDI inflows. 7 They reported significant negative correlation between corruption and FDI inflows, under a log linear specification, which indicates that a higher level of corruption would reduce the level of FDI inflows into a host country. The main problem with their and Drabek and Payne s (2001) analysis is that they used a simple pooled OLS regression without any concern for the possibility of host country fixed effect and serial correlation, or heteroscedasiticity of the error, which would render the analysis to be not very reliable. 3 In the paper Wei did not specify which year or years the data for FDI stocks were from. 4 The difference between their observations could also be a result of their different estimation methods, where by using panel data Wheeler and Mody (1992) have taken possible individual effects into account that also include corruption in the host country, while Wei (2000) did not. 5 Wei defines net FDI incentive as the difference between the FDI incentive provided by a host government and the restrictions, for details of calculation see Wei (2000b). This measure was included to control for the possible correlation between corruption and FDI incentives provided. Omission of this may cause biased estimation. 6 They define non-transparency as a set of government policies that increases the risk and uncertainty to international investors. It is caused by the presence of bribes and corruption, unstable economic policies, weak and poorly enforced property rights and inefficient government institutions. This measure of non-transparency is similar to the variable RISK that is used by Wheeler and Mody(1992). 7 They also performed Probit estimations on the relative under/over investment of the source into a particular host country. 3

4 Smarzynska and Wei (2002) used a unique cross-section firm-level data set on FDI in Eastern Europe and the former Soviet Union, between 1989 and 1995, including MNEs from both the U.S. and other countries and various measures of corruption in a host country from Country Risk Group (ICRG), GCR and Transparency International (TI), to study the effect of corruption on FDI at individual firm level. They found that corruption in the host country dramatically reduces the probability of inward FDI into the host country. What is more, if FDI takes place, the ownership of the firm will be shifted towards joint ventures, with a higher level of corruption in the host country. One of the most recent studies of the relationship between FDI and corruption was carried out by Egger and Winner (2005). Using the aggregate inward FDI stocks in 73 host countries and the corruption index from TI, between 1995 and 1999, they found corruption to be positively correlated with the level of FDI in the host country, both in the short-run and long-run. What is more, this positive relationship between FDI and corruption is stronger in short-run than in long-run, and is robust to changes in corruption measures. 8 However, in a different study, Egger and Winner (2006) used data from bilateral outward FDI stocks of 21 OECD countries into 59 host countries, between 1983 and 1999, and found that corruption is negatively correlated with FDI, when measured by the TI index. 9 This negative impact is more significant for FDI between OECD countries than between OECD and non-oecd countries and the negative effect of corruption has weakened over the years in their sample, while the effects of economic growth and changes in factor endowments have strengthened. 10 One of the possible reason for the mixed result could be that as addressed by Zhou (2007) that corruption in a host country can be distinguished into two different types according to its different purposes. They are called non-discriminatory and discriminatory corruption, which would affect the MNEs FDI decisions differently. Nondiscriminatory corruption means the government takes remuneration to do what it is supposed to do, such as provide a licence to a business, and discriminatory corruption means in exchange for a bribe the government provides the briber with a service that is not supposed to be provided. Using game theoretical models, Zhou (2007) show that 8 Two different measures were used, which are the indices provided by International Country Risk Guide and World Bank. 9 There are potential problems with the corruption measure used in this study. The authors pointed out that TI does not provide historical data pre-1995 at an annual base and they estimated the values for corruption when the data were missing, which could put the reliability of their analysis into question. What is more, as has been pointed out by Transparency International, the Corruption Perception Index (CPI) before 1995 only includes very limited countries and was composed from different sources and used different methods to these measures post-1995, which could put the consistency and comparability of the data over years into question. 10 The difference in these two studies could result from the different data and methods used. In their 2006 study, bilateral FDI data were used, while in their 2005 study aggregate FDI data were employed. The effect of corruption on aggregate and bilateral FDI might not be the same, which could partly explain the different results. Alternatively, the different results could result from the extensive use of the CPI data before 1995, where most of the data for corruption need to be filed by the authors, which puts the reliability of the results from their 2006 study into question. 4

