WCO Research Paper No.

27

What Happened to the Selected?

Performance Measurement of the
Korea Customs Service Selectivity
System

(December 2012)

Chang-Ryung Han and Robert Ireland

Abstract

Customs uses risk management as the core philosophy to balance regulatory control
and trade facilitation. To deal with emerging and evolving risks, Customs should
regularly adjust their risk management techniques and develop new approaches when
necessary. Improving such efforts can benefit by applying rigorous risk management
practices from other disciplines and measuring the results empirically. This paper
examines the performance of the Korea Customs Service (KCS) selectivity system,
drawing on practices used in the fields of taxation and insurance that deal with similar
kinds of risk or fraud. This paper focuses on the relationship between selection and
detection rather than scrutinizing selection and detection rates independently in order to
alleviate concerns about smugglers possibly exploiting revealed selectivity performance
data. The KCS currently uses three selection methods: manual selection, rule-based
selection, and random selection. During the study, manual selection surprisingly
showed the highest detection rate and the best efficiency. The rule-based selection
appears to play an important role in deterring fraud techniques and opens the way for
Customs officers to use their experience, knowledge and skills to adopt manual selection
to keep up with smugglers' evolving approaches. Random selection also appears to
have a deterrent effect because its unpredictability overcomes smugglers evolving
predictions on selection patterns. The paper analyzes these results and concludes that
the three selection methods are complementary for the detection and deterrence of
emerging and evolving risks.

Key words

Risk Management, Rule-based Selection, Manual Selection, Random Selection, and

Korea Customs Service (KCS)

Acknowledgements

This paper was written by Chang-Ryung Han and Robert Ireland of the WCO’s Research
and Strategies Unit in the Office of the Secretary General. The authors are grateful to
KCS staff for providing the selectivity performance data and other information in
interviews. The authors also express their appreciation to WCO officials Thomas
Cantens, Min Han, Jong-Ho Kim, Mariya Polner, and Tadashi Yasui for their comments.

Disclaimer

The WCO Research Paper Series disseminates the findings of work in progress to
encourage the exchange of ideas about Customs issues. The views and opinions
presented in this paper are those of the authors and do not necessarily reflect the views
or policies of the WCO or WCO Members.

Note

All WCO Research Papers are available on the WCO public website: [Link].
The authors may be contacted via research@[Link].
-----------------------
Copyright © 2012 World Customs Organization.
All rights reserved. Requests and enquiries concerning translation, reproduction and adaptation rights should be
addressed to copyright@[Link].

2
1. Introduction

During the past 60 years, the volume of international trade has increased
exponentially. Resources that Customs can mobilize to handle the increasing trade
have been, however, limited and their reinforcement have not caught up with the pace of
international trade growth (Keen, 2003). The expansion of international trade and
pressure from the international trading community to minimize government intervention
increased Customs’ interest in trade facilitation. To harmonize regulatory control and
trade facilitation 1 , many Customs have adopted risk management as their dominant
philosophy and this was embedded in the WCO’s Revised Kyoto Convention (Harrison
and Holloway, 2007). The extensive use of risk management in Customs administration
was reflected in a risk management survey conducted by the WCO in 2010 where 84
percent of respondents answered they have adopted risk management in their
administration (WCO, 2010).

For Customs, the most prominent risks are non-compliant or illegitimate trade
transactions undermining government revenue and dangerous goods threatening society,
although risks that Customs should address are all events and activities that hinder
Customs from achieving their objectives (WCO, 2011; Widdowson, 2005). Risk
management to Customs is a process to mitigate and avoid risks 2 hampering
achievement of Customs objectives. The introduction of risk management in Customs
administrations has brought a change in Customs’ approach to risk. While putting risk
management in place, many Customs have kept away from their traditional stringent
gatekeeper role at borders. They have lessened their regulatory control-oriented
administration and minimized unnecessary disruption and cost to legitimate trade and
passengers (Widdowson, 2007). Customs have re-engineered their strategies and
tactics against risk so to concentrate their resources on the high-risk end of the risk
continuum (WCO, 2011) and have launched various forms of risk management
programs fitting their environments, such as cargo selectivity systems, post-clearance
audit programs, and Authorized Economic Operator programs (Closs and McGarrell,
2004). Yet, whereas there have been many upper-level discourses advocating the
adoption of risk management in Customs administration, there have been few studies to
present substantial methodologies necessary to operate the introduced risk
management programs (Hinsta et al, 2011). The lack of research to examine the
feasibility, effectiveness, and efficiency of risk management programs (Laporte, 2011)
has hampered improving the existing programs and devising new ones. The problem
may be influenced by many factors, such as political and administrative culture, fiscal
situation, and information and telecommunication infrastructure where Customs
undertake their daily tasks.

From a research perspective, however, two reasons seem to be prominent. First,

few theories have been formulated so as to assist the embedding of risk management in
Customs administrations. Theoretical frameworks are important in operating risk
management programs because theories play a role in seeking and providing logically

1
The harmonization of the two distinct objectives does not just take into account trade facilitation
more and regulatory control less than before and achieve an appropriate balance between them
but also prevent the influx of high risk cargo without interfering with trade facilitation.
2
According to Deloach (2000), there are four ways in dealing with risk: avoid, reduce, transfer,
and retain or accept.

3
plausible explanations necessary for the operation of risk management programs and
helping to predict their outcomes based on the formulated mechanisms (Babbie, 2007).
Second, limited data is available for research on Customs risk management. Customs
are reluctant to share their data with researchers because they fear traders are likely to
exploit such statistics to avoid examination and inspection3. Thus, risk management
techniques have not been much vetted theoretically and empirically, despite their
importance to all Customs administrations.

