Corruption Accusations and Bureaucratic Performance: Evidence from Pakistan


 This study assesses the effectiveness of anti-corruption policies of Pakistan by relating the corruption of government officials to the actual and perceived bureaucratic hurdles faced by formal businesses. It offers a unique perspective by focusing on cases in which the accused officers voluntarily disclosed the misuse of public money (or gains acquired through corruption) in order to avail the option of plea-bargaining. The empirical analysis estimates the effect of these policies on the responses of the managers of business firms in Enterprise Surveys. The number of accusations seems to reduce the incidence of bureaucratic corruption. While the amount that is being offered in bargaining has a negative effect overall, it switches sign overtime suggesting the possibility that it may increase corruption in the long run. The core results are robust across various measures of firm-level corruption. The instrumental variable estimates produce similar results.


Introduction
A potent cause behind the misallocation of resources in developing countries is the misuse of authority and public funds by government officials. Particularly, at the microeconomic level, bureaucratic holdups and delays are not only ethically inappropriate but also cause a significant drag on private businesses. Unsurprisingly, the efforts to conquer corruption are not confined to regulatory reforms and incentive mechanisms but also include harsh punishments like even the death penalty.¹ Ultimately, a society's tolerance for corruption, and the choice and effectiveness of anti-corruption measures are context specific and require careful analysis (Graaf, 2007).
The fight against corruption is usually operationalized through independent anti-corruption agencies. However, it is difficult to separate the good from bad intentions in these campaigns as those in public offices have every incentive to belittle the political statures and reputations of their rivals through fabricated cases.
In recent decades, aggressive anti-corruption campaigns have been a feature of many developing countries. Thus, a number of high-profile politicians and ex-public officials were convicted in South Korea.² In China, there was a drive to purge the ruling Communist Party of corrupt officials. Corruption has been cited as one of the main factors in the decline and fall of the Indian National Congress, the political party that ruled India for most of its post-independence history.³ In Pakistan too the issue has been a source of political uncertainty since 1990s when elected governments were dismissed in quick succession on grounds of corruption. In 1997, the Bureau of Accountability (called Ehtisab Bureau at that time) was established to formally investigate political corruption. In 2002, through various legal amendments, Ehtisab Bureau was replaced by the National Accountability Bureau (NAB). In the 11th Five-Year Plan (2013-2018 formulated by the Pakistan Planning Commission, corruption free governance is set as a national goal. The efforts of NAB against corruption are highly debated in Pakistan and have caused political polarization. During the period of this study (i.e. 2002-2013), NAB maintained a conviction rate of 51% and has decided 1,762 cases. Perhaps the most controversial aspect in the NAB's fight against corruption is the use of procedures like plea-bargaining and voluntary disclosure. These are accepted legal procedures in many other countries like Canada, the UK, the USA, and India.⁴ At least theoretically, these procedures cannot be dismissed as ineffective in the context of a developing country where judicial processes are painfully slow and replete with all types of inertia that favor the culprit. However, in the case of NAB the main criticism is that its policy allows easy escape to those involved in corruption by surrendering a proportion of the embezzled funds (currently 80%). Thus, without empirical analysis, the net benefits of the policy are unclear.
Using the case of Pakistan, this study gathers unique data about all the corruption cases involving government servants (GS) where the accused volunteered to return embezzled funds and was granted the option of plea bargaining by the court. With below potential rate of economic growth, and a slow but consistent adoption of democratic norms, the emerging political parties of Pakistan are sensitive to the administrative weaknesses and bureaucratic excesses of past regimes and have vowed to strengthen accountability mechanisms.
The empirical enquiry in this study focuses on the corruption cases involving GS and bureaucrats from all the four provinces of Pakistan. This data set is merged with the World Bank Enterprise Surveys (ES) data to see the effect of the number of plea-bargained cases and the amount used in the plea bargaining on the relevant responses of business managers. The responses of business managers to corruption-specific questions in the ES data set are interpreted here as a proxy for business-government relations. The study, therefore, helps us to understand the efficacy of the measures to curb corruption and how they affect the perceptions of firm managers.
