WPS4394 Policy ReseaRch WoRking PaPeR 4394 The Incidence of Graft on Developing-Country Firms Alvaro González J. Ernesto López-Córdova Elio E. Valladares The World Bank Financial and Private Sector Development Vice Presidency Enterprise Analysis Unit November 2007 Policy ReseaRch WoRking PaPeR 4394 Abstract This paper measures the extent to which firms in as likely to be asked for bribes as are firms in Latin developing countries are the target of bribes. Using America, although there is substantial variation within new firm-level survey data from 33 African and Latin each region. Last, we show that graft appears to be American countries, we first show that perceptions more prevalent in countries with excessive regulation adjust slowly to firms' experience with corrupt officials and where democracy is weak. In particular, our results and hence are an imperfect proxy for the true incidence suggest that the incidence of graft in Africa would fall by of graft. We then construct an experience-based index approximately 85 percent if countries in the region had that reflects the probability that a firm will be asked for levels of democracy and regulation similar to those that a bribe in order to complete a specified set of business exist in Latin America. transactions. On average, African firms are three times This paper--a product of the Enterprise Analysis Unit, Financial and Private Sector Development Vice Presidency--is part of a larger effort to study and promote reforms in the business environment. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The corresponding author may be contacted at jlopezcordova@ifc.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team The Incidence of Graft on Developing-Country Firms Alvaro González, J. Ernesto López-Córdova, and Elio E. Valladares1 Keywords: Corruption; graft; regulation; firms; democracy JEL Classification: D21, D73, H11, K42, L20, L50 1 Financial and Private Sector Development Vice-Presidency, World Bank Group. Email: agonzalez4@worldbank.org; jlopezcordova@ifc.org; and evalladares@worldbank.org. 1 Introduction Corruption is a serious burden for firms in the developing world. In 2006, two out of every five firms in Africa and Latin America reported that unofficial payments were required "to get things done," and one in six said they were expected to present informal gifts when meeting with tax inspectors. On average, informal gifts or payments "to get things done" were equivalent to 2.1 percent of firm sales, which, taken at face value, would not appear to be excessive, especially in comparison with applicable tax rates around the world. Nevertheless, the uncertainty and illegal nature associated with corruption makes it more burdensome on firms than official taxation (Shleifer and Vishny 1993; Fisman and Svensson 2007). More worrisome is that the incidence of bribes is higher precisely in the poorest countries, where development needs are most pressing. For example, whereas 9 percent of firms in Chile believe informal gifts are required to "get things done," 87 percent of firms in Burkina Faso are of that view. Similarly, two out of every three firms in Cameroon and the Democratic Republic of Congo state that they must pay bribes when meeting with tax officials. Finally, firms in Africa report having to pay higher bribes, as a percent of sales, than their counterparts in relatively-affluent Latin America, 2.7 vs. 1.4 percent, respectively.2 Since growth is unlikely without a vibrant private sector, measuring and understanding how corruption affects firms is an important research area. However, efforts in that direction are thwarted by the lack of reliable information about the incidence of corruption. By its very nature, it is difficult to come by objective data on the pervasiveness of graft. Work on the subject often relies on perceptions on the extent of corruption, but there is evidence that perceptions are a poor reflection of the prevalence of corrupt practices (Olken 2007; Weber 2007).3 In addition, existing cross-country measures of corruption are often based on surveys of a limited number of experts, a non-representative sample of firms (e.g., multinational corporations), or households, and hence may not necessarily reflect the experience of the average enterprise. In this paper we exploit a novel dataset of nearly 10,000 firms in 33 countries in Latin America and Africa to compute objective measures of the incidence of corruption.4 The data come from the 2 All figures come from the World Bank's Enterprise Surveys and are available at http://www.enterprisesurveys.org . 3 Even the figures in the opening paragraph can be criticized for reflecting firms' views on how widespread corruption and not necessarily their own experience. 4 Our sample consists of firms in 18 African and 15 Latin American countries. In Africa they include: Angola, Botswana, Burkina Faso, Burundi, Cameroon, Cape Verde, Democratic Republic of Congo, Gambia, Guinea, Guinea-Bissau, Malawi, Mauritania, Namibia, Niger, Rwanda, Swaziland, Tanzania, and Uganda. In Latin America they are: Argentina, Bolivia, Chile, Colombia, Ecuador, El Salvador, Guatemala, Honduras, Mexico, Nicaragua, Panama, Paraguay, Peru, Uruguay, and Venezuela. 2 Enterprise Surveys collected by the World Bank that cover business conditions in most major economies across the globe. These surveys capture firm perceptions about the quality of the business environment, as well as objective information on firm characteristics and the problems firms must deal with when transacting with the public or private sector. These problems include delays or difficulties in gaining access to electricity or credit, the extent of obligations from complex taxes, and frequent inspections, among others. The Enterprise Surveys also contain information on firms' perceptions about the problems that corruption poses for their performance, as well as records on whether firms were asked to make "an informal gift or payment" when requesting access to basic infrastructure services or government permits.5 The latter form the basis of the analysis in this paper. We first matched data on perceptions and on instances of bribery and show that firms that solicit services or licenses and are not asked for bribes hold a more optimistic view about the effect of corruption on firm performance, relative to both firms that are victims of extortion and those that did not request services and hence are beyond the reach of corrupt officials. We take this as evidence suggesting that perception-based measures of corruption are an imperfect proxy for the true incidence of graft. We then use the Enterprise Survey data to construct a Graft Index of Firm Transactions (GIFT). The index reflects the probability that a firm will be asked for an informal gift or bribe when requesting access to infrastructure services or permits. The proposed index has several advantages over alternative measures of corruption. Most notably, the index relies on "hard" data -- firms' encounters with corrupt practices -- and not on managers' or experts' perceptions about the extent of corruption in a country. Another advantage is that our primary data come from nationally representative surveys and hence capture the experience of the typical firm's dealing with dishonest government officials. The fact that we focus on a common set of transactions guarantees that our results are comparable across countries. Admittedly, the index is based on a narrow definition of graft that focuses on petty bribes and we do not directly account for several other forms of corruption (e.g., in the procurement of government contracts) that could potentially have a bigger effect on firm performance. The index strongly indicates that firms in Africa are particularly vulnerable to graft. Entrepreneurs in the region had on average a 19 percent chance of being asked for bribes; among the comparatively wealthier Latin American countries, the figure was only a third as high (7 percent). Within each of the two regions, and even among neighboring countries, there is substantial variation in the incidence of bribery, suggesting that corruption is not necessarily explained by cultural or historical factors. The index also shows that bribery is more common when requesting licenses and permits than when soliciting infrastructure services. 5 Specifically, access to electricity, water, telephone services, construction permits, operating licenses, and import licenses. 