This paper focuses on Customs selectivity systems, which is generally regarded

as a type of Customs risk management technique. In particular, this study analyzes the
effectiveness and efficiency of selection methods that the selectivity systems employ
and derives important factors in operating the selectivity systems, with data supplied by
the KCS and drawing on theories developed in other domains.

2. Theory

Several international organizations have produced Customs risk management

guides. The WCO has developed practical guides including on Customs risk indicators
(WCO, 2011). The Organization for Economic Cooperation and Development (OECD)
and the World Bank, have released publications advocating Customs’ adoption of risk
management to facilitate trade (De Wolf and Sokol, 2004; McLinden et al, 2010; OECD,
2003). There is a gap, however, between the strategic justification and operational
guides for risk management. To embed risk management in Customs administration,
intermediates filling the chasm are necessary. The role is fulfilled by theories for risk
management in the sense that theories sort out plausible factors influencing risk
management programs and provide logical explanations to improve the programs. Yet,
few theories that play such an expected role have been known in the field of Customs
administration.

Unlike Customs administration, the fields of taxation and insurance, which have
similar mechanisms to that of Customs administrations in terms of fraud control, have
analyzed how to detect and deter tax and insurance fraud (Derrig, 2002; Viaene and
Dedene, 2004; Andreoni et al, 1998). Tax and insurance fraud control theories have
focused on three categories: (1) why fraud is committed; (2) why selection is needed;
and (3) how selection should be made.

First, analysis of the causes of tax and insurance fraud occurrence has focused
on fraud opportunities rather than fraudsters’ individual characteristics. Viaene and his
colleagues (2004) contend that, like other kinds of crimes, fraud is also committed by a
combination of motivated offenders and criminogenic opportunities, drawing on Felson’s
routine activity theory 4 . They suggest that it is important to remove fraud-ridden

3
In this paper, whereas the term examination refers to probing the veracity of information on
documents, such as manifests, declarations, invoice, bill of lading, and packing lists, the term
inspection refers to investigation of the physical status of cargo. Most cases concerning valuation
issues need examination of documents without physical inspection of cargo. Cases concerning
classification of products, however, require not only examination of documents but also physical
inspection of products.
4
According to Felson (2009), crime occurs when a motivated officer meets a suitable target that
is not protected by a guardian. His routine activity theory emphasizes the convergence of a
motivated officer, a suitable target, and absence of guidance.

4
opportunities to prevent occurrence of fraud. Game theorists analyze fraud with the
principal-agent model, focusing on how prospective offenders have motives for fraud
and exploit fraud-ridden opportunities. According to the principal-agent model, whereas
tax payers and insurance claimants (agent) have full information about their status of
income, health, and accidents, tax agencies and insurance companies (principal) do not
know about the agent as much as the agent. The information asymmetry leaves the
principal with no option but to trust the agent; which provides the agent with an incentive
to defraud tax agencies and insurance companies. Thus, tax agencies and insurance
companies have devised various methods to narrow the information gap between the
principal and the agent (Tennyson and Salsas-Forn, 2002; Viaene and Dedene, 2004).
Some scholars focusing on tax compliance have tried to explain tax non-compliance with
variables regarding tax policies and administration, such as, tax rates, tax payer
perception of fairness in taxation, tax audit methods, and non-compliance penalties (Alm,
1999; Andreoni et al, 1998; Kirchler et al, 2007).

Second, tax and insurance fraud control theorists are interested in case selection
mechanisms in the sense that selection is a prerequisite for detection. It is important to
establish a system that a crime committed is known to the authorities for crime
prevention and control. The detection process of tax and insurance fraud is somewhat
different from that of ordinary commercial frauds. Whereas detection of ordinary
commercial frauds is usually triggered by victim reports to the authorities, detection of
tax and insurance fraud is barely gleaned by individual clients’ reports. In other words,
whereas the victims of ordinary commercial frauds take their cases to the authorities,
such as criminal justice systems and civil courts, to try to restitute their loss, tax
authorities and insurance companies play a role in enforcing laws and contracts for their
clients, such as tax payers and insured person, rather than being entities protected by
laws and contracts. However, the fraud victims are tax agencies and insurance
companies themselves. They usually do not recognize their own victimization until they
proactively investigate fraud to discover it (Piquet et al, 2007). Thus, tax and insurance
fraud control theories focus on the selection of incoming tax returns and insurance
claims as a process of detecting their victimization (Graetz et al, 1986). Even though tax
agencies and insurance companies undertake a screening process of incoming tax
returns and insurance claims to identify frauds, they cannot do an in-depth examination
of all incoming tax returns and insurance claims. Tax payers and insurance claimants,
meanwhile, want their reports to be processed and confirmed quickly. Fraud control
entails costs not only to taxpayers and claimants but also to tax agencies and insurance
companies. Thus, tax agencies and insurance companies should concentrate their
limited resources on a small number of high risk targets. This situation leads tax
agencies and insurance companies to select a small number of tax returns and claims to
verify whether their reports are accurate. For this, tax agencies have formulated tax
audit programs, which select the most suspicious tax turns as subjects of tax audit (Alm,
1999; Andreoni et al, 1998). Insurance companies have established automated fraud
detection systems, which sort out incoming claims into two bins where (1) a group of
non-suspicious routine claims is put through a simplified process and claims are paid
quickly and (2) the group of suspicious claims is processed through intensive verification,
such as special identification units’ investigation and external audit (Derrig, 2002; Graetz
et al, 1986; Viaene et al, 2007). Recent automated detection systems evaluate the
gravity of incoming claims with lists of fraud indicators or flags representing fraud
detection expertise.