The study fills out the following important gaps in the existing literature on anti-corruption policies in developing countries: • It contributes by assessing the efficacy of anti-corruption policies by using both the perceptions and experiences of firm managers as indicators of business-government relations. This adds to the existing studies that rely mostly on the public perception-based measures of corruption (e.g., Donchev & Ujhelyi, 2014). • Using information about actual corruption cases involving GS, it provides evidence that a tradeoff exists between recovering looted funds and the long-term objective of eliminating corruption. Thus, the strategies or incentives that promise easy and quick recovery of looted funds may increase corruption in the long term.
Using multiple questions related to bribery incidents in the ES data set, we discovered a negative correlation between the number of plea-bargained cases and the perception and experience of bureaucratic corruption of business firms. Interestingly, the effect of the size of the bargained amount is insignificant in baseline analysis. This finding is interpreted here as the relative importance of the punishment effect (of conviction) vis-á-vis the disincentive caused by surrendering of the gains. However, incorporating the possibility of time dependence of the effect, we find that the coefficient of the bargained-amount switches sign thus supporting the view that the availability of the plea-bargaining option may actually encourage, rather than discourage, corruption over time. This is in line with the existing finding that the policy of plea bargaining is bound to be abused (Aniche, Alumona, & Obiwulu, 2020).
Section 2 reviews the relevant literature. In Sections 3 and 4, we present data and empirical results. Section 5 concludes and notes the limitations of the research.

Related Literature
The economic approach to crime relies on the framework first formalized by Becker (1968). In this framework, rational agents prefer illegal exchanges if the expected net benefits from such transactions are positive. This elegant cost-benefit framework is extended to understand the underlying logic of white-collar crimes (Draca & Machin, 2015;Rose-Ackerman, 2010). Accordingly, a firm is more inclined to pay bribes if there are net benefits in avoiding complex and strict procedures, the unscrupulous court system and incompetent government service (Huang & Rice, 2012;Jiang & Nie, 2014;Wu, 2009). Thus, bribing and lobbying is an important way to avoid regulatory delays and competitive pressures (Oh, 2009;Tonoyan, Strohmeyer, Habib, & Perlitz, 2010;Zhou & Peng, 2012). In developing countries, formal sector firms view corrupt practices as "the ways things are done" (Brass, Butterfield, & Skaggs, 1998;Hallward-Driemeier & Pritchett, 2015).
The link between corruption and firm performance is influenced by the strategic activities managed by a firm in response to corruption (Galang, 2012). In the presence of regulations replete with bureaucratic hold ups, corruption may help businesses by greasing up the process of starting a new business (Dreher & Gassebner, 2013;Jiang & Nie, 2014;Wang & You, 2012)and by reducing delays in public service availability (Van Vu, Tran, Van Nguyen, & Lim, 2018;Zhou, Han, & Wang, 2013). Bribes, by winning political favors, can also serve as an instrument to increase company value (Faccio, 2006). In contrast, corruption has a negative impact on firm growth if it hurts competitiveness (Fisman & Svensson, 2007;Gaviria, 2002).
The negative effects of corruption on a firm's productivity are more prevalent in those countries where the legal framework is weak (De Rosa, Gooroochurn, & Görg, 2015). In contrast, Bai, Jayachandran, Malesky, and Olken (2017) have estimated a reverse effect, i.e., corruption reducing the effect of growth especially for firms who can easily relocate.
The characteristics of firms also play a role in determining their preference for informal payments. For instance, bigger firms are more likely to pay bribes to get government contracts and preferential tax benefits (Delavallade, 2012;Mishina, Dykes, Block, & Pollock, 2010;Rand & Tarp, 2012). However, the cost of corruption is higher for small firms if it taxes a large proportion of their resources (Zhou & Peng, 2012). Paunov (2016) argues that small firms are negatively affected by corruption if they are adhering to international quality standards. Dutta and Sobel (2016) suggest that bribery is not a "cost of doing business," rather corruption impairs entrepreneurship.