3 Finally, we use the index to shed light on some of the factors that lie behind the incidence of graft. In particular, we study whether excessive regulation is associated with corruption. For that we use data from the World Bank's Doing Business project.The latter ranks countries according to the extent and nature of the regulatory and legal obligations that firms have to meet to be able to operate in an economy; more obligations translate into a lower rank.6 We also consider whether democratic governments, more accountable to their citizenry, do a better job in containing petty bribery. We find that both excessive regulation and weak democracies increase the likelihood that firms will be the target of bribes. Our quantitative results imply that differences in the incidence of graft in Africa and Latin America would disappear if the former had levels of democracy and regulation similar to those that exist in the Western Hemisphere. The rest of the paper is organized as follows. In Section 2, we address the relationship between experience with corruption and the perception of corruption. We show that perceptions adjust only gradually to changes in the true extent of graft. In Section 3, we explain how we calculate the Graft Index of Firm Transactions, apply the methodology to our sample of countries, and present our estimates. In Section 4, we look at some of the correlates of the index at the country level. Although the data do not allow us to identify the causes behind graft, findings here regarding the link between regulation, democracy, and corruption are in line with those suggested in the literature and explored by other authors. Section 5 concludes. 2 Do perceptions of corruption match incidences of graft? In this section, we discuss the relationship between perceptions and incidences of corruption. We begin by discussing the difficulty with measuring the extent of corruption using perceptions data --which are at present almost exclusively what is used to measure corruption. By comparing firm-level data on the incidences of bribes with firm-level data on perceptions of corruption, we show that perceptions and incidences are imperfectly matched but likely to be updated depending on recent experience. With respect to updating of corruption perceptions, the data show that when firms have a positive experience with honest, uncorrupt officials, these same firms are prone to have more positive perceptions about the extent of corruption than firms that had no record of transactions with officials or those that had transactions and were asked for bribes to complete them. 6 See www.doingbusiness.org. 4 2.1 Why is corruption hard to measure and compare across countries? In order to assess the nature and extent of corruption, economists prefer transaction data --who businesses bribe, how much they bribe, how often they bribe, and what is gotten in return for those bribes. Given the illegal nature of these activities, such data are difficult to get. In their absence, many studies have had to settle for perceptions and opinions about corruption to determine its prevalence and nature (Lambsdorff 2006; Kaufmann et al 2007). While perceptions data are easier to get than direct reports of corrupt deals, using perceptions is problematic. First, it is difficult to pin down how perceptions are formed because corruption refers to activities that are the hidden and largely unobservable. Therefore, perceptions about corruption are likely formed by what people believe to be generally taking place and less so on what is personally experienced. In any one city or country, what people believe to be taking place may start to converge since people read the same newspapers, are exposed to the same political rhetoric and may hear others' opinions about corruption (Cábelková and Hanousek 2004; Moyal et al. 2004). In the end, people may end up repeating those personally unverified opinions until they become well-ingrained folklore (Andvig 2004). Second, it is not clear what type of corruption one is measuring when constructing a corruption index (Knack 2002, Weber 2007). Some indexes are based on experts' assessments of overall corruption in a country. Most often, these experts are either managers of multinational firms operating in these countries or financial analysts who study the investment risk of several countries across the globe. Managers and financial analysts are unlikely to have specific personal experience with having to give petty bribes to get things done. It is also more likely that a CEO has personal experience with a different type of corruption, political corruption, for example, and may refer to that when asked about corruption. In sum, experts in one country may refer to political corruption and experts in another county may refer to the practice of doling out protection money to the local thugs to keep shipments safe and on time. Third, there is a contextual problem with perceptions as each respondent has his or her own point of reference and it is unlikely to be shared by many (Bertrand et al. 2001). It is possible that individuals do not share the same point of reference even when experiencing the same incident. For example, it is not clear that if a manager considers a country to be "very corrupt" another manager that works in other countries will share the same relative measure of what is "very corrupt." Among many things, the personality or state of mind of the respondent may affect responses, some seeing the glass half empty and others half full. Most of these criticisms point out why there is likely to be a gap between perceptions of corruption and direct experience with it. Convincing as these arguments are, there is little direct evidence about the nature or size of this difference between perceptions and experience. Exploiting the perceptions 5 and the direct experience data contained in the Enterprise Surveys, we have the ability to explore the existence and nature of the gap between the two. 2.2 Are perceptions correlated with the incidence of bribes? In likely response to the criticisms and limitations of perceptions data on corruption, there is a nascent empirical literature developing on the relationship between experience with corruption and perceptions of corruption.7 Olken (2006) takes a detailed, micro approach and examines corruption perceptions and incidences of skimming off the top from road construction in Indonesia. He finds that perceptions on the extent of corruption and objective levels are a good reflection of each other when the extent of corruption can be easily and objectively confirmed. However, when corruption is carried out in ways that are easily hidden and unverifiable, perceptions and the extent of corruption begin to diverge. Two country studies, one in Ukraine (Cábelková and Hanousek 2004) and one in Uruguay (Moyal et al. 2004), show that media sources are influential in forming perceptions of corruption and, most importantly, these studies show how negative perceptions of corruption may reinforce people's willingness to offer bribes. However, these two studies do not have data that tally direct experiences with bribing or other forms of corruption. While these micro studies show that it is difficult to disentangle the relationship between perception and experience, Tirole (1996) provides a theoretical model as to the dynamic nature of perceptions and experience and under what conditions perceptions and experience of corruption diverge. With respect to the dynamics, collective reputations are difficult to change. With respect to their divergence, Tirole points out that the corrupt acts of others stick to all officials, even when there may be only a handful of corrupt officials. Furthermore, this collective reputation provides few incentives for even honest civil servants to maintain their integrity and remain incorruptible. In sum, both the nature of the dynamics and of the collective reputation of corruption requires many repeated acts of honesty from public officials to wipe out the perception that corruption is prevalent. Informed from what we know of the empirical and theoretical literature, we fill the gap of research on the nature and (possibly) its dynamics on the relationship between perception and experience with corruption. We do so by matching perceptions data on corruption to the transaction data to determine if perceptions differ when a firm is or is not a victim of graft and for the first time, to the best of our knowledge, determine the relationship between the two. For perceptions, we use data from cases where firms are asked to rank the top three obstacles, out of a list of sixteen, that affect the operations of the establishment. Answers from all entrepreneurs asked 7 For a cross country examination of the relationship between perception of and experience with corruption, see Weber (2007). 