Journal of Development Economics 99 (2012) 282291

Contents lists available at SciVerse ScienceDirect

Journal of Development Economics

journal homepage: www.elsevier.com/locate/devec

Mutual supervision in preshipment inspection programs

V. Dequiedt a,, A.-M. Geourjon b, G. Rota-Graziosi a, c
a
b
c

CERDI, Universit d'Auvergne, 65 Boulevard F. Mitterrand, 63000 Clermont-Ferrand, France

FERDI, 63 Boulevard F. Mitterrand, 63000 Clermont-Ferrand, France
International Monetary Fund, Fiscal Affairs Department, Tax Policy Division, Washington DC, United States

a r t i c l e

i n f o

Article history:
Received 17 December 2009
Received in revised form 5 January 2012
Accepted 14 February 2012
JEL:
D82
F13
L33
Keywords:
Preshipment inspection
Mutual supervision
Corruption
Customs administration

a b s t r a c t
Preshipment inspection (PSI) programmes are implemented in many developing countries to assist governments in the collection of revenue at their borders. These programs consist of the delegation of the inspection
of imports to a private rm. To study these PSI programmes, we develop a hierarchical agency model in
which the government authority can rely on two supervisors, namely the private inspection rm and the customs administration, to control importers' declarations. The government's optimal contract is fully characterized. We devote some attention to the inspection policy and its comparative statics properties. In particular,
we identify the situations in which PSI programmes are revenue enhancing. We also discuss the reconciliation
policy, i.e. what to do in case of conicting inspection reports by the inspection rm and the customs administration. In the optimal mechanism, mutual supervision between the inspection rm and the customs administration is used to provide adequate incentives to all parties.
2012 Elsevier B.V. All rights reserved.

1. Introduction
In many developing countries, customs revenue remains an important source of public nancial ressources. 1 However, in such
countries customs tariff collection is suspected to be particularly
inefcient. 2 Several factors may explain this inefciency. Most developing countries encounter major difculties in controlling porous borders. Despite ambitious reforms to modernize customs,
This paper beneted from the support of the FERDI (Fondation pour les Etudes et
Recherches sur le Dveloppement International) and of the programme d'investissements d'avenir of the french government.
Corresponding author.
1
More generally, the customs environment has a signicant impact on trade ows
and economic well-being. See Wilson et al. (2005) and the World Bank working papers
by Hoekman and Nicita (2008) and Njinkeu et al. (2008) for further details.
2
For instance, Bhagwati (1964) established that the discrepancies between a country's reported imports and the corresponding exports reported by its trading partners
may be explained by opportunistic undervaluation or misclassication of imports at
the border in order to reduce the tariff burden (see also de Wulf, 1981). Pritchett and
Sethi (1994) analyze the relation between the revenues collected and the ofcial rates
of taxation in some developing countries (Jamaica, Kenya, and Pakistan). They compare
the statutory ad valorem tariff rates (ofcial rates) with the ratio of tariff revenue to
import value (collected rates). They highlight that the collected rate is weakly related
to the ofcial rate. More recently, Fisman and Wei (2004) focus on evasion in China's
imports from Hong Kong at a higher level of disaggregation than the preceding paper.
They study empirically how tax evasion responds to tax rates, and they show a strong
positive relationship between them: products taxed at higher rates are more prone to
be affected by evasion.
0304-3878/$ see front matter 2012 Elsevier B.V. All rights reserved.
doi:10.1016/j.jdeveco.2012.02.005

administrative capacities in these countries are still weak and ex

post controls absent or ineffective. In addition, it is acknowledged
that corruption of customs administrations is widespread and
costs a non-negligeable share of potential customs revenue.
Frequently encouraged by international institutions, some countries have resorted to Pre-Shipment Inspection (PSI) programmes to
improve the efciency of tariff collection, and to ght customs administration corruption. 3 These programs consist of the delegation of an
inspection of imports to a private surveillance company (the PSI
rm) at embarkation ports, airports, in the exporting rms' premises
or at destination. Pre-shipment inspection complements information
provided in customs declarations and can be used to evaluate customs duties. This additional information should limit the discretionary power of customs ofcers and thus be an efcient means of
reducing customs corruption. The market for PSI services is oligopolistic: only four multinational rms share more than 90% of the market, which was estimated to be more than 650 millions dollars in
2007 (see Table 3).
The aim of this paper is to build a simple theoretical model that
captures the essential ingredients of PSI programmes and that can
be used to derive positive and normative implications for their design. We formalize the relationship between the four different parties
involved in those programs, namely the regulator, the customs

See the World Bank discussion paper by Low (1995) for a general presentation.

V. Dequiedt et al. / Journal of Development Economics 99 (2012) 282291

Table 1
PSI Programmes for Customs purposes (Revenue Protection).

283

Table 2
Customs support services (including Destination inspection and/or Selective PSI).

Country

Mandated member(s)
of IFIA PSI committee

Basis of member
choice

Country

Mandated member(s)
of IFIA PSI committee

Basis of member
choice

Angola
Bangladesh
Benin
Burkina Faso
Central African Rep.
Chad
Congo
Dem. Rep. Congo
Iran

BIVAC, Cotecna, SGS

BIVAC, Intertek, SGS, OMIC
BIVAC
Cotecna
BIVAC
BIVAC
Cotecna
BIVAC
BIVAC, Cotecna, Intertek,
SGS, OMIC
SGS
Cotecna
CU International, Cotecna,
Intertek, OMIC, SGS

Importer
Geographical
Monopoly
Monopoly
Monopoly
Monopoly
Monopoly
Monopoly
Importer

Burundi
Cameroon
Comoros
Cote d'Ivoire
Equatorial Guinea
Ghana
Guinea(Conakry)
Haiti
Liberia
Mali
Mozambique
Nigeria
Senegal
Sierra Leone
Tanzania
Togo

SGS
SGS
Cotecna
BIVAC
Cotecna
BIVAC, Cotecna
BIVAC
SGS
BIVAC
BIVAC
Intertek
Cotecna, SGS
Cotecna
Intertek
Cotecna
Cotecna

Monopoly
Monopoly
Monopoly
Monopoly
Monopoly
Air and land-/sea-freight
Monopoly
Monopoly
Monopoly
Monopoly
Monopoly
Port of arrival
Monopoly
Monopoly
Monopoly
Monopoly

Mauritania
Niger
Uzbekistan

Monopoly
Monopoly
Importer/exporter

Source: International Federation of Inspection Agencies (IFIA), June 2010.