5 non-discriminatory corruption discourage FDI, while discriminatory corruption could encourage FDI into a corrupt host country. This highlights the opposite effects of nondiscriminatory and discriminatory corruption, which have been largely neglected in both theoretical and empirical studies. These observations regarding the opposite effects of non-discriminatory and discriminatory corruption on FDI could provide us with some possible theoretical reasons for why both positive and negative correlations between corruption and FDI have been observed by various empirical studies. Even though the difference between these two types of corruption can be distinguished at a theoretical level, in reality, as observed by numerous studies such as Aidt (2003) and Svensson (2005), they tend to coexist with each other and are positively correlated. For instance, TI s Global Corruption Barometer 2004 notes that countries such as France, Greece, Italy, Portugal and Spain, with high levels of non-discriminatory corruption, are also reported to have high levels of discriminatory corruption practices. This makes the separation of them in practice quite difficult. What is more, nearly all measures of corruption that have been used for empirical studies, such as TI s Corruption Perception Index (CPI), World Bank s Government Indicator and corruption sores from Country Risk Group (ICRG) and Global Competitive Report (GCR), do not distinguish between these two types of corruption. This means that the results observed from any empirical studies that use these measures show the net impact of these two types of corruption. Therefore, depending on whether non-discriminatory or discriminatory corruption has dominating effect, it is not surprising to have mixed results regarding the impact of corruption on FDI. The aims of this study will be threefold. First we will test the hypothesis on the possible impact of the level of corruption in the host country on MNEs incentives for FDI, predicted by Zhou (2007). As mentioned in the above paragraphs, distinguishing between discriminatory and non-discriminatory corruption is important for our understanding of the potential underlying relationship between corruption and FDI decisions, however unfortunately in reality, due to the way corruption measures were constructed, it is impossible to separate these two components of corruption. As a result, it is not possible to test the individual effect of non-discriminatory and discriminatory corruption directly. Instead here we will test the overall effect of non-discriminatory and discriminatory corruption on MNEs incentives to undertake FDI in a particular country. This study will be one of the few studies that pays attention to the question of the possible effect of corruption on the probability of FDI taking place. 11 Hopefully our study will shed some light on this rarely studied phenomenon. Also, in this paper we will try to determine the direction and magnitude of the effect of corruption on the level of FDI invested in the host country, after MNEs have decided to actually invest there. It is worth noting that, even though empirical research on the possible effect of corruption on FDI has been undertaken for years, there still is not a uniform framework for studying this issue. For instance, some studies concentrated on the effect of corruption in the host country on the level of FDI inflows, such as Wheeler 11 This problem has only be studied by Smarzynska and Wei (2002) using firm level data. 5

6 and Mody (1992), Drabek and Payne (2001) and Habib and Zurawick (2002), while others focused on its effect on the level of FDI stocks, such as Wei (2000a, 2000b, 2001) and Egger and Winner (2005, 2006). It is well understood that because FDI stock is an accumulation of FDI inflows over the past, it has been viewed as the long-term decision of MNEs, while inflows of FDI have been viewed as the short-term decisions of MNEs (Hines, 1995 and Wei, 1997). Due to the difference in time-span, their reactions to the level of corruption in the host country might not be the same. As a result, to fully understand the effect of corruption on MNEs, both their short-term and long-term decision need to be analysed. Here we will study the effect of corruption on MNEs short-term decision (i.e. FDI inflows) first, before we analyse its effect on MNEs long-term decision (i.e. FDI stocks). All the earlier studies except these by Wei (2000a, 2000b, 2001) and Smarzynask and Wei (2002), have ignored the possible sample selection problem in studying the effect of corruption on FDI. 12 To do so, they effectively assumed that the chance of a country being selected to become a host country is randomly determined. If this were the case, explaining the effects of corruption on the level of FDI in the population by using only the data from countries with positive FDI would provide consistent results. However, in reality the countries that have positive FDI are not randomly selected, instead they have been selected by the MNEs based on the factors that could also affect the level of FDI invested. As will be shown in more detail in the next section, in this case the estimate of the population parameter would be biased when the selection process is ignored (see Heckman, 1979). In other words, the relationships between corruption and FDI that were estimated without taking into account the location selection process would be incorrect. As distinct from the majority of the previous studies in the literature, in this chapter we