The role of customs in securing containerized global supply chains

Accepted Manuscript

The Role of Customs in Securing Containerized Global Supply Chains M. Pourakbar, R. Zuidwijk PII: DOI: Reference:

S0377-2217(18)30406-5 10.1016/j.ejor.2018.05.012 EOR 15125

To appear in:

European Journal of Operational Research

Received date: Revised date: Accepted date:

9 October 2016 3 April 2018 8 May 2018

Please cite this article as: M. Pourakbar, R. Zuidwijk, The Role of Customs in Securing Containerized Global Supply Chains, European Journal of Operational Research (2018), doi: 10.1016/j.ejor.2018.05.012

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT

Highlights • Container inspection rate is optimized using a game between Customs and antagonist. • The trade-off is economic impact on public and private parties and security.

CR IP T

• Containers Informational and physical inspection rates are functions of risk levels.

AC

CE

PT

ED

M

AN US

• Incentive mechanisms to build public private partnership for security are developed.

1

ACCEPTED MANUSCRIPT

Supply Chains

CR IP T

The Role of Customs in Securing Containerized Global

M. Pourakbar, R. Zuidwijk

Rotterdam School of Management, Erasmus University

AN US

Burgemeester Oudlaan 50, 3000DR Rotterdam, The Netherlands [email protected], [email protected]

AC

CE

PT

ED

M

Corresponding author: Morteza Pourakbar, [email protected], Phone: (+31) 10 - 40 82775

2

ACCEPTED MANUSCRIPT

Abstract Customs has to deal with a massive number of containers arriving at ports. This massive flow of cargo provides an opportunity for organized crime infiltration. Risk management and the security of the supply chain has become a top priority for Customs administrations and for

CR IP T

private firms. In this paper, we develop models that allow Customs to optimize its inspection process to target high-risk containers without hindering the flow of safe containers with extra delays at ports. The model characterizes optimal informational and physical inspection rates as a function of the risk factors attributed to containers. We use this model to analyze how an effective public-private partnership for risk and security management can be established

AN US

between Customs and private firms.

Key words: supply chain management; security management; public-private partnership; cus-

AC

CE

PT

ED

M

toms; game theory

3

ACCEPTED MANUSCRIPT

1

Introduction

Seaports handle billions of metric tons of import/export cargo annually. This massive flow of cargo provides a huge opportunity for antagonists (including smugglers, traffickers, terrorists, and fraudsters) to infiltrate this channel. Customs plays a prominent but challenging role in stopping the

CR IP T

flow of illicit products. The operating environment of Customs administrations is characterized by a constantly changing risk landscape. Customs antagonists are dynamic enemies who review and reshape their operations constantly. These illicit operators seek high profits and low-risk business opportunities, much as their legal counterparts do, while adopting new ways to conceal illicit

AN US

items, forge trading documents, or solicit Customs officers. Customs is concerned with all the risks related to the cross-border movement of goods, with the following characteristics: 1) Belong to Customs responsibilities as stated in the Customs law; 2) Have negative impacts on citizens and/or nations such as security, safety, health, and economics etc.; 3) Cause problems in costs, liabilities and reputational issues of supply chains (Hintsa et al. [2011]).

M

Transport security is largely dependent on Customs efforts. Prior to September 11, 2001 most

ED

discussions on transportation security focused on controlling theft and reducing contraband. After September 11, the highest-order definition of freight security changed from theft-proof to tamperproof. Suddenly, intermodal containers became potential weapon delivery systems. Since then,

PT

security issues have become one of the primary concerns for both Customs and supply chains.

CE

Customs efforts to secure supply chains, society and the public are not limited to anti-terrorist activities. A high level of attention is paid to trafficking (both human and drug), smuggling, and

AC

counterfeiting. Customs main approach towards preventing illicit and threatful shipments from entering society is informational or physical inspection. Cargo inspections might take place at the time of stuffing the container at the point of origin. Accordingly, for some countries carriers are required to submit manifest information even before cargo is loaded onto a vessel. Customs then performs a background screen on the manifest, carrier, and shipper to determine if the shipment poses a risk to society. Customs performs additional screening prior to shipment arrival via a computerized targeting system. Physical inspections also take place at the port of destination. 4

ACCEPTED MANUSCRIPT

Given the capacity limitations of Customs, a percentage of incoming containers are selected for additional scrutiny. It consists of scanning via an X-ray or a gamma-ray scanner. Inspectors examine the scanned images for discrepancies with the manifest and for other signs of risk. In some cases, physical inspection is required in which the container is opened and its contents unloaded

CR IP T

and examined (Moore [2007]). Inspections are time-consuming and therefore costly for both Customs and supply chains. Bakshi and Gans [2010] report that unanticipated container delays at an inspection facility can cause costly supply chain disruptions equal to 0.5% of the value of a container per day. One of the main challenges for Customs is to design an inspection policy so that the positive impact of mitigating

AN US

the risk of a threatful container escaping Customs inspection outweighs the negative impact of extra delays, costs, and inefficiencies imposed on supply chain parties. Any changes in inspection policy raises several issues such as purchasing, operating, and maintaining additional facilities, as well as possible changes in container delivery time. Therefore, the main trade-off for Customs, given its

M

limited resources, is how to balance the cost of inspection incurred by both Customs and firms and to mitigate security risks. In order to address these concerns, we develop an optimal inspection

ED

policy, which addresses the trade-off between cost efficiency and security. Such a policy should utilize Customs’ limited resources to target containers with high-risk profiles and to prevent them

PT

from entering legitimate supply chains. This should be done effectively and efficiently without creating an operational obstacle for supply chain firms. As the first contribution of this paper, we

CE

model this as a sequential game between Customs and the antagonist, in which the antagonist sets the rate of infiltration and Customs reacts accordingly and sets inspection policies. In this setting,

AC

we consider the impact of Customs information and physical inspection policies on the possible cost incurred by business firms and on the mitigation of security risks. We show that the optimal inspection policy should take into account the containers’ risk profile and allocate resources accordingly. More specifically, we find the optimal percentage of containers with a specific risk factor that should be inspected through informational and physical inspection processes. We show that, as intuition dictates, containers with higher risk factors should be inspected more frequently.