5
 Third, tax and insurance fraud control theories have examined selection
methodologies in light of effectiveness and efficiency. Some assert that selection should
focus on minimizing total costs, including fraud control cost and losses incurred by fraud.
Fraud control is conducted to prevent financial and reputational damage from fraudulent
claims. Fraud control per se, however, incurs costs to tax agencies and insurance
companies as well. Thus, they contend that fraud control should be done to the extent
that the expected marginal costs of fraud control are equal to its marginal savings from
the control (Tennyson and Salsas-Forn, 2002).

Other theorists assert that suspicious tax returns and insurance claims should be
addressed from the perspective of deterrence. In fraud control, it is important to detect
fraudulent tax returns and insurance claims, but the ultimate goal of fraud control is to
prevent recidivism and deter new fraudsters. According to deterrence theory, the
deterrent effect5 increases in three situations: the severity of the punishment increases,
the probability of being punished increases, or the temporal gap between crime
commitment and punishment decreases 6 (Becker, 1968; Nagin, 1998). In this
connection, in the field of insurance, scholars recommend that insurance companies
carry out intensive audits frequently to raise the rate of fraudulent claims detection
(Picard, 1996). Fraud control theorists contend that fraudsters adjust techniques so as
to exploit the inertia of complex systems and evade fraud control. Because fraudsters
adapt their tactics, fraud control should be unpredictable. Thus, these scholars support
the use of random selection for intensive audits (Tennyson and Salsas-Forn, 2002).

Customs also have concerns about fraud control similar to tax agencies and
insurance companies. Like tax agencies and insurance companies, Customs (the
principal) are disadvantaged to traders (the agent) in terms of quantity and quality of
information about cargo, traders, and consignment payment terms. This is because
Customs are not involved in the trade transactions and are unaware of the specific
agreements between the exporter and the importer. Customs usually simply accept the
information that traders submit as it is for speedy clearance, so long as information on
declarations matches that of trade documents and is not explicitly suspected of
document manipulation.

Even though Customs can close in on the substance of a trade transaction by

examining invoices produced by the exporter and the importer, if the invoices are
manipulated, either individually or by collusion between the exporter and the importer,
then Customs get deceived. Customs can uncover the veracity of a trade transaction by
in-depth investigation of trade documents (e.g., invoice, packing list, and Bill of Lading)
and historic databases of traders, gathering intelligence from informants, and physical
cargo inspection. Traders are aware, however, that Customs cannot examine all
declarations and cargo because the cost not only impacts traders (e.g., inspection fee,
container devanning/revanning cost, and delaying cost) but also Customs (e.g.,

5
Many empirical studies suggest double-barreled findings on deterrence theory; whereas macro-
level deterrence studies show that programs based on deterrence perspective have a weak
influence on crime rates, the perceptual deterrence literature shows that deterrence measures
working on perceptions of deterrence are likely to have some effects on crime (Nagin, 2011; Pratt
et al 2008).
6
This proposition suggests that the shorter the interval between crime commitment and
punishment, the greater the deterrent effect of the punishment. For instance, immediate
punishment is likely to be more deterrent than delayed punishment.

6
inspection facility operating costs and inspection officers’ travel expenses). The
information asymmetry between Customs and traders provides a motive for traders to
engage in illegal trade, including Customs fraud.

In this connection, Customs have endeavored to gather more comprehensive

data about traders and their trade transactions as early as possible in the supply chain in
order to reduce the information gap between traders and themselves. To have the
obtained information used immediately for examination of trade transactions and
inspection of cargo, many Customs have established automated selectivity systems,
which screen incoming reports (e.g., manifests and declarations) by freight forwarders
and traders and decide which reports are sufficiently low-risk to be cleared without any
intervention and which ones are suspicious and merit intensive examination. However,
some Customs administrations have not clarified what purposes their selectivity systems
are designed to achieve, which method is most efficient, and which is best fit for their
trade environment and administrative culture. This study, drawing on studies from
taxation and insurance, describes why and how Customs can strengthen the
establishment and operation of their selectivity systems.

3. Data

It is essential to establish a theoretical foundation to run a Customs selectivity

system effectively and efficiently. It is difficult, however, for external researchers to
acquire data concerning the performance of selectivity systems for theory formation and
testing. Many Customs have low examination and inspection rates, taking into account
trade facilitation objectives and limited resources for regulatory control (Closs and
McGarrell, 2004; Geourjon et al, 2012; Harrison and Holloway, 2007). Some Customs
fear that if their risk management performance data and methodologies are open to
outside stakeholders, then prospective non-compliant traders can exploit the information.

The concern is based on the assumption that traders are unaware of the
selection and detection rates of Customs selectivity systems. Traders are aware,
however, of the limitations of Customs controls. Some traders know how many of their
declarations and cargo are examined and inspected. What traders do not know is the
overall selection and detection rates. Hence, what Customs should be concerned about
is not the likelihood that criminal traders evade the selectivity systems or exploit
publicized data but whether they can improve selectivity systems through constant
evaluation. Nonetheless, there is a need to accept that selection and detection rates
can be generally understood by experience. Furthermore, given the likelihood that once
risk management performance statistics are made public, even reliable traders are
attracted to circumvent Customs selectivity systems, it is necessary to reach a
compromise between Customs’ concern about the publicity of the statistics and the
necessity of research on Customs selectivity systems. The resolution can begin with the
questions of what information Customs are reluctant to divulge and what information is
truly needed to study selectivity systems. The information that Customs does not want
to share are the selection and detection rates of their selectivity systems. Studies
should, however, focus on the relationship between selection rates and detection rates.
Hence, the authors of this study asked the KCS to provide its selectivity system
performance statistics and obtained permission to use its statistics under the condition
that selection and detection rates are not specified in the final research product. To
have heighted understanding on the KCS’s selectivity systems and their performance

7
data, the authors interviewed three key officers are in charge of operating the selectivity
systems.