6 this question is the left-hand side, dependent, variable. For the experience variable, the Enterprise Surveys contain information on whether businesses had to provide a bribe to complete any one of six different transactions: requests for an electrical connection, a water connection, telephone service, an import license, a construction-related permit, or an operating license. With this perceptions dependent variable and direct experience independent variable, we specify an econometric equation of the form: (1) Yijk = = G Gijk + GGijk + Xijk +ijk 10 nb nb b b The dependent variable, Yijk , is a binary variable that equals one if the firm ranked corruption as one of the top three obstacles and zero if it did not. The are two dummy variables, Gijk and Gijk , that represent b nb firms that solicited a service or license and reported bribing or no bribing, respectively. These dummy variables of graft are the objective or experiential measure of corruption that is of interest here. The Gijk nb variable takes on a value of one when a firm solicited any of the services but no bribe was solicited or expected and zero otherwise. The Gijk variable takes on a value of one when a service was solicited and a b bribe was asked for or expected and zero otherwise. The omitted category is firms that did not solicit services or licenses. In estimating Equation 1, we include a matrix Xijk of control variables. It includes country and industry fixed effects to account for the unobserved parameters at the country and industry levels. It also includes firm size and age, and a binary variable indicating whether the firm has experienced arson, robbery or theft in the last year, as another control variable. We include the latter variable since corruption and crime are often symptoms of governance systems that are not functioning well. We expect that crime and corruption go hand in hand and our econometric specification would suffer from omitted variable bias had we not included some control for some measure of the quality of governance systems in the business environment. The term ( ) is an error term that is potentially heteroskedastic and that may be correlated across all firms within each country. Therefore, we calculate robust standard errors and allow for clustering by country. Of interest are the signs of coefficients G and G . If the true extent of corruption were nb b common knowledge to all firm, then we could reasonably expect to see no difference in the way in which firms perceive corruption (when G = 0 or G = 0), after controlling for observables, irrespective of nb b whether they were bribed or not. In this situation, we would expect to see a very close relationship between corruption perceptions and its incidence. On the other hand, if the extent of corruption is 7 imperfectly observed, firms will update their views on the problems posed by corruption based on their experience in dealing with corrupt officials. In particular, a firm that solicits a service or license and is not asked for a bribe will be more sanguine about the problems posed by corruption than firms that do not solicit any service or license ( G < 0). Similarly, a firm that requests a service or license and is asked for nb a bribe will become more cynical than firms that do not make any requests ( G > 0). A corollary of this b is that firms that do not request any licenses services will adjust their views on corruption more gradually than firms that do engage public officials and face the possibility of being asked for bribes. Table 1 presents Probit regression estimates of Equation 1. The results show that firms that undertake some transaction and are not victims of bribes are less likely to rank corruption as a top-three obstacle to enterprise performance ( G < 0). These results hold among firms in both regions as well as nb for the pooled data and are independent of the size or age of the firm. One hypothesis consistent with this empirical finding is that there is updating taking place. Firms begin the transaction with a similar perception as to the severity of corruption that most other firms share. However, once firms complete a transaction in which they did not have to resort to bribing, they do not have their initial perceptions verified and, as a result, become more optimistic about the extent of corruption than their peers. This positive updating of perceptions on corruption is independent of the initial average level of corruption. In other words, even in countries, in this sample, that are relatively bribe-free, firm's perceptions of corruption are on average improved when they are not asked for a bribe to complete their transaction. Regarding firms that are asked for bribes, we see that the relevant coefficient, G , has a positive b sign but is not statistically different from zero. That is, updating does not seem to take place when a firm reports being asked for a bribe. For firms that were asked for a bribe to complete a transaction, their perception of the severity of corruption is statistically indistinguishable from the perception of firms that did not solicit any services. This indicates that there is inertia in perceptions about corruption. This inertia is consistent with what Tirole pointed out in his theoretical model on the collective reputations of corrupt officials. Taken together, the result that updating only takes place when firms complete transactions without having to bribe, coupled with the result that firms that had to resort to bribing are statistically indistinguishable from firms that did not deal with officials at all, tell us that perceptions are likely to lag objective measures on corruption in cases where the extent of corruption is in flux. For the aims of the present paper, the evidence herein supports the argument that perceptions-based measures of corruption are an imperfect proxy for the true incidence of graft. 8 3 Measuring the incidence of graft We have presented evidence showing that using perceptions about the severity of corruption is an inaccurate gauge of the true extent of corruption. Since it is safe to say that corrupt practices have serious economic consequences and, hence, that the study of corruption based on reliable data merits careful attention, in this section we propose a measure of graft that is based on actual firm encounters with dishonest officials. To that end, we take advantage of detailed survey data on approximately 10,000 firms in 33 African and Latin American countries to calculate what we call a Graft Index of Firm Transactions (GIFT). The data come from the World Bank Group's Enterprise Surveys, which collect information on whether businesses had to provide a bribe to complete a series of six different transactions: requests for an electrical connection, a water connection, telephone service, an import license, a construction-related permit, or an operating license. 3.1 How is the Graft Index of Firm Transactions calculated? Formally, the index is defined as the sample probability that a firm in country k will be asked to provide a bribe, conditional on the firm undertaking one of the six aforementioned transactions. Mathematically, 6 njk xijk (1) GIFTk = j=1 i=1 , 6 njk j=1 is the GIFT for country k , where the sub-indices i and j represent firm i and transaction type j , respectively. The binary variable xijk is equal to 1 if firm i was requested to make an informal payment when undertaking transaction type j in country k and 0 otherwise. The denominator is the sum of all transactions of type j in country k that occurred between the time the survey took place and up to two years prior. In words, the Graft Index of Firm Transactions is the proportion of instances in which firms were either expected or requested to pay a gift or informal payment over the number of total solicitations for public services, licenses or permits for that country. We emphasize that the index is based on the respondent's direct experience with corruption. As such, this index does not have the disadvantages that are present in perceptions indexes. In addition, the index can be compared across countries. All firms 9 across the globe must undertake the transactions listed above at one point or another in the life and operation of the business. This index can be criticized for being based on the self-reporting of illegal activities. Interviewees' may fear the consequences of answering honestly, especially if they have themselves been directly involved in corrupt transactions. However, the questions asked puts the interviewee in the role of victim and not promoter of corruption. It would be very different to worry about receiving an honest answer when asking about how much and who a firm had to bribe to be granted a lucrative government contract than it is when asking the same firm to tell the interviewer if the firm was compelled to provide a bribe to get a service or license. The index can also be criticized for its narrow focus on the bribery of officials delivering infrastructure services and licenses. The index does not measure corruption that may take place in large- scale business transactions such as favorable deals on government contracts, the granting of government licenses or rights of use of public goods to insiders, rigged participation in public tenders, or lax enforcement of regulations or terms of government contracts because of a payoff. The index also does not measure corruption in situations where an economic transaction is not concerned, such as in the legal system where a court is asked to look the other way or to rule in favor of a party that paid a bribe. Lastly, the index also does not deal with political corruption; that is corruption associated with manipulated and non-transparent elections, the buying of legislative votes, or political nepotism. These kinds of corruption may involve both greater amounts of money and represent larger economic distortions than the common, petty corruption that our index measures. 