Paper decentralized bribery the costs, the benefits and the taming


Published on

NES 20th Anniversary Conference, Dec 13-16, 2012
Article "Decentralized Briber: The Costs, The Benefits and The Taming" presented by Sergey Popov at the NES 20th Anniversary Conference
Author: Sergey V. Popov

  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

Paper decentralized bribery the costs, the benefits and the taming

  1. 1. Decentralized Bribery: The Costs, The Benefits andThe Taming1 1 Popov: National Research University Higher School of Economics, Moscow,Sergey V. Popov I thank Dan Bernhardt, Mattias Polborn, John Nye and the par-February 28, 2012 ticipants of a seminar in the Laboratory of Institutional Analysis at HSE for use- I propose a bribery model in which bureaucratic decisionmaking is de- ful suggestions and thought-provoking centralized. I establish that bribe extortion is economically nonneutral, discussions. The support of the Basic and that capital markets in corrupt economies exhibit higher returns. Research Program at HSE is gratefully There are multiple stable equilibria: high levels of bribery reduce the acknowledged. The usual disclaimer applies. economy’s productivity due to suppression of small businesses. Com- petition among bureaucrats might improve the outcome, but does not necessarily decrease the total graft. The choice of corruption fighting tactics and the choice of whom to blame provide nontrivial outcomes. Keywords: corruption, bribery, decentralization. JEL: D73. No man is an Iland, intire of it selfe; every man is a peece of the Continent. John DonneThe russian language distinguishes between two different classesof bribery: likhoimstvo is bribery for doing things that an officialshould be preventing; and mzdoimstvo is bribery for doing things thatan official should be doing for free2 . Both are corruption, using pub- 2 Prahab Bardhan. Corruption andlic office for personal gain. An example of the first kind of bribery development: a review of issues. Journal of Economic Literature, 35(3):1320–1346,can be taking a bribe to overlook hazardous working conditions in 1997; Toke S. Aidt. Economic analysisa factory. An example of the second kind of bribery can be gouging of corruption: a survey. The Economic Journal, 113(491):F632–F652, 2003; andthe bribe by threatening to close a factory due to nonexistent viola- Jakob Svensson. Eight questions abouttions. The first kind of bribery can be prosecuted ex-post, and it’s corruption. The Journal of Economicclearly detrimental to the welfare of the economy. The second kind of Perspectives, 19(3):19–42, 2005bribery is simply a transfer, and is therefore perceived as innocuous.I concentrate on the second kind of bribery in this study, and I showthat this “transfer bribery”3 has significant economic consequences. 3 Shleifer and Vishny [1993], calls this corruption without theft, Bliss and Di Tella [1997] calls this surplus-shifting,This paper proposes a model of bribery that does not require the and Drugov [2010] calls this extortion.influence of centralized government4 . Decentralized corruption is 4 The Independent. Bungs and bribesrelevant when many decisions are made simultaneously by different football can’t kick this habit, March 15 1995. URL http://www.independent.people. Most people face corruption everywhere: it is never the case,; Thefor example, that the police are corrupt, but educators are not. More- Register. Apple accused of extortion by rival tablet biz, January 11 2012.over, a corrupt policeman will eventually interact as a client with a URL corrupt educator, who in turn will be a client of a potentially uk/2012/01/11/apple_extortion/;corrupt doctor. Most of the time, corrupt officials would rather pay and The St. Petersburg Times. U.S. travel agency accuses Aeroflot ofsmaller bribes themselves. But individual changes in bribe-taking extortion, February 9 2012. URLbehavior will not change the bribe amount that bribe-givers expect action_id=2&story_id=35138to give, and this critical issue is not captured by a single-bureaucrat
  2. 2. decentralized bribery: the costs, the benefits and the taming 2approach. Decentralization of decisionmaking does not solve thecorruption problem: one needs competition of inspectors for eachparticular corruption opportunity to lower or remove the bribe. There are both benefits and costs of bribery. Corruption destroysless lucrative projects, but frees up resources that can be used inother more productive projects, or abroad. Corruption serves as aredistributive mechanism: governmental officials are notoriouslyunderpaid for the services they provide. But, most importantly, cor-ruption destroys incentives for investment: smaller projects mightnot be able to feed both the investor and the bribe. In this case,small project investors do not enter the market, and inspectors ex-pect larger projects, and react by further increase of the expectredbribe. Introducing competition among inspectors can allow smallerprojects to start up, but does not necessary lower the total graft, si-multaneously adding the red tape costs.The literature has reached an empirical consensus that corruptionis detrimental to welfare, and significantly reduces both long-termgrowth and near-term investment. Corrupt economies are mostlyclosed and heavily regulated. Corruption is enforced by a lack of ed-ucation, low income levels and ethnic heterogeneity, and ex-coloniesare more prone to corruption5 . 5 Paolo Mauro. Corruption and growth. There is a vast theoretical literature. Pioneered by the rent-seeking The Quarterly Journal of Economics, 110 (3):681–712, 1995; Alberto Ades andliterature of Tullock [1967] and Krueger [1974], it includes the queue Rafael Di Tella. Rents, competition,model of Lui [1985], where bribes are taken for advancing customers and corruption. The American Economic Review, 89(4):982–993, 1999; Rafaelin a queue and actually improve allocations; Alesina and Angele- La Porta, Florencio Lopez-de-Silanes,tos [2005] models theft from government coffers, arguing that more Andrei Shleifer, and Robert W. Vishny. The quality of government. Journal ofredistribution does not necessarily bring more equality because of Law, Economics, Ond organization, 15(1):corruption; Aghion et al. [2009] builds a model of endogenous regu- 222, 1999; and Simeon Djankov, Rafaellation, arguing that societies with little social conscience invite more La Porta, Florencio Lopez-de-Silanes, and Andrei Shleifer. The regulation ofregulation; and many others. The closest model to mine is Bliss and entry. The Quarterly Journal of Economics,Di Tella [1997]. They argue that corruption can make the economy 117(1):1, 2002less competitive, move to a monopoly outcome and that bureaucratscan gouge all the monopoly profits away. Svensson [2003] uses a sim-ilar model to accompany a survey from Uganda to illustrate that thesize of bribe depends on a firm’s prospects. He predicts that becauseof bribes, investment in a less profitable sector that features moreliquid assets might be preferred to investment in a more profitablesector that features less investment reversibility precisely becauseofficials require bigger bribes in the second scenario. Mauro [2004]incorporates corruption into a growth model, bringing attention tomultiple equilibria as a potential cause of differences in developmenttrajectories. In contrast to my model, his model focuses on govern-mental provision of a public good.