5

ACCEPTED MANUSCRIPT

These results are formulated in theorems 4 and 5 in Section 5. We also examine, as the second contribution of this paper, how a public-private partnership for security enhancement between Customs and business firms can be established. Even though Customs administrations are in the front-line of battling illicit activities, the physical flow of goods

CR IP T

and its related information is in the hands of private companies that own the shipments, containers, and ships. Therefore, Customs needs private sector information on the containers’ histories to assess how secure these containers are, and private companies need the benefits of the intelligence and enforcement resources owned by government to reduce the probability of disruptions in their supply chains. Moreover, private parties need to be assured that Customs inspections will

AN US

not cause major disruptions in the flow of containers, and might be even willing to invest in facilitating Customs inspections. This explains why public-private partnerships (PPP) have become a major point of discussion in businesses and government agencies concerned with homeland security (Busch and Givens [2012]). Furthermore, PPP is a significant and enduring part of critical

M

infrastructure protection initiatives and plays a paramount role in achieving supply chain resilience (Sheffi [2007]). There have been various mechanisms for establishing a public-private partner-

ED

ship between Customs and business firms to enhance security. Some private companies invest in Customs’ inspection facilities to ensure a smooth flow of their own cargo. For example, APM

PT

Terminals has invested in a $10 million container inspection facility in Nigeria for the physical examination of containers by Customs operatives. This facility introduces measures to aid prompt

CE

service delivery at the port (APM-Terminals [2011]). There are also other forms of partnerships aiming at sharing information between Customs and private firms. They are mainly based on

AC

the fact that firms have access to valuable sources of information that Customs can use to target high-risk containers. In addition, private firms can also piggyback on Customs’ risk management systems to improve their security. For example, the EU 7th Framework Program project CASSANDRA - Common Assessment and Analysis of Risk in Global Supply Chains - aims at introducing a data pipeline for the exchange of required information along the entire supply chain. This enables open, flexible and standardized communication amongst all partners (CASSANDRA [2013]). An-

6

ACCEPTED MANUSCRIPT

other example of PPP through data and information sharing is Interface Public Member (IPM), which is a global anti-counterfeiting tool currently used by one third of WCO Members (WCO [2012]). Using this platform, Customs officers can access IPM anywhere in the world via a simple and secure interface available in their national language. Customs officers can log on at any time

CR IP T

to learn about a variety of products, brands, and the distinguishing features between counterfeit and genuine products. IPM is freely available to all of member Customs administrations. Sharing counterfeit related information by both Customs and supply chains through IPM can help both parties in implementing an effective anti-counterfeiting strategy. Using a principal-agent modeling approach, we model two viable scenarios, namely, investment in physical inspection facilities and

AN US

information technology enhancements. We are able to characterize efficient investment thresholds that could lead to collaboration and partnership between public and private firms resulting in

2

Literature Review

M

improved security management.

ED

Although governments and governmental agencies play a significant role in securing supply chains, and their actions impact supply chains’ performance significantly, their role has not received the at-

PT

tention it deserves in the operations management/supply chain management literature. Few papers have studied the role of government’s security initiatives on the operational aspects of cross-border

CE

supply chains. Lee and Whang [2005] describe how the principles of total quality management can actually be used to design and operate processes to assure supply chain security. They argue

AC

that higher quality can be attained at lower cost by proper management and operational design, and that this central theme of the quality movement is also applicable in supply chain security. They consider the Smart and Secure Trade-lanes initiative (SST) and show how supply chain security can be obtained at lower cost by using simple quantitative tools. They show that real time information of containers while in transit can be utilized to significantly lower inventory costs. Wein et al. [2006] develop a mathematical model to find the optimal inspection strategy for the detection of an

7

ACCEPTED MANUSCRIPT

illicit nuclear weapon. Considering an 11-layer security system, the study suggests that employing high-energy x-ray radiography and elongating passive neutron tests at overseas ports may provide cost savings. Kantor and Boros [2010] argue that the best detection rates are achieved when the available budget is allocated between screening and unpacking using a mixture of strategies that

CR IP T

maximize detection rate. They show that this yield increases as much as 100% in detection, without a significant increase in inspection costs. Bakshi et al. [2011], using discrete event simulation based on actual data, empirically show that the current screening regime is not effective, and a twostep screening process (with a rapid primary screening of all cargo, followed by a more detailed screening of those containers that failed the first test) might be better. The most related paper to

AN US

ours is the one by Bakshi and Gans [2010] that develops an economic model in which the government provides incentives to firms in the form of reduced inspections if the firms decide to comply with Customs-Trade Partnership Against terrorism (C-TPAT) security initiative. Our paper differs from [Bakshi and Gans, 2010] as we consider a sequential multi-level inspection process and we

M

characterize the optimal inspection rate at each level as a function of risk levels. Additionally, we develop incentive mechanisms to initiate public private partnership (PPP) between customs and

ED

supply chain firms. Moreover, we find investment thresholds to establish an effective partnership. Research in the area of homeland security has also examined how government strategies can

PT

directly impact adversary’s actions such as terrorist attacks. For example, Haphuriwat et al. [2011] study the 100% inspection policy imposed by the US government at either US or foreign ports.

CE

They quantify a game-theoretical model of terrorist decision making to understand the role of nuclear detection technologies in deterring nuclear terrorism. Based on the level of retaliation

AC

imposed by the terrorist, they are able to determine whether a 100% inspection should be conducted or whether partial inspection is sufficient. This stream of research is broadly categorized as defender-attacker problems and explores the efficiency of government risk mitigation strategies against terrorist infiltration risk (Zhuang and Bier [2007], Bier [2007], Pinker [2007], Bakir [2011], Wang and Zhuang [2011], Nikoofal and Zhuang [2012], and Rios and Insua [2012] ). Given that this paper focuses on finding an optimal inspection policy taking into account se-

8

ACCEPTED MANUSCRIPT

curity considerations, another relevant body of literature concerns the passengers airline security inspection. After September 11 attacks, there has been a growing body of literature on designing and optimizing passengers security inspections at airports. There is a number of inspections done on passengers as well as on baggages to check if there are any threats entering the system. Some

CR IP T

of the studies investigate the advantages of classifying passengers into different groups and find the degree of inspection required for each group such that the security risks are minimized. Using queuing theory, Babu et al. [2006] develop a model determining the number of groups, the fraction of passengers and their assignment to check points such that the passenger inconvenience is minimized. Nie et al. [2009] extend this work by relaxing the assumption that the threat probability is

AN US

constant across all passengers. They assume that passengers are classified into several risk classes via a pre-screening system. Additionally, the staffing level at each check point is considered in the model. McLay et al. [2010] develop a model to sequentially and optimally assign passengers to airport security checkpoints. In their setting, the authors assume that risk levels are determined

M

through a pre-screening system and these levels are used to optimally assign passengers to inspection resources such that the total security is maximized subject to capacity and assignment

ED

constraints. The problem is modeled as a Markov decision process and the assignment policy is found using dynamic programming. Nie et al. [2012] explain that there are two kinds of screening

PT

lanes in a typical US airport, namely, a normal lane and a selectee lane. They argue that a selectee lane is not effectively utilized and only a small number of passengers are selected to go through

CE

it. They develop a simulation based model to assign passengers from different risk classes to the selectee lane based on the number of passengers already waiting in the selectee lane.