3.2 GIFT estimates We estimate the GIFT for all six transactions taken together (Table 2). We also grouped transactions into two separate subsets, infrastructure (electricity, water, telephone --- Table 3) and licensing (import licenses, construction permits, operating licenses --- Table 4), and estimate the GIFT for each subset. Finally, we estimate the GIFT for each transaction separately in each country, but we warn that in many instances our confidence intervals become large (Tables A.1 to A.6). In all cases, we estimated the standard error of our estimate and its 95-percent confidence interval. In countries with few transactions, the confidence intervals can be substantial, making it hard to definitively rank several countries. We first note that, pooling data from all 33 countries, a total of 9519 requests for licenses or infrastructure services were registered in the Enterprise Surveys. Firms reported being asked for bribes in 933 instances. Thus, on average, firms in these countries are the target of bribery one out of 10 times they perform any of the six transactions included in the survey. Nevertheless, the difference in the incidence of 10 graft among African and Latin American firms is substantial. The former are subjected to bribery more than 19 percent of the time, compared to less than 7 percent among their Latin American peers. In other words, African firms are three times as likely to be the victims of corruption relative to firms in Latin America.8 Table 2 reports GIFT estimates for each country taking all transactions together; Figure 1 depicts countries ordered from less to more corrupt, according to the point estimate of the graft index. Namibia stands out as the least corrupt country in our sample. Out of 166 transactions recorded, no instances of requests for bribes were recorded. The 95-percent confidence interval suggests that only as many as 2.7 percent of all firms would be targeted by corrupt officials in that country. The next four least graft-prone countries in our sample are all in Latin America (Uruguay, Chile, Colombia, and El Salvador). The probability that a firm is the target of bribes in those countries lies between one and 4.4 percent. At the opposite end, the five most corrupt countries in the sample are all in Africa. In the Democratic Republic of Congo, the most corrupt country in our sample, a firm will be asked for bribes 53 to 72 percent of the time with a 95 percent probability, whereas in Guinea, Cameroon, and Mauritania, more than half of all firms will be asked for bribes. It is important to keep in mind that our index measures graft imprecisely and, therefore, that one cannot simply take the point estimates behind Figure 1 to make statements about whether graft is more pronounced in one country than in another. In order to say something about the relative incidence of graft between two countries, we calculated whether their corresponding GIFT estimates are statistically different. Results appear in Table 5. For example, although Namibia has the lowest estimated incidence of corruption, it is statistically as uncorrupt as Uruguay and less corrupt than all other countries. Uruguay, in turn, displays the same level of graft as Namibia, but one could not reject the hypothesis that graft in that country is the same as in Chile, Colombia, El Salvador, Rwanda, Botswana, Argentina, and Panama. Rwanda's GIFT estimate is particularly noisy, given that the number of observed transactions and instances of corruption are very low (see Table 2); hence, despite its low GIFT of 0.031, only eight countries appear to have unambiguously lower or higher levels of corruption than Rwanda. Some differences among neighboring countries are interesting on their own. El Salvador, for instance, is significantly less corrupt than Mexico and other countries in Central America -- Guatemala, Nicaragua and Honduras; the latter, in contrast, is significantly more corrupt than the other four countries. Likewise, 8 The 95 percent confidence interval of the odds ratio of the GIFT of the two regions goes from 2.52 to 3.21. 11 in the Andean region, Colombia stands out as less corrupt than the rest, while Paraguay and Ecuador are distinctly more corrupt.9 Among the most graft-prone countries in the sample, Guinea, Cameroon, Mauritania and DR Congo are statistically more corrupt than the 29 countries in our sample with lower graft indices. It should be noted that Paraguay and Ecuador, the two most corrupt countries in Latin America, appear to be less corrupt only when compared to the latter four extreme cases (Guinea, Cameroon, Mauritania, and DR Congo), and are equally corrupt, or even more so, than the rest of the African countries in our sample. That the index may vary so widely among countries in the same region suggests that corruption is unlikely to be explained by historical or cultural traits, but rather by the institutional environment that exists in each country. We explore that possibility in the next section. We turn now to the incidence of graft by type of transaction; see Tables 3 and 4. Looking at our sample as a whole, bribery is more prevalent when soliciting licenses or permits than when requesting infrastructure services. The data show that 11.3 percent of firms are asked for bribes in the former case, three percentage points more than when requesting electricity, water, or telephone connections; the gap is statistically significant at the 95 percent level. Nevertheless, the difference is driven primarily by Latin American firms. Whereas in Africa we do not find any statistically significant difference in the incidence of graft between licensing or infrastructure transactions, in Latin America obtaining licenses puts firms at a higher risk of being asked for bribes, 8.3 percent vs. 5.3 percent relative to requests for infrastructure services. On a country by country basis, the probability of being asked for bribes in licensing vis-à-vis infrastructure is statistically higher in Argentina, Bolivia, El Salvador, Peru, and DR Congo; the converse is only true in Malawi and Niger. Three hypotheses come to mind in trying to explain differences in graft incidence across licensing and infrastructure. First, a number of countries in the world have privatized the provision of infrastructure services, primarily in telecommunications, but also in water and electricity provision. Private providers of such services would have greater incentives to setup mechanisms that prevent their employees from requesting informal payments from their customers, while perhaps increasing formal fees that would accrue to profits. Second, at least in the case of telephony, competition, especially from mobile telephones, would appear to be stiffer, which would reduce the ability to extract rents from firms. Third, government regulation and red-tape is more common in obtaining licenses and permits and, as we show in the following section, excessive regulation creates opportunities for corrupt officials to extract bribes from firms. 9 The Andean region is comprised of Bolivia, Colombia, Ecuador, Paraguay, Peru, and Venezuela. 