will model MNEs locations choice and their desired level of FDI in a host country simultaneously, to overcome the possible sample selection bias. 13 Our study will be different from the earlier studies on this issue by Wei (2000a, 2000b, 2001) in terms of estimation methods. In his study Type I Tobit estimations were employed, which does not provide a theoretical reason for why some data were censored. This makes Type I Tobit not an efficient way to study sample-selection problem. In our case, Heckman s (1979) method will be employed, which not only takes the probability of data being censored into ac- 12 The studies by Wei (2000a, 2000b, 2001) and Smarzynska and Wei (2002) would not suffer from this potential selection bias problem, as they have taken this particular issue into account by using Tobit (in Wei s studies) and Probit (in Smarzynask and Wei s study) estimations. They modeled the effects of corruption on investment in a host country simultaneously with the selection process of a country being selected to become a host country with positive FDI. 13 Sample selection error correction is very rare in international economics, because some believed that by using fixed-effect panel estimation, we would be able to correct the possible correlation between the selection processes for location and FDI inflows/stocks. But as pointed out by Verbeeck and Nijman (1992), fixed-effect panel estimation can only correct sample selection bias if the correlation between the error terms from the selection process for location and FDI inflows/stocks are correlated through timeinvariant unobservable variables. This means that if the error terms are not correlated in this particular fashion, the results observed from fixed-effect panel estimation will still be biased. As will be show later, in our data fixed-effect panel estimation cannot correct sample selection bias. 6

7 count, but also estimates the underlying relationship between corruption and the selection process. This should provide us with a more accurate picture of the determinants of FDI stocks in a particular host country. 14 Furthermore, unlike most of the studies in the literature such as Wei (2000a, 2000b), Smarzynska and Wei (2002), Habib and Zuzwicki(2002) and Drabek and Payne (2001), which used either cross-section or pooled panel data in estimation, here in our study panel data methods will be employed. 15 In their studies, by using cross-section or pooled data, they have greatly simplified the estimation process but at the expense of losing efficiency or even consistency. In our study of the effects of corruption on FDI levels, fixed/random unobserved individual and time effect panel data estimation will be adopted, in order to control for the possible unobserved source-host specific effect and the possible time specific effect resulting from shocks in the sample years to all host countries. Doing so would enable us to utilize the dynamic features of our data, thereby improving the accuracy of our estimation and providing us with a clearer picture of the complicated relationship between corruption and FDI. Furthermore, in recent years more attention has been paid by international organizations such as the World Bank and OECD to combating corruption around the world. For instance, the OECD Anti-Bribery Convention requires all 30 OECD members and 6 non-members who signed up for the convention to prevent, detect, prosecute and sanction bribery of foreign officials. Since it was implemented, there have been 30 convictions for bribery of foreign government officials in various parties to the convention. This would increase the cost of bribes for MNEs based in OECD countries, and therefore might alter their attitude towards corruption in the host country. However, according to the new report regarding the Anti-bribery Convention enforcement by Transparency International since its implementation in February 1999, there are still only a few international businesses and MNEs are fully aware of this convention, so as a result its implementation has had very limited effect on international businesses. In this study we shall be the first to test the impact of the OECD Anti-Bribery Convention on the effect of corruption for MNEs, empirically. The rest of the paper is organised as the follows. First, an introduction to the econometric methods that will be used to analyse the impact of corruption on FDI decisions (i.e. the decision of whether to invest in a certain country and how much to invest) and the data that is used in this study will be provided. Next, the results from our estimations will be examined and discussed, followed by a conclusion of all the main results from our study. 14 Smarzynaka and Wei (2002) used double probit heckman s model to study the probability of FDI into a particular country and the probability of fully owned enterprise through FDI in the host country, simultaneously. 15 There are only a few studies that employed proper panel data estimation methods, by Wheeler and Mody (1992) and Egger and Winner (2005,2006). 7