AC

What differentiates our work from the previous studies are as follows: First it develops a holis-

tic view integrating all Customs’ inspection processes. The proposed model includes all primary informational, secondary informational, and physical inspections and their inherent errors such as the risk of false detention of safe containers as well as false release of risky containers. Second, this model does not focus on a specific risk and it can be used to address several categories of risks be it terrorist, counterfeiting, or smuggling, because it incorporates the impact of risk mitigation

9

ACCEPTED MANUSCRIPT

strategies on both society and business firms. Third, we quantify the impact of the delay caused by increased inspections on business firms. We consider the trade-off between inefficiency due to delays and improved risk and security status. Fourth, we also model real-world public private partnership mechanisms for enhancing the security of maritime logistics. Using this modeling

CR IP T

framework, we obtain investment thresholds to promote the establishment of an effective PPP. As previously emphasized, it has been advocated that maritime security management requires a high level of collaboration between governments and supply chain firms and our model helps to identify and signify the efficient collaboration mechanisms.

The rest of this paper is structured as follows: We describe the problem in the next section

AN US

and model Customs’ and the antagonist’s problems in Section 3. In Section 4, we develop optimal inspection policies for both primary informational and physical inspections. Section 5 presents a numerical analysis of the base case and Section 6 extends the model to account for public-private

Problem definition

ED

3

M

partnership mechanisms. Section 7 concludes the paper.

Essentially, all Customs inspections are done in two stages, as outlined in Figure 1. The first stage

PT

is about informational inspection of the data and documents of the containers and at the second stage, if needed, the contents of the containers are physically inspected. The first stage consists

CE

of two substages, namely, primary informational and secondary informational inspection. Once a preliminary declaration is placed and prior to the arrival of a container, a container undergoes

AC

primary informational inspection. This inspection is based on processing container related data including manifest documents, and bill of lading, etc. Some containers belonging to trusted parties are considered to be low-risk. These are green-flagged and are not subject to further inspection. The output of inspection assigns a risk score z to each container. This risk score signifies the level of risk associated with the container and is defined as P{no alarm | threat}. Hence, 1 − z signifies the discovery of a threat immediately after primary informational inspection. According

10

ACCEPTED MANUSCRIPT

release α2, β2

no signal

First stage: secondary informational inspection

flag green

𝑧 no alarm

alarm 1−𝑧 detain

release

PR{z} flag red

P{y}

CR IP T

threat assign risk score 𝑧 no threat

First stage: primary informational inspection

alarm 1−𝑦

detain

1-PR{z}

1-P{y}

𝛼,𝛽

Second stage: physical inspection

AN US

ℎ(𝑧)

assign risk score 𝑦

signal

flag orange α1, β1

no alarm 𝑦

𝑔(𝑦)

release

detain

Figure 1: Customs inspection processes

M

to the value of z, Customs might identify some high-risk containers. These are red-flagged and

ED

must undergo physical inspection. However, the value of z might not be decisive enough and further informational inspection might be required. In such a case, the container is medium-risk and is orange-flagged. This is time-consuming and therefore costly for both Customs and business

PT

firms because the containers need to wait at Customs until the authorities decide on further action.

CE

Depending on the attributed risk score, z, the authorities decide which percentage of containers with a risk score of z, are to be sent for further secondary informational inspection (orange-flagged)

AC

and which percentage are to be physically inspected (red-flagged). These percentages are denoted by PO {z} and PR {z}, where PO {z} = 1 − PR {z}. Moreover, h(z) and H(z) are defined as the density and cumulative functions of risk levels after primary informational inspection. Orange-flagged containers are sent for a more sensitive secondary informational inspection,

where further documentation screening takes place. If the container triggers an alarm signal after the secondary informational inspection, the container is detained. If no alarm signal is triggered, a risk factor y is attributed to the container. y basically means that secondary informational inspec11

ACCEPTED MANUSCRIPT

Table 1: Table of notaion

CR IP T

Definition Risk level assigned after primary informational inspection Risk level assigned after secondary informational inspection Density and cumulative functions of risk levels at primary informational inspection Density and cumulative functions of risk levels at secondary informational inspection Type I errors of, respectively, primary informational, secondary informational and physical inspections Type II errors of, respectively, primary informational, secondary informational and physical inspections Percentage of containers sent for physical inspection after primary informational inspection Percentage of containers sent for physical inspection after secondary informational inspection Antagonist’s penalty if a threat is detected The delay cost of a safe container caused by a false detention The waiting cost of a safe container while undergoing physical inspection Antagonist’s profit if a risky container escapes inspections Customs cost of inspecting a container per time unit Customs cost of detaining a container per time unit Arrival rate of containers at primary informational inspection Rate of antagonist’s container infiltration

AN US

Notation z y h(z) and H(z) g(y) and G(y) α1 , α2 and α β1 , β2 and β PR {z} P{y} Lf Ld Lw Le ws wd µ γ

tion does not raise any alarm, even though the container might carry a threat. Thus (1 − y) signifies

M

the discovery of a threat. Similar to primary informational inspection, g(y) and G(y) are defined as the density and cumulative functions of risk levels after secondary informational inspection.

ED

According to the output of secondary informational inspection, the system decides whether to send the container for physical inspection or to release it. The higher the risk factor attributed to the

PT

container, the more likely it is to be flagged for physical inspection. P{y} denotes the percentage of containers with risk factor y flagged for physical inspection.