  3. 3. decentralized bribery: the costs, the benefits and the taming 3The paper is organized as follows. First, I introduce the generalmodel and define the equilibrium. I then look at the model’s predic-tions: I study capital market outcomes; I study how to combat cor-ruption with exit facilitation; I illustrate that transfer bribery mightkeep the economy in a bad equilibrium where small entrepreneursdo not start up their businesses. Then I study how the organizationof inspection industry affects bribe-taking behavior. Finally, I discussmy model’s limitations and potential extensions, and conclude.1 The ModelThink of a story of a person who comes to a police inspector to passa driving test. The police inspector can clearly see a bad driver, andperfectly understands the welfare costs of allowing bad drivers onthe street; but denying a driver’s license to a good driver does notproduce a welfare externality, assuming away congestion issues.There are always obscure things, like missing a look in a rear-viewmirror, that can be used to fail a candidate with a comment of “reck-less driving”. These things make the implicit threat of failing a can-didate safe for the inspector. The question becomes: would a bribe inthis situation make a difference?Agents interact in a single-period game. There is a continuum I postpone discussion of model varia-measure 1 of ex-ante identical agents, whose preferences are defined tions to the conclusion.over a single good, which can be consumed or invested. There aretwo possibly overlapping subsets of agents: investors and inspectors,both of positive mass; those who do not belong to either set are out-siders. An agent becomes an investor with probability γ, and η is theprobability that one becomes an inspector. Each investor decides R observed; some Roles are assigned whether to start up his projects are cancelled; randomly project payoffs realize Time Investors observe Each inspector gouges realizations of their K bribes Figure 1: Timing of the gameInvestors draw a project of size K, where K is the required numberof units that must be invested, from a distribution with pdf f K (·)and cdf FK (·). A project returns a random, idiosyncratic payoff R ≥ The heterogeneity of K could be moti-0 per unit of investment, drawn from pdf f R (·) and cdf FR (·) that vated not only by technology, but also by the pledgeable income of investors.
  4. 4. decentralized bribery: the costs, the benefits and the taming 4is not observable by the investor at the time of investment, and isindependent of K. After investment, each investor is assigned a random inspector,who is supposed to approve the project, but instead attempts togouge a bribe, a sum of money s, from the project’s profits. If therealized project’s profits after paying a bribe are too small, the projectis cancelled, and the investor loses 1 − φ of his investment; φ is therecovery rate of investment. Since the probability of turning up as aninspector for the investment of the agent who serves as the inspectorfor your own project is zero, the decision of an agent in the role of ainspector does not interact with the decision of the same agent in therole of an investor. Each investor must decide whether to pursue his investmentproject. Starting a project of size K earns the expectation of s Here a ∨ b is an operator of max( a, b). [ RK − s ∨ φK ] − K = [ R − ∨ φ] − 1 K. KIf the after-bribe net return is less than φ, the investor cancels theproject. The investor cannot be liable for the project that is not lucra-tive enough to pay for the bribe that it faces. In other words, an in-vestor cannot be forced to pay a bribe; the investor can instead chooseto take everything he can and walk away. An investor starts his project if his expected net return is positive,i.e., if s E[ R − ∨ φ] − 1 ≥ 0. (1) K ˆLet s denote the value of s such that (1) holds with equality (implic-itly indexed by K). It is profitable for an investor to start a project of ˆsize K if and only if s < s. Inspectors informed about size K will charge bribes as a function of K; IResult 1 If participation constraint (1) is satisfied for some value of K (φ), postpone this issue to Subsection 1.3. Most of the results are derived withoutit is satisfied for the same bribe size for projects with larger K (φ) too. uncertainty about K, and are extended where possible to cases of uncertaintyIf returns are drawn from an exponential distribution, the borderline and imperfect information. This is theˆs is governed by principal difference between my model and one of Bliss and Di Tella [1997]:  1/α− φ  they use heterogeneity in fixed costs ˆ s ˆ s +φ 1 ln 1−φ and get the result that less efficientE [ R − ∨ φ ] = φ + e−α( K ) − φ ≥ 1 ⇒ s ≤ φ + ˆ  K. firms exit the market. In my model, K α α imperfect information of inspectors generates corruption that makes small firms less efficient after a bribe, evenAn inspector observes neither K nor R, so his bribe demand can- if they were equally efficient before anot depend on either. However, in equilibrium, each inspector knows bribe.which projects are deemed good enough for participation by in-vestors. An inspector’s problem is to choose bribe demand s to solve +∞ max sP ( R > s/K + φ) = s (1 − FR (s/K + φ)) f K (K )dK. (2) s 0
  5. 5. decentralized bribery: the costs, the benefits and the taming 5 The first-order condition is +∞ +∞ (1 − FR (s/K + φ)) f K (K )dK = s 1/K f R (s/K + φ) f K (K )dK, 0 0 +∞ (1 − FR (s∗/K + φ)) f K (K )dK EK [1 − FR (s∗/K + φ)] s∗ = 0 +∞ 1 = . (3) /K f R (s∗/K + φ) f K (K )dK EK [1/K f R (s∗/K + φ)] 0 When there is no uncertainty about K, Equation (3) can be rewritten as = 1− F(s/K++) ) . Assuming the R ( /K φ s s KAn equilibrium (pure strategy perfect Bayesian) is a collection of fR φ right-hand side to be decreasing (the increasing hazard rate assumption is sat-• s∗ ∈ R+ : the size of bribe, amount of money taken out of the isfied by a large family of distributions) project’s profits if the project is pursued; will guarantee the inspector’s problem to have a unique solution. However, in• K ∗ ∈ R+ : the critical level of investment such that investors with Subsection 1.3 I’ll give an example of distributions of K and R that produce projects of size K ≥ K ∗ decide to pursue them; an increasing right-hand side of Equa- tion (3) even though the distribution ofsuch that R has a nonincreasing hazard rate.• s∗ solves the inspector’s problem (2) given rational beliefs that only projects above K ∗ are implemented, and• an investor with a project of size K ∗ is weakly better off starting the project, and all owners of projects with K < K ∗ in support of distribution of projects find it suboptimal to pursue the project, given rational beliefs about the bribe size s∗ . There are three classes of outcomes, depending on the investors’ 6 If a restriction equilibrium existsparticipation: where only the best type of projects start up, then an argument similar toabundance: all projects are started; the intuitive criterion of Cho and Kreps [1987] rules out the autarky equilib-restriction: only a subset of project sizes is started; rium: the bribe cannot be expected to be so big that the best possible projectautarky: no projects are started.6 is not executed, because what type of projects would support these bribes?..Result 2 An equilibrium exists if the support of K is compact. I next study the properties of equilibria. The model is compact, yetit allows me to convey the main result: corruption can be so rampantthat small projects are not viable, and only big projects can start.This only increases the bribe size, securing the separation betweenequilibria. Hence, even decentralized transfer bribery can harm theeconomy, and the harm is not limited to the less lucrative projectsbeing cancelled due to a too small outcome of R.