8 2 Methodology and Data In this section the econometric models that will be employed for our study of the impact of corruption on MNEs FDI decisions will be introduced, which will be followed by a brief discussion of the data used Estimating the Probability of FDI From the theoretical models of Zhou (2007), we know that the incentive for the foreign firm to invest in a country can be represented by its incentive function below: I = f(π F Π E ) (1) where I indicates the incentive for a foreign firm to engage in FDI and Π F and Π E indicate the levels of profits from FDI and exporting respectively. The incentive for a foreign firm to invest in a country is determined by the difference in its profits from FDI and its alternative option, namely exporting, given the market structure in the host country. From the equation above it is clear that we have assumed that the foreign firm views FDI and exporting as two alternative methods of supply, which are substitutes to each other. The incentive for FDI is increased when the profits from FDI increase relative to profits from its alternatives. In the case of one foreign firm, its incentive for FDI has an easy and clear interpretation. However, the incentive for FDI can be difficult to interpret when we look at aggregated data and deal with the decisions made by large numbers of foreign investors, as we are going to do here. In our case, the incentive for FDI (i.e. I) will be interpreted as a measure of the likelihood that the foreign investors from country j will invest in the host country i at time t. Here it is worth mentioning that MNEs normally make their investment decisions at time t based on information about past corruption and other control factors. Therefore, it is reasonable to assume that the probability of a source country j investing in a host country i at time t is a function of explanatory variables in the last period (i.e. t 1). 17 Therefore, the econometric model estimated can be written as follows: I i.jt = 1 if Ii.jt > 0 I i.jt = 0 otherwise (2) where Ii.jt = b 1 C it 1 + π 1 X i.jt 1 + v i.jt We further assume that the error term of the above equation satisfies the follow condition: 16 An illustration of the possible sample selection bias is provided in the appendix. 17 This is true for all variables except the measure for the effectiveness of the OECD Anti-Bribery Convention, which is assumed to affect FDI decisions from 1999 onwards, hence enters the estimation equation in levels and not in lags. 8

9 E(v i.jt X i.jt ) = E(v i.jt ) = 0 v i.jt N(0, σ 2 v) where I as before represents the likelihood of the foreign investors investing in the host country, which is an indicator variable, taking the value 1 if the latent variable Ii.jt is positive and taking the value of zero otherwise. The latent variable Ii.jt represents the difference between the potential profits from FDI and the alternative choice for investors from country j, who try to enter market i at time t. While the latent variable is not directly observable, whether FDI actually take place in a host country can be directly observed. 18 The latent variable also depends on factors other than corruption in the host country, which are taken as control variables and included in the matrix X i.jt 1. From the theoretical models in the previous chapters, we know that the control variables would include measures for the costs of production, the level of imports in the host country from the source country and host market size. What is more, the former two would be expected to be negatively correlated with the probability of FDI, while the latter would be positively correlated. In the above estimation equation, C it 1 represents the level of corruption in the host country i at time t 1, therefore the coefficient b 1 measures the effect that changes in the level of corruption in a host country would have on the probability of the foreign investor from country j investing in this country (country i), given that all other factors remain constant. From the theoretical model we know that Ii.jt would be negatively related to the level of non-discriminatory corruption and positively correlated to the presence of discriminatory corruption in the host country i, therefore the effect of C it 1 on I (i.e. the sign of b 1 ) will be determined by the type of corruption that has the dominant effect. In the above equation, v i.j is a combined error term that includes an unobserved time invariant heterogeneity κ i.j to reflect the possible characteristics between a particular source-host relation (i.j); an unobserved individual invariant heterogeneity ξ t to represent the possible yearly shocks within the sample years for all countries and e i.jt a random shock term, which represents the possible shocks for the source-host combination i.j at time t. From the above equations (2) and the error term distribution, we know that: P (I i.jt = 1 C i.jt 1, X i.jt 1, v i.jt ) = P [v i.jt (b 1 C it 1 + π 1 X i.jt 1 )] = Φ(b 1 C it 1 + π 1 X i.jt 1 ) The above equation (3) is the equation that is adopted for our estimation analysis of the effect of corruption on the probability of a certain source country j investing in a host country i at time t. 18 Which will be 1 if the level of FDI inflows/stocks between source-host combination i.j is positive, and will be 0 otherwise. 9 (3)