CE

We assume that the antagonist’s affordability to penetrate in legitimate channels is limited. This is denoted by γ as a fraction of the total number of containers. The antagonist has to choose which

AC

containers with a certain risk level to infiltrate. We model the interaction between the antagonist and Customs as a Stackelberg game and use backward induction to determine the optimal inspection policies. From the antagonist’s perspective, it is optimal to infiltrate containers that offer the highest profit. If a container is detained by Customs, a penalty Lf (Lf < 0) applies depending

on the type of risk. If a container escapes Customs inspections, the antagonist gains a benefit that amounts to Le . Moreover, it is worth noting that some adversaries such as counterfeiters, do not 12

ACCEPTED MANUSCRIPT

only benefit from diluting the stream of legitimate cargo, but also take advantage of perturbing the stream of licit cargo. This is because any delay in the delivery of genuine products might provide opportunities for counterfeiters to sell counterfeit products in the market. Thus, we incorporate the cost of detention delays, Ld , and the waiting cost of safe containers while undergoing physical

CR IP T

inspection, Lw , in the profit function of the antagonist. In our setting, during primary informational inspection the authorities need to set PR {z} and consequently PO {z}. For physical inspection, P{y}, the rate of containers flagged for physical inspection with a risk level y has to be decided. We also assume that there are inspection error inherent to each inspection process. The errors included in the model are type I and II errors.

AN US

Type I error happens when a safe container is detained and type II error occurs once a risky container is released after inspection. For primary and secondary informational inspections we define αi and βi where i ∈ {1, 2} as the type I and II errors, respectively. They are defined as αi = P{threat detected | safe} and βi = P{no threat detected | threat}. For physical inspection,

Modeling Customs’ and Antagonist’s Problems

ED

4

M

we similarly define these error terms as α and β. All notation are summarized in Table 1.

PT

First, we model the antagonist’s profit function associated with primary informational inspection. Given a stream of containers with risk score z and inspection probability PR {z}, the antagonist’s

CE

profit function, if she decides to infiltrate these containers, is expressed by Π1a = (1 − z) [(1 − γ)α1 E{Ld } + γ(1 − β1 )Lf ] 

AC h



+z PR {z} α1 (1 − γ)(E{Lw } + αE{Ld }) + (1 − β1 )γ(βLe + (1 − β)Lf ) + (1 − PR {z})Π2a (1)

The first term in relation (1) expresses the antagonist’s profit if a container with risk score z is detained with probability (1 − z). If the antagonist is, for example, a counterfeiter she gains an expected benefit of (1 − γ)α1 E{Ld } from the delay caused in the flow of legitimate containers but 13

i

ACCEPTED MANUSCRIPT

incurs a cost if containers are found to be counterfeit. This cost is equal to γ(1 − β1 )Lf . A risky container will not trigger an alarm signal with probability z. This leads to a fraction of containers with risk score z, PR {z}, to be flagged for physical inspection. If the container is legitimate but is flagged for physical inspection because of an error during primary informational inspection,

CR IP T

α1 , the expected profit for the antagonist equals α1 (1 − γ)(E{Lw } + αE{Ld }). However, if an infiltrated container is not red-flagged, because of a type II error during primary informational inspection, β1 , it will be detected during physical inspection with probability (1 − β) and the antagonist incurs a cost of Lf . Otherwise, it escapes physical inspection and the antagonist gains a profit of Le . A container with risk score z is flagged for secondary informational inspection with

AN US

probability (1 − PR {z}) and the antagonist gains a profit of Π2a . The antagonist’s profit is Π2a after the container is flagged orange for secondary informational inspection. Π2a will be modeled later in this section. For the sake of simplicity, we assume that the profit of the antagonist is equal to the collective loss of Customs and business firms. Thus, the government decides on PR {z} such that

M

expected total cost for Customs and business firm is minimized as follows: min Π1a = (1 − z) [(1 − γ)α1 E{Ld } + γ(1 − β1 )Lf ]

ED

PR {z}

  i 2 +z PR {z} α1 (1 − γ)(E{Lw } + αE{Ld }) + (1 − β1 )γ(βLe + (1 − β)Lf ) + (1 − PR {z})Πa (2)

PT

h

Customs decides on PR {z} such that the benefits for the antagonist is minimized. We now calculate

CE

the benefits for the antagonist after physical inspection. Depending on the risk factors attributed to the containers, y, distributed according to g(y), and inspection errors α2 and β2 , the expected

AC

benefit for the antagonist by targeting γ percent of containers is as follows  Π2a = (1 − y)[γ(1 − β2 )Lf + (1 − γ)α2 E{Ld }] + y (1 − P{y})γβ2 Le

   i h  +P{y} γβ2 (1 − β)Lf + βLe + (1 − γ)(1 − α2 ) αE{Ld } + E{Lw }

14

(3)

ACCEPTED MANUSCRIPT

α2 and β2 denote errors associated with secondary informational inspection. The profit function of the antagonist after physical inspection (3) follows the same rationale as that of (1). If the container triggers an alarm signal during inspection, with probability (1−y), the antagonist benefits (1 − γ)α2 E{Ld } if it is a safe container because of a type I error, α2 , during the inspection process.

CR IP T

However, if it is a risky container, the antagonist will incur a cost of γ(1−β1 )Lf . If no alarm signal is triggered, with probability y, then a percentage of (1 − P{y}) is released and the antagonist gains γβ2 Le which is proportional to a type II error during secondary informational inspection, β2 . P{y} percent of containers with a risk score of y are flagged for physical inspection after the secondary informational inspection. If the container is safe and flagged for physical inspection

AN US

because of a type I error during secondary informational inspection, the antagonist’s expected benefit is (1 − γ)(1 − α2 )[αE{Ld } + E{Lw }]. Similarly, if a container is infiltrated and flagged for physical inspection, the antagonist’s expected profit is γβ2 [(1 − β)Lf + βLe ]. In order to determine Customs equilibrium strategy, given capacity limitations, the admin-

minimized. The expected total cost is

M

istration sets P{y} such that the expected loss to the society and the legitimate supply chain is

ED

 minΠ2a = (1 − y)[γ(1 − β2 )Lf + (1 − γ)α2 E{Ld }] + y (1 − P{y})γβ2 Le P{y}

(4)

PT

   i h  +P{y} γβ2 (1 − β)Lf + βLe + (1 − γ)(1 − α2 ) αE{Ld } + E{Lw }

CE

Inspections are costly for business firms. They are willing to participate as long as the cost incurred by the antagonist due to inspections outweighs the cost incurred by them because of perturbations

AC

in the legitimate channel due to increased levels of inspections. In other words, to avoid trivial solutions, we assume that the benefits of avoiding a threat should exceed the inspection costs incurred by firms. The non-triviality assumptions are stated for all physical, secondary and primary informational inspections, respectively, as follows:

γβ2 (1 − β)(Le − Lf ) ≥ (1 − γ)(1 − α2 )(αE{Ld } + E{Lw }) 15

(5)