12 4 What lies behind the incidence of corruption? Having estimated measures of the incidence of corruption, in this section we explore some of its correlates. We do this by running regressions of the Graft Index of Firm Transactions on a number of different regressors, motivated by the existing literature. Specifically, we consider whether firms are more likely to fall prey to corrupt officials in overly-regulated economies and in less democratic countries. Admittedly, cross-country data make it difficult to identify the causal links between graft and potential explanatory variables. With this caveat in mind, our aim is to shed light on some of the factors that are believed to be drivers of corruption. The existence of burdensome business regulations stands out as a potential driver of graft.10 While some degree of regulation could be justified under the argument that it is required to safeguard the public interest, a competing explanation, the tollbooth view (Shleifer and Vishny 1993), is that regulations are put in place in order to extract rents in favor of specific business interests or government officials. Djankov et al. (2002) explore such alternative explanations and conclude that, rather than protecting the public interest, regulation --in their study, business entry rules-- is associated with greater levels of corruption. In the same vein, Svensson (2005) presents econometric evidence showing a positive link between greater regulation and more corruption. More recently, Olken and Barron (2007) look at bribe payments by truck drivers at checkpoints along Indonesian roads and find support for Shleifer and Vishny's (1993) tollbooth hypothesis. Both the study by Djankov et al. (2002) and that of Svensson (2005) rely mainly on corruption perception measures which, as we have argued, are only an imperfect approximation to actual corrupt practices. Thus, it is worth asking whether regulation might be behind corruption when we use our index of the actual incidence of graft. Figure 2 shows that there is a clear positive correlation between the incidence of graft and the extent and nature of the regulatory and legal obligations that firms face. The latter is obtained from the "Ease of Doing Business" indicator in the Doing Business dataset, with a higher measure indicating a less benign business environment. The GIFT allows us to delve into the subject. For example, in Figure 3 we show that the probability of being the victim of graft when requesting an operating license or a construction permit are positively correlated, respectively, with Doing Business measures of restrictions on starting a business and problems in dealing with licenses in construction projects. Econometric results in Table 6 confirm what we see graphically: excessive regulation is associated with more graft even after taking into consideration other factors that may explain the level of corruption. The results in column 2 suggest that a one-standard deviation decline in obstacles to doing 10 See Bardhan (1997) for a discussion. 13 business reduces the probability of being the victim of graft by 8 percentage points. Likewise, from column 3, reducing constraints in starting a business by one-standard deviation results in a 7.5 percent- point lower probability that firms will be asked for bribes when requesting an operating license. Last, the likelihood that firms will be hit by graft when requesting construction permits is six percentage points lower following a one-standard deviation reduction in the Doing Business "Dealing with licenses" measure (column 4). Therefore, our results confirm previous evidence, based on perceptions data, linking regulation and corruption, with the added benefit that we are able to focus more narrowly on specific regulations and transactions affected by corrupt practices. Firms are also more susceptible to graft in countries where the institutional environment is weak and, in particular, where the accountability of government officials is limited. In particular, democratically-elected governments are more open to public scrutiny and hence are more likely to adopt anti-corruption efforts (Bardhan 1997; Treisman 2000; Svensson 2005). We use data from the Polity IV Project to study the relationship between democracy and corruption.11 Our measure of democracy is based on the "polity score," which provides a measure of competitiveness in the process of executive recruitment, constraints on the chief executive, and the competitiveness and regulation of political participation. The polity score takes values from -10 to 10, with increases in the score reflecting a more democratic political regime. As we report in Table 6, firms in democratic countries are less likely to be asked for bribes. In column 5, the estimated coefficient for the polity score is negative and significant at the 10 percent level. When we include "Ease of doing business" as a measure of regulation (column 6), both the latter and the polity coefficient have the expected sign and are significant at the 10 percent level; the hypothesis that both of them are jointly equal to zero is rejected at the five-percent level. The fact that both coefficients are still statistically significant is of interest. Djankov et al (2002) show that democratic governments are less likely to adopt costly regulations. In Table 5 we observe that the estimated coefficient for regulation falls when we account for the level of democracy, which is consistent with the evidence in Djankov et al (2002). In addition, in our sample, even holding constant the level of democracy, regulation is still associated with more graft. Columns 7 and 8 show that the positive association between bureaucratic constraints in starting a business or obtaining a construction license, on the one hand, and the incidence of graft in obtaining and operating license or construction permits, on the other, remains statistically significant. In order to put our previous results in perspective, let us consider how much graft in Africa would decline if both the levels of democracy and of regulation moved to those that exist in Latin America. 14 Countries in the former region are characterized by weaker democracies and more regulation, as well as more pervasive graft. Among countries in our sample, the median polity level is -1 in Africa and 8 in Latin America. Moreover, the "Ease of doing business" percentile rank is .67 in Africa and .49 in Latin America. The estimates in column 6 of Table 6 imply that strengthening democracy and reducing regulation from their respective median levels in Africa to those of Latin America would reduce the probability that firms are victims of graft by 16.2 percentage points from its average level of 19 percent; that is, the incidence of bribery in Africa would fall by 85 percent under this scenario. Thus, our back-of- the-envelope calculation suggests that fostering democracy and reducing excessive regulation would go a long way in improving Africa's business climate by reducing corruption. 5 Conclusions In this paper we argue that existing measures of corruption around the world are an inaccurate gauge of the true incidence of graft on the typical firm. Such measures are often based on surveys of experts, of specific types of firms, or of households, which do not necessarily match one-to-one with the views held by the typical firm. Moreover, existing indicators are often based on perceptions and not necessarily on hard data. Yet, we present evidence showing that average firm perceptions adjust only gradually to changes in the business environment. We show that firms that request licenses or infrastructure services and are not asked for bribes hold a more sanguine view about the pervasiveness of graft, relative not only to firms that did fall prey to corrupt officials but also to firms that did not request such services and hence would have not been affected by bribery. Then, for example, if a country were to effectively launch an anti-corruption campaign, firms' views on corruption would change, but only gradually, to the improved business climate. In order to remedy the shortcomings of existing corruption measures, we introduced an experience-based index, the Graft Index of Firm Transactions, which measures the probability that a firm will be asked for a bribe in order to complete a specified set of business transactions. We estimated the index using data on approximately 10,000 firms from the World Bank's Enterprise Surveys in 33 countries in Africa and Latin America. Our index has several advantages: It is based on firms' direct encounters with corruption and not on perceptions; it is free of ambiguities as it focuses on a common set of business transactions in all countries; and it reflects the incidence of graft on the typical firm of a country since it is based on nationally representative data. On the downside, our index focuses on petty bribery and does not capture other possible forms of corruption. 11 See Marshall and Jaggers (2005). Data and documentation available at http://www.cidcm.umd.edu/polity/ 15 The Graft Index of Firm Transactions shows that African firms are three times as likely to be victims of bribery than their Latin American counterparts. Within each region, though, there is substantial variation. Namibia, along with Uruguay, stands out as the least corrupt country. Paraguay and Ecuador, the most corrupt Latin American countries in our sample, lag behind several African countries. Corruption is gravest in four African countries -- Guinea, Cameroon, Mauritania, and the Democratic Republic of Congo. In those countries, around one in two firms is the victim of bribery. Our index also indicates that bribery is more common when requesting licenses or government permits than when requesting infrastructure services such as telephone, water, or electricity connections. In order to shed light on the factors that lie behind corruption, we run country-level regressions with our graft index as dependent variable. We find a strong correlation between excessive regulation and corruption, with a one-standard deviation in the ease of doing business reducing graft by approximately one third of a standard deviation. Likewise, democratic governments do a better job in curtailing corruption. As a back-of-the-envelope application of these findings, our results imply that bribery in Africa would fall by 85 percent if it had levels of democracy and regulation similar to those that exist in Latin America, closing the gap in the incidence of graft between the two regions. 