If corruption is a way to avoid complex administrative procedures, then voluntary disclosure and plea bargaining can also be interpreted as possible means to avoid complex and time-consuming legal procedures (Feeley, 1982). However, it is important to see how the legal option of plea-bargaining changes the usual calculus of crime by eliminating the possibility of punishment (or at least harsh punishment). For instance, Reingnaum (1993) argues that plea bargaining reduces the defender's expected cost and induces crime. Thus, plea bargaining dilutes the deterrence effect (Miceli, 1996).
There are studies supporting the deterrence effect associated with apprehension and detection risk (e.g., Barnum, Nagin, & Pogarsky, 2020;Becker & Stigler, 1974;Rose-Ackerman, 1978). On the other side, there exist evidence that a high risk of detection has a deterrence effect but kills the intrinsic motivation for honesty (Schulze & Frank, 2003). The net effect on overall corruption is undetermined.
Another factor is the impact of public perception on corruption control, as it shows how much people are willing to support anti-corruption policies (Li, Gong, & Xiao, 2015). Thus, any state seeking a positive outcome of an anti-corruption instrument should gain public support for strongly backing it up with policy enforcement.
By contrast, corruption in law enforcement limits the deterrence effect (Polinsky & Shavell, 2001). In a weak institutional environment, public officials misuse their power and distort institutions by reducing detection and punishment to facilitate their own embezzlements (Boly & Gillanders, 2018).
This strategy (of voluntary disclosure and plea bargaining) is known for its mixed implications (Aniche et al., 2020). Theoretically, it increases welfare by motivating the truly guilty to self-select themselves and contributes to the accuracy of the legal procedure (Grossman & Katz, 1983). On the negative side, it may help institutionalize corruption and in Pakistan it is heavily criticized on this aspect. Thus, it is necessary to see whether this strategy offers any benefits to society other than having a cost-effective route to recovering embezzled funds.
Framing the discussion in the context of Pakistan we find only limited scholarship on the causes and cures for corruption in Pakistan. For instance, Farooq, Shahbaz, Arouri, and Teulon (2013) argue that corruption hinders economic growth in the case of Pakistan. In contrast, the policies and strategies adopted by the NAB, the country's foremost anti-corruption agency, are not welcomed and debated extensively. For many, the drive against corruption is ineffective and good only for political ends (Ali, 2018). However, in the absence of scientific evidence such accusations are mere perceptions. Given the huge amount of resources at stake, it is necessary to check the extent to which the nabbing of corrupt officials temper economic outcomes.
Considering the above discussion, this inquiry intends to add two novel angles to the existing literature: (i) the study focuses on the variation in the responses of business managers to questions related to bureaucratic corruption. In particular, it estimates the effect of the number of past plea-bargained cases on the responses of business managers. It thus provides a broader assessment of the efficacy of the anti-corruption campaign; (ii) it also estimates the effect of funds recovered through plea bargaining on the outcome. Thus, by comparing the effect of recovered funds over time, it checks for the long-term consequences of the policies aimed at easyrecovery of looted public funds.
Benefiting from the above discussion we can formulate the following two hypotheses: H1 Nab effect hypothesis: A higher number of plea-bargaining cases in the past reduce the government official's incentive to indulge in corrupt transactions, ceteris paribus. H2 Bargain effect hypothesis: (H2.1) The possibility that one can bargain one's way out of corruption related charges by surrendering a part of the illicit gains may encourage corruption; (H2.2) the probability that one may have to surrender misappropriated funds in case of conviction may reduce the net expected gains from such transactions and thus reduce corruption.