Selection

All declarations and reports to the KCS are submitted electronically and go
through a process of decision-making of whether they are legitimate enough to be
accepted. The decisions are made by Customs officers who screen incoming
declarations and reports that are assigned to them. Customs officers proactively accept
non-suspicious declarations and reports. However, if Customs officers do not make any
decision on the assigned declarations and reports within a certain period of time, the
KCS’s automated clearance system 7 accepts declarations and reports as they are in
order to prevent the non-decision making from hampering trade facilitation. Selection is
a kind of decision-making to withhold acceptance of declarations and reports. The KCS
conducts regular selections8 in seven channels for examination and inspection: manifest
reports 9 , import declarations, export declarations, import express parcels, incoming
passengers’ luggage, cleared declarations (for post-clearance audits) and drawback
claims. All selection processes and results reporting are managed by the selectivity
systems of each channel. In each selectivity system, sample selections are categorized
into two methods: (1) computerized selection and (2) Customs officers’ manual selection.

The computerized selection comprises two types: rule-base and randomness.

The rules, or selection criteria, of the rule-based selection draw from several sources.
The dominant criterion is based on successful detections made by manual selection.
Data-mining on the accumulated declarations extracts patterns (Geourjon et al, 2012;
Laporte, 2011) of legitimate trade transactions and builds the computer selection criteria
that sort out irregular declarations that do not meet the patterns. Some criteria incarnate
trade regulations of the government and automatically screen out declarations that do
not meet the regulations. The rule-based selection criteria that do not detect anything
during a certain period of time (e.g., six months) after being mounted on a selectivity
system are destined to be automatically dismounted from the selection system. The
random selection is to select declarations whereby all declarations have an equal
chance of being selected, not just unplanned or unconscious selection. Manual
selection is conducted by Customs officers based on their own experience and
knowledge, not with the use of computerization or by pre-programmed criteria. The
process for the manual selection is, however, supported by the selectivity systems that
record which officer selected which declaration and what offense was charged for the
selected declaration.

7
The KCS has an automated clearance system named Uni-Pass; which processes a variety of
data from freight forwarders, carriers, shippers, bonded warehouses, traders, Customs brokers,
and trade-related governmental agencies. The processed data are shared with its selectivity
system which has been developed along with each part of the automated clearance system and
whose origin dates back to the early 1990s.
8
The KCS does not necessarily rely on regular selections for inspection on cargo and audit on
traders. When the KCS has solid intelligence about certain cargo and transactions, it undertakes
inspections and audits without the assistance of the selectivity systems.
9
The KCS selects manifests to inspect containers in which high risk cargoes are likely to be
included in order to detect smuggling committed without import declarations.

8
Figure 1: Composition of selectivity systems of the KCS

Source: the authors

KCS management monitors and manages the overall selection rate based on its
strategic risk assessment. For instance, when the Korean government began to
prioritize trade facilitation, the KCS lowered the selection rate. The KCS has also
adjusted the selection rate as a result of health risks of some imported goods. Its
working-level officers allot the overall selection rate to each selectivity systems as their
target selection rates, taking into account their selection speeds, detection rates, and
responses from traders, and fine tune the ratios among the selection rates.

Table 1: Composition ratios (%) of the selection methods in the import selectivity
system
Year 2009 2010 2011 2012 (Q1 to Q3)
Rule-based selection 80 84 79 79
Random selection 7 4 6 7
Manual selection 13 12 15 13
Total 100 100 100 100

In the import selectivity system that plays the most import role in the KCS’s
selectivity systems, the rule-based selection accounts for about 80 percent of KCS’s
overall selection rate. Originally the rule-based selection was made by risk indicators,
which suspicious declarations and cargoes have in common. For example, if a certain
kind of products manufactured by company A in country B had been detected as
infringing on intellectual property rights, the same kind of products of company A are
registered to the import selectivity system for inspection10. The rule-based selection,

10
If the trader who imports a kind of products manufactured by company A is charged with an
IPR infringement, the importer usually replaces company A with another suppler to avoid his or
her cargo being selected again. If the infringement on IPRs was a mistake and unusual, the

9
together with an increase in the collected data quantity and advancement of database
technology, has evolved to the current advanced form in which Customs officers
establish not only verification criteria (If-Then Select) but also falsification ones (Unless-
Then Select). For example, when a certain kind of goods are produced only in Country
B that has concluded a Free Trade Agreement with Korea, if declarations to import the
same kind of goods from countries other than Country B are submitted to Customs, then
the declarations are suspected to evade Customs duty and should be selected.

The manual selection accounts for about 10 percent of the overall selection rate.
According to the interviewed officers, even though KCS management has wanted to
have a little higher rate in the manual selection, the number has not increased. The
officers attribute its low selection rate to its labor intensive methodology that the manual
selection is not simply based on their instant intuition but made through risk analysis on
declarations with their own expertise and knowledge, which takes 3 to 4 hours per
declaration. They cannot conduct many analyses because of clearance time
requirements11. In addition, cargo inspection entails costs not only to traders but also to
Customs. Customs officers tend to be circumspect in manual selection and concentrate
on a small number of apparently problematic declarations in order to avoid blame for
“selection-but-no-detection”. Customs officers’ manual intervention is not confined to
selection. Even if the programmed selections winnow out some declarations or cargo for
examination and inspection, Customs officers can override the selections when they are
confident the selected declarations or cargo are not suspicious.