16 References Ades, A. F., and R. di Tella 1997. "National champions and corruption: Some unpleasant interventionist arithmetic." The Economic Journal 107(443):1023­42. Ades, A. F., and R. di Tella 1999. "Rents, competition, and corruption." American Economic Review 89(4): 982­93. Bardhan, P. 1997. "Corruption and development: A review of issues." Journal of Economic Literature, Vol. XXXV (Sept.): 1320-1346. Bertrand, M., and S. Mullainathan 2001. "Do people mean what they say?: Implications for subjective survey data." American Economic Review 91 (2), (May) pp. 67-72. Cábelková, I., and J. Hanousek 2004. "The power of negative thinking: corruption, perception and willingness to bribe in Ukraine." Applied Economics, 36:4, 383-397. Djankov, S., R. La Porta, F. Lopez-de-Silanes, and A. Shleifer. 2002. "The Regulation of Entry." Quarterly Journal of Economics vol. 117(1): 1-37. Fisman, R. and J. Svensson 2007. "Are corruption and taxation really harmful to growth?: Firm level evidence." Journal of Development Economics, 2007, 83 (1): 63-75. Kaufmann, D., A. Kraay, and M. Mastruzzi. 2007. "Governance Matters VI: Aggregate and Individual Governance Indicators for 1996-2006." Policy Research Working Paper 4012, World Bank, Development Research Group, Washington, D.C. Knack, S. 2006. "Measuring corruption in Eastern Europe and Central Asia: A critique of the cross- country indicators." Policy Research Working Paper 3968, World Bank, Development Research Group, Washington, D.C. Lambsdorff, J.G. 2006. "The methodology of the TI Corruption Perceptions Index 2006." http://www.icgg.org/downloads/CPI_2006_Methodology.pdf Lanyi, A. 2004. "Measuring the economic impact of corruption: A survey." The IRIS Discussion Papers on Institutions and Development 04/04, Center for Institutional Reform and the Informal Sector, University of Maryland. Marshall, M., and K. Jaggers. 2005. "Polity IV project: Dataset users' manual." Center for Global Policy, School of Public Policy, George Mason University. Mauro, P. 1995. "Corruption and growth." Quarterly Journal of Economics 110 (August): 681­712. 17 Moyal, P., M. Rossi, and T. Rossi 2004. "De la percepción de la corrupción a la coima: un puente invisible." Universidad de la República, Facultad de Ciencias Sociales, Departamento de Economía, Documento de Trabajo 09/04. Olken, B.A. 2007. "Corruption perceptions vs. corruption reality." Working Paper 12428 (March), NBER, Cambridge, MA. Olken, B.A., and P. Barron 2007. "The simple economics of extortion: Evidence from trucking in Aceh." Working Paper 13145 (May), NBER, Cambridge, MA. Shleifer, A., and R.W. Vishny. 1993. "Corruption." Quarterly Journal of Economics, 108: 599-617. Svensson, J. 2005. "Eight questions about corruption." Journal of Economic Perspectives, vol. 19 (3): 19- 42. Tirole, J. 1996. "A theory of collective reputations (with applications to the persistence of corruption and to firm quality)." The Review of Economic Studies, vol. 63(1): 1-22. Treisman, D. 2000. "The causes of corruption: a cross-national study." Journal of Public Economics 76(3, June): 399­457. Weber Abramo, C. 2007. "How much do perceptions of corruption really tell us?" Economics Discussion Papers, Discussion Paper 2007-19 (Available at: http://www.economics- ejournal.org/economics/discussionpapers). 18 Tables and Figures Table 1 Perception vs Incidence of Corruption: Probit Regression Results (Dependent variable: Firm ranked corruption a top three obstacle) (1) (2) (3) Latin America Africa Pooled Employment (log) -0.013 -0.004 -0.011*** (0.004)*** (0.007) (0.004) Age of establishment 0.000 0.001 0.000 (0.000) (0.001) (0.000) Omitted category: No service solicited Solicited service and was asked for bribe 0.014 0.010 0.014 (0.024) (0.026) (0.019) Solicited service and was not asked for bribe -0.025 -0.037 -0.028 (0.012)** (0.016)** (0.010)*** Observations 6510 1877 8387 Notes: Robust standard errors in parentheses * significant at 10%; ** significant at 5%; *** significant at 1% 19 Table 2 Graft Index of Firm Transactions All transactions (Probability that a firm will be asked for bribes when undertaking any of six business transactions) Number of Number of 95% Confidence Interval transactions bribes Country Index recorded requested Standard error Lower bound Upper bound Namibia 0.000 166 0 0.000 0.000 0.027 Uruguay 0.021 387 8 0.007 0.010 0.041 Chile 0.023 744 17 0.005 0.014 0.037 Colombia 0.023 603 14 0.006 0.014 0.039 El Salvador 0.026 508 13 0.007 0.015 0.044 Rwanda 0.031 32 1 0.031 0.000 0.171 Botswana 0.037 163 6 0.015 0.015 0.080 Argentina 0.042 744 31 0.007 0.029 0.059 Panama 0.045 199 9 0.015 0.023 0.085 Mexico 0.055 490 27 0.010 0.038 0.079 Guatemala 0.064 453 29 0.012 0.045 0.091 Nicaragua 0.071 434 31 0.012 0.050 0.100 Bolivia 0.072 498 36 0.012 0.052 0.099 Peru 0.087 494 43 0.013 0.065 0.115 Venezuela 0.090 156 14 0.023 0.053 0.146 Burundi 0.098 61 6 0.038 0.042 0.202 Uganda 0.103 340 35 0.016 0.075 0.140 Burkina Faso 0.109 46 5 0.046 0.043 0.235 Malawi 0.120 275 33 0.020 0.086 0.164 Honduras 0.121 390 47 0.016 0.092 0.157 Angola 0.127 189 24 0.024 0.086 0.183 Swaziland 0.143 70 10 0.042 0.078 0.245 Paraguay 0.143 370 53 0.018 0.111 0.183 Cape Verde 0.152 33 5 0.062 0.062 0.314 Guinea-Bissau 0.154 78 12 0.041 0.089 0.251 Ecuador 0.159 666 106 0.014 0.133 0.189 Tanzania 0.167 233 39 0.024 0.125 0.221 Gambia 0.183 60 11 0.050 0.104 0.301 Niger 0.201 179 36 0.030 0.149 0.266 Guinea 0.454 108 49 0.048 0.363 0.548 Cameroon 0.466 176 82 0.038 0.394 0.540 Mauritania 0.514 74 38 0.058 0.402 0.624 Congo, Dem. Rep. 0.630 100 63 0.048 0.532 0.718 Notes: Information about requests for water connections was not collected in Venezuela. Standard errors and confidence intervals were calculated pooling all data togethere and assuming that the request for bribes follows a binomial distribution. 20 Table 3 Graft Index of Firm Transactions Infrastructure services (Probability that a firm will be asked for bribes when requesting electricity, water or telephone connections) Number of Number of 95% Confidence Interval transactions bribes Country Index recorded requested Standard error Lower bound Upper bound Namibia 0.000 84 0 0.000 0.000 0.052 El Salvador 0.007 274 2 0.005 0.000 0.028 Bolivia 0.008 243 2 0.006 0.000 0.031 Uruguay 0.015 197 3 0.009 0.003 0.046 Chile 0.017 404 7 0.006 0.008 0.036 Argentina 0.022 461 10 0.007 0.011 0.040 Colombia 0.025 363 9 0.008 0.012 0.047 Panama 0.035 115 4 0.017 0.011 0.089 Peru 0.035 254 9 0.012 0.018 0.067 Botswana 0.036 56 2 0.025 0.003 0.128 Mexico 0.049 366 18 0.011 0.031 0.077 Guatemala 0.060 248 15 0.015 0.036 0.098 Swaziland 0.067 30 2 0.046 0.008 0.224 Burundi 0.067 30 2 0.046 0.008 0.224 Nicaragua 0.077 221 17 0.018 0.048 0.120 Venezuela 0.082 85 7 0.030 0.038 0.163 Uganda 0.101 89 9 0.032 0.052 0.183 Angola 0.122 90 11 0.035 0.068 0.207 Honduras 0.146 144 21 0.029 0.097 0.213 Guinea-Bissau 0.152 33 5 0.062 0.062 0.314 Burkina Faso 0.154 26 4 0.071 0.055 0.341 Cape Verde 0.154 13 2 0.100 0.031 0.435 Paraguay 0.158 171 27 0.028 0.110 0.220 Malawi 0.161 143 23 0.031 0.109 0.230 Tanzania 0.170 100 17 0.038 0.108 0.256 Ecuador 0.192 271 52 0.024 0.149 0.243 Gambia 0.238 21 5 0.093 0.102 0.455 Niger 0.309 68 21 0.056 0.211 0.427 Guinea 0.462 65 30 0.062 0.346 0.581 Cameroon 0.475 40 19 0.079 0.329 0.625 Congo, Dem. Rep. 0.484 31 15 0.090 0.320 0.652 Mauritania 0.569 51 29 0.069 0.433 0.695 Notes: Information about requests for water connections was not collected in Venezuela. Standard errors and confidence intervals were calculated pooling all data togethere and assuming that the request for bribes follows a binomial distribution. 21 Table 4 Graft Index of Firm Transactions Licensing (Probability that a firm will be asked for bribes when soliciting import, operating or construction licenses) Number of Number of 95% Confidence Interval transactions bribes Country Index recorded requested Standard error Lower bound Upper bound Namibia 0.000 82 0 0.000 0.000 0.054 Colombia 0.021 240 5 0.009 0.008 0.049 Uruguay 0.026 190 5 0.012 0.010 0.062 Chile 0.029 340 10 0.009 0.015 0.054 Rwanda 0.034 29 1 0.034 0.000 0.186 Botswana 0.037 107 4 0.018 0.012 0.095 El Salvador 0.047 234 11 0.014 0.026 0.083 Burkina Faso 0.050 20 1 0.049 0.000 0.254 Panama 0.060 84 5 0.026 0.022 0.135 Nicaragua 0.066 213 14 0.017 0.039 0.108 Guatemala 0.068 205 14 0.018 0.040 0.112 Mexico 0.073 124 9 0.023 0.037 0.134 Argentina 0.074 283 21 0.016 0.049 0.111 Malawi 0.076 132 10 0.023 0.040 0.135 Venezuela 0.099 71 7 0.035 0.046 0.193 Uganda 0.104 251 26 0.019 0.071 0.148 Honduras 0.106 246 26 0.020 0.073 0.151 Burundi 0.129 31 4 0.060 0.045 0.295 Paraguay 0.131 199 26 0.024 0.090 0.185 Angola 0.131 99 13 0.034 0.077 0.213 Bolivia 0.133 255 34 0.021 0.097 0.181 Niger 0.135 111 15 0.032 0.083 0.212 Ecuador 0.137 395 54 0.017 0.106 0.174 Peru 0.142 240 34 0.023 0.103 0.192 Cape Verde 0.150 20 3 0.080 0.044 0.369 Gambia 0.154 39 6 0.058 0.069 0.301 Guinea-Bissau 0.156 45 7 0.054 0.074 0.291 Tanzania 0.165 133 22 0.032 0.111 0.238 Swaziland 0.200 40 8 0.063 0.102 0.350 Mauritania 0.391 23 9 0.102 0.221 0.593 Guinea 0.442 43 19 0.076 0.304 0.589 Cameroon 0.463 136 63 0.043 0.382 0.547 Congo, Dem. Rep. 0.696 69 48 0.055 0.579 0.792 Notes: Standard errors and confidence intervals were calculated pooling all data togethere and assuming that the request for bribes follows a binomial distribution. 