Deals and Delays: Firm-level Evidence on Corruption and Policy Implementation Times

Can corruption accelerate policy implementation? This question is part of the larger debate regarding the role of corruption in private sector development. Most studies conclude that corruption retards economic growth by distorting incentives, increasing transaction costs and aggravating uncertainty, leading to misallocation and underinvestment (Murphy et al. 1991; Shleifer and Vishny 1993; Rose-Ackerman 1997; Mauro 1995; Fisman and Svensson 2007). Yet some have argued that corruption can enhance efficiency by enabling entrepreneurs to circumvent burdensome business regulation or by incentivizing bureaucrats to work harder with bribes working as a piece rate (Leff 1964; Huntington 1968; Lui 1985; Lein 1986).

This paper revisits this issue by examining the relationship between requests for bribes and the time it takes to complete various regulatory requirements—getting construction permits, operating licenses, electrical connections, and clearing customs—using firm-level data from the World Bank Enterprise Surveys. More specifically, the paper tests three predictions implied by the “grease the wheels” hypothesis, which postulates that (i) firms confronted with bribes should get things done faster ceteris paribus. It also predicts that this relationship is heterogeneous both across firms and countries, with (ii) firms with a higher opportunity cost of waiting being willing to pay more and consequently facing shorter wait times, and (iii) bribing being more beneficial when regulation is onerous, such that the relationship between bribing and wait times should be starker in countries with more protracted de jure regulations.

In spite of extensive theoretical analysis of the relationship between regulation, corruption, and bureaucratic efficiency (see, e.g., Rose-Ackerman, 1975, 1999; Cadot 1987; Schleifer and Vishny 1993; Acemoglu and Verdier 2000; Aidt 2003), examining the relationship between corruption and the efficacy of policy implementation itself has largely been neglected to date. Previous studies using firm-level data have studied the determinants of the incidence of corruption (Svensson 2003; Clarke and Xu 2004; Cai, Fang, and Xu 2011; Ayyagari, Demirguc-Kunt, and Maksimovic 2014) and its (indirect) impact on firm performance in terms of employment and sales growth (Fisman and Svensson 2007; Aterido, Hallward-Driemeier, and Pages 2011) as well as investment (O' Toole and Tarp 2012), but not on the duration of policy implementation associated with the provision of particular government services. A notable exception is a study by Kaufmann and Wei (1999), which demonstrated that firms that pay bribes tend to spend more, not less, time with government officials overall. While suggestive, their paper neither examines actual implementation times nor controls for firm-fixed effects. Moreover, it does not systematically assess heterogeneity, which is an important but often overlooked corollary of the “grease the wheels” hypothesis.

Policy implementation times directly affect firm performance and are a very suitable metric for assessing government effectiveness both within and across countries (which perhaps explains why they feature prominently in cross-country comparisons of regulations such as the World Bank's Doing Business database). In addition, there is ample evidence that the impact of regulation on economic outcomes is contingent on its implementation, and that protracted policy implementation impedes trade (Djankov et al. 2010), job creation (Freund and Rijkers 2014), firm entry (Klapper, Laeven, and Rajan 2006), productivity (Nicoletti and Scarpetta 2003), and growth (Djankov et al. 2005).

The World Bank Enterprise Surveys are well suited for analyzing the association between bribing and policy implementation duration. In addition to containing detailed information on firm characteristics, the surveys include questions on a set of government services or permits that might have been sought, whether extra payments or gifts were requested, and the time it took to complete the process. Having data on multiple transactions for the same firms not only allows for the testing of the robustness of results across services, it also allows for controlling for firm fixed effects, such that identification is based on variation in the incidence of bribe requests across different types of permit applications made by the same firm. The data also enable us to assess heterogeneity by means of difference-in-difference designs and quantile regressions which allow for a differential impact of bribing across the distribution of policy implementation times.