Source: International Federation of Inspection Agencies (IFIA), June 2010.

administration, the PSI rm and the importer. In the spirit of Tirole

(1986) (see also Tirole, 1992), we develop a model of hierarchical
agency with two supervisors to capture the interactions between preshipment inspection and customs administration inspection. Equiped
with this model, we discuss properties of the optimal inspection policy
(i.e. when to inspect and who should inspect imports) and of the
optimal reconciliation policy (i.e. what to do with conicting pieces
of information).
Despite the importance of PSI programmes for public nance in
developing countries, the academic literature on PSI issues remains
very restricted. Only a few papers focus directly on PSI programmes:
Anson et al. (2006), Johnson (2001), Yang (2008a) and Yang (2008b).
Apart from the notable exception of Anson et al. (2006) which we discuss further below, those papers are empirical and evaluate the profitability of PSI programmes. Clearly, PSI programmes can help to
improve tax collection through outsourcing to the private sector:
the inspection rm (the PSI rm) contributes to the denition of
the tax base. 4 In this spirit, Yang (2008a) considers that PSI programmes offer the opportunity to hire integrity from the private
sector. He found that these programs increased import duties by
1530% during the rst ve years. On the other hand, Anson et al.
(2006) obtained less clear-cut results and according to their estimations, PSI programmes may decrease (in the case of the Philippines)
or increase (in the case of Argentina) fraud.
As the whole literature on contract theory suggests (see for instance Bolton and Dewatripont, 2005 or Laffont and Martimort,
2001), it is likely that the details of the design affect the efciency
of a whole program. However, in order to identify good practices,
an empirical strategy seems inappropriate because available data
are insufcient. This leads us to adopt a theoretical perspective and
to construct a simple model of PSI programmes.
We use a hierarchical agency framework in which the regulator
(i.e. the government authority which acts as a Principal) may employ
two different supervisors, the PSI rm and the customs administration, to control the declaration of imports by an agent (the importer).
Our model captures the following four central elements: (i) the regulator must deal with asymmetric information, as the agent knows
more than he does about the value of imports; (ii) the agent is opportunistic and will underdeclare if the fear of being detected is not
strong enough; (iii) the customs administration can inspect imported

4
The partial privatization of tax collection is not new. Tax farming in France before
the 1789 Revolution is certainly the most famous example. It may be used even today
in developed countries: e.g. in 2006, the Internal Revenue Service (IRS) planed to outsource the collection of some private debts (in fact 12,500 taxpayers whom owe $25
000 or less in back taxes) to three private companies (see IRS Enlists Help in Collecting
Delinquent Taxes, The New York Times, August 20, 2006).

goods but the customs ofcer in charge of this control may be corrupt
and enter into a side-agreement with the agent; (iv) the PSI rm can
also inspect the imports. This rm is not subject to corruption with
other parties but is nevertheless subject to opportunistic behavior
and will not hesitate to make false reports if it is in its interest to do
so.
We therefore model the PSI rm as an opportunistic but uncorrupted supervisor. The no corruption assumption may be questionable but is standard in the PSI literature. Johnson (2001), for instance,
argues:
Normally, given their geographical separation, there is little contact between customs authorities and PSI agents, thereby making
collusion between them extremely costly.
On the other hand, private and public actors agree that PSI rms
have a tendency to signal errors only when this increases the payment they receive. To deal with this opportunism they must receive
adequate incentives.
When studying corruption in customs administration, we abstract
extortion practices from the problem and concentrate on underinvoicing. It is plausible that doing so underestimates the benets of a PSI
program, because by limiting the discretionary power of customs
ofcers, PSI certainly makes extortion more difcult. However, our
optimal inspection policy for PSI programmes does not call for inspection by the customs administration on the equilibrium path: it is also
the optimal way to ght extortion.
Our model is reminiscent of double supervision models developed
by Bac and Bag (2006), Kofman and Lawarre (1993) or Mishra and
Anant (2006). Following those papers, we study the interaction between collusive and non-collusive supervisors. However, to take
into account specic aspects of PSI programmes, we are led to consider a timing that slightly differs from the one considered in those
Table 3
Revenues of the four biggest rms on the PSIs market in 2007.

SGS()
BIVAC
Intertek
Cotecna
Total

Total revenues
millions $

Total revenues
millions $

3751
2637
1095
216
7699

182
208
74
195
658

Sources: Union des Banques Suisses (UBS) and Safra Bank (Lux.) Estimations.
FS1: Socite Gnrale de Surveillance. From 1991 until 1997, Cotecna was an afliated
rm of SGS.

284

V. Dequiedt et al. / Journal of Development Economics 99 (2012) 282291

models. In our special context, the non-collusive supervisor (the PSI

rm) intervenes rst, before the collusive supervisor (i.e. the customs
administration).
On the positive side, our results help understand under which
conditions PSI programmes can be efcient, i.e. increase the regulator's revenue. It turns out that a very high level of corruption in the
customs administration may undermine the functioning of the PSI
contract. While a high level of corruption increases the desirability
of customs ofcer monitoring by the PSI rm, it also increases the
cost of such monitoring because customs ofcers cannot be used to
discipline the PSI rm.
On the normative side, our results stress the importance of commitment by the regulator to an inspection and a reconciliation policy.
The benets of commitment are well known in the agency literature,
but this is worth emphasizing in our context as it seems that many
countries decide, if not inspection at least reconciliation, ex post.
We also stress that the optimal reconciliation policy must solve a double moral hazard problem and must be designed to provide adequate
incentives both to the agent and to the PSI rm. It is clearly suboptimal to always trust one party; and mutual supervision is the key for
the discipline of both supervisors.
Anson et al. (2006) propose a different model to study PSI programmes. They focus on the moral hazard aspect of customs inspection, whereas our model is a pure adverse selection one. Their game
theoretic model is aimed at being positive, and providing testable implications. Our approach uses contract theory and is more normative.
As a consequence, their model and ours are complementary and can
be used to address different questions.
This paper is organized as follows. PSI programmes and their
prominent features are presented in Section 2. In Section 3, we construct step-by-step our model of PSI contracts and discuss several
useful benchmark cases. Section 4 is devoted to the characterization
of the optimal contract. There we derive the optimal probabilities of inspection. We also compare the different reconciliation policies that the
regulator can implement and their impact on inspection costs. Section 5
discusses positive and normative implications of our model.
2. How PSIs work
The rst PSI programme was implemented in Zaire (now Democratic
Republic of Congo) in 1963. It was then followed by more than fty
countries around the world (see Tables 1 and 2). At the beginning, the
main objective of PSI programmes was to reduce capital evasion due to
overinvoicing. As capital controls progressively disappeared, the objectives of PSI programmes changed, and they turned into an instrument
to ght import tariff evasion. Even today and despite some tax transition,5 customs declarations remain crucial for developing countries because a large proportion of nal consumption goods are imported and
thus a large proportion of the VAT is collected at the border.6
When a country implements a PSI program, it delegates to a private surveillance rm (the PSI rm) an inspection of imports. The
PSI rm has ofces all over the world and usually conducts inspection
in the exporting country. It can also inspect at destination. In both
cases (preshipment inspection or destination inspection), inspection
by the PSI rm occurs before customs inspection.
Under a PSI program, the importation process can be described as
follows: To activate preshipment inspection, the importer is required to
inform the PSI rm and to provide it with forms that include detailed information about the operation. Once informed, the PSI rm controls the
merchandise. It can either systematically inspect all imports or conduct
a risk analysis for control selectivity and target suspicious shipments. As