10 2.2 Estimation of the level of FDI Having completed the first stage of our estimation, which studies the probability of investing from source country j to host country i at time t, the next stage involves the estimation of the level of FDI that will be invested in the host country. In the theoretical literature about FDI, there are two main streams of theory regarding the motivations for FDI and the key factors that could determine the level of FDI in a host country. One is called the Vertical Integration theory of FDI, which assumes that MNEs break the production process into stages and allocate different stages to different locations in order to take advantage of the different factor endowments. Some of the best known studies in this spirit are carried out by Helpman (1984, 1985), and suggest that the differences in production costs such as labour costs could determine the level of FDI. The other stream is called the Horizontal Integration theory of FDI, developed by economists such as Horstmann and Markusen (1992) and Brainard (1993). These models aim to explain the high level of FDI between similar or even identical countries, and assume that the primary motivation for MNEs is to gain market access rather than to take advantage of differences in factor endowments. It has been predicted that in this case the host country market size and trade costs would be vital in determining the level of FDI. Here, we will follow the majority of the empirical studies in the literature and control for the factors that are suggested by both streams of theory. We will specify the level of FDI to be a function of the level of corruption and the measures of production costs, trade costs and host market size in the host country. 19 Therefore, the model we are going to estimate can be written as: logk i.jt = b 2 C it 1 + π 2 Z it 1 + u i.jt (4) We further assume that the error term of the above equation satisfies the condition below: E(u i.jt Z i.jt ) = E(u i.jt ) = 0 u i.jt N(0, σ 2 u) Where logk i.jt represents the log value of FDI invested by country j in country i at time t and Z i.jt 1 is a matrix containing the control variables that have been discussed above. u i.j is a composite error term including an unobserved factor θ i.j to represent the time invariant factors for the specific source-host combination (i.j) that could affect an MNE s decision regarding its optimal capital stock; ɛ t represents the individual invariant year specific effect and a i.jt is an idiosyncratic error term that represents the random shock for country j s investment into host country i at time t. As before, C it 1 indicates the level of corruption in the host country i and b 2 measures the overall effect of corruption on the level of FDI stocks holding all other factors constant, which is the main object of investigation in this study. 19 It is worth noting that, as has been mentioned earlier, the factors that might determine a country becoming a host country could also determine the level of FDI in the host country. 10

11 Similarly to the case of FDI location choice, here we will again assume that the two types of corruption affect the level of FDI in opposite directions. Discriminatory corruption implies that the briber would be granted favorable treatment such as market position/access from the host government in exchange for the bribe paid, it could increase the profitability of investment and therefore encourage the MNE to invest more in this host country. 20 If this effect dominates, then corruption would be positively correlated to the FDI level overall. On the other hand, non-discriminatory corruption means that an investor would not benefit in terms of market position or access by bribing the host government. More often, it is required merely to induce the government to provide the service it should provide for free. It can be seen as the everyday corruption that takes place at implementation of existing laws, rules and regulations, which is similar to petty corruption defined by U4 (Anti-corruption Resource Center) or passive bribe defined by the UN. Bribes are demanded in order to obtain basic access to services provided by the government. This type of corruption would increase the costs of the MNE, in turn reducing the return from its investment and discouraging the level of investment in the host country. 21 If this effect dominates then the overall effect of corruption would be negative. The above discussion shows that b 2 could take a positive or negative sign, depending on which type of corruption has a larger effect on MNEs FDI decisions. 2.3 Selection bias correction It is worth noting that the data for FDI stocks can only be observed for those host countries with positive FDI investment. In other words, we can observe the level of FDI if and only if the foreign investors choose to invest in that country and a particular sourcehost combination (ij) becomes active, i.e. I i.jt = 1. This means the sample used to estimate the econometric relationship between corruption and FDI capital using equation (4) is not random. Instead it is selected by equation (2), so the estimation errors of these two models might be correlated. If this is the case, the estimation for FDI capital without considering the location selection process might suffer from sample selection bias. 22 Hence the parameter which explains the impact of corruption on FDI that we estimated might not represent the population parameter. In the following analysis we will show that the assumption of the previous studies about the independence of the FDI level choice to the host country selection process can be rejected. This means that the relationship observed 20 Such corruption normally involves top government officials and large amount of bribe payments. 21 Zhou (2007) found that it is possible for the presence of non-discriminatory corruption to increase the profit of FDI by inducing exit of the domestic competitor in an industry. This is possible due to changes in industry structure. However, when the industry structure is constant, non-discriminatory corruption cannot increase profits from FDI. As we do not expect the overall industry structure of a host country to change from year to year, therefore it is rational to assume that non-discriminatory corruption would reduce the profitability of FDI at the aggregate level. Hence, in our aggregate data, the coefficient for corruption would be negative if the effect of non-discriminatory corruption dominates. 22 Please refer to the appendix for a more detailed discussion of how sample selection bias can arise and the consequences of it. 11