Previewing our main findings, the results of this paper consistently demonstrate that demand for bribes are not associated with accelerated implementation and instead tend to be associated with protracted policy implementation. According to our preferred estimates, firms confronted with demands for bribes or expectations that a gift must be paid on average take 1.5 times as long to get a construction permit, operating license or electrical connection ceteris paribus, and 1.2 and 1.4 times as long to clear customs for exporting and importing respectively. The positive association between demands for bribes and the duration of policy implementation remains when controlling for firm and entrepreneurs' (unobserved) characteristics, as well as using alternative bribe proxies, and helps explain the remarkable heterogeneity in de facto policy implementation times demonstrated by Hallward-Driemeier and Pritchett (2011).

The data are also at odds with the second and third testable predictions of the grease hypothesis, notably heterogeneity in the relationship between demands for bribes and policy implementation times across firms as well as across countries. Firms that are likely to pay higher bribe amounts do not report shorter delays when confronted with a request for a bribe ceteris paribus. Moreover, quantile regressions demonstrate that the association between bribes and delays holds across the distribution of policy implementation times; even for firms that get things done relatively quickly bribe demands are associated with delays. If those firms would have been able to get things done faster by paying bribes, one would have expected bribe demands to be associated with expedited implementation instead.¹ Last but not least, we find no evidence for the proposition that corruption enhances efficiency when the de jure business environment is burdensome.

It should be noted that we cannot establish the direction of causation.² Even so, the association between bribes requests and delays is consistently significantly positive, which is at odds with the grease the wheels hypothesis predicting bribes would be associated with accelerated policy implementation times.

The results are consistent with a world where some government officials intentionally delay government services to extract side payments. However, this could not be a stable equilibrium if targeted firms cognizant of the possibility of corruption were able to offer to speed up the process by offering bribes earlier on in the process; waiting can only be an optimal equilibrium outcome in the presence of information frictions.

The remainder of this paper is organized as follows; the next section reviews related literature, the second section then distils testable predictions from it and explains our econometric strategy. The third section describes the data. Results are presented in section four. A final section concludes.

I. Conceptual Considerations and Econometric Strategy

Two competing hypotheses underlie the relationship between corruption and its effects on firm outcomes. Testable predictions are drawn from these hypotheses and motivate the estimation strategy.

Competing Hypotheses: “Grease vs Sand”

Whether and if so, when, corruption enhances efficiency continues to be the subject of significant controversy (see, e.g., Bardhan [1997] and Svensson [2005] for reviews of the literature). While the cross-country correlation between corruption, underdevelopment and underinvestment has been appealed to as evidence for the conventional wisdom that corruption is typically detrimental to economic performance (Mauro 1995), a substantial body of literature has been devoted to the idea that there are instances in which corruption may enhance welfare. A particularly prominent view is that in a second-best world corruption can compensate for the consequences of deficient institutions, which is at the root of the “grease the wheels” hypothesis stipulating that corruption enhances efficiency and especially so in environments characterized by bad bureaucracies and poor policies (Méon and Sekkat 2005; Méon and Weill 2010), even if some resources have to be invested into corrupt activities.

The efficiency enhancing impact of corruption can work through various channels. To start with, corruption may limit the adverse effects of inefficient regulations (Leff 1964). In addition, the opportunity to receive side payments may enhance the quality of bureaucrats by raising their effective compensation and provide incentives to speed up service times in sluggish bureaucracies if bribes work as a piece rate. Moreover, corruption may enhance allocative efficiency if bribery is competitive and bureaucrats prioritize serving those with the highest willingness to pay, who presumably derive the greatest surplus from being served. The bribery models of Beck and Maher (1986) and Lien (1986), for example, replicate the efficiency outcomes of competitive bidding.

In regards to policy implementation duration, the “grease the wheels” hypothesis predicts that ceteris paribus (i) bribe paying agents are likely to “get things done” faster. An important but often neglected implication of the “grease the wheels” hypothesis is heterogeneity in the relationship between bribes and implementation times, both across firms and countries. To start with, it predicts that (ii) firms with the highest opportunity cost of waiting, or the lowest ability to evade bureaucratic harassment, should be expected to be more willing to pay to avoid delays.³ Firms with deeper pockets would also be more able to pay; the speed of policy implementation would likely increase with the size of the bribe payment. Moreover, (iii) the beneficial impact of corruption might be expected to be larger the more burdensome de jure regulations are. Where the formal requirements for compliance allow for long processing times and multiple procedures, the scope for speed money should be all the greater.

