Snitches Get Riches: a Comprehensive Proposal to

Combat Bribery with Asymmetric Punishment

MacKenzie Pantoja

Keywords: Bribery, Corruption, Asymmetric Punishment, Crime,

Law, Economics.

Abstract: I propose three separate legal regimes built on asymmetric

punishment to deter bribery. In the first regime, a citizen will receive
immunity and a refund of any bribe money they pay if they report the
official who solicited the bribe. In the second regime, the official is
entitled to immunity and can keep the bribe money that they received
if they report the citizen who paid the bribe. While both regimes can
deter bribery, my model shows that the latter regime tends to deter
public service provision. Consequently, I argue the latter regime should
be implemented when the public service is socially harmful, and the
former regime should be implemented when the public service is
beneficial. The third and final regime I argue for combines elements of
both regimes when the parties are missing information that determines
the citizen’s eligibility for the public service, and from that, unsure of
whether the public service is beneficial. Finally, I argue that giving
both parties the right to report eventually can serve as a powerful tool
to combat bribery featuring repeat players.

Dedication

I dedicate this draft to Matthew Burtell, pursuant to a silly promise I made to him during
a game of Catan.

Acknowledgements

I would like to thank Robert Krzyzanowski, Justin Guo, Sebastian Quaade, my brother
Spencer, Zachary Liscow and Susan Rose-Ackerman for their invaluable feedback on this paper.
Most of all, I would like to thank Haydée Marino for helping me turn a good idea into a
(hopefully) comprehensible idea.

1. Introduction

Bribery is a major concern across the world. The UN estimates over 1 trillion dollars in
bribes are paid every year, and the global cost of corruption is about 5% of world GDP.1
1
Global Cost of Corruption at Least 5 Per Cent of World Gross Domestic Product, United Nations (Sept.
10, 2018) [Link]
Countries with high corruption levels tend to be less successful by a number of metrics, and
institutions such as the World Bank sponsor anti-corruption programs. And yet, there is little
evidence that the problem is shrinking worldwide.2
Yet simply referring to all illegal payments to government officials as “bribes” fails to capture
nuances in the nature of corrupt transactions. The Russian language has two words for bribery:3
mzdoimstvo, accepting a bribe to do what you are supposed to do anyway,4 and likhoimstvo,
accepting a bribe to do something that should not be done.5 Kaushik Basu proposes that, in
the case of harassment bribes (mzdoimstvo—bribes for public services that the constituent is
entitled to receive without a bribe), a constituent should be able to get a refund on their bribe
money, keep the public service, and receive immunity if they report the official to authorities.6
The logic of Basu’s proposal is simple. Under the status quo, if a bribe is detected, both the
official and the citizen are punished, regardless of whether one party reports the other. So if I
am an official who solicits a bribe from a constituent, I can be confident that the constituent
will not report me. If they report me, we are both facing a significant potential punishment, so
we each have an incentive to conceal any evidence of the corrupt transaction. Consequently, we
can trust each other. However, under Basu’s proposal, it is harder for me to trust the citizen:
they have a strong incentive to report me if I accept the bribe, so I am less confident that I
will be able to retain the surplus from the deal.
While the proposal is novel, the logic is not. It is an example of asymmetric punishment.
Broadly speaking, asymmetric punishment describes any legal regime in which multiple par-
ties complicit in criminal activity face nonidentical punishments, typically because one party
cooperated with authorities in exchange for leniency (or even a reward). The SEC rewards
whistle-blowers who report financial crimes.7 The United States and Europe implement asym-
metric punishment regimes to deter cartels.8 Even the classic description of the prisoner’s
dilemma is just an example of asymmetric punishment.
Yet Basu limited his suggestion to harassment bribes. Few have seriously entertained imple-
menting his policy to combat collusive bribery (likhoimstvo—bribes for illegal public services).
Even Basu is against it. What is wrong with implementing Basu’s policy to deter collusive
bribes? And if Basu’s policy does not work, is there a way to use asymmetric punishment to
2
Robbie Gramer, Bribery Is on the Rise Worldwide, and It Costs A Lot More Than Just Money, For-
eign Policy (Dec. 1, 2016) [Link]
index-infographic/
3
Susan Rose-Ackerman & Bonnie J. Palifka, Corruption and Government 51 (2nd ed. 2016).
4
For instance, refusing to provide a tax return without a bribe.
5
For instance, paying a bribe to avoid lawful arrest.
6
Karna Basu et al., Asymmetric Punishment as an Instrument of Corruption Control, 18 J. of Pub. Econ.
Theory 831-856 (2016).
7
Office of the Whistleblower, Securities and Exchange Commission, [Link] (last
visited June 29, 2020)
8
Catarina Marvao & Giancarlo Spagnolo, What Do We Know About the Effectiveness of Leniency Poli-
cies? A Survey of the Empirical and Experimental Evidence, SSRN Electronic Journal (Oct 20, 2014)
[Link]/10.2139/ssrn.2511613.

2
fight collusive bribery?
In this paper, I present a framework to deter all forms of bribery with various regimes built
on asymmetric punishment. In the case of harassment bribes, I agree with Basu’s proposal,
which I call the Refund regime. The constituent offering the bribe should be entitled to a refund
of their bribe money, immunity, and the public service if they report the official who solicited
the bribe. The official’s punishment should double.
In the case of collusive bribes, I argue for what I call the Keep regime—a mirror image of
the Refund regime. Under the Keep regime, the official who received the bribe is entitled to
keep the bribe money and receive immunity if they report the constituent who offered it to
them. The constituent’s punishment should double.
While both the Refund regime and the Keep regime can deter bribery, I show the Keep
regime is more likely to deter both collusive and harassment public service provision. Because
only collusive public services are socially harmful, this makes the Keep regime better suited to
fight collusive bribery, and the Refund regime better suited to combat harassment bribery.
Each policy seems to be the converse of the other, so the suggestion that they would have
heterogeneous effects seems counter-intuitive. The asymmetries arise because in my model of
“symmetric” punishment (as well as the real world), punishments are not actually symmetric.
Even if both parties are subject to an identical term of imprisonment, etc.,9 the official must
also forfeit any bribe money that they have received. The citizen must also forfeit the public
service that they received if the bribe was collusive.10 When I enrich my model to reflect these
asymmetries in punishment that exist in the status quo, asymmetries in the regimes emerge.
For instance, suppose the bribe is collusive. Perhaps a firm seeks a permit to operate a power
plant that does not comply with government regulations. Both policies could deter bribery, in
which case, both policies could prevent the permit from being issued. However, bribery is not
always deterred: for some parameter values, the party that does not have the right to report
may be willing to accept the risks from the other party reporting. Here, the Refund regime
creates problems.
Suppose we implement the Refund regime and the official decides the bribe money is worth
the risk they face from the firm potentially reporting. Note that the firm must value the bribe
less than the permit; otherwise, the firm would not have paid the bribe in the first place.
Consequently, the Refund regime would not have any effect unless the firm was allowed to
keep the permit after reporting the official. Consequently, while the Refund regime could deter
bribery, if it does not, the Refund regime could prevent law enforcement from confiscating the
permit. This is sub-optimal, because social welfare is reduced when the firm keeps the permit.
This is not an issue with the Keep regime: not only could it deter bribery, but if the firm gambles
and the official reports, law enforcement has a better chance of detecting and confiscating the
9
18 U.S.C. 201
10
For instance, if the public service was police protection, the citizen does not forfeit the right to be protected
by the police if they are caught paying the bribe. But if the public service is a government contract, they have
to forfeit the contract if they are caught.

3
permit than they would have had in a world without reporting. Thus, the Keep regime both
deters bribery and makes it easier for the official to confiscate the public service, thus making
the Keep regime ideal for collusive bribes.
Now suppose the bribe is harassment. Perhaps an official seeks a bribe from a citizen in
exchange for the citizen’s tax return. If the official will not provide the tax return in a world
without bribery–perhaps it requires too much effort–both policies could deter the bribe and
prevent the tax return from being provided. This an ambiguous outcome, since ideally, the
citizen would receive their tax return. However, when the official would provide the tax return
in a world without bribery–perhaps it enhances their job security–deterring bribery guarantees
that the citizen will receive the tax return.
The latter case seems bizarre: if the official would provide the tax return without bribery,
why is the citizen paying a bribe? It can only occur when the official is a repeat player. In the
first period, the official incurs a cost by refusing to provide the tax return without a bribe in
order to develop a corrupt reputation. This encourages citizens to pay a bribe. If the official is
successful, they get the best of both worlds: bribe money and job security. However, if we can
deter the official from accepting bribe money, society gets the best of both worlds: the official
provides the tax return (to preserve their job) and no bribe is paid.
Basu likely envisioned this outcome when he first proposed his policy. The Refund regime,
in fact, could achieve this outcome: reporting is attractive to the citizen, because if it deters
bribery, the citizen could receive the tax return for free. The Keep regime, however, would fail.
If the official did report, bribery might be deterred, but this is unlikely for the same reason
bribery exists in the first place: the official is a repeat player! So if reporting would deter
bribery, the official almost certainly will not report in order to preserve the corrupt reputation
they have worked hard to cultivate. Consequently, while the Refund regime has ambiguous
effects on public service provision in the harassment context,11 the Keep regime seems to deter
it. This is harmful, because social welfare increases when the tax return is provided.
While the results of my study of the Refund regime against harassment bribes largely repli-
cates prior research,12 to the best of my knowledge, this is the first paper to analyze the Keep
regime in significant detail or argue that it is optimal for collusive bribes. Nevertheless, I believe
the most important contribution this paper makes to the literature lies in my proposal to deter
bribery featuring repeat players. Under both the Refund regime and the Keep regime, bribery
can plausibly persist when the party with the right to report is a repeat player that hopes to
benefit from more corrupt transactions in the future. To combat this, I argue the other party
should eventually be given the right to report if the first party has not done so within some
time frame. The threat of the other party reporting enhances the incentive of the first party to
report. Unless both parties are repeat players, it is challenging to sustain bribery. Moreover,
I will prove that by implementing the Refund regime and the Keep regime sequentially rather
11
It depends on how motivated the Bureaucrat is to provide the public service without a bribe, e.g., does job
security outweigh the urge to be lazy. If it does, the Refund is perfect.
12
Martin Dufwenberg & Giancarlo Spagnolo, Legalizing Bribe Giving, 53 Econ. Inquiry 836-853 (2015).

4
than implementing both immediately, we can accentuate the effects of the preferred policy be-
fore we realize the ambiguous results of the sub-optimal policy. Finally, even if both parties are
repeat players, we could imagine that being reported may lead to forced exit from the bribery
market (e.g., official loses their job, firm can no longer bid on government contracts, etc.). If
this is true, giving both parties the right to report creates significant barriers to maintaining
corrupt relationships with a large number of parties. One party’s exit could have a domino
effect on other corrupt relationships, as they snitch on their past partners in order to preserve
their ill-gotten gains.
The line between collusive and harassment bribes may be blurry—particularly when the
official can provide or withhold the public service as lawful exercise of their discretion. How-
ever, I note that larger bribes are more likely to be collusive, for an official must incur more
costs—both personal and professional—to provide a public service that ought not be provided.
Consequently, I argue that a threshold bribe size should be used to determine bribe type for
discretionary public services. While this rule will not always accurately classify discretionary
public services, it will still often succeed. Without it, legal ambiguities could cripple these
regimes.
Nevertheless, I acknowledge a third type of bribe, never before identified in the literature.
Using the terminology above, a “mixed” bribe describes a scenario in which the official could not
have known what they were “supposed” to do because they were missing relevant information
about the citizen (e.g., performance on a driving test). In the terms of the model, if certain
characteristics that determine a citizen’s eligibility for a public service are unknown, a bribe for
the public service is a mixed bribe. These characteristics can be learned by administering some
sort of test or exam. In the case of mixed bribes, I argue for a policy I call the Dual regime that
combines elements of both regimes. The Dual regime converges toward the preferred regime
if the parties have similar private information concerning the citizen’s eligibility for the public
service.

2. Literature Review
Literature on the use of asymmetric punishment as a means to deter bribery dates back to
a 2011 working paper by Kaushik Basu. Basu proposed that for harassment bribes—bribes for
something that the citizen offering the bribe is already legally entitled to receive—the citizen
should not be punished and should receive a refund on their bribe if the bribe recipient is
prosecuted.13 Dufwenberg & Spagnolo proposed a modification which Basu later accepted—
that the citizen’s actions should only be legal if they report the bribe.14
Further investigation of the Refund regime has yielded mixed—but mostly positive—results.
13
Kaushik Basu, Why, for a Class of Bribes, the Act of Giving a Bribe Should Be Treated as Legal, DEA
Ministry Of Fin., Gov’t, Working Paper 172011, (2011).
14
Martin Dufwenberg & Giancarlo Spagnolo, Legalizing Bribe Giving, 53 Econ. Inquiry 837 (2015); Karna
Basu et al., Asymmetric Punishment as an Instrument of Corruption Control, 18 J. of Pub. Econ. Theory
831-856 (2016).

5
Dufwenberg and Spagnolo find that, in a one-shot game, the Refund regime either has no effect
(and bribes are not paid) or it deters bribery at the cost of the public service being provided.
The latter occurs when government institutions are sufficiently weak. In a repeated game, the
Refund regime deters bribery, but it may encourage public service provision if institutions are
sufficiently effective.15
Abbink (2014) finds that the Refund regime decreases bribery incidence in a lab setting.
However, if the official can retaliate if the citizen is not believed, the effect is weakened.16 In
another lab study, Abbink (2017) finds that allowing both parties to report each other deters
bribery more than allowing just one party to report the other, particularly in repeated games.17
Abbink’s policy resembles the policy I refer to as the Dual regime, which I believe is optimal
for mixed bribery, as detailed below.
Oak proposes a model in which the citizen can choose the type of bribe to offer.18 In this
model, if the Refund regime is implemented for harassment bribes only, there are potential
spillover effects in the collusive domain that make the welfare effects of the Refund regime
ambiguous. Imagine the Refund regime is implemented and an entrepreneur has the choice of
building a plant that is compliant with regulations or non-compliant. A bribe for a compliant
plant is harassment, while a bribe for the non-compliant plant is collusive. Basu’s policy makes
extracting harassment bribes more difficult. This deters harassment bribery, and it could deter
collusive bribery if it encourages entrepreneurs to build compliant plants so they do not have
to pay a bribe. However, the bureaucrat may respond by obstructing compliant plants and
encouraging entrepreneurs to build non-compliant plants. Once a non-compliant plant is built,
the bureaucrat receives a collusive bribe. To the extent this occurs, not only does the policy fail
to deter bribery: it leads to harmful, non-compliant plants. Which effect is stronger depends
on how easily an entrepreneur can appeal the improper denial of a compliant plant. As I will
argue later, implementing the Keep regime in the collusive regime is likely to solve this issue:
in fact, it would likely improve public good quality.
Engel provides experimental evidence that implementing a policy similar to the Refund
regime in the collusive setting could increase bribery.19 In Engel’s setup, there is a potential
hold-up problem: the official may collect the bribe money and nevertheless refuse to provide
the public service. The Refund regime may offer a way around this hold-up problem: a citizen
may attempt to develop a reputation for reporting bribes only when the official takes the money
and does not provide the public service. As I will argue later, implementing the Keep regime
15
Martin Dufwenberg & Giancarlo Spagnolo, Legalizing Bribe Giving, 53 Econ. Inquiry 836-853 (2015)
16
Klaus Abbink et al., Letting the Briber Go Free: An Experiment on Mitigating Harassment Bribes, 111
SSRN Electronic Journal 17-28 (2014).
17
Klaus Abbink & Kevin Wu, Reward self-reporting to deter corruption: An experiment on mitigating collu-
sive bribery, 133 J. of Econ. Behav. & Org. 256-272 (2017).
18
Mandar Oak, Legalization of Bribe Giving when Bribe Type Is Endogenous, 17 J. of Pub. Econ. Theory
580-604 (2015).
19
Christoph Engel et al., Symmetric vs. Asymmetric Punishment Regimes for Bribery, 18 Am. L. & Econ.
Rev. 506-556 (2016).

6
in the collusive setting does not run into this issue. In fact, it can be used to make the hold-up
problem an even larger challenge for the parties to overcome.
Finally, Basu later wrote a more comprehensive paper addressing the Refund regime.20 Basu
finds that the Refund regime generally deters bribery. Whether it increases or decreases public
service provision depends on the bureaucrat’s intrinsic motivation to provide the public service
in the absence of a bribe. While not modeled in Basu’s paper, we could imagine it is a function
of the amount of effort required to provide the public service, the strength of government
institutions monitoring the bureaucrat, moral considerations, etc.

