RISK MANAGEMENT IN SUPPLY CHAINS

by

Sanjay Kumar

APPROVED BY SUPERVISORY COMMITTEE:

___________________________________________Kathryn E. Stecke, Chair

___________________________________________Holly S. Lutze

___________________________________________Divakar Rajamani

____________________________________________Thomas G. Schmitt

Copyright 2009

Sanjay Kumar

All Rights Reserved

Memories of Vishnu

RISK MANAGEMENT IN SUPPLY CHAINS

by

SANJAY KUMAR (B.E, M.TECH., M.S.)

DISSERTATION

Presented to the Faculty of

The University of Texas at Dallas

in Partial Fulfillment

of the Requirements

for the Degree of

DOCTOR OF PHILOSOPHY IN

MANAGEMENT SCIENCE

THE UNIVERSITY OF TEXAS AT DALLAS

December, 2009

UMI Number: 3391748

All rights reserved

INFORMATION TO ALL USERS The quality of this reproduction is dependent upon the quality of the copy submitted.

In the unlikely event that the author did not send a complete manuscript

and there are missing pages, these will be noted. Also, if material had to be removed, a note will indicate the deletion.

UMI 3391748

Copyright 2010 by ProQuest LLC. All rights reserved. This edition of the work is protected against

unauthorized copying under Title 17, United States Code.

ProQuest LLC 789 East Eisenhower Parkway

P.O. Box 1346 Ann Arbor, MI 48106-1346

v

PREFACE

This dissertation (or thesis) was produced in accordance with guidelines which permit the

inclusion as part of the dissertation (or thesis) the text of an original paper or papers

submitted for publication. The dissertation (or thesis) must still conform to all other

requirements explained in the “Guide for the Preparation of Master's Theses and Doctoral

Dissertations at The University of Texas at Dallas.” It must include a comprehensive

abstract, a full introduction and literature review and a final overall conclusion. Additional

material (procedural and design data as well as descriptions of equipment) must be provided

in sufficient detail to allow a clear and precise judgment to be made of the importance and

originality of the research reported.

It is acceptable for this dissertation (or thesis) to include as chapters authentic copies of

papers already published, provided these meet type size, margin and legibility requirements.

In such cases, connecting texts which provide logical bridges between different manuscripts

are mandatory. Where the student is not the sole author of a manuscript, the student is

required to make an explicit statement in the introductory material to that manuscript

describing the student's contribution to the work and acknowledging the contribution of the

other author(s). The signatures of the Supervising Committee which precede all other

material in the dissertation (or thesis) attest to the accuracy of this statement.

vi

ACKNOWLEDGEMENTS

I would like to express my deep and sincere gratitude for everyone who helped me in

completing this dissertation. Foremost, it is difficult to overstate my gratitude to my Ph.D.

supervisor Kathryn E. Stecke for her invaluable guidance and encouragement. I am also

thankful for her unswerving support, patience, and temperament, which were keys in

completion of this dissertation.

I am deeply indebted by the guidance and support of Dr. Thomas G. Schmitt of the

University of Washington Seattle. His wide knowledge and logical way of thinking have

been of great value for me. His personal guidance has provided a good basis for the present

thesis. A significant portion of this dissertation is based on my work with Dr. Thomas G.

Schmitt and Dr. Fred G. Glover of the University of Colorado at Boulder. I am grateful to

Dr. Fred Glover for helping me learn adaptive search methods.

Other members of my dissertation committee have served varied and extremely useful roles

in my development as a scholar and the development of this thesis. I thank Dr. Holly Lutze

for helping me with the job searching process. I also appreciate her help in wholeheartedly

sharing all her teaching material and tips. I have also benefited from Dr. Divakar Rajamani’s

class on practical and applied aspects of supply chain management.

I would like to acknowledge the financial support from the School of Management, the

University of Texas at Dallas. The support provided by Sophia and Barbara is also

appreciated.

vii

Most importantly, none of this would have been possible without the love and support of my

family and wife Renee. My very special thanks to my parents whom I owe everything. Thank

you for everything.

My studies in Dallas would not have been the same without the diversions provided by all

my student-colleagues in the School of Management. I am particularly thankful to Anna,

Arnold, Di, Emre, Gan, Peter, Sam, Satya, Sirong, Subodha, and Yue.

Finally, the acknowledgements would not be complete without a heart felt thanks to Vishnu

Manikumar, my dearest friend to whom this dissertation is dedicated. He was the most

valuable friend I will ever have. Vishnu, you will always be with us.

August, 2009

viii

RISK MANAGEMENT IN SUPPLY CHAINS

Publication No. ___________________

Sanjay Kumar, Ph.D.The University of Texas at Dallas, 2009

Supervising Professor: Kathryn E. Stecke

Many supply chain models and studies have been designed for a certain world. In reality,

uncertainty caused by disruptions and their consequences have underlined the need for

models that address planning and operational decisions under disruptions. This dissertation

studies ordering and planning decisions in a multi-stage supply chain under disruptions.