These efficiency arguments, however, may fail if officials inflict deliberate administrative delays to extract bribes (as suggested by Myrdal 1968). This ability to target specific firms to induce them to pay bribes is at the heart of the “sand in the wheels” hypothesis which predicts that bribe demands will be associated with longer delays. Kaufmann and Wei (1999), for example, present an endogenous model of bribery and harassment which they model as a Stackelberg game in which the official moves first and chooses an optimal level of red tape to maximize bribe intake. Firms act as price-takers maximizing profits. The model predicts a positive, rather than a negative, association between bribery and red tape because bureaucrats will impose delays based on firms' ability to pay, even though bribe payments themselves reduce effective red tape conditional on bureaucratic harassment.

However, this “sand in the wheels” hypothesis is not entirely compelling since one might anticipate that informed firms cognizant of the possibility that bureaucrats will impose delays on them might offer to pay bribes sooner to avoid additional delays. In addition, the “sand” hypothesis assumes bureaucrats are somehow able to price discriminate and detect which firms are willing to pay more than others. Cadot's (1987) model of the allocation of permits by officials systematically analyses the implications of different informational assumptions; when information is perfect and agents are perfectly informed about each other's type and discount rate, a unique separating equilibrium emerges in which agents benefitting from bribing, pay instantly when confronted with dishonest officials. However, uncertainty about the characteristics of the bureaucrat and the entrepreneur can result in multiple equilibria, in which delays serve as useful signals about one's type, akin to the literature on stalling in bargaining (Crampton 1992; Kennan and Wilson 1993; Babcock and. Loewenstein 1997; Abreu and Gul 2000; Thanassoulis 2005).

Thus, the main challenging in discriminating between these competing “grease” and “sand” hypotheses is identification of counterfactual outcomes. One would ideally run an experiment in which identical firms varying only in their willingness to bribe were exposed to different bureaucrats varying only in their propensity to engage in corrupt practices (and perhaps the de jure regulations which they are supposed to enforce) in the exact same location at the exact same time, as this would provide a plausible identification of counterfactual outcomes. With the available cross-sectional data, we unfortunately cannot establish causality. We do not have precise information under what circumstances bribe proposals are being initiated, by whom, and when in the process the requests are made, and lack information on firms' and bureaucrats' subjective expectations. This especially limits our ability to test specific predictions of the “sand in the wheels” hypothesis which relies on strong informational assumptions on the part of the bureaucrat as (s)he is assumed to engage in price discrimination.

Nonetheless, the data contain rich detail on both bribes and wait times and enable us to control for an extensive set of observable determinants of corruption. The data thus allow for testing of various predictions of the “grease the wheels hypothesis,” according to which bribes are associated with expedited implementation and according to which the relationship between bribes and policy implementation times should be heterogeneous across firms, with the extent to which bribery accelerates implementation times being correlated with both firms' willingness to pay and the severity of regulation. By contrast, the “sand” hypothesis predicts a positive association between bribes and wait times. Not finding such an association would amount to a rejection of this hypothesis.

Econometric Framework

Our empirical strategy is to sequentially test the three predictions of the “grease” hypothesis discussed above. The next section discusses how we test the first prediction that demands for bribes ought to be associated with reduced implementation times ceteris paribus, and presents a number of robustness tests including a set of regressions in which we control for firm fixed effects as well as bribe proxies that are less vulnerable to endogeneity. The following section explains how we test for heterogeneity in the association between bribe demands and policy implementation times, which underpins the second and third testable predictions of the “grease” hypothesis. It presents both difference-in-difference designs to test for heterogeneity associated with the magnitude of bribe payments and the severity of regulation, as well as quantile regressions which examine differences across the distribution of wait times.