12 between corruption and FDI might be incorrect. To correct for the sample selection problem, the method developed by Heckman (1979) will be adopted. We will estimate the incentive and FDI level decisions of foreign firms simultaneously. The econometric model used here can be written as: I i.jt = 1 if I ijt > 0 I i.jt = 0 otherwise where I i.jt = b 1 C it 1 + π 1 X i.jt 1 + v i.jt logk i.jt = b 2 C it 1 + π 2 Z it 1 + θ i.j + u i.jt if I = 1 where all the variables are as defined above. 2.4 The Data Some of the previous studies on this issue used FDI inflows as their dependent variable, including Wheeler and Mody (1992), Drabek and Payne (2001) and Habib and Zurawick (2002), while others used the level of FDI stocks as their dependent variable, including Wei (2000a, 2000b, 20001) and Egger and Winner (2005, 2006). In this study, to understand the effect of corruption on MNEs short-term and long-term FDI choices more fully, both FDI inflows and stock will be used in turn as the dependent variable in the regressions. FDI data is drawn from a sample of 20 OECD source countries and 52 host countries over the period from 1996 to 2003, including both developed and developing countries all over the world. 23 We measure the dependent variable logk i.jt the capital inflow (and stocks) invested from country j to country i at time t, by the total bilateral FDI inflows (and stocks) at 1999 US dollar price measured using Purchase Power Parity (PPP). 24 This is observed from OECD International Direct Investment statistics. The likelihood that investors in country j would invest in country i at time t (i.e. I i.jt ) is measured by an indicator variable that takes the value of 1 if the level of bilateral FDI inflows (and stock) is positive and takes the value of 0 otherwise. The main explanatory variable we are interested in, both in the incentive and capital stock regressions is the level of corruption in host country i at time t 1, C it 1, which is measured by TI s CPI and the World Bank s Governance Indicator for Control of Corruption. Both measures are averages of results form surveys on corruption. They are chosen as the measures for corruption here partly because by including a number of surveys, they can reduce measurement error, if all surveys are independent; and also partly because both are widely available to the public. TI s CPI ranges from 0 to 10 and the World Bank s Governance Indicator ranges from -2.5 to 2.5, with a lower grade value represents a more corrupt country in each case, while a higher grade represents a cleaner host government. For explanatory simplicity, here both corruption measures are multiplied by 1, so in our case a larger corruption score 23 A full list of countries used is provided in the appendix. 24 It is worth noting that here the FDI data is not deflated as there is not any reliable bilateral deflator available. 12 (5)

13 represents a more corrupt host country. This means that if non-discriminatory corruption has a dominant effect on MNEs investment choices, the overall effect of corruption on the incentive for FDI and the level of FDI inflows/stocks would be negative and the coefficients for C it 1 (b 1 and b 2 ) should be negative as a result. What is more, the effect of the OECD Anti-Bribery Convention on MNEs attitudes toward corruption in the host country can be tested by exploiting the time-dimension of our data. This convention was implemented in February 1999, which means that we would expect MNEs to become more averse to corruption attempted in the host country after the year The effectiveness of this convention can be tested by testing the significance of the coefficient for the interaction term constructed by multiplying the indicator variable for post 1999 (this indicator equals 1 if the data was observed post 1999, and equals 0 otherwise) with the measure of corruption. The coefficient of this interaction variable would be negative if the implementation of this convention makes the negative effect of corruption stronger after 1999, and would be positive otherwise. The primary control variables in the regression for the incentive for FDI include the host country s market size measured by the host country s log-gdp at 1999 US dollar deflated by GDP deflator, the growth rates of total GDP and fixed capital and the log of the country s total population, observed from World Bank s World Development Indicators. The measures for production costs of the foreign firm are the real average wage rate in the manufacturing section in each country observed from OECD Labour Force Statistics and the maximum marginal corporation tax rate observed from the World Tax Database. The incentive for exporting form the source country into the potential host country is proximated by a ratio of source-host country bilateral imports to total import of the host country. A higher value of this ratio would be taken as evidence that it is relatively easy for the source country to export to the host country, while a lower ratio value would mean the opposite. The bilateral import data are observed from the OECD trade database and total imports are derived from the World Bank s World Development Indicators. The openness of the country is taken from the Penn World index, which is used as a proxy for the trade restrictions of the host country. This is expected to be negatively correlated with the probability of FDI, as a lower level of trade restriction would increase the profit from exporting, which in turn would reduce the relative attractiveness of FDI as the MNEs mode of entry into a host country. In this study we also control for the effect of the host country s attitude towards FDI on MNEs investment decisions. This is approximated by an index called net restrictions observed from Wei (2000b). The index for net restriction ranges from -4 to 4, where a more positive number indicates lower net restrictions in the host country. 