In the first of the three essays, we collect and compile data and show that there has been a

marked increase in both the frequency and economic losses from catastrophes. U.S. business

interests are particularly vulnerable to terrorist attacks. A catastrophe classification

framework that matches different types of catastrophes to a variety of infrastructural

components of supply chains is proposed. We propose strategies that can decrease the

vulnerability of a supply chain. Potential benefits from mitigating strategies during normal

times are also investigated. We also identify 16 research problems and areas that should be

addressed to create resiliency in supply chains. In second essay we address one of these

problems, which is related to relastic modelling and understanding the long term

consequences of disruptions and popular disruption mitigation strategies.

ix

We conduct a study of disruptive effects as part of a large-scale study launched by Sandia

National Laboratories. The essay involves case studies of electronics firms to reveal the

structure and complexity of supply chains, prior disruptions, and mitigation strategies in use.

We identify three vital metrics as drivers of performance. Experiments demonstrate that an

objective function, which applies costs to the performance metrics, can be quite ill-behaved.

We find that order expediting, often used to mitigate disruptions, can induce bullwhip effects

that hurt system performance. While increased information and flexibility to react are

generally desirable, we observe that it is also easy to overreact, with undesirable

consequences. The study motivates the use of metaheuristic methods as an alternative to

analytical modelling. In third essay develop these metaheuristic methods.

Adaptive and evolutionary metaheuristics search methods are explored for their effectiveness

in searching ordering decisions in a multi stage supply chain. Adaptive search was shown to

be superior to evolutionary or line search methods. The performance of these methods under

various scenarios related to disruptions and seasonality is also reported.

x

TABLE OF CONTENTS

Preface..................................................................................................................................v

Acknowledgements..............................................................................................................vi

Abstract..............................................................................................................................viii

List of Figures....................................................................................................................xiii

List of Tables......................................................................................................................xv

CHAPTER 1 INTRODUCTION .......................................................................................1

1.1 Sources of Supply Chain Disruptions, Factors that Breed Vulnerability, and Mitigating Strategies .............................................................................2

1.2 Modeling and Analyzing Economic Consequences of Supply Chain Disruptions...................................................................................................3

1.3 Adaptive Search Methods for Ordering Decisions in Multi-stage Supply Chains.. ........................................................................................................3

CHAPTER 2 SOURCES OF SUPPLY CHAIN DISRUPTIONS, FACTORS THAT BREED VULNERABILITY, AND MITIGATING STRATEGIES...........5

2.1 Introduction..................................................................................................5

2.2 Catastrophes.................................................................................................8

2.2.1 Intentional Acts ................................................................................9

2.2.2 Accidents........................................................................................13

2.2.3 Natural Calamities..........................................................................14

2.3 Factors that Cause Vulnerability in a Supply Chain..................................18

2.4 Catastrophe Classification .........................................................................20

2.5 Catastrophe Mitigation...............................................................................24

2.5.1 Proactive Strategies ........................................................................24

2.5.2 Advance Warning Strategies..........................................................26

2.5.3 Coping Strategies ...........................................................................27

2.5.4 Cost/Benefit Trade-offs of Mitigation Strategies...........................30

xi

2.5.5 Selection of Mitigation Strategies ..................................................31

2.6 Research Problems in Disruptions Management .......................................33

2.7 Conclusions................................................................................................41

CHAPTER 3 MODELING AND ANALYZING ECONOMIC CONSEQUENCES OF SUPPLY CHAIN DISRUPTIONS............................................................43

3.1 Introduction................................................................................................43

3.2 Literature Review.......................................................................................45

3.3 Bridging Macroeconomic and Operational Decision-making ...................49

3.4 Key Modeling Issues..................................................................................51

3.4.1 Key Issue 1: A Reasonable Baseline for the Supply ChainStructure........ .................................................................................51

3.4.2 Key Issue 2: Inventory and Disruption Logic ................................55

3.4.3 Key Issue 3: Performance Metrics .................................................59

3.4.4 Key Issue 4: System Performance with a Disruption.....................60

3.4.5 Key Issue 5: System Performance under Expediting .....................64

3.4.6 Key Issue 6: System Performance and Applicability of Analytics, Heuristics ......................................................................66

3.4.7 Key Issue 7: Genetic Search versus Line Search ...........................69

3.5 Conclusions and Future Directions............................................................71

CHAPTER 4 ADAPTIVE SEARCH METHODS FOR ORDERING DECISIONS IN MULTI-STAGE SUPPLY CHAINS.........................................................77

4.1 Introduction................................................................................................77

4.2 Literature Review.......................................................................................80

4.3 The Supply Chain Model ...........................................................................85

4.3.1 Supply Chain Structure and Activities ...........................................85

4.3.2 Supply Chain Parameters and the Inventory Logic........................86

4.3.3 Ordering Policies............................................................................91

4.3.4 Objective Function and Decision Variables...................................92

4.4 Description of Metaheuristic Search Methods...........................................93

4.4.1 Fibonacci Search ............................................................................94

4.4.2 Genetic Search................................................................................94

4.4.3 Tabu Search Strategies for Supply Chain Applications .................96

4.5 Results......................................................................................................102

xii

4.5.1 Implementation of Metaheuristics................................................103

4.5.2 Cost Comparison between Search Methods under Static Demand........................................................................................106