  6. 6. decentralized bribery: the costs, the benefits and the taming 61.1 Inspector’s Problem and Capital MarketsConsider an outcome where all investors have the same project sizeequal to K. The bribe size chosen by inspectors from (2) is then s 1 − FR (s/K + φ) = . (4) K f R (s/K + φ)Hence, the bribe is a fixed proportion of the size of capital. Assum-ing the decreasing hazard rate of f R would produce a unique bribe ¯size choice s. Instead, I assume the existence of a positive bribe size.Then, the expected return of the investment project of size K is +∞ ¯ sEπ (K ) = E[ RK − s ∨ φK ] − K = K ¯ R− − φ dFR + φ − 1 . ¯ s K +φ K Conditional on starting up and not cancelling a project, the returnon projects in corrupt societies is higher than in less corrupt societies.Unconditionally, the return would be lower, but this would not beobservable in data.What is the return rate to capital? The derivative of ex-pected profit with respect to K from the point of view of an investorwho takes the bribe size as given is Average return Extra chance of not cancelling∂Eπ (K ) +∞ ¯ s ¯ s ¯ s = R− − φ dFR − 1 +K 2 1 − FR +φ . ∂K ¯ s K +φ K K KTobin’s marginal Q, the ratio of marginal productivity to the aver-age productivity of capital, exceeds 1. The expected profit per unit ofinvestment has to be nonnegative, or investors would not start theirprojects. Since the imposed bribe is a fixed cost, even with a linearproduction function, projects exhibit increasing returns to scale. Thisprovides investors incentives to merge projects, or to attract foreigninvestment. Aggregating projects by some agents increases the bribesize expectations, which in turn indirectly imposes an externality onother investors, discouraging smaller investors from starting theirsmaller projects. Such megalomania, however, is fruitless strategi-cally. In the current model, the bribe constitutes a fixed proportion ofthe project’s size. The bribe size will increase if investors merge theirprojects7 , resulting in no change of average productivity of a merged 7 Here I think about a hypotheticalproject. Another argument against the scale increase is that it is eas- splitting of the set of investors into a continuum of non-overlapping constel-ier for corrupt inspectors to target a bigger project to leech a bribe lations of the same finite size.from, but the strategic search by inspectors is outside of the scope ofcurrent paper.
  7. 7. decentralized bribery: the costs, the benefits and the taming 7 This result hinges upon the assumption of constant returns toscale, represented by independence of R and K. If real enterprises ex-hibit decreasing returns to scale, a similar argument will demonstratethat the Tobin’s marginal Q in data can be higher than what wouldbe in the world without corruption.1.2 Recovery Rate Affects BriberyThe rate of recovery, defining when an investor decides thatthe bribe demand is too high and decides not to pursue the project, 8 One can see that in Subsection 1.3, with the uncertainty about the projecthas a strong effect on corruption. Consider a function H ( x |φ) = size distribution, the recovery rate EK [1− FR ( x/K +φ)]EK [1/K f R ( x/K +φ)] , a reformulation of (3). The bribe size demanded by change does not matter. The right-hand side of Equation (5) is exactly zero.inspectors will be the intersection of the 45◦ line with H ( x |φ).Result 3 Suppose there is no uncertainty about the project size K. Thenan increase in φ reduces the bribe level s = H (s|φ) as long as FR (·) has anincreasing hazard rate.If returns are exponentially distributed, with no uncertainty aboutinvestment size, the recovery rate has no effect on the equilibriumbribe demand.8 However, most common distributions feature an s 45◦ line H (s|φ = 0.2)increasing hazard, and the result is very intuitive: a better recovery Bribe at φ = 0rate makes it less unattractive for the investor not to pursue a project,which causes inspectors to reduce their demands. H ( s | φ = 0)Result 4 Either (i) if f R (·) > 0 on (φ, φ + s∗/K∗ ]; or (ii) f R (·) is close touniform; are sufficient to guarantee that increases in φ weakly reduce the Bribe at φ = 0.2bribe level. sThese are sufficient conditions for∂H ( x |φ) EK [ f R ( x/K + φ)] E[1/K f ( x/K + φ)] Note: f R is Beta(0.1, 0.3), K H = 2, = − − H ( x |φ) 1 R x <0 K L = 1, K = K L with probability ∂φ EK [1/K f R ( x/K + φ)] E[ /K f R ( /K + φ)] λ = 0.5. (5)to hold. Figure 2 illustrates the logic. A cleaner but significantly Figure 2: Increase in recoverystricter assumption that f (·) is positive on the whole support of R is rate lowers the bribe.clearly violated by any distribution of R with unbounded support.Better recovery rates can reduce bribes because inspectors real-ize that investors have a better outside opportunity, and will toleratebribes less. This can motivate corruptioneers to demand industry-specific investments from potential investors before they can applyfor a permit. This also suggests that industries with better recoveryrates should suffer less from corruption, especially when one endog-enizes decisions by corrupt officials to choose the industry to target.For example, software development, which is easy to set up and sell
  8. 8. decentralized bribery: the costs, the benefits and the taming 8out, is likely to be less rife with corruption than highway construc-tion, which features significant investment into industry-specificequipment. This seems to be a strong and intuitive recipe for fightingcorruption: make investment more recoverable, possibly by focus-ing on industrial development with more recoverable investments.This does not mean that government should subsidize cancellations,because then less lucrative projects may start up just to get canceled. To ease presentation, φ will be set to 0 for the rest of the paper.1.3 The Squandering Virulence of Corruption Only H types Both typesThe environment that I will use for this part features hetero-geneity with respect to project size and exponential returns. LetK ∈ {K L , K H } with probabilities λ and 1 − λ correspondingly, andR ∼ Exp(α), so that P( R > t) = e−αt . Then, if both types of projectsare getting started up, the utility of the inspector as a function of thebribe amount s is −α Ks −α Ks s sP( RK > s) = s λe L + (1 − λ ) e H . (6) Figure 3: The right-hand sideTo solve for equilibrium, first consider the best response of in- of Equation (7) (solid line) andspectors. The first-order condition of the inspector’s problem (6) is Equation (8) (dashed line). KL α Ks   −α Ks −α Ks 1− e H 1 λe L+ (1 − λ ) e H KL  KH s= s s = 1+ s . (7) α λ e − α K L + 1− λ e − α K H α λ + KL α KH KL KH 1− λ KH e The right-hand side is an increasing function9 of s, starting from a 9 This is increasing because the likeli-value above K L and converging to KαH . Therefore, there is a solution. α hood of a continuing project to be of K H type is increasing with the bribe When only K H project size is started up, the inspector’s first-order size. Increasing hazard rate assumptioncondition’s right-hand side changes: on f R (·) is no longer sufficient for the uniqueness of a solution to (7). −α s −α s 1 0 × e KL + 1 × e KH K s= s s = H. (8) α 0 e−α K L + 1 e−α K H α KL KHThere might be multiple equilibria under reasonable assump-tions. Figure 4 shows an example of such an outcome. Both equilibriaare stable: a tiny change in the fundamentals of both investors’ andinspectors’ problems do not make either equilibrium go away. It isobvious from Figure 4 that the abundance equilibrium needs smallα to exist for every given λ. Lowering K H also lowers the bribe size KLwithout affecting the threshold bribe for participation constraint.The welfare costs of bribery are not as much in the loss of lesslucrative projects, but in squandering small projects by staying in a
  9. 9. decentralized bribery: the costs, the benefits and the taming 9 s Bribe s 45◦ line 45◦ line KH KH α α Inspector’s ˆ s choice ˆ s Inspector’s KL KL choice α α Bribe ∗ s ∗ s s1 s2 (a) Only K H projects start up (b) Both project types start up ˆ Note: α = 0.2, K L = 1, K H = 2, λ = 0.8. Since K L = 1, s is the bribe that agents with type L projects can pay and be indifferent between starting up the project or not; see Equation (1). For K H projects, 2ˆ > ρ KαH , so restricted equilibrium exists. s Figure 4: Multiple equilibriarestricted equilibirum. Even if ? temporary effort in lowering bribescannot remove bribes completely, it can be strong enough to movethe economy into an equilibrium where more projects are started up.Both inspectors and investors are interested in this outcome—largeproject investors start paying smaller bribes, small project investorsnow find bribes small enough to start their small projects, and in-spectors collect bigger graft. Unfortunately, they cannot coördinateon such deviation since the bribe-taking decision is decentralized.If inspectors had perfect information about every project’ssize, this shortcoming would not be an issue, as inspectors couldcharge bribes proportional to the size of the project. Even imperfect The inspector can convey his bribeinformation would ease the participation constraint on the small expectations by citing the “violated” statute codes: a corrupt traffic officerprojects’ investors, potentially inviting them to participate. This could threaten a driver with KY plateswould not, however, destroy the restricted equilibirum. Assume with a speeding ticket, and a driver with NH plates with reckless driving.that the inspector obtains a correct signal with probability q, and withprobability 1 − q he gets an incorrect signal. In a two-state worldwhere both types of projects start up, the inspector would update anapriori belief in probability of observing a low-type project condi-tional on the signal: (1 − q ) λ qλ λH = , λL = . q (1 − λ ) + (1 − q ) λ qλ + (1 − q)(1 − λ)Based on λ L and λ H , each inspector will solve two different problemsto choose the bribe size. These problems are illustrated in Figure 5.This creates two bribe levels, s∗ and s∗ . When q > 1/2, λ L < λ < λ H H Land s∗ < s∗ < s∗ , where s∗ is the outcome of a model with q = 2 . L H 1