25 The potential effect of the host country s political system on the probability of FDI is also takes into consideration. This is done by including the indices of overall democracy in the host country and the durability of the host government s regime, obtained from the Polity IV project. The measure for democracy, called Polity2, ranges from -10 to 10, where a higher number indicates a more democratic host country, and a lower number 25 It is worth noting that in Wei s paper this is called net incentive. 13

14 implies a more autocratic host country. It is expected to be positively correlated with the probability of FDI, as more democratic countries tend to provide more protection to private property and investment from exploitation by the host government. The durability of a host government s regime is measured by the number of years since the most recent regime change, which is included to measure the stability of the host government and is expected to be positively correlated with the probability of FDI into the host country. 26 The main control variables for the FDI inflow/stock regressions include real GDP, total population, growth rate of total GDP and fixed capital of the host country, maximum marginal corporate tax, the average manufacturing wage, openness of the host country, restrictions on and incentives for FDI, democracy of the country and the durability of the host country s regime, which are as explained before. These variables are used in order to control for the costs of labour, market sizes, export potential, trade costs and technology in the host country suggested by the Vertical Integration theory of FDI. Here we will use the percentage of exports in total real GDP to measure the potential for exporting, which is an important factor in the Horizontal Integration theory of FDI, is observed again from the World Bank s World Development Indicators. Table 1 summarises the key variables in our study, which are the probability of FDI from country i to country j takes place, the level of FDI inflows/stocks and two different measures of corruption in the host country. Table 1: Summary Statistics M ean Std.Dev. M in. M ax. no.obs. log Inflow log Stock Prob. Inflow Prob. Stock TI s CPI Control of Corruption Estimation Results In this section the results from this study are reported and discussed in relation to the observations from the previous studies in the literature. We will first study the effect of corruption on FDI inflows, using TI s CPI as the measure for corruption in the host government. We shall use the World Bank s Governance Indicator of Control for Corruption to check the robustness of our estimation regarding changes in corruption measurements. Later, the same analysis will be carried out for FDI stocks in the host country. This should provide us with insights about the possible impact of corruption on both MNEs short-term (FDI inflows) and long-term (FDI stocks) investment choices. 26 Please see the Polity IV data manual for detailed definition and explanation. 14

15 3.1 Results from the FDI probability estimation First we study the effect of corruption on the probability of a country becoming a host country for FDI inflows, in a particular year. To this end, as a base line model, we pooled all the data together and performed pooled Probit estimation on our data. The results are show in Table 2, which reveals that the probability of a country becoming a host country for FDI inflows can be negatively affected by the level of corruption in the country, at the 1% significance level. In order to utilise the time-dimension of our data and control for possible individual specific effects and time specific effects, panel (random effect) probit estimations were performed. The results are provided in Tables 3, which show that the coefficients for corruption are statistically significantly negative at the 5% level, in all specifications. This means there is strong empirical evidence to suggest that the negative effect of nondiscriminatory corruption dominates the overall effect of corruption on the probability of a country becoming a host country for FDI inflows, in our sample. The results also reveal that the marginal effect of corruption on the probability of FDI inflows taking place in a country ranges from 5.7% to 12.2%. In other words, holding all other factors constant, on average, if corruption in a country worsened from the level of France (6.9 in 2003) to that of Italy (5.3 in 2003), the probability of FDI inflows taking place in the country would be reduced by about 9.12 to percent, under various specifications. The above discussion implies that the negative impact of corruption is an important determinant on the probability of FDI inflows taking place in a country, both in terms of statistical significance and in terms of magnitude. 