3. Set-up

3.1. A Bureaucrat, a Constituent, and a Public Service

Suppose there are two actors: the Constituent and the Bureaucrat. These will be abbreviated
with C and B in equations. They are bargaining over a Public Service, which will henceforth
be referred to as binary variable P . P = 1 only if the Public Service is provided. Not all public
services should always be provided, however. For example, if the Public Service is a driver’s
license, factors like the Constituent’s age, their performance on a driving test, whether they
own insurance, etc., may affect the utility society gets when the license is provided. I define
y = {y1 , y2 , ...} as the set of characteristics which determine the Constituent’s eligibility for the
license as well as the utility society gets (or loses) if the license is provided. The size of the
bribe exchanged, if any, is denoted by χ.
Social welfare is the sum of utility from the Public Service—zero, when the Public Service is
withheld—as a function of y , and any change in social welfare from a bribe χ being paid. This
is written mathematically as U (P, y, χ) = P w(y) + g(χ). Note that w(y) can be either positive
or negative, depending on the nature of the public service and the circumstances in which it is
(or would be) provided. I assume that all else equal, social welfare is reduced if a bribe is paid,
i.e., ∂U
∂χ
< 0. It is possible that bribery might still be beneficial if it facilitates the provision of a
sufficiently desirable public service (i.e., greasing the wheels). Even then, society would prefer
that the public service be provided without a bribe.
I also assume that the law is well defined. By this, I mean that if the set of eligibility
characteristics y is known, we can place the Public Service into one of three categories:

Required: the Bureaucrat is legally required to grant the Public Service to the Constituent.

Prohibited: the Bureaucrat is legally prohibited from granting the Public Service to the Con-
stituent.

Discretionary: the Bureaucrat can lawfully grant or deny the Public Service at their discre-
tion.
20
Karna Basu et al., Asymmetric Punishment as an Instrument of Corruption Control, 18 J. of Pub. Econ.
Theory 831-856 (2016).

7
 Past literature notes that the line between harassment bribery and collusive bribery is gray.
I agree that it is often unclear whether or not it is socially optimal for an official to provide a
public public service. However, for the purposes of the policies I propose, all we need to know
is whether an official is allowed to provide (or withhold) the public service. This judgment is
much easier than a normative judgment about the social consequences of the official’s actions.

3.2. Types of Bribes

Harassment Bribes and Collusive Bribes
If the Bureaucrat would increase social welfare by providing the Public Service (w(y) > 0),
then a bribe for the Public Service is a harassment bribe. I will also refer to this type of public
service as a harassment public service. I assume that any bribe for a required public service is a
harassment bribe. As an example, suppose that police officers solicit a bribe from a constituent
in exchange for police protection. By law, the police are supposed to protect the constituent
regardless of whether a bribe is paid. For that reason, I presume that social welfare increases
if the citizen is protected, all else equal.
On the other hand, if the Bureaucrat would decrease social welfare by providing the Public
Service (w(y) < 0), then a bribe for the Public Service is a collusive bribe. I will also refer to
this type of public service as a collusive public service. I assume that any bribe for a prohibited
public service is a collusive bribe. As an example, suppose police discover a constituent with
stolen goods, and the constituent offers a bribe in exchange for freedom. By law, the constituent
should not be free. For that reason, I presume that letting the constituent retain their freedom
lowers social welfare, all else equal.
For simplicity’s sake, I will use H as a dummy variable that is equal to 1 when the bribe is
harassment (i.e., when w(y) > 0). Otherwise, H will be 0.

Mixed Bribes
Suppose the Constituent offers a bribe in exchange for a driver’s license prior to taking a
driving test. If the Constituent had passed the test, any subsequent bribe would have been a
harassment bribe—perhaps the Constituent paid the bribe because the Bureaucrat would not
administer the test in good faith. If the Constituent had failed the test, any subsequent bribe
would have been a collusive bribe—perhaps the Constituent paid the bribe because they knew
they would fail the test. But until the test has been taken, we do not know if the Constituent
would have been entitled to the license. I refer to this as a mixed bribe. By assumption, a
mixed bribe only occurs if factor(s) necessary to determine the Constituent’s eligibility for the
Public Service, y , are unknown to at least one party. Thus, the sign of w(y) is unknown. This
does not violate the well-defined-law-assumption: we can determine whether the Constituent is
entitled to the license by administering the exam. Until then, I will say a bribe for the license
is mixed.
Note that if the Constituent or the Bureaucrat has the power to establish the relevant char-
acteristics without some sort of test or inspection, the bribe is not mixed. For instance, suppose

8
the Constituent needs to be a certain age to receive the Public Service. If the Constituent is old
enough, any claim of ignorance by the Bureaucrat is probably a lie and the bribe is likely harass-
ment. The Constituent has no reason to withhold their age if they are old enough. But if they
are not old enough, perhaps the Constituent did withhold their age, and the attempted bribe
was collusive. However, if something that neither party could have known with certainty—like
performance on a test or inspection—determines the Constituent’s eligibility, the bribe is mixed.

A Note on Discretionary Public services

Discretionary public services theoretically fall under a gray area that receives surprisingly
little attention in the literature. Standard practice is to define harassment bribes as bribes for
something a citizen is legally entitled to receive and collusive bribes as bribes for anything else.21
Under this definition, any bribe for a discretionary public service is collusive. Nevertheless, this
classification might be wrong. Suppose a firm bribes a politician to vote “Yea” on a pending bill.
This could be a collusive bribe: perhaps the bill is a means by which the firm can extract rent at
society’s expense. But, under my definition (and the Russian one), it could also be a harassment
bribe: perhaps the bill increases social welfare, but it particularly increases the welfare of the
firm, and the politician demanded a bribe to extract rent from the firm. Consequently, it is
more challenging to justify an assumption about the sign of w(y).
Nevertheless, I agree with the (implicit) assumption in the literature that bribes for dis-
cretionary public services should generally be assumed to be collusive. As I will explain later,
treating discretionary public services as collusive could have positive externalities on public
good quality. Moreover, there are political reasons to favor treating these public services as
collusive. Nevertheless, larger bribes are more likely to be collusive, so a numerical monetary
cutoff could be an imperfect but still useful (and clear) proxy for bribe type. Such a cutoff is
only needed for discretionary public services, of course.

3.3. Utility Functions

Let JB (y) and JC (y) represent all factors affecting the utility of the Bureaucrat and the
Constituent, respectively, from providing or receiving the Public Service under circumstances
y . These include moral considerations, the effect of providing the Public Service on the Bu-
reaucrat’s job security, effort costs associated with providing the Public Service, etc. They will
disappear from the utility functions if the Public Service is not provided. I assume that the
Constituent receives positive utility from the Public Service, i.e., JC (y) > 0. As in Dufwenberg
& Spagnolo, I also assume that the Bureaucrat gets negative utility from expending effort to
provide a socially harmful and illegal public service that could cost them their job if it was

21
Karna Basu et al., Asymmetric Punishment as an Instrument of Corruption Control, 18 J. of Pub. Econ.
Theory 831-856 (2016); Mandar Oak, Legalization of Bribe Giving when Bribe Type Is Endogenous, 17 J. of
Pub. Econ. Theory 580-604 (2015); Klaus Abbink et al., Letting the Briber Go Free: An Experiment on
Mitigating Harassment Bribes, 111 SSRN Electronic Journal 17-28 (2014); Thomas Carson, Bribery, Extortion,
and “The Foreign Corrupt Practices Act”, 14 Phil. & Pub. Aff. 66-90 (1985).

9
discovered.22 By this, I mean JB (y) < 0 if H = 0 (i.e., the Public Service is collusive).
Let FBL and FCL represent the Fine imposed on the Bureaucrat and the Constituent respec-
tively if they are convicted of bribery in a given legal regime L. L is a categorical variable that
represents the four possible legal regimes studied in this paper. For instance, if the Keep regime
is in place, then L = K . The different regimes will be defined later in terms of how they affect
the potential Fines for each party. GB and GC represent dummy variables that reflect whether
the Bureaucrat and the Constituent are guilty of bribery. These variables are not necessarily
equal: if the Bureaucrat asks for a bribe and the Constituent refuses, the Bureaucrat is guilty
of bribery but the Constituent is not.
Finally, pB (GB ) and pC (GC ) represent the probability of being convicted of bribery. If
GB = GC , I assume pB (GB ) = pC (GC ) = p(G). By this, I mean that if one is convicted, the
other must also be convicted, unless one is guilty and the other is not. I assume p(1) > p(0),
meaning that all else equal, one is more likely to be convicted of bribery if they are actually
guilty of it. I do not assume p(0) = 0: later, p will be determined in part by accusations made
by the Bureaucrat and the Constituent, and some regimes may create an incentive for false
accusations.

Incorrupt Utility
First, if no bribe is offered or solicited, the incorrupt outcome is realized. Without bribes,
the Constituent does nothing: the Bureaucrat maximizes their utility by choosing whether or
not to provide the public service.

UBL = JB (y)P − FBL p(0) (1)

UCL = JC (y)P − FCL p(0) (2)

Disagreement Utility
Second, if one or both sides express an interest in exchanging a bribe, but no actual bribe
occurs, one or both parties receive their disagreement payoffs. Simply by making a proposal,
at least one party is guilty of bribery and has increased their probability of punishment, even
before any monetary exchange.

VBL = JB (y)P − FBL p(1) (3)

VCL = JC (y)P − FCL p(1) (4)

Additionally, suppose the Bureaucrat demands a bribe, the Constituent refuses to comply,
and the Bureaucrat is convicted of bribery. At the time of conviction, suppose the Bureaucrat
had not yet provided the Public Service. The Bureaucrat will be removed from their position and
22
A bribe may offset this cost, but it is still a cost

10
be replaced by a new bureaucrat who will provide the Public Service for free with probability
µ if H = 1.23 In short, even if the Bureaucrat doesn’t provide the Public Service, there is a
chance that the Constituent will receive it anyway. If the Bureaucrat demanded a harassment
bribe and got in trouble for it, maybe a new official will come along and provide it for free.
The Constituent’s incorrupt payoff is now the following.

UCL = HGB pB (GB )µ|1 − P |JC (y) + JC (y)P − FCL p(0) (5)

The first term is 0 unless P = 0, H = 1 and GB = 1; otherwise, it is the Constituent’s utility

from the public service weighted by the probability they will receive it when the Bureaucrat
demands a bribe and does not provide the Public Service. For the sake of brevity, I will refer
to it as d, to reflect a potential utility boost the Constituent receives if the Bureaucrat is in the
disagreement game.

Bribe Utility
Lastly, suppose a bribe is exchanged. I assume a bribe will only be offered if JB (y) < 0.
In a one-shot interaction, the Constituent will not offer to pay a bribe if they know that the
Bureaucrat would maximize their utility by providing the Public Service even without a bribe.
This assumption will be relaxed in a repeat player setup. As in Basu, I assume there is no
hold-up problem: P = 1 if χ > 0. Each player’s utility will be denoted by the following:

zBL = JB (y) − FBL p(1) + χ (6)

zCL = JC (y) − FCL p(1) − χ (7)

A bribe will be exchanged if the minimum bribe necessary to make the Bureaucrat better
off than they would be in an incorrupt scenario is smaller than the maximum bribe the the
Constituent is willing to pay for the Public Service. In evaluating this decision, both consider
the potential Fine. These conditions are detailed in Appendix A.1.

3.4. Accusations
Now, suppose each side has the option to accuse the other of accepting or paying a bribe. By
doing so, they raise the probability of conviction but also may change the Fine, FL depending
on the regime in place. To levy an accusation, they must pay administrative cost AB or AC .
The Bureaucrat and the Constituent maximize their utility with respect to accusation
dummy variables aB (γ), aC (γ). γ is a state variable which can take one of three values: t (true),
f (false), or d (true in the disagreement game). For instance, if aC (t) = 1, the Constituent would
23
As before, I assume that an official has no reason to provide a harmful and illegal public service without
the influence of bribe money.

11
make a true accusation given the opportunity. The fact that there are two types of true accu-
sations may seem odd. I separate them because, for instance, if the Constituent is considering
whether or not to report a Bureaucrat who wants a bribe, their decision may be a function
of whether or not they have paid a bribe. As an example, under symmetric punishment, the
Constituent will not report the Bureaucrat if they have already exchanged bribe money. By
doing so, they admit their own guilt and could be punished. Consequently, aC (t) = 0 under
symmetric punishment. However, if the Bureaucrat demanded a bribe and the Constituent has
not paid, the Constituent might report the Bureaucrat’s demand. The Constituent will not be
arrested since they did not pay a bribe, and perhaps reporting the Bureaucrat could get them
replaced. Consequently, perhaps aC (d) = 1. Thus, whether or not it is optimal to levy a true
accusation may be a function of whether a bribe has been paid.
Only one outcome will be realized: I assume one cannot levy a true accusation when the
other side is innocent or a false accusation when the other side is guilty. The truth of the
allegations are a function of the guilt of the other party: so if the Bureaucrat asks for a bribe and
the Constituent refuses, any accusation made by the Bureaucrat is false and any accusation made
by the Constituent is true (although it falls under category d). The Fine and the probability
of conviction are functions of these dummy variables. For instance, the probability that the
Constituent is convicted given accusations aB and aC is denoted by pC (GC , aB , aC ). In these
terms, pB (0, 0, 1) would represent the probability that the Bureaucrat would be convicted given
that they are innocent but the Constituent levied a false accusation. Utility functions with
accusations are given in Appendix A.2.

3.5. Legal Regimes Defined

Recall that L is a categorical variable that represents the legal regime in place and aB and aC
represent whether an accusation was made by the Bureaucrat or the Constituent respectively.
L can take one of four values:

Symmetric Punishment
Symmetric punishment (shortened to S in equations): Both players face an identical baseline
punishment if caught. In addition to the baseline punishment, the Bureaucrat must also forfeit
any bribe money they receive. If the bribe was collusive, the Constituent must also forfeit the
public service. This is the status quo.
Symmetric punishment can be represented mathematically as the following: FCS = F +
JC (y)P |1 − H| and FBS = F + χ. F represents the baseline punishments. The second term
in each equation represents the potential forfeiture of the public service and the bribe money,
respectively, as part of the punishment. FBS and FCS are independent of any accusations.

Refund Regime
The Refund regime (shortened to R in equations): If the Constituent reports the Bureaucrat
for accepting a bribe, the Constituent is immune from prosecution and is entitled to a refund on

12
any bribe money they paid. They are also entitled to keep the public service. The Bureaucrat’s
punishment doubles if the Constituent reports them. The Constituent has no obligation to
report anything if a bribe did not actually occur.24 Consequently, the Constituent cannot
receive their disagreement payoffs. This is Dufwenberg and Spagnolo’s “modified Basu policy.”
I define the Refund regime as follows: FCR (aC = 1) = −χ, FBR (aC = 1) = 2FBS and GC = 0 if
χ = 0. In all other respects, the Refund regime is identical to Symmetric punishment. Note the
first term is negative—the Constituent’s reward is modeled as a negative punishment, in that
they receive money that they did not have at the time of the report.

Keep Regime
The Keep regime (shortened to K in equations): If and only if the Bureaucrat reports the
Constituent’s bribe, the Bureaucrat is immune from prosecution and is entitled to keep whatever
bribe money they received. If the Bureaucrat asks for a bribe and the Constituent refuses, the
Bureaucrat has no obligation to report anything. Thus, the Bureaucrat cannot receive their
disagreement payoffs. The Constituent’s punishment doubles if the Bureaucrat reports them.
I define the Keep regime as follows: FBK (aB = 1) = 0, FCK (aB = 1) = 2FCS and GB = 0 if
χ = 0. Otherwise, the Keep regime is identical to Symmetric punishment.

Dual Regime
The Dual regime (shortened to D in equations): If both parties report each other (within
some time-frame), the bribe is refunded, the Public Service is forfeited if it is not a harassment
public service, and neither party is prosecuted. If only the Constituent reports, then the Refund
regime is implemented. If only the Bureaucrat reports, then the Keep regime. All disagreement
payoffs are eliminated: no one can be punished unless money is exchanged. If the bribe is
mixed, create a procedure/deadline to determine y and potentially award the Public Service.
I define the Dual regime such that it adopts the changes described in the previous two
regimes, with two additional changes: FBD (aB = aC = 1) = χ and FCD (aB = aC = 1) =
JC (y)|1 − H| − χ. If both sides report each other, the Bureaucrat’s only “punishment” is that
the bribe money must be returned, and the Constituent’s only “punishment” is that they must
return the public service if they are not entitled to it—although they still get a refund of any
bribe money they have paid. Additionally, in the case of a mixed bribe (i.e., y is unknown),
the Constituent can reapply for the public service and the interaction restarts, although both
sides have already paid administrative costs.

4. Benchmark Analysis

Assuming symmetric reporting costs, I argue that the Refund regime is the appropriate
policy to deal with harassment bribes, the Keep regime is the appropriate policy to combat
24
Under Symmetric punishment, if the Constituent offers a bribe and the Bureaucrat refuses, the Constituent
is guilty of bribery as soon as they make the offer. However, forcing the Constituent to report this interaction
in order to receive immunity under the Refund regime would accomplish nothing.