  10. 10. decentralized bribery: the costs, the benefits and the taming 10 s 45◦ lineResult 5 If restricted equilibrium exists, for large enough q the abundanceequilibrium exists.This does not remove the existence of the restricted equilibrium. λHEven if the q is big enough, so that the abundance equilibrium exists,a belief that small businesses do not start up will lead inspectors λLto disregard their signals. Even if investors could coöperate andstart up a positive mass of small projects to manifest their collectivepotential, the decentralization of decisionmaking would neither allowindividual inspectors to comprehend the organized deviation nor toattempt lowering the bribe to attract small businesses. Unless q = 1, s s∗ L s∗ Hthe problem of squandering small projects persists. Figure 5: Two bribe levels.2 Bribery and PunishmentAnti-corruption effort is taken worldwide in order to suppressthe rent-seeking behavior. Lowering bribes even temporarily canlet the economy to move from a worse equilibrium to a better one.However, economies often have different responses to the same treat-ment. How do punishments affect the incentives of the agents in thismodel?2.1 Multiple InspectorsHaving to pass just one inspector is a simplification. Thissubsection studies the change in equilibrium in case investors facemultiple inspectors. Obviously, when inspectors can conspire, theybehave as in the case of a single inspector.When approvals of two inspectors are necessary for the This part mirrors the findings ofinvestor to harvest returns, this will bring more project cancellations Shleifer and Vishny [1993].into the economy. Since inspectors cannot conspire, each inspector Similar findings are reported inwill have less effect on a change in probability of soliciting a bribe by Arbatskaya [2007] and Janssen andchanging his own bribe size, hence the total graft will grow. Roy [2002]. Drugov [2010] argues that competition in bureaucracy might induce more investment inResult 6 When the soultion to (3) is unique, increasing the number of avoiding externalities; this is not ainspectors per project weakly increases the graft. problem here, but this is a valuable argument for a mechanism design problem of convincing inverstors toWhen one inspector out of many is sufficient for the project choose better projects. Subsectionto continue, and investor can appeal the decision of a previous in- 2.2 considers this problem while answering why do we need to keepspector, the outcome becomes more interesting. Each investor has a inspectors around.sequence of inspectors, assigned independently and anonymouslyex-ante, and the negative decision of each inspector can be overruled
  11. 11. decentralized bribery: the costs, the benefits and the taming 11by the next one. There is always an honest inspector at step T + 1,honest inspector never takes bribes. The investor makes a decision ofwhether to pay a bribe or to go to a next inspector for an appeal, andhe has a rational belief that these inspectors will expect dependingon how long the investor kept appealing10 . Each change of an in- 10 I think here about a person who triesspector will shrink the return by a factor δ < 1 due to accomodating again and again to pass the driving test. Inspectors from the traffic authoritiesto unreasonable demands of the previous inspector. The timing is know the quantity of previous trials,summarized in Figure 6. but do not know which inspector will be approached next, and the quantity of inspectors is big enough to make Payoff δ0 RK − s1 history-dependent threats and promises Pay s1 Yes inefficient. Investor to in- observes Payoff δ1 RK − s2 spector Pay s2 Yes R No 1? to in- Payoff δ T −1 RK − s T spector ...Pay s T Yes 2? No to in- Payoff δ T RK spector Yes T? No Inspector T + 1 is honest T Figure 6: Timing of the Bargain- The belief in a sequence of st t=1 constitutes a part of equilibrium ing for the Bribedefinition now, and it should coincide with true choices of inspectors T s∗ t=1 to be rational. When st+1 > st for every t > 1, there is no treason to go to the second inspector, and the previous results apply. I will use s∗ to denote the bribe size inWhen st+1 < st , the problem of an inspector who is approached on the baseline model, and I will assume the increasing hazard rate in this part ofthe step t is the subsection. max sP ( RK − s > max(0, δRK − st+1 )) . s Since there is an honest inspector, the investor will earn at leastδ T RK, so zero does not matter. s − s t +1 st = arg max sP R > . s (1 − δ ) K This should not be interpreted as “su- Obviously, picking st < st+1 is a waste of opportunity: all investors perior corrupt bureaucrats make lesswould agree to a bribe lower than what the next inspector demands. money in bribes than their corrupt underlings”. It means that a superiorTherefore, in optimum st > st+1 for every t. The first-order condition will take a smaller bribe for the sameis then job. This is a no-persecution scenario; with an active anti-corruption effort, superiors have more to lose, and there- +∞ s∗ −s∗+1 t t s∗ −s∗+1 t t fore might be motivated to bribe even s∗ 0 1 − FR (1− δ ) K f K (K )dK P R> (1− δ ) K smaller (possibly zero!) amounts. This t = s∗ −s∗+1 = s∗ −s∗+1 . can be the story behind the assumption (1 − δ ) +∞ 1 t t t t 0 /K f R (1− δ ) K f K (K )dK E[1/K f R (1− δ ) K ] of existence of honest inspectors. (9)
  12. 12. decentralized bribery: the costs, the benefits and the taming 12 When there is no uncertainty about K, the first-order condition canbe rewritten as s∗ −s∗ s∗ 1 − FR (t1−δt)K+1 s∗ − s∗+1 t t t = s∗ −s∗+1 ≡ HR . (10) (1 − δ ) K f R (t1−δt)K (1 − δ ) K The separation across inspectors 4 3 2 1 If the second inspector is honest, s2 = 0, then s1 ≤ s∗ , with equal- ∗ity when δ = 0. Competition among inspectors can lower bribes,and potentially might make the participation constraint for investors HR ( R|·)slacker. This result does not depend on not having uncertainty aboutK, but it has an implicit assumption that the participation decision H ( R |0)by investors does not change. Since the bribe is lower, allowing forparticipation decision to change will replace the equality sign with ≤ H ( R |2) H ( R |1)sign. R ˆ ˆ R3 R2 ˆ R1 ∗ ∗ If s2 > 0 and δ is close enough to zero, s1 > s∗ . The ability toscreen investors, albeit imperfectly, allows inspectors to gouge bigger ˆ Hazard rate ordering: st imply Rtbribes from more lucrative projects. This effect might be big enough Q(s|t) 45◦ line t = 2to actually increase the total graft. t=1 To solve for the bribery outcomes, observe first that investors are ¯self-selecting into groups based on R. If an investor with R = R for ∗ s1 t=0 ¯some R decides to pay the bribe at period t, all investors with R > R ¯too decide to pay the bribe before period t + 1. ∗ s2Less lucrative projects end up in a chain of appeals andchanges of the inspectors. Lucrative projects find it more costly to ∗ s3appeal. The estimates of corruption based on reports of those who ∗ s4 = 0 scomplain about losses due to red tape are likely to underestimate ˆ − R2 1− δthe true losses, and the bribes the complainers report are likely tobe smaller than what compliant agents pay. Denote the return of ˆ − R1 1− δa project whose investor is indifferent between paying the bribe at ˆperiod t and appealing to the next inspector by Rt ; investors withR≤R ˆ t projects complain. Therefore, the distribution of R gets worse ˆ Inspectors’ best response: Rt implywith every iteration in a hazard rate stochastic dominance sense, st .lowering HR (·). Denote the hazard rate of the distribution that thet-period inspector faces by HR (·|t). Note: FR ( x ) = 1 − exp(−(0.4x )2 ), ˆ ˆ δ = 0.5. R1 = 1.3885, R2 = 0.5697; these Let K = 1 for brevity. Rewrite equation (10) to get are equilibrium in a game where the fourth inspector is honest. s∗ s∗+1 = s∗ − (1 − δ) HR 1 t t − t t ≡ Q ( s ∗ | t ). t (11) 1−δ Figure 7: Solving (11).Since HR (·|t) is decreasing, the right-hand side of the previous equa-tion is increasing. Since the hazard rate is positive, the right-handside is less than zero in the neighborhood of s∗ = 0. t This system of equations, combined with s∗ +1 = 0, can be solved Tas shown on Figure 7.