27 The OECD Anti-bribery Convention, which criminalised any OECD MNEs bribing foreign governments, was implemented in February This would have made corruption more costly for MNEs, and therefore we expect the level of corruption in a country to have a stronger negative effect on the probability of FDI inflows taking place after the year In other words, as discussed previously, after the implementation of the convention we would expect MNEs to be more averse to the level of corruption in the host country, hence the coefficient for the Anti-Bribery convention should be negative. The regression results in Tables 2-3 reveal just the opposite of what was expected. They show that the coefficient for the anti-bribery measure is significantly positive, which indicates that since the implementation of OECD Anti-Bribery Convention, the OECD MNEs have became much less averse to the level of corruption in the host country, rather than more averse. This provides empirical support to TI s claim of the limited usefulness of the convention in altering the investment choices of MNEs into corrupt host countries. 28 The possible reason for this counter-intuitive observation might be that since 1999, there has been a slight fall in the average corruption level worldwide, therefore when making their FDI inflow decisions MNEs might worry about corruption less than they did 27 This observation is similar to Smarzynaka and Wei s (2002), where they found the corruption level in a country has a negative and significant effect on the probability of FDI stocks taking place in that country. 28 This observation is similar to Egger and Winner (2006), where they found the negative effect of corruption has weakened over time, between year 1983 and

16 before. As a result, MNEs have been less averse to corruption in the host country since Alternatively, the results above might occur because since the implementation of the convention in 1999, it has become harder for the host government to demand corruption under the bright day light, therefore corruption in the form of non-discriminatory corruption falls. 29 On the other hand, in the case of discriminatory corruption, both bribee and briber would gain from the corrupt deal, which makes it much easier to keep it secret than non-discriminatory corruption. As a result, bribes in the form of discriminatory corruption would not be reduced as significantly as non-discriminatory corruption after the implementation of the Anti-Bribery convention. This indicates that after the implementation of the Anti-Bribery convention in 1999, the proportion of corruption that is of the discriminatory form would be higher than before. Because it was predicted by the theocratical model that discriminatory corruption has a positive impact on MNEs incentive to conduct FDI, a higher proportion of discriminatory corruption in total corrupt deals would mean a larger positive effect of corruption on the probability of FDI inflows taking place, hence the coefficient for the Anti-Bribery measure would be positive. Furthermore, as suggested by economic theories, the larger the host country market size, the more likely the foreign investor will supply its market through FDI. The results in Tables 2-3 show that the host country s market size, measured by total population and total GDP, has a positive impact on the probability of FDI inflows taking place at the 5% significance level, which are in line with observations from earlier studies in the literature. The estimation results also provide empirical support for the assumption made in the majority of the international economics theory (including mine) that MNEs view FDI and exporting as substitutes for each other, therefore the more open a host country is, the less would be the trade costs involved and hence the less likely would MNEs be to invest in that country. What is more, the results show that in our data, it is more likely for MNEs to invest in a country that has already established trade links with their home country. In particular, the higher the volume of trade between the host and source country relative to the rest of the world, the more likely FDI inflows are to take place. Despite the importance of this effect in terms of its statistical significance at the 1% level, it is small in terms of magnitude. Tables 2-3 show that the other traditional economic factors, such as the potential for host market expansion, measured by the growth rate of total GDP and fixed capital, the production costs of MNEs in the host country measured by corporate tax and labour costs are not significant factors in determining the probability that FDI inflows taking place between a particular source-host pair. The non-economic factors, such as the host government s attitude towards FDI, measured by the net restriction of FDI, and the political system in the host country, measured by democracy of the regime, are also insignificant factors in determining the probability that FDI inflows taking place in the country. However, there is some evidence to suggest that the stability of the host country s political system, measured by the durability of the host government regime, can positively affect the probability of FDI inflows taking place. 29 Non-discriminatory corruption means the briber does not get direct benefit in return for the bribe paid. As a result, non-discriminatory corruption is costly for any briber. 16