13
collusive bribes, and the Dual regime is the appropriate policy for mixed bribes. These results
hold because, while both policies likely deter bribery, they have different effects on public
service provision. The Keep regime is more likely than the Refund regime to deter public service
provision in both settings. In the case of harassment bribes, this is sub-optimal: social welfare
increases when the Public Service is provided. In the case of collusive bribes, this is optimal
outcome: social welfare decreases when the public service is provided. In a repeated interaction,
the Refund regime has ambiguous effects on the rate of public service provision—although it
certainly does not deter public service provision as much the Keep regime. This is better suited
to the harassment setting, where a potential increase in public service provision is desirable.
In the case of mixed bribes, I argue for the Dual regime. If the parties privately believe that
the bribe would be harassment, the Dual regime actually becomes the Refund regime under
conservative assumptions. If they privately believe that the bribe would be collusive, the Dual
regime tends to converge toward the Keep regime, although mixed strategy equilibria may
emerge. In the event there is sufficient disagreement over what kind of bribe is being offered,
both sides may report and we can establish a procedure to determine y so we can identify which
regime is appropriate.

4.1. One-shot Game

The first order conditions with respect to all possible accusations are given in Appendix A.3.
For simplicity’s sake, I assume p(0, 0, 0) = 0, i.e., if no bribe took place and no false accusation
was levied, the probability of conviction is 0.

4.1.1. Symmetric Punishment

Under Symmetric punishment, all accusations lower the utility of all parties, with one
possible exception: if the Bureaucrat demands a harassment bribe and the Constituent refuses,
the Constituent may report the demand with the hope that the Bureaucrat will be replaced by
someone less corrupt.25 Bribery will persist if the following inequality holds:

−JB (y) + F p(1, 0, 0)(2 − p(1, 0, 0))

≤ JC (y) − d − JC (y)|1 − H|p(1, 0, 0) (8)
1 − p(1, 0, 0)

See Appendix A.1.1 for details.

4.1.2. The Refund Regime

Under the Refund regime, the same potential accusation outlined in section 4.1.1 remains a
possibility, subject to identical terms. However, under the Refund regime, both true and false
accusations by the Constituent become a possibility.26 Since FCR (1) = −χ and p(0, 0, 0) = 0 by
assumption, if the following inequality holds, the Constituent will levy a false accusation:
25
See Appendix A.3.1
26
See Appendix A.3.2.

14
 (χp(0, 0, 1) − AC > 0) =⇒ aC (f ) = 1 (9)

Under equation 9,27 if the potential “refund” of a fabricated bribe weighted by the probability
it is realized exceeds the administrative costs of making the accusation, the Constituent will
falsely accuse the Bureaucrat of bribery.
If the following inequality holds, the Constituent will truthfully report the Bureaucrat under
the Refund regime:

(χp(1, 0, 1) + (F + JC (y)|1 − H|)p(1, 0, 0) − AC > 0) =⇒ aC (t) = 1 (10)

Here, if the reward from a true accusation and the potential punishment avoided from an
unreported bribe exceeds the administrative costs of making an accusation, a true accusation
will be levied.
Consider the strength of the above incentives. Equation 10, the Constituent’s incentive
to levy a true accusation, is strictly larger than equation 9, their incentive to levy a false
accusation. The second term in 10 is positive and the first term is larger because a true
accusation is presumed to be more likely to result in a conviction than a false accusation.
Consequently, under the Refund regime, the incentive to levy a true accusation is strictly larger
than the incentive to levy a false one.28
Suppose only true accusations persist. Bribery will persist if the following holds:

(−JB (y) + 2F p(1, 0, 1))(1 − p(1, 0, 1))

+ AC ≤ JC (y) − d (11)
1 − 2p(1, 0, 1)

Note that p(1, 0, 1) < 0.5 for bribery to persist; otherwise, only negative bribes would satisfy
the Bureaucrat’s participation constraint.29 This is because the Bureaucrat’s baseline punish-
ment includes forfeiture of the bribe, so if the punishment doubles, the Bureaucrat will not
accept a bribe unless they have at least a 50% chance of keeping it.

27
In reality, if a false accusation is believed, χ will be zero but some manufactured number will be recovered
anyway. However, I still refer to the manufactured number as χ because I assume it is subject to the same con-
straints as χ: appendix equations A.1 and A.2, the Bureaucrat and the Constituent’s participation constraints.
In essence, if the Constituent makes a false accusation, they still have to choose a realistic number.
28
In this model, increasing AC could be an effective mechanism to filter out false accusations. But we could
enrich the model, such that the Constituent can pay a cost to preserve evidence to increase the credibility of
their accusation. If preserving real evidence is less costly than fabricating fake evidence, real accusers would
preserve more evidence. This could act as a filter against false accusations that does not have the drawback of
potentially discouraging real accusers from coming forward.
29
See Appendix A.1.2.

15
4.1.3. Comparing the Refund Regime and Symmetric Punishment
How does this compare to Symmetric punishment? For brevity’s sake, say p(1, 0, 0) = p̂ and
p(1, 0, 1) = (p̂ + ν) (since an accusation makes detection more likely). Equation 11 reflects the
conditions required for bribery to persist under the Refund regime with reporting. Equation
8 represents the same for Symmetric punishment without reporting. How do they compare?
If the following inequality holds, asymmetric punishment deters bribery more effectively than
Symmetric punishment.

−JB (y) + F p̂(2 − p̂) (−JB (y) + 2F (p̂ + ν))(1 − (p̂ + ν))
+ p̂JC (y)|1 − H| ≤ + AC
1 − p̂ 1 − 2(p̂ + ν)
JB (y)(p̂2 + ν + p̂ν) − F (p̂2 + 2ν(1 − (1 − p̂)(ν + p̂))
=⇒ + p̂JC (y)|1 − H| ≤ AC (12)
(1 − p̂)(1 − 2p̂ − 2ν)

Consider the harassment setting, where the second term is equal to 0. The first term
represents factors that would increase bribery surplus under the Refund regime. However, it
is negative for p̂ > 0, even when ν = 0.30 Since AC > 0, the inequality always holds, so in
the one-shot harassment setting, the Refund regime does a better job of deterring harassment
bribery than Symmetric punishment, albeit at the cost of the public service provided. This
replicates the results of Dufwenberg & Spagnolo, who find Basu’s policy deters bribery at the
cost of the public service provided in a one-shot interaction.
In the collusive context, the effects are more ambiguous. The second term represents one fac-
tor that could increase post bribery surplus: the Constituent no longer has to forfeit a collusive
public service if they are detected. Nevertheless, under conservative parameter assumptions,
this inequality is negative.31 However, even if the Refund regime generally deters bribery in the
collusive setting, it has ambiguous effects on the rate of public service provision. This differs
from Dufwenberg & Spagnolo. The difference lies in the fact that, in my model of Symmetric
punishment, if a bribe is detected, punishment includes forfeiture of any collusive public service.
So under Symmetric punishment, even when bribery persists, the collusive public service will
be rescinded with probability p̂. The Refund regime essentially discards this possibility: by
reporting the bribe, the Constituent can keep a collusive public service even when the bribe
would have been detected without the Constituent’s accusation. Since a potential increase in
collusive public service provision is harmful, this outcome is sub-optimal.
Naturally, one might ask: why must the Constituent be allowed to keep collusive public
services after reporting the bribe under the Refund regime? The issue is the Constituent will
not report unless they can keep the public service. This can be demonstrated mathematically,
30
Intuitively: if the Constituent reports, the Refund regime increases equilibrium bribe size because the
Bureaucrat faces more risk, since their punishment doubles. But since the punishment includes forfeiture of
the bribe, increasing bribe size increases expected punishments, even under the conservative assumption that
reporting has no effect on punishment probability. Additionally, JB (y) is negative.
31
See Appendix B.

16
but the intuition is obvious: the Constituent must value the public service at least as much or
more than the bribe money, or they would not have paid the bribe in the first place.32 So a
simple refund of the bribe money at the cost of the public service will not lead the Constituent
to report the bribe.
If false accusations persist in equilibrium, the Refund regime is still a more effective deter-
rent against harassment bribery than Symmetric punishment if true accusations increase the
probability of detection more than false accusations.33 This seems reasonable, although it is
not assumed in the model. While true accusations are more likely to be believed, the baseline
detection probability of real bribes is higher, so the effect of true accusations relative to false
accusations is ambiguous. Still, it is probably easier for the Constituent to bolster true accusa-
tions with evidence, and this probably will make true accusations more powerful. Nevertheless,
even when true accusations are more powerful than false accusations, the deterrent effect will
be weaker if false accusations are present.

4.1.4. A Note on Bribe Size

The Refund regime, even when it deters bribery, seems to have dubious effects on the Con-
stituent’s welfare, even in the harassment setting, where the Constituent’s welfare is a priority.
The regime deters public service provision, although I will show this does not necessarily hap-
pen in a repeated game. When there is reporting, the Constituent might be better off if they
can get a refund on their bribe. But the Constituent might also be worse off: because of the
reporting risk, the Bureaucrat can demand a larger bribe.
However, it is not necessarily the case that bribe size will increase. If reporting under
equation 10 would deter bribery, the Bureaucrat might respond by asking for a smaller bribe. If
the Bureaucrat asks for a sufficiently small bribe, the Constituent may decide not to report on
the grounds that reporting is not worth the effort for such a small refund. Such an equilibrium
may improve the Constituent’s (as well as social) welfare relative to Symmetric punishment—
even though it does not actually deter bribery.

4.1.5. The Keep Regime

Under the Keep regime, the only accusation that might maximize anyone’s utility is a true
accusation by the Bureaucrat.34 If the following inequality holds, the Bureaucrat will truthfully
report the Constituent if the Constituent pays a bribe:

32
The model would need to be adjusted to reflect the fact that, if the Constituent reports the bribe, the public
service will be confiscated no matter what. E.g., it’s impossible for the Constituent to claim that they bribed
the Bureaucrat for a government contract while concealing the existence of the government contract. Moreover,
allowing the Constituent to keep the government contract only if the accusation is not believed creates an absurd
incentive to make truthful accusations sound as ridiculous as possible.
33
See Appendix A.1.2.
34
See Appendix A.3.3.

17
 ((F + χ)p(1, 0, 0) − AB > 0) =⇒ aB (t) = 1 (13)

Intuitively, this means that the Bureaucrat will levy an accusation if the expected costs of
a potential punishment exceed the administrative costs of reporting the bribe.
When the Bureaucrat reports, bribery will persist if the following inequality holds:

−JB (y) + AB + 2F p(1, 1, 0) ≤ JC (y) − JC (y)|1 − H|2p(1, 1, 0) (14)

See Appendix A.1.3 for details.

4.1.6. Comparing Symmetric punishment and the Keep regime

How does this compare to Symmetric punishment? Once again, say p(1, 0, 0) = p̂ and
p(1, 0, 1) = p̂ + ν . Now suppose p(1, 1, 0) = p̂ + φ, i.e., when the Bureaucrat makes an accusation,
detection probability rises by φ.35 Compare equations 8 and 14, which govern bribery in Sym-
metric punishment and the Keep regime (given reporting), respectively. The Keep regime is a
more effective deterrent against bribery if the following inequality holds:

−JB (y) + F p̂(2 − p̂)

+ p̂JC (y)|1 − H| + d ≤ −JB (y) + AB + 2F (p̂ + φ) + JC (y)|1 − H|2(p̂ + φ)
1 − p̂

−JB (y)p̂ + F p̂2

=⇒ − 2φF − (p̂ + 2φ)JC (y)|1 − H| + H(p̂ + aC (d)ν)JC ≤ AB (15)
1 − p̂

In the collusive setting, the Keep regime deters bribery. The first term will be positive, the
second and third terms will be negative, and the fourth term will be 0 for all collusive bribes.
The third term will always offset the first term. Notice that post-bribe surplus is a function
of detection probability and the size of the punishments. Even if the increase in the former is
negligible, the latter will increase. For collusive bribes, the Keep regime increases punishments
by the value of the public service while decreasing them by the value of the bribe. The former
will dominate: otherwise, the Constituent is paying a bribe for a public service that they value
less than the bribe money that they paid. The Keep regime deters public service provision
because it deters bribery. Further, reporting makes it more likely that law enforcement will be
able to confiscate the Public Service.
The Keep regime also eliminates the issue Engel objected to in his experimental critique of
Basu’s policy in the collusive setting.36 Engel presents a model in which Basu’s policy can be
35
I do not use ν for the Bureaucrat’s accusations, because I do not assume the Constituent’s accusations and
the Bureaucrat’s accusations are equally credible. In fact, since the Bureaucrat has no incentive to levy a false
accusation, I suspect φ > ν.
36
Christoph Engel et al., Symmetric vs. Asymmetric Punishment Regimes for Bribery, 18 Am. L. & Econ.

18
used as a way around a hold-up problem. If the Refund regime (or something like it, as in Engel)
is used in the collusive setting, the Constituent could develop a reputation for only reporting
bribes when the Bureaucrat pockets the money and does not provide the Public Service. The
Constituent could use the threat of reporting to encourage the Bureaucrat to actually provide
the Public Service. Thus, the Refund regime could increase collusive bribery. But the Keep
regime eliminates this threat: if the Bureaucrat decides to take the money and withhold the
Public Service, the Constituent has no way to retaliate. We could, in fact, even modify the
Keep regime so as to make the hold-up problem even more significant. If the Bureaucrat accepts
a bribe for a collusive public service, perhaps the Bureaucrat should not even be required to
report the Constituent as long as the Bureaucrat does not provide the public service. As an
example, if somebody, for instance, tries to bribe a judge for an unjust ruling in a court case,
perhaps the judge should not even be required to report the bribe as long as they do not grant
the unjust ruling. The hold-up problem is assumed away in my paper, so this is an area for
future research. Nevertheless, it is still worth acknowledging the Keep regime’s potential to
further increase frictions in the collusive bribery market.
In the harassment setting, the effects of the Keep regime are ambiguous. The first term in
equation 15 is positive, the second term is negative, the third term is zero, and the fourth term
(aka d) is positive. The ambiguity stems from two factors.
First post-bribe surplus is a function of punishment likelihood and the size of the punishment
if it is imposed. For harassment bribes, the Keep regime increases the former at the expense
of the latter. The total punishment no longer includes the bribe money, but if this is offset by
the increased probability of punishment, then the Keep regime could deter harassment bribes.
The second factor contributing to the ambiguity is that the Keep regime eliminates d, the
fourth term in equation 15. Under the Keep regime, even if the Bureaucrat demands a bribe,
they are not guilty of bribery unless they have accepted money. Under Symmetric punishment,
the Bureaucrat could have been fired and replaced for making this sort of demand, which is
reflected in d. Consequently, under Symmetric punishment, d acts as a lite version of the
Refund regime.37 Even if the Constituent and the Bureaucrat are both better off when a bribe
is exchanged, before that can happen, the Constituent must be given the opportunity to defect
and report the Bureaucrat with the hope that the Bureaucrat’s replacement will provide the
Public Service without asking for a bribe. If defection is a sufficiently attractive option for the
Constituent, the Bureaucrat will not cooperate in the first place. The Keep regime prevents
this from happening by allowing the Bureaucrat to demand bribes. In doing so, it may increase
harassment bribery.
Nevertheless, I suspect it will ultimately deter harassment bribery. It is difficult to justify
a low value for φ, precisely because the Bureaucrat has no incentive to levy a false accusation.
Of course, it would still come at the cost of the public service being provided.

Rev. 506-556 (2016).

37
It is strictly less powerful: see Appendix C.

19
4.2. Repeated Game
Suppose we are in a repeated game. Let δ represent each party’s discount factor. There
are five strategies in total that parties could adopt which may not be utility maximizing in a
one-shot game, but may be utility maximizing in a repeated game because it alters the behavior
of the other party.38

4.2.1. Harassment Bribery when JB (y) > 0

In the one-shot game, when JB (y) is positive, if the Bureaucrat demands a bribe, the Con-
stituent can call the Bureaucrat’s bluff. The Constituent will refuse to pay with the knowledge
that the Bureaucrat would maximize their utility by providing the Public Service even without
a bribe. However, in the repeated game, the Bureaucrat might be able to establish a corrupt
reputation by refusing to provide the Public Service without a bribe (even though establishing
said reputation will hurt them in the short term). If they do, the Constituent cannot call the
the Bureaucrat’s bluff and may acquiesce to the Bureaucrat’s demands. Call this “Strategy 1.”
The Bureaucrat will implement Strategy 1 if the following inequalities hold:

JB (y) + FBL (0)(p̂ + νaC (d)) <

δ
(χ − FBL (aB (t))p(1, aB (t), aC (t)) − AB aB (t) + FBL (0)p(0, 0, aC (f )))
1−δ
(16)

JC (y) − FCL (aC (t))pC (1, aB (t), aC (t)) − AC (aC (t) − aC (f )) − χ >
d(aC (d))−FCL (aC (f ))p(0, 0, aC (f ))
(17)

The first inequality is the Bureaucrat’s participation constraint. The left-hand side of the
first inequality represents the forgone utility in period one that the Bureaucrat loses by refusing
to provide the Public Service. This is the sum of the utility from the Public Service plus the
potential punishment the Bureaucrat is facing if they are caught (unsuccessfully) demanding a
bribe. The right hand side represents the potential utility gain the Bureaucrat would receive
from bribes in all future interactions.
The second inequality is the Constituent’s participation constraint. Even if the Bureaucrat
is willing to provide the public service when given a bribe, the Constituent will not pay the
bribe regardless unless equation 17 holds. The left-hand side represents the Constituent’s utility
if a bribe is paid and a public service is provided, minus any extra administrative costs if the
Constituent decides to report. The right-hand side represents the Constituent’s baseline utility
when no bribe is exchanged.