  13. 13. decentralized bribery: the costs, the benefits and the taming 13Social welfare can be affected via two channels. First and fore-most, some agents suffer the red tape costs δR. These agents toleratethe red tape costs precisely because these costs are small. This oppor-tunity however allows inspectors to price discriminate investors, po-tentially extracting bigger total amount of bribes by “taxing” highlyproductive entrepreneurs more. The total output might simultane-ously grow due to lower bribes on less productive firms, hence biggerparticipation of less lucrative enterprises. When switching from theworld of the baseline model to the world of an opportunity to switchyour inspector, one will see more projects starting up, positive usageof the opportunity to switch, less reports of shutting down projectsdue to bribe and more reports of exposure to bribes per capita; butit is hard to predict whether the total graft will increase or decrease.The amount of surplus left for investors might either increase due tomore projects starting up and potentially lower bribes, or decreasedue to potentially bigger bribes and price discrimination by corruptofficials.2.2 Why Have Inspectors?I assume inspectors are kept in the economy to prevent harmful Cadot [1987], Acemoglu and Verdierprojects from happening. In order to obtain the independence of [2000], Drugov [2010] make similar arguments in similar frameworks.findings of the paper from the interactions of bribe-taking inspectorswith harmful projects, I need to show that there is an equilibriumin a metagame, where investors choose not to participate in badprojects. In other words, I design a punishment mechanism thatmakes the choice of the project type efficient. Consider a modification of the original setup where investors canchoose between executing a good project of size K = 1 with return R,or a bad project of the same size with return (1 + θ ) R and a negativeexternality of −kθR imposed on all other agents in the economy.I assume k > 1, so that maximization of the total output calls forstarting up good projects instead of bad ones. Since projects areinfinitesimal, the effect of unleashing a marginal bad project on thewelfare of individual inspector’s utility is negligible. Inspectors cansee the nature of the project perfectly11 . 11 Imperfect observability obtains similar results.When only the inspector is punished, one can imagine a detec-tion and punishment technology that takes away the collected bribes with probability G (s/s∗ ), where s∗ is the average bribe in the econ-omy. What G (·) can ensure an equilibrium where only good projectsare started up? The inspector’s problem is therefore twofold: onebribe level for good projects, another bribe level for bad ones.
  14. 14. decentralized bribery: the costs, the benefits and the taming 14 1 s∗ = arg max(1 − FR (s))(1 − G (s/s∗ ))s = g ¯ f R (s∗ ) . (12) s g g(s∗/s∗ )1/s∗ g ¯ ¯ 1− FR (s∗ ) + 1− G (s∗/s∗ ) g ¯ g 1 s∗ = arg max(1 − FR (s/(1 + θ )))(1 − G (s/s∗ ))s = (1 + θ ) ¯ . (13) b f R (s∗/1+θ ) s ∗ ¯∗ 1 ¯∗ s b 1− FR (s∗/1+θ ) + (1 + θ ) g( bG(s )//∗s) 1− /s ¯ s b b I assume that the right-hand sides of Equations (12) and (13) havea unique intersection with 45◦ line. The equilibrium therefore isadjusted to include:• s∗ , the average bribe in equilibrium, ¯• s∗ and s∗ , bribe levels for bad and good projects, respectively. b g• and each investor’s decision whether to pursue a “bad” project or a “good” project. g(t)Result 7 When g(·) = 0, s∗ = (1 + θ )s∗ . When 1−G(t) is increasing, and b gright-hand side of Equation (13) is decreasing12 , s∗ ≤ (1 + θ )s∗ . In both b g 12 The first assumption is the increasingcases, all investors pursue bad projects. hazard rate assumption for G (·); the second one is implied by increasing hazard rates of both G (·) and FR (·), but If taking bigger bribes is riskier, bad projects are becoming even does not necessarily require them.more profitable because inspectors are taking away smaller share ofthe rents by bribes. Punishing inspectors more will lower the bribestaken away from the investors, but this will only improve the badinvestors’ post-bribery returns. 13 Imagine observing citizens’ incomePunishing investors for bribery seems cruel in the context of or consumption with noise, whichthe baseline model of the paper, where all projects are beneficial and could be caused by the family structure. This allows to detect inspectors withnot paying a bribe means your project is not happening, but this is unusually large income, but veryan effective tool to make bad projects less attractive. To make bad imperfectly. Bigger difference between s∗ and s∗ will make the detectionprojects more detectable13 , a benevolent planner should not impose b g more reliable. Simultaneously, sincea penalty on inspectors. Holding an investor liable increasingly more the planner would want to inducewith the bribe size might make the participation constraint for bad an outcome where no investors are pursuing a bad project, a large bribe forprojects worse. Assume for the rest of this subsection that inspectors the bad project is not affecting anyone’sare not punished (so s∗ = s∗ (1 + θ )), whereas the investor loses her b g welfare.project altogether if the bribery is detected in the interactions withthe inspector. The participation decision is then (1 − G (s∗/s∗ )) E[ R − s∗ ∨ 0] − 1 > 0 g ¯ gfor good projects, and (1 − G (s∗/s∗ )) E[(1 + θ ) R − s∗ ∨ 0] − 1 > 0 b ¯ bfor bad projects. The equilibirum when only good projects start upimposes s∗ = s∗ , and it should be the case that good projects have ¯ g
  15. 15. decentralized bribery: the costs, the benefits and the taming 15higher expected return. Therefore, in such equilibrium two condi-tions should hold:(1 − G (1)) E[ R − s∗ ∨ 0] > max[(1 − G (1 + θ ))(1 + θ ) E[ R − s∗ ∨ 0], 1]. g g Good project return Bad project return (14)Result 8 The equilibrium where only good projects start up exists when(1 − G (1)) is big enough and (1 − G (1 + θ )) is small enough.One would need a more increasing G (·) if inspectors were prosecutedfor the size of taken bribe to make bad projects go away. Nothing yetprevents G (1) from being equal to 0 to stay in line with the body ofthe paper. The equilibirum where only bad projects start up might exist atall times when the “good” equilibrium exists. In this case, s∗ = s∗ . b ¯Manipulating G (·) to induce exit from such an equilibrium requiresmaking the following equation to hold: 1(1 − G ( )) E[ R − s∗ ∨ 0] > max[(1 − G (1))(1 + θ ) E[ R − s∗ ∨ 0], 1]. g g 1+θ Good project return Bad project return (15)Result 9 An equilibrium where only bad projects start up does not exist if G (t) = 1 − ctρ−1 for ρ < 0 satisfy 1(1 − G ( 1+θ ) is big enough, and (1 − G (1)) is small enough. incentive constraint conditions in both Results 8 and 9.More tolerance towards petty bribery might induce the transitionfrom an equilibrium where only harmful projects start up to theequilibrium where only good projects are started up.Summing up, an arrangement when inspectors are expected to makediscretionary decisions about which projects should be allowed tostart up can only work out if investors are also punished for bribery.For k, the parameter of destructiveness of bad projects, big enough,the negative consequences of allowing transfer bribery to go onmight be better than negative effects of investors switching to badprojects. The penalty for bribery should be harsh enough to haveenough freedom to discourage bad projects from running, but shouldnot be too harsh, so that investors are still starting up their projects.Inspectors are pivotal in this setup: it’s their ability to distinguishgood projects from bad and solicit bigger bribes from latter that al-lows the policymaker to prevent the spread of bad projects underthe form of “fighting the corruption”. This is why corruption is hardto weed out: it serves a beneficial purpose of creating incentives for