Among the seven KCS selectivity systems, this study examines the import
selectivity system. The seven selectivity systems are respectively operated in the three
divisions of the KCS. The two divisions that maintain selection data regarding manifest
reports, passengers, and express parcels declined to participate in this research project.
This study is provided with the 15 quarterly selection data from 2009 to the third quarter
of 2012 concerning import declarations, export declarations, post-clearance audits, and
drawback claims, but decided to focus on the import selectivity system. This decision
was based on the higher volume generated by the import selectivity system.

Detection

Selection is intended to detect offenses. There are many types of Customs

offenses and they vary country to country. Customs duty and tax evasion is the primary
offense that Customs targets. Other offenses relate to illegal drugs, counterfeit goods,
and endangered fauna and flora. Customs is also concerned with manipulation of origin
of goods and defects in import/export requirements. The KCS pays special attention to
these problems as core targets that need criminal penalties or administrative sanctions.
Errors in the quantity and quality of goods that are not related to Customs duty and taxes,
however, are regarded as minor problems for which the KCS gives traders demerits of

importer needs to endure inspections on subsequent consignments until the KCS gets sure that
Company A’s export cargo is no longer problematic and dismount the selection criterion from its
selectivity system.
11
The KCS’s automated clearance system measures the elapsed time of each declaration and
evaluates clearance performance of each Customs house with its averaged elapsed time monthly,
quarterly, and annually.

10
being selected for comprehensive compliance audits12. Even though the errors harm the
accuracy of trade statistics, the KCS refrains from sanctions on traders and goods
because enforcement against minor violations would hamper trade facilitation. In this
respect, in this study, such minor problems that do not need criminal penalties or
administrative sanctions are excluded13.

4. Analysis

Hypotheses

The study has two key objectives. First, it examines which selection method
among the three selection methods is the most efficient. There are two lines of thoughts
on selection methods. Many field officers, who are inclined to give more credit to their
intuition and accumulated experience, claim that manual selection is more effective and
efficient than rule-based selection 14 . Some management-level officers who seek
predictable and reliable Customs practices prefer the rule base selection method,
believing it is more objective approach because it is mainly based on historical data.
Moreover, there is the view that it prevents officer from being involved in corruption. As
shown in the table 1 presenting the lowest portion of the random selection in the overall
selection rate, field officers and management-level officers regard the random selection
as a supplement instrument for the other selection methods rather than an independent
detection tool. To test the hypotheses, this study seeks a way of comparing detection
rates of each selection method employed in the import selectivity system.

Second, this study addresses the question of how a change in selection rates
affects detection rates. Customs is destined to detect harmful foreign products and
illegitimate trade transactions. On the other hand, Customs is regularly pressured to
facilitate trade. Customs that emphasizes regulatory control more than trade facilitation
may increase selection rates. However, traders may be dubious of whether an increase
in selection rates leads to an increase in detection rates. This study examines the
relationship between selection rates and detection rates with simple bivariate correlation
analysis.

12
The KCS has a comprehensive compliance audit program as a regular check-up of traders.
For comprehensive compliance audits, officers visit traders’ main offices and examine their
businesses related to Customs administration all together, such as import declarations, export
declarations, drawback claims, management of origins, and foreign exchange transactions. To
select subjects for the comprehensive compliance audit program, the KCS maintains a trader
profile database combining all examinations, inspections, and investigation results and minor
errors.
13
In the KCS, the detection rates influence field officers’ bonus, transfer to better positions, and
promotion. The officers had paid excessive attention to minor errors to get higher detection rates
until the KCS established a policy of zeroing in on core targets in evaluation of performance of
Customs houses and officers. This practice had brought about unnecessary tension with traders
and had been thought of as a factor against trade facilitation. The KCS decided to consider
detection rates concerning core targets in performance evaluation, even though it maintains
records of minor violations.
14
According to the officers interviewed with the author, even though Customs field officers
believe that manual selection is the most effective and efficient method, they do not think that
most selections should be made by the manual selection.

11
Effectiveness of selection

Selection does not guarantee detection. Customs endeavors to improve

selectivity to raise detection rates. This study employs a relative risk formulation to
examine how more strongly a selection method is associated with detection than another
selection method and determine which selection method is most effective. The relative
risk15 refers to the ratio of probabilities in which an event occurs in two different groups
(Agresti and Finlay, 1999; Miller, 2005). For instance, the relative risk of the manual
selection to the random selection is calculated by dividing the detection rate of the
manual selection by the detection rate of the random selection. The relative risks do not
show how well each selection method targets but can indicate which selection method is
more accurate.

Table 2 shows that the detection probability by the manual selection (MN) is
about six to eight times higher than that of the random selection (RD) (MN > RD). The
rule-based selection (RB)’s detection probability is about two times higher that of the
random selection (RB > RD). The manual selection is about three times better than the
rule-based selection in detection probability (MN > RB). According to the relative risks
the manual selection is more effective than the other methods (MN > RB > RD). The
random selection has the worst detection performance16.

Table 2: Relative risks for the selection methods of the import selectivity system
2009 2010 2011 2012 (Q1 to Q3)
Rule-based selection to Random selection 1.7 2.1 2.1 2.4
Manual selection to Random selection 5.8 8.0 7.4 9.5
Manual selection to Rule-based selection 3.3 3.8 3.5 3.9

The difference in detection performance between the manual selection and the
rule-based selection is intriguing because it would imply that officers’ discretion shows
better performance than risk-based selection drawing from historical data. In particular,
some scholars criticize that manual selection, which is dependent upon human judgment,
is static and rigid and is vulnerable to fraudsters’ ever-changing techniques, compared to
statistical scoring techniques (Geourjon et al, 2012; Laporte, 2011). However, there are
a number of reasons for why this may be the case. First, the rule-based selection may
not be as swift as the manual selection in keeping up with evolving fraud techniques. As
criminologists have said about crime displacement 17 (Welsh and Farrington, 1999),
criminals change their techniques when they think the government is successfully

