Vous êtes sur la page 1sur 58

DP RIETI Discussion Paper Series 17-E-010

Corruption, Market Quality and Entry Deterrence in


Emerging Economies

Krishnendu Ghosh DASTIDAR


Jawaharlal Nehru University

YANO Makoto
RIETI

The Research Institute of Economy, Trade and Industry


http://www.rieti.go.jp/en/
RIETI Discussion Paper Series 17-E-010
February 2017

Corruption, Market Quality and Entry Deterrence in Emerging Economies

Krishnendu Ghosh DASTIDAR *

Jawaharlal Nehru University

YANO Makoto

Research Institute of Economy, Trade and Industry/Kyoto University

Abstract

We extend the Yano (2009) concept of market quality to analyze the effects of corruption in a three-
stage game of entry deterrence. The incumbent has incomplete information on the entrant's costs but
can increase these costs by resorting to unfair means (e.g., bribing a politician who harms the entrant).
The effectiveness of the bribe depends on the fairness index prevailing in the economy. The entrant
observes these costs and decides whether or not to enter. We completely characterize the optimal bribe
and show that this depends on the index of fairness. We also show that zero bribes need not maximize
welfare and market quality. Our results seem to be compatible with anecdotal evidence from emerging
economies such as India.

Keywords: Market quality, Entry deterrence, Bribe, Cournot


JEL Classification: L11, L13, O16, O17

RIETI Discussion Papers Series aims at widely disseminating research results in the form of professional
papers, thereby stimulating lively discussion. The views expressed in the papers are solely those of the
author(s), and neither represent those of the organization to which the author(s) belong(s) nor the Research
Institute of Economy, Trade and Industry.

* Corresponding Author: Krishnendu Ghosh Dastidar, Jawaharlal Nehru University, New Delhi-110067, India.

Email: krishnendu.dastidar@gmail.com.
Makoto Yano, Research Institute of Economy Institute of Economic Research, Kyoto University, Yoshida-

Honmachi, Sakyo-ku, Kyoto, 606-8501, Japan. E-mail:yano@kier.kyoto-u.ac.jp, Tel: 81-75-753-7185. Fax: 81-75-
753-7170.
1 Introduction
In many emerging economies corruption and poor governance lead to restricted entry. Es-
tablished rms often use dubious means to deter entry of other rms who are perhaps more
e cient than the incumbent in the sense that they have lower marginal costs. A similar
phenomenon is observed in developed economies where lobbying leads to restricted entry.
The common perception is that rms in developing countries are more likely to pay bribes
to get their desired objective, whereas rms in developed countries are more prone to lobby
the government to change the rules. Note that lobbying is a legal and regulated activity in
many countries, whereas bribery is not. However, all such activities tend to lower market
quality.1
If an incumbent can successfully impose cost increases on its rivals, it can prot out of it.
Salop and Scheman (1983,1987) point out that imposing higher costs on a rival can tame or
kill it as eectively as predatory pricing, and possibly at a lower cost to the dominant rm.
This increase in rivalscost can be achieved through the abuse of government procedures,
including sham litigation and the misuse of licensing and regulatory authorities. There may
be many types of regulatory hurdles for new entrants in a market, including controls by
licensing authorities, health and building inspectors and planning boards and an established
rm often can bribe government o cials (especially in emerging economies) to impose large
costs on a potential entrant. In such economies big inuential domestic rms may use sham
proceedings under import relief laws, e.g. an unwarranted claim of dumping, to engage
in non-price predation against a foreign rival. It is interesting to note that even physical
violence (bombings and beatings) had been used by incumbent rms in the past to tame
rivals (see Martin, 2010).2
As noted before, incumbent rms in a developed economy use lobbying to create entry
barriers. A recent example is as follows. Founded in 2008 and based in San Francisco,
California, Airbnb is a web based marketplace for people to discover and book unique ac-
1
As Yano (2009) observes- In general, free entry is the most fundamental rule that should be observed
in a market. To the extent to which freedom of entry is restricted, the market becomes less eective.
2
In pages 253-255 of his book, Martin (2010) provides a summary of this theory of Raising Rivals Costs
and also gives specic real life examples of such cost-raising strategies chosen by the incumbent.

1
commodations around the world. Under pressure from the hotel industry and a populace
concerned with the surge of foreigners in their neighborhoods, Prime Minister Shinzo Abes
government in Japan released guidelines this year for home sharing - called minpaku in
Japanese - that makes most Airbnb rentals in the country illegal. Under the new guidelines
Airbnb hosts would only be allowed to rent to guests who stay for a week or longer, a mi-
nuscule slice of the market (for details see Nakamura, Y. and M. Takahashi in Bloomberg,
February 19, 2016). This is clearly a case where lobbying by the incumbent industry has
successfully created entry barriers for the new entrant, Airbnb.3
In the present paper we analyze the above scenario by using a three-stage game of entry
deterrence in a horizontally dierentiated duopoly. In the rst-stage the incumbent (rm
1) decides on the level of bribeto prevent a technologically superior potential competitor
from entering the market. It may be noted here that the term bribealso indicates lobbying
expenses that are used to inuence policy decisions in such a way so as to increase the cost
of any potential entrant. The eectiveness of the bribe depends on the fairness index
prevailing in the economy. Lower is the fairness index, higher will be the eectiveness of
such a bribe to increase the entrants costs and hence lower will be the incentive to enter. In
the second stage the entrant (rm 2) observes its own cost and then decides to enter or not
to enter. In case it chooses to enter, rm 2 incurs an additional entry cost. Firm 2 decides
to enter i it expects strictly positive prot in the third stage. If rm 2 enters, then in the
third stage the rms play an incomplete information Cournot game. If rm 2 does not enter
then rm 1 gets monopoly payo.
By characterizing the optimal bribe in equilibrium this study demonstrates that the equi-
librium outcome crucially depends on the fairnessindex and the dierentiationparameter
(whether goods are substitutes or complement). When products are substitutes, if the fair-
ness index is su ciently high, the incumbent has no incentive to pay bribe, which ensures
competition and possibly a higher market quality. If the fairness index is su ciently low,
3
Ruling Liberal Democratic Party lawmaker Masaaki Taira, involved in drafting the Japanese govern-
ments guideline acknowledges, the new rules may create severe restrictionsfor hosts of the 26,000 prop-
erties in the country, but says the guidelines are a reection of the competing interests of his constituency.
The hotel industry had very serious concerns, so we set the minimum number of nights at a level that lowers
the chances for competition.

2
then in contrast, the incumbent has an incentive to pay bribe and such bribery leads to
restricted entry and consequently, a possible lower market quality. When products are com-
plements, then we show that regardless of the level of fairness index, the incumbent has no
incentive to pay any bribe and free competition is ensured.
It is often the case that although the incumbent knows that it possibly has an inferior
technology (higher marginal cost) than the potential entrant, it does not know the extent
of its own cost disadvantage. That is, the incumbent cannot observe the entrants type
(cost). This uncertainty (or incompleteness of information) is highlighted in the present
study. It leads to three dierent ndings: First, if the fairness index is su ciently high,
the optimal bribe is zero regardless of whether there is incomplete information or complete
information. Second, if the fairness index is neither too high nor too low, equilibrium bribe
is positive with complete information but it is zero with incomplete information. In this
case, entry is deterred under complete information and not under incomplete information.
As such, incomplete information appears to be bribery-preventive. This result is somewhat
similar to Maskin (1999), who proves in a dierent context, that uncertainty reduces the
entry barrier. Third, if the fairness index is su ciently low, our model predicts completely
blockaded entry under both complete and incomplete information. However, in this case,
the amount of equilibrium bribe is larger with incomplete information than with complete
information. That is, incompleteness of information seems to be bribery-promotive rather
than being bribery-preventive. This result stands somewhat in contrast to Maskin (1999).
Some of our results seem to be compatible with anecdotal evidences from an emerging
economy like India. For example, when the fairness index is low enough then we get a positive
level of bribe in equilibrium and entry is deterred with positive probability. In emerging
economies like India, where the fairness index is low due to poor law and order enforcement,
bribery is very common. Our results also indicate that positive bribe in equilibrium (which
in turn implies restricted entry) is more likely if the market size is large. It may be noted that
the market size will be larger when income levels are higher. Anecdotal evidences suggest
that bribes are higher in places that are relatively more prosperous. For example, the real
estate business in the rich neighborhoods of major cities in India are marked by extreme
corruption and bribery. Moreover, in such prosperous pockets the real estate industry is

3
dominated by big rms. Only in less prosperous places one normally nds smaller local
rms. Our results provide a possible theoretical explanation behind such evidences (see the
report by Ernst and Young, 2012).

1.1 E ciency, fairness and market quality

This study is the rst attempt to deal with the endogenous determination of market quality
as dened by Yano (2009), who refers to market quality as a measure of e ciency in
allocation and fairness in pricing in a market.4 E ciency refers to Pareto e ciency.
Fairness may be stated as fairness in dealing or in the process in which the terms of trade
are formed. A price formed through fair dealing is a fair price.5 Fair dealing should be
measured against a set of rules and laws imposed so as to maintain the well functioning of a
market. According to Yano (2008a), one such rule may be the non-discriminatory treatment
of actual and potential trading partners or, in other words, to ensure free entry and exit in
the market. However, such fairness cannot be guaranteed when one party (say one seller)
has signicant and unilateral powers (for instance, the power to use violence) enabling it
to unilaterally set the terms and change the rules of the game. This is often observed in
emerging economies.
In this paper we follow the Yano (2009) approach and dene market quality to be a
weighted average of total welfare(consumer surplus plus producer surplus) and total fair-
ness. Total fairness is dened to be the product of fairness indexand bribe paid. Lower
is the fairness index, higher will be the eectiveness of any bribe paid by the incumbent to
increase the entrants costs and hence lower will be the incentive to enter. In our set-up the
maximum possible fairness index is zero and in this case bribe has no inuence on the total
4
There are a couple of papers that study the endogenous determination of market quality (although in
a much looser sense). Dei (2011) analyses the dynamic development of a high quality labor market where
unskilled and skilled workers are properly distinguished . The paper by Furukawa and Yano (2014) studies
market quality by focusing on fairness in handling intellectual properties.
5
Actions in a particular market are competitively fair if they are conducted in compliance with the set
of generally accepted rules. Moreover, a state of that market is competitively fair if it is formed through
competitively fair actions and if there are no prot opportunities left available for competitively fair actions
(Yano, 2009).

4
fairness. This will be true when the governance is perfect. Some of the Scandinavian coun-
tries (for example, Denmark) possibly has a very high fairness index. However, when the
governance is relatively poor (which is very likely in emerging economies like India, Pakistan,
Bangladesh etc.), the fairness index is likely to be negative.
In many emerging economies the law enforcement agencies are often weak and very
corrupt and this means that big rms can get away by harming others illegally. Such an
act increases the marginal cost of the potential entrant. A low fairness index means that
fairness in dealing is crucially hampered and the incumbent is able to unilaterally set the
terms of the game to its advantage. A low fairness index together with a high bribe means
that total fairness will be low and consequently the market quality will also be low.6
In a country like India, the above scenario is often observed in many industries (for
example, the real estate industry). Such an industry is dominated by a few large rms, who
are, in general, close to powerful politicians and function under their patronage.7 These
rms take recourse to illegal means to discourage entry. Since the owners of such rms are
well connected, they get away with their illegal actions. Even if the entrant (which may be
a relatively small local rm or a new foreign rm, with little contacts with local politicians)
were to lodge a complaint against the big incumbent rm, nothing is likely to happen. In
countries like India, Pakistan, Bangladesh etc. it is almost taken for granted that the rich
and powerful will not be touched even if they are on the wrong side of law. In fact, ordinary
persons and entities are often hounded if they take on powerful organizations or persons.
This fact is quite well known and there are several media reports on this as well.8 This
implies that a potential entrant will not waste its time and resources in pursuing a legal
case against the big incumbent as this will simply increase its costs further without any
6
As noted earlier, the incumbent rm can bribe the local politician to harm the potential entrant. This
can take the form of physical violence or some other forms of harassment. This bribe can also be interpreted
as cost of hiring goons to threaten and even physically harm the entrant and since the law and order enforcing
agencies do not function properly, this strategy is often very successful in countries like India.
7
Sometimes politicians themselves or their family members are among the largest shareholders of such
companies. The Competition Commission of India in its several annual reports have documented unfair
practises by large Indian rms. All such reports are available on the website <<http://www.cci.gov.in/>>).
8
The report by Ernst and Young (2012) provides details on bribery and corruption in the construction
sector.