4.6 Comparison between Static and Dynamic Policies .................................110

4.6.1 Adaptive Search Methods under Seasonal Demand ....................111

4.6.2 Adaptive Search Methods under Disruptions...............................115

4.7 Extensions to a Global Supply Chain ......................................................118

4.8 Conclusions..............................................................................................119

CHAPTER 5 CONCLUSIONS......................................................................................121

APPENDIX A: THE SANDIA AGENT-BASED MODEL............................................125

APPENDIX B: CASE STUDIES OF ELECTRONICS COMPANIES IN THE PACIFIC NORTHWEST.......................................................................132

APPENDIX C: CASE INTERVIEW PROTOCOL ........................................................148

BIBLIOGRAPHY.............................................................................................................157

VITA

xiii

LIST OF FIGURES

Number Page

Figure 2.1 Total number of attacks worldwide...............................................................10

Figure 2.2 Percentage attacks on various classes of targets from 1991 to 2003.............10

Figure 2.3 Percentage attacks on U.S. business interests to total attacks on U.S.Interests.........................................................................................................11

Figure 2.4 Average yearly accidents and the yearly damage caused..............................14

Figure 2.5 Average natural catastrophes reported and the yearly economic losses caused............................................................................................................15

Figure 2.6 Total natural disasters in the U.S...................................................................16

Figure 2.7 Total number of disasters in the U.S. with economic losses exceeding onebillion dollars.................................................................................................17

Figure 2.8 Components at various stages of a supply chain............................................21

Figure 3.1 Prototypical supply chain...............................................................................55

Figure 3.2 Performance effects of a disruption...............................................................63

Figure 3.3 Shape of total cost function........................................................................... 68

Figure 4.1 Sample convergence....................................................................................109

Figure 4.2 Average convergence in 100 runs............................................................... 109

Figure 4.3 Seasonal demand function...........................................................................113

Figure 4.4 Order-up-to levels for echelon 1 under seasonal demand........................... 115

Figure 4.5 Demand function with disruptions in final customer demand.....................117

Figure 4.6 Order-up-to levels for echelon 1 under a disruption demand......................118

Figure A.1 The N-ABLE enterprise agent....................................................................128

xiv

Figure A.2 Example of regional shortages caused by N-ABLE simulation of HurricaneKatrina.........................................................................................................130

Figure B.1 Case study supply chain structures .............................................................133

xv

LIST OF TABLES

Number Page

Table 2.1 International terrorist attacks by region ...........................................................11

Table 2.2 Supply chain practices and their effect on vulnerability causing factors.........18

Table 2.3 Severity and possibility of effects of a catastrophe..........................................23

Table 2.4 Interrelationship of strategies and catastrophe types .......................................34

Table 2.5 Interrelationship of strategies and vulnerability breeding factors ...................36

Table 3.1 Case summaries ...............................................................................................54

Table 3.2 Various costs at two levels each ......................................................................69

Table 3.3 Results of Line and Genetic search under various cost combinations.............71

Table 4.1 Cost comparison of various metaheuristics solution methods........................105

Table 4.2 Percentage cost comparison...........................................................................106

Table 4.3 Consistency of metaheuristics methods.........................................................107

Table 4.4 Sample results ................................................................................................108

Table 4.5 Performance of ordering policies under seasonal demand ............................114

Table 4.6 Percentage improvements in using dynamic policies ....................................114

Table 4.7 Performance of ordering policies under disruptions.......................................117

Table 4.8 Percentage improvements in using dynamic policies ....................................118

1

CHAPTER 1

INTRODUCTION

Real-world applications of supply chains often experience significant economic

repercussions as a result of disruptions. There have been specualtions that such disruptions

are increasingly affecting supply chain performance. Despite disruptions having a low

probability of occurrence, there is a good chance overall of something big and unexpected to

happen. These events could be costly and can impact the performance for a long time. This

underlines the need to develop supply chain models to manage disruptions.

This thesis includes three essasy on closely related aspects of supply chain

management under disruptions. In this dissertation we start with identification of various

sources that have resulted in increased disruptions in supply chains. We also identify

strategies that can be used to mitigate these disruptions. We collaborte with SANDIA

national laboratories to study disruptions in north-western United States. The study using

literature and case studies develops a rellastic supply chain model and identify performance

metrics critical for modelling. We also reveal properties of objective function that motivate

the use of metaheristic search methods for making ordering decisions in supply chains. In the

last chapter we develop metaheuristic tools for making ordering decisions in a multi-stage

supply chain. These tools are shown to be effective in a dynamic enviroment. In this rest of

this chapter we summarise the three essays covered in this thesis.

2

1.1 Sources of Supply Chain Disruptions, Factors that Breed Vulnerability, and

Mitigating Strategies

The focus of business towards increasing efficiency and reducing costs has resulted in supply

chains that are efficient during normal times, but at the cost of being vulnerable to

disruptions. From time to time frequent as well as rare catastrophes also disrupt supply chain

operations. We collect and compile data from many sources and show that there has been a

marked increase in both the frequency and economic losses from natural and man-made

catastrophes. We find that business losses constitute a major percentage of the total losses

caused by these catastrophes. The statistics suggest that for terrorist attacks, the vulnerability

of U.S. business interests is much higher than others. Examination of the geographical and

chronological distributions of catastrophes provides useful information for managers

concerned about such disruptions.

We develop a catastrophe classification framework that matches different types of

catastrophes to a variety of infrastructural components of supply chains. The framework also

connects a variety of mitigating strategies to appropriate catastrophe types. We identify

factors that can be used to assess the vulnerability of a supply chain. They can also be useful

to compare possible alternative decisions based on the vulnerability they may cause in the

supply chain.

To manage vulnerability in supply chains, we propose strategies that can be

implemented by a company to decrease the possibility of occurrence, provide advance

warning, and cope after a disturbance. We reveal potential benefits from mitigating strategies

during normal times, which indicates that well-developed strategies can also result in better

3

efficiency. We identify many future research areas concerning disruption handling in supply

chains.

1.2 Modeling and Analyzing Economic Consequences of Supply Chain Disruptions

Supply chains often experience significant economic disruptions, as in the case of facility

breakdowns, transportation mishaps, natural calamities, and terrorist attacks. We collaborated

in a study of such disruptive effects as part of a large-scale initiative by Sandia National

Laboratories. For our part, we began by conducting case studies of three electronics firms

and their suppliers to reveal the industry’s supply chain structure and complexity, nature of

historical disruptions, and possible mitigation strategies. We identify three vital metrics

(final-echelon service levels, system inventory, and system expediting) as drivers of

performance. Exploratory experiments indicate that a cost function based on these drivers

can be quite ill-behaved, and metaheuristics capable of looking beyond local optima are

warranted. We found that genetic search over inventory system parameters yielded better

solution quality than unimodal search. We also observe that bullwhip effects induced by

disruptions can severely affect service levels and system inventory for long periods. We find

that order expediting, often used to mitigate disruptions, can also induce bullwhip effects,

and hurt rather than help overall performance. While we support the notion that increased

information and flexibility are generally desirable, it is easy to overreact, with undesirable

consequences.

1.3 Adaptive Search Methods for Ordering Decisions in Multi-stage Supply Chains

This chapter focuses on the problem of determining periodic review inventory management

policies in multi-echelon supply chains. A four-echelon, six-stage decentralized supply chain

4

is modeled. We propose a novel objective function that enables us to find a dynamic, time

variant policy. The objective is to minimize discounted costs of holding inventory,

backorders, lost sales, and expediting. The decision variables are order-up-to quantity,

percentage of orders expedited, and expediting triggers. We consider two triggers: the first is

based on on-hand inventory. A second trigger results in expedited orders when the pipeline

inventory falls below a threshold.

We propose evolutionary and adaptive search methods to solve the problem.

Fibonacci and genetic search methods are used to establish a baseline for comparison. We

also use properties of search space to increase the effectiveness of search routines. The study

focuses on establishing adaptive search method as a means to make decisions in a supply

chain. Under the cost parameters considered, the results suggest that adaptive search can

significantly outperform evolutionary and gradient search methods for ordering decisions.

We test the robustness of method by considering a variety of cost combinations. We also test

the method under non stationary demand scenarios. Under a seasonal and disrupted demand

we show that Adaptive search methods can be effectively used to make ordering decisions.

Results for a local objective are presented and an extension to a global objective is suggested.

5

CHAPTER 2

SOURCES OF SUPPLY CHAIN DISRUPTIONS, FACTORS THAT BREED

VULNERABILITY, AND MITIGATING STRATEGIES

2.1 Introduction

In recent years, there has been a surge in articles addressing supply chain disruptions. The

addressed disruptions range from a small plant fire that may affect a small number of local

companies to global catastrophes such as 2002-2003 SARS, which affected global economy.

Regardless of the scale of a catastrophe, the consequences could be severe for a company.

Some recent documented major supply chain disruption causes include the West

Coast port strike in 2002, the 1999 Taiwan earthquake, hurricanes Katrina and Rita in 2005,

Asian tsunami in 2004, and September 11 attacks. Other smaller catastrophes such as snow

storms, heavy rain, excessive wind, fire, industrial and road accidents, strikes, and changes in

government regulations regularly interrupt normal operations in supply chains. These

disruptions motivated researchers to address various aspects of disruptions management,

including severity (Craighead et al. 2007), causes (Kleindorfer and Saad 2005), resilience

(Sheffi and Rice 2005), risk management (Chopra and Sodhi 2004), consequences

(Hendricks and Singhal 2003), vulnerability (Stauffer 2003 and Svenson 2000), and business

continuity (Zsidisin et al. 2005). The purpose of this research is to develop a framework for

identifying and evaluating the risk sources and the mitigation strategies that are suitable for a

company. We also critically analyze the structural properties of modern supply chains to

reveal vulnerability causing factors.

6

The increased emphasis on disruptions management is sometimes based on

postulation that present supply chains undergo higher risk of disruptions and associated

losses than those in the past. A 2006 McKinsey global survey revealed that one-third of

company executives believe that supply chain risks are the same or have reduced in the past

five years. Have the risks really increased or decreased? We address this by collecting and

analyzing past catastrophe and economic losses data so as to investigate for trends in the

losses and number of disruptions. We also investigate existing supply chains to identify

vulnerability-causing aspects that may have evolved in recent times.

We collected a large amount of data on man-made and natural catastrophes and their

associated economic losses from a wide variety of sources, which include research articles,

Department of State, Center for Research on the Epidemiology of Disasters, Federal

Emergency Management Agency, U.S. Department of Commerce, National Climate Data

Center, and National Counterterrorism Center. Using this data we contribute to the literature

by statistically confirming the speculation that the number of supply chain disruptions are

increasing. Our analyses suggest that economic losses are increasing at a fast rate. Data

compilation also revealed geographical and chronological inferences on catastrophe

concentrations and the need of mitigation strategies. The findings can potentially help

managers make robust decisions about location of facilities and suppliers, transportation

modes, contingency measures, inventory policies, and levels of flexibility and redundancy.

Supply chains have evolved to maximize efficiency and speed, motivated by the

desire of businesses to compete cost effectively in today’s markets. This focus on efficiency

has resulted in supply chains that are more vulnerable to disruptions. We critically analyze

various common supply chain practices, designed for greater efficiency, to reveal their

7

unfortunate consequence of increasing vulnerability. We argue that resilient supply chains

can be created by managing these vulnerability-causing factors.

Different catastrophes have widely varying consequences for supply chains.

Managers cannot anticipate all. We propose a classification framework that outlines possible

effects and the severity with which different supply chain components are affected by

different catastrophes. This helps provide a greater understanding of causes and effects of

disruptions.

After the 9/11 terrorist attacks, research has focused primarily on increased security

measures to reduce both the occurrence of terrorist attacks and decrease supply chain

disruptions (Rosoff and von Winterfeldt 2006; Sheffi 2003; Rice and Caniato 2003; Lee and

Wolfe 2003). However, security cannot eliminate the possibility of a terrorist attack or help

avoid or manage natural catastrophes. Supply chain disruptions are unavoidable (Craighead

et al. 2007). Therefore, in addition to catastrophe avoidance, supply chain managers also

need strategies and tactics for catastrophe mitigation.

We provide what we think is the most complete set of strategies to date to avoid or

mitigate a wide variety of catastrophes. Proactive strategies are identified that can help a

company avoid or decrease the possibility of certain types of disruptions. To gain benefits

from advance information, advance warning strategies that can help forecast a catastrophe

are proposed. Flexibility and redundancy in various supply chain components help define

coping strategies to help mitigate catastrophes.

Recent research has investigated the effectiveness of various mitigation strategies

(Tomlin 2006; Albeniz and Simchi-Levi 2005; Swinney and Netessine 2008). To effectively

manage disruptions, a company may need to employ several mitigation strategies. Moreover,

8

some mitigation strategies can be effective against several different catastrophes. We develop

a qualitative framework to identify the appropriateness and effectiveness of various

disruption mitigation strategies to different catastrophe types. The framework can be an

effective and practical tool for identification of best mitigation strategies. Our framework

focuses on a typical manufacturing-oriented supply chain that has suppliers/manufacturers

located overseas, distributors/retailers/customers located in the U.S., and uses sea and land

ports to import raw materials and finished goods into the U.S. Despite our focus, many of the

findings and insights are applicable to a wide variety of supply chains. Service industries can

also benefit from the findings.

Common belief is that mitigation strategies are expensive and inefficient. By drawing

parallels from the quality research of the 1970’s, we emphasize that well-developed

strategies can increase efficiency. In particular, similar to Philip Crosby’s notion that “quality

is free,” we reveal some efficiency-improving effects of mitigation strategies.

The plan of the chapter is as follows. In Section 2.2, catastrophe data is analyzed.

Section 2.3 describes how current practices in supply chains cause vulnerability. A

catastrophe classification scheme is provided in Section 2.4. Various catastrophe mitigation

strategies are discussed in Section 2.5. Finally, Section 2.6 identifies future research areas

that should be addressed by researchers and managers to encourage robust supply chains.

2.2 Catastrophes

Natural and man-made catastrophes disrupt supply chain operations. In this section, we

present and analyze terrorist events, accidents, and natural calamities data.

9

2.2.1 Intentional Acts

Disruptions in supply chains can be caused by conscious acts by a person or a group.

Depending on the intention, these acts can be classified as terrorist or non-terrorist. The

statistics presented in this section are compiled from data published by the National

Counterterrorism Center and 13 annual reports on Patterns of Global Terrorism published by

the Department of State in the years 1991 through 2003. From 2004 onwards, reports on

Patterns of Global Terrorism were replaced with Country Reports on Terrorism reports. In

these new reports we did not find enough relevant information to be useful for this research.

Terrorist Acts

Typical terrorist organizations of the past (before the 1990s) operated in small areas, with

little impact on the economy. In contrast, today’s terrorist organizations threaten both the

international community and supply chains. Using statistical data, we analyze terrorist

attacks worldwide and on U.S. interests. We define “U.S. interests” as U.S. residents,

infrastructure, and businesses that are partially or fully owned by U.S. companies. We also

study the attacks on targets that are partially or fully owned by businesses. We refer to such

targets as “business interests.”

Figure 2.1 shows the total number of terrorist attacks worldwide between 1993 and

2003. To our surprise, the figure does not show any clear trend, indicating that the risks to

supply chains from terrorist events may not be increasing. However, examination of terrorist

targets revealed that from 1991 to 2003, 62% of total attacks were on business interests

(Figure 2.2). In 2003, of the total terrorist attacks worldwide, 50% were on U.S. interests

(Figure 2.3). From 1996 to 2001, the percent attacks on U.S. interests relative to total

10

worldwide attacks increased steadily. In 1996, for every 100 attacks on U.S. interests, 67

were on business interests. The percentage steadily increased to 94% in 2001.

Figure 2.2. Percentage attacks on various classes of targets from 1991 to 2003.

Figure 2.1. Total number of attacks worldwide.

Business62%

Diplomat9%

Government4%

Military2%

Others

23%

11

Figure 2.3. Percentage attacks on U.S. business interests to total attacks on U.S. interests.

After 9/11, many businesses around the world and in the U.S. increased security

around their premises and transport carriers. The investments paid off. In 2002 and 2003,

attacks on business interests decreased (Figure 2.3). The total number of attacks also

decreased sharply in 2002 and 2003 (Table 2.1).

Table 2.1. International terrorist attacks by region.

Even with no apparent trend in the number of terrorist attacks, statistics show that

businesses are increasingly preferred terrorist targets. Businesses give disaster preparedness a

International terrorist attacks by region

Year Africa Asia EurasiaLatin

AmericaMiddle

EastNorth

AmericaWestern Europe

1996 11 11 24 84 45 0 1211997 11 21 42 128 37 13 521998 21 49 14 111 31 0 481999 53 72 35 122 26 2 852000 55 98 31 192 20 0 302001 33 68 3 201 29 4 172002 6 101 8 46 35 0 92003 6 80 2 20 67 0 33

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

1996 1997 1998 1999 2000 2001 2002 2003

Per

cen

tag

eAttacks on non-UStargetsAttacks on US non-business interests

Attacks on USbusiness interests

12

low priority (Helferich and Cook 2002). Until recently the data also shows that the U.S., and

specifically U.S. businesses, are highly vulnerable to terrorist attacks. Since 2001, despite

investing over $500 million on homeland security, the U.S. remains extremely vulnerable to

terrorist attacks (Flynn 2004).

To understand the geographical spread of terrorist targets, we collected data on

terrorist attacks in various regions of the world, summarized in Table 2.1. North America is

relatively safe for supply chains as it suffered less than 1% (19/2463 =0.7%) of the total

terrorist attacks that happened throughout the world. However, outsourcing and globalization

make American companies susceptible to disruptions that happen elsewhere. U.S. supply

chains are increasing outsourcing operations to Asia, which has seen a many-fold increase in

number of attacks from 1996 to 2003. Since the early 1990s, all forms of violence, except

terrorism, have decreased over 40% (HSC 2005). In 1997, the Department of State

designated 30 organizations as terrorist organizations. Since then the number has steadily

increased to 43 in 2007.

With the increasing vulnerability, supply chains need strategies to mitigate risks

originating from terrorist attacks. After 9/11, many businesses around the world increased

security around their premises and their transport carriers. The investments paid off. In 2002,

attacks on business interests decreased considerably, from Figure 2.3. The total number of

attacks also decreased sharply in 2002 and 2003, from Figure 2.1.

Non-Terrorist Intentional Acts

Helferich and Cook (2002) identified union strikes, lifestyle changes, government spending

shifts, economic downturns, competitive service improvements, hostile corporate takeovers,

changes in operations technology, and changing product technology as intentional human

13

acts that disrupt supply chains. Management manipulations by Enron and accounting

scandals in WorldCom showed the extent of disruptions caused by non-terrorist intentional

acts.

Although some of these non-terrorist intentional acts happen over time (lifestyle

changes, operational technology advancements), others such as strikes or management

problems can happen suddenly. In September 2000 in the UK, fuel prices rose sharply

resulting in an impact on almost every industry (Chapman et al. 2002). The event severely

disrupted over 34% of the companies in the UK (Peck and Juttner 2003). After 9/11, only

21% of companies suffered severe disruptions. This suggests that relatively small unintended

disruptions in common markets (fuel price rise) can cause disruptions in supply chains more

severe than dramatic intentional acts such as 9/11.

2.2.2 Accidents

Accidents are unintentional man-made catastrophes. Examples include Three Mile Island,

Chernobyl, Union Carbide, Bhopal, and Exxon Valdez. More recently, a series of blackouts

in the U.S., Italy, and the UK, caused by accidents at power generating stations, revealed the

vulnerability of supply chains to disruption in utilities. For U.S. industries, the blackout of

August 2003 was the most disruptive event after 9/11 (Zwirn 2003). The automobile industry

was paralyzed. More than 50 assembly plants operated by GM, Ford Motor Co., and Daimler

Chrysler were shut down. Wal-Mart shuttered about 200 stores.

The data collected from the Centre for Research on the Epidemiology of Disasters

shows that during the 20th century, the number of technological and transport accidents that

affected a large population and resulted in declarations of emergency has increased from 1

per year in the first 3 decades to 220 in the last decade. See Figure 2.4. Along with the count,

14

the resulting economic damages have also increased many-fold. Over the years, industrial

systems have increased in number. With advancements in technology, the complexity of

industrial systems increases, which in turn increases the potential for breakdowns (Perrow

1984). Supply chains must effectively cope with disruptions whether they are intentional or

accidental.

0

50

100

150

200

250

1901

-191

0

1911

-192

0

1921

-193

0

1931

-194

0

1941

-195

0

1951

-196

0

1961

-197

0

1971

-198

0

1981

-199

0

1991

-200

0

Num

ber

of a

ccid

ents

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Dam

age in billion US

dollars

Average yearlyaccidents

Average yearlylosses

Figure 2.4. Average yearly accidents and the yearly damage caused.

2.2.3 Natural Calamities

Natural catastrophes have always been a threat for humans and their property. Similar to

terrorist attacks, early natural catastrophes (before the 1990s) had strong regional but

relatively limited global effects. Recently, hurricane Katrina, Asian tsunami, SARS, and Bird

Flu resulted in severe global economic consequences. SARS disrupted businesses worldwide.

It caused huge economic damages in the Asia-Pacific region. Australia, with no diagnosed

case of SARS, suffered an estimated loss of $1 billion (O’Malley 2003).

15

To reveal trends in natural catastrophes, we compiled and analyzed data maintained

by the Centre for Research on the Epidemiology of Disasters. Figure 2.5 summarizes the

findings. In the last 100 years, the frequency of natural catastrophes reported has increased

over 40-fold. During the same period, the economic damages have also increased

considerably. The increasing trend in number and damages is much more prominent in the

last quarter of the 20th century. The trend can be attributed partly to widespread geographic

human presence and advancements in information technology. Nevertheless, the increasing

trend is alarming because it shows the vulnerability and economic and human losses that can

be caused by natural catastrophes.

Figure 2.5. Average natural catastrophes reported and the yearly economic losses caused.

Major natural catastrophes can cause economic disasters; however, the smaller

weather events cause the most damage to supply chains. In the U.S., heat, tropical storms,

floods, severe weather, blizzards, wild fires, and ice storms cause over 90% of the total

economic losses from natural catastrophes (Ross 2003).

16

Unlike terrorist attacks, natural disasters in the U.S. are increasing. Data collected from

various publications from the U.S. Department of Commerce, National Climate Data Center,

Federal Emergency Management Agency, and Lott and Ross (2006), compiled in Figures 2.6

and 2.7, shows that U.S. supply chains faced a greater threat from disasters in the 1990s than

in the eighties. In the nineties, both the numbers of disasters, as well as those resulting in

losses of over a billion dollars, were considerably higher.

The distribution of population and the infrastructure has further increased the risks.

According to the U.S. Census Bureau, the population of coastal areas in Florida, which are

highly vulnerable to hurricanes and tropical storms, has increased from one million in 1940

to over 10 million in 1990. In 2001 in the U.S., about 48 million people live on the hurricane-

prone coast line. By 2010, the population in these areas is expected to rise to 73 million

(Helferich and Cook 2002). Similarly, in many earthquake-prone areas such as San Francisco

and Tokyo, population has increased steadily.

Figure 2.6. Total natural disasters in the U.S.

17

Figure 2.7. Total number of disasters in the U.S. with economic losses exceeding one billion

dollars.

Occurrence of natural calamities is highly erratic. From 1940 to 1960, the U.S.

suffered 18 major hurricanes. Surprisingly, for the next 30 years, only one major hurricane

struck the U.S. However, natural phenomena are cyclic (Helferich and Cook 2002). A return

of active hurricane seasons, along with increased population along the coast, provides

conditions that may result in severe economic disasters. Pielke and Landsea (1998) state that

“it is only a matter of time before the nation experiences a $50 billion or greater storm, with

multi-billion dollar losses becoming increasingly more frequent.” Economic losses from

hurricane Katrina were over $125 billion (Finkle 2005). Former UN Secretary General Dr.

Annan says, “A wide variation in the number and intensity of natural hazards is normal and

to be expected. What we have witnessed over the past decades, however, is not nature’s

variation but a clear upward trend” (Annan 1999).

18

2.3 Factors that Cause Vulnerability in a Supply Chain

Traditionally, supply chain designs primarily focus on cost efficiency (Nahmias 2009). Also,

supply chains and business practices are designed for a stable world (Monahan et al. 2003).

This has resulted in supply chains that are vulnerable to disruptions. We identify four

vulnerability causing factors, which when altered by a supply chain structural change or

decision may affect vulnerability. These factors and their association with modern supply

chain management practices are summarized in Table 2.2.

Number of exposure points: Raw materials, before reaching a customer in the form of

finished product, may travel through various geographic/political regions, change

ownerships, and are transported by different modes of transportation. All of these are

potential points where a supply chain is exposed to disruptions. For example, a product that is

manufactured in Asia and shipped to customers in the U.S. is susceptible to man-made and

natural catastrophes that may happen in the U.S. or Asia. Similarly, probability of disruptions

from causes such as bankruptcies, worker strikes, and accidents may depend on the number

of hands (ownerships) a product passes through.

Table 2.2. Supply chain practices and their effect on vulnerability causing factors.

Vulnerability Causing FactorsIncrease in the

Number of Exposure Points

Increase in Distance/Time

Decrease in Flexibility

Decrease in Redundancy

Sup

ply

Cha

in

Man

agem

ent P

ract

ices Globalization X X

Decentralization X XOutsourcing X XSole Sourcing XJIT X XProduct/Process Complexity XLitigation X

19

Modern supply chain trends such as globalization, decentralization, and outsourcing

make supply chains efficient at the cost of increase in the number of exposure points.

Globalization, which includes offshore sourcing, manufacturing, and assembly, increases the

number of geographical regions that a product passes through, thereby increasing the number

of climatic and political catastrophes that may disrupt a supply chain. Similarly, outsourcing

may increase vulnerability by increasing the number of ownerships. Decentralization

increases the number of companies involved in a supply chain. These companies can be

located far apart, increasing geographical and political exposure.

With increasing complexity of products and processes, today’s supply chains are not

always a linear chain of companies, but often a complex web of interconnected companies. In

order to achieve efficiency, this complexity is motivating supply chains to transform into

systems that are decentralized (to achieve expertise and efficiency) and use global sourcing,

which in turn increase the vulnerability by increasing the number of exposure points. Recent

Homeland Security measures such as Customs-Trade Partnership Against Terrorism,

Container Security Initiative, and the 24 hour manifest rule have served to introduce

additional stages and transactions, thereby increasing the vulnerability.

Distance/time: Outsourcing, globalization, and decentralization also add to the vulnerability

of a supply chain as they increase the distance (and/or time) the material take to travel

between the supply chain echelons. A shipment made over the ocean from Asia to the U.S.

may take 45 days. The shipment can be disrupted by a catastrophe that happens during any of

these 45 days. In contrast, a shipment within the U.S. may take only a day or a week. The

increase in the distance or time makes the control and coordination of a supply chain

difficult, making it easy for disruptions to occur.

20

Flexibility: Other supply chain practices such as sole sourcing increase the vulnerability by

decreasing flexibility in the supply chain. Using very few (or sole) suppliers can create a

dependence on suppliers. UPF Thompson became insolvent at the end of 2001. The impact

on Land Rover, which had UPF as its only supplier of chassis, was sudden and severe as it

had to lay off 1,400 workers at its Solihull UK plant (Chapman et al. 2002).

Redundancy: JIT and lean policies have intuitive economic rationale. However, the savings

has a cost. The savings are achieved by reducing buffer and redundancy, which increase

vulnerability as supply chains lose the ability to absorb unusual system disturbances. “If a

company … zealously go[es] after efficiency [to] be lean, having less inventory, … and just-

in-time manufacturing, then I think it is definitely more vulnerable to the major disruptions

that can be man-made or natural” (Lee 2001).

The four vulnerability causing factors identified can be helpful in assessing a supply

chain’s vulnerabilities. They can also be used to understand the impact of various supply

chain decisions on their vulnerability.

2.4 Catastrophe Classification

For a supply chain, there are different types of catastrophes that can occur, each with a

different amount of risk and with different consequences. For example, a biological attack

can have the severe consequence of inducing large human losses, while a winter storm may

affect supplies. Also, different catastrophes (for example, a hurricane and an earthquake)

may have similar consequences. It is difficult for companies to plan for every catastrophe. A

classification of catastrophes may help plan mitigation strategies.

21

Figure 2.8. Components at various stages of a supply chain.

SUPPLIER

Equipment

Lab

or

Util

ities

Com

mun

icat

ion

Fin

ance

/Ban

ks

Law

s/R

egul

atio

n

Raw Material MANUFACTURER

Equipment

Lab

or

Util

ities

Com

mun

icat

ion

Fin

ance

/Ban

ks

Law

s/R

egul

atio

n

Value AddedDISTRIBUTOR

Equipment

Lab

or

Util

ities

Com

mun

icat

ion

Fin

ance

/Ban

ks

Law

s/R

egul

atio

n

Value AddedCUSTOMER

Equipment

Lab