38
Two of these strategies are given in Appendix D.1.

20
 These inequalities are of great importance. If we can prevent either of them from holding,
we can deter bribery without deterring public service provision—an unambiguously positive
outcome in the harassment context. But we can go further. Suppose there are two types of
constituents. One has a high willingness-to-pay (WTP) for the Public Service, while the other
has a low WTP. Suppose that only the former is willing to pay a bribe, but the Bureaucrat
cannot price discriminate between the two. The Bureaucrat must refuse to provide the Public
Service to low WTP constituents in order to keep receiving bribes from those with a high
WTP. Equation 16, the Bureaucrat’s participation constraint, would need to be modified to
reflect the cost low WTP constituents impose on the Bureaucrat. Essentially, the Bureaucrat
faces an additional cost when they have to deny the Public Service to those who cannot afford.39
However, the Bureaucrat may still demand bribes if there are enough high WTP constituents
to offset them. Yet if a regime can violate either player’s participation constraint (equations
16 and 17), that regime will not just deter bribery—it will increase public service provision for
low WTP consumers. Deterring bribery while increasing public service provision is the ideal
outcome in the harassment context.

4.2.2. Protecting a Corrupt Reputation

In the Refund regime, we can imagine a scenario where reporting the Bureaucrat is worth
the effort for the potential refund, but if the Constituent reports, other officials will not ask for
bribes from them in the future. This is a problem when JB (y) < 0, because the Bureaucrat will
not provide the Public Service without a bribe. Mathematically, when equation 9 holds but
equation 11 does not, and the Constituent is a repeat player, they might set aC (t) = 0 because
they want other officials to trust them. Call this “Strategy 2.”
The Constituent will implement Strategy 2 when the following inequality holds:

δ
χ(p̂ + ν) − AC + (F + |1 − H|JC (y))p̂ < (JC (y) − (F + |1 − H|JC (y))p̂ − χ) (18)
1−δ

The left-hand side represents the utility loss from not reporting the Bureaucrat in the first
period. It is the sum of the potential bribe money that they may have been able to recover and
the potential punishment they are facing if the bribe is detected anyway, minus administrative
costs they avoided. The right-hand side represents the discounted utility gain from being able
to offer bribes to the Bureaucrat in the future.
Finally, in the Keep regime, we can imagine a scenario where reporting is a good decision
for the Bureaucrat because it helps protect themselves from a potential punishment. But if the
Bureaucrat reports, they lose the opportunity to get bribes from the Constituent in the future.
So the Bureaucrat may hide the bribe with the hope of soliciting more bribes from the citizens
39
Alternatively, perhaps those who cannot afford to pay a bribe do not even both applying for the Public
Service, in which case, the Bureaucrat faces no additional cost.

21
in the future. Mathematically, if equation 13 holds but equation 14 does not, the Bureaucrat
could set aB (t) = 0 to make sure citizens feel comfortable bribing them. This is essentially
a mirror image of Strategy 2. Call it “Strategy 3.” The Bureaucrat will implement it if the
following holds:

δ
(F + χ)p̂ − AB < (min(JB (y), 0) − (F + χ)p̂ + χ) (19)
1−δ

The left-hand side represents the costs of not levying an accusation in the present. It is
equal to the potential punishment that the Bureaucrat is facing minus the administrative costs
the Bureaucrat would pay by reporting. The right-hand side represents the utility gain from
all future bribes. The first term may seem confusing. But the right-hand side is not simply the
Bureaucrat’s utility in a bribery equilibrium: it is their utility in a bribery equilibrium minus
any utility they would have received in the incorrupt equilibrium. This represents the gain
from future bribes. If providing the Public Service lowers the Bureaucrat’s utility, JB (y) is a
cost that the Bureaucrat must incur only when they accept a bribe. If providing the Public
Service raises the Bureaucrat’s utility, it is a benefit the Bureaucrat receives regardless. The
second term represents potential punishments, while the third term represents potential bribe
money.
Theoretically, each party could also commit to levying false accusations so as to weaken
the effect of their true accusations in the context of equations 11 and 14 (joint participation
constraints under the Refund regime and the Keep regime). However, this seems silly. The
other party could respond to this strategy by refusing to cooperate with people who make
false accusations—the false accusations hurt them as well. Moreover, if someone levies false
accusations to the point that they develop a reputation for it, law enforcement may intervene.40

4.2.3. Fighting Strategy 1 with Symmetric punishment, the Refund regime, and the
Keep regime
When JB (y) < 0, the repeated game and the one shot game differ in just one respect: for
all parameter values in which the Refund regime and the Keep regime deter bribery and public
service provision, the effect can be mitigated by implementing strategies 2 and 3, respectively.
Consequently, the effects of each policy are weakened, although there is little reason to expect
a sign change.
However, when JB (y) > 0 and the Bureaucrat is implementing Strategy 1, the results are
more interesting. Under Symmetric punishment, the only accusation which may maximize
anyone’s utility is aC (d) = 1. Bribery will persist if the following holds:

40
Even if law enforcement is lazy and ineffective, this individual might be prosecuted because law enforcement
finds them annoying.

22
 F (1 − δ)(JB (y) + F (p̂ + νaC (d)))
JC (y)(1 − µ(p̂ + νaC (d))) + AC aC (d) − F p̂ > + (20)
1 − p̂ (1 − p̂)δ

See Appendix D.2 for details.

How does the Refund regime perform? Remember that the Refund regime and Symmetric
punishment are identical unless the Constituent decides to report a bribe. Recall that the
Bureaucrat’s punishment only doubles if the Constituent reports—even when no bribe money
is exchanged.41 Bribery will persist if the following holds:

JC (y)(1 − µ(p̂ + νaC (d))) − (1 − aC (d))AC

>
1 − p̂ − ν
2F (1 − δ)(JB (y) + F p̂(1 + aC (d)) + 2F νaC (d))
+
1 − 2(p̂ + ν) δ(1 − 2(p̂ + ν))
(21)

See Appendix D.3 for details.

How does this compare to Symmetric punishment? The presence of aC (d) makes the math
complex, but I show in Appendix D.4 that the Refund regime deters bribery more effectively
than Symmetric punishment. By doing so, the Refund regime both deters bribery and increases
public service provision when Strategy 1 is in play, an unambiguously positive outcome.
How does the Keep regime perform? Recall that aC (d) = 0 under the Keep regime—or
even null, because the Constituent cannot accuse the Bureaucrat of doing something that the
Bureaucrat is allowed to do. Moreover, recall that the Keep regime is otherwise indistinguishable
from Symmetric punishment unless the Bureaucrat reports.42 . Suppose the Bureaucrat does
report. Bribery will persist if the following holds:43

1−δ
JC (y) − 2F (p̂ + φ) > JB (y) + AB (22)
δ

How does this compare to Symmetric punishment, given by equation 20? The results are
ambiguous. Because χ is no longer included in the punishments, the total pool of punishments
shrink, which could increase bribery. At the same time, if the increase in the probability of
punishment is sufficiently large, it will offset the fact that the punishments themselves are
41
I also assume no false accusations in equilibrium. As before, if false accusations persist, these results hold
as long as true accusations increase the probability of detection at least as much as false accusations.
42
Otherwise is relevant: given sufficiently low administrative costs, these accusation can combat Strategy 1,
and by eliminating them the Keep regime could increase bribery and lower public service provision. See, e.g.,
Strategy 4 in Appendix D.1.
43
See Appendix D.5.

23
smaller. Yet the Constituent had some incentive to levy a truthful accusation under Symmetric
punishment (in the disagreement case), which weakens the potential increase in the probability
of punishment. Still, because the Bureaucrat has no incentive to levy a false accusation, the
Bureaucrat’s accusations are probably more credible.44
I suspect φ is large enough, that when reporting occurs, the Keep regime would deter
harassment bribery, although perhaps less than the Refund regime. But there is a bigger
concern: reporting almost certainly will not occur if it would deter bribery. This is because
Strategy 1 is predicated on the Bureaucrat being a repeat player that wants bribes in the future.
As a repeat player, the Bureaucrat likely will not report if it prevents them from receiving future
bribes.
If reporting would deter bribery, bribery must exist when there is no reporting. For that to
happen, the Bureaucrat’s participation constraint under Symmetric punishment must hold.45
This is given in Appendix D.2. Recall that equation 19 represents whether the Bureaucrat will
choose not to report under the Keep regime with the hope of soliciting more bribes in the future.
This is also referred to as “Strategy 3.” I reproduce the two equations below.

The Bureaucrat’s Strategy 1 participation constraint under Symmetric punishment:

δ
JB (y) + F (p̂ + νaC (d)) < (χ − (F + χ)p̂)) (23)
1−δ
The Bureaucrat’s Strategy 3 participation constraint:
δ
(F + χ)p̂ − AB < (min(JB (y), 0) − (F + χ)p̂ + χ) (24)
1−δ

If the first equation holds, then the Bureaucrat is willing to take a risk by not providing the
Public Service in period one in order to solicit more bribes in the future. If the second equation
holds, the Bureaucrat will take a risk by not reporting the Constituent in order to solicit more
bribes in the future.
Note the similarity between equations 23 and 24. For Strategy 1 to even be on the table
JB (y) is positive by assumption, so the right hand sides are identical. If equation 23 holds,
equation 24 will also hold unless bribe size is sufficiently large. This is unlikely given that the
Keep regime with reporting lowers bribe size because the Bureaucrat faces no risk. Thus, if
equation 23 holds—and it must hold if reporting would deter bribery—it is likely that equation
24 will hold as well. If equation 24 holds, the Bureaucrat will not report if reporting would
44
While not in the model, I can imagine one plausible counterexample. Suppose a police officer arrests a
citizen. The officer notices that the citizen has money on their person. The officer may benefit from confiscating
the money and claiming the citizen tried to use the money to bribe them. Much of this could have happened
under the status quo, but under the Keep regime, the officer may not have to pretend that they do not have
the citizen’s money. This is a legitimate concern, but it is also an extreme scenario. The Bureaucrat will rarely
have the ability to take money from the Constituent with physical force.
45
The Keep regime without reporting is just Symmetric punishment.

24
deter bribery.
This is why the Keep regime fails to prevent Strategy 1, thus failing to deter bribery.
Intuitively, the premise behind this sort of harassment bribery is that the Bureaucrat is a repeat
player who is willing to pay a cost by not providing the Public Service in order to develop
a corrupt reputation. But if the Bureaucrat decided to report the constituents who bribed
them—and reporting deterred scared the constituents away—that would defeat the purpose
of developing a corrupt reputation in the first place! So if reporting would deter bribery, the
Bureaucrat will almost certainly not report, and the Keep regime will fail.
Making matters worse, the Keep regime could increase bribery and reduce public service
provision by eliminating aC (d).46 Under Symmetric punishment, the Constituent still has some
recourse if the Bureaucrat demands a harassment bribe and the Constituent refuses. The threat
of the Constituent reporting the Bureaucrat’s demand might reduce bribery. However, the Keep
regime eliminates this option by allowing the Bureaucrat to shamelessly beg for a bribe without
risking punishment. So, under reasonable conditions, the Keep regime could increase bribery
and deter public service provision, the worst possible outcome when dealing with harassment
bribes. The Refund regime, however, performs much better.

4.2.4. Bribe Size when JB (y) > 0

Note that the Bureaucrat’s minimum acceptable bribe (given by equations D.3, D.5, and
D.11 in the Appendix) are increasing in JB (y), which would suggest a higher minimum bribe size
for public services that impose negative costs on the Bureaucrat. This is strange: why would
the Bureaucrat have a higher minimum bribe size when they are providing a public service that
they want to provide? The Bureaucrat has to incur a cost by not providing the public service
to establish their corrupt reputation. The larger that cost is, the larger the bribe they need
to offset it. However, these equations probably should not be used to determine equilibrium
bribe size. In real life negotiations, there is likely an inter-temporal bargaining constraint: the
Bureaucrat cannot use costs incurred in the past as leverage to demand a larger bribe in the
present. Consequently, the Nash bargaining solution should still be governed by equations seen
in the one-shot game, which are decreasing in JB (y).
It is possible that the Nash bargaining solution will violate the Bureaucrat’s period 1 partic-
ipation constraint, in which case, I overstated the parameter values under which bribery may
persist. For instance, suppose there are two periods, and the Bureaucrat must incur a cost of
6 by not providing the Public Service in period 1 to establish a corrupt reputation, and the
Constituent values the public service at 8. Both parties would be satisfied with a bribe of 7
in period 2. But the Bureaucrat cannot use the cost they incurred in period 1 as leverage to
demand a larger bribe in period 2: perhaps the Bureaucrat is better off with any bribe larger
than 0 in period 2. If this is true, they may “meet in the middle” and exchange a bribe of
46
Note that given sufficiently low administrative costs, the joint participation constraint under Symmetric
punishment is more likely to be violated when aC (d) = 1, i.e., equation D.8 in the appendix is less likely to hold
than equation D.10.

25
4 in period 2. But this would not be enough to offset the cost that the Bureaucrat incurred
in period 1. In such a scenario, bribery may not occur even if equations 20, 21, or 22 hold.
However, this has little effect on my results: I show in Appendix E that this effect appears to
be homogeneous across regimes.

4.3. Bolstering the Policies

The above shows the following: in a repeated interaction for a harassment public service, the
Refund regime deters bribery but has ambiguous effects on public service provision. Whether
it increases or decreases public service provision depends on the sign of JB (y). In the collusive
context, the Refund regime deters bribery and it still has ambiguous effects on the rate of
public service provision. While bribes are less frequent, when they do happen, public service
provision increases, because the Public Service is less likely to be confiscated. Conversely, the
Keep regime has ambiguous effects on harassment bribery but it seems likely to deter public
service provision. When JB (y) > 0, it seems likely that it will increase harassment bribery and
deter public service provision, the worst possible outcome. When JB (y) < 0, it seems likely
it will deter both. But, in the collusive context, the Keep regime has unambiguously positive
effects: it deters both bribery and public service provision.
As Basu intended, I believe this shows the Refund regime is a better policy than the Keep
regime and Symmetric punishment to combat harassment bribery. Since the Keep regime
achieves unambiguously positive results in the collusive context, I believe it is the optimal
policy to combat collusive bribes.
However, each policy can be weakened when the reporting party is a repeat player. To
combat this, I make another novel proposal: if the party with the right to report has not
done so within some fixed period of time, the other policy is implemented simultaneously. For
instance, if the Keep regime is implemented to combat collusive bribes, and the Bureaucrat has
not reported the Constituent within, say, a week of the bribe, the Constituent should be given
the right to report the Bureaucrat, and if they do so before the Bureaucrat reports them, the
Refund regime is implemented.
To be clear, I will not argue that each policy should be implemented simultaneously as soon
as the bribe is paid. They shall be implemented sequentially. Even if they were implemented
“simultaneously,” I assume, in practice, they would be implemented sequentially. For instance,
if the Bureaucrat, due to their prior experience and connections, is able to make a phone call and
accuse the Constituent faster than the Constituent is able to accuse the Bureaucrat, the Keep
regime is functionally being implemented before the Refund regime. Rather than allowing such
arbitrary factors determine the order in which the policies are implemented, I argue that they
should be implemented sequentially according to a predetermined procedure that strengthens
the preferred policy for each type of bribe.
To determine the outcome of said policy, we must work backward. Suppose we implement
the Keep regime first, and after a fixed period of time passes with no reporting by the Bureaucrat,

26
the Refund regime is added to the mix. Adding the Refund regime obviously has no effect unless
the Constituent decides to report the Bureaucrat. Our first question: for parameter values under
which the Bureaucrat does not report, may it nevertheless be optimal for the Constituent to
report? Recall that the Constituent and the Bureaucrat’s decision to report are governed by
equations 10 and 13, respectively. They are reproduced below:

(χ(p̂ + ν) + (F + JC (y)|1 − H|)p̂ − AC > 0) =⇒ aC (t) = 1 (25)

((F + χ)p̂ − AB > 0) =⇒ aB (t) = 1 (26)

Assuming symmetric reporting costs, the Constituent’s incentive to levy an accusation under
the Keep regime is strictly stronger than the Bureaucrat’s under the Refund regime, since
equation 25 is strictly larger than equation 26. This is not to suggest the Constituent’s reporting
incentive will always be stronger—asymmetries in reporting costs would probably favor the
Bureaucrat, who may have experience navigating the judicial process. Still, adding the Refund
regime to the mix absolutely could induce reporting when there was none under the Keep
regime.
The effect becomes more significant if we consider repeat games. Suppose the Bureaucrat
only chooses not to report because they are a repeat-player implementing Strategy 3, hoping
to get more bribes in the future. If the Constituent is not a repeat player, the Constituent
is guaranteed to report under symmetric reporting costs. If the Bureaucrat would report in a
one-shot game (a premise underlying Strategy 3), then the Constituent will report if they are
in a one-shot game, since their incentive to report in a one-shot interaction is at least as strong
as the Bureaucrat’s given symmetric reporting costs. So there are certainly scenarios in which
the Constituent would report given the opportunity, even while the Bureaucrat stays silent.
Suppose such a scenario comes to pass and the Bureaucrat is a one-shot player. The Bureau-
crat’s decision to report was governed by equation 25. However, the Bureaucrat’s return from
reporting will change if the Constituent will eventually report them. If the Refund regime is im-
plemented after the Keep regime, and the Constituent will report the Bureaucrat once the Refund
regime is in place, the Bureaucrat will report if the following inequality holds:

(2(F + χ)(p̂ + ν) − AB > 0) =⇒ aB (t) = 1 (27)

Relative to equation 26, the Bureaucrat is facing twice the punishment and a larger proba-
bility of conviction if they do not report. In fact, the Bureaucrat’s incentive to report will be
strictly stronger than the Constituent’s (given by equation 25) for any harassment public service
(assuming symmetric reporting costs), which is to say that if it is optimal for the Constituent to
report a harassment bribe, it must be also be optimal for the Bureaucrat to report it.47 In the
47
While the Keep regime is not designed for harassment bribes, it is still helpful to know what might happen

27
collusive case, the Bureaucrat’s incentive might be weaker than the Constituent’s, depending
on the value of the public service provided and the credibility of the Constituent’s accusation.
However, it seems likely that the Bureaucrat will report.48 At the minimum, the Bureaucrat’s
incentive to report is twice as strong as it was before.
How does this policy compare to the Keep regime alone? Adding the Refund regime affects
bribery in two scenarios. If the Bureaucrat does not report and the Constituent does, then all
of the effects of the Refund regime—good and bad—are realized.49 However, if the Bureaucrat
decides to report only because of their fear that the Constituent will, any effect that comes from
the Bureaucrat reporting is accentuated. Given that the Bureaucrat will almost certainly report
under equation 27 (since we are assuming that the Constituent will report under equation 25),
when the Bureaucrat is a one-shot player, the latter seems much more likely than the former.
Thus, we accentuate the Keep regime.
Now suppose the Bureaucrat is a repeat-player. Assume the Constituent will report the
Bureaucrat when given the opportunity. The Bureaucrat can still implement Strategy 3 if the
following holds:

δ
2(F + χ)(p̂ + ν) − AB < (min(JB (y), 0) − 2(F + χ)(p̂ + ν) + χ) (28)
1−δ

The left-hand side represents the Bureaucrat’s payoffs from reporting the Constituent in the
first game. It is identical to equation 27 for the one shot reporting decision. The right hand
side represents all future bribery surplus (if any) the Bureaucrat lost by choosing to report.
Recall that Strategy 3 was previously governed by equation 24, reproduced below:

δ
(F + χ)p̂ − AB < (min(JB (y), 0) − (F + χ)p̂ + χ) (29)
1−δ

The results are similar to the one shot game. Compare equations 28 and 29. By doubling
punishments and increasing the probability that they will happen, the Bureaucrat is much more
willing to report even though it will damage their corrupt reputation. Their corrupt reputation
is also less valuable to them, in that continuing to accept bribes is much less attractive to the
if a discretionary public service is misclassified.
48
See appendix equation E.3, which denotes the Nash Bargaining Solution for equilibrium bribe size for under
the Refund regime with reporting. This is almost certainly at least half the value of the public service—probably
much higher, given how much risk the Bureaucrat is facing relative to the Constituent. If F , too, is at least half
the value of the public service, equation 27 is larger than equation 25, so the Bureaucrat’s incentive to report is
at least as strong as the Constituent’s.
49
I would add one caveat: it seems less likely that the Refund regime would lead to more collusive public
service provision. Even if the Bureaucrat doesn’t report, because the Refund regime is not implemented until,
say, a week has passed, law enforcement has one week to discover the bribe before the Constituent even has a
chance to report. If this happens, the potential increase in public service provision is mitigated.

28
Bureaucrat if the Constituent eventually reports them.
The Refund regime’s effects will be realized if both sides of equation 28 are negative, i.e.,
the Bureaucrat’s administrative costs are so high that they would rather skip the bribe entirely
than deal with administrative costs. This is unlikely for the same reasons I argued equation 27
would almost certainly be positive. If the left-hand side is positive and equation 28 does not
hold, the Bureaucrat will report and all of the Keep regime’s effects will be realized. Otherwise,
the Bureaucrat accepts the fact that the Constituent will report but decides it is worth the risk,
and bribery is not deterred.
Surprisingly, this policy seems to mitigate against the Keep regime’s biggest weakness.
Recall that the Keep regime alone produces the worst results when JB (y) > 0. If the Bureaucrat
reported, the Keep regime could have positive effects, but the Bureaucrat almost certainly will
not report for the same reason bribery existed in the first place: the Bureaucrat is a repeat
player. Moreover, giving the Constituent no recourse when the Bureaucrat shamelessly asks
for a bribe could increase bribery when JB (y) > 0. The Keep regime, ideally, should not be
implemented when JB (y) > 0. But suppose a harassment public service is misclassified as
collusive and JB (y) > 0. Adding the Refund regime could solve the problems associated with
the Keep regime. Even if the Bureaucrat is a repeat player, if the Constituent reports the bribe,
the Bureaucrat’s participation constraint may be violated, forcing the Bureaucrat to report or
more likely preventing the bribe from happening in the first place. Moreover, even when the
Bureaucrat does not report, bribery almost certainly does not increase. For while the Keep
regime effectively stripped the Constituent of their right to report any bribe demand made by
the Bureaucrat, this right was effectively replaced by giving the Constituent the right to report
the Bureaucrat a week after the bribe.50 Thus bribery could be deterred and harassment public
services could be provided even under the Keep regime.
Note that the Bureaucrat could have an incentive to levy a false accusation in this scenario.
If there is considerable risk that the Constituent will falsely accuse them once the Refund
regime is implemented, the Bureaucrat may incur the administrative costs of a false accusation
simply to hedge against this risk. However, this risk seems unlikely. If this is a problem, the
presence of false accusations is harmful to both parties: the Bureaucrat incurs administrative
costs by levying them and the Constituent faces legal risk once they are levied. Consequently,
the Constituent has an incentive to credibly pre-commit, if possible, to not levying a false
accusation. Perhaps they develop a reputation for not doing so: alternatively, the Bureaucrat
brings a recording device and turns it on in front of the Constituent. Both parties would be
happy to see it there. Moreover, many constituents will not levy false accusations—but the
Bureaucrat may not be able to differentiate those who will from those who will not. Potential
type I and type II errors weaken the attractiveness of false allegations by the Bureaucrat as a
hedge against the risk of a false allegation by the Constituent. Consequently, it seems unlikely
50
See Appendix C, although recourse given after the bribe might be a slightly weaker deterrent, because
there is a chance the Bureaucrat will get to keep the money. Moreover, evidence might be lost over the course
of a week.

29
that the Bureaucrat will levy false accusation.
What if the Refund regime is implemented, then the Keep regime? The effects are similar,
but much weaker. Generally, the Constituent’s incentive to report is at least as strong or
stronger than the Bureaucrat’s, especially in light of the fact that the Bureaucrat is much
more likely to be a repeat player. Consequently, adding the Keep regime will usually have
no effect: if the Constituent decided not to report, the Bureaucrat probably will not report
either. If the Bureaucrat would report, adding the Keep regime will generally have the same
effects described above: it will accentuate the Refund regime by increasing the Constituent’s
incentive to report.51 This could plausibly happen if the Bureaucrat has lower reporting costs:
perhaps the Bureaucrat’s experience in government may make it easier for them to navigate
the bureaucratic process.
There are more benefits to implementing both regimes. Even though implementing both
policies can blow-up corrupt relationships when only one side is a repeat player, bribery could
plausibly persist if both parties are repeat players. But we can enrich the model. If both
parties are repeat players, it seems plausible that they have corrupt relationships with multiple
parties. Assume the Bureaucrat takes bribes from many people, including the Constituent, and
the Constituent gives bribes to multiple officials, including the Bureaucrat. Moreover, suppose
that if a party is reported and convicted, they may be forced to exit the bribery market with
some positive probability; maybe a firm is no longer eligible for licenses, maybe an official loses
their job, etc.
Now suppose the Bureaucrat is aging, and decides to retire. Before doing so, they report
all of the parties from whom they have solicited bribes in the past, so they can ensure that
they will retain all of the money from their prior corrupt bargains. One of the people they
report is the Constituent, who is convicted and forced to exit the bribery market. Post-exile,
the Constituent reports the other officials they have bribed so they can keep the ill-gotten
gains from their prior transactions and receive a significant refund of bribe money. A cascade
emerges: once one person exits the bribery market and reports their past corruption, people
who have no connection to that person may end up being reported as other people start exiting
the bribery market.
We can do better. Perhaps the Constituent has not been convicted yet: they have been
accused by the Bureaucrat and may be forced to exit the bribery market. The accusations
against the Constituent send chills down the spines of all of the officials who the Constituent
has previously bribed. What if the Constituent is convicted? Will the Constituent turn on
them? Out of (perhaps rational) fear, a few report their bribes with the Constituent. Alas,
other citizens who have bribed these officials see that they have reported the Constituent.
Are they next? They certainly hope not, and they decide to report these officials. The cycle
continues, and the inevitability of it makes it difficult to sustain corrupt relationships with
multiple people and introduces even more frictions into the bribery market.
51
See Appendix F.

30
5. Mixed Bribes and the Dual Regime

In this section I will argue that the Dual regime is optimal for mixed bribes. I show that the
Dual regime is actually identical to the Refund regime if the parties privately believe the bribe
is harassment under conservative assumptions. As established previously, the Refund regime is
preferred for harassment bribes. I also show that, with subtle manipulations, the Dual regime
can be made to converge toward the Keep regime when the bribe is collusive—again, an ideal
result. However, when the Constituent believes that they are eligible for the Public Service, and
the Bureaucrat has doubts, it is feasible that both parties will report each other. When this
happens, the record of both parties reporting can be used as a weapon against bureaucratic
obstructionism.
In many scenarios, the Dual regime is identical to the Refund regime or the Keep regime.
It differs, however, when both parties report.52 Each party is, for instance, given three days to
report, and if both parties report within that interval, then the Dual regime takes effect.
Recall that, given symmetric reporting costs, the Constituent’s incentive to levy a true
accusation under the Refund regime is stronger than the Bureaucrat’s incentive to levy a true
accusation under the Keep regime. Suppose equation 25 holds, so the Constituent’s levies a
true accusation. If the Bureaucrat reports, the probability of detection is 1. What will the
Bureaucrat do?
Given aC (t) = 1
(2(F + χ)(p̂ + ν) − χ − AB > 0) =⇒ aB (t) = 1 (30)

This equation may hold for sufficiently large F , p̂, or ν . Suppose it does. Will it change the
Constituent’s decision to report?

Given aB (t) = 1

(2(F + JC (y)|1 − H|)(p̂ + φ) − JC (y)|1 − H| + χ − AC > 0) =⇒ aC (t) = 1 (31)

Note that for any harassment bribe, this is strictly larger than equation 25. Consequently,
the Constituent’s incentive to report becomes even stronger if the Bureaucrat reports a harass-
ment bribe.
This is not a stable equilibrium: if both parties will report, the Bureaucrat, at the minimum,
has paid an administrative cost to achieve the status quo. Consequently, if equation 30 holds,
there will be no bribe for a harassment public service: the Bureaucrat can achieve the same
outcome (or better) without paying an administrative cost simply by not accepting a bribe.
This leads to an interesting result: given symmetric reporting costs, for any harassment
52
Note that it does not matter who reports first, which is how it differs from the policies I advocate for in
4.3.

31
public service, the Dual regime becomes the Refund regime, which, as established previously, is
the preferred policy to combat harassment bribes! Intuitively, both parties reporting is not a
stable: at least one party is paying an administrative cost to achieve an outcome that could
have been achieved without incurring administrative costs.53 Since the Constituent’s incentive
to report is at least as strong as the Bureaucrat’s regardless of what the Bureaucrat decides to
do (when the Constituent does not need to forfeit the public service), the Bureaucrat will never
report. We return to the Refund regime. The Constituent will report if equation 25 holds.
Bribery can persist if equation 25 does not hold or if the Bureaucrat is willing to gamble that
the Constituent’s accusation will not be believed.
What if the bribe is collusive? The outcome is murkier. Equation 31 might not hold, and
even if it did, the Constituent is paying an administrative cost to achieve an outcome that they
could have had in the status quo.54 Unless administrative costs are high enough that nobody
will report, a mixed strategy equilibrium emerges. Sometimes, it will be optimal for one party
to report no matter what (i.e., equations 26 and 30 or 25 and 31 hold), in which case the
other party will either not participate or not report and hope that the accusation will not be
believed.55 Some bribery deterrence seems likely: both policies generally deter collusive bribery
when there is reporting, and some reporting will happen given sufficiently low administrative
costs.
What happens if both parties do not report within the specified time interval? It may have
been optimal for each party to report if they knew the other was not going to do so. If the
Refund regime and the Keep regime are implemented simultaneously, we could realize the sort
of arbitrary race to the phone described in section 4.3. I would again prefer to implement
the policies sequentially. In order to make the Dual regime resemble the Keep regime in the
collusive context, I believe the Refund regime should be implemented first, and then the Keep
regime.
This adds an extra layer of complexity to the Dual regime. To review, under the Dual
regime both parties have the right to report for a few days. Who reports first is irrelevant.
Then only the Constituent has the right to report for a few days. Then both are given the
right to report; whoever reports first receives immunity, etc. Under this regime, choosing not
to report in the first few days suddenly becomes costly to the Bureaucrat, for the same reasons
outlined in 4.3. If the Refund regime alone violates the Bureaucrat’s participation constraint,
the Bureaucrat will either report during the first interval or not accept a bribe at all. This
gives the Bureaucrat an incentive to report during the first interval at least as strong as it is in
4.3.56 The policy converges toward the Keep regime described in section 4.3. If the Constituent
53
Both parties, if H = 0.
54
The Constituent gets their bribe money back but loses the public service.
55
They will not report, because that means paying an administrative cost for an outcome that could have
been achieved without an administrative cost.
56
See equation 27, the Bureaucrat’s decision to report under the Keep regime followed by the Refund regime.
Equation 30 is irrelevant: the Bureaucrat will always report in the first interval if the Refund regime would
violate their participation constraint. Knowing this, the Constituent will not pay the bribe unless they do not

32
would report in the second period, then the Bureaucrat will report in the first period under
the conditions described in section 4.3. The Constituent, knowing the Bureaucrat would almost
certainly report under 4.3, will not report during the first period—they would be paying an
administrative cost to achieve the status quo. The Constituent, therefore, will only pay a bribe
if they feel the risk from the Bureaucrat reporting is acceptable because it is the only way they
might be able to get the collusive public service. The Keep regime, ideal for collusive bribes,
has returned.
Consequently, under the Dual regime, for harassment public services, we will realize the
Refund regime: given symmetric reporting costs, if there is any reporting, it is only done by
the Constituent, and the Constituent’s incentive to report is as strong as it is under the Refund
regime. Conversely, for collusive public services, given symmetric reporting costs, we realize the
Keep regime described under 4.3. If there is any reporting, it is likely done by the Bureaucrat,
although the Constituent may report if the Bureaucrat does not do so in the first period.
What if there is uncertainty as to whether the Constituent would be able to keep the public
service? This is only the case when y is unknown to one or both parties. y includes all factors
relevant for determining the Constituent’s eligibility for the public service. By assumption, the
Constituent’s eligibility is only unknown when y is unknown. I refer to a bribe in such a scenario
as a mixed bribe. To see what happens to mixed bribes under the Dual regime, it would help
to imagine what would happen if each party had different beliefs about the probability that
the bribe is harassment.
Suppose the Bureaucrat privately believes the bribe is probably harassment and the Con-
stituent does not. There will likely be no bribe at all. The Bureaucrat believes the Refund regime
is probably in place because the Dual regime becomes the Refund regime when the bribe is ha-
rassment. The Refund regime could violate the Bureaucrat’s participation constraint, which
would make the Bureaucrat hesitant to accept any bribe offer. Meanwhile, the Constituent
(falsely) believes the Bureaucrat will report them, and the Constituent cannot help themselves
by reporting, because if both parties report, the Constituent does not expect that they will be
able to keep the public service. Consequently, the Constituent is unlikely to make any bribe
offer if they are uncomfortable with the Keep regime.
However, if the Bureaucrat thinks the bribe is collusive and the Constituent thinks it is
harassment, a bribe may be exchanged followed by both parties reporting it. The Bureaucrat
believes the Constituent will not report and is just gambling that the Bureaucrat’s allegation
will not be believed, and vice versa. They are both wrong, and they both report, thereby
incurring and administrative cost for an outcome that could have been achieved for nothing.
Suppose this scenario occurs. The Constituent bribed the Bureaucrat for a driver’s license
prior to taking a test, and they both reported each other thinking the other would not. The
license is returned and the bribe is refunded. The test still needs to be administered. Now

plan on reporting the first interval, i.e., they’re willing to accept the risk of punishment for a chance at the
public service.