5
possible benet. In short, in such economies, if the incumbent harms the entrant through
illegal means and increases the entrants operating costs, the entrant just accepts this as
fait acompli and then decides whether to enter or not. By analysing this aspect, this study
shows how the level of fairness in dealing and of e ciency in allocation, or in short, market
quality, are determined simultaneously.
We now proceed to provide the following: (i) a brief overview of our framework (ii) a
summary of our main results and (iii) a discussion of the literature related to our framework.

1.2 Overview and some denitions

We consider the following model in a dierentiated product market. There is one incumbent
and one potential entrant. The incumbent rm can bribe the local politician to harm the
potential entrant. We capture the eect of such a bribe in the following way. If the
incumbent rm pays bribe b 0 then this increases the marginal cost of the entrant by hb,
where h > 0.
We dene = h to be the index of fairness. Higher is h (i.e. lower is ), higher will
be eectiveness of the bribe (b) in increasing the entrants marginal cost. Total fairness is
dened to be b= hb. When b > 0, a cost is inicted on the entrant through unfair (and
illegal) means and the total fairness is negative. The maximum possible fairness is attained
when bribe is zero.
Market quality (Q) is dened to be a weighted average of welfare ( W = consumer surplus
plus producer surplus) and total fairness ( b). That is, we have

Q (b) = W (b) + (1 ) b where 2 (0; 1)

It is often the case that in the absence of bribe such potential entrants have lower marginal
costs than the incumbent. Absence of any bribe means that total fairness is at its maximum.
In other words, in our framework, when the market is completely fair, the entrant has lower
marginal cost than the incumbent. We capture this in the following way. Let the incumbent
rms marginal cost be c. In the absence of any bribe (i.e. when b = 0) the potential entrants
marginal cost is c , where is the ability level of the entrant. Higher is the entrants
ability, lower will be its marginal cost. The incumbent cannot observe the entrants type

6
( ). We assume that is distributed over [0; ] with distribution function F (:) and density
function f (:).
Some possible reason behind such asymmetric costs and incomplete information are as
follows. The potential entrant may be a foreign rm with a superior (low cost) technology.
This may also be true in an emerging economy where the potential entrant is a small local
rm. The incumbent rm is typically very large, has a large bureaucracy and it draws labour
from the formal sector, where wages are higher as compared to the informal sector. This
pushes up the per unit cost of the incumbent. On the other hand, the small local rm has
access to the informal labour market and consequently it can pay lower wages.9 Also, the
size of its bureaucracy is smaller and this means its per unit costs are lower as compared to
the incumbent.
Often, the incumbent rm does not know the true cost of operation of such an entrant.
That is, is private information to the entrant. If the potential entrant is a foreign rm with
a superior technology the incumbent may not be aware of the extent of the technological
superiority. When the potential entrant is a small local rm then also the incumbent may
face incomplete information. As noted before the small local rm has access to the informal
labour market. Such a labour market is completely unregulated and wages are often decided
by informal bargaining. Consequently, wages are known only to the small local rm and the
labourer. Moreover a small local rm is more likely to have a better idea of the cultural
aspects like caste/community equations. Consequently, it may be able to get labour (and
even other physical inputs like bricks, sand, cement) cheaply. Since the incumbent is an
established entity, it is required (by law) to purchase inputs and hire labour from the formal
sector where prices and wage rates are typically known. As such, its costs are generally
known to everybody.
Since the entrants marginal cost is lower, it means that if no bribe was paid by the
incumbent (i.e. b = 0), then entry would be certain. In this paper we try to capture this
aspect, derive some results and oer some policy prescriptions to improve market quality.
9
In a country like India an overwhelming fraction of the labour force (about 90%) is employed in the
informal sector. Often, small local rms have much better access to this labour force than the large established
rms.

7
1.3 Summary of our ndings

We solve the three-stage game by backward induction. We rst compute the Bayesian-Nash
equilibrium of the third stage game.
Then we characterize rm 2s decision in the equilibrium of the second stage. We show
that if the bribe (which has been chosen by rm 1 in the rst stage) is below a threshold (b)
then all types of rm 2 enter. That is, probability of entry is one. If the bribe level lies in
between b and b (where b > b) then some types of rm 2 enter. We show specically that 2
will enter i its ability ( ) is higher than a critical type . This depends on the level of
bribe and higher is the level of bribe, higher will be . In this case the probability of entry
is 1 F( ). If the bribe chosen in the rst period is greater than or equal to b, then no
type of rm 2 will enter.
Thereafter, we solve the rst stage game when rm 1 chooses the optimal level of bribe.
Note that 1 will choose a bribe to maximize its expected payo after taking into account the
possible equilibrium outcomes in the second and third stages. Since a bribe b increases 2s
per unit cost by hb, 2s prot will be decreasing in b. This means that for b high enough 2s
expected payo will be zero and 2 will not enter. However bribing is also costly for 1 since
b is like a sunk cost to rm 1. If h is high (fairness index, , is low), then the eectiveness
of bribe is high and this allows player 1 to deter entry with a relatively lower level of bribe.
We show that the optimal bribe depends crucially on the fairness index ( = h).
We show that when goods are substitutes (i.e. > 0) then, if h is below a threshold level
(i.e. is above a threshold level) then the optimal bribe is zero. If h is above a certain level
(i.e. is below a certain level) then the optimal bribe lies in the interval b; b . Under no
circumstances will the optimal bribe lie in the interval (0; b). This means, if there is a strictly
positive bribe in equilibrium, then it must be greater than or equal to b. When goods are
complements (i.e. < 0), we show that the optimal bribe is always zero, regardless of the
value of (or h).
We also compute the expected welfare (consumer surplus plus producer surplus). We
provide an example to show that while zero bribe always maximizes total fairness (the
maximum possible value of total fairness, b = hb is zero), it need not maximize total

8
welfare when goods are substitutes. This result is somewhat related to Mankiw and Whinston
(1986), who introduced the concept of business stealing. The business-stealing eect exists
when the equilibrium strategic response of existing rms to new entry results in their having
a lower volume of sales - that is, when a new entrant steals businessfrom incumbent rms.
Put dierently, a business-stealing eect is present if the equilibrium output per rm declines
as the number of rms grows. (Mankiw and Whinston, 1986). In their paper it is shown
that when entrants incur a xed set-up cost of entry and when there is business stealing
eect then free entry is not welfare maximizing. It may be noted that when > 0 (goods
are substitutes) we have the business stealing eect in our model. In our model b = 0
implies certain entry but this need not maximize total welfare when goods are substitutes.
Note that in our set-up market quality is a convex combination of total welfare and total
fairness. Since zero bribe (certain entry) need not maximize total welfare, this implies zero
bribe may not always maximize market quality when goods are substitutes. This seems to
suggest the following: There may be a possible trade-o between total welfare and total
fairness and it raises intriguing questions as to how bribe aects market quality when goods
are substitutes.
However, when goods are complements, we show that under some parametric restrictions,
zero bribe maximizes total welfare and market quality. Note that the incumbent rm always
choose zero bribe in the rst-stage when goods are complements. An interesting policy
prescription that emerges is as follows: In order to curb bribery and improve market quality,
the government should foster competition in goods that are complements to each other. In
the real estate sector an example of such complements would be where one rm provides
residential housing and the other rm provides shopping malls in the same locality.
In our model rm 2s costs are private information (as is known only to rm 2).
A natural question that arises is as follows. How will our results change if there is no
incompleteness of information ( is common knowledge)? With complete information, when
rm 1 takes a decision regarding b it knows with certainty whether entry will be deterred
or not (that is, probability of entry is either zero or one). However, when is private
information then the equilibrium probability of entry lies between zero and one. In this case
rm 1 is not exactly sure whether a bribe b will deter entry or not (unless b b). There is

9
an uncertainty regarding the impact of bribe on 2s entry decisions. This is compatible with
evidences from real life. When an incumbent takes some irrevocable decision to deter entry,
the incumbent is often not too sure whether entry will be successfully deterred or not. It
only knows that higher b is more likely to deter entry. We have tried to capture this aspect
in our model.

1.4 Related Literature

The idea of market qualityis borrowed from Yano (2009).10 In this paper we focus on a
particular aspect of market quality namely, how corruption aects entry decisions and overall
market quality.
Our approach is somewhat similar to Dixit (1980). In Dixit (1980) an irrevocable invest-
ment by the incumbent rm allows it to alter its own marginal cost curve and thereby the
post entry equilibrium. The role of such an irrevocable investment in entry deterrence is to
alter the initial conditions of the post-entry game to the advantage of the established rm.
In our paper the incumbent pays some bribe which is similar to an irrevocable investment
(since the bribe is like a sunk cost to the incumbent in the third stage). However, unlike
Dixit (1980), in our model, such a bribe alters the marginal cost of the potential entrant
and not the marginal cost of the incumbent. Moreover, unlike Dixit (1980) the post entry
game in our exercise is an incomplete information game. It may be mentioned here that
Dixit (1980) focusses entirely on the role of investment in impeding entry and his exercise
is not linked to market quality. Our objective is to analyze how bribe aects overall market
quality.
Salop and Scheman (1983, 1987) demonstrate that imposing higher costs on a rival
can tame or kill it as eectively as any predatory pricing, and often at lower cost to the
dominant rm. A rival that has higher cost will typically reduce output (in a quantity-
10
Market quality economics is a eld that was born directly out of Yanos research. Its basic idea is that,
just as there are high-quality and low-quality products, there are also high-quality and low-quality markets.
High-quality markets are enriching; low-quality markets are impoverishing. Yano (2006, 2008a and 2008b)
provide some major results on several aspects on market quality. The recent paper by Dastidar and Dei
(2014) provides a short introduction to the basic ideas on the concept of market quality.

10
setting market) or raise price (a price-setting market with product dierentiation). This will
permit a dominant rm to expand its own output or raise its own price. If a rivals costs are
made high enough, it may shut down entirely. In our model the incumbent chooses bribe to
inict harm upon the entrant.
The fact that zero bribe need not always be welfare maximizing in our framework is
related to the work of Mankiw and Whinston (1986). In their model entrants incur a xed
set-up cost of entry. They introduce a concept called business stealing. The presence of
the business-stealing eect drives a wedge between the marginal entrants evaluation of the
desirability of entry and the social planners. This means that free entry (which results in
zero prot for all rms) need not be welfare maximizing. Similarly, in our model there is
business stealingwhen goods are substitutes. The bribe, b = 0 implies certain entry in our
model,but this need not always be welfare increasing.
Maskin (1999) analyses a model where capacity installation by an incumbent rm serves
to deter others from entering the industry. The uncertainty about demand or costs forces
the incumbent to choose a higher capacity level than it would under certainty. This higher
level diminishes the attractiveness of deterrence under uncertainty. In our model, we show
that incompleteness of information may result in less or more bribes as compared to the case
of complete information depending on the value of the fairness index.
Shleifer and Vishny (1993) show that structure of government institutions and the politi-
cal process are important determinants of corruption. In their framework government o cials
have some discretion over some economic activities (for example, issuing licence/permits to
produce) and this enables them to collect bribes from private agents. In our model, weak
law enforcement machinery allows the incumbent to harm potential entrants through un-
scrupulous means and deter entry.
There is a huge literature on corruption but very few papers are directly related to the
approach taken in the present exercise. Some papers that are somewhat related to our paper
are discussed below. It may be noted that the models and the focus of these papers are very
dierent from ours.
Harstad and Svensson (2011) analyze a model where faced with a regulatory constraint,
rms can either comply, bribe the regulator to get around the rule, or lobby the government

11
to relax it. This model explains the common perception that bribery is relatively more
common in poor countries, whereas lobbying is relatively more common in rich ones.
Campos, Estrin and Proto (2010) argue that corruption matters not so much because
of the value of the bribe, but because of another less studied feature of corruption, namely
bribe unavoidability. The paper by Emran and Shilpi (2000) constructs simple asymmetric
information models to analyze the eects of bureaucratic corruption on entry conditions and
output. Sequeria and Djankov (2013) analyze how the structure of public bureaucracies
determines the way in which public o cials set bribes. Broadman and Recanatini (2000)
develop an analytical framework for examining the role basic market institutions play as
determinants of rent-seeking and illicit behavior in transition economies. The papers by
Athreya and Majumdar (2005) and Lambert-Mogiliansky, Majumdar and Radner (2008)
deal with some aspects of petty bureaucratic corruption and its e ciency implications. The
paper by Kunieda and Shibata (2014) considers an economy with credit market imperfections
(that typically characterize a low quality nancial market with weak enforcements rules) and
analyses how changes in the degree of credit constraints aect economic uctuations.