a result of inspection, the PSI rm issues a certicate of conformity or a

certicate of non-conformity. The customs administration intervenes
after the PSI rm. Once the customs declaration is accepted, the administration can decide to control by physical or non-physical inspection. At
this stage the control can again be systematic or selective. The customs
administration then issues the nal declaration. Duties and taxes to be
paid by the importer are those reported on this nal declaration. The
goods are released after the payment. They can still be subject to postrelease control. Ex post controls are well-developped in OECD countries
but in developing countries they are embryonic.
PSI programmes also organize reconciliation of information provided by the PSI rm with the information provided by the customs
administration. Such reconciliation is aimed at detecting potential
differences and undertaking follow-up actions.
Beyond these common elements, there is a great diversity in PSI
programmes. For instance, in some countries importers have to pay
for the PSI services, in others it is the government itself (see
Tables 1 and 2 for a presentation of current programs). The authority
that mandates a PSI may be the Ministry of Finance (Angola, Chad),
the Ministry of Trade (Ivory Coast), the Central Bank (Iran), or the Institute of Standards and Metrology (Jordan). The price of the PSI service varies, although it is usually based on the f.o.b. value of imports
and is xed at between 0.5 and 0.8% of this value.
3. The model
3.1. A model of corrupt customs administrations
The model is based on a simple version of the PrincipalSupervisor-Agent models that were developed following the pioneering contribution of Tirole (1986). A government authority or regulator, denoted R, is supposed to collect customs duties from an
importer, denoted A. These duties are proportional to the quantity
of good q, which can be thought of as being equivalent to the imports
value. The duties are xed at a level q where is the tariff rate. The
true quantity q is the importer's private information, and we assume
for simplicity that it takes value in {1, 2}.
The importer's decision to import the good is inuenced by many
factors, like the size of the market, the business cycle, etc., that are beyond the scope of our simple model; and we assume therefore that
q = 1 with an exogenous probability equal to 1/2 (again this value is
taken for simplicity) that is common knowledge. Our model is one
with adverse selection. All parties are risk neutral.
3.1.1. Complete information
Under complete information, i.e. when the regulator observes the
true q, customs duties collection is a trivial task and the regulator obtains a transfer equal to
C

3
:
2

3.1.2. Incomplete information and no supervision

Under incomplete information, i.e. when the regulator does not
observe the true q, customs duties collection is more problematic because of the importer's opportunistic behavior. In the absence of any
supervisory inspection, A would minimize tariff payment and would
always declare 1 because the regulator is unable to verify the importer's declaration. In that case, the regulator would collect
I

R b R :

Tax transition consists in reducing import tariffs and increasing revenues from domestic taxation such as Value Added Tax.
6
Keen (2003), chapter 1, provides gures showing that in many developing countries, more than one half of the total VAT is collected on imports.

Underinvoicing would dramatically reduce tariff collection by the

regulator. This benchmark plays some role in the analysis that follows
and we refer to it as the no inspection policy.

V. Dequiedt et al. / Journal of Development Economics 99 (2012) 282291

3.1.3. Uncorrupted supervision

To avoid underinvoicing and to enforce the collection of customs
duties, the regulator employs the customs administration, which is
modeled as a single supervisor S1. Upon receiving A's customs declaration q~ , the regulator can ask S1 to inspect the merchandises. At an
opportunity cost c1, S1 will learn the true value q and must be reimbursed for those inspection costs by the regulator. The variable c1 is
a measure of the efciency of the customs administration. When the
customs administration is uncorrupted, the information obtained by
S1 is truthfully transmitted to the regulator. If the report of the customs ofcer differs from the imports declaration, the regulator can
impose a ne FA on the importer. To be effective the ne must be
such that FA > : the regulator must impose a penalty on the fraudulent importers in addition to the payment of truthful duties. This is
what we assume in the following.
In that situation, customs administration inspection is used to discipline the importer and the probability of inspection must be adjusted to provide incentives for truthful declaration. Let us denote p1()
the contingent probability of inspection. Of course, costly inspection
is unnecessary when the report of the importer q~ is equal to 2: in
equilibrium, p1(2) = 0. Now we denote p1(1) = p1.
Incentive compatibility for the importer's declaration imposes
p1 F A ;
where the left hand side is the prot of misreporting, while the right
hand side is the expected cost of doing so. Because inspection is costly, this incentive constraint is binding in a truthful equilibrium and
the regulator optimally sets
p1

:
FA

The equilibrium probability of inspection p1 increases with the

tariff rate and decreases with the ne that can be imposed on the
importer FA. However, whether the regulator prefers inducing a
truthful declaration or accepting equilibrium underinvoicing depends
on the cost c1 of inspection. The expected prot of the regulator is



3

R max ;
c1 ;
2
2F A
and the regulator prefers inducing a truthful declaration whenever
c1 FA.
3.1.4. Corrupt supervision
As argued above, the hypothesis of an uncorrupted customs administration is not suitable for most countries that implement a PSI
program. We now introduce corruption in our simple model. We
still assume that the customs administration (S1) is paid according
to a xed wage that reimburses its cost c1. However, we consider
now that S1 may be corruptible. With probability (1-s), the customs
ofcer that controls the merchandise is honest and sends to the
regulator the true value q that he observed. With probability s, the
customs ofcer is dishonest and colludes with the importer A. 7 The
variable s measures the degree of corruption of the customs administration. Collusion is valuable for the importer/customs ofcer coalition
when the importer declares q~ 1 while the true value, observed by
the ofcer, is q = 2. In that case, we assume that bargaining between
the importer and the ofcer takes place under complete information
and is Pareto efcient.
7
This is a shortcut. The rationale may be that the regulator imposes a constant ne
FS1 to corrupt ofcers, and that heterogeneous ofcers differ in their subjective evaluation of this ne. A proportion 1 s of ofcers consider the ne sufcient and prefer not
to enter collusive agreements. A proportion s consider it is worth being corrupt.

285

The outcome of this bargaining may depend on the perishability of

the product or on the cost related to the release time of the merchandise. In the sequel, we assume that the importer can secure a report
q = 1 from the ofcer by paying him /2 which is to be compared to
the ne FA. 8
The timing of events is given below:

At
At
At
At
At
At

date t = 0, the importer A learns its type q.

date t = 1, A declares q~ to R.
date t = 2, with probability p1 q~ , S1 inspects the merchandises.
date t = 3, S1 and A may collude.
date t = 4, S1 sends a report q~ 1 to R.
date t = 5, transfers take place.

The only tool available to the regulator remains the probability of

inspection p1(). Of course, in our two-type model, there is no equilibrium inspection when the importer declares q~ 2 and we still denote
p1 = p1(1). Let us dene the incentive compatibility constraint of an
importer that should declare 2. In order to prevent the importer
from declaring 1, the probability of being inspected must satisfy


p1 1sF A s :
2
The left hand side of this equation characterizes again the potential gains of the importer from undervoicing (declaring 1 instead of
the truthful report q~ 2). The right hand side is the expected cost
of doing so: if it is inspected, which occurs with probability p1, the importer must pay FA if the customs ofcer is honest, which occurs with
probability 1-s, while it must pay the bribe /2 if the ofcer is dishonest (which occurs with probability s).
Under the optimal truthful contract, the regulator minimizes the
inspection costs subject to the incentive constraint and sets:
p1

;
F A sF A =2

which is an increasing function of s, the degree of corruption of the

customs administration. With a more corrupt customs administration, it is necessary to inspect imports more often, so that the probability of being inspected by an honest ofcer is high enough.
It is possible to determine whether it is worth providing incentives for truthful declaration by comparing RI and RC minus the
expected cost of inspection.
Proposition 1. When the regulator can only use a corrupt customs
administration to collect duties, his payoff is



3

c1 :
R max ;
2
2F A s2F A
From this expression we can compute the threshold c^1 over which
the regulator prefers to let underinvoicing occur in equilibrium. This
threshold c^1 depends now on the tariff rate . It is given by


c^1 F A 1s s :
2
Notice that it increases with the tariff rate and decreases with
the degree of corruption of the customs administration s. We can
state the same property differently and isolate s instead of c1. In so
doing, we identify a threshold in the corruptibility of the customs administration above which the regulator prefers to implement the no

8
The ex post stake in bargaining is FA, but the ex ante stake is . If the share of the ex
post stake that the importer can secure is too low, it would anticipate a hold up problem and would never underinvoice in the rst place. So for simplicity we arbitrarily assume that bargainers equally share the ex ante stake.