If bribes are a means to avoid burdensome business regulation, one would expect their incidence to be correlated with shorter wait times (i.e. $βB<0$⁠). Conversely, if corruption impedes policy implementation, one might expect bribe incidence to be associated with delays $(i.e.βB>0)$⁠. Note that to account for cases in which firms reported to have waited less than a day, we use the log of 1 plus the number of days it took to get service as the dependent variable.

• HO: $βB=0$ Corruption is not correlated with policy implementation times

• HA1: $βB>0$ Corruption is associated with delayed policy implementation (“sand”)

• HA2: $βB<0$ Corruption is associated with expedited policy implementation (“grease”)

There are a number of potential problems with this testing strategy. The relationship could be driven by omitted variables impacting both the duration of policy implementation and the prevalence of corruption. This could be the case, for example, when bribing only occurs in certain circumstances, such as when excess demand for particular services leads to congestion. This could be at a sector level, to the extent requirements vary across sectors, or geographically if many firms in the same location seek the same license or permit at the same time. An alternative example of a “third factor” that could cause wait times and bribery to be (spuriously) correlated is complexity; it is feasible to imagine more complex projects being both more susceptible to corruption and facing longer policy implementation times.

Second, there is a possibility of reverse causation; one might worry that firms might be more likely to offer a bribe if they have had to wait for a while. Although the bribe request proxy we use asks specifically about demands for bribes and expectations thereof—and not about bribe offers initiated by firms— respondents could nonetheless confuse the two. Alternatively, one might expect a positive correlation between policy implementation times and bribe incidence, simply because entrepreneurs who have to wait longer are at heightened risk of being asked for bribes.

The use of subjective perceptions of bribe incidence and wait times aggravates these concerns. It is possible that unobserved firm characteristics, such as the entrepreneur's innate pessimism, affect both perceptions of bribe incidence and policy implementation times. Alternatively, firms that face longer delays to get things done might become more dissatisfied with regulatory agencies and consequently more likely to complain about corruption (even if there was none).

To mitigate these endogeneity concerns, we conduct a number of robustness checks. To start with, we include sector-country-year effects; if licenses are sector-specific, this would control for congestion within sectors. As a robustness check, we also allow for city-sector-year effects to control for city-sector specific congestion, and the patterns remain.⁴

Second, we pool applications for different types of government services by the same firm and control for firm fixed effects. This should control for characteristics of the firm that could make it a target for bribes, as well as the complexity of the projects the firm is undertaking and needing permits for. Identification now comes from variation in the incidence of bribing across different type of government services provided to the same entrepreneur.

Third, we also address the potential for endogeneity from the use of firms' subjective measures of bribes. In addition to firms' own responses, we use alternative proxies such as the leave-out-mean of firms in the same sector, same country and same size class. Since this measure is not “polluted” by entrepreneurs' idiosyncratic characteristics, it likely mitigates endogeneity bias (Escribano and Guash 2005).

We also construct a measure of firms' latent pessimism. Following Van Praag and Ferrer-I-Carbonnel (2004), we first run a number of ordered probit models of firm's perceptions of the severity of various constraints not directly connected with obtaining a construction license (and clearing customs, getting an operating license or an electrical connection) on firm size, sector, country, and year dummies. Subsequently, a principal component analysis is performed on the residuals of these regressions. The first principal component is used as a proxy for entrepreneurs' latent pessimism. Relatedly, firms were asked how severe a constraint they consider corruption. To the extent that this is a proxy for their likelihood to complain and exaggerate policy implementation times, one might expect that when confronted with bribes, such firms would be particularly likely to exaggerate. We will show that this is not the case.

Accounting for Heterogeneity

If corruption is particularly efficient grease when rules are more burdensome one would expect $βBR$ to be negative.

An alternative means to examine whether OLS regressions, which provide estimates of average effects, obscure heterogeneity, whereby some firms (perhaps those that do not pay) suffer from corruption, but others (perhaps those who do pay) benefit, we run quantile regressions, using as dependent variable policy implementation times demeaned by the country-sector-year average, such that the resulting estimates are not driven by cross-country differences. If firms that get things done rapidly do so because they pay bribes, one would expect the coefficient associated with bribe demands to be negative at lower parts of the distribution.