ACCEPTED MANUSCRIPT

, γ(1 − βi )(−Lf ) ≥ (1 − γ)(αi E{Ld } + E{Lw }), i ∈ {1, 2}

(6)

γ(1 − β1 )(1 − β)(Le − Lf ) ≥ (1 − γ)α1 (αE{Ld } + E{Lw })

(7)

and

CR IP T

These constraints, corresponding to physical, secondary informational, and primary informational inspections, respectively, basically mean that the direct benefits of seizing a risky container should outweigh the indirect costs caused by increased inspections. The left-hand side of expression (5) indicates the expected benefits of detaining a threatful container flagged for physical inspection af-

AN US

ter secondary informational inspection. Customs benefits from penalizing the antagonist and from not allowing the antagonist to benefit from Le . The right-hand side of this expression indicates the costs of such a secondary informational inspection policy. The cost includes the waiting, delay and false detention costs imposed on business firms. As indicated in the inspection process diagram

M

in Figure 1, both primary and secondary informational inspections can lead to an immediate detention, (6) can ensure that the benefits of detention of rightly detected threat outweighs the cost

ED

of false detention of safe containers during both primary and secondary informational inspections. Relation (7) is similar to relation (5) expressed for primary informational inspection.

PT

However, we should note it is practically impossible for Customs to detect all risks. Eventually, counterfeiters, smugglers, and other adversaries are able to conceal some parts of their activities

CE

from Customs inspection. We assume that the antagonist determines her activities so that the gains through orange stream is non-negative. i.e., (Π2a ≥ 0). It is highly unlikely that a container

AC

flagged for physical inspection leaves the inspection facility undetected. The antagonist’s main profit comes from orange-flagged containers released before being sent for physical inspection.

4.1

Inspection Processes

The effectiveness of inspection procedures highly depends on the time and care with which they are conducted. For example, the amount of time and care spent on analyzing the data associated 16

ACCEPTED MANUSCRIPT

with containers, interpreting X-ray images or physically inspecting containers play a vital role in identifying threat. Following the same approach as Bakshi and Gans [2010] we define Si as the time spent during the inspection procedure, and we model inspection time as a function of errors

Si = ψ(κi , αi , βi ) + φi

CR IP T

and inspection capacity as follows:

(8)

where φi is a random variable to incorporate the randomness involved in the procedure. An instance for ψ(κi , αi , βi ) to model the physical inspection process is ln(α) ln(β) − . κ κ

AN US

ψ(κ, α, β) = −

(9)

κ ≥ 1 indicates the inspection capacity i.e. the number of staff conducting inspection or the number or capacity of machines used in the process. We assume that the inspection process is

M

rigorous and as a measure for rigorousness we assume that error terms are not very high and are

Congestion at Physical Inspection Facility

PT

4.2

ED

such that α, α1 , α2 , β, β1 and β2 ≤ 0.5.

To make the problem analytically tractable, we assume that orange-flagged containers arrive at the

CE

physical inspection facility according to a Poisson process. Thus, the inspection process can be modeled as an M/G/1 queue. Additionally, we assume that there is only one inspection facility.

AC

These assumptions allow us to include an analytical expression for the number of containers at the inspection facility. θ defined as follows indicates the fraction of orange-flagged containers, which

are subject to physical inspection.

θ=

Z

0

1

P{y}h(y)dy

17

(10)

ACCEPTED MANUSCRIPT

Similarly, the fraction of containers flagged red after primary informational inspection and are subject to physical inspection is calculated according to

θR =

Z

1

0

PR {z}g(z)dz.

(11)

CR IP T

Assuming that containers arrive for primary informational inspection according to a Poisson process with arrival rate µ, the arrival rate at the physical inspection facility, for tractability reasons, is also assumed to follow a Poisson process with mean λ = (θ + θR )µ. Using the properties of M/G/1 queue, the expected number of containers either undergoing inspection or waiting to be processed

AN US

is given according to

Expected # of containers in physical inspection = N{S p } = λE{S} + (λ)2 where ρ2 is the utilization rate of the inspection facility and is defined as

E{S 2 } 2(1 − ρ2 )

λ . E{S}

(12)

Although the

M

objective functions in (2 and 4) incentivize Customs to keep the error terms as small as possible, concern for the viability of its own administration prevent it from simply setting them equal to

ED

zero. Specifically, Customs is willing to reduce errors as long as inspection induced congestion

E{S } λE{S} + (λ)2 2(1−ρ ≤n 2) 2

(13)

CE

PT

does not surpass a certain threshold given by the following:

This can be interpreted as Customs’ physical and capacity constraints. For example, it might have

AC

limited capacity that can only process up to n containers in a certain time period. Customs by setting PR {z} and P{y} influences θ and θR and accordingly λ. Thus, Customs has to set PR {z}

and P{y} such that (13) holds.

18

ACCEPTED MANUSCRIPT

4.3

Congestion at Primary informational Inspection

Assuming that containers arrive at the primary informational inspection facility according to a Poisson process, we can follow the same approach as physical inspection. However, in this case, the expected time a container undergoes primary informational inspection should not exceed a

CR IP T

certain threshold.

Expected waiting time of a container for primary informational inspection = E{S1 }+µ Thus, we have 2

AN US

E{S1 } ≤τ E{S1 } + µ 2(1−ρ 1)

E{S1 2 } 2(1 − ρ1 ) (14)

(15)

Given the vast number of containers arriving at port, a speedy processing of information is required.

Analysis of the Base Case

M

5

In order to determine the optimal percentage of containers sent to physical inspection after primary

ED

and secondary informational inspections, we use a backward dynamic programming approach with two stages, namely, primary informational and secondary informational inspections. We start back-

PT

ward and first find optimal decision variables for the secondary informational inspection. We can then characterize the optimal primary informational inspection policy. First, we show some prop-

CE

erties that are used to determine the optimal inspection policies.