33
there is public record that the Constituent desires a license. To prevent the Bureaucrat from
retaliating against the Constituent by refusing to administer the test or creating other obstacles,
we force the Bureaucrat and the Constituent to pick a date to administer the exam. Any bribes
that happen before that exam are presumed to be collusive—they suggest the Constituent
believes they will fail the exam. If the Bureaucrat does not show-up to administer the exam,
they are facing legal punishment (barring extraordinary circumstances). After the exam is
administered, implement the Refund regime or the Keep regime, depending on the Constituent’s
performance.
One may object: how does an outsider know what happened during the exam? It may
often be tough to tell. But one factor that helps in these regimes is that at least one party will
have an incentive to preserve evidence of what happened during the test. If the Constituent
performed well on the test, they have an incentive to preserve evidence of their success in order
to ensure that they will receive a refund on any harassment bribe they pay. If the Constituent
fails the test, the Bureaucrat has an incentive to preserve evidence of their failure to ward
off false allegations of harassment bribery and perhaps use as leverage for extracting collusive
bribes. Under the status quo, neither party would have any incentive to preserve evidence of
what happened.

5.1. Experimental Work on the Dual regime

Note that Abbink (2017) studies a policy similar to the Dual regime in a laboratory setting.57
In his model, each party can receive a reward if they report and the other does not, although
each party is punished if both report, creating a true prisoner’s dilemma. Both parties always
have an opportunity to report; considering the fact that, in the real world, the probability that
both parties report each other at the exact same moment is zero, implicitly, each party has
some amount of time to report before a prosecution is initiated. This is why it resembles the
Dual regime, as opposed to what I describe in section 4.3.
There are some differences, however. Both parties receive a reward if they report and the
other party does not. This is different from my conception of the Keep regime; in my model, the
Keep regime allows the Bureaucrat to keep their bribe money but not receive additional money
on top of it. Moreover, in Abbink’s model, if both parties report, both parties are punished,
whereas in my model, they return to the state they were in before a bribe occurred (although
they have paid administrative costs). Additionally, Abbink’s model contained some favorable
assumptions: probability of detection goes from 0 to 1 if either side reports, there are no false
accusations, and no administrative costs. All of these assumptions are dubious in light of the
fact that each party would receive a reward if a false accusation was believed.58 The presence
of false accusations probably would make true accusations less credible, and the difficulty of
sorting false accusations from true accusations probably increases the administrative costs of
57
Klaus Abbink & Kevin Wu, Reward self-reporting to deter corruption: An experiment on mitigating collu-
sive bribery, 133 J. of Econ. Behav. & Org. 256-272 (2017).
58
Recall that in my conception of the Keep regime, the Bureaucrat has no incentive to levy a false accusation.

34
levying accusations.
Nevertheless, in a repeated game, the Dual regime was extremely effective at deterring
collusive bribery. I still prefer a conception of the Dual regime where, if both sides report, they
are not both punished, but instead return to the status quo, for a number of reasons.
Public record of the fact that the Constituent seeks a public service from the Bureaucrat
can act as a hedge against bureaucratic obstructionism in the harassment context. By forcing
the Bureaucrat to publicly identify what information is needed from the Constituent in order
to determine the Constituent’s eligibility for the public service, we may be able to prevent
the Bureaucrat from imposing other hurdles on the Constituent. In this sense, a scenario in
which both sides report could actually be better for society than no one reporting at all. But
by punishing each side when they both report, the Constituent will not receive this sort of
recognition, and the Bureaucrat’s ability to obstruct the Constituent (out of either malice or
laziness) remains unchecked.
Additionally, I believe social stigma is one of the biggest real-world obstacles to the success of
these policies. When explaining my proposal to friends and colleagues, I found some react with
mild disgust, even if they concede that it might ultimately achieve its desired end. Regardless
of whether or not the proposal “works,” some are uncomfortable with, e.g., letting an official
solicit and keep collusive bribes as long as they report the people who pay them. I believe these
people conceive of the official as a co-conspirator who nevertheless would win in this world by
betraying those work with him—even if those people are also in the wrong. I, however, view
the official as a heroic whistle-blower whose efforts can help purge distortions of the democratic
process. I hope that norm can be instilled in the general public, and that there thus will be no
stigma against whistleblowing. But by implementing a prisoner’s dilemma regime, where both
parties are punished if they both report, we treat each party as co-conspirators, and preserve
stigmas against reporting. In the context of the model, I believe punishing both parties when
they report each other creates reputational costs to whistleblowing, which may discourage
whistleblowing.
I also prefer to implement the Keep regime followed by the Refund regime, as opposed to the
Dual regime alone, in the collusive setting. Recall that the Refund regime leads to incentives to
levy false accusations, but the Keep regime does not. This is an advantage for the Keep regime
generally. However, the Constituent has a stronger incentive to levy true accusations under the
Refund regime than the Bureaucrat has under the Keep regime (assuming symmetric reporting
costs). This is an advantage for the Refund regime. The fact that neither policy has both
advantages is no coincidence: the Refund regime allows the Constituent to receive something
that they did not have at the time of the accusation, which creates an incentive to levy false
accusations but also strengthens the incentive to levy true accusations.
Yet implementing the Refund regime after the Keep regime leads to a remarkable result:
under conservative assumptions, we achieve the best of both worlds. The threat of the Con-
stituent reporting once the Refund regime is implemented strengthens the Bureaucrat’s incentive

35
to report to the point that their reporting incentive is almost certainly stronger than the Con-
stituent’s.59 And yet, as noted in 4.3, the Bureaucrat likely still does not have an incentive
to levy false accusations. Strengthening the Bureaucrat’s incentive to report to the point that
it is as strong as the Constituent’s without creating an incentive to levy false accusations is
incredible. Accusations are likely and credible, and yet punishments do not become draconian.
False accusations by the Constituent, however, may persist under the Dual regime, which would
undermine true accusations.

6. Implementation Considerations

This idealized model, while helpful, ignores important practical implementation considera-
tions. I will note some of them here.

6.1. The Fugitive Problem

Suppose the Constituent bribes the Bureaucrat for help escaping prison. Law enforcement
learns of the Constituent’s escape and is actively searching for the Constituent, but they are
unaware of the Bureaucrat’s involvement. If the Bureaucrat reports the bribe, the Bureaucrat’s
report is not useful for law enforcement. Law enforcement knows of the Constituent’s escape
and is actively searching for them, and the Bureaucrat knew that when they reported the
Constituent. Common sense dictates the Bureaucrat should not get immunity for reporting the
bribe: reporting a wanted fugitive does not help law enforcement.
We can take this example further. Suppose the Constituent escapes from prison. The Bu-
reaucrat, who worked at the prison, does not show up to work and flees the country shortly after
the Constituent’s escape. The Bureaucrat is wanted for questioning regarding the Constituent’s
escape. After a week has passed and the Refund regime is implemented, the Constituent comes
out of hiding and insists that Irene, widely thought to be innocent, not the Bureaucrat, is the
government employee who helped the Constituent escape. The Constituent demands immu-
nity and insists on keeping their freedom. The accusation is all too convenient, because the
Constituent would not receive immunity if they accused the Bureaucrat, since the Bureaucrat
is a wanted fugitive. Law enforcement finds Irene, although few take the Constituent’s allega-
tions seriously and Irene is subsequently acquitted. Nevertheless, under the Refund regime, the
Constituent retains their immunity unless the Bureaucrat is subsequently convicted.
These are absurd results, and common sense and political pressures may prevent them from
coming to pass. But why do they occur? Where do the Refund regime and the Keep regime
go wrong? The fugitive example is a specific instance of issues that arise when the probability
that the Public Service is confiscated diverges from the probability the bribe is detected.
These probabilities are assumed to be the same in my model. But when the Constituent is
already a wanted fugitive, the probability that the Public Service (the Constituent’s freedom) is
confiscated has nothing to do with the detection of the bribe. If these outcomes were tethered
59
Less certain if the Bureaucrat is a repeat player.

36
together, i.e. the Constituent would only keep the Public Service if the bribe was undetected,
a false report would be probably be counterproductive, since it would attract unwanted law
enforcement attention to the Public Service. But if there is no tether, the Constituent can create
one by levying an accusation, true or false: if the real bribe is not discovered, the Constituent
retains their freedom. Thus creating a tether may lower the probability that the Public Service
is confiscated. Moreover, when the tether is severed, the Bureaucrat’s accusations may not be
helpful and should not be rewarded.60
Moreover, for the Constituent to levy a false accusation, there must be some reason the
Constituent cannot or does not want to levy the accusation against the party that is guilty.
In the example above, the Constituent cannot accuse the Bureaucrat because the Bureaucrat
is a wanted fugitive: the tether has been severed both ways, and the Constituent’s accusation
would not be helpful. We could also imagine scenarios where the Constituent wants to maintain
a collusive relationship with the Bureaucrat. Perhaps the Bureaucrat is a family member.
Otherwise, levying a true accusation is more attractive than levying a false accusation, because
levying a true accusation allows the Constituent keep the public service with probability 1.
These circumstances are unusual, but the law must give the government enough flexibility
to deny the Bureaucrat and the Constituent immunity in ridiculous circumstances such as these.
However, giving the government flexibility can create hold-up problems, particularly when the
Refund regime is implemented in the presence of collusive bribes. Under the Refund regime,
there will be plenty of political pressures to find an excuse to not give the Constituent a collusive
Public Service. These pressures must be resisted when the Constituent reports in good faith.
Otherwise, the Refund regime will not bolster the Keep regime.61 Finding the balance between
laws that create enough flexibility to prevent abuses without creating hold-up problems is a
tough task for the policy maker, but it is a crucial one for these regimes.

6.2. Discretionary Public services

As Li notes, leaving ambiguity in the legal regime when officials have discretion can greatly
weaken the effects of these systems.62 I believe that if these policies are to be effective, these
ambiguities must be eliminated, even if some public services are misclassified by virtue of a
bright-line rule that is not always accurate. I also argue that discretionary public services
should generally be treated as collusive.
60
Another example: the Constituent pays a bribe from a country that will not extradite the Constituent to
the Bureaucrat’s country. Meanwhile, the Bureaucrat could provide the Public Service and report the bribe with
the knowledge that the Constituent will not be prosecuted and could even pay more bribes to the Bureaucrat
in the future. Even here, if the Public Service is not provided or it can be rescinded, the Bureaucrat should not
be prosecuted.
61
As noted in Dufwenberg & Spagnolo, perhaps, instead of receiving the public service, the Constituent can
receive something of equivalent value. This will often help, but there will still be hold-up problems, and for
some public services no amount of cash will make the Constituent as happy as, e.g., release from prison.
62
Xingxing Li, Guest Post: Bribery and the Limits of Game Theory—the Lessons from China, Financial
Times (May 1, 2012) [Link]
theory-the-lessons-from-china

37
 The strongest point in favor of classifying discretionary public services as collusive stems
from an offshoot of Oak’s concern with Basu’s policy.63 Using Oak’s terminology, suppose
the Constituent seeks a permit to build a power plant. The Constituent can endogenously
choose whether to build a compliant plant or a non-compliant plant. If the Constituent builds
a compliant plant and pays a bribe for the permit, the bribe is harassment. Otherwise, it is
collusive. Oak noted that if Basu’s policy is implemented for harassment but not collusive public
services (as Basu proposed), the policy could have spillover effects for collusive public services.
The harassment regime favors the Constituent, and since the Constituent endogenously chooses
what kind of plant to build, the Constituent may be tempted to build a compliant plant to avoid
paying bribes. Conversely, perhaps the Bureaucrat has the power to obstruct compliant plants,
encouraging the Constituent to build a non-compliant plant so that it is easier for bribes to be
exchanged. It is not clear which effect will dominate; it depends on how easily the Constituent
can appeal an improper denial of a compliant plant.
Note that the proposal above already mitigates this concern. Under my proposal, the
Constituent has a stronger incentive to resist the Bureaucrat’s push toward a non-compliant
plant because the Keep regime makes the collusive regime appear extremely unattractive to the
Constituent. In fact, if bribery would be deterred by the Keep regime, the Constituent has no
reason to build a non-compliant plant, because the Bureaucrat has no reason grant a permit
to a non-compliant plant unless they receive bribe money. Thus, the Constituent would only
build a non-compliant plant if they were willing to pay a bribe and accept the risks from the
Bureaucrat reporting under the Keep regime—which seems like an unlikely scenario.
But we can do better. Suppose there is a third type of plant, an “arguably compliant”
plant. The Bureaucrat is allowed to but not required to give a permit to this plant. If we
classify arguably compliant plants as collusive public services, we encourage the Constituent
not to settle for building a plant that is arguably compliant. Rather, the Constituent should
strive to build a plant of such high quality that the Bureaucrat cannot legally reject it. Moreover,
the Bureaucrat has no reason to push the Constituent to build socially harmful, non-compliant
plants; arguably compliant plants are fine with them, and perhaps impose fewer moral costs than
non-compliant plants. So when public service type is endogenously chosen by the Constituent,
treating discretionary public services as collusive will generally improve the quality of the public
services provided.
Moreover, the hold-up problems associated with the Refund regime continue to weaken it in
the discretionary context. Suppose the public service is a government contract, and a firm pays
a bribe to receive it, although they clearly did not deserve it. Technically, it may be within the
official’s power to grant them the contract as a discretionary matter even if a bribe had not been
paid. But the firm might be worried—rightfully so—that even if they are supposed to be able
to keep the contract after when they report (as the Refund regime requires), the government

63
Mandar Oak, Legalization of Bribe Giving when Bribe Type Is Endogenous, 17 J. of Pub. Econ. Theory
580-604 (2015).

38
may succumb to political pressures and search for an excuse to void the contract. Perhaps the
government will claim non-performance. For discretionary public services, the promises of the
Refund regime appear less credible to the party that supposedly benefits from it.
Finally, implementing the Keep regime then the Refund regime just seems to be a little bit
more effective. As explained in section 5.1, when we implement the Keep regime followed by
the Refund regime, we may be able to achieve the best of both worlds by giving the Bureaucrat
an incentive to report that is probably at least as strong as the Constituent’s without creating
an incentive for the Bureaucrat to levy a false accusation. If we implement the Refund regime
followed by the Keep regime, the Constituent may report but false allegations could persist,
which could undermine the effectiveness of the true accusations.
Nevertheless, I would add a caveat. My model suggests that harassment bribes will gener-
ally be smaller than collusive bribes. Providing collusive public services generally require the
Bureaucrat to incur more costs than providing a harassment public service. They have to incur
more moral costs, it may negatively impact their job security, and even effort costs are prob-
ably higher.64 Because collusive public services are generally more costly to the Bureaucrat,
the Constituent generally must pay the Bureaucrat a larger bribe in order to induce them to
provide the public service. Consequently, it may be reasonable to implement a policy where
any bribe for a discretionary public service larger than X is presumed to be collusive, whereas
any bribe smaller than X will be presumed to be harassment.
There is another benefit to this approach. The potential refund of fake bribe money is the
only reason the Constituent would ever levy a false accusation under equation 9. This policy
would effectively cap the size of any refund for a discretionary public service—at least until a
week has passed with no reporting. Consequently, the Constituent’s incentive to levy a false
accusation for discretionary public services becomes significantly weaker. The Constituent may
also be discouraged from levying true accusations, but the risk is smaller, for the potential
refund is not the sole reason the Constituent would levy a true accusation. Under equation
10, the Constituent might levy a true accusation because they are interested in avoiding the
potential punishment if they are caught committing a real crime.
We could enrich the model. The Constituent may fabricate a bribe accusation in exchange
for a potential $1000 refund: they likely would not do so for a $1 refund. Perhaps ν , the effect
of an accusation by the Constituent on the probability of punishment, should be a function of χ,
the size of the alleged bribe. Consequently, the Constituent’s allegations are more credible when
bribe size is small. This makes the Refund regime generally more attractive when bribe size is
small—which proves to be yet another benefit of using a threshold bribe size in determining
which policy to implement for discretionary services.65
This is a rough and imperfect heuristic. Many discretionary public services will be misclas-
sified. Still, the rule comes with little ambiguity. A citizen must ask two questions to determine
64
The Bureaucrat might need to put effort into hiding the fact that the public service was provided.
65
Since harassment bribes should generally be smaller, as argued above, this is yet another benefit of imple-
ment the Refund regime to combat harassment bribes.