15
The term risk in the relative risk, which is usually used in epidemiology, would be better to be
interpreted as a probability of the occurrence of an event (selection) in this study.
16
Some Customs officers present an opinion that the primary objective of the random selection is
not to detect illegitimate transactions but to capture the prevalence of illegitimate cargoes or
transactions. Whereas the detection rates from the rule-based selection and the manual
selection cannot be used to infer the prevalence of illegitimate cargoes and transaction in all
declarations, the detection rate by the random selection shows how many illegitimate cargoes
and transactions are likely to be in all declarations.
17
There are five different types of crime displacement: temporal (change in time), tactical (change
in method), target (change in victim), territorial (change in place) and functional (change in type of
crime). They have concerned that crime prevention and control measures can be compromised
by these displacements and have sought to solutions to secure intended outcomes (Welsh and
Farrington, 1999).

12
targeting those techniques. However, for Customs officers, it takes time to capture
criminal traders’ adjustments. Moreover, it even takes more time to implement
responses to the changed techniques because they are made through a process of
extracting effective manual selections. Hence, when new selection criteria of the rule-
based selection are applied to declarations, criminal traders may have already changed
methods again. Second, the selection criteria of the rule-based selection include not
only the extractum of the effective manual selections but also trade regulations 18, which
broadly indicate kinds of goods requiring special handlings rather than pinpoint suspects.
Such requirements cannot be adjusted quickly to chase ever-changing fraud techniques.
Thus, if low effective criteria based on government’s trade regulations were dismounted
from the selectivity system, the rule-based selection would have better detection rate.
The selection criteria embodying trade regulations cannot be removed from a pool of
selection criteria (the reason is presented later). The factors seem to have influenced
that the criteria tend to become deadwood to skillful fraudsters in detection.

Efficiency of selection

Detection rates are a good indicator to evaluate the performance of the selectivity
systems. However, detection rates per se do not provide a clue regarding costs incurred
by selections. The costs involved in selection are an important concern to both Customs
and traders. When there are selection methods with the same level of detection rates, if
a selection method entails a lower level of costs than the others, it is natural that the
selection is preferred by Customs and traders. In other words, detection rates or relative
risks may not be a sufficient indicator to evaluate selection methods. Hence, this study
introduces selection efficiency, which takes into account the effectiveness of detection
and the costs derived by selection simultaneously. This can be expressed in a ratio of
the detection rate to the selection rate. This is intended to encompass both detection
and selection rates in examining the performance of selection methods on the
assumption that an increase in the number of selections leads to a proportional increase
in clearance costs to traders and Customs, provided the cost per selection is the same
across the selection methods.

A hypothetical example shown in Table 3 illustrates how different Method A and

Method B that have the same detection rate are in their selection efficiency. Suppose
that Method A and Method B have the same detection rate of 5 percent; yet, their
selection rates differ. Method A selected fewer declarations than Method B and
obtained the same level of detection rate. This means that Method A is more efficient
than Method B.

Table 3: Selection efficiency of two different methods with the same detection rate
Selection Detection
Population Selection efficiency
case rate (%) case rate (%)
Method A 4,000 100 2.5 5 5 2
Method B 4,000 200 5 10 5 1

18
The KCS also makes selection criteria, consulting trade regulations regarding food, sanitation,
and health and receiving other ministries and agencies’ requests.

13
 When comparison between the three selection methods is made, Table 4 19
shows that the manual selection is the most efficient, and is followed by the random
selection and the rule-based selection. This finding sounds somewhat counterintuitive.
It is predicted that the manual selection based on Customs officers’ own experience,
intelligence and intuition produces greater detection with a small number of selection.
The finding that the random selection is more efficient than the rule-based selection,
however, is surprising.

There are a number of possible reasons for the finding. First, as Table 1 shows,
the KCS emphasizes the rule-based selection in operating the selectivity system, even
though the rule-based selection’s effectiveness is inferior to the manual selection. The
KCS’s course of having the rule-based selection method the highest selection rate,
regardless of its low detection, has contributed to the low efficiency of the rule-based
selection. Second, the random selection’s selection rate (denominator) is the lowest but
its detection rate (numerator) is not much lower than that of the rule-based selection. In
other words, the random selection is structured to select less and get a detection rate
corresponding to the population parameter (i.e., illegitimate trade rate). In particular, the
KCS uses random selection for inferring the prevalence of illegitimate cargo in
declarations from the detected samples and capturing illegitimate cargo that the manual
selection and the rule-based selection fail to detect. Unlike the other two selection
methods, random selection does not have predictable selection patterns. The
unpredictability of the random selection may counteract criminal traders’ adjustment of
their fraud techniques to predictable selections and to contribute to better efficiency of
the random selection.

A caveat in interpreting the finding is that the rule-based selection is not as

efficient as the other two methods do not necessarily mean that rule-based selection is
not useful. The value of the rule-based selection should be sought from the
perspectives of not only detection but also deterrence20. The essence of deterrence
theory is to sanction rule-breakers and prevent them and even others from engaging in
crime and deviance again (Nagin, 2011).

Table 4: Selection efficiency of the selection methods of the import selectivity system
2009 2010 2011 2012 (Q1 to Q3)
Rule-base selection 3.1 3.6 5.3 5.0
Random selection 20.9 33.9 32.1 22.6
Manual selection 62.1 96.5 100.3 118.4

In this respect, the rule-based selection may play a role in deterrence in two
ways. First, important trade-related regulations are factored into the selectivity system in

19
The selection efficiency (= a detection rate / a selection rate) does not say a probability, or
percentage. This measure can provide information of which method is superior compared to the
others. This does not say how more efficient method A is than method B. For instance, in
interpreting Table 4, it should not be said that the manual selection was three times as efficient as
the manual selection in 2009.
20
The three selection methods all may have deterrent effect. The manual selection’s deterrent
effect is quite collateral. In other words, whereas the manual selection is more punishment-
oriented deterrence than the others, the rule-based selection is more prevention-oriented than the
manual selection.