286

V. Dequiedt et al. / Journal of Development Economics 99 (2012) 282291

inspection policy and let underinvoicing occur. Such a threshold ^s 1 is

dened by:
^s 1

F A c1
:
F A 2

One may wonder whether it is possible to decrease the impact of

corruption and therefore to increase the payoff of the regulator by
using an incentive wage policy at the customs administration, i.e. in
our model, by offering to S1, a higher wage in case it reports 2. Such
a possibility is beyond the scope of our analysis which considers
that the degree of corruption s is an exogenous parameter. Moreover,
as shown in Besley and McLaren (1993), an increasing efciency
wage might be too costly to implement if the proportion of honest
customs ofcers is high enough and the regulator might prefer to
let corruption occur and pay a constant reservation wage. In the
same vein, Kofman and Lawarre (1996b) prove, in an optimal contract setting, that deterring corruption might be too costly for the
regulator if he cannot screen the supervisor on the corruptibility
dimension. Let us nally mention that in our model information is
soft (a supervisor can overestimate the f.o.b. value) and nothing prevents S1 from declaring 2 even if the true value is 1. For the regulator,
the costs associated with such a strategy are not explicit in the model.
However, it is reasonable to think that reputation effects would lead
the importers to change their plans in the future in reaction to such
extortion and that the regulator would be hurt in the long run.
For further reference, when the regulator chooses to implement
inspection by the customs administration with p1 dened in Eq. (3),
we denote it as the customs-only inspection policy.
3.2. The PSI game
Consider now the possibility that the regulator uses another
source of information, namely a preshipment inspection rm S2.
This second supervisor can carry out inspection of the merchandises
and intervenes before the customs administration. We shall assume that
this supervisor is not subject to corruption. When the regulator asks S2
to inspect, S2 observes the true value of q (without cost, for simplicity)
and reports q~ 2 . The regulator pays it q~ 2 , where [0;]. This wage prole is the result of bargaining between the different countries interested
in PSI programmes and the small number of rms offering this kind of service. We assume for simplicity that it is exogenously xed.9 Even if this
supervisor is not corruptible, the problem with S2 comes from the fact
that it has an incentive to always exaggerate q~ 2 in order to receive a
higher wage. If the regulator uses S2 alone, it would not acquire any valuable information about the true value q. Therefore, it has to combine
inspections by S1 and S2, taking into account the propensity of S1 to minimize q, because of corruption, and of S2 to exaggerate q because of opportunistic behavior. We assume that in case the regulator is persuaded that
S2 exaggerates the quantity q~ 2 , it can impose a ne FS2 >.10 The second
supervisor is risk neutral and maximizes its expected revenue.
The timing is now as follows:
At date t = 0, the importer A learns its type q,
At date t = 1, A declares q~ to R,
At date t = 2, with probability p2 q~ , R asks S2 to inspect the
merchandise,
At date t = 3, S2 sends a report q~ 2 ,
9
The linearity of S2's wage prole is assumed to stick with real world practices
where S2 is paid between 0.5 and 0.8% of the inspected merchandise value. This assumption could be relaxed. What is actually important for our results is that S2 is paid
more when it reports 2 than when it reports 1. This hypothesis could be justied in the
rst place by the consideration of a moral hazard problem in the delegation of the supervisory task to the PSI rm. For simplicity, we ignore this moral hazard problem and
we simply assume that S2's wage is linear and increasing.
10
This static ne FS2 may have a dynamic interpretation and correspond to the
expected loss incurred by the PSI rm when its contract is not renewed.

At date t = 4, with probability p1 q~ ; q~ 2 R asks S1 to inspect the

merchandise,
At date t = 5, S1 and A may collude,
At date t = 6, S1 sends a report q~ 1 to R,
At date t = 7, transfers take place.
When a supervisor is not asked to inspect, its report is by convention. The instruments available to the regulator are the two contingent probabilities of inspection (p1(), p2()). Because inspection
is costly, it is of course optimal to set p2(2) = 0, p1(2, ) = 0 and
p1(1, 1) = 0. The regulator must still choose p2(1) = p2, p1(1, 2) = p1
and p1(1, ) = p1(). These endogenous variables characterize
what we call the inspection policy. The no inspection policy is one
particular case with p1 = p2 = 0. The customs-only inspection policy
is another particular case with p2 = 0.
It is also possible for the regulator to choose which report determines the administrative truth, i.e. which report will be used to calculate duties and/or nes. By selecting one value, the regulator
determines which parties are to be ned in case of conicting reports
(the parties that made a report conicting with the truth). Of particular importance are the decisions to adopt rst, in cases q~ 1, q~ 2 2
and q~ 1 , second, in cases q~ 1, q~ 2 2 and q~ 1 1. Such decisions
characterize what we will call the reconciliation policy.
Theoretically in our model it would be optimal for the regulator to
ne both A and S2 in case two reports conict. However, it would be
problematic to assume that the regulator can implement such a punishment strategy because it would rely on the threat of punishing
both parties in case of a conict between A and S2. 11 In other words,
the regulator would have to use the same proofs both against A and
S2. Moreover, in the context of customs duties and due to regulatory
constraints, the decision of the regulator is not generally to choose
the parties to be punished but rather to choose which report determines the true value. To stick with this interpretation, we will assume
in what follows that the regulator can only punish one party at a time.
The reconciliation policy may be stochastic but in any case, the regulator can only ne A or S2.

4. The optimal PSI contract

4.1. Regular truthful equilibrium
We rst characterize the optimal contract that induces truthful
reporting in the continuation equilibrium, both by A and S2.
Suppose in addition that p1(0, ) =0 and that R imposes a ne on A
when A's report conicts with S2's report, unless S1's report conrms A's
report. In that latter case (which occurs when q~ q~ 1 1 and q~ 2 2), R
imposes a ne on S2. In the following, a contract such that the continuation equilibrium satises these properties will be called a regular truthful
contract.12 A complete justication for focusing on this type of continuation equilibria is postponed to Section 4.3.13
11
In a PrincipalSupervisorAgent setting with two supervisors (Kofman and Lawarre,
1996a) construct a mechanism that possesses prisoners' dilemma features to eliminate
corruption. However they aknowledge that such a mechanism is hardly seen in the real
world.
12
Note that under a regular truthful contract, the nes are imposed only out of the
continuation equilibrium. They are just threats. This means that honest (equilibrium)
behavior is never punished: certainly a desirable property, even if its long term benets are not modeled here.
13
Regular truthful contracts seem to be desirable from the regulator's point of view.
An intuition can be gained by recognizing that both the agent and the PSI rm must be
incentivized and therefore must be punished in some state of nature. Moreover, it only
makes sense to punish S2 when q~ q~ 1 1 and q~ 2 2 because it is the only circumstance in which it is reasonable to suspect a misbehavior. The most efcient way to incentivize S2 therefore seems to be to impose the ne on S2 in that circumstance. And
then, the incentives to A are provided by punishing it when q~ 1, q~ 1 and
q~ 2 2. Therefore, it seems likely that regular truthful contracts minimize the inspection costs. Such a result is rigorously established below in Section 4.3.