39
what legal regime is in place:
1. Is the official required to provide the public service?
2. Is the official allowed to provide the public service, and, if so, is the bribe less than X ?
If the answer to either of those questions is yes, then the Refund regime is in place.

6.3. Evidentiary Issues

Suppose we implement the Keep regime, and after a week has passed, we implement the
Refund regime. The Bureaucrat makes an accusation during the first week. We could imagine
law enforcement sitting on the information, hoping that the Constituent reports and confirms
the allegation when they are given the right to report. But if law enforcement makes a habit of
doing this, the Constituent may not report out of fear of being set up. This defeats the purpose
of implementing the Refund regime. Consequently, as a rule, any evidence gathered while the
Constituent while was confessing should be legally inadmissible if the Constituent was led to
believe that the Bureaucrat did not report the bribe. A similar rule should be implemented if
the Bureaucrat makes a confession under similar circumstances.
We could take this further. Suppose the Bureaucrat does report in the first week, and law
enforcement executes a search warrant on the Constituent’s property. A recording device which
contains audio confirmation of the bribe is found. The Constituent admits its contents and
claims they saved it in case they had a chance to make an allegation against the Bureaucrat in
the future. As a rule, said recording device should not be admissible against the Constituent.
Using it as evidence against the Constituent dissuades them from preserving the evidence at
all, thus weakening the Bureaucrat’s incentive to report in the first place (while also making it
more difficult to differentiate false allegations from true allegations).
This seems to be a tough pill to swallow. What evidence can be used against the Con-
stituent? The relevant test is simple: does the evidence exist because the Constituent wanted
to preserve it for use against the Bureaucrat? If so, it should not be admitted. In these regimes,
no sane participant would save an audio recording of a bribe unless they planned on using it
against the counterparty. The only reason that evidence exists is because the Constituent hoped
to use it against the Bureaucrat. The Constituent should not be punished for preserving it. But
plenty of evidence might exist not because the Constituent tried to preserve it, but simply be-
cause destroying all evidence of a crime might be difficult. All such evidence should be subject
to ordinary evidentiary standards. There will be false positives and false negatives, and this
paper takes no position on what the burden of persuasion should be. However, parties should
not be punished for willfully preserving evidence for use against the other.
Some may still worry that the decision to investigate or prosecute may be influenced by
whether both sides eventually reported each other, even if one side’s decision to report is
legally inadmissible. A potential solution: the government could bifurcate judicial responsibil-
ities. Every police officer and prosecutor can document and pursue allegations made by bribe
recipients or bribe payers, but not both. For instance, if officer A is designated for bribe recip-
ients, then they can document and pursue accusations made by bribe recipients but not bribe

40
payers. They also cannot communicate case details with any officer or prosecutor designated
for bribe payers until charges are filed. The same restrictions apply to officers designated for
bribe payers. No restrictions are added to the officers’ freedom to pursue other crimes.

6.4. Ignorance of the Law

What if y is known, but either the Bureaucrat or the Constituent is ignorant as to the state
of the law? This can lead to sub-optimal outcomes. Suppose the Constituent pays a bribe that
they suspect to be harassment, but due to their own inexperience they are not entirely sure
whether they are entitled to receive the thing for which they paid a bribe.66 The Constituent’s
uncertainty may make them hesitant to report the bribe during the one-week interval. If the
bribe is actually collusive, they are not reporting under the Refund regime. Rather, they are
hurting themselves by confessing without the right to report under the Keep regime. But after
the week passes, the Bureaucrat may report, and if the Bureaucrat reports first we realize sub-
optimal outcomes. Moreover, even if the Constituent does report, evidence may have spoiled
during the wait.
The solution: “preporting.” One side may not be sure whether they have the right to report
the other until a week has passed. Under a “preporting” regime, they can report on day one,
and if they do have the right to report, the accusation becomes effective immediately. If they
do not have the right to report immediately, the accusation becomes effective on day eight if the
other side does not report. If the other side does report within the week, then the “preporting”
party can be prosecuted, but their “preport” cannot be used against them in a court of law.
If they are found guilty, they are free to cite the “preport” as a reason they deserve a reduced
punishment.

6.5. Other Considerations

What if the parties use a middle-man? In such a scenario, there are two bribes: the citizen
bribed the middle-man, and the middle-man bribed the official. The bribes are for the same
public service, so they are of the same nature. So if the public service is collusive, the bureaucrat
can turn in the middle-man, pocket the bribe money, and receive immunity, while the middle-
man can turn in the citizen, pocket the money,67 and receive immunity from that transaction.
But the middle-man does not receive immunity for any bribe they paid to the bureaucrat: just
the bribe they received from the citizen. Moreover, the bureaucrat is not required to turn in
the citizen, if they even know who it is. The middle-man will suffice.
This framework does little to solve the problem posed by Abbink (2014). Abbink’s lab study
notes that the deterrence effects from Basu’s policy are weakened significantly when the official
can retaliate against a citizen who reports them—although it still does not perform worse than
66
The Bureaucrat has little reason to be honest and probably wants the Constituent to believe the bribe is
collusive under my regimes. This can lead to sub-optimal outcomes—but this issue exists in the status quo as
well. For if the Bureaucrat can convince the Constituent that they are paying a bribe for something illegal, the
Bureaucrat may be able to extract a larger bribe.
67
They do not get to keep whatever they paid to the bureaucrat, so it might just be some fee.

41
Symmetric punishment.68 This could be weakened by trying to preserve the anonymity of the
whistle-blower in the harassment setting. Note that anonymous whistle-blowing seems less
helpful in the collusive context—it seems harder for the citizen to retaliate, plus releasing the
names of whistle-blowers may make it difficult for officials to develop a corrupt reputation.
Given that the rights of each party change after a certain amount of time has passed, the
reporting party might have an incentive to lie about when the bribe took place. For instance,
if we are in the Keep regime, the Constituent may have an incentive to report the bribe on
day 1 and claim that it actually took place 7 days ago, thereby subject the Bureaucrat to
immediate prosecution. But perhaps the Bureaucrat intended to report the bribe promptly,
and was arrested before they had the opportunity (and cannot prove they intended to report
the bribe).
Note that the only party who has an incentive to potentially lie about the timing of the
bribe is the party that must wait a week for their reporting right to vest. This makes the
solution to the problem easy; if the party reporting is not endowed with the right to report
(e.g., a Bureaucrat self-reporting a harassment bribe, claiming a week has passed), then law
enforcement should wait at least one week before initiating a prosecution. If the other party
does not report within that time frame, the prosecution proceeds. If they do report and claim
the bribe happened within the week, they’re almost certainly telling the truth; after all, why
would they wait so long to report the bribe? Still, law enforcement will have to make a judgment
call about the timing of the bribe.

7. Conclusion

This paper presents a comprehensive proposal to combat all forms of bribery using regimes
built on asymmetric punishment. In my model, the regimes I propose deter bribery with at
worst ambiguous effects on public service provision. They likely improve public service quality
when that can be endogenously chosen by the citizen,69 and they do not aid in solving hold-up
problems associated with collusive bribery.70 Moreover, implementing them simultaneously acts
as a powerful deterrent against repeat players, especially those who have corrupt relationships
with multiple parties. Finally, I note that while bureaucratic discretion is theoretically a gray
area in these cases, bribe size is a strong indicator of bribe type, and can be used to remove
ambiguity from the law that can hamper these regimes.
For as interesting as many of the results are, note, however, that these policies are not
a panacea. They all deter harassment public service provision when the bureaucrat is not
motivated to provide the public service without a bribe. Perhaps the reduction in bribery is
68
Klaus Abbink et al., Letting the Briber Go Free: An Experiment on Mitigating Harassment Bribes, 111
SSRN Electronic Journal 17-28 (2014).
69
Mandar Oak, Legalization of Bribe Giving when Bribe Type Is Endogenous, 17 J. of Pub. Econ. Theory
580-604 (2015).
70
Christoph Engel et al., Symmetric vs. Asymmetric Punishment Regimes for Bribery, 18 Am. L. & Econ.
Rev. 506-556 (2016).

42
worth it. Still, for as much power as these tools have to fight corruption, they do nothing to
solve the problem of the official who shirks their responsibilities not because they are greedy,
but because they are apathetic. For that, there is no substitute for stronger institutions that
punish those who ignore their responsibilities and reward those who take their job seriously.

43
Appendix A. Supplemental Equations

Appendix A.1. Acceptable Range of Bribes

zBL ≥ (UBL |P = 0)

=⇒ JB (y) − FBL p(1) + χ ≥ −FBL p(0)

=⇒ χ ≥ −JB (y) + FBL (p(1) − p(0)) (A.1)

zCL ≥ (UCL |P = 0)

=⇒ JC (y) − FCL p(1) − χ ≥ d − FCL p(0)

=⇒ χ ≤ JC (y) − d − FCL (p(1) − p(0)) (A.2)

There may be a temptation to combine equations A.1 and A.2 into a single inequality.
It must be avoided. For χ is not simply a lump sum transfer between agents; it actually
endogenously determines the Fine, FL in all but one of the regimes I analyze. Even under the
status quo, part of the punishment for receiving a bribe is forfeiture of the bribe itself. With
that in mind, even if the combined surplus of the agents is higher when a bribe is paid, there
may not be a way to execute a lump sum transfer that ensures each individual agent has higher
surplus.71

Appendix A.1.1. Symmetric Punishment

We essentially return to equations A.1 and A.2, the Bureaucrat and the Constituent’s par-
ticipation constraints before accusations were introduced, since no accusations are levied.72
Bribery will persist if the following inequalities hold:

χ ≥ −JB (y) + (F + χ)p(1, 0, 0)

=⇒ χ(1 − p(1, 0, 0)) ≥ −JB (y) + F p(1, 0, 0)

−JB (y) + F p(1, 0, 0)
=⇒ χ ≥ (A.3)
1 − p(1, 0, 0)
χ ≤ JC (y) − d − (F + JC (y)|1 − H|)p(1, 0, 0) (A.4)

Appendix A.1.2. The Refund Regime

If the Constituent only makes true accusations, the minimum and maximum acceptable
bribes are given by the following:
71
This leads to an interesting possibility where asymmetric punishment actually leads to increased bribery
if the reporting party can credibly commit to lying about the size of the bribe. The lie can help preserve the
lump-sum transfer. In this paper, I will assume this sort of coordination is impossible.
72
With the possible exception of aC (d) = 1, but this is included in d.

44
 The Bureaucrat’s participation constraint:

χ ≥ −JB (y) + 2(F + χ)p(1, 0, 1)

=⇒ χ(1 − 2p(1, 0, 1)) ≥ −JB (y) + 2F p(1, 0, 1)

−JB (y) + 2F p(1, 0, 1)
=⇒ χ ≥ (A.5)
1 − 2p(1, 0, 1)
The Constituent’s participation constraint:

χ ≤ JC (y) − d + χp(1, 0, 1) − AC

=⇒ χ(1 − p(1, 0, 1)) ≤ JC (y) − d − AC

JC (y) − d − AC
=⇒ χ ≤ (A.6)
1 − p(1, 0, 1)

If false accusations persist in equilibrium, these equations are the exact same, with one caveat:
all instances of p(1, 0, 1) must be replaced with p(1, 0, 1) − p(0, 0, 1), the latter term reflecting
the probability of a wrongful conviction given a false accusation by the Constituent. This is
because these constraints depend not on the absolute expected punishments/rewards, but on
the difference between expected punishments/rewards relative to the incorrupt outcome. If the
Constituent will make false accusations, the difference becomes smaller.
How do false accusations affect equation 12? The math need not be changed significantly:
simply re-imagine ν as the effect of a true accusation on detection probability minus the effect
of a false accusation on detection probability. If ν is still positive, the inequality still holds.
However, it is not necessarily positive.
Suppose p̂ = 0.3. Now suppose a false accusation has a 5% chance of leading to a conviction.
Unless a true accusation has at least a 35% probability of leading to a conviction, the Refund
regime performs worse than symmetric punishment. If, for instance, a true accusation only
increased detection probability by 4% (from 30% to 34%), then Symmetric punishment triumphs
over the the Refund regime. This is because accepting a bribe under Symmetric punishment
increases detection probability by 30% (from 0 to 0.3), while accepting a bribe under the Refund
regime only increases it by 29% (from 0.05 to 0.34).

Appendix A.1.3. The Keep Regime

If Bureaucrat reports, the minimum and maximum acceptable bribes are given by the fol-
lowing:

45
 The Bureaucrat’s participation constraint:

χ ≥ −JB (y) + AB (A.7)

The Constituent’s participation constraint:

χ ≤ JC (y) − 2(F + JC (y)|1 − H|)p(1, 1, 0) (A.8)

−JB (y) + AB ≤ JC (y) − 2(F + JC (y)|1 − H|)p(1, 1, 0)

Appendix A.2. Utility Functions with Accusations

Incorrupt:

UBL = JB (y)P − FBL (aB (f ))p(GB = 0, aB (f ), aC (f )) − AB aB (γ) (A.9)

UCL = JC (y)P − FCL (aC (f ))p(GC = 0, aB (f ), aC (f )) − AC aC (γ) + d (A.10)

Disagreement:

VBL = JB (y)P − FBL (aB (f ))pB (GB = 1, aB (f ), aC (d)) − AB aB (f ) (A.11)

VCL = JC (y)P − FCL (aC (f ))pC (GC = 1, aB (d), aC (f )) − AC aC (f ) (A.12)

Bribe:
zBL = JB (y) − FBL (aB (t))pB (GB = 1, aB (t), aC (t)) − AB aB (t) + χ (A.13)
zCL = JC (y) − FCL (aC (t))pC (GC = 1, aB (t), aC (t)) − χ − AC aC (t) (A.14)

Appendix A.3. First Order Conditions with Respect to Accusations

The following denote the utility change that the Bureaucrat and the Constituent receive if
they make the three possible accusations: false (f ), true in the disagreement game (d), or true
in the bribe game (t):

46
 ∂UB
= −FBL (1)pB (0, 1, aC (f )) + FBL (0)pB (0, 0, aC (f )) − AB (A.15)
∂aB (f )
∂UB
= −AB (A.16)
∂aB (d)
∂UC
= −FCL (1)pC (0, aB (f ), 1) + FCL (0)p(0, aB (f ), 0) − AC (A.17)
∂aC (f )
∂UC ∂d
= − AC = HGB (pB (1, aB (f ), 1) − pB (1, aB (f ), 0))µ|1 − P |JC − AC (A.18)
∂aC (d) ∂aC (d)
∂VB
= −FBL (1)p(1, 1, aC (d)) + FBL (0)p(1, 0, aC (d)) − AB (A.19)
∂aB (f )
∂VC
= −FCL (1)p(1, aB (d), 1) + FCL (0)p(1, aB (d), 0) − AC (A.20)
∂aC (f )
∂zB
= −FBL (1)p(1, 1, aC (t)) + FBL (0)p(1, 0, aC (t)) − AB (A.21)
∂aB (t)
∂zC
= −FCL (1)p(1, aB (t), 1) + FCL (0)p(1, aB (t), 0) − AC (A.22)
∂aC (t)

If an equation is positive, then one maximizes their utility by levying an accusation against
the other. For simplicity’s sake, assume p(0, 0, 0) = 0, i.e., if no bribe took place and no false
accusation was levied, the probability of conviction is 0. Note that A.16 is always negative.

Appendix A.3.1. Symmetric punishment

Under Symmetric punishment, because F and A are positive and accusations are presumed
to increase the probability of punishment, every equation with the possible exception of A.18
is negative. A.18 represents the Constituent’s payoff from reporting the Bureaucrat if they
demand a harassment bribe and the Constituent refuses. Under A.18, the Constituent may
report if there is a sufficiently high probability that the Bureaucrat will be replaced by someone
less corrupt.