22 23 ,TR M,R ,YRP N,IG Z, R,EN ZAR,TR ZAR,TR RAZ uptr SW, B, M,R r r r r r r r r r r r cor ZAR,TRM M,R ZAR,TRM T,R M R, e or othellA othe othe All All rehtollA CM,NIG, MG othe ZAR AGO, othe othe othe othe othe othe othe NER All A,ZT, ZAR,TR M R, R, MC m B, HND R, All All All All All All All CM,NIG, MC CM,NIG, MC N, ... GM A, ECU CM NER NER A,ZT U, UG,REP IN,G NB,G ZA,T A,ZT IN,G GI,REN EC PV,C CU,E CU,E ,L , ... A, ANP I, V V Z, PAN , , yll BOL,, BO ,M CP CP VEN BW PAN G, BD, ca A, A, RG,A PAN ARG AR A, A, VEN isti uptr RW BW A, A, BW BW R, NIC,MTG NIC,, VPC GT, BFA EX I, M EXM, BFA, ,MTG,XE BFA, PER, A, M A, L, stat e orc A, BW r r A, A, PE BFA, UG N, BO r r r r Y BD N, UG WS,OGA,DNH art UR SLV.,L PAN, RW A, BW,A heto RW All hetollA PA I, RW MTG,X L, RW PAN, I, I, NBG BO BFA PAN,, PA I,DB, G, BD W heto M hat ME VEN, G, BD, hetollA hetollA All hetollA rehtollA equally CO ARG esir SLV.,L V, WR,L SLV. SLV. ARG AR C,IN,M CPV, R, AR VEN ... HL,C, CO NIC,, ARG A, VEN A, VEN, GT, BFA, A, CH BW A, PE PE 5 Count AMN CO,YRU SL,L RY,U L,HC,YRU L,OC,YRU L,OC,YRU MTG,XEM A, PE A, BW R, R, A, WB,A R,EP WB,A L, EXM, UG I, RW WR WR BO WR C,IN PAN BD , , Table )T ARG ARG ARG ARG ARG A, ARG ARG A, ARG A, L (GIF SLV SLV BW BW BW BO ns uptr L,OC L,OC SLV, SLV, SLV, SLV, M SLV, SLV, SLV, SLV, C,IN, actio cor None None NAM MAN LHC, LHC, M NAM NAM NAM HL,C, L,HC, L,OC L,OC L,OC L,OC L,OC ansrT less NAM NAM ... URY, URY, L,HC,Y L,HC,Y L,OC L, CH, L,HC,Y L,HC,Y L,OC GT,XEM L, CH, PAN, L,HC,Y L,OC L, GT,XEM UR, UR, UR, UR, UR, CH, Firm NAM NAM of URY, PAN, NAM NAM NAM NAM URY, NAM URY, ex NAM NAM NAM Ind aftrG FT GI 0.000 0.021 0.023 320.0 0.026 0.031 0.037 0.042 0.045 .0550 0.064 0.071 0.072 0.087 0.090 0.098 .1030 0.109 .1200 e th in Y L L A A odeC AMN UR CH CO SLV RW BW GRA NAP EXM MTG I ICN L A BO ERP ENV BDI UG BFA WM encesref dif o cali ry biai aib or a a naa a ala Fas uela i atist guayu St Count Nam Ur Chile moloC adlvaS and am coixe ia ez da lawa El Rw Botsw Argentina Pan M Guatem Nicaragu Boliv Peru Ven Burundi Ugan Burkina M RAZ uptr T,R M cor R, ZAR,TRM ZAR,TRM ZAR,TRM ZAR,TRM ZAR,TRM ZAR,TRM ZAR,TRM ZAR,TRM ZAR,TRM e or MC R, R, R, R, R, R, R, R, R, ZAR ZAR oneN None m N, MC MC MC MC MC MC MC MC MC ... GI,REN IN,G IN,G IN,G IN,G IN,G IN,G IN,G IN,G IN,G . GIFT het ... gniylr yll unde ca n isti uptr tiou stat e orc art rehtollA TR ZAR ribt her her her her her her her her her R, ot ot ot ot ot ot ot ot ot M,R TRM TR M All All All All All All All All All hat equally MC IN,G CM dislai IN,G m esir ... bino the of Count tioar (continued)5 , , , , , , , , , , , , the )T ARG ARG ARG ARG ARG ARG ARG VEN ARG VEN ARG ARG of A, L A, L A, L A, PER A, A, L A, (GIF Table BW BO BW BO BW BO BW L, BW BW BO BW R,EP A, BW R,EP A, PER A, alv BW L, BW ns uptr SLV, C,IN, SLV, C,IN, SLV, C,IN, BO SLV, SLV, SLV, C,IN, L, L, BO DNH,I inter e SLV, BO SLV, BO SLV, SLV, M M M M MW, r r r r actio cor L,OC A A ansrT less L, ... CH, GT,XEM L,OC L, CH, GT,XEM L,OC L, CH, GT,XEM L,OC C,IN,M L,OC XEM, L,OC L,OC C,IN, UG L,OC C,IN, UG L,OC C,IN,M L,OC L, GT, L, PAN L, L, M L, M L, GT, L, MTG,X heto heto heto heto All All All All onfidencc CH, EXM, CH, CH, GT,XEM CH, 95% e Firm of URY, PAN, URY, PAN, URY, PAN, URY, PAN URY, URY, PAN, URY, GT,XEM, CH, URY, GT,XEM, CH, EXM, CH, URY, PAN URY, ME,NAP thfo ex NAM NAM NAM NAM NAM NAM NAM PAN NAM PAN NAM NAM tion Ind aftrG FT timase GI .1210 0.127 0.143 0.143 0.152 0.154 0.159 0.167 0.183 .2010 .4540 0.466 0.514 0.630 na e on th in D odeC HN OGA Z B R SW PRY PVC NBG ECU ZAT MG NER GIN MC TRM sed ZAR ba e encesref ars iont dif u Rep.. cali ry sa y r ia Dem ur iland Verde ola ania bia eroon o, Calcula:s atist az uado St Count Hond Ang Sw Paragua Cape Guinea-Bissa Ec anzT auritan Gam Niger Guinea Cam M Cong Note 24 25 n - 0 )4 4 )0 )**6 1 9) )8 TF ti )9 GI nstructiooC perm -0.03 (0.02 -0.04 (0.04 0.325 (0.15 -0.01 0.00( 0.276 (0.19 30 0.384 - gin ** ** )(7( FTIG eratpO 021 license 0. 0.017)( -0.023 0.036)( 322 0. (0.107)* -0.010 (0.003)* -0.179 0.167)( 454 30 0. llA- )6 TF 07 .022) .000 .035) 80 .203)* .011 GI transactions 0.0 (0 -0 (0 0.3 (0 -0 006)*.0( 094.0- .243) 25 (0 30 0.4 llA- sactions 16 19) 17 36) 13 07)* 8 38)** 6 )(5 FTIG tran -0.0 (0.0 -0.0 (0.0 -0.0 (0.0 0.29 (0.1 30 0.36 onti - 8 5)** 1 8) )*8 gulations. 6 )(4 TF GI ucrtsnoC ti 9)** re perm -0.08 0.03( -0.00 0.04( 0.322 (0.16 0.651 0.30( 33 0.446 e m Table - gin * ) ) )** ) burdenso )(3( GIFT Operat license 0.017- (0.024 .0070 (0.052 423.0 (0.137 -0.008 (0.196 33 356.0 more tes llA- deno * se )2 TF 06 .024) 68 .041) 46 .252)* GI transactions 0.0 (0 0.0 (0 0.6 (0 305.0- .269) 88 1% (0 33 0.3 increa at na llA- erehw sactions 56 21)** 5 68)*** 3 y )(1( FTIG significant***; ranks, tran -0.0 (0.0 0.53 (0.1 33 0.22 acrc %5 ile eses at emoD renth nta rcentep pa ific in + and, )e in sign ** + gs) + tion es scory errors pressed (lo ex laug business 10%; Variable: y ns are nt business ceil polit( dard att Re capita doing a dumm lesb per of aft, pende Gr De GNI Africa Ease Starting gininatbO acy ionsta stan : mocreD nstant tes Co Observ R-squared No Robust significan* Varia + Appendix Table A.1 Graft Index of Firm Transactions Electrical Connections (Probability that a firm will be asked for bribes when requesting an electrical connection) Number of Number of 95% Confidence Interval transactions bribes Country Index recorded requested Standard error Lower bound Upper bound Namibia 0.000 26 0 0.000 0.000 0.152 Venezuela 0.000 28 0 0.000 0.000 0.143 Botswana 0.000 14 0 0.000 0.000 0.251 Peru 0.011 89 1 0.011 0.000 0.067 Chile 0.014 141 2 0.010 0.001 0.053 El Salvador 0.019 104 2 0.013 0.001 0.072 Bolivia 0.023 86 2 0.016 0.001 0.086 Argentina 0.034 177 6 0.014 0.014 0.074 Colombia 0.035 114 4 0.017 0.011 0.090 Mexico 0.036 137 5 0.016 0.013 0.085 Uruguay 0.040 75 3 0.023 0.009 0.116 Panama 0.044 45 2 0.031 0.004 0.156 Burundi 0.063 16 1 0.061 0.000 0.303 Nicaragua 0.067 89 6 0.027 0.028 0.142 Guatemala 0.076 105 8 0.026 0.037 0.145 Angola 0.104 48 5 0.044 0.041 0.226 Swaziland 0.125 8 1 0.117 0.001 0.492 Burkina Faso 0.125 8 1 0.117 0.001 0.492 Paraguay 0.129 62 8 0.043 0.064 0.237 Guinea-Bissau 0.133 15 2 0.088 0.025 0.391 Uganda 0.140 43 6 0.053 0.062 0.276 Honduras 0.164 55 9 0.050 0.086 0.285 Ecuador 0.165 103 17 0.037 0.105 0.249 Malawi 0.182 44 8 0.058 0.092 0.322 Niger 0.208 24 5 0.083 0.088 0.409 Tanzania 0.243 37 9 0.071 0.132 0.403 Cape Verde 0.250 4 1 0.217 0.034 0.711 Gambia 0.286 7 2 0.171 0.076 0.648 Cameroon 0.417 12 5 0.142 0.193 0.681 Guinea 0.529 34 18 0.086 0.367 0.686 Congo, Dem. Rep. 0.615 13 8 0.135 0.354 0.824 Mauritania 0.667 15 10 0.122 0.415 0.850 Rwanda 0 Notes: Standard errors and confidence intervals were calculated pooling all data togethere and assuming that the request for bribes follows a binomial distribution. 26 Appendix Table A.2 Graft Index of Firm Transactions Water Connections (Probability that a firm will be asked for bribes when requesting a water connection) Number of Number of 95% Confidence Interval transactions bribes Country Index recorded requested Standard error Lower bound Upper bound Uruguay 0.000 27 0 0.000 0.000 0.148 Bolivia 0.000 25 0 0.000 0.000 0.158 Botswana 0.000 11 0 0.000 0.000 0.300 Swaziland 0.000 5 0 0.000 0.000 0.489 Chile 0.000 26 0 0.000 0.000 0.152 El Salvador 0.000 24 0 0.000 0.000 0.163 Namibia 0.000 16 0 0.000 0.000 0.227 Argentina 0.026 39 1 0.025 0.000 0.144 Nicaragua 0.029 34 1 0.029 0.000 0.162 Colombia 0.043 23 1 0.043 0.000 0.227 Honduras 0.056 18 1 0.054 0.000 0.276 Uganda 0.056 18 1 0.054 0.000 0.276 Paraguay 0.080 25 2 0.054 0.011 0.261 Guinea-Bissau 0.083 12 1 0.080 0.000 0.375 Panama 0.083 12 1 0.080 0.000 0.375 Guatemala 0.130 46 6 0.050 0.057 0.260 Angola 0.130 23 3 0.070 0.037 0.330 Mexico 0.161 31 5 0.066 0.066 0.331 Tanzania 0.167 24 4 0.076 0.061 0.365 Burkina Faso 0.167 6 1 0.152 0.011 0.582 Ecuador 0.171 35 6 0.064 0.077 0.331 Gambia 0.200 5 1 0.179 0.020 0.640 Peru 0.222 27 6 0.080 0.103 0.411 Malawi 0.263 19 5 0.101 0.115 0.491 Niger 0.286 14 4 0.121 0.113 0.550 Congo, Dem. Rep. 0.375 8 3 0.171 0.135 0.696 Guinea 0.391 23 9 0.102 0.221 0.593 Cape Verde 0.500 2 1 0.354 0.095 0.905 Cameroon 0.500 8 4 0.177 0.215 0.785 Mauritania 0.750 12 9 0.125 0.462 0.917 Rwanda 0 Burundi 0 Venezuela n.a. Notes: Information about requests for water connections was not collected in Venezuela. Standard errors and confidence intervals were calculated pooling all data togethere and assuming that the request for bribes follows a binomial distribution. 