V. Dequiedt et al. / Journal of Development Economics 99 (2012) 282291

287

In a regular truthful contract, the relevant incentive constraint for

A can be written

4.2. Optimal inspection policy



p2 1p1 sF A p1 s :
2

In this section, we check whether implementing the optimal regular truthful equilibrium identied in Proposition 2 is indeed the optimal inspection policy available to the regulator. The regulator
basically faces three options. He can decide to implement the optimal
regular truthful equilibrium identied in Proposition 2, i.e. decide to
use the PSI rm; he can decide to use the customs administration
only and implement the inspection policy identied in Eq. (3); or he
can simply choose to let underinvoicing occur and decide to use neither the PSI rm, nor the customs administration. In order to compare
the last two options, we have already introduced the threshold ^s 1 dened in Eq. (4). For the two other pairs of options, we will need

The right hand side characterizes the expected cost of declaring 1

when q = 2. In that case, the agent is ned FA when S2 inspects (which
occurs with probability p2 in a truthful continuation equilibrium), unless S1 inspects also and colludes with A (which occurs with probability p1s). When the latter occurs, A must nevertheless pay a bribe /2
to S1. The left hand side of Eq. (5) is the benet of declaring 1 instead
of 2.
In a truthful equilibrium, S2 must have an incentive not to exaggerate its report if it observes q = 1. The corresponding incentive constraint is

^s 2

FS

p1 F S2 ;

the benet from lying must be less than the expected cost of doing so.
Because inspection is costly, R wants to minimize the probabilities
p1 and p2 and constraints (5) and (6) will bind at the optimum. 14
Proposition 2. The optimal regular truthful contract is characterized by
the probabilities of inspection


p1

;
F S2

which denes the indifference cutoff between the optimal regular

truthful equilibrium identied in Proposition 2 and the no inspection
policy; and
^s 3

FA


F A 2

c1

c1

S2

which denes the indifference cutoff between the optimal regular

truthful equilibrium and the inspection policy identied in Eq. (3).
Proposition 3. The corruptibility thresholds ^s 1 , ^s 2 and ^s 3 satisfy the
following properties:

and
p2

F A

;
F A 2

sF
FA FA 2
S

Some comparative statics are illuminating. First, and as intuition

suggests, p2 increases with s the corruptibility of the customs administration, and decreases with FA, the ne that can be imposed on a dishonest importer. More interestingly, p2 also decreases with FS2 the
ne that can be imposed on a dishonest PSI supervisor. This is so because a higher FS2 allows a reduction in p1, which in turn means that A
will have fewer opportunities to collude and escape the penalty FA. It
is therefore easier to provide incentives to A. Finally, p2 increases with
. This means that a badly negotiated PSI contract leads to higher
costs for two reasons. The rst one is a direct effect, a higher
means that it is more costly for the regulator to implement a given
probability of inspection by S2. The second effect is indirect, a higher
makes it more difcult to provide incentives to S2. It calls for a
higher p1, which in turns means that A will be able to collude more
often. Therefore, inducing truthful reporting by A necessitates a
higher p2.
Note that in a regular truthful equilibrium, the customs administration never inspects because S2's report never conicts with A's report. Therefore, the equilibrium value p1 has no direct impact on the
regulator's prot. It has only an indirect impact through its inuence
on p2 .
Note also that implementing the probabilities of inspection characterized in Proposition 2 requires commitment by the regulator. It
is well known from the literature on auditing (see Khalil, 1997 or
Strausz, 1997) that the optimal inspection strategy that can be played
with commitment may not be subgame perfect if the regulator cannot
commit.
14
There is another incentive constraint that we must take into account. Out of equilibrium, if S2 observes q = 2, it must prefer reporting of the truth rather than 1. This
path of events does not occur in a truthful equilibrium, but we implicitly assume this
constraint is satised when we wrote Eq. (5). This additional constraint is p1sFS2.
Provided, inequality (6) binds, it is automatically satised.

(i) If c1, then ^s 1 ^s 2 ;

(ii) If > c1, then the thresholds ^s 1 , ^s 2 and ^s 3 are such that either
^s 3 b^s 1 b^s 2 or ^s 2 b^s 1 b^s 3 , or ^s 1 ^s 2 ^s 3 ;
The optimal inspection policy is characterized by:
(j) If c1, the regulator implements the optimal regular truthful
equilibrium characterized in Proposition 2 when s is lower
than ^s 2 and prefers the no inspection policy when s is higher
than ^s 2 .
(jj) If > c1 and ^s 3 ^s 1 ^s 2 , the regulator chooses the customsonly inspection policy when s is lower than ^s 3 , implements
the optimal regular truthful equilibrium characterized in
Proposition 2 when ^s 3 s^s 2 , and prefers the no inspection
policy when s^s 2 ,
(jjj) If > c1 and ^s 2 ^s 1 ^s 3 , the regulator chooses the customsonly inspection policy when s is lower than ^s 1 , and prefers
the no inspection policy when s^s 1 .
Proof. See Appendix A.
It is illuminating to illustrate graphically the content of
Proposition 3. We show in Fig. 1 the optimal inspection policy in
the corruptibilityinefciency plan, s; c1 R  R .
To each value of s and c1, it associates the preferred policy (i.e. PSI,
customs only or no inspection) of the regulator. Entering a PSI programme is clearly not optimal for all countries. Proposition 3 claries
the impact of the different parameters of our model on the desirability of such a program. In case (jjj) for instance, entering a PSI programme is always dominated, whatever the degree of corruption of
the customs administration. Such a case occurs when > c1, i.e. an
uncorrupted customs administration would cost less than the PSI
rm, and is lower but close enough to FS2, which means that S2 is
difcult to incentivize.
If all other things being equal, c1 grows and gets closer to , we
turn to case (jj). It is optimal to enter a PSI programme for intermediate degrees of corruption of the customs administration. In this case,
the customs-only inspection policy is preferred for low degrees of

288

V. Dequiedt et al. / Journal of Development Economics 99 (2012) 282291

s
No Inspection
s 1
C

A
s 2

Customs
only

PSI
s 3

c1

Fig. 1. Optimal inspection policy.

administration and one could think that S2's report is more reliable
than S1's report.
It is important to realize that the optimality of a reconciliation policy that is unfavorable to S2 is an equilibrium phenomenon, i.e. it is
true if the regulator carefully designs all other aspects of the program.
Intuitively, in equilibrium, inspection by the customs administration would be used to discipline the opportunistic supervisor S2 and
not directly to discipline the agent. Moreover, in our model, corruption of the customs administration is supposed not to occur in equilibrium if S2 is used sufciently frequently. Actually, with an
imperfect inspection technology, corruption and underinvoicing are
controlled through the reconciliation policy implemented when
q~ 1, q~ 2 2 and q~ 1 , while the oppportunisitc behavior of S2 is
controlled through the reconciliation policy implemented when q~
q~ 1 1 and q~ 2 2. This explains why it is optimal to believe S2
when q~ 1, q~ 2 2 and q~ 1 , and to believe A when q~ q~ 1 1
and q~ 2 2.
5. Discussion

corruption while the no inspection policy is optimal for very high degrees of corruption.
Finally, when the customs administration is very inefcient, with
c1 > , it is never optimal to use the customs-only inspection policy.
This is case (j) and for low degrees of corruption, it is optimal to
enter a PSI programme, while for high degrees of corruption, the no
inspection policy is preferred.
A discussion of the positive and normative content of Proposition
3 follows in Section 5. Before that, we scrutinize the reconciliation
policy in the next subsection.