AC

Lemma 1 We define ` = {y|y offers the maximum benefit to the antagonist} with y ∗ = inf ` and ξ = {z|z offers the maximum benefit to the antagonist} with z ∗ = inf ξ, then • P{y} is strictly increasing over y in ` • PR {z} is strictly increasing over z in ξ. Lemma 1 shows that the riskier the containers are, the higher the percentage that should be flagged for physical inspection. Thus, Customs should allocate a bigger share of its inspection capacity to 19

ACCEPTED MANUSCRIPT

high-risk containers. Moreover, from the antagonist’s perspective, setting higher inspection levels for high-risk containers makes all containers equally likely to be infiltrated. Besides inspection frequency, inspection rigour also plays a key role in establishing an effective inspection process. The more time spent on rigorously inspecting the containers, the less likely it becomes that a risky

CR IP T

container can escape detection or that a safe container is falsely detained. In this model, we do not optimize for inspection errors, but the following lemma states that despite the fact that Customs seeks to minimize these errors, it would never be optimal to set them equal to zero. Lemma 2 If ∃ α, β, α1 , β1 such that ψ(κ, α, β) < ∞ and ψ1 (k, α1 , β1 ) < ∞ then

AN US

• It is never optimal to have α = 0 or β = 0

• It is never optimal to have α1 = 0 or β1 = 0 • It is never optimal to have θ = 0 or θR = 0.

M

The main intuition behind Lemma 2 is that Customs avoids setting very low values for error terms as this imposes huge operational inefficiencies through excessive delays. Moreover, as long as

flagged containers to zero.

ED

there is a threat from an antagonist, Customs should never set the rate of orange-flagged or red-

PT

Lemma 3 Customs space availability constraint is binding in the optimal solution provided that

CE

antagonist benefits, Π1∗ a > 0.

Moreover, Customs processing time constraint for primary informational inspection is binding

AC

in the optimal solution provided that antagonist benefits, Π2∗ a > 0. As intuition dictates, Lemma 3 shows that as long as the antagonist’s profit is positive, Customs

should utilize its capacity to the fullest. Using these properties, we can characterize the optimal physical inspection policy as follows: Theorem 4 P{y}, the percentage of container with risk factor y sent for physical inspection right after secondary informational inspection, can be characterized as follows: 20

ACCEPTED MANUSCRIPT

• if 0 ≤ y ≤ y ∗ then P{y} = 0 • if y ∗ ≤ y < 1 then y∗ γ(1 − β2 )Lf + (1 − γ)α2 E{Ld } − γβ2 Le (1 − ) γ(1 − β2 )[(1 − β)Lf + βLe ] + (1 − γ)α2 [αE{Ld } + E{Lw }] − γβ2 Le y (16)

CR IP T

P{y} =

Theorem 4 states that there is a threshold, y ∗ , on the risk score. As long as a container’s estimated risk is below this threshold, it should not be subject to physical inspection but once the risk score goes beyond this threshold, it should be inspected with a probability P{y}, which is strictly in-

AN US

creasing in y. Similar to physical inspection, we can determine the optimal primary informational inspection as follows:

Theorem 5 Optimal primary informational inspection policy, PR {z}, can be characterized as

• if 0 ≤ z ≤ z ∗ then PR {z} = 0

M

follows:

ED

• if z ∗ ≤ z < 1 then PR {z} is strictly increasing in z and PR {z} should be found such that

CE

PT

the following equation holds

Γ1 − E[Π2a ] z∗ PR {z} = (1 − ) Γ2 − E[Π2a ] z

(17)

AC

where

Γ1 = (1 − γ)α1 E{Ld } + γ(1 − β1 )Lf

and

Γ2 = α1 (1 − γ)(E{Lw } + αE{Ld }) + (1 − β1 )γ(βLe + (1 − β)Lf )

As observed in relations (16) and (17), further characterization of the optimal PR {z} and P{y} is very cumbersome. It is because the choice of PR {z} and P{y} affects waiting and congestion

21

ACCEPTED MANUSCRIPT

Table 2: Parameter setting of the base case scenario Definition Antagonist’s penalty if a threat is detected Antagonist’s profit if a risky container escapes inspections Customs cost of inspecting a container per time unit Customs cost of detaining a container per time unit Arrival rate of containers at primary informational inspection Rate of container infiltration

Value -100 200 15 15 5 0.1

CR IP T

Notation Lf Le ws wd µ γ

related costs in equations (3) and (4). Therefore, for analytical tractability, using Lemma 1, we assume that PR {z} and P{y} are both linear and strictly increasing functions in z and y, respectively.

AN US

More particularly, we define PR {z} = δ1 (z − z ∗ ), z ≥ z ∗ , and P{y} = δ2 (y − y ∗ ), y ≥ y ∗ where δ1 and δ2 > 0 are slopes of lines corresponding to primary and secondary informational inspection policies. We try to find the optimal values for δ1 , δ2 and z ∗ , y ∗ such that the profit for the antagonist

6 Numerical Analysis

M

is minimized.

ED

In this section, we conduct a numerical analysis to achieve a better understanding of the role of costs and other parameters on optimal inspection policies. The parameter setting of the base case

PT

is shown in Table 2.

First, we study the impact of the antagonist’s infiltration rate. When this increases, Customs

CE

expects to detain more high-risk containers. But Customs faces space availability and capacity restrictions and would rather allocate available space to high-risk containers only. As a consequence,

AC

when the infiltration rate increases, the inspection rate of low-risk containers decreases. This can be observed in Figure 2 for both primary informational and secondary informational inspections. An increase in the magnitude of type I and II errors during inspection makes the process more

costly and accordingly less attractive. Therefore, if errors increase sharply, Customs is better off decreasing inspection levels to avoid the costs incurred due to inspection inherent errors. This result is depicted and observable in Figure 3 for primary informational inspection. 22

AN US

CR IP T

ACCEPTED MANUSCRIPT

AC

CE

PT

ED

M

Figure 2: Impact of γ on inspection policies

Figure 3: Impact of changing α1 and β1 on primary inspection policy

23

ACCEPTED MANUSCRIPT

7

Extensions: Public-Private Partnership Models to Enhance Customs Security Management

As discussed in the introduction section, enhancing containerized supply chain security requires

CR IP T

effort and investment by both private firms and public administrations. In this section, we extend the results of the base setting to model two common and viable public-private partnership mechanisms. The first one models a private firm’s investment in public physical inspection facilities to speed up the inspection process. Examples of such investments are training Customs ground officers to improve the detection efficiency of high-risk containers or investing in Customs inspection

AN US

equipment and state-of-the-art technologies. The second case models a private firm’s investment in supply chain visibility, by providing Customs with more accurate and quality container-related information. This allows Customs to piggyback on business data and private firms to benefit from

7.1

M

Customs’ risk management expertise.