Data
This study relies on two sources of information. The number of corruption cases involving government officials in each province of Pakistan is extracted from the NAB's annual reports. The second source is the World Bank Enterprise Surveys (ES) which provides information on business firms operating in the formal sector of Pakistan in its four provinces. The following paragraphs explain these data sets in detail (Tables 1-2).

NAB Data Set
This data set comprises corruption cases involving GS where, under some agreed conditions, the accused has voluntarily returned a percentage of the embezzled funds to the government. The information is gathered from the annual reports and documents available on the website of the NAB (http://nab.gov.pk/). (Table 3 at the end provides the distribution of cases involving government bureaucrats over the years in Pakistan.) A corruption case, in the context of this study, is defined as an instance of wrongdoing in which a government official is convicted for misappropriation of government resources for personal gains and the case is settled through the procedure of plea bargaining or voluntary  disclosure. The process of plea bargaining allows the convict to deposit an agreed upon amount in the government exchequer to avoid a jail term. The cases with pending status or with doubtful convictions (e.g., those involving political leaders) are not considered. Similarly, the cases where the individual availed the option of plea bargaining before the start of the enquiry are also ignored. The information is collected separately for all provinces for the period 2002-2013. Thus, in our data set, the province of Punjab has the largest number, i.e., 118, of plea-bargained cases. It means an average of 13 pleabargained cases per year. In terms of bargained amount, the largest amount offered by a convicted GS is PKR 853,000,000 in sector EOBI (the pension, old age benefits, and social insurance institution of the Government of Pakistan) of the province of Sindh in the year 2003.
Importantly, the total value of the funds voluntarily surrendered or plea bargained is worth a total of PKR 2,270,717,484 over our sample period. If we scale this amount by the annual per capita income of an average Pakistani in the year 2013 (which was $1386 or PKR 145,530), the amount equals the annual income of around 15,600 Pakistanis! It highlights the importance of the problem.

ES Data Set
The ES data set is an authentic source of information on various actual constraints that formal sector business firms face. The ES data set covers 146,000 firms in 143 countries. The surveys are conducted using similar questionnaire and instruments which make the analysis of this study comparable and replicable in different contexts. The manager responses cover a broad array of questions relating to a firm's operations including financial issues, dealings with government regulators, quality of infrastructure, workforce characteristics, and performance. The analysis in this study uses ES data set from the survey waves of 2007 and 2013 in Pakistan. Given our focus, we exploit the variation in the responses of the managers of business firms to questions asking them about government regulators and other aspects of corrupt dealings. The questions used in the empirical analysis are given in Appendix II at the end of the study.

Empirical Model
The empirical model relies on Becker's (1968) model of crime as extending by Rose-Ackerman (1978). In the context of this study, the theoretical model predicts that a selfinterested government officer demands bribe adjusting the returns with the probability of being sentenced. The recent history of corruption accusations and convictions involving government officials will serve as the subjective probability of being accused and raises the risk of demanding a bribe. This deterrent effect would reduce incidents of corruption, ceteris paribus.
In contrast, the option of plea bargaining may have a positive or negative effect on corrupt transactions. If it reduces the net return from misappropriation, then it can have a depressing effect. But the possibility of avoiding prison by surrendering a part of the illegal gains, if convicted, may reduce or completely overpower its deterrent effect. Ultimately the consequence of such a provision is an empirical question.
To determine the impact of arresting corrupt officials due to incidents of corruption, we developed a pooled least squares model with time and spatial fixed effects. We follow the recommendation of Wooldridge (2010) in estimating the model with sector, time, and province effects and employ standard errors that are robust against the effects of clustering due to the stratified random sample in the ES data:  Additional Controls = (ManExp; Exporter; ISOCertificate), where Corruption is the indicator of corruption. In the baseline analysis, it is the perception of business firm managers of the likelihood of encountering a corrupt official. In addition to this perception-based measure, we have also used alternative indicators. Two of these indicators (RegTime and Bribe3) are experience based and the other (Bribe2) is perception based. Thus, our results contribute to the literature dealing with perceptions as well as the experience of corruption as both aspects offer distinct insights.⁵ The No. of Accused GS denotes the average number of corrupt GS being accused and plea-bargained in the last 4 years in the given province. In estimations, the value of the No. of Accused GS for each province-year is relative to the total number of accusations by the NAB.