5.1. Understanding the efciency of PSI programmes

Proposition 5. When S2 's inspection technology is imperfect, the regulator

minimizes the inspection costs while ensuring truthful reporting by A and S2
when he implements the reconciliation policy that is favorable to the importer
when q~ q~ 1 1 and q~ 2 2.

Let us come back to Fig. 1 and its positive interpretation. Entering

a PSI programme is not optimal for all countries. In particular, when
the level of corruption in the customs administration is too high, it
may be preferable to simply let underinvoicing occur. This is what
happens for a country in A for instance. For those countries with
high level of corruption, PSI programmes are not the solution and it
may be preferable to tackle the customs corruption problem more directly. By contrast, below a critical level of corruption and above a
customs' cost of control, the PSI programmes are optimal and so justied. This is the case for a country in B.
In the corruptibilityinefciency plan, a vertical move to the bottom (from A to B for instance) corresponds to a reduction in the
parameter s. It can be interpreted as an ethical modernization of
customs. Depending on the initial situation of the country, a PSI programme can become the optimal policy if the customs administration
is rather inefcient. If the cost of customs control (c1) is sufciently
low (an efcient customs administration), this ethical modernization increases the likelihood that the customs-only inspection policy
is optimal. To illustrate what is meant by ethical modernization,
let us mention that in 1993, in recognition of the signicant problems
of corruption in customs administrations, the World Customs Organization (WCO) adopted the Arusha Declaration. The declaration sets
out the main features that should be present to reduce corruption.
Practical measures to promote integrity in customs modernization
programs, often designed with technical assistance from international
institutions (IMF, World Bank, WCO), aim at providing a framework
for changing the incentive structure and for establishing the legal
and administrative procedures that are necessary to detect, punish,
and reduce corruption (de Wulf and Sokol, 2004; Keen, 2003). 15
Moving horizontally to the left (from the point A to C or from B to
D for instance) corresponds to a reduction in the parameter c1. This
move can be interpreted as a technical modernization of customs.
Some components of this technical modernization are usually included in PSI proposals by inspection companies to obtain the contract.
Moving from B to D in Fig. 1 increases the likelihood that the
customs-only inspection policy is optimal. Contracts that include customs' modernization services are clearly problematic because if modernization is successful, the contract will not be renewed.
Morocco and the Philippines are regarded as two examples of successful customs modernization (see Keen, 2003). In 1996, in Morocco,

Proof. See Appendix C.

The result stated in Proposition 5 may seem counterintuitive because PSI programmes are used to ght corruption in the customs

15
Relying on an honest ex post audit to discipline customs ofcers might be an effective measure and result in a decrease in s. We thank an anonymous referee for suggesting this point.

4.3. Optimal reconciliation policy

How to reconcile conicting information provided by the PSI rm
and the customs administration is a hotly debated practical issue.
Translated into the terms of our model, this amounts to knowing
how the regulator should behave when q~ q~ 1 1 whereas q~ 2 2,
and when q~ 1, q~ 2 2 and q~ 1 . Should he ne A because this certainly corresponds to a case of underinvoicing with or without corruption, or should he ne S2 because this certainly corresponds to an
opportunistic strategy of that party? So far we have assumed that facing
the rst situation, the regulator chooses to ne S2 while he chooses to
ne A when facing the second. This turns out to be an optimal reconcialiation policy as we show in the next two propositions.
Proposition 4. The regulator strictly prefers to always punish A when
q~ 1, q~ 2 2 and q~ 1 ; and he is indifferent between the various
incentive compatible reconciliation policies when q~ q~ 1 1 and q~ 2 2.
Proof. See Appendix B.
In order to discuss more accurately what is optimal when q~ q~ 1
1 and q~ 2 2, we modify slightly the model in order to introduce some
imperfection in the inspection technology of S2. We assume that
when q = 2, if S2 inspects, it observes 2 with probability 1, while, if
q = 1, S2 observes 2 with probability and 1 with probability (1 )
( being small). Here again we are interested in determining the reconciliation policy that minimizes the expected inspection costs and that
induces truthful reporting by A and S2 on the equilibrium path.

V. Dequiedt et al. / Journal of Development Economics 99 (2012) 282291

the average delay for containers in Casablanca was 16 days, of which

10 days were attributable to customs controls. The government realized how corrupt the customs administration was. The customs CEO
was red and replaced by a manager from the private sector. The
King fully supported the customs modernization project that this
new CEO carried out. After three years, the processing time for customs declarations had fallen from ten days to 3 h, and after ve
years 85% of declarations were processed in less than an hour and a
half; all with no apparent cost in terms of customs revenue collected.
Morocco never signed a PSI contract. In broad outline, Morocco
moved from point A to point D in Fig. 1 without crossing the zone
where PSI is optimal.
In the Philippines, despite the huge increase in import volumes
(by more than 160% between 1990 and 1996) cargo clearance times
fell from an average of approx. one week to less than 48 h for selected
shipments and less than 15 min for super green lane shipments. After
this successful modernization the Philippines did not renew its PSI
contract (see Tables 1 and 2). In broad outline, the Philippines
moved from B to D.
On the other hand, in most Subsaharan African countries (e.g.
Benin, Mali, Niger among others) PSI programmes have been implemented for decades and are systematically renewed. In those countries, the level of corruption is too high and/or the technical
modernization is too weak to abandon the PSI programmes.
5.2. Are existing PSI programmes optimally designed?
The positive reading of Fig. 1 that we performed above is based
upon the assumption that existing PSI programmes are optimally
designed. A careful interpretation of our results also requires comparison of the functioning of existing PSI programmes with the prominent features of optimal programs as identied in Propositions 2
and 3 above. To organize the comparison, we summarize these features in four properties of optimal programs: (i) some selectivity of
controls, (ii) a reconciliation procedure, (iii) the primacy of customs
information in this reconciliation procedure, and nally (iv) strong
commitment by the regulator. Let us discuss these points in turn.
(i) Selectivity of controls is necessary both at the private surveillance company level and at the customs administration level
in order to minimize the cost of these controls. The optimal
contract identied in Proposition 2 involves inspection probabilities strictly lower than 1 for both supervisors. PSI rms as
well as customs administrations have long been reluctant to
introduce selectivity. For PSI rms, this can be understood by
looking at their remuneration schedule which has always
been a percentage of the inspected merchandise value. Nevertheless more and more PSI contracts include selectivity requirements to lower the cost for the regulators. The evolution
towards selectivity by customs administration is much slower.
Some PSI contracts include the provision, to the customs administration, of a risk analysis conducted by the PSI company
based on its own controls data. Customs administrations seldom use the results of these analyses. Corrupt customs ofcers
clearly prefer to inspect all merchandise to maximize their private benets. They argue that risk analyses based on PSI rm's
data are unreliable because they do not use ofcial customs litigation data. They also question the impartiality of PSI companies when they provide a risk analysis for customs use.
Recently, customs administrations in Burkina Faso, Ivory
Coast, Mali, and Senegal started to develop their own risk analysis based on the results of both the PSI rms and the customs
administration controls.
(ii) The reconciliation of information provided by the PSI rm and
the customs administration is an important element of optimal
PSI contracts. In most PSI programmes reconciliation policy is