Investment in Customs Physical Inspection Facilities

ED

In this section, we model the interaction between supply chain actors and Customs as a sequential game. We assume that the antagonist always decides to infiltrate γ percent of containers with

PT

a specific risk factor. Then, incorporating the antagonist’s decision, the interaction between the private firm and Customs is modeled as a Stackelberg game in which the firm acts as the leader

CE

and decides whether to invest in Customs inspection facilities. First, Customs decides on the level of capacity increase k, k > 1 and investment requirement by firms Ik . In return for this investment,

AC

inspection error terms are also updated to αp and βp . The contract offered by Customs is denoted

by {Ik , k, αp , βp }. The firm’s decision of whether to accept the proposal depends on the cost of executing the

program. The firm faces two choices: Whether to reject the offer and incur the same inspection induced costs as before or to invest in inspection facilities and benefit from an enhanced inspection process. 24

ACCEPTED MANUSCRIPT

7.1.1 The Agent’s Problem The business firm decides to accept the offer if the operational efficiencies gained through enhanced inspection compensates the investment requested. An enhanced inspection process will affect the inspection and detention rate of Customs, and consequently the inspection induced costs incurred

CR IP T

by the firm. The condition that must be satisfied for the firm to accept the offer is as follows:

Ik + ws N{Snp } + wd N{Dnp } ≤ ws N{S p } + wd N{Dp }

(18)

N{Snp } and N{S p } are, respectively, the expected number of containers per time unit either waiting

AN US

for or under inspection after investment in inspection facilities and before that. Similarly, N{Dnp } and N{Dp } denote the expected number of containers in detention per time unit after and before investment. Customs incurs inspection costs but the firm incurs other costs including labor, instruments and facilities. ws and wd are, respectively, inspection and detention cost per container per

M

time unit. Ik denotes the investment on a per time unit basis. It is an increasing function in k, meaning that higher levels of capacity require higher investment levels. The following proposition

ED

formulates a threshold for the investment level.

Proposition 6 If Customs inspection procedure follows α and β errors, there is a threshold for the

PT

firm to accept the principal’s offer {Ik , k, αp , βp }, and invest in increasing inspection facilities (κ)

CE

by a factor k, which is unique and given by

(19)

AC

ζp = ws (N{S p } − N{Snp }) + wd (N{Dp } − N{Dnp })

7.1.2

The Principal’s Problem

The principal tries to minimize the expected costs of the antagonist’s activities. The solution to the principal’s problem should be to provide the appropriate incentives for the firm to participate. The capacity constraint for Customs remains similar to (13), meaning that the total number of containers being processed should not exceed Customs’ space availability and capacity. 25

ACCEPTED MANUSCRIPT

7.2

Customs’ Improvement on Business Data Availability

If supply chain business actors decide to invest in their supply chain visibility, Customs can benefit by using business information systems to piggyback on both business data and business risk management systems. This implies that Customs would have a better risk assessment and therefore the

CR IP T

risk scoring mechanism would be improved. As a result, the number of orange-flagged containers would decrease.

The contract offered by Customs is denoted by {Iν , η} where Iν denotes the investment that supply chain firms need to make on their own and in Customs information infrastructures to allow Customs access to supply chain databases. η is the new rate of data availability. The motivation

AN US

behind this setting is as follows: If the firm decides to invest in data acquisition infrastructures, it can offer Customs more business data. As a result, Customs can access various data sources from several supply chain stakeholders and can therefore make a better risk assessment. Better risk assessment is reflected in the definition of g(z) and h(y) functions. After investing in data

M

availability, the risk density function is altered by gv (z) and hv (y) where Gv (z) ≥ G(z) and Hv (y) ≥ H(y), i.e. the risk after investment is stochastically dominated by the risk before invest-

ED

ment. As a result, Customs can enhance its risk assessment mechanism and target risky containers

PT

more accurately and efficiently 7.2.1 The Agent’s Problem

CE

The firm would accept the offer if the congestion cost reduced through improved container targeting outweighs the investment required. If such investment is made, primary informational inspec-

AC

tion processing might take longer because of higher levels of information availability, but arrival rates to secondary informational inspection will decrease. It should be noted that improving supply chain visibility through data sharing not only affects Customs’ primary informational inspection but also directly influences the arrival rate to the physical inspection facility. Therefore, we need to consider the effect of inspection and detention costs in both inspection processes. Denoting the investment on supply chain visibility per time unit by Iν and letting subscript ν denote the cost 26

ACCEPTED MANUSCRIPT

parameters and decision variables after making the supply chain visible, the condition that should be met to induce the firm’s investment in visibility is stated as follows:

Iν + ws (N{Sνi } + N{Sνp }) + wd (N{Dνi } + N{Dνp }) ≤ ws (N{S i } + N{S p }) + wd (N{Di } + N{Dp }) (20)

CR IP T

ws and wd are defined similar to the previous case. N{Sνi } and N{Sνp } denote the number of containers either waiting for or undergoing primary informational inspection and physical inspection, respectively, after investment in the supply chain visibility has taken place. N{S i } and N{S p } denote the same measures but before investment in supply chain information sharing infrastructure.

AN US

N{Dνi } and N{Dνp } denote the number of containers in detention after primary informational and physical inspections after investment. Similarly, we define N{Di } and N{Dp } as the same measures but before investment. This constraint basically ensures that the firm’s investment compensates for the additional operational costs caused by improved inspection and detention processes.

M

Proposition 7 If Customs inspection procedure follows α1 and β1 errors, there is a threshold for the firm to accept the principal’s offer {Iν , η, α1n , β1n }, and increase data sharing by a factor η,

ED

which is unique and given by

The Principal’s Problem

CE

7.2.2

PT

ζi = ws (N{Sνi } + N{Sνp } − N{S i } − N{S p }) + wd (N{Dp } + N{Dnp } − N{Di } − N{Dp }) (21)

The principal’s problem remains similar to the previous case. The solution to the principal’s prob-

AC

lem should incentivize the firm to participate.