II. Data

The World Bank Enterprise Surveys are the main data source used in this article. The surveys have been conducted in 107 countries and contain detailed information on firm characteristics, such as their sector, location, sales, capital stock, size, ownership structure, management, and their manager's subjective perceptions of the business climate and obstacles to doing business.⁶ Figure 1 shows how entrepreneurs rated the severity of various constraints, and demonstrates that corruption is a salient concern, with 41 percent of firm managers considering it a major or severe constraint—the most common of the potential constraints to be ranked as such alongside electricity access.⁷

The data also contain information on how long it took firms to obtain construction permits, operating licenses, electrical connections, and to clear customs for importing and exporting. These self-reported policy implementation times will be our key dependent variables. The surveys furthermore asked entrepreneurs whether “In relation to that application for a construction permit [operating license/electrical connection/clearing customs] was an informal payment or gift expected or requested?” The answers to these questions will be our key explanatory variables. The descriptive statistics presented in Table 1 suggest that demands for such side payments are prevalent; 19, 17, and 14 percent of firms reported having been either expected to, or explicitly asked to, pay a bribe when applying for a construction permit, operating license, electrical connection, respectively. In a subset of Middle Eastern and Northern African countries, firms were also asked about whether they were asked for bribes when dealing with customs, where corruption appears even more rife. As many as 38 and 44 percent of firms in our data reported having been asked to pay a bribe when dealing with customs to export and import respectively.

We complement the data on de facto policy implementation times with information on the de jure time it should take to get things done from the Doing Business Indicators. Figure 2 plots the 10th percentile, median, and 90th percentile de facto log policy implementation times in a given country-year reported by entrepreneurs surveyed by the World Bank Enterprise Surveys against the amount of time it should take to get these things done according to the Doing Business indicators.⁸ A striking feature of the data is the tremendous variability in policy implementation times, which is only loosely correlated with the amount of time it should take to get these things done. Interestingly, within-country variability appears to dominate cross-country variability (see also Hallward-Driemeier and Pritchett 2011).

That this heterogeneity is related to the incidence of demands for bribes is demonstrated in Figures 3a–3c, which plot kernel density estimates of the amount of time it takes to get various things done, demeaned by the country-sector-year average in the relevant year, separately for firms that report being asked for bribes and ones that do not. While there is substantial overlap across these distributions, the distribution for the firms that indicated being confronted with bribes tends to lie to the right of that of firms that were not. Bribe requests thus appear to be associated with protracted policy implementation on average. Also note that the kernel density graphs for firms reporting bribes are unimodal. If corruption benefitted a few privileged firms but not others, one might expect to see a bimodal distribution.

Thus, prima facie, the data show that corruption is a salient concern for entrepreneurs, that policy implementation times vary across firms and are at best loosely correlated with de jure policy implementation times, and, moreover, that firms that were confronted with demands for bribes on average take longer to get things done.

III Results

After showing which firms are asked to pay bribes, the basic results between requests for bribes and wait times are reported, along with several robustness checks.

Which Firms Are Asked for Bribes?

To set the stage for the analysis that follows, we first examine which firms are most likely to be confronted with bribe requests by running a simple probit model with a binary indicator of whether or not a firm was asked for a bribe when applying for a construction permit as the dependent variable. In the first specification, presented in Table 2, we only include controls for firm size and its square, as well as country-sector-year dummies. Inclusion of these dummies is motivated by the desire to compare firms that are as similar as possible; identification comes from comparing firms within the same country within the same sector who applied for a permit the same year. In the second specification we add controls for capital intensity proxied by capital per worker, the log of the age of the firm, ownership dummies, whether or not the firm exports, and the manager's experience, education, and gender. These dummies proxy for managerial competence and potential differential treatment by gender. The third column adds output per worker as an additional control variable to assess whether more productive firms are more likely targets. Since this variable is potentially endogenous (more productive firms may have more resources to bribe and/or corruption may make firms more productive), it is entered separately. Finally, we examine the relationship between bribe requests, firm visibility, and level of interaction with public officials, by controlling for whether or not a firms is International Organization for Standardization (ISO) certified, formal, has attempted to secure a government contract in the past year, was visited by tax officials, and the share of the manager's time that is spent dealing with regulations, variables that are all potentially endogenous. Since not all explanatory variables are available for all firms, the sample gets progressively smaller as we add explanatory variables.⁹ Standard errors are clustered by country and year.