289

unsuccessful, even if some reconciliation procedure is included. There are two main reasons for this failure. First, data crosschecking suffers from the lack of computer connection
between PSI rm and customs administration. Second, the
two pieces of information are hard to compare since the
goods listed in the certicate issued by the PSI rm may be
imported in separate shipments and generate several customs
declarations. There have been some experiences, in Mauritania
for instance, of reconciliation by a high level steering committee of the Ministry of Finance using data provided by the PSI
rm and the customs administration. Unfortunately these experiences remain sporadic, and ministers in charge of import
verication programs are reluctant to undertake concrete
follow-up actions.
(iii) In a few countries, the regulator imposes customs duties that
are based on the import value certied by the PSI rm. However, in most countries, in case there is a conict between the information provided on the PSI rm certicate and that coming
from customs control, the customs duties are based on the latter: a consequence of the sovereign power of the customs
administration.
(iv) The regulator's involvement in the design and implementation
of PSI contracts is necessary to improve reconciliation practices
and to enforce selectivity in customs and PSI rm controls.
Multilateral institutions try to encourage this involvement,
and governments seem to realize more and more that PSI contracts must not be managed by the customs administration
alone. For instance, according to the last Memorandum of Economics and Financial Policies, the Malian government intends
to introduce an automated targeted system for customs inspection by 2012 (see Mali: Letter of Intent, IMF, May 2011 16).
To summarize this discussion, existing PSI contracts certainly do
not share all the characteristics of optimal contracts as identied in
our model. However most of them possess one or another of these
characteristics. Most importantly, they seem to be evolving broadly
in the right direction, i.e. so as to get closer to the optimal contract.
5.3. Where to go with the PSIs?
Our model suggests several conclusions about the future of PSI
contracts, which have recently been questioned during WTO negotiations on exchange facilitation. Our theoretical analysis shows under
which circumstances, especially the corruption level of customs administration and its degree of modernization, a PSI contract can be efcient, i.e. can increase customs revenue. We have already mentioned
in a previous paragraph the necessity for the regulator to actively implement the PSI program. The regulator's commitment is crucial both
for the design and for the day-to-day operating of the contract.
The regulator should also pay attention to potential conicts in the
objectives of PSI contracts. A PSI contract should not incorporate a
skill transfer program from the PSI rm to the customs administration
for at least two reasons. First, as illustrated in Fig. 1, the PSI rm has
no incentive to help the customs administration modernize. By
doing so it would help the country move in a direction where PSI programmes are useless. Second, in a PSI contract, the mechanics of mutual supervision works correctly only if the two supervisors do not
communicate. Otherwise, collusion between them might hurt the
regulator. Skill transfer programs necessitate communication for instance to share knowledge about risk analysis or to develop the use
of computers. It is likely that this also facilitates collusion between
the PSI rm and the customs administration. An improvement for future programs concerning customs in developing countries would be
to separate the objective of customs modernization and the objective
16

Available at http://www.imf.org/external/np/loi/2011/mli/052611.pdf.

290

V. Dequiedt et al. / Journal of Development Economics 99 (2012) 282291

of monitoring customs corruption and address them via two separate

contracts.

administration. More precisely, the regulator chooses to set p1()

= 0 when

Appendix A. Proof of Proposition 3

0

1
F A Fs F A 2
S2

 A;
b c1 @
F A s F A 2

First, let us compare what the regulator obtains in a regular truthful equilibrium and what he obtains if he decides simply not to ask for
inspection and let underinvoicing occur. In the absence of any inspection, the expected loss is /2, while the expected gains come from
what is saved in terms of the inspection costs 2 p2 . Accordingly, we
^ over which the regulator prefers to let
can dene a threshold
underinvoicing occur, i.e. prefers to set p2 = p1 = 0. Straightforward
computations give the equation
^

1 Fs

S2

FA

:
F A 2

^ decreases with s. Equivalently, it is possible

It is easily seen that
^
to dene a threshold s 2 for the corruptibility of the customs administration. Above that threshold, optimal inspection by the PSI rm is too
costly and the regulator prefers not to inspect the merchandise at all.
^s 2

FS

F A

:
F A 2

while he chooses to set p2 = 0 in case of a strict reverse inequality. If

the left and the right hand sides in Eq. (A.2) are equal, he is indifferent
between the different inspection policies as long as Eq. (A.1) is binding. It can easily be seen that Eq. (A.2) is satised for all admissible
values of the parameters as soon as b c1, i.e. as soon as the PSI rm
is less costly than an uncorrupted customs administration. However,
it can still be satised when the reverse inequality holds, > c1, i.e.
when the PSI rm is more costly than an uncorrupted customs administration. If we want to isolate the impact of s, the corruptibility
of the customs administration, on the regulator's choice, we can
state that the regulator chooses to set p2 0, i.e. to use the PSI rm
when

FA



1 b s

FA :
c1
c1 F S2
2

This last inequality will have different consequences, depending





on the signs of c1 1 and c1 FS . Let us dene a third threshold
2

for s, namely ^s 3 by
Intuitively when the degree of corruption of the customs administration increases, inspection by the PSI rm must be more frequent in
order to prevent underinvoicing, and this is more costly for the regulator who prefers to let underinvoicing occur. Next, we check when it
is optimal for the regulator to x p1() = 0.
In other words, we check when it is optimal to use inspection by
the customs administration for the sole purpose of providing a
disciplining device for the PSI rm. When p1() is different from 0,
in a truthful equilibrium, constraint (6) must still hold and will be
binding in equilibrium. Taking this into account, the incentive constraint corresponding to Eq. (5) is written now as

p2 F A

!


s 


:
FA
1p2 p1 F A s F A
F S2
2
2

A:2

A:1

The left hand side is, as before, the benet of an importer that
declares 1 instead of the true value 2. The right hand side is the associated cost of doing so. With probability p2, S2 inspects the merchandises and discovers the lie; the importer is then ned and pays FA,
unless S1 is also asked to inspect and colludes with A to contradict
S2's report. This latter event occurs with probability p1 s (where p1 is
given by Eq. (6) binding) and leads A to pay /2 instead of FA. With
probability (1 p2)p1(), S1 carries out the only inspection of the
merchandises and A is ned unless S1 is corruptible, this case corresponds to the second term in the right hand side. In a truthful equilibrium, the objective of the regulator is to minimize the inspection costs
given by
1p2 1p2 p1 c1 ;
subject to constraint (A.1). Let us label x the variable p2 and y the variable (1 p2)p1(). It can be clearly seen that both the objective
function and the constraint are linear in x and y. Due to this linearity
of the objective and the constraint, we conclude that the solution will
be in a corner, i.e. either x = 0 or y = 0 which implies that either p2
will be set equal to zero or p1() will be set equal to zero. The regulator will either choose to use S2 and then will set p1() = 0 or will
choose not to use S2 and to rely only on S1, the customs