Plan of the paper We provide the model of our exercise in section 2. Thereafter, in
section 3 we derive the equilibrium of our three-stage game and provide the major results
on optimal bribe. Here we also give examples to illustrate our main results. Section 4 deals
with welfare and market quality. In section 5 we give our concluding remarks. Lastly, in the
appendix we provide the details of our equilibrium computations and the proofs of all the
results.

2 The Model
We consider the following scenario in an emerging economy market. There is one incumbent
(rm 1) and one potential entrant (rm 2). Firm 1s per unit cost is c. When rm 1 does
not resort to any unfair means (it pays no bribe) rm 2s per unit cost is c , where is
the e ciency level of rm 2. This is private information to rm 2. Firm 1 cannot observe
. We assume that is distributed over [0; ] (where < c) with distribution function F (:)

12
and density function f (:). This means if rm 1 does not resort to any unfair means then
rm 2 has a cost advantage.
However, the incumbent can increase the cost of operation of rm 2 by paying some bribe
(say b) to the politician, who in turn can harm the potential entrant. If rm 1 pays b then
this increases the marginal cost of rm 2 by hb, where h > 0. We noted before that = h
is the index of fairness. Lower is , higher will be eectiveness of bribe and consequently,
lower will be the market quality.
Let k 2 (where k > 0) be the entry cost of rm 2. We provide the cost function of the
incumbent (rm 1) and the entrant (rm 2) below. Let rm 1 choose b 0. This bribe is
like a sunk cost for rm 1. Then we have the following:

C1 (q1 ) = cq1 + b

C2 (q2 ; ) = (c + hb ) q2 + k 2

can be interpreted as e ciency of the entrant. This is private information to rm 2. We


consider the following three-stage game.

1. In the rst-stage the incumbent (rm 1) decides on a level of b (bribe).

2. In the second stage the entrant (rm 2) observes its own marginal cost and then
decides to enter or not to enter. It may be noted that observing its own marginal cost
is equivalent to observing and hb. Typically, in emerging economies when an entrant
takes an entry decision, it is based on its cost which it knows clearly. In case it chooses
to enter, 2 incurs an entry cost, k 2 . 2 decides to enter i it expects strictly positive
prot in the third stage. This is a standard assumption in the IO literature (see Dixit,
1980 and Tirole, 1988).

3. If 2 enters, then in the third stage the rms play an incomplete information Cournot
game in a dierentiated good market. If 2 does not enter then 1 produces monopoly
output. Note that in this stage b is sunk cost for 1 and k 2 is a sunk cost for 2.

13
For the third-stage competition we consider a representative consumers utility function
based on Dixit (1979). Scores of papers in the literature have used this. A small sample
of such papers is as follows: Singh and Vives (1984), Hackner ( 2000), Bester and Petrakis
(1993), Zanchettin (2006), Pal (2010) and Alipranti, Milliou and Petrakis (2014).
On the demand side of the market, the representative consumers utility function of two
dierentiated products, q1 and q2 , and a numeraire good, q0 ; is given by the following:

1 2
U = a (q1 + q2 ) q + q22 + 2 q1 q2 + q0 :
2 1
The parameter measures the degree of product dierentiation and 2 [ 1; 1]. When
< 0 the goods are complements and when > 0 the goods are substitutes. Note that
when is unity then the products are homogeneous (perfect substitutes) and when is zero
the products are independent. We will consider cases where 6= 0.
The utility function generates the following system of inverse demand functions:

p1 = a q1 q2

p2 = a q1 q2

2.1 Notations and Assumptions

Let
2
A=a c; B = 2 and D = 4

Note that since 2 [ 1; 1] we have B 2 [1; 3] and D 2 [3; 4]. We can interpret A to be
a proxy for market size.
Let Z
1
(x) = f ( )d
1 F (x) x

Note that (0) is the expected e ciency of the entrant and (x) is the expected e ciency
given that it is more than x. We assume the following.

Assumption 1 A > (0) and A 3k.

0 00
Assumption 2 (x) 2 (0; 1) and (x) 0 for all x 2 (0; ).

14
Remark The rst assumption holds true if the market size A is high enough relative
to entry cost and the expected e ciency of the entrant. It may be noted that the rst
assumption ensures that in the absence of any bribe being paid all types of rm 2 will enter.
That is, with b = 0 entry occurs with probability one (see Lemma 1). The second assumption
puts some restrictions on the distribution function of 2s types. This is required for some
of our major results on optimal bribe. It may be noted that the second assumption always
holds if is uniformly distributed over [0; ]. It also holds for many other distributions.

3 Equilibrium analysis
We will solve the game backwards. We will rst analyze the third-stage game. We now
proceed to provide the equilibrium derivations.

3.1 Third stage equilibrium

If rm 2 chooses to enter in the second stage, then in the third stage the rms play an
incomplete information Cournot game and earn duopoly prots. If rm 2 had chosen not to
enter in the second stage, then rm 1 chooses monopoly output.
Now suppose 2 has chosen to enter in the second stage. In the appendix we provide the
computation the Bayesian-Nash equilibrium of the third stage. In this stage, bribe level b
has been determined previously and known to both rms, whereas the entrants ability, ;
is known only to the entrant. For any given b, let the Bayesian-Nash equilibrium be q1 (b)
(quantity choice by rm 1, which does not know ) and equilibrium choice by rm 2 (with
type ) be q2 ( ; b).
Note that we construct an equilibrium such that in the second-stage 2 will enter i its
ability ( ) is higher than a critical type . That is, rm 2 enters i 2( ; ]. If 2 enters,
then in the third stage rms 1 and 2 play an incomplete information Cournot duopoly game.
The equilibrium outcome is provided below. The computation details are provided in the
appendix.

15
Third-stage Bayesian Nash equilibrium
8
< AB+ hb ( ) if 2 enters
D
q1 (b) =
: A if 2 does not enter
2
8
< 2AB+ D 4hb+ 2 ( )
if 2 ( ; ]
2D
q2 ( ; b) =
: 0 if 2 [0; ] (2 does not enter for such )

Equilibrium prots at a Bayesian Nash equilibrium Routine computations show


that the equilibrium duopoly prots are as follows.

1 (b) = [q1 (b)]2 b

2 ( ; b) = [q2 ( ; b)]2 k 2 where 2( ; ]

3.2 2nd stage equilibrium

Given the equilibrium outcome in the third stage, we next proceed to analyze the second
stage game. In the second stage rm 2 makes a move and chooses either to enter or not to
enter. We analyze the relationship between bribe paid in the rst-stage and rm 2s entry
decision in the second stage. The proofs are given in the appendix.
We rst analyze the behavior of rm 2 when its type is = 0 (i.e. the lowest possible
type). First suppose that b = 0. That is, rm 1 does not resort to any bribing. Then, we
show that even the lowest possible type ( = 0) will get strictly positive payo if it enters.

Lemma 1 2 (0; 0) > 0.

Remark Note that q2 ( ; b) is strictly increasing in . This implies that 2 ( ; b) is strictly


increasing in . Hence, 2 ( ; 0) > 2 (0; 0) for all > 0. This means that if b = 0 (rm 1
does nothing), all types (even the lowest ability type) will choose to enter. This fact is also
intuitively obvious.

We now analyze the case where b > 0. Let


2
2AB +
(0) 2Dk
b =
4h
2AB + 4 2Dk
b =
4h
16
2
Since 1 and > (0), we clearly get that b > b. We proceed to our major results in
the second-stage equilibrium.

Proposition 1 When b 2 (0; b) then the following is true. (i) All types enter and the
@ 2( ;b) @ 2( ;b)
probability of entry is one. (ii) @b
< 0 and @
> 0.

Proposition 2 When b 2 b; b then the following is true. (i) 2 (0; ) and when rm
2s type is 2 [0; ] it does not enter. When rm 2s type is 2( ; ] it enters. The
@ 2( ;b) @ 2( ;b)
probability of entry is 1 F( ). (ii) @b
< 0 and @
> 0.

Remark Note that rm 2 enters i it expects strictly positive prot in the post-entry
game. Suppose b = b. Then, if rm 2 when type = 0 (the lowest type) enters, its payo
would be 2 (0; b) = 0. This implies, when b = b the lowest type will not enter. Note that
2 ( ; b) > 2 (0; b) for all 2 (0; ). This means if b = b then all types > 0 will enter.
Now suppose b = b. Then, if rm 2 when type = (the highest type) enters, its payo
would be 2 ; b = 0. This means type will not enter if b = b. Since 2 ;b < 2 ;b
for all 2 [0; ), no type will enter if b = b. In short, propositions 1 and 2 and the discussion
above imply the following:

1. b 2 [0; b] then = 0. All types enter. Here the probability of entry is one.

2. If b 2 b; b then 2 (0; ). Here is such that 2 ( ; b) = [q2 ( ; b)]2 k 2 = 0. 2


enters i its type is 2( ; ]. Here The probability of entry is 1 F( ).

3. If b 2 b; 1 then = . No type will enter. Entry is completely blockaded and the


probability of entry is zero.

The next result extends the points stated above. It identies conditions under which
2 (0; ).

2
4 2A 2A 4
Proposition 3 2 (0; ) i D
hb 2 2+
2k + D
(0) ; 2+ 2k + D
.

17
3.2.1 Preliminary results

We now provide some preliminary results. These will help us in deriving the major results
on optimal bribe in the next section.

Lemma 2 q1 (b) is continuous for all b 2 0; b .

Lemma 3 8
>
<
h
if b 2 (0; b)
D
q10 (b) = h 1
d ( )
>
: d

2d ( ) if b 2 b; b :
D+ d

Lemma 4 If f (0) > 0 then


dq1 (b) dq1 (b)
lim > lim+ i >0
b !b db b !b db
and if f (0) = 0 then
dq1 (b) dq1 (b)
lim = lim+ :
b !b db b !b db
Lemma 5
A k dq1 (b) h
lim q1 (b) = and lim = :
b !b 2 b !b db D+4

Remark Lemma 2 shows that q1 (b) is continuous for all b 2 0; b . Lemma 4 demonstrates
that q1 (b) is not dierentiable at b unless f (0) = 0. However, q1 (b) is dierentiable at all
b 2 (0; b) [ b; b .
We now proceed to demonstrate that both q1 (b) and q2 ( ; b) are strictly positive for all
b 2 0; b . We need this for non-triviality of our results. Note that assumption 1 implies
AB (0)
that A > (0). Since B 2 [1; 2] and 2 [ 1; 1] this implies that q1 (0) = D
> 0.
d ( )
First take > 0 (goods are substitutes). Note that assumption 2 states d
2 (0; 1). This
implies from lemma 3 we get that q10 (b) > 0 for all b 2 (0; b) [ b; b i > 0. Also, lemma
2 shows that q1 (b) is continuous for all b 2 0; b . Since q1 (0) > 0 and q10 (b) > 0 when >0
then we must have q1 (b) > 0 for all b 2 0; b . Now take < 0 (goods are complements).
A k
Lemma 5 shows that limb !b q1 (b) = 2
> 0 (since < 0). Note that from lemma 3
we have if < 0 then q10 (b) < 0 for all b 2 (0; b) [ b; b . This means that if < 0 then
q1 (b) > limb !b q1 (b) for all b 2 0; b . This ensures that q1 (b) > 0 for all for all b 2 0; b .
Using a similar logic we can show that q2 ( ; b) > 0 for all b 2 0; b .

18
3.3 First stage equilibrium

We now analyze the rst-stage game and solve for the equilibrium level of bribe chosen by
rm 1. In the rst stage rm 1 chooses the optimal level of bribe b to maximize its expected
payo (anticipating the equilibrium outcomes in stage 2 and 3). Let the optimal level of
bribe be b .
Note that if 1 chooses b 2 [0; b] then all types enter ( = 0). In this case 1 gets duopoly
2
payo 1 (b) = (q1 (b)) b. When 1 chooses b 2 b; b then 2 (0; ). 2 does not enter if its
type is . The probability of this event is F ( ). In this event 1 will be monopolist and
A2
his payo will be 4
b. Note that 2 will enter if its type is > . The probability of this
event is (1 F( )). Here 1 will be a duopolist and his payo will be 1 (b) = (q1 (b))2 b.
When 1 chooses b 2 b; 1 no type of 2 will enter ( = ). In this event 1 will be monopolist
A2
and his payo will be 4
b.
Hence, 1s expected payo, denoted by E1 (b), is as follows:

8
>
> (b) if b 2 [0; b]
>
< h 1
i
A2
E1 (b) = F( (b)) b + (1 F( (b))) 1 (b) if b 2 b; b
>
>
4
>
: A2
4
b if b 2 b; 1
where 1 (b) = (q1 (b))2 b.