The location is denoted by k, and t denotes time. The baseline model assumes that the arrest of corrupt officials has a linear effect on corruption. We relax this assumption by conditioning our analysis separately on the survey year. Moreover, the baseline model also includes the average bargained amount (Plea Bargain Amount) to estimate its effect separately, as is required by Hypothesis 2.
Among the control variables we include various firmlevel determinants that can make it more or less attractive to bribe-seeking bureaucrats like its age, size, exporting status, and its sector and location. Additional controls are added to take account of the managerial capital (ManExp) and firm productivity (ISOCertificate); the effect of the level of economic development is captured through the human development index (HDI) of the province 4 Results Table 4 reports the baseline estimates and extensions that include additional controls. All the models employ standard errors that are robust against the clustering effects at the level of a province. The dependent variable is dichotomous which explains the use of the logit estimator. The diagnostics reported toward the end of Table 4 suggests that each of the models is highly significant overall.
In terms of the primary relationship of interest, our results suggest a strong and significant effect the number of (past) plea bargained cases has on the likelihood of corruption.⁶ The coefficient of interest (i.e., No. of Accused GS) is negative and statistically significant. More importantly, the estimated effect in the baseline regression (Column 1) amounts to an 80% reduction in the odds of a corrupt encounter with each 1 standard deviation (SD) increase in the number of accusations. Importantly, the coefficient of Plea Bargain Amount is negative but insignificant in all cases.
Among the controls we find a significant negative effect of SME on corruption likelihood. The same holds true for HDI which is a proxy for the level of development of a province. Our results also suggest a positive effect of the firm's age on corruption likelihood although the coefficient has no statistical support. Among the sector effects we find a significantly greater likelihood of corruption across industries broadly categorized as leather products, machinery and electronics, motor vehicles, and nonmetallic minerals. These sectors are prone to corrupt encounters compared to the benchmark "textiles-garments" sector.
Among the additional controls, we can see the insignificant effect of the top manager's sector specific experience with corruption likelihood. The likelihood of corruption is significantly higher if a firm is earning at least more than 10% of its sales from exports (Exporter) or is having an international quality certification (ISO Certificate). It may suggest the greater interaction of exporting firms and those having superior quality products that make them easy targets of corrupt government officials.
The location and time effects are also significant. In the case of the former, the probability of corruption is higher in relatively developed provinces like Punjab, Sindh, and KPK compared to the base category represented by Baluchistan (the least developed province in Pakistan). While in the case of time effects, the binary variable representing 2013 suggests that the likelihood of corruption has decreased relative to the earlier time period.


6 For the purposes of interpretation, we use percentage change in the odds ratio estimated using "listcoef" routine in STATA 16 version.

Robustness
In this section, we first check our findings using alternative measures of corruption (both perception and experience based). Second, we check the time dependence of the effect of anti-corruption strategy (of plea bargaining) by splitting our sample into two time periods for which firm surveys were available. Finally, we address endogeneity issues using an instrumental variable (IV) estimator.
In Table 5, we change the dependent variable to estimate the effect of accusations on the other dimensions of corruption. In Column 1 of the table, the dependent variable measures the percentage of the senior management's time spent in dealing with government regulations (RegTime). Unlike the perception-based indicator in the baseline analysis, RegTime captures the manager's actual corruption experience. Moreover, if government bureaucrats were abusing their regulatory powers to extract bribes from the firms, then this indicator is likely to capture any changes in this behavior due to the anticorruption policy of the NAB.