Appendix A.3.2. The Refund regime

Under the Refund regime, the sign of FB is unchanged relative to Symmetric punishment
so the Bureaucrat still will not levy any accusations. Equations A.16 and A.20 are moot
because they involve the Constituent’s disagreement payoffs, which are impossible because the
Constituent can legally offer a bribe. A.18, which represents the Constituent accusing the
Bureaucrat of demanding a bribe, is unchanged and the sign remains unknown. That leaves
A.17 and A.22 as the only equations that could have become positive under the Refund regime.
These are given by equations 9 and 10 in the body of the paper.

47
Appendix A.3.3. The Keep regime
Under the Keep regime, equation A.18 and A.19 are moot because they involve the Bu-
reaucrat receiving their disagreement payoffs. This is impossible because the Bureaucrat can
legally ask for a bribe. The sign of FC is unchanged relative to Symmetric punishment, and
with A.18 gone, the Constituent will not levy any accusations. Because Constituent will not
levy a false accusation and FBL (1) = 0, the first two terms in equation A.15 are 0 and thus A.15
is still negative. A.16 is also negative. The only accusation which might be utility maximizing
is given by A.21, which would represent the Bureaucrat levying a true accusation. This is given
by equation 13 in the body of the paper.

Appendix B. Testing the Refund Regime against Collusive Bribes

If inequality 12 holds, the Refund regime does better than Symmetric punishment at deter-
ring bribery. Whether inequality 12 holds is ambiguous when the bribe is collusive (H = 0).
However, under conservative parameter assumptions, the equation holds. A small increase in
detection probability (ν ), small administrative costs (AC ), and a large value for the public good
JC make inequality 12 less likely to hold, and thus constitute conservative assumptions. For
p̂ = 0.4, ν = 0.05 and 20AC = JC (y), the Refund regime is more likely to deter bribes that Sym-
metric punishment for 2F ≥ JC (y). Alternatively, if p̂ = 0.1 and the other parameters are left
the same: bribery is eliminated if 2.5F ≥ JC (y). In both scenarios, the Refund regime triumphs
if the baseline fine is at least half the value of the public service, which is conservative. Thus,
even if accusations increase detection probability by just 5%, the Refund regime likely does
better than Symmetric punishment at deterring collusive bribery.

Appendix C. Is the Refund regime redundant?

Note that aC (d) and aC (t) cannot both be equal to one at the same time—if a bribe is
demanded, the Constituent cannot accuse the Bureaucrat both before and after a bribe has
been paid (or the second accusation has no effect). However, if the former is preferable to the
latter, the Refund regime is redundant. Intuitively, asymmetric punishment actually exists in
the status quo in some real sense. If an official demands a bribe, the Constituent can report
the demand and receive no punishment as long as they did not pay the bribe. They get to keep
any bribe money that they would have paid, and they might receive the Public Service from
the next official without a bribe (if H = 1). If this option is as attractive as reporting after the
bribe has been paid, the Refund regime has no effect.
Is the Refund regime redundant? The Constituent gets more utility from reporting after the
corrupt transaction if the following holds:

JC (p̂ + ν)µ < JC − χ(1 − p̂ − ν)

48
 Note that this is strictly less likely to hold for larger values of χ. Suppose χ is equal to the
maximum bribe the Constituent is willing to pay for the public service under equation A.6.

(JC − JC (p̂ + ν)µ − AC )(1 − p̂ − ν)

JC (p̂ + ν)µ < JC −
1 − p̂ − ν

This reduces to 0 < AC , which is true by assumption. Consequently, the Refund regime is
not redundant: paying a bribe and reporting it is more attractive that reporting before the
bribe has been paid. Intuitively, if the Constituent reports the bribe before exchanging the
money, they risk not receiving the Public Service but they get to keep any potential bribe
money. If they report after the bribe, they risk the bribe money but they get to keep the Public
Service. The latter is more attractive: otherwise, the Constituent paid a bribe that violates
their participation constraint.

Appendix D. Fighting Strategy 1

Appendix D.1. The Constituent’s Response

When JB (y) > 0,73 the Constituent could set aC (d) = 1 even if they are better off in a one-
shot bribery regime (i.e., equation 9 holds). Intuitively, if an official asks a citizen to pay a $5
bribe in exchange for a license, it may require less hassle to just pay the bribe. But the citizen
may report the demand just to discourage other officials from making this demand from them
in the future. The citizen will only do this, however, if future officials will provide the license
even without the bribe. Call this “Strategy 4.”
Suppose the Constituent can ensure that at least one of the participation constraints for
Strategy 1 (equations 16 and 17) will be violated if and only if Strategy 4 is implemented. the
Constituent will implement Strategy 4 if the following holds:

δ
JC (y) − JC (y)µν + AC − χ < (χ + FCL (aC (t))p(1, aB (t), aC (t)) + AC aC (t)) (D.1)
1−δ

The left hand side of this inequality represents the potential cost incurred today if the public
service is not provided, although the Constituent still saves by avoiding bribe money. It might
be negative, in which case, reporting is a good decision regardless of how it may change the
Bureaucrat’s behavior. The right hand side represents the long run savings in terms of bribe
money that did not need to be paid, punishments that may not be levied because no bribe was
exchanged, and administrative costs saved in the event the Constituent’s optimal decision post
bribe is to report.74
73
Implying that H = 1
74
We could imagine a mirror scenario: when JB (y) < 0 and H = 1, the Constituent could set aC (d) = 0

49
 Additionally, when JB (y) > 0, the Constituent could set aC (t) = 1 even when reporting is
suboptimal in a one-shot game (equation 10 does not hold). This is similar to Strategy 4, except
the bribe has already been paid and the license has already been provided. It still might not
be worth the effort, but by reporting the bribe, the Constituent could dissuade future officials
from asking for a bribe. Call this “Strategy 5.”
Strategy 5 is limited to the Refund regime because Strategy 4 dominates Strategy 5 under
Symmetric punishment. Under Symmetric punishment, if the Bureaucrat demands a bribe from
the Constituent, reporting the Bureaucrat before giving them any money is strictly better than
reporting them after giving them money, and the deterrent effect on the Bureaucrat is identical.
Strategy 5 dominates Strategy 4 under the Refund regime.75 Neither strategy is viable under
the Keep regime.76
Now suppose the Constituent can ensure that at least one of the participation constraints
for Strategy 1 (equations 16 and 17) will be violated if and only if Strategy 5 is implemented.
the Constituent will implement Strategy 5 if the following holds.

δ
−χ(p̂ + ν) + AC − F p̂ < (χ + F p̂) (D.2)
1−δ

Note that, in the right hand side of the equation, I assume aC (t) = 0. Intuitively, this
inequality is meant to test whether the utility loss from reporting is worth the benefit of not
being in the bribery regime in the future. But, if the optimal decision in the bribery regime is
to report regardless, the premise falls apart. There is no “utility loss” from reporting because
equation 10 holds.

Appendix D.2. Symmetric punishment

The Strategy 1 participation constraints of the Bureaucrat and the Constituent are gov-
erned by equations 16 and 17. Under the Symmetric punishment, they reduce to the follow-
ing:

even when the Constituent is worse off in the bribery regime compared to aC (d) = 1 (i.e., equation A.2 does not
hold). Perhaps reporting the official is “worth the hassle,” if it could get the official replaced. But if they do, they
will not be able to report in the future, since future officials will not ask for a bribe. However, this seems self
contradictory: replacing the official probably is not worth the effort unless there’s a reasonable probability the
new official will provide the Public Service without a bribe. µ, in all likelihood, is close to zero when JB (y) < 0.
75
See Appendix C
76
Under the Keep regime, Strategy 4 is impossible because the Bureaucrat is free to ask for bribes, so Strategy
5 theoretically becomes viable. But if the Constituent actually reported bribes under the Keep regime, the
Bureaucrat could neutralize the strategy by also reporting the bribe. Unless the Bureaucrat faces extraordinary
reporting costs relative to the Constituent, this response renders Strategy 5 moot.

50
 The Bureaucrat’s participation constraint:
δ
JB (y) + F (p̂ + νaC (d)) < (χ − (F + χ)p̂))
1−δ

F (1 − δ)(JB (y) + F (p̂ + νaC (d)))

+ <χ (D.3)
1 − p̂ (1 − p̂)δ
The Constituent’s participation constraint:

JC (y) − F p̂ − χ > JC µ(p̂ + νaC (d)) − AC aC (d)

=⇒ JC (y)(1 − µ(p̂ + νaC (d))) + AC aC (d) − F p̂ > χ (D.4)

Appendix D.3. The Refund regime

The Strategy 1 participation constraints of the Bureaucrat and the Constituent are governed
by equations 16 and 17. Under the Refund regime, they reduce to the following:

The Bureaucrat’s participation constraint:

δ
JB (y) + F p̂(1 + aC (d)) + 2F νaC (d) < (χ − 2(F + χ)(p̂ + ν))
1−δ

2F (1 − δ)(JB (y) + F p̂(1 + aC (d)) + 2F νaC (d))

=⇒ + <χ (D.5)
1 − 2(p̂ + ν) δ(1 − 2(p̂ + ν))
The Constituent’s participation constraint:

JC (y) − χ(1 − p̂ − ν) − AC < JC µ(p̂ + νaC (d)) − AC aC (d)

JC (y)(1 − µ(p̂ + νaC (d))) − (1 − aC (d))AC

=⇒ >χ (D.6)
1 − p̂ − ν

Note that bribe size still must be positive. Consequently p̂ + ν < 0.5 for the Bureaucrat’s
participation constraint (equation D.5) to hold.

Appendix D.4. The Refund Regime Compared to Symmetric Punishment

Recall that equation 20 determines whether bribery will persist under Symmetric punish-
ment due to Strategy 1. Equation 21 does the same for the Refund regime. To see how the
Refund regime compares to Symmetric punishment, we must compare equations 20 and 21. To
simplify the math, note that the Constituent receives the same reward for setting aC (d) = 1
in both settings.77 Suppose aC (d) = 1 under both regimes. How do the joint participation

77
For Strategy 4 purposes, the Bureaucrat receives a larger punishment under the Refund regime. Still, this
exercise is meant to demonstrate that even under conservative assumptions, the Refund regime is superior to
Symmetric punishment

51
constraints of Symmetric punishment and the Refund regime (inequalities 20 and 21) compare?
Inequality 21 can be rewritten as follows:

JC (y)(1 − µ(p̂ + ν)) 2F (1 − δ)(JB (y) + 2F (p̂ + ν))

> +
1 − p̂ − ν 1 − 2(p̂ + ν) δ(1 − 2(p̂ + ν))
(1 − δ)(JB (y) + 2F (p̂ + ν)) 1 − p̂ − ν
=⇒ JC (1 − µ(p̂ + ν)) > (2F + )( ) (D.7)
δ 1 − 2(p̂ + ν)

Inequality 20 can be rewritten as well:

(1 − δ)(JB (y) + F (p̂ + ν)) 1

=⇒ JC (1 − µ(p̂ + ν)) > (F (1 − p̂2 + p̂) + )( ) − AC (D.8)
δ 1 − p̂

Note that the right-hand side of equation D.7 is strictly larger than the right-hand side of
equation D.8: AC disappears, the right most term of D.7 is larger than 1−1 p̂ for any detection
probability that leads to a positive bribe size,78 and (1 − p̂2 + p̂) < 2. Therefore, when aC (d) = 1,
the Refund regime deters bribery and increases public service provision relative to Symmetric
punishment when JB (y) > 0, an unambiguously positive outcome for harassment bribes.
If we set aC (d) = 0, inequality 21 (joint participation constraint under the Refund regime)
reduces to the following:

JC (y)(1 − µp̂) − AC 2F (1 − δ)(JB (y) + F p̂)

> +
1 − p̂ − ν 1 − 2(p̂ + ν) δ(1 − 2(p̂ + ν))
(1 − δ)(JB (y) + F p̂) 1 − p̂ − ν
=⇒ JC (1 − µp̂) > (2F + )( ) + AC (D.9)
δ 1 − 2(p̂ + ν)

And inequality 20 (joint participation constraint under Symmetric punishment) reduces to

the following:

(1 − δ)(JB (y) + F p̂) 1

JC (1 − µp̂) > (F (1 − p̂2 + p̂) + )( ) (D.10)
δ 1 − p̂

Once again, the right hand side of equation D.9 is strictly larger than the right hand side of
equation D.10. Ergo, the Refund regime deters bribery and increases public service provision
for JB (y) > 0.

78
See Appendix D.3

52
Appendix D.5. The Keep Regime
The Strategy 1 participation constraints of the Bureaucrat and the Constituent are governed
by equations 16 and 17. Under the Keep regime, they reduce to the following:79

The Bureaucrat’s participation constraint:

δ
JB (y) < (χ − AB )
1−δ
1−δ
=⇒ JB (y) + AB < χ (D.11)
δ
The Constituent’s participation constraint:

JC (y) − 2F (p̂ + φ) > χ (D.12)

Appendix E. Nash Bargaining Solution

The Nash bargaining solution is the value for χ that maximizes the product of the difference
in utility values in the bribery game and the disagreement game. Under Symmetric punishment,
for harassment bribes, it is given by the following:80

max(JC (y) − d − F p̂ − χ)(JB (y) − (F + χ)p̂ + χ)

χ

= max(JC (y) − d − F p̂ − χ)(JB (y) − F p̂ + χ(1 − p̂)) (E.1)

χ

The left hand side represents the Constituent’s utility from the public service minus bribe
money, potential punishments, and d, which represents the possibility that the Constituent
receives the public service even in the disagreement game. The right hand side represents the
Bureaucrat’s utility in the bribery game: the Bureaucrat’s disagreement utility is 0 because
they do not provide the public service, so there is nothing to subtract.
Ignoring terms which do not have χ, since they will be eliminated when we take the first
order condition, this equation reduces to the following:

max(χ(1 − p̂)(JC (y) − d − F p̂) − χ2 (1 − p̂) + χ(F p̂ − JB (y)))

χ

Deriving with respect to χ, we get the following:

79
Recall that φ is the increase in detection probability from the Bureaucrat levying a true accusation.
80
See equations A.9-A.14

53
 (1 − p̂)(JC (y) − d − F p̂) − 2χ(1 − p̂) + F p̂ − JB (y) = 0

Solving for χ, we get the following:

JC − d − F p̂ F p̂ − JB (y)
χ= + (E.2)
2 2(1 − p̂)

Compare to equations A.3 and A.4, which represent minimum and maximum bribe size
under Symmetric punishment: equilibrium bribe size is the midpoint between the Bureaucrat’s
minimum acceptable bribe and the maximum the Constituent is willing to offer.
By symmetry, equilibrium bribe size under the Refund regime with reporting is equal to the
following:81

JC − d − AC 2F (p̂ + ν) − JB (y)
χ= + (E.3)
2(1 − p̂ − ν) 2(1 − 2(p̂ + ν))

Under the Keep regime with reporting:82

JC − 2F (p̂ + φ) AB − JB (y)
χ= + (E.4)
2 2

If any of these values violate equations D.3, D.5, or D.11 (which represent minimum bribe
size when JB (y) > 0 under Symmetric punishment, the Refund regime with reporting, and the
Keep regime with reporting respectively), bribery under Strategy 1 will not persist.
While it is possible for the Bureaucrat’s participation constraint to be violated for all of these
regimes given a sufficiently large JB (y) and δ , no regime seems to be preferred. For instance,
while it can be shown that equilibrium bribe size is lower under Symmetric punishment than the
Refund regime with reporting (E.2<E.3), it can also be shown that minimum bribe size is lower
under Symmetric punishment than the Refund regime with reporting (D.3<D.5). Consequently,
while it presents an interesting avenue for further research, it does not affect the rest of the
paper’s conclusions.

Appendix F. The Refund Regime, Followed by the Keep Regime

If the Refund regime is implemented first, then, after a week with no reporting, the Keep
regime is added, nothing new happens unless the Bureaucrat decides to report. Suppose they
81
See equations A.5 and A.6, which represent minimum and maximum bribe size under the Refund regime
with reporting
82
See equations A.7 and A.8

54
do. Would the Constituent regret choosing not to report?

(χ(p̂ + ν) + 2(F + JC (y)|1 − H|)(p̂ + φ) − AC > 0) =⇒ aC (t) = 1 (F.1)

Compared to equation 25, the Constituent has a much stronger reason to report, as it will
allow them to avoid a punishment that is at least twice as risky as before.
What if the Constituent is a repeat player implement Strategy 2? The Constituent will
abandon Strategy 2 and report unless the following holds:

δ
χ(p̂+ν)−AC +2(F +|1−H|JC (y))(p̂+φ) < (JC (y)−2(F +|1−H|JC (y))(p̂+φ)−χ) (F.2)
1−δ

This is far less likely to hold than equation 18: the Constituent’s decision to not report is
much riskier given the potential punishment, and protecting a corrupt reputation is far less
valuable considering the risks from the Bureaucrat’s reporting.

55