Overall, these models explain a modest share of the variance, with pseudo R²s ranging from 0.145 to 0.169. The results yield a number of interesting associations. To start with, larger firms are more likely to be asked for bribes. Second, international orientation matters; the marginal effects calculated at the sample mean suggest that foreign owned firms are ceteris paribus approximately 5 to 7 percent less likely to be confronted with bribe requests, whereas exporting firms are some 6 to 7 percent more likely to be asked for a bribe.¹⁰ Third, there is a statistically significant association between bribe demands, time spent with government officials, and inspections. Firms that were visited by tax officials are ceteris paribus 9.5 percent more likely to have been asked for a supplemental payment or gift. Note that the other variables are not systematically statistically significant predictors of bribe incidence. In particular, we do not find evidence for differential treatment of firms based on the characteristics of the manager. We also do not find evidence for the notion that more capital intensive firms are easier targets, though this finding might (in part) reflect the quality of our capital stock proxy.

Basic Results of Bribe Requests and Implementation Times

Now that we have demonstrated which firms face the highest incidence of bribe requests, we turn to the results of focal interest, notably the relationship between bribe requests and policy implementation times. This section focuses on examining whether bribe demands are correlated with accelerated implementation on average, whereas the next section analyzes whether there is heterogeneity in the relations between bribe demands and policy implementation times depending on firm's ability to pay and the complexity of the de jure regulatory environment.

The baseline results are presented in Table 3, which regresses the self-reported time it took to obtain a construction permit on a host of firm-characteristics. The specifications broadly mimic those presented in Table 1; all specifications include country-sector-year dummies to ensure identification is based on comparing firms within the same sector and country in a given year. The second specification includes additional characteristics for the size of the firm and its square, which proxy inter alia for the firm's visibility and the potential complexity of the requested permit, its age, capital intensity, ownership structure, whether or not it exports, and characteristics of the manager. The third and preferred specification adds output per worker as a proxy for productivity and firm's ability to pay, while the fourth adds controls for visibility and proxies for the intensity of interaction with government officials. Standard errors are again clustered by country and year.

Starting with the results of focal interest, requests for bribes are associated with strongly and significantly protracted implementation times. This association is statistically significant at the 1 percent level, robust to including various proxies for various firm characteristics (column 2), firm productivity (column 3), as well as proxies for visibility and interaction with public officials (column 4). Moreover, the associated effect is large and economically meaningful; on average, firms that are confronted with bribes have to wait approximately 1.5 to 1.8 times as long for their construction permit as firms that were not asked for bribes; note that we take the exponent of the coefficient associated with bribe demands to arrive at these comparisons.

Size is also correlated with policy implementation,¹¹ with larger firms facing especially long delays. However, none of the other variables are consistently statistically significant, individually or jointly; neither government nor foreign owned firms, nor exporters benefit from expedited policy implementation all else being equal. Younger firms appear more likely to wait longer, although the association between firm age and wait times is not always statistically significant. Overall, these specifications explain between 32 and 36 percent of the observed variance across firms.