14
the form of the selection criteria of the rule-based selection. If a trader violates a
regulation that is enrolled in as a selection criterion in the selectivity system, their
declaration or cargo is selected for examination or inspection and they are less likely to
engage in such an offense again. Second, the rule-based selection may reduce
criminogenic opportunities by enrolling the verified selection criteria in their detection
effectiveness. Even though the rule-based selection’s primary objective is to tackle
illegitimate trade, it is also meaningful to prevent criminal traders from employing the
same techniques and narrow down the number of techniques that criminal traders can
rely on.

Correlation between selection and detection

When the general public has a heightened concern about illegitimate trade,
especially with respect to health and security issues, Customs are pressed to conduct
more selections to detect more. Given that not every selection leads to detection,
however, an increase in selection does not guarantee an increase in detection. This
study examines the correlation between selection and detection to probe what changes
in detection are induced by changes in selection.

In Table 5, the correlation coefficient21 for the overall selection rate shows that
there is nearly no relationship between selection and detection (ro = -0.001); which
means that a change in the overall selection rate is not likely to lead to a change in its
detection rate.

The correlation analysis on each selection method, however, shows somewhat

different pictures. The manual selection has a stronger relationship between the
selection and detection rates (rm = - 0.41) than the other selection methods (rru = 0.11
and rra = 0.32). The low coefficient for the rule-based selection (rru = 0.11) implies that its
selection and detection rates are barely related. When it comes to the directions of the
coefficients for each selection method, the coefficient of the random selection are
positive. In other words, an increase in the selection rate of the random selection tends
to lead to an increase in its detection, whereas for the manual selection, the more
selected, the less detected. This finding suggests that a simple increase in selection
rate is likely to end up imposing costs to Customs and traders without an increase in
detection rate. In particular, when the KCS is asked to raise its selection rate, it would
be better to raise the random selection rate rather the manual selection rate.

Table 5: Correlation coefficients to show the association between selection and

detection rates of selection methods
Overall Rule-based selection Random selection Manual selection
-0.001 0.11 0.32 -0.41

5. Conclusion

This study shows that the three selection methods have different outcomes.
However, they do not work independently but are intertwined and complement one
another. Without the manual selection, the rule-based selection cannot stand. As

21
The subject of this study is the population of import declarations submitted to the KCS from
2009 to 2011. Significance tests are not necessary to infer population parameters from sample
statistics.

15
advocates for statistical scoring techniques admit, the results from manual selections are
a crucial ingredient of rule-based selections. The results from random selections show
clues of which areas Customs officers pay attention to when they manually select
suspicious declarations or cargoes. The performance of rule-based selections plays a
role in presenting guides to develop powerful manual selections applicable to region-
wide or nationwide.

There is no doubt, therefore, that good manual selections benefit rule-based

selections. Even though the manual selection shows a good outcome, if the rule-based
selection produces poor performance, it should be suspected that there is something
wrong with the process of transforming refined manual selections to the selection criteria
of the rule-based selection. Three kinds of problems may occur in the migration process.
First, some skillful selectors do not have motivation to share their expertise with their
colleagues because their own expertise is the result of their investment for better
information and knowledge. Their selection expertise is tightly associated with
promotion and merit bonus. Second, it takes time to make powerful selection criteria
applicable to nationwide or region-wide with individual manual selection results. Hence,
a division may need to be in full charge of creating and managing selection criteria.

It is necessary to share responsibilities between the three selection methods in

order to make the import selectivity system work well as a whole. The manual selection
needs to concentrate on complicated fraud techniques and cunning traders and make
better achievement, when the rule-based selection treats relatively simple frauds and
casual fraudulent traders. If Customs rely on only the manual selection to obtain a
current level of deterrence without employing the rule-based selection, Customs need to
recruit more Customs officers for selection work or cannot help taking sloppy selections.
The random selection also complements the other two selection methods through its
unpredictable selection pattern. The manual selection and the rule-based selection are
based on the analysis of criminal traders’ fraud technique patterns and in turn the
selections are carried out with certain patterns. Hence, the manual selection and the
rule-based selection have difficulties in tackling unknown and irregular techniques. Such
obscure techniques can come to light by the random selection. Therefore, a selectivity
system can operate by an exquisite combination among the three selection methods.

This study does not take into account various factors in its analyses on the
performance of the import selectivity system, such as the attention or will of the
commissioner or directors on selection work, the proficiency of selectors and inspectors,
and changes in incentive for higher detection rate. This study’s subject is the import
selectivity system of the KCS and does not contend that the results are generalizable to
other Customs’ selectivity systems. That is because each Customs may have different
contexts and mechanisms in operating its selectivity system. Nonetheless, this study is
expected to function as a stepping stone for subsequent empirical research on risk
management techniques.

16
References

Agresti. A. and Finlay, B. (1999) Statistical Methods for the Social Sciences. Upper
Saddle River: Prentice Hall.

Alm, J. (1999) Tax Compliance and Administration. In: W. Hildreth and J. Richardson
(eds.) Handbook on Taxation. Boca Raton: CRC Press.

Andreoni, J., Erard, B., and Feinstein, J. (1998) Tax Compliance. Journal of Economic
Literature, 16: 818-860.