  16. 16. decentralized bribery: the costs, the benefits and the taming 16inspectors to utilize their specialized skills of distinguishing projectsproperly, stimulating investors to stay away from bad projects.2.3 Why Some Investors Are Not Asked For A Bribe?Assume no heterogeneity of projects. The distribution of projectreturns FR (·) features an increasing hazard rate. Assume inspectorsget punished (their bribe collection get taken away) with probabilityG (Γ/Γ∗ ), where Γ is the total collected graft and Γ∗ is the averagegraft per inspector in the economy. This is in contrast to the previous,where the probability of punishment depended on the size of eachindividual bribe. Just as before, G (·) has properties of a cdf, andfeatures increasing hazard rate.Assume there is less inspectors than investors in the econ-omy, and therefore each inspector has to deal with γ > 1 investors ηon average. Assume that the number of investors for each inspectoris assigned randomly, according to a distribution with mean γ/η , afterthe initial investment. If inspector is assigned just one investor, his bribe level choice willbe, as before, ∗ 1 s1 = f ( s ∗ ) ∗ ∗ . (16) R 1 1 g(s1 /Γ ) 1− F ( s ∗ ) + Γ∗ 1− G (s∗ /Γ∗ ) R 1 1 ∗Under increasing hazard rate assumption, s1 exists and is strictlypositive.Assume now that an inspector faces two investors. If theinspector is approached by investors in sequence, and the first in-vestor chose not to pay bribe and not to go forward with the project,then the bribe charged from the second invesor can be obtained from(16). Now assume the first investor has paid Γ. What is the paymentthe second investor is going to get charged? ∗ s2 = arg max(1 − G (Γ+s/Γ∗ )(s + Γ)(1 − FR (s)) + (1 − G (Γ/Γ∗ )ΓFR (s). s (17)Rewrite the first-order condition to get ∗ 1− G (Γ/Γ∗ ) f R ( s2 ) 1+ ∗ 1− G (s2 +Γ/Γ∗ ) 1− FR (s2 )∗ Γ ∗ s2 +Γ = ∗) f R ( s2 ∗ s2 +Γ/Γ∗ ) ⇒ 1 g( ∗ + Γ∗ 1− G (s∗ +Γ/Γ∗ ) 1− FR (s2 ) 2 Due to higher risk of punishment Due to already having Γ of graft. Γ g(s2 +Γ/Γ∗ ) ∗ ∗ ∗ G (s2 +Γ/Γ∗ )− G (Γ/Γ∗ ) f R (s2 ) 1 − ∗ ∗ Γ 1 − G (s2 +Γ/Γ∗ ) ∗ ∗ 1− G (s2 +Γ/Γ∗ ) ∗ 1− FR (s2 ) s2 = ∗ ∗ + ∗ ∗ Γ . (18) f R ( s2 ) 1 g(s2 +Γ/Γ∗ ) f R ( s2 ) 1 g(s2 +Γ/Γ∗ ) ∗ 1− FR (s2 ) + Γ∗ 1− G (s2 +Γ/Γ∗ ) ∗ 1− FR (s2 )∗ + Γ∗ 1− G (s∗ +Γ/Γ∗ ) 2
  17. 17. decentralized bribery: the costs, the benefits and the taming 17 The importance of the second term depends upon which hazard f (·) g(·)rate grows faster, 1−R (·) or 1−G(·) . The question is: is there a big FRenough bribe that might interest the inspector? Γ g(Γ/Γ∗ ) ∗ s Γ = 2s1 ∗Result 10 If Γ∗ 1− G (Γ/Γ∗ ) < 1, s2 > 0. Γ = 3s1 ∗ 45◦ lineThe right-hand side of (18) obviously does not have to be above zero, Maximumas the second term might be not big enough to dominate the firstterm, and the first term eventually will become negative; particularly, g(Γ/Γ∗ ) ∗it is always negative when 1−G( g(Γ/Γ∗ )) Γ > 1). There might not be an Γ ∗interior solution, and instead one can get a solution at s2 = 0. Minima ∗ Γ g(Γ/Γ∗ )Result 11 s2 = 0 when Γ/Γ∗ is big enough ( Γ∗ 1−G(Γ/Γ∗ ) > 1) and (18) has sat most one solution. Note: FR (t) = 1 − exp(−0.4t), G (t) = 1 − exp(−0.5t), Γ∗ = ∗ ∗ ∗ 2, s1 = 1.5385. For Γ = 3s1 the inspector will choose s2 = 0: the ∗ ∗ ∗ first term of (18) is negative. For Γ = 2s1 , s2 ≈ 0.48s1 .Zero bribe backlashes on the first period, where the inspector has Figure 8: Solving (18).to take into account that another person will come when choosingΓ, and being too greedy with the first applicant might prevent fromtaking a bribe from the second applicant. The problem facing the inspector when the first investor is ap-proaching him is even less analytical. However, one can clearly see ∗already that taking s1 from all investors becomes less probable. Thisnot only creates some variance in the coerced amounts even with-out any project heterogeneity, but also can lower Γ∗ to less than γ s1 , η ∗which would happen if punishment were happening on a case-by-case basis. Therefore, assuming away the costs of decisionmaking, away of fighting corruption might be to increase the amount of deci-sions per inspector, against the intuition of decentralization. This alsowill make the participation constraint of investors slacker. On theother hand, insufficiently increasing hazard rate of G (·) might putthe economy in the situation where bribe-takers after taking a lot ofbribes are asking for a very big bribe, because the second term of (18)dominates.3 DiscussionCompetition among investors is assumed away to illustrate thatthe multiplicity of equilibria is not driven by strategic complemen-tartities. One could assume that the returns’ distribution is stochas-tically improving if there are fewer projects starting up. This couldobtain two equilibria, one with high profits and high bribes, and onewith low profits and low bribes, depending upon the functional formof stochastic improvement. On the other hand, more competitioninduces more innovation, and hence in the long run the total graft