Robustness

Table 4 repeats the exercise for other regulatory transactions, using as the base specification the one used in Table 3 column 3, that is, with a rich set of firm characteristics including labor productivity. Standard errors are clustered by country and year. Column 1 reports the effects of a bribe request on the time it takes to receive an operating license, column 2 for an electrical connection, column 3 for clearing customs to export and column 4 for clearing customs for imports. In each case the effect of a bribe request is significantly and positively associated with increased delays. The coefficient is smaller for exports but still significant at the 5 percent level. Across all the different transactions, the same pattern emerges. For ease of interpretation, Table 5 summarizes the delays in implementation associated with bribe demands by taking the exponent of the coefficient on bribe demands. The coefficient estimates imply that firms confronted with a request for a bribe when applying for an operating license or electrical connection tend to wait 1.5 times as long as firms that were not confronted with such demands ceteris paribus. Firms confronted with demands for bribes when obtaining customs clearances have to wait 1.1 times as long as firms that were not asked for bribes when exporting and 1.4 times as long when importing. Thus, the delays associated with corruption are sizeable.

Table 6 then exploits the fact that we have information on multiple transactions per firm. By pooling the transactions together, we can control for firm fixed effects. Standard errors are clustered by firm-year. Note that firms that only applied for one type of government service are excluded from the estimation sample. We also exclude firms that made multiple applications but either reported never to have been asked for a bribe or that reported to have been confronted with demands for bribery in each instance since such firms do not exhibit variation in bribe propensity across applications; Identification now comes from differences within firms across the transactions and whether bribes were requested or not. This approach also helps control for differences in the susceptibility of firms to bribe request, potential complexity of the activities firms are involved in and other (unobserved) time-invariant characteristics correlated with bribing that might otherwise also explain wait times. This is a more stringent test that the earlier approaches. The pattern of results remains robust.

We further test the robustness of the results using the Z-score of the delay times (calculated by country sector and year) in column 2 of Table 6. This controls for the differences in the distribution of wait times across the various types of transactions. We also test additional measures that help control for potential endogeneity. Column 3 reports the results from using the ‘leave out the mean’ measure of bribes. And finally, column 4 expands the set of fixed effects to include application-country-year effects rather than separate dummies for each type of application. The results are robust: bribe requests continue to be correlated with significantly slower implementation. Also note that the estimated delay in policy implementation associated with bribe requests is similar in magnitude to those documented in Table 5; conditional on firm fixed effects and service type, bribe demands are associated with an increase in service time by a factor of 1.5 (column 3) to 1.8 (column 1).

Table 7 presents specifications that assess the robustness of the striking association between bribes and slower policy implementation, using specifications very similar to those presented in column 3 of Table 3 but with some modifications. First, the specification presented in column 1 uses the number of days in levels as opposed to logarithms as the dependent variable. The results are qualitatively robust and the estimates imply that firms confronted with demands for bribes have to wait 33 days longer to get a construction permit than firms that were not ceteris paribus.

Second, the specification presented in column 2 includes an additional control for firms' latent pessimism.¹² Although this indicator enters statistically significantly at the 1 percent level suggesting pessimistic entrepreneurs are more likely to report longer wait times, the coefficient associated with bribe incidence remains positive, large and statistically significant.

Third, column 3 includes as an explanatory variable the self-reported severity of corruption as a constraint to doing business. This variable enters statistically significantly, but its inclusion hardly impacts the coefficient estimate associated with bribe requests.

Fourth, to assess whether geographic congestion might be driving the results, in column 4 we include city-sector-year (as opposed to country-sector-year) dummies and the results remain robust.

In summary, although we cannot establish causality, the association between protracted policy implementation and demands for bribes or expectations thereof appears fairly robust across the various robustness checks.¹³

IV Heterogeneity

Now that we have established that on average bribe demands are associated with delays, we assess whether the data demonstrate heterogeneity in the relationship between bribe incidence and speed of policy implementation.

Table 8 explores whether the association between bribe incidence and policy implementation is heterogeneous across firms. All specifications control for firm characteristics and country-sector-year dummies unless otherwise indicated. To start with, in column 1 the bribe indicator is interacted with output per worker demeaned by the country average to assess whether the impact of bribes on policy implementation times varies with firm productivity (perhaps because they are more willing to pay). Although the estimated interaction effect is positive, it is not statistically significant; the null hypothesis that more productive firms are not able to get things done faster cannot be rejected.