Since q1 (b) is continuous for all b 2 0; b and (b) is continuous in b, we must have
that E1 (b) is also continuous for all b 2 0; b . Note that if b 2 b; b

A2
E1 (b) = F ( ) (q1 (b))2 + (q1 (b))2 b
4

Using computations derived in the appendix we get if b 2 b; b


2 d ( )
3
d 4 2
A [1 F( )] q1 (b) 1 d
E10 (b) = f( (b)) (q1 (b))2 + 15
db 4 2

Note that limb !b (b) = 0. Using the previous lemmas we get,

dE1 (b) h
lim = [2A (0) 2k ] 1
b !b db 2D

19
2 3
2
dE1 (b) h 4A 4k 2 (0) 4f (0) k
lim+ = 4 5 1
db 2
4 [D + f (0) (0)]
b !b +f (0) ( (0))2 + 8Af (0) k

Therefore,

dE1 (b) dE1 (b)


lim+ lim
b !b dbb !b db
2 3
h f (0) 4Dk 2 + 4k 3
(0) + 8ADk + 2 3
( (0))2
= 4 5
4D [f (0) (0) 2 + D] 4A 2
(0) + D ( (0)) 2

Note that the above implies if f (0) > 0 then E1 (b) may not be dierentiable at b = b.
However, E1 (b) is dierentiable at all b 2 (0; b) [ b; b .

3.4 Optimal bribe

We now provide the main results regarding the optimal level of bribe, b , which is chosen by
the incumbent. We will rst consider the case when the goods are substitutes ( > 0).
Note that the bribe, b, is like a sunk cost for the incumbent. It chooses this sunk cost
in the rst stage. While higher b increases the sunk cost, it also increases 2s per unit cost
of production. This in turn increases 1s prot and decreases 2s prot in the third stage
(when they play a incomplete information quantity choice game). That is, higher b has two
opposing eects. In equilibrium, therefore, 1 must choose an optimal b where these two
eects balance each other out.
We noted earlier that if b = b and if rm 2s type is (highest possible type) and if
this type enters then it will get zero payo ( 2 ; b = 0). Since 2 enters only when it gets
strictly positive payo it means that if b = b then no type of 2 will enter. Firm 1 is then
A2 A2
a monopolist and produces output 4
and gets a prot 1 b = 4
b. Note that 1 will
A2
never choose a b > b as it will then get E1 (b) = 4
b < E1 b . Hence, in all equilibria, we
get b b. We will show that the optimal level of bribe, b , depends crucially on index of
fairness, = h.

20
3.4.1 Optimal bribe when goods are substitutes ( > 0)

D
Proposition 4 If h A[f ( )(2k+ )+1]
for all 2 [0; ] then b = 0.

8
Proposition 5 If h [A (0)]
then b b.
n o
12 8
Proposition 6 If h max 5f ( )k A
; [A (0)]
for all 2 [0; ] then b = b.

Remark Proposition 4 shows that when fairness index is relatively better ( high enough
, h low enough) then we do not observe bribe in equilibrium. This happens when the
governance is very good and the eectiveness of any bribe is very low. Note that given any
and h, the su cient condition in proposition 4 is more likely to hold if the market size, A,
is relatively small.
On the other hand, if the fairness index is low enough ( is low enough , h is high
enough) then we observe a positive level of bribe in equilibrium (propositions 5 and 6). In
emerging economies like India, where the fairness index is low due to poor law and order
enforcement, bribery is very common. Note that that given any and h, the su cient
conditions in propositions 5 and 6 are more likely to hold if the market size, A, is large. The
market size will be larger (smaller) when income levels are higher (smaller). This means,
given any and h, equilibrium bribe is likely to be positive in a poorly governed economy if
the market size, A, is high enough. Higher is the bribe, lower will be the probability of entry.
Bribing shoots up the production cost of potential entrants and they nd it unprotable
to enter. Anecdotal evidences from India suggest that bribes are higher in places that are
relatively more prosperous. For example, the real estate business in the rich neighborhoods
of major cities in India are marked by extreme corruption and bribery. In such prosperous
pockets the real estate industry is typically dominated by big rms. Only in less prosperous
places one tends to nd smaller local rms along with the bigger rms. Our results provide
a possible theoretical justication to such evidences.
D
Propositions 4 and 6 describe two extreme possibilities. If h A[f ( )(2k+ )+1]
we observe
n o
zero bribe and certain entry. If h max 5f ( 12)k A ; [A 8 (0)] we observe the highest possible
level of bribe, b, and entry is completely blockaded (zero probability of entry). This means
that a necessary condition for partial entry (where entry occurs with probability between zero

21
n o
D 12 8
and one) is h 2 A[f ( )(2k+ )+1]
; max 5f ( )k A
; [A (0)]
. We now provide an illustrative
example to explore this possibility.

Example Let be uniformly distributed over 0; 29 . then f ( ) = 2


9
and F ( ) = 2
9
.
Let A = 5; = 1; h = 1; k = 1. This means
9 1
( ) = + ; b = 1:562 5; b = 5:5
4 2
8 25
= b for b 2 b; b
7 14
In this example the incumbents expected payo, E1 (b), is the following
8
< 1 b2 7
b + 121
if b 2 [0; b]
9 18 144
E1 (b) =
: 16 3
b 184 2
b + 2153
b 577
if b 2 b; b
3087 3087 3087 1372

We plot below E1 (b).

1's expected payoff


1.0

0.9

0.8

0.7

0.6

0.5
0 1 2 3 4 5
bribe

E1 (b) (Player 1s expected payo)

A2
Routine computations lead to the following. E1 (0) = 0:840 28; 4
b = 0:75 and
E1 (b) = 0:503 91. Optimal bribe, b = 3:8834 and E1 (b ) = 1:085 5. Since b 2 b; b entry
is partially blockaded. The equilibrium probability of entry is 1 F( (b )) = 0:410 56.

The next proposition demonstrates that regardless of the fairness index, = h, it is


not possible to have 0 < b < b.

22
Proposition 7 b 2 f0g [ b; b .

Remark Proposition 7 is interesting. It states that if there is an equilibrium with strictly


positive bribe then the level of such a bribe will be at least b. Hence, in equilibrium we will
2AB+ 2 (0) 2kD
observe either a zero bribe or a bribe which is at least b. Since b = 4h
, it is
strictly increasing in the market size. Therefore, given a fairness index ( = h), bribes are
likely to be higher if the market size (A) is larger. This nding is consistent with anecdotal
evidences from India.
h i
36 9
Proposition 8 If h max A[5A+12 ]
; A[2A+3 ]
then b < b:

Remark The above result indicate that if h is below a threshold level then in equilibrium
rm 1 chooses b < b and hence always allows entry of some type of entrants. That is,
equilibrium probability of entry is always positive. For any given and h, this case is likely
to appear if the market size, A, is relatively small.

3.4.2 Incomplete information vs complete information

Suppose that 2s costs are know to the incumbent. That is ^ 2 (0; ) is observed by 1 and
rms play a complete information Cournot game in the third stage. Routine computation
yield the following.
8 8
< AB+ hb ^ if 2 enters < 2AB+4^ 4hb
if 2 enters
D 2D
q1 (b) = and q2 (^ ; b) =
: A if 2 does not enter : 0 if 2 does not enter
2
h
Let ^b = 2AB+4^ 2Dk
4h
.It can be easily shown that 2 always enters if b 2 0; ^b and does
h
not enter, if b 2 ^b; 1 . First note that ^b < b. Let optimal bribe for the case of complete
n o
information be b . Note it can be easily shown that b 2 0; ^b . That is, in equilibrium,
rm 2 is either clearly in (b = 0) or clearly out b = ^b . There cannot be any partial
entry deterrence (as in the case of incomplete information). Note that
2
AB ^ A2 AB + 2^ Dk
b = ^b () ^b () h h i:
D 4 A2 AB ^ 2
2 4 D

23
Comparing with propositions 4-6 we get the following. If the fairness index ( ) is
su ciently high (i.e. h is su ciently low), the optimal bribe is zero regardless of whether
there is incomplete information or complete information. Note that if is small enough
and positive it is possible to have h AB+2^
2
Dk i
2 h D
. This means b = 0
2 A4 ( ABD ^ ) A[f ( )(2k+ )+1]

(proposition 4) and b = ^b (see above). That is, if the fairness index ( = h) is neither
too high nor too low, equilibrium bribe is positive and entry is completely blockaded with
complete information whereas the equilibrium bribe is zero with incomplete information. In
this case incomplete information seems to be bribery-preventive. This result is somewhat
similar to Maskin (1999), who proves in a dierent context, that uncertainty reduces the entry
12 8 h AB+2^ Dk
barrier. Note that if h max 5f ( )k A
; [A
;
(0)] 2 2
2
i then b = b (proposition
( ABD ^ )
A
4

6) and b = ^b. That is, if the fairness index is su ciently low (h is su ciently high), our
model predicts completely blockaded entry under both complete and incomplete information.
However, in this case, the amount of equilibrium bribe is larger with incomplete information
than with complete information (since ^b < b). That is, incompleteness of information is
bribery-promotive rather than being bribery-preventive. This result stands somewhat in
contrast to Maskin (1999).
Till now we have provided results when goods are substitutes ( > 0). We now provide
the main result with complements ( < 0).

3.4.3 Optimal bribe when goods are complements ( < 0)

Proposition 9 If goods are complements (i.e. < 0) then b = 0.

Remark When goods are complements then regardless of the level of fairness index (or
market quality), we observe zero bribe in equilibrium. The intuition is as follows. Bribe
pushes up costs of entrants and discourages entry. However, with complements, it better
for the incumbent rm to allow entry as it pushes up its demand and hence its prot.
Consequently, the incumbent chooses to opt for zero bribe. This has interesting policy
prescription in the following sense. In emerging economies like India, where market quality
is poor, it is better for the government to foster competition in complementary goods. This
is likely to reduce bribery. Since competition in substitutes tends to eat away prots, the

24
incumbent nds it advantageous to resort to bribe and discourage entry. To counter this
tendency, the governments should choose to encourage competition in complementary goods.

4 Welfare and market quality


We now provide some basic results on welfare and market quality.

4.1 Welfare

Note that on the demand side of the market, the representative consumers utility function
of two dierentiated products, q1 and q2 , and a numeraire good, q0 is given by

1 2
U = a (q1 + q2 ) q + q22 + 2 q1 q2 + q0 :
2 1

The inverse demand functions are

pi = a qi qj ; i; j = 1; 2; i 6= j:

Here consumer surplus is

CS = U (q1 ; q2 ) p1 q1 p2 q2
1 2
= a (q1 + q2 ) q + q22 + 2 q1 q2 + q0 p 1 q1 p2 q2
2 1

The producer surplus is

PS = 1 + 2 = [p1 q1 cq1 b] + p2 q2 (c + h (b) ) q2 k2

Therefore, total welfare is as follows:

1 2
CS + P S = (a c) q1 + (a c hb + ) q2 q1 + q22 + 2 q1 q2 + q0 b k2
2
1 2
= Aq1 + (A hb + ) q2 q1 + q22 + 2 q1 q2 + q0 b k2
2

Note that in equilibrium 1s output is q1 (b) and 2s output (when type ) is q2 ( ; b). So
we have to look for expected welfare. Note that when rm 2s type is , rm 2 does

25
A
not enter (q2 = 0) and rm 1 is a monopolist q1 = q1m = 2
. Such types ( ) do not
enter and do not incur the entry cost k 2 . For this case,
1 2
(CS + P S) = Aq1 q + q0 b
2 1
3 2
= A + q0 b
8
The probability of the above case is F ( ). Therefore the expected (CS+PS), which is the
expected welfare is as follows:
3 2
W (b) = A + q0 b F( )
8
2 3
Z
Aq1 (b) + (A hb + ) q2 ( ; b)
+ 4 5 dF ( )
1
2
fq12 (b) + q22 ( ; b) + 2 q1 (b) q2 ( ; b)g + q0 b k 2

For the welfare analysis we now consider the case of substitutes.

4.1.1 Case of substitutes ( > 0)

For substitutes we can show that W 0 (b) < 0 at b = 0 (see the appendix for the computations).
This means b = 0 is a local maximizer for W (b). A natural question that arises is the
following: Is zero bribe (which implies certain entry) always welfare maximizing? Our answer
is surprisingly negative. We demonstrate this in terms of an example.