  18. 18. decentralized bribery: the costs, the benefits and the taming 18might be higher in a more competitive allocation. This ambiguityis hard to resolve in general scenario, but empirical evidence doessuggest that a lack of competition is part in parcel with corruption14 . 14 Alberto Ades and Rafael Di Tella. Rents, competition, and corruption. The American Economic Review, 89(4):Risk-aversity is not modeled explicitly, but the results are robust. 982–993, 1999Risk-averse investors will have a stronger participation constraint,but will not change investors’ behavior after investment, since I as-sume no uncertainty about R at the point of decision to pay the bribe.Hence, the bribe amount will not be affected unless the set of par-ticipating projects is affected. The risk-aversity of inspectors willsomewhat change the inspector’s problem. Particularly if the utilityof s dollars of bribe is (s + µ)ρ − µρ for ρ ∈ (0, 1) and µ > 0, theinspector’s choice becomes increases in s, but <s 1− ρ EK [1 − FR (s/K )] s s=ρ +µ 1+ −1 . EK [1/K f R (s/K )] µWhen µ is zero, only the first term remains. The second part of right-hand side increases slower than left-hand side for big enough s, sothe optimal solution exists if the solution existed originally. Solutionis continuous in µ and ρ, particularly around ρ = 1. If, in addition,ρµρ−1 ≤ 1, this can be interpreted as a wasteful bribe-pocketingtechnology, where the transfer of s produces (s + µ)ρ − µρ < s ofcash in inspector’s pocket. Other forms of utility functions are alsopossible. But even the simplest risk-aversity in inspectors make thebribe size depend upon the bribe opportunities available: a largernumber of projects per inspector will enable the inspector to riskmore, and charge bigger bribes. Risk-aversity would, however, make a difference if we discuss thewelfare implications of transfer bribery. Even in the abundance equi-librium some projects do not continue because of the bribe threshold,so some value is lost, hence the transfer bribery worsens the totalwelfare in the risk-neutral world. However, in the risk-averse world,it is possible that the utility ex-ante, before the role assignment, willbe higher in the world with corruption, since corruption redistributessome income across agents in the economy. This is a plausible ar-gument for economies where the proportion of investors is small,the return of a project features a fat tail, and the disposable incomeof an outsider or a non-bribing official is low. However, rather thanallowing corruption to run rampant, a better policy recommenda-tion in this scenario would be to look for reasons why the fraction ofinvestors is so small.Private information about returns is a seemingly innocuous
  19. 19. decentralized bribery: the costs, the benefits and the taming 19assumption. One could have two distributions of return, a stochas-tically better one and a stochastically worse one, and disregard theuncertainty about the investment size. Then an argument similar tothe one presented in Subsection 1.3 will be applicable: projects withworse return might be socially optimal to implement, but the bribemight be too large to start up these projects. However, if it’s the inspectors who have private information aboutthe prospects of individual projects, the outcome changes somewhatfrom the baseline scenario. Indeed, if the inspector could crediblycommunicate his good signal to an investor, the inspector couldcount on a larger bribe. But there is no reason to report such signaltruthfully when the inspector cannot credibly prove that this is thetrue signal. Whether it is possible to harvest bigger bribes dependson whether investors will believe the inspector’s signals. In fact,Subsection 2.1 can be perceived as inspectors collecting informationabout projects’ returns. In a restricted equilibrium, investors would be interested in cred-ibly revealing their investment size, hoping to get a smaller bribe,which would overcome the squandering problem. Revealing R, how-ever, is never optimal: the inspector will immediately demand thewhole profit, since it’s likely that he cannot commit to a bribe in ad-vance due to the illegal nature of bribery.Honest inspectors that do not ask for bribes will relax the partic-ipation constraint, creating a more hospitable atmosphere for smallbusinesses, but simultaneously they will let big fish go away non-squeezed. The body of corrupt officials might actually be interestedin cleansing the ranks to induce better participation of investors, de-pending upon the shape of f R (·) and f K (·), but not necessarily to thesocially optimal levels.Income inequality and growth are the usual determinants ofeffective governance. Income inequality in societies where investorsearn more on average than workers is likely to be positive. If smallbusinesses are suppressed, and the heterogeneity with respect to Kis induced by limited pledgeable income of the poor, this is likelyto create a version of the poverty trap, where poor entrepreneurscan never earn enough to start up a business large enough to feedthe corrupt, whereas rich dynasties run large-scale enterprises eventhough they might suffer from decreasing returns to scale. Growth isnot modeled explicitly for brevity. As Alesina and Angeletos [2005]emphasize, it is important to recognize that the problem of bad equi-librium with high corruption being stable is not that there is just asequence of beliefs that lead to a bad equilibrium. The bad stable
  20. 20. decentralized bribery: the costs, the benefits and the taming 20equilibrium is correpsonding to a long historical experience of highcorruption that suffocates small-scale businesses. Short-term lossesare multiplied by long-term iterations, generating cross-countrygrowth divergence.A general equilibrium model—featuring the choice of the role,the decision of each investor to run a “good” project (like the oneI discussed) or a “bad” project (with unfavorable properties likenegative externalities), the decision of the informed inspector to pre-vent bad projects or ask for a bribe, and decisions of a policymakerregarding the inspectors’ renumeration package, taxation and super-vision over the inspectors—would be richer, and would provide abetter view of ways in which corruption hurts the society. One couldcontemplate the wage effects: higher wage of outsiders would ac-tually lower the coerced bribe amount, because the projects wouldbecome less profitable. One could also see that squandering of smallprojects would lead to lower demand for labor, and therefore lowerwages, having an indirect effect on ex-post inequality. However, thiswill complicate the mathematics, and obscure the main interaction Iwant to study: between “good” project starters and corrupt inspec-tors.4 ConclusionIn this study, I find that transfer bribery is not economically neu-tral. Not only does it destroy some of the output of less productiveprojects, it can shrink the set of projects that are started up. Bribesthat are too high might not only kill the less lucrative projects, but alsocan discourage small businesses from opening up, since bureaucratscannot distinguish the investment size from the investment’s return.This is not due to credit constraints of the investors, but rather due totheir limited liability with respect to bribes and an ability to choosewhether to pursue a small project. As with any transfer, bribes can be beneficial if the alternative tobecoming an investor is too grim: a society with high unemploy-ment might be eager to forego a less lucrative portion of investmentprojects to redistribute the benefits obtained by entrepreneurs. I donot believe this benefit is effective in the long run, but it does explainwhy “baksheesh” corruption is only frowned upon in developingcountries. It also makes it clear why entrepreneurs become interestedin social responsibility and donating a portion of the incomes to thepoor: they would rather give up a portion of their income before theneedy become an authority and get the opportunity to issue permits.One example of such self-organization might be the trade unions in