27 Appendix Table A.3 Graft Index of Firm Transactions Telephone Connections (Probability that a firm will be asked for bribes when requesting a telephone connection) Number of Number of 95% Confidence Interval transactions bribes Country Index recorded requested Standard error Lower bound Upper bound El Salvador 0.000 146 0 0.000 0.000 0.031 Rwanda 0.000 3 0 0.000 0.000 0.617 Cape Verde 0.000 7 0 0.000 0.000 0.404 Bolivia 0.000 132 0 0.000 0.000 0.034 Namibia 0.000 42 0 0.000 0.000 0.100 Uruguay 0.000 95 0 0.000 0.000 0.047 Guatemala 0.010 97 1 0.010 0.000 0.062 Argentina 0.012 245 3 0.007 0.002 0.037 Peru 0.014 138 2 0.010 0.001 0.055 Panama 0.017 58 1 0.017 0.000 0.100 Colombia 0.018 226 4 0.009 0.005 0.046 Chile 0.021 237 5 0.009 0.008 0.050 Mexico 0.040 198 8 0.014 0.019 0.079 Swaziland 0.059 17 1 0.057 0.000 0.289 Botswana 0.065 31 2 0.044 0.008 0.217 Uganda 0.071 28 2 0.049 0.009 0.237 Burundi 0.071 14 1 0.069 0.000 0.335 Nicaragua 0.102 98 10 0.031 0.055 0.179 Tanzania 0.103 39 4 0.049 0.035 0.242 Venezuela 0.123 57 7 0.043 0.058 0.236 Malawi 0.125 80 10 0.037 0.067 0.217 Honduras 0.155 71 11 0.043 0.087 0.258 Angola 0.158 19 3 0.084 0.047 0.384 Burkina Faso 0.167 12 2 0.108 0.035 0.460 Paraguay 0.202 84 17 0.044 0.129 0.301 Ecuador 0.218 133 29 0.036 0.156 0.296 Gambia 0.222 9 2 0.139 0.053 0.557 Guinea-Bissau 0.333 6 2 0.192 0.093 0.704 Guinea 0.375 8 3 0.171 0.135 0.696 Congo, Dem. Rep. 0.400 10 4 0.155 0.167 0.688 Niger 0.400 30 12 0.089 0.246 0.577 Mauritania 0.417 24 10 0.101 0.244 0.612 Cameroon 0.500 20 10 0.112 0.299 0.701 Notes: Standard errors and confidence intervals were calculated pooling all data togethere and assuming that the request for bribes follows a binomial distribution. 28 Appendix Table A.4 Graft Index of Firm Transactions Construction Permits (Probability that a firm will be asked for bribes when requesting a construction permit) Number of Number of 95% Confidence Interval transactions bribes Country Index recorded requested Standard error Lower bound Upper bound Namibia 0.000 21 0 0.000 0.000 0.182 Uruguay 0.000 70 0 0.000 0.000 0.062 Swaziland 0.000 5 0 0.000 0.000 0.489 Burkina Faso 0.000 4 0 0.000 0.000 0.546 Colombia 0.000 41 0 0.000 0.000 0.102 Chile 0.037 136 5 0.016 0.014 0.085 Guinea-Bissau 0.071 14 1 0.069 0.000 0.335 Venezuela 0.091 11 1 0.087 0.000 0.399 Botswana 0.100 10 1 0.095 0.000 0.426 El Salvador 0.103 58 6 0.040 0.045 0.211 Argentina 0.106 132 14 0.027 0.063 0.171 Mexico 0.108 37 4 0.051 0.037 0.253 Panama 0.118 34 4 0.055 0.041 0.272 Nicaragua 0.119 42 5 0.050 0.047 0.255 Cape Verde 0.125 8 1 0.117 0.001 0.492 Guatemala 0.136 59 8 0.045 0.068 0.248 Ecuador 0.167 60 10 0.048 0.091 0.282 Honduras 0.175 40 7 0.060 0.084 0.323 Uganda 0.188 16 3 0.098 0.058 0.438 Angola 0.190 21 4 0.086 0.071 0.406 Peru 0.195 87 17 0.043 0.125 0.292 Malawi 0.200 35 7 0.068 0.097 0.362 Paraguay 0.213 75 16 0.047 0.135 0.320 Bolivia 0.222 63 14 0.052 0.136 0.340 Rwanda 0.250 4 1 0.217 0.034 0.711 Gambia 0.286 7 2 0.171 0.076 0.648 Tanzania 0.364 22 8 0.103 0.196 0.571 Niger 0.389 18 7 0.115 0.202 0.615 Cameroon 0.429 7 3 0.187 0.158 0.750 Guinea 0.556 9 5 0.166 0.266 0.812 Congo, Dem. Rep. 0.625 8 5 0.171 0.304 0.865 Mauritania 0.700 10 7 0.145 0.392 0.897 Burundi 1.000 1 1 0.000 0.167 1.000 Notes: Standard errors and confidence intervals were calculated pooling all data togethere and assuming that the request for bribes follows a binomial distribution. 29 Appendix Table A.5 Graft Index of Firm Transactions Import Licenses (Probability that a firm will be asked for bribes when requesting an import license) Number of Number of 95% Confidence Interval transactions bribes Country Index recorded requested Standard error Lower bound Upper bound Rwanda 0.000 11 0 0.000 0.000 0.300 Namibia 0.000 22 0 0.000 0.000 0.175 Malawi 0.000 36 0 0.000 0.000 0.115 Guinea-Bissau 0.000 4 0 0.000 0.000 0.546 Botswana 0.000 19 0 0.000 0.000 0.198 Colombia 0.009 117 1 0.009 0.000 0.052 Guatemala 0.009 113 1 0.009 0.000 0.053 Chile 0.013 75 1 0.013 0.000 0.079 Mexico 0.019 52 1 0.019 0.000 0.111 El Salvador 0.021 94 2 0.015 0.001 0.079 Peru 0.031 32 1 0.031 0.000 0.171 Uruguay 0.034 87 3 0.020 0.008 0.101 Argentina 0.041 98 4 0.020 0.013 0.104 Panama 0.042 24 1 0.041 0.000 0.219 Venezuela 0.045 22 1 0.044 0.000 0.235 Burundi 0.050 20 1 0.049 0.000 0.254 Nicaragua 0.051 59 3 0.029 0.012 0.145 Bolivia 0.070 71 5 0.030 0.027 0.158 Niger 0.083 60 5 0.036 0.032 0.185 Burkina Faso 0.091 11 1 0.087 0.000 0.399 Uganda 0.097 31 3 0.053 0.026 0.257 Paraguay 0.098 92 9 0.031 0.050 0.178 Ecuador 0.111 199 22 0.022 0.074 0.162 Mauritania 0.111 9 1 0.105 0.000 0.457 Tanzania 0.125 32 4 0.058 0.044 0.287 Honduras 0.133 45 6 0.051 0.059 0.266 Angola 0.176 17 3 0.092 0.054 0.418 Gambia 0.200 10 2 0.126 0.046 0.521 Swaziland 0.231 13 3 0.117 0.075 0.509 Cape Verde 0.286 7 2 0.171 0.076 0.648 Guinea 0.286 7 2 0.171 0.076 0.648 Cameroon 0.459 37 17 0.082 0.310 0.616 Congo, Dem. Rep. 0.846 13 11 0.100 0.565 0.969 Notes: Standard errors and confidence intervals were calculated pooling all data togethere and assuming that the request for bribes follows a binomial distribution. 30 Appendix Table A.6 Graft Index of Firm Transactions Operating Licenses (Probability that a firm will be asked for bribes when requesting an operating license) Number of Number of 95% Confidence Interval transactions bribes Country Index recorded requested Standard error Lower bound Upper bound Cape Verde 0.000 5 0 0.000 0.000 0.489 Namibia 0.000 39 0 0.000 0.000 0.107 Panama 0.000 26 0 0.000 0.000 0.152 Burkina Faso 0.000 5 0 0.000 0.000 0.489 Rwanda 0.000 14 0 0.000 0.000 0.251 Chile 0.031 129 4 0.015 0.009 0.080 Paraguay 0.031 32 1 0.031 0.000 0.171 El Salvador 0.037 82 3 0.021 0.008 0.106 Botswana 0.038 78 3 0.022 0.009 0.112 Colombia 0.049 82 4 0.024 0.015 0.123 Malawi 0.049 61 3 0.028 0.011 0.140 Nicaragua 0.054 112 6 0.021 0.022 0.114 Argentina 0.057 53 3 0.032 0.013 0.160 Uruguay 0.061 33 2 0.042 0.007 0.206 Honduras 0.081 161 13 0.021 0.047 0.134 Niger 0.091 33 3 0.050 0.024 0.243 Gambia 0.091 22 2 0.061 0.013 0.290 Uganda 0.098 204 20 0.021 0.064 0.147 Angola 0.098 61 6 0.038 0.042 0.202 Mexico 0.114 35 4 0.054 0.039 0.265 Bolivia 0.124 121 15 0.030 0.075 0.196 Tanzania 0.127 79 10 0.037 0.068 0.219 Venezuela 0.132 38 5 0.055 0.053 0.278 Peru 0.132 121 16 0.031 0.082 0.205 Guatemala 0.152 33 5 0.062 0.062 0.314 Ecuador 0.162 136 22 0.032 0.109 0.233 Burundi 0.200 10 2 0.126 0.046 0.521 Guinea-Bissau 0.222 27 6 0.080 0.103 0.411 Swaziland 0.227 22 5 0.089 0.097 0.439 Mauritania 0.250 4 1 0.217 0.034 0.711 Guinea 0.444 27 12 0.096 0.276 0.627 Cameroon 0.467 92 43 0.052 0.369 0.569 Congo, Dem. Rep. 0.667 48 32 0.068 0.525 0.784 Notes: Standard errors and confidence intervals were calculated pooling all data togethere and assuming that the request for bribes follows a binomial distribution. 31 Figure 1 Graft Index of Firm Transactions (GIFT) Probability that a firm will be asked for bribes when undertaking any of six transactions Namibia Uruguay Chile Colombia El Salvador Rwanda Botswana Argentina Panama Mexico Guatemala Nicaragua Bolivia Peru Venezuela Burundi Uganda Burkina Faso Malawi Honduras Angola Swaziland Paraguay Cape Verde Guinea-Bissau Ecuador Tanzania Gambia Niger Guinea Cameroon Mauritania Congo, Dem. Rep. 0 .2 .4 .6 GIFT SOURCE: Authors calculation based on World Bank's Enterprise Surveys data. 32 Figure 2 Graft and Excessive Regulation snoict brfodeksa )seibr ZAR .6 MRT CMR sa GIN an .4 Tr mriF gineb ofx m fir denIt fo a .2 GMB NER CPV TZA GNB afrG ilitybaborP( SWZ PRYECU AGO UGA MWI HND BFA PER VEN BDI MEX NIC GTM BOL ARG CHL 0 BWAURY SLVPAN COL RWA NAM .3 .4 .5 .6 .7 .8 Ease of Doing Business (Percentile Rank - Higher is worse) SOURCE: Authors calculation based on Doing Business 2007 and Enterprise Surveys. 33 Figure 3 Graft and Excessive Licensing Requirements Operating Licenses Construction Permits 1 1 BDI ess )seibrbrof .8 sit )seibrbrof .8 MRT enciL ksagine de ZAR .6 mreP deksa ZAR .6 gni GIN at b noitcurt gineb erpO- CMR rmif GIN sn CMR .4 TFIG afoy .4 Co- rmif afo NER TZA ilitbaborP( GMB MRT RWA SWZ BOL .2 GNB BDI .2 HNDUGA MWI PER PRYAGO ECU GTMECU GTM PER MEX TZA VEN BOL FTIG yilitbaborP( MEX PANNICCPV SLV ARGBWA UGA VEN GMB AGO HND NER GNB NICMWIARG COL URY CHL CHL BWA SLV PRY 0 0 SWZ NAM URY COL BFA PAN RWA NAM BFA CPV 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1 Starting a Business Dealing with Licenses (Percentile Rank - Higher is worse) (Percentile Rank - Higher is worse) SOURCE: Authors calculation based on Doing Business 2007 and Enterprise Surveys. 34