An illustrative example Note that when b = 0, all types of rm 2 enter (i.e. proba-
bility of entry is one). In this case = 0 and the expected welfare is as follows:
2 3
Z
Aq1 (0) + (A + ) q2 ( ; 0)
W (0) = 4 5 dF ( )
1
0
2
fq12 (0) + q22 ( ; 0) + 2 q1 (0) q2 ( ; 0)g + q0 k 2

Note that when b = b then no type of rm 2 enters. This means = and rm 1 is a


monopolist. In this case, we have
3
W b = A2 + q0 b (47)
8
Consider the following values of the parameters.

A = 3; k = 1; = 1; h = 1 and is uniformly distributed over [0; ] .

2 (0; 0:219)

26
All the assumptions of our model are satised here and we have

(0) = ; b= and b = .
2 8
Using routine computations we can show that
53 2 2 27
W (0) = + + 3 + q0 and W b = + q0
288 3 8
Note that from above we have
53 2 5 3
W b W (0) = +
288 3 8
It may be noted that
W b W (0) > 0 for all 2 (0; 0:219)

Our example shows that zero bribe is not the global maximizer of W (b) when 2 (0; 0:219)
(since W b W (0) > 0). In fact, it can be shown that when = 0:2, the welfare maximizing
bribe is b = 0:2. We plot below W (b) over the range 0; b = [0; 0:2] to illustrate this.11 Note
that when = 0:2 then
8
< 11 2
b 221
b + 22 613
if b 2 [0; b]
18 90 7200
W (b) =
: 300 3
b + 1300 2
b 92
b + 42 313
if b 2 b; b
343 343 343 13 720

welfare
3.16

3.14

3.12

3.10

3.08

3.06
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20
bribe

Welfare: W (b)
11
When b = b, rm 1 is a monopolist and chooses a monopoly output in equilibrium. Its net monopoly
A2 9
prot is 4 b= 4 > 0 when 2 (0; 0:219).

27
We state this result in terms of a proposition.

Proposition 10 When goods are substitutes ( > 0) zero bribe may not be welfare maximiz-
ing.

Remark As noted in the introduction, this result is somewhat related to Mankiw and
Whinston (1986). In their paper it is shown that when entrants incur a xed set-up cost of
entry and when there is business stealing eectthen free entry is not welfare maximizing.
We provide a discussion of this point below.

Business stealing in our model According to Mankiw and Whinston (1986) the
business-stealing eect exists when the equilibrium strategic response of existing rms to
new entry results in their having a lower volume of sales-that is, when a new entrant steals
business from incumbent rms. Put dierently, a business-stealing eect is present if the
equilibrium output per rm declines as the number of rms grows. Note that this does not
mean that total output decreases with entry. In fact, in our model, where there are two
rms, with complete information and substitutes ( > 0) the incumbentsoutput decreases
with entry while the total output goes up. With incomplete information the concept of
business stealing is as follows:
When b 2 b; b we have 2 (0; ) and probability of entry is 1 F( ). Note that
as b increases the probability of entry goes down. In our paper when b 2 b; b then the
incumbents equilibrium output, q1 (b), increases with b provided > 0 (substitutes). More
b means lower probability of entry and incumbents equilibrium output increases. In other
words, more entry (lower b) would imply that incumbents equilibrium output decreases.
This is business stealing eectin our model with incomplete information.
A
Note that without entry the incumbents output is 2
. If b = 0 then entry is certain.
With zero bribe the incumbents equilibrium output is AB D (0) and entrants output is
2AB+ D+ 2 (0) AB (0)
A
2D
. Now 2
> D
i > 0. That is, with b = 0 and certain entry the
incumbents output goes down. However, total output with entry (incumbents output plus
A
entrants output) is strictly greater than 2
. That is, total output increases with entry when
@
there is zero bribe. Also note that both for b 2 b; b we have @b
(q1 (b) + q2 ( ; b)) < 0. That

28
is, total output q1 (b) + q2 (b) decreases with b. That is, lower entry (more b) implies lower
total output. In other words, more entry (lower b) implies greater total output.

Mechanism behind increase in welfare with increase in bribe We have shown


that zero bribe (certain entry) need not be welfare maximizing. Take the case of b = 0
(incumbents optimal bribe is zero). In our example welfare is maximized at b = b (no-entry
at all). If an entrant causes incumbent rms to reduce output, entry may be more desirable
to the entrant than it is to society. In our model, lower b means more probability of entry
but lower output for the incumbent. Also, incumbents prot will be lower than its prot as
a monopolist. While more entry (lower b) means additional output of the entrant and some
additional prots as well, it may be insu cient to make up for the loss in output and prot
of the entrant. As a result, total welfare may go down with lower b (more entry).

4.1.2 Case of complements ( < 0)

Proposition 11 When goods are complements ( < 0), if A k then zero bribe maxi-
mizes total welfare.

Remark For complements 2 [ 1; 0). Now A k will hold true if the market size,
A, is large relative to entry cost. Since A 3k (assumption 1), A k will also hold if
1
3
.

4.2 Market quality

It may be noted that market quality (Q) is dened to be a weighted average of total welfare
(W ) and total fairness ( b). That is, we have

Q (b) = W (b) + (1 ) b, where 2 (0; 1) :

Note that
Q (b) = W (b) (1 ) hb

Conventional wisdom suggests that b = 0 should maximize Q (:). But in our example we
demonstrated that with substitutes ( > 0) it is possible to have W b > W (0). This implies

29
that if is close enough to unity then zero bribe need not maximize overall market quality.
While zero bribe maximizes total fairness (the maximum possible value of b = hb is
zero), it need not maximize total welfare. This seems to suggest that there may be a trade-
o between total welfare and total fairness. Note that b = 0 makes entry certain. As
noted before, presence of the business-stealing eect drives a wedge between the entrants
evaluation of the desirability of entry and the social planners. With substitutes there is
business stealing eect in our model. This means that entry of a new rm need not always
be welfare increasing. This in turn implies that if is high enough, entry of a new rm entry
need not increase market quality, Q (:) when goods are substitutes.
In proposition 9 we had shown that when goods are complements then the incumbent
optimally chooses zero bribe in equilibrium (regardless of the fairness index). In proposition
11 we demonstrated that if the market size is large enough zero bribe maximizes welfare
when goods are complements. Hence, zero bribe maximizes overall market quality also. This
reinforces our policy prescription that governments should foster competition in complements
to reduce bribery and increase market quality.

5 Conclusion
In this paper we analyzed the eects of corruption in an entry deterrence game. Low fairness
index (or market quality) allows the incumbent to raise the costs of the potential entrant
by resorting to dubious means. We completely characterized the optimal bribe level in
equilibrium and also showed zero bribe need not be welfare maximizing. Our results seem
to be compatible with anecdotal evidences from an emerging economy like India. A more
rigorous empirical study is required to check whether our theoretical results hold true or not.
Some simple questions that arise are as follows.

1. We have shown that while zero bribe maximizes total fairness, it need not maximize
total welfare and market quality. This raises intriguing questions on the relationship
between bribe, market quality and welfare. For example, in our model, h is inversely
related to fairness. Suppose now that the government can choose h as a policy variable.
Then, what is the optimal value of h? Our example suggests that if the objective is

30
to maximize total market quality (which is a convex combination of welfare and total
fairness) then zero h need not always be optimal. Clearly more research is needed on
this front.

2. In our exercise if the incumbent pays bribe, b, it incurs a sunk cost equal to b . What
happens to equilibrium outcomes if the the cost of paying the bribe (b) is some function
(b), where (:) is strictly increasing? For example, suppose that the incumbent
generates funds for the bribe amount by taking money away from the other protable
activities. The optimal way to do this is to take it out from the least protable
alternatives rst. The resulting opportunity cost of paying a bribe, b, may thus be
00 00
convex in b. That is, we may have > 0 (in our model = 0). The analysis of
equilibrium outcomes for such bribe costs will be an interesting exercise.

3. Will the results change if instead of Cournot competition we have Bertrand competi-
tion in the third stage? We know that in the industrial organization literature, there
are papers that show that equilibrium outcomes depend crucially on the nature of
competition (Cournot or Bertrand).12 It would be interesting to recast our exercise
with price competition in the third stage.

12
See Vives (1999) for a succinct summary of the classic results around this point. Alipranti et al (2014)
provides some recent results.

31
References
1. Alipranti, M., Milliou, C. and E. Petrakis (2014) Price vs. quantity competition in
a vertically related market Economics Letters vol. 124, pp. 1226. doi:10.1016/
j.econlet.2014.05.002 (accessed August 10, 2016).

2. Athreya, K.B. and M. Majumdar (2005) A note on closing the windows: some lessons
from queuing theory International Journal of Economic Theory vol. 1, pp. 73-81.
DOI: 10.1111/j.1742-7363.2005.00006.x (accessed August 10, 2016).

3. Bester, H. and E. Petrakis (1993) The incentives for cost reduction in a dierentiated
industry, International Journal of Industrial Organization vol. 11, pp. 519534.
doi:10.1016/0167-7187(93)90023-6 (accessed August 10, 2016).

4. Broadman, H.G. and F. Recanatini (2000) Seeds of corruption: do market institutions


matter?Policy Research Working Paper 2368 The World Bank, Europe and Central
Asia Region, Poverty Reduction and Economic Management Sector Unit, Washington
DC, U.S.A. http://dx.doi.org/10.1596/1813-9450-2368 (accessed August 10, 2016).

5. Campos, N.F., S. Estrin and E. Proto (2010) Corruption as a barrier to entry: theory
and evidence IZA Discussion Paper No. 5243 IZA, Bonn, Germany. http://ftp.iza.
org/dp5243.pdf (accessed August 10, 2016).

6. Dastidar, K.G. and F. Dei (2014) Introduction to the special issue of Pacic Economic
Review on market quality economicsPacic Economic Review, vol. 19, pp. 531-536.
DOI:10.1111/1468-0106.12081 (accessed August 10, 2016).

7. Dei,F.(2011)Quality of labor markets in a developing country, Review of Interna-


tional Economics,19(4),626-633. DOI:10.1111/j.1467-9396.2011.00970.x (accessed Au-
gust 10, 2016).

8. Dixit, A. (1979) A Model of duopoly suggesting a theory of entry barriers, The Bell
Journal of Economics vol. 10, pp. 20-32. DOI: 10.2307/3003317 (accessed August 10,
2016).

32
9. Dixit, A. (1980) The role of investment in entry deterrence, The Economic Journal
vol. 90, pp. 95-106. DOI: 10.2307/2231658 (accessed August 10, 2016).

10. Emran, M.S. and F.J. Shilpi (2000) Corruption and entry deterrence, The Bangladesh
Development Studies, vol. 26, pp. 69-8. Stable URL: http://www.jstor.org/ sta-
ble/40795602 (accessed August 10, 2016).

11. Ernst and Young (2012) Managing bribery and corruption risks in the construc-
tion and infrastructure industry EYG no. AU1402. http://www.ey.com/Publication
/vwLUAssets /Managing_bribery_and_corruption_risks_in_the_construction_and
_infrastructure_industry/$FILE/Assurance_ %20FIDS _sector_paper_Construction.
pdf (accessed August 10, 2016).

12. Furukawa, Y. and M. Yano (2014) Market quality and market infrastructure in the
South and technology diusion, International Journal of Economic Theory 10 (1),
139-146. DOI: 10.1111/ijet.12032 (accessed August 10, 2016).

13. Hackner, J. (2000) A Note on price and quantity competition in dierentiated oligopolies,
Journal of Economic Theory vol. 93, pp. 233239. doi:10.1006/jeth.2000.2654 (ac-
cessed August 10, 2016).

14. Harstad, B. and J. Svensson (2011) Bribes, Lobbying, and Development, Ameri-
can Political Science Review vol. 105, pp. 46-63. doi:10.1017/S0003055410000523
(accessed August 10, 2016).

15. Kunieda, T. and A. Shibata (2014) Credit market imperfections and macroeconomic
instability, Pacic Economic Review vol. 19, pp. 592-611. doi: 10.1111/1468-
0106.12085 (accessed August 10, 2016).

16. Lambert-Mogiliansky, A., M, Majumdar and R. Radner (2008) Petty corruption: a


game-theoretic approach International Journal of Economic Theory vol. 4, pp. 273-
297. DOI: 10.1111/j.1742-7363.2008.00078.x (accessed August 10, 2016).