  21. 21. decentralized bribery: the costs, the benefits and the taming 21US in 1950s (see Hutchinson [1969]); another is “patent trolling”, in-tellectual property protection run amok (see Magliocca [2006], Diessel[2007]). On general, I argue, for corruption to be harmful to the societythere is no need for governmental coffers or public good provision.Hidden fees in hotels and airports, yearly tuition raises, medicalinsurance plans with uncovered treatments, a required gratituityof 18% in some restaurants, scams regarding small recurrent creditcard charges,—a lot of transactions put agents in a hold-up situationwhere they might end up sacking their investment. Corruption bygovernmental officials is the most unethical, as other economic agentshave little success arguing that their functioning requires a monopoly.I argue that there is a problem of small businesses staying away fromparticipation. I argue that competition among inspectors can makeparticipation more attractive for small businesses. This does notmean that there should be a lot of inspectors, but it should be thecase that the next inspector is able to overrule the rejection of theprevious one. Fighting corruption with penalties requires some sophistication.Penalties that are too big will incentivize away the bribe-taking be-havior, but are also likely to scare away people from taking the in-spector job. Designing a system that detects too large bribes andpunishes the inspectors might only lower the bribe size, but will notprovide incentives to stay socially responsible (i.e. not use harmfulpractices). The only way that bad projects can be prevented is whenthe residual claimants, the investors, are punished too, and punishinginspectors only makes the available signals—bribes—less informative.Unfortunately, this would never eliminate bribes. One way to makean inspector take zero bribes from at least some investors is to designa system that would informatively detect the aggregate graft for eachinspector.A Proofs of Results sResult 1: Let K > K, and for K E[ R − K ∨ φ] ≥ 1. Observe that s s s sK < K , and therefore R − K > R − K for every R. Therefore, s s[ R − K ∨ φ] ≥ [ R − K ∨ φ]. Take expectations to obtain the result. s s Let φ > φ. Observe that [ R − K ∨ φ] < [[ R − K ∨ φ] ∨ φ ] < s[ R − K ∨ φ ] for every R. Take expectation to obtain the result.Result 2: The best response of inspectors to the boundary K is anincreasing function s∗ (K )—the bribe given that all investors withprojects above K start up their projects, the best response of investors
  22. 22. decentralized bribery: the costs, the benefits and the taming 22is K ∗ (s)—the smallest project size that satisfies the participation con-straint given bribe s. Their composition, K ∗ (s∗ (K )), either is definedfrom the support of K to the support of K, in which case Kakutani’sTheorem obtains the result, or is not defined at some point of K, inwhich case s∗ (K ) constitutes a part of an autarky equilibrium, whereno projects are started up. R 1− F ( s + φ )Result 3: When there is no uncertainty, HR (s|φ) = f (s+φ) . When RR distribution features a decreasing hazard rate, HR (·) is decreas-ing. An increase in φ means a shift of HR (·) to the left; hence, theintersection is happening at a smaller value of s.Result 4: Both conditions are sufficient for (5) to hold.Result 5: When the restricted equilibrium exists, it means that for 1− FR (s∗ /K H )H-type projects, s∗ = H H f R (s∗ /K H ) KH satisfies the participation con- H 1− F (s∗/K )straint, or that E[ R − H ∨ 0] > 1. When q = 1, s∗ = f (s∗/K )L K L , s∗ /K H L R L R L Land therefore s∗ /K H = H s∗/K L , L since the solution is unique by assump-tion. Because of this, E[ R − s∗/K L ∨ 0] > 1, and therefore participation Lconstraint is satisfied. Finally, one can see that s∗ (q) is continuous in Lq around q = 1, therefore, there is q big enough that supports theexistence of abundance equilibrium.Result 6: Consider a problem of one of the two inspectors in thiscase. Let s∗ denote the bribe that is going to be charged by anotherinspector: s 1 − FR (s+s∗/K ) max sP( R > s+s∗/K ) ⇒ = . s K f R (s+s∗/K ) ∗ 1− F (2s∗/K )In equilibrium, s = s∗ , and therefore 2s = 2 f (2s∗/K) . Thus, the K R Rtotal bribe of a duümvirate, 2s ∗ , is an intersection of a 45◦ degreeline and a curve above the first-order condition curve (4). Similarresult holds for an arbitrary number of inspectors, provided that thesolution is unique.Result 7: First, let’s establish that when g(·) = 0, s∗ = (1 + θ )s∗ , and b gall investors pursue bad projects. The first half of the result followsfrom (12) and (13). The second half follows from the fact that when itpays off to invest in a good project, it also pays off to invest in a badproject:E[(1 + θ ) R − s∗ ∨ 0] = E[(1 + θ ) R − (1 + θ )s∗ ∨ 0] = (1 + θ ) E[ R − s∗ ∨ 0] > E[ R − s∗ ∨ 0] > 1. b b g g
  23. 23. decentralized bribery: the costs, the benefits and the taming 23Observe that substituting s∗ = (1 + θ )s∗ into (13) produces b g 1 s∗ g f R (s∗ ) . g g((1+θ )s∗/s∗ )1/s∗ g ¯ ¯ 1− FR (s∗ ) + (1 + θ ) 1−G((1+θ )sg/s¯∗ ) gThe increasing hazard rate of g(·) and θ > 0 imply that g((1+θ )s∗/s∗ ) g ¯ g(s∗/s∗ ) g ¯ (1 + θ ) ∗ ∗ > , 1 − G ((1 + θ )s g/s ) ¯ 1 − G (s∗/s∗ ) g ¯and therefore the right-hand side of Equation (13) is less than theleft-hand side when s∗ = (1 + θ )s∗ . The assumption that the right- b ghand side of Equation (13) is decreasing therefore suggests that s∗ bshould be less than (1 + θ )s∗ , and therefore bad projects are more gbeneficial for the investor than just (1 + θ ) times the outcome of thegood project.Results 8 and 9: Direct consequence of Equations (14) and (15).Result 10: The derivative of (17) at s = 0 yields: ((( ((((1 − G (Γ/Γ∗ ))(1 − FR (0)) − g(Γ/Γ∗ )Γ/Γ∗ (1 − FR (0)) + (((1 − G (Γ/Γ∗ ))Γ − (((1 − G (Γ/Γ∗ ))Γ. f R (0)( (((( f R (0)( ((((If this is positive, s = 0 cannot be a local maximum, much less aglobal one.Result 11: The first assumption is satisfied when the negative part Γ g(Γ/Γ∗ )of the first term of (18) is dominating ( Γ∗ 1−G(Γ/Γ∗ ) > 1), and so thevalue of the right-hand side of (18) with ordinata is below 0. Theintersection “from below”, like on Figure 8, will produce a localminimum.References Daron Acemoglu and Thierry Verdier. The choice between market failures and corruption. American Economic Review, 90(1):194–211, 2000. Alberto Ades and Rafael Di Tella. Rents, competition, and corruption. The American Economic Review, 89 (4):982–993, 1999. Philippe Aghion, Yann Algan, Pierre Cahuc, and Andrei Shleifer. Regulation and distrust. Technical report, NBER, 2009. Toke S. Aidt. Economic analysis of corruption: a survey. The Economic Journal, 113(491):F632–F652, 2003. Alberto Alesina and George-Marios Angeletos. Corruption, inequality, and fairness. Journal of Monetary Economics, 52(7):1227–1244, 2005. Maria Arbatskaya. Ordered search. The RAND Journal of Economics, 38(1):119–126, 2007.