Given that ES data do not provide separate indicators on the different aspects of regulation, the RegTime is an overall indicator of taxes, customs, labor regulations, licensing, and registration. Therefore, these results allow us to detect whether there is any improvement in efficiency in the overall regulatory machinery due to anti-corruption policies. Unlike our baseline model this variable is not dichotomous and so we use Ordinary Least Squares to estimate the model.
The results in Column 1 ( Table 5) suggest a strong negative effect of the nabbing of government bureaucrats on firm managers' time. In other words, the greater is the number of plea-bargained corruption cases in the past, the lesser is the time private businesses have to spend in dealing with government regulations. In quantitative terms, a typical firm's management needs 5% less time per week to deal with GS with each one SD increase in the number of plea-bargained cases in the province where the firm is located. Unlike baseline results, the bargaining effect is significant and negative in this case. Thus, an increase of one SD in the amount of plea-bargaining saves only 0.3% of the weekly time managements spent Notes: Standard errors in parentheses are robust against heteroskedasticity and province specific clustering effects; ***p < 0.01, **p < 0.05, *p < 0.1. Constant is included but not reported.
in dealing with bureaucrats. This result also suggests that corruption is the cause of inefficiency in Pakistan and that the anti-corruption policy of plea-bargaining helps improve the regulatory apparatus of the government.⁷ In Column 2 (Table 5), the dependent variable (Bribe2) is again dichotomous measuring the incidents of bribery if a firm experiences at least one bribe payment across six different types of public transactions dealing with access to public utilities, building permits, operating licenses, and taxes. The effect of detection is again negative and significant. In specific terms, the result suggests that the odds of corrupt dealings decline by 92% with each single SD increase in the cases of plea-bargaining involving government officials. The bargaining effect is again insignificant.
The dependent variable in Column 3 of Table 5 measures corruption perceptions rather than experience. More specifically, the variable labelled Bribe3 assumes a value of 1 if a formal payment is expected to secure a government contract. The coefficient of interest is again negative and significant although only at 10% level, whereas the bargaining effect is negative and significant.
In the last two columns of Table 5, we estimate our baseline model separately for business firms surveyed in 2007 and 2013. Although the results still suggest the negative influence of the number of plea-bargained cases, the effect of the plea bargained amount is negative for the firms surveyed in 2007 and becomes positive in the case of 2013. This suggests a change in the effect of the plea bargaining option on corruption over time. However, we refrain from inferring much from this result because of the non-experimental nature of our data. Nonetheless, this novel finding provides an interesting avenue for future research. Table 6 addresses the issue of endogeneity. Admittedly, the demand and supply of "corrupt transactions" is simultaneous which makes it difficult to establish causality. To handle the issue of endogeneity we used twostage least squares (2SLS) which uses the IV estimator to capture the exogenous impact of endogenous regressor. We instrument No. of Accused GS using a measure of political will.

Endogeneity
The rationale behind the use of political will is not difficult to discern. It can be argued that the strength and independence with which anti-corruption agencies pursued corruption in the government's ranks reflect the government's resolve and will to get rid of this menace (Ankamah & Manzoor E Khoda, 2018). One possible determinant of the government's resolve to avoid antigrowth practices is the number of popular votes that the ruling party receives in the last elections. As Besley, Persson, and Sturm (2010) show, the lack of political competition is associated with anti-growth policies. Therefore, greater popular support provides greater political will to the ruling party to take strong measures to prevent the misuse of public authority. Thus, we use the provincial level share of the ruling party's vote in the election as a proxy for its will to fight administrative corruption in government departments. This variable is denoted as PopVote and it comes from Chaudhry and Mazhar (2019) data set constructed specifically for Pakistan. We use this variable and its squared form to instrument the grabbing variable. The 2SLS results using PopVote as instrument for No. of Accused GS are reported in Table 6. Both the columns include location, time, and sector effects in line with the baseline specification. In Column 1 we report only the PopVote as an instrument, while in Column 2 we also include its square, i.e., PopVoteSq. It is important to check whether our instruments are satisfying the statistical requirements necessary for a valid inference in the 2SLS procedure. To this end, we have reported a number of diagnostic tests toward the bottom of the table. For instance, first-stage F-statistics in both the regressions exceed the benchmark value of 10, suggesting a strong correlation between the No. of Accused GS and the Pop-Vote. The Hausman test of endogeneity of the No. of Accused GS provides strong evidence that we cannot treat it as exogenous. Finally, the over identifying restrictions imposed by our instruments (in Column 2) are not valid as the value of the Sargan test suggests.