17. Mankiw, N.G. and M.D. Whinston (1986) Free entry and social ine ciency, Rand

33
Journal of Economics vol. 17, pp. 48-58. http://scholar.harvard.edu/les/mankiw
/les/free_entry.pdf (accessed August 10, 2016).

18. Martin, S. (2010) Industrial Organization in Context, Oxford University Press, New
York, U.S.A.

19. Maskin, E.S. (1999) Uncertainty and entry deterrence, Economic Theory vol. 14,
pp. 429-437. DOI: 10.1007/s001990050302 (accessed August 10, 2016).

20. Nakamura, Y. and M. Takahashi (2016) Airbnb faces major threat in Japan, its fastest
growing marketBloomberg Business, February 19, 2016. http://www.bloomberg.com/

news/articles/2016-02-18/fastest-growing-airbnb-market-under-threat-as-japan-cracks-
down (accessed August 10, 2016).

21. Pal, R. (2010) Technology adoption in a dierentiated duopoly: Cournot vs Bertrand,


Research in Economics vol. 64, pp. 128-136. doi:10.1016/j.rie.2009.11.003 (accessed
August 10, 2016).

22. Salop, S. C. and D.T. Scheman (1987) Cost-raising strategies, Journal of Industrial
Economics vol. 36, pp. 19-34. DOI: 10.2307/2098594 (accessed August 10, 2016).

23. Salop, S. C. and D.T. Scheman (1983) Raising rivals costs American Economic
Review vol. 73, pp. 267-271. Stable URL: http://www.jstor.org/stable/1816853 (ac-
cessed August 10, 2016).

24. Sequeira, S. and S. Djankov (2013) Corruption and rm behavior, Working Paper
54321 The London School of Economics and Political Science, London, UK. http:
//eprints.lse.ac.uk/54321/ (accessed August 10, 2016).

25. Shleifer, A. and R.W. Vishny (1993) Corruption, The Quarterly Journal of Eco-
nomics vol. 108, pp. 599-617. doi: 10.2307/2118402 (accessed August 10, 2016).

26. Singh, N. and X. Vives (1984) Price and quantity competition in a dierentiated
duopoly, Rand Journal of Economics vol. 15, pp. 546554. Stable URL: http:
//www.jstor.org/stable/2555525 (accessed August 10, 2016).

34
27. Tirole, J. (1988) Industrial Organization Theory, MIT Press, Cambridge USA.

28. Vives, X. (1999) Oligopoly pricing: old ideas and new tools, MIT Press, Cambridge
USA.

29. Yano, M. (2006) A price competition game under free entry, Economic Theory vol.
29, pp. 395414. DOI: 10.1007/s00199-005-0028-8 (accessed August 10, 2016).

30. Yano, M. (2008a) Competitive fairness and the concept of a fair price under Delaware
Law on M&A, International Journal of Economic Theory Vol. 4, pp. 175190. DOI:
10.1111/j.1742-7363.2008.00082.x (accessed August 10, 2016).

31. Yano, M. (2008b) Economic analysis of market quality, in The Japanese Economy
A Market Quality Perspective, M. Yano, ed. Keio University Press: Tokyo, pp. 320.

32. Yano, M. (2009) The foundation of market quality economics, Japanese Economic
Review vol. 60, pp. 1-32. DOI: 10.1111/j.1468-5876.2008.00471.x (accessed August
10, 2016).

33. Zanchettin, P. (2006) Dierentiated duopoly with asymmetric costs, Journal of


Economics and Management Strategy vol. 15, pp. 9991015. DOI: 10.1111/j.1530-
9134.2006.00125.x (accessed August 10, 2016).

35
Appendix
In the appendix we provide the equilibrium computations and the proofs of all the results
mentioned in the paper.

6 Computation of third stage equilibrium


If rm 2 chooses to enter in the second stage, then in the third stage the rms play an
incomplete information Cournot game and earn duopoly prots. If rm 2 had chosen not to
enter in the second stage, then rm 1 chooses monopoly output.
Now suppose 2 has chosen to enter in the second stage. We will now compute the
Bayesian-Nash equilibrium of the third stage. In this stage, bribe level b has been determined
previously and known to both rms, whereas the entrants ability, ; is known only to the
entrant. For any given b, let the Bayesian-Nash equilibrium be q1 (b) (quantity choice by
rm 1, which does not know ) and equilibrium choice by rm 2 (with type ) be q2 ( ; b).
Note that in the second-stage 2 will enter i its ability ( ) is higher than a critical type
. That is, rm 2 enters i 2( ; ]. If 2 enters, then in the third stage rms 1 and 2
play an incomplete information Cournot duopoly game.
Hence, in the third-stage equilibrium, 1 knows that 2 enters i its type . That is,
in equilibrium, 1 knows that it is facing an opponent with type . Since 1 knows it is
facing an opponent with type , then in equilibrium it computes 2s expected output
R f( )
to be Exp.(q2 ( ; b) j )= q2 ( ; b) 1 F( )
d .
In a Bayesian-Nash equilibrium, where 2[ ; ] we have the following.

q2 ( ; b) = arg max q2 fA q1 (b) q2 hb + g k2 , (1a)


q2 0
Z
f( )
q1 (b) = arg max q1 A q1 q2 ( ; b) d b (1b)
q1 0 1 F( )

Using (1a) and (1b) we get that

@
q2 fA q1 (b) q2 hb + g k2 = 0 at q2 = q2 ( ; b) (2a)
@q2
Z
@ f( )
q1 A q1 q2 ( ; b) d b = 0 at q1 = q1 (b) (2b)
@q1 1 F( )

36
Let Z
f( )
q2Exp: ( )= q2 ( ; b) d :
1 F( )
Using (2a) and (2b) we have

1
q2 ( ; b) = [A q1 (b) hb + ] (3a)
2
1h i
q1 (b) = A q2Exp: ( ) (3b)
2

Using (3b) in (3a) we get

1h n o i
q2 ( ; b) = A A q2Exp: ( ) hb +
2 2
1h i
= AB + 2 q2Exp: ( ) 2hb + 2 (4)
4

From (4) above we get


Z
f( )
q2 ( ; b) d
1 F( )
Z
1h 2 Exp:
i f( )
= AB + q2 ( )
2hb + 2 d
4 1 F( )
1h i 1Z f( )
= AB + 2 q2Exp: ( ) 2hb + d
4 2 1 F( )
1h i
= AB + 2 q2Exp: ( ) 2hb + 2 ( ) (5)
4
R
Since as per our denition q2 ( ; b) 1 fF( ( ) ) d = q2Exp: ( ) the above implies that

[AB 2hb] + 2 ( ) [AB 2hb] + 2 ( )


q2Exp: ( )= 2
= (6)
4 D

Using (6) in (3b) we get

AB + hb ( )
q1 (b) = (7)
D

Using (7) in (3a) we get


2
2AB + D 4hb + ( )
q2 ( ; b) = (8)
2D

It may be noted that if rm 2 does not enter, then rm 1 is a monopolist in this stage and
A
it produces q1m = 2
.

37
Third-stage Bayesian Nash equilibrium
8
< AB+ hb ( ) if 2 enters
D
q1 (b) =
: A if 2 does not enter
2
8
< 2AB+ D 4hb+ 2 ( )
if 2 ( ; ]
2D
q2 ( ; b) =
: 0 if 2 [0; ] (2 does not enter for such )

Equilibrium prots at a Bayesian Nash equilibrium Routine computations show


that the equilibrium duopoly prots are as follows.

1 (b) = [q1 (b)]2 b

2 ( ; b) = [q2 ( ; b)]2 k 2 where 2( ; ]

7 Computation and proofs of 2nd stage equilibrium


outcomes
Given the equilibrium outcome in the third stage, we next proceed to analyze the second
stage game. In the second stage rm 2 makes a move and chooses either to enter or not to
enter. We analyze the relationship between bribe paid in the rst-stage and rm 2s entry
decision in the second stage.

Proof of Lemma 1 Suppose b = 0 and suppose that rm 2 when its type is = 0 decides
to enter. Then, following our earlier computations we get 2 (0; 0) = [q2 (0; 0)]2 k 2 > 0 i
q2 (0; 0) > k. Now

2AB + 2 (0)
q2 (0; 0) = >k
2D
() 2AB + 2 (0) > 2kD (9)

Note that D = B (2 + ) and 2 [ 1; 1]. Note = 0 implies that B = 2 and D = 4.


2
Then 2AB + (0) = 4A > 2kD = 8k (see assumption 1). Now suppose 6= 0. Then
2
> 0. Note that 2AB 2kD () A k (2 + ). From assumption 1 we have A 3k.
2
This means 2AB + (0) > 2kD.

38
We now analyze the case where b > 0 but is small enough. Since 2 (0; 0) > 0 and 2 (0; b)
is continuous in b, we have that for b small enough 2 (0; b) > 0. This means all types of rm
2 will enter when b is small enough. That is, = 0 for b small enough.
Note that for b small enough we have = 0 and hence

2 (0; b) = [q2 (0; b)]2 k2


2 2
2AB 4hb + (0)
= k 2 > 0: (10)
2D

(10) above clearly shows that 2 (0; b) is strictly decreasing in b and there exists b > 0 s.t.
2 (0; b) = 0. This implies 2 (0; b) > 0 for all b 2 [0; b). This in turn means that if 1 chooses
b 2 [0; b) all types of rm 2 will enter. Clearly when b b, we have = 0. Note that

2 (0; b) = 0 (11)

2AB 4hb+ 2 (0)


Now note that 2 (0; b) = 0 () q2 (0; b) = k. This implies 2D
= k.

Hence
2
2AB + (0) 2kD
b= (12)
4h

Proof of Proposition 1 Straight forward and follows from discussion above. Note that
when b 2 (0; b) all types enter. That is, = 0. Here
2 2
2AB + D 4hb + (0)
2 ( ; b) = k2
2D
@ 2 (:) 2h @ 2 (:) 1
Clearly @b
= D
< 0 and @
= 2
> 0.

We now provide some discussion that will lead to the proofs of propositions 2 and 3.
Since 2 (0; b) = 0 it means that if b > b and if rm 2, when its type is = 0 were to
enter, its payo would be 2 (0; b) < 0. This implies that when b > b some types will choose
not to enter. That is, > 0 for b > b.
Now note that Z R
f( ) f ( )d
( )= d =
1 F( ) 1 F( )

39
This means
R
d ( ) (1 F( )) f( )+f( ) f ( )d
= 2
d (1 F( ))
f( )
= [ ( ) ] (13)
1 F( )
Note that ( ) 0 (Since ( ) is the expected value given that it is more than
d ( )
). This means d
0.
Now take b > b. Note that for b close enough to b the critical type, will be lower than
the highest type . Note that 2 ( ; b) = [q2 ( ; b)]2 k 2 = 0. Now [q2 ( ; b)]2 k 2 = 0 i
q2 ( ; b) = k. This means
2
2AB + D 4hb + ( )
k=0
2D
From above (and using (13)) we can compute that
d 4h
= (14)
db D+ 2 f( ) [ ( ) ]
1 F( )

d
Note that db
> 0 as ( ) .
For b > b and we have

2 ( ; b) = [q2 ( ; b)]2 k2
2 2
2AB + D 4hb + ( )
= k2 (15)
2D
Note that
@q2 ( ; b) 1 ( )d 2d
= 4h +
@b 2D d db
1 f( ) d
= 4h + 2 [ ( ) ] (16)
2D 1 F( ) db
From (14) we get that

2 f( ) d
4h + [ ( ) ]
1 F( ) db
d
= D < 0. (17)
db
Using (16), (17) and (14) we get that
@q2 ( ; b)
<0 (18)
@b
40
@ 2( ;b)
From (15) we know that @b
= 2q2 ( ; b) @q2@b
( ;b)
. Hence from (18) we have when b > b,
2 ( ; b) is strictly decreasing in b. Also, in this case, 2 ( ; b) is strictly increasing in .
That is, 2 ( ; b) > 2 ( ; b) for b > b.
Now note that using Lhospitals rule we get

lim ( )= (19)
!

2
Since D = 4 ; by using (15) we get,
2 2
2AB 4hb + D +
2 ( ; b) = k2
2D
2
2AB 4hb + 4
= k2 (20)
2D

Note that 2 ( ; b) is strictly decreasing in b and there exists b s.t. 2 ; b = 0. This means,
for b = b, = .
Hence
2
2AB 4hb + 4
2 ;b = k2 = 0
2D
2AB 4hb + 4
i =k
2D
From above we get that

2AB + 4 2kD
b= (21) :
4h
2
Since 1 and > (0), comparing (21) with (12) we clearly get that

b>b (22)

Proof of Proposition 2 Straightforward and follows from discussion above ( see equations
13-21).