^s 3

FA


F A 2

c1

c1

S2

We are now ready to establish the proposition. Property (i) is an

immediate consequence of the expressions of ^s 1 and ^s 2 . To prove
(ii), observe that the right hand side of Eq. (A.2) increases with s
and that this strict inequality is veried for ^s 1 if and only if ^s 1 b^s 2 .
Therefore, we know that ^s 3 b^s 1 ^s 1 b^s 2 , which is equivalent to (ii).
Properties (j), (jj) and (jjj) are direct consequences of the analysis
we have conducted so far.
Appendix B. Proof of Proposition 4
For simplicity, we consider a situation in which it is optimal for the
regulator to use the two supervisors. Suppose now that when q~
q~ 1 1 whereas q~ 2 2, the regulator can play a mixed strategy and
nes A with probability 1 and nes S2 with probability 1 1.
To keep things comparable, we assume that in this case, the bribe
which a dishonest customs ofcer receives is (1 1)/2, i.e. (1 1)
times the bribe he received when 1 was equal to 0.
Assume also that when q~ 1, q~ 2 2 and q~ 1 , the regulator
nes S2 with probability 2 and A with probability 1 2.
In that case, the two relevant incentive constraints for A and S2 are
respectively



;
p2 1p1 1 2 F A p1 1sF A p1 s 1 F A 1 1
2
B:1
and
1p1 2 p1 1 1 F S2 :

B:2

We can rearrange the incentive constraint (B.1) to obtain:




F
p2 F A 11p1 2 p1 1 1 p1 1 1 1 s A 2 : B:3
FA

V. Dequiedt et al. / Journal of Development Economics 99 (2012) 282291

Constraints (B.2) and (B.3) are binding at the optimum. From this
we can characterize the two equilibrium probabilities p1 (1, 2) and
p2 (1, 2) that minimize the regulator's total inspection costs. From
the fact that Eq. (B.2) is binding, we deduce that p1 1 1 FS ,
2
with equality when 2 = 0. We can also rewrite Eq. (B.3) binding as
p2 F A 1


!
F

;
p1 1 1 1 s A 2
F S2
FA

B:4

from which we deduce that in order to minimize p2, the regulator

must choose 2 = 0, i.e. p1 1 1 F .
S2
Once we know that it is optimal to set 2 = 0, the choice of 1 has

no impact on the optimal p2. The only thing that changes is p1 which
is higher when 1 is different from 0. However, in equilibrium, this
has no impact on the regulator's payoff because S1 never inspects in
a truthful equilibrium. Incentive compatibility nevertheless imposes

that F S 1
1.
2

Appendix C. Proof of Proposition 5

With this information structure and for sufciently small, the incentives are not changed. Indeed, for the agent, the relevant incentive
constraint is still the one that states that an agent with q = 2 should
not prefer to declare 1. In this case, the inspection technology is actually perfect and the constraint is exactly the same as before: it is
Eq. (B.1).
Concerning the second supervisor S2, when he observes 1, he
knows for sure that the true value of q is 1, and he anticipates that
if the customs administration inspects the merchandises, it will discover that the true value is 1 with probability 1. Therefore, the incentive constraint of a supervisor S2 that observes 1 and considers the
possibility of reporting 2 is again the same as before: it is Eq. (B.2).
For the second supervisor, the imperfection in the inspection technology could have an impact on the other incentive constraint: the one
that states that when S2 observes 2, it prefers to report 2 rather
than 1. However, this constraint was not binding in equilibrium
when = 0, i.e. with a perfect inspection technology, and will not
be binding either for sufciently small, as a simple continuity argument shows.
The only difference is that, with an imperfect inspection technology, inspection by S1 occurs with a positive probability on the equilibrium path. On the equilibrium path, A reports 1 with probability 1 ,
and this is followed by a truthful report 2 by S2 with probability . Inspection by S1 therefore occurs with the ex ante probability p1/2. In
this case, the regulator strictly prefers a reconciliation policy that
minimizes p1 all other things being equal.

291

References
Anson, J., Cadot, O., Olarreaga, M., 2006. Tariff evasion and customs corruption: does
preshipment inspection help? Contributions to Economic Analysis and Policy 5
(1), 16001623.
Bac, M., Bag, P.K., 2006. Benecial collusion in corruption control: the case of nonmonetary penalties. Journal of Development Economics 81 (2), 478499.
Besley, T., McLaren, J., 1993. Taxes and bribery: the role of wage incentives. The Economic Journal 103, 119141.
Bhagwati, J., 1964. On the underinvoicing of imports. Oxford Bulletin of Economics and
Statistics 26 (4), 389397.
Bolton, P., Dewatripont, M., 2005. Contract theory. MIT Press, Cambridge.
de Wulf, L., 1981. Statistical analysis of under- and overinvoicing of imports. Journal of
Development Economics 8 (3), 303323.
de Wulf, L., Sokol, J., 2004. Customs modernization handbook. World Bank Publications,
Washington D.C.
Fisman, R., Wei, S.-J., 2004. Tax rates and tax evasion: evidence from missing imports in
China. Journal of Political Economy 112 (2), 471500.
Hoekman, B., Nicita, A., 2008. Trade policy, trade costs and developing country trade.
Policy Research Working Paper. The World Bank.
Johnson, N., 2001. Committing to civil service reform: the performance of preshipment
inspection under different insitutional regimes. Policy Research Working Paper.
The World Bank.
Keen, M., 2003. Changing Customs: Challenges and Strategies for the Reform of Customers Administration. International Monetary Fund, Washington D.C.
Khalil, F., 1997. Auditing without commitment. The RAND Journal of Economics 28 (4),
629640.
Kofman, F., Lawarre, J., 1993. Collusion in hierarchical agency. Econometrica 61 (3),
629656.
Kofman, F., Lawarre, J., 1996a. A prisoners' dilemma model of collusion deterrence.
Journal of Public Economics 59, 117136.
Kofman, F., Lawarre, J., 1996b. On the optimality of allowing collusion. Journal of Public Economics 61, 383407.
Laffont, J.-J., Martimort, D., 2001. The Theory of Incentives: the Principal-agent Model.
Princeton University Press.
Low, P., 1995. Preshipment inspection services. World Bank Discussion Paper.
Mishra, A., Anant, T.C.A., 2006. Activism, separation of power and development. Journal
of Development Economics 81 (2), 457477.
Njinkeu, D., Wilson, J., Powo Fossa, B., 2008. Expanding trade within Africa. Policy Research Working Paper. The World Bank.
Pritchett, L., Sethi, G., 1994. Tariff rates, tariff revenue, and tariff reform: some new
facts. World Bank Economic Review 8 (1), 116.
Strausz, R., 1997. Delegation of monitoring in a principal-agent relationship. Review of
Economic Studies 64 (3), 337357.
Tirole, J., 1986. Hierarchies and bureaucracies: on the role of collusion in organizations.
Journal of Law Economics and Organization 2 (2), 181214.
Tirole, J., 1992. Collusion and the theory of organizations. In: Laffont, J.-J. (Ed.), Advances in Economic Theory, Sixth World Congress. Elsevier.
Wilson, J., Mann, C.L., Otsuki, T., 2005. Assessing the benets of trade facilitation: a
global perspective. The World Economy 28 (6), 841871.
Yang, D., 2008a. Can enforcement backre? Crime displacement in the context of
customs reform in the Philippines. The Review of Economics and Statistics 90
(1), 114.
Yang, D., 2008b. Integrity for hire: an analysis of a widespread customs reform. Journal
of Law and Economics 51 (1), 2557.