The coefficient of the No. of Accused GS in 2SLS estimates in both columns in Table 6 is significant beyond 1%. Because of the dichotomous nature of the original dependent variable we avoid interpreting the 2SLS coefficients. The results on control variables follow the earlier pattern. Thus, the firm being of small or medium size decreases the probability that it will be asked to pay a bribe by a government official. The age of the firm is insignificant and the overall level of development of a province is negatively correlated with corruption occurrence.
In sum, the results as per various specifications and after using the different forms of corruption as dependent variable clearly show that the nabbing of corrupt officials reduces incidents of corruption at the level of the business firm. Its weaknesses aside, the evidence from 2SLS regressions overcome the endogeneity in the hypothesized link between nabbing and corruption incidents and give support to our baseline results. In sum, we can claim that different statistical procedures yield consistent results.

Conclusion
An important issue in the literature on corruption involves the best way to deal with it. Huge monetary resources, besides time and effort, are devoted to designing effective institutions to get rid of this menace. Gathering scientific evidence on the efficacy of such measures directly informs the policy makers and is no less instructive for academicians.
This study tests the efficacy of anti-corruption measures in the context of Pakistan. Using theoretical insights, e.g., Becker (1968) and Rose-Ackerman (2010), it derives two hypotheses to test the impact of the plea-bargaining policy on corruption perceptions and corruption experiences of formal business firms.
The study gathers a unique data set on the number of accused government officials who availed the option of plea-bargain and voluntary disclosure of illicit gains. So unlike other corruption cases, these cases are unique in attracting public attention (and ire). The risk of detection and scrutiny by the NAB with the positive probability of Overidentifying restrictions test reports the value of Sargan score test. The null hypothesis is that overidentifying restrictions are valid. *** p < 0.01, ** p < 0.05, * p < 0.1.
surrendering the ill-gotten wealth influence the costbenefit ratio of potential bribe seekers. Two waves of the World Bank Enterprise Surveys (2007 and 2013) provide a firm-level data set of more than 2,000 firms covering the four main provinces of Pakistan to empirically test these hypotheses. Although the econometric results support the hypotheses, the effect of the number of accusations (H1) on reducing corruption is more consistent. The results are robust against various specifications and hold as well if we divide our sample into individual surveys. With some qualifications, the results hold for the 2SLS estimator as well. Importantly, the independent effect of the amount of plea-bargaining on firm-level corruption although negative overall comes out positive if we restrict the sample to focus only on the latter half of the sample period, thus supporting hypothesis H2.1.
In terms of good effect the practice of plea-bargaining has on the perceptions and experiences of business firms and the value of resources recovered over the period under consideration, this policy may offer net social benefits. But the positive effect it exercises on corruption in the second time period suggests that it may increase corruption in the long term. Thus, it is likely that long-term expected costs exceed short-term gains.
This research is not without limitations. First, it focuses specifically on Pakistan. Although governance issues in Pakistan are not very different from other developing countries, it is difficult to tease out many generalizations. Second, the empirical analysis relies on a data set that is pooled crosssectional and is observational in nature. This reduces the reliability of the causality analysis. Nevertheless, the study serves as a motivation for future researchers to go beyond this simple analysis and add refinements to gather more reliable insights about this important area of public administration.