Proof of Proposition 3 Note that from the discussion in the main body of our paper it
is clear that.
0< < () b < b < b

41
From (12) and (21) ) we get
2
2AB +
(0) 2kD
b =
4h
2AB + 4 2Dk
b =
4h

Therefore,
2
2AB + (0) 2kD 2AB + 4 2kD
b < b < b () <b<
4h 4h
2
2AB + (0) 2kD 4hb 2AB + 4 2kD
() < <
D D D
2
Note that since B = 2 and D = 4 = B (2 + ) the above is equivalent to the
following:

2
2A 4 2A 4
b < b < b () 2k + (0) < hb < 2k +
2+ D D 2+ D
This completes the proof.

8 Notes on preliminary results and their proofs


Note that 1 (b) = [q1 (b)]2 b. Hence,

d 1(b) dq1 (b)


= 2q1 (b) 1 (23)
db db

When b b then = 0. This means for b 2 (0; b)

AB + hb (0)
q1 (b) = (24)
D
dq1 (b) h
= (24a)
db D

This implies when b 2 (0; b)

d 1(b) 2 [AB + hb (0)] h


= 1 (25)
db D2

From (12) and (24) we get that

2A (0) 2k
q1 (b) = (26)
4

42
AB+ hb (0)
Proof of lemma 2 Note that when b 2 (0; b) we have q1 (b) = D
. Clearly
AB+ hb ( )
q1 (b) is continuous for all b 2 (0; b). When b 2 b; b we have q1 (b) = D
.
Since is continuous in b, clearly q1 (b) is continuous for all b 2 b; b . Note that as
b ! b we have ! 0. Hence, limb!b (b) = 0. Routine computation shows that
limb!b q1 (b) = limb!b+ q1 (b) = q1 (b). Hence, q1 (b) is continuous for all b 2 0; b .

Note that when b 2 b; b we have


AB + hb ( )
q1 (b) = (27)
D
where 2 (0; ) : Hence, for b 2 b; b we have
dq1 (b) d ( )d
= h
db D d db
" #
f( ) 4h
= h [ ( ) ]
D 1 F( ) D+ 2 f( ) [ ( ) ]
1 F( )
h 2 f( ) 4f ( )
= D+ [ ( ) ] [ ( ) ]
D D+ 2 f( ) [ ( ) ] 1 F( ) 1 F( )
1 F( )

h f( )
= D 1 [ ( ) ]
D D+ 2 f( ) [ ( ) ] 1 F( )
1 F( )
f( )
h 1 1 F( )
[ ( ) ]
= (28)
D+ 2 f( ) [ ( ) ]
1 F( )

d ( ) f( )
Since d
= 1 F( )
[ ( ) ] using above we get that when b 2 b; b
d ( )
dq1 (b) h 1 d
= (28a)
db D+ 2d ( )
d

Note that when b = b we have = 0. Hence, using (28) we get


dq1 (b) h (1 f (0) (0))
lim+ = 2f
(29)
b !b db (D + (0) (0))

AB+ hb (0)
Proof of lemma 3 Note that when b 2 (0; b) we have q1 (b) = D
. Hence,
h
when b 2 (0; b) we have q1 (b) = D
. From (28a) we get that when b 2 b; b we have
d ( )
h 1 d
dq1 (b)
db
= 2d ( ) .
D+ d

43
dq1 (b) h
Proof of lemma 4 Note that from the previous lemma we have limb !b db
= D
. From
dq1 (b) h (1 f (0) (0))
(29) we get that limb !b+ db
= (D+ 2 f (0) (0))
. Hence,

dq1 (b) dq1 (b)


lim lim+
b !b db b !b db
h D (1 f (0) (0))
= 1
D (D + 2 f (0) (0))

Note that since (0) > 0 we get that if f (0) > 0

D (1 f (0) (0)) D (1 f (0) (0))


2f
< 1 =) 1 2f
> 0.
(D + (0) (0)) (D + (0) (0))

This means when f (0) > 0

dq1 (b) dq1 (b)


lim > lim+ i > 0.
b !b db b !b db
dq1 (b) dq1 (b)
Also, when f (0) = 0 then limb !b db
limb !b+ db
= 0.

Now let

d ( ) f( )
J = lim = lim [ ( ) ] (30)
! d ! 1 F( )

Note that
( )
J = f ( ) lim (31)
! 1 F( )
Using (19) and the fact that is distributed over [0; ] with distribution function F (:) we
know that

lim [ ( ) ] = 0
!

and lim [1 F( )] = 0 (32)


!

44
Hence, by using Lhospitals rule we get
d[ ( ) ]
d
J = f ( ) lim d[1 F ( )]
!
d
d ( )
d
1
= f ( ) lim
! f( )
f( )
1 F( )
[ ( ) ] 1
= f ( ) lim (using (13))
! f( )
f( )
lim ! 1 F( )
[ ( ) ] 1
= f( )
f( )
f( )
= 1 lim [ ( ) ]
! 1 F( )
= 1 J (see (30)) (32a)

(32a) implies that


1
J= (33)
2
Using (28a) and routine computations show that

dq1 (b) h
lim = (34)
b !b db D+4

and
A k
lim q1 (b) = (35)
b !b 2

Proof of lemma 5 Directly follows from (34) and (35).

9 Computations and proofs of rst stage equilibrium


outcomes
We now analyze the rst-stage game and solve for the equilibrium level of bribe chosen by
rm 1. In the rst stage rm 1 chooses the optimal level of bribe b to maximize its expected
payo (anticipating the equilibrium outcomes in stage 2 and 3). Let the optimal level of
bribe be b . We will rst deal with substitutes ( > 0).

45
9.1 Optimal bribe when goods are substitutes ( > 0)

We now provide some preliminary computations for the proofs of propositions 4-9.
Note that combining (13) and (14) we get
d 4h
= (36)
db D + 2 d d( )

d ( ) d 4h d ( )
Since by assumption 2 d
0 we must have db D
. Again by assumption 2 d
is
d ( )
non-decreasing in . From (30) to (33) we showed J = lim ! d
= 12 . This means
d ( ) 1 d 4h 8h 8h
d 2
for all 2 [0; ]. Hence, from (36) we have db D+ 12 2 = 2D+ 2 = 8 2 h
2
(as 1). From the above discussion we get that
d 4h
h (37) :
db D
d ( )
h 1 d
From lemma 3 we get that when b 2 b; b , q10 (b) = 2d ( ) . Using (36) we have
D+ d

d d ( )
q10 (b) = 1 (38)
db 4 d
From the main body of the paper we know that if b 2 b; b
A2
E1 (b) = F ( ) (q1 (b))2 + (q1 (b))2 b
4
This means if b 2 b; b
d A2
E10 (b) = f ( ) (q1 (b))2 + 2q1 (b) q10 (b) [1 F( )] 1
db 4
Using (38) we get for all b 2 b; b
8 h i 9
d < =
A2
0
f( ) 4
(q1 (b))2
E1 (b) = 1 (39)
db : + 1 q (b) 1 d ( )
[1 F( )] ;
2 1 d

Routine computations show that


A2
(q1 (b))2 = [2k + (0)] [4A 2k (0)] (40)
4 16
Note that (0) < and 2 [0; 1] as goods are substitutes. This means

If > 0
2
A
(q1 (b))2 < [2k + ] A (41)
4 4

46
Proof of Proposition 4 From previous lemmas and discussion we get that when >0
AB+ hb (0) dq1 (b) h
(goods are substitutes) and b 2 (0; b) we have q1 (b) = D
and db
= D
> 0.
When b 2 (0; b) we have E1 (b) = 1 (b) = (q1 (b))2 b. From (26) we have q1 (b) =
2A (0) 2k
4
. As q1 (:) is strictly increasing in b we get for all b 2 (0; b)

h h
E10 (b) = 0
1 (b) = 2q1 (b) 1 < 2q1 (b) 1
D D
2A (0) 2k h hA
= 1 1.
2 D D
D
The above shows that E10 (b) < 0 if h A
. By hypothesis of the proposition we have
D D
h A[f ( )(2k+ )+1]
. This implies h A
. Hence, we get E10 (b) < 0 for all b 2 (0; b).
Using (37) and (39) we get that for all b 2 b; b
8 h i 9
4h < =
A2 2
f( ) 4
(q1 (b))
E10 (b) 1 (42)
D : + 1 q1 (b) 1 d ( )
[1 F( )] ;
2 d

d ( )
Note that since q1 (b) is strictly increasing for all b 2 b; b ,1 F( ) 1 and 1 d
1
(from assumption 2) we get from above

4h A2 1
E10 (b) < f( ) (q1 (b))2 + lim q1 (b) 1 (43)
D 4 2 b!b

Since > 0 by using (35) and (41) in (43) we get that

4h 1 A k
E10 (b) < f ( ) [2k + ] A + 1
D 4 2 2
4h 1
< f ( ) [2k + ] A + A 1
D 4 4
h A
= ff ( ) [2k + ] + 1g 1 (44)
D
D h A
Note that h A[f ( )(2k+ )+1]
for all 2 [0; ] implies D
ff ( ) [2k + ] + 1g 1 0.
From (44) this implies E10 (b) < 0 for all b 2 b; b . We have already demonstrated that
E10 (b) < 0 for all b 2 (0; b). Also, E1 (b) is continuous for all b 2 0; b . All these together
D
imply that E1 (0) > E1 (b) for all b 2 0; b . Hence, if h A[f ( )(2k+ )+1]
for all 2 [0; ]
then b = 0.

47
Proof of Proposition 5 From (25) we get that when b 2 (0; b)
d 1(b) 2 [AB + hb (0)] h
= 1
db D2
2[AB+ hb (0)] h 2[AB (0)] h
Since > 0, we get D2
> D2
. Since 1, D 2 [3; 4] and B 2 [1; 3]
8 D2 2[AB (0)] h
we get h [A (0)]
implies that h 2 (AB (0))
and this turn implies D2
1 0
d 1 (b) 2[AB+ hb (0)] h
and this means db
= D2
1 > 0 for all b 2 (0; b). Note that E1 (b) = 1 (b)
8
for all b 2 (0; b). Hence we get that if h [A (0)]
then E10 (b) > 0 for all b 2 (0; b). Also,
E1 (b) is continuous for all b 2 0; b . Consequently, b b.

Proof of Proposition 6 Since > 0 implies that q1 (b) is strictly increasing in all b 2 b; b
we must have
A2 A2
(q1 (b))2 > lim (q1 (b))2
4 4 b!b
2 2
A A k
=
4 2
k
= (2A k )
4
k
(2A k) (since 1) (45)
4
d
From (37) we have h db
. Also, 1 F( ) 0. Using these facts and (45) in (39) we
have for all b 2 b; b
8 h i 9
d < f( ) A2
(q1 (b))2 =
4
E10 (b) = 1
db : + 1 q1 (b) 1 d ( )
[1 F( )] ;
2 d
k
> hf ( ) (2A k) 1 (46) :
4
Now note that from assumption 1 we have A 3k. This means
A 5A
2A k 2A = (47) :
3 3
Using (47) in (46) we get for all b 2 b; b
5k hf ( )
E10 (b) > 1 (50) :
12
n o
12 8 12
If h max 5f ( )k A
; [A (0)]
for all 2 [0; ] then h 5f ( )k A
. But this implies (from
(50)) for all b 2 b; b , E10 (b) > 0.

48
8
From proposition we know that if h [A (0)]
then b b. Consequently, h
n o
max 5f ( 12)k A ; [A 8 (0)] implies that b = b.

Proof of Proposition 7 Note that from propositions 4, 5 and 6 we get su cient conditions
under which b = 0 or b 2 b; b . We now rule out the possibility that b 2 (0; b). On the
contrary suppose that 0 < b < b. Note that when b 2 (0; b), E1 (b) = 1 (b). From (25) we
know that when b 2 (0; b)
d 1(b) 2 [AB + hb (0)] h
= 1
db D2
d 1 (b)
Clearly db
is strictly increasing in b when b 2 (0; b). This implies 1 (b) cannot achieve a
maximum in (0; b). Hence, b 2
= (0; b).