Babbie, E. (2007) The Practice of Social Research. Belmont: Thomson-Wadsworth.

Becker, G. (1968) Crime and Punishment: An Economic Approach. Journal of Political

Economy: 169-217.

Closs, D. and McGarrell, E. (2004) Enhancing Security throughout the Supply Chain.
Washington D.C.: the IBM Center for the Business of Government.

De Wolf, L. and Sokol, J. (eds.) (2004) Customs Modernization Handbook. Washington,

D.C.: The World Bank.

Deloach, J. (2000) Enterprise-Wide Risk Management: Strategies for Linking Risk and
Opportunity. London: Financial Times/Prentice Hall.

Derrig, R. (2002) Insurance Fraud. The Journal of Risk and Insurance, 69(3): 271-287.

Felson, M. (2009) Crime and Everyday Life. Thousand Oaks: Sage Publication.

Geourjon, A., Laporte, B., Coundoul, O., and Gadiaga, M. (2012) Inspecting Less to
Inspect Better. In: T. Cantens, R. Ireland, and G. Raballand (eds.) Reform by
Numbers: Measurement Applied to Customs and Tax Administrations in
Developing Countries. Washington D.C.: The World Bank.

Graetz, M., Reinganum, J. and Wilde, L. (1986) The Tax Compliance Game: Toward an
Interactive Theory of Law enforcement. Journal of Law, Economics, and
Organization, 2(1): 1-32.

Graetz, M., and Wilde, L. (1985) The Economics of Tax Compliance: Fact and Fantasy.
National Tax Journal, 38: 355-363.

Harrison, M and Holloway, S. (2007) Customs and Supply Chain Security – “The Demise
of Risk Management?” Paper presented at Annual Conference of APEC Centres:
18 April, Melbourne, Australia.

Hintsa, J. , Hameri, A., Thibedeau, C., Sahlstedt, J., Tsikolenko, V., Finger, M., and
Granqvist, M. (2011) Customs Risk Management (CRiM): A Survey of 24 WCO
Member Administrations. Lausanne: Cross-Border Research Association.

17
Keen, M. (2003). The Future of Fiscal Frontiers and the Modernization of Customs
Administration. In: M. Keen (eds.) Changing Customs: Challenges and Strategies
for the Reform of Customs Administration. Washington D.C.: The World Bank.

Kirchler, E., Muehlbacher, S., Kastlunger, B., and Wahl, I. (2007) Why Pay Taxes? A
Review of Tax Compliance Decisions. International Studies Program Working
Paper 07-30. Atlanta: Georgia State University.

Knowles, J., Persico, N., and Todd, P. (2001) Racial Bias in Motor Vehicle Searches:
Theory and Evidence. Journal of Political Economy, 109(1): 203-229.

Laporte, B. (2011) Risk Management Systems: Using Data Mining in Developing

Countries’ Customs Administrations. World Customs Journal, 5(1): 17-27.

McLinden, G., Fanta, E., Widdowson, D., and Doyle, T. (eds.) (2010) Border
Management Modernization. Washington D.C: The World Bank.

Miller, J. (2005) The Chicago Guide to Writing about Multivariate Analysis. Chicago: The
University of Chicago Press.

Nagin, D. (2011) Deterrence – Scaring Offenders Straight. In: F. Cullen and C. Jonson
(eds.) Correctional theory: Context and Consequences. Thousand Oaks: Sage
Publication.

Nagin, D. (1998) Criminal Deterrence Research at the Outset of the Twenty-First

Century. Crime and Justice, 23: 1-42.

Organization for Economic Cooperation and Development (OECD). (2003) Policy Brief -
Trade Facilitation: The Benefits of Simpler, more Transparent Border Procedure.
Paris: The OECD.

Picard, P. (1996) Auditing claims in the insurance market with fraud: the creditability
issues. Journal of Public Economics, 63:27-56.

Pinquet, J., Ayuso, M. and Guillen, M. (2007) Selection Bias and Auditing Policies for
Insurance Claims. Journal of Risk and Insurance, 74(2): 425-440.

Pratt, T., Cullen, F., Blevins, K., Dagie, L., and Madensen, T. (2008). The Empirical
Status of Deterrence Theory: A Meta-Analysis. In: F. Cullen, J. Wright, and K.
Blevins (eds.) Taking Stock: The Status of Criminological Theory. New
Brunswick: Transaction Publishers.

Tennyson, S. and Salsas-Forn, P. (2002) Claims Auditing in Automobile Insurance:

Fraud Detection and Deterrence Objectives. Journal of Risk and Insurance,
69(3): 289-308.

Viaene, S., Ayuso, M., Guillen, M., Van Gheel, D., and Dedene, G.. (2007) Strategies for
detecting fraudulent claims in the automobile insurance industry. European
Journal of Operational Research, 176: 565-583.

18
Viaene, S. and Dedene, G. (2004) Insurance Fraud: Issues and Challenges. The
Geneva Papers on Risk and Insurance, 29(2): 313-333.

Welsh, B. C. and Farrington, D. P. (1999) Value for money? A review of the costs and
benefit of situational crime prevention. The British Journal of Criminology, 39(3):
345-368.

Widdowson, D. (2007) The Changing Role of Customs: Evolution or Revolution? World

Customs Journal 1(1): 31-37.

Widdowson, D. (2005) Managing Risk in the Customs Context. In: L. de Wulf and J.
Sokol (eds.) Customs Modernization Handbook. Washington D.C.: World Bank.

World Customs Organization (WCO). (2011) WCO Customs Risk Management

Compendium. Brussels: The WCO.

World Customs Organization (WCO). (2010) Security and Facilitation of Trade:

Consolidated Results of Safe Framework Implementation Survey. Brussels: The
WCO.

19