7.3 Analysis We first analyze the investment thresholds developed in Section 6 to induce PPP for security improvement. The first PPP mechanism discussed earlier is investment in physical inspection facilities. As observed in Figure 4.a, the firm’s investment threshold increases for a higher level of 27

CR IP T

ACCEPTED MANUSCRIPT

Figure 4: Investment in physical inspection facility vs. business data availability

AN US

inspection capacity. The intuition is that increasing capacity imposes higher investment cost, but results in higher operational efficiency for firms. With more inspection facilities, Customs can handle containers more quickly and efficiently, which reduces congestion related costs for firms. As mentioned earlier, if business parties share business data and information with Customs, the

M

administration can improve risk assessment and as a result the pdf of risk score would change to gν (z) such that Gν (z) ≥ G(z). An example of this is shown in Figure 5, where the pdf of z is a

ED

discrete function. η = 1 represents the status quo of the probability distribution function. As η increases, the pdf of z also changes such that Gν (z) ≥ G(z). As a result of the higher level of

PT

supply chain visibility provided by business firms, the risk score attributed to containers decreases. According to Figure 4.b, corresponding to the second PPP mechanism, if supply chain visibility

CE

increases and higher levels of data and information are provided to Customs, the firm’s investment threshold increases. This is because even though such a higher load of information might require

AC

extra processing, it allows Customs to target high-risk containers more effectively and efficiently. Better targeting of high-risk containers results in less congestion in the flow of licit containers. Therefore, the firm would be willing to make a higher level of investment in IT infrastructure in order to provide Customs with more supply chain visibility.

28

CR IP T

ACCEPTED MANUSCRIPT

Figure 5: g(z) for different values of η

Summary and Conclusion

AN US

8

Massive flows of containers provide an opportunity for terrorists, smugglers, and counterfeiters to infiltrate this mode of transport. Customs is in the front-line of battling antagonists. The ad-

M

ministration’s main weapon against the antagonists is inspection, either informational or physical. However, inspections are costly not only to Customs but also to firms and supply chains as delays

ED

can have a detrimental economic impact on firms. Moreover, Customs has limited resources and their utilization should be optimized. Even though the decisions of Customs can have a significant

PT

impact on public and private parties, its role has not been extensively investigated in the academic literature. In this paper, we develop models that can be used to determine Customs’ optimal in-

CE

spection policies considering its resource constraints and the cost incurred on firms because of inspection related delays. We show that Customs need to take into account container risk factors,

AC

and that high-risk containers should be assigned more inspection capacity. The model can be exploited to address different categories of risks faced by Customs. Securing containerized supply chains is the joint responsibility of private firms and public agencies because information is owned by private firms, whereas intelligence is owned by Customs. An effective public-private partnership model could enhance Customs’ risk and security management and benefit both public and private firms. In this paper, we consider two common practices in the real world, namely, private

29

ACCEPTED MANUSCRIPT

investment in Customs’ inspection facilities and the creation of supply chain visibility by allowing Customs to have access to business data. Using our model, we are able to find the optimal investment threshold for private firms in order to develop an effective public-private partnership.

CR IP T

References APM-Terminals. APM terminals apapa berth productivity up 50%. Technical report, 2011. URL www.apmterminals.com/media/Files/Corporate/News/PressReleases. V. L. L. Babu, R. Batta, and L. Lin. Passenger grouping under constant threat probability in an

AN US

airport security system. European Journal of Operational Research, 168(2):633–634, 2006. N. O. Bakir. A stackelberg game model for resource allocation in cargo container security. Annals of Operations Research, 187(1):5–22, 2011.

M

N. Bakshi and N. Gans. Securing the containerized supply chain: Analysis of government incen-

ED

tives for private investment. Management Science, 56(2):219–233, 2010. N. Bakshi, S. E. Flynn, and N. Gans. Estimating the operational impact of container inspections at

PT

international ports. Management Science, 57(1):1–20, 2011. V. Bier. Choosing what to protect. Risk Analysis, 27(3), 2007.

CE

N. E. Busch and A. D. Givens. Public-private partnerships in homeland security opportunities and

AC

challenges. Homeland Security Affairs, 8(1), 2012. CASSANDRA. risk

in

global

Cassandra supply

=

common

chains.

assessment

Technical

report,

and

analysis

2013.

of URL

http://www.cassandra-project.eu/mainmenu/home.html. N. Haphuriwat, V. M. Bier, and H. H. Willis. Deterring the smuggling of nuclear weapons in container freight through detection and retaliation. Decision Analysis, 8(2):88–102, 2011. 30

ACCEPTED MANUSCRIPT

J. Hintsa, T. Mannisto, A.P. Hameri, C. Thibedeau, J. Sahlstedt, V. Tsikolenko, M. Finger, and M. Granqvist. Customs risk management (crim): A survey of 24 wco member adminstrations. Technical report, World Customs Organization and Cross-Border.org, 2011.

gets: The testing power index. Risk Analysis, 30(4), 2010.

CR IP T

P. Kantor and E. Boros. Deceptive detection methods for effective security with inadequate bud-

H. Lee and S. Whang. Higher supply chain security with lower cost: Lessons from total quality management. International Journal of Production Economics, 96(3):289–300, 2005.

L. A. McLay, A. J. Lee, and S. H. Jacobson. Risk-based policies for airport security checkpoint

AN US

screening. Transportation Science, 44(3):333–349, 2010.

J.E. Moore. The economic impacts of terrorist attacks. Edward Elgar Publishing, 2007. X. Nie, R. Batta, C. J. Drury, and L. Lin. Passenger grouping with risk levels in an airport security

M

system. European Journal of Operational Research, 194(2):574–584, 2009.

ED

X. Nie, G. Parab, R. Batta, and L. Lin. Simulation-based selectee lane queueing design for passenger checkpoint screening. European Journal of Operational Research, 219(1):146–155, 2012.

PT

M. E. Nikoofal and J. Zhuang. Robust allocation of a defensive budget considering an attacker’s private information. Risk Analysis, 32(5):930–943, 2012.

CE

E. J. Pinker. An analysis of short-term responses to threats of terrorism. Management Science, 53

AC

(6):865–880, 2007.

J. Rios and D. R. Insua. Adversarial risk analysis for counterterrorism modeling. Risk Analysis, 32(5), 2012.

Y. Sheffi. The resilient enterprise: overcoming vulnerability for competitive advantage. The MIT Press, paperback edition, 2007.

31

ACCEPTED MANUSCRIPT

X. Wang and J. Zhuang. Balancing congestion and security in the presence of strategic applicants with private information. European Journal of Operational Research, 212(1):100–111, 2011. WCO. Interface public members. Technical report, 2012. URL http://www.wcoipm.org/.

materials in shipping containers. Risk Analysis, 26(5), 2006.

CR IP T

L. M. Wein, A. H. Wilkins, M. Baveja, and S. E. Flynn. Preventing the importation of ilicit nuclear

J. Zhuang and V. M. Bier. Balancing terrorism and natural disasters-defensive strategy with en-

AC

CE

PT

ED

M

AN US

dogenous attacker effort. Operations Research, 55(5):976–991, 2007.

32