2AB+4 2kD AB (0)


Proof of Proposition 8 Note that b = 4h
(from (21). Also, q1 (0) = D
.
Routine computations show that
A2
b (q1 (0))2
4
0 1
2 2 2 2
1 @ 4hA B + hA D + 8hAB (0)
= A (51)
4hD2 2ABD 2 2 2
4h ( (0)) + 2kD 3
4 D 2

Using the fact that D = B (2 + ) and some algebraic manipulations, we get from above
A2
b (q1 (0))2
42 3
4A2 2 8A (0) 4 2( (0))2
14 (2+ )2
+A + D2
= D(2+ ) 5 (52)
4 1
(2AB 2kD + 4 )
h

Note that B 2 [1; 2] and D 2 [3; 4] and 2 [0; 1] (as goods are substitutes). Assumption 1
implies that 2AB 2kD = 2B [A k (2 + )] > 0. Also note that (0) < . This means
from (52)
A2 1 4A2 8A 1
b (q1 (0))2 < + A2 + (4 )
4 4 9 6 h
2
1 5A 4A 4
= + (53) :
4 9 3 h
Note that
36 5A2 4A 4
h =) + 0 (54) :
A [5A + 12 ] 9 3 h

49
Using (54) in (53) we get that

36 A2
h =) b (q1 (0))2 < 0 (55) :
A [5A + 12 ] 4

Now note that

A2
b (q1 (b))2 b
4 2 3
2
2 2A (0) 2k
A 2AB + 4 2kD 4 4 5
=
4 4h 2AB+ 2 (0) 2kD
4h
0 1
2 2 2
1 @ 4hk 4hk (0) + 8Ahk
= A (56)
16h h 2
( (0))2 + 4 2
(0) + 4Ah (0) 16

Using some algebraic manipulations, we get from above

A2
b (q1 (b))2 b
4 2 3
2
1 4 4k (k + (0)) + 4A (2k + (0))
= 5 (57)
16 2
( (0))2 + h4 ( 2
(0) 4 )

2 A
Note that 2 [0; 1], 1, k 3
(assumption 1) and (0) < . hence, from (57) we have

A2 1 4
b (q1 (b))2 b < 4A (2k + ) + ( 4 )
4 16 h
1 3
= A (2k + )
4 h
1 2A 3
A + (58)
4 3 h

Note that from (58) we have that

9 A2
If h then b (q1 (b))2 b <0 (59) :
A [2A + 3 ] 4

From (55) and (59) we get that

36 9
h max ;
A [5A + 12 ] A [2A + 3 ]
8 h 2 i 9
>
> A 2
b (q1 (0)) < 0 >
>
>
< 4 >
=
=) or (60)
>
> h 2 i >
>
>
: A >
b (q (b))
4
2
b <0 ; 1

50
Note that the optimal bribe, b b. If 1 chooses b = b entry is completely blockaded and it
A2
gets E1 b = 4
b. If 1 chooses b = 0, then entry is certain and it gets E1 (0) = 1 (0) =
(q1 (0))2 . If If 1 chooses b = b, then also entry is certain and it gets E1 (b) = 1 (b) =
(q1 (b))2 b.
From (60) we get that then

36 9
h max ; =) E1 b < max fE1 (0) ; E1 (b)g :
A [5A + 12 ] A [2A + 3 ]
h i
36 9
Hence h max A[5A+12 ]
; A[2A+3 ]
=) b < b:

9.2 Optimal bribe when goods are complements ( < 0)

Proof of Proposition 9 Note that when b 2 (0; b) we have E1 (b) = 1 (b) = (q1 (b))2 b.
dE1 (b)
Therefore when b 2 (0; b) we get db
= 2q1 (b) dqdb
1 (b)
1. From (24a) we get that for
dq1 (b) h
b 2 (0; b), db
= D
< 0 (since < 0). Hence, < 0 implies that E10 (b) < 0 for all
b 2 (0; b).
Now note that if b 2 b; b

A2
E1 (b) = F ( ) (q1 (b))2 + (q1 (b))2 b
4

From (44) we know that for all b 2 b; b

h A
E10 (b) < ff ( ) [2k + ] + 1g 1 (61)
D

Since < 0 from (61) we get that E10 (b) < 0 for all b 2 b; b . Also, E1 (b) is continuous for
all b 2 0; b . Hence, b = 0.

10 Computations and proofs on welfare


Note that in section 4.1 we have already derived the welfare function. We reproduce it below.

3 2
W (b) = A + q0 b F( )
8
2 3
Z
Aq1 (b) + (A hb + ) q2 ( ; b)
+ 4 5 dF ( ) (63)
1
2
fq12 (b) + q22 ( ; b) + 2 q1 (b) q2 ( ; b)g + q0 b k2

51
From above we can write the expected welfare as follows:
3 2 1 2
W (b) = A F( ) + Aq1 (b) q (b) k 2 (1 F( ))
8 2 1
Z Z
+ [A hb q1 (b)] q2 ( ; b) dF ( ) + q2 ( ; b) dF ( )
Z
1
q22 ( ; b) dF ( ) + q0 b (64)
2
Note that for b 2 [0; b) we have = 0 (all types enter). This means F ( ) = 0. Hence
for b 2 [0; b)
Z
1 2
W (b)jb2[0;b) = Aq1 (b) q (b) + [A
hb q1 (b)] q2 ( ; b) dF ( )
2 1 0
Z Z
1
+ q2 ( ; b) dF ( ) q22 ( ; b) dF ( )
0 2 0
+q0 b k 2 (65)

Hence, for b 2 [0; b)

Z
0
W (b)jb2[0;b) = (A q1 (b)) q10 (b) [h + q10 (b)] q2 ( ; b) dF ( )
0
Z
@q2 ( ; b)
+ [A hb q1 (b)] dF ( )
0 @b
Z Z
@q2 ( ; b) @q2 ( ; b)
+ dF ( ) q2 ( ; b) dF ( ) 1 (66)
0 @b 0 @b
To compute W 0 (b) for b 2 [0; b), note the following. From our earlier derivations we know
that Z
f( ) [AB 2hb] + 2 ( )
q2 ( ; b) d =
1 F( ) D
When b 2 [0; b), = 0 and we have
AB + hb (0)
q1 (b) = (67a)
D
2
2AB + D 4hb + (0)
q2 ( ; b) = (67b)
Z 2D
AB 2hb + 2 (0)
q2 ( ; b) f ( ) d = (67c)
0 D
From (67) we get that when b 2 [0; b)
h @q2 ( ; b) 2h
q10 (b) = and = (68)
D @b D
52
Hence, for b 2 [0; b) (using (66) and (68))

Z
0 h h 2
W (b)jb2[0;b) = (A q1 (b)) h+ q2 ( ; b) dF ( )
D D 0
Z
2h
[A hb q1 (b)] dF ( )
0 D
Z Z
2h 2h
dF ( ) + q2 ( ; b) dF ( ) 1 (69)
0 D 0 D
Using the fact that D + 2 = 4, B = 2 and using (69) we get that for b 2 [0; b)
2 R 3
h 4 (A q 1 (b)) 2 0 2
q ( ; b) dF ( )
W 0 (b)jb2[0;b) = 5 1
D 2 [A hb q1 (b)] 2 (0)
Z
h
= AB + q1 (b) + 2 [hb (0)] 2 q2 ( ; b) dF ( ) 1 (70)
D 0

We now compute W 0 (b) at b = 0.


Z
0 h
W (b)jb=0 = AB + q1 (0) 2 (0) 2 q2 ( ; 0) dF ( ) 1
D 0
h AB (0) AB + 2 (0)
= AB + 2 (0) 2 1 (71)
D D D

10.1 Case of substitutes ( > 0)

For substitutes since 2 (0; 1] using (71) above we can show that W 0 (b) < 0 at b = 0. This
means b = 0 is a local maximizer for W (b). Now we will show that b = 0 need not be the
global maximizer of W (b).
Note that when b = 0, all types of rm 2 enter (i.e. probability of entry is one). In this
case = 0 and the expected welfare is as follows:

2 3
Z
Aq1 (0) + (A + ) q2 ( ; 0)
W (0) = 4 5 dF ( ) (72)
1
0
2
fq12 (0) + q22 ( ; 0) + 2 q1 (0) q2 ( ; 0)g + q0 k 2

Note that when b = b then no type of rm 2 enters. This means = and rm 1 is a


monopolist. In this case, we have
3
W b = A2 + q0 b (73)
8
53
Now consider the following values of the parameters.

A = 3; k = 1; = 1; h = 1 and is uniformly distributed over [0; ] .

2 (0; 0:219)

All our assumptions are satised here and we have (0) = 2


. In this specic example we
also have
B = 1; D = 3; b = and b =
8
Hence when b = 0, we have

3 2
q1 (0) =
3
6+3 + 2
q2 (0; ) = (74)
6

Using (72), (73) and (74) routine computations show that

53 2 2
W (0) = + + 3 + q0
288 3
27
W b = + q0 (75)
8

Note that from (75) we have

53 2 5 3
W b W (0) = +
288 3 8

It may be noted that W b W (0) > 0 for all 2 (0; 0:219).

10.2 Case of complements ( < 0)

Note the following which follows from our earlier computations. Since < 0 (complements)
d ( )
and d
< 1 for all 2 [0; ] (assumption 2)
8
@q2 ( ; b) < 4h
< 0 for b 2 (0; b)
2D
= (76a)
@b : 1 d < 0 for b 2 b; b
2 db
8
>
<
h
< 0 for b 2 (0; b)
dq1 (b) D
= h 1
d ( ) (76b)
db >
: d

2d ( ) < 0 for b 2 b; b
D+ d

54
When < 0 by using (69) it is straightforward to show that W 0 (b) < 0 for all b 2 (0; b).
Now suppose b 2 b; b . Then from (63) we have

3 2
W (b) = A F( ) + q0 b
8
2 3
Z
Aq1 (b) + (A hb + ) q2 ( ; b)
+ 4 5f ( )d (77)
1
2
fq12 (b) + q22 ( ; b) + 2 q1 (b) q2 ( ; b)g k2

Note that 2 ( ; b) = (A hb + ) q2 ( ; b) k 2 = [q2 ( ; b)]2 k 2 . Using this in (77) we get


that when b 2 b; b

3 2
W (b) = A F ( ) + q0 b
8
2 3
Z
Aq1 (b) + 21 [q2 ( ; b)]2 k 2
+ 4 5f ( )d (78)
1 2
q (b)
2 1
q1 (b) q2 ( ; b)

From (78) we get that when b 2 b; b

3 2 d
W 0 (b) = A f( ) 1
8 db
2 3
1 2 2
d Aq1 (b) + [q2 ( ; b)] k
f( )4 2 5
db 1 2
q (b) q1 (b) q2 ( ; b)
2 1
2 3
Z
A dqdb
1 (b)
+ q2 ( ; b) @q2@b
( ;b)
+ 4 5f ( )d (79)
dq1 (b)
q1 (b) db
q1 (b) @q2@b
( ;b)
q2 ( ; b) dqdb
1 (b)

Note that 2 ( ; b) = [q2 ( ; b)]2 k 2 = 0 and q2 ( ; b) = k. Using these facts and (79) we
have that when b 2 b; b

3 2 d
W 0 (b) = A f( ) 1
8 db
d 1 2 1 2
f ( ) Aq1 (b) k q (b) kq1 (b)
db 2 2 1
2 3
Z dq1 (b)
(A q1 (b) q2 ( ; b))
+ 4 db 5f ( )d (80)
@q2 ( ;b)
+ @b (q2 ( ; b) q1 (b))

dq1 (b) @q2 ( ;b)


Note that from (76a) and (76b) we get that db
; @b < 0. This means

dq1 (b) @q2 ( ; b)


(A q1 (b) q2 ( ; b)) + (q2 ( ; b) q1 (b)) < 0 (81)
db @b
55
Also note that

d 1 2 1 2
Aq1 (b) k q (b) kq1 (b)
db 2 2 1
dq1 (b)
= [A q1 (b) k] < 0 (82)
db

(82) implies that

1 2 1 2
Aq1 (b) k q (b) kq1 (b)
2 2 1
1 2 1 2
> lim Aq1 (b) k q (b) kq1 (b)
b!b 2 2 1
3 1 2
= (A k)2 k (83)
8 2

Now note that

3 1 2 3 2
(A k)2 k A () 4A k 4 3 2
(84)
8 2 8

Proof of Proposition 11 Note that since for complements 2 ( 1; 0) we get that


A k =) 4A k (4 3 2 ). Now using (80) and (84) we get that W 0 (b) < 0 for
b 2 b; b . We also know that W 0 (b) < 0 for b 2 (0; b).

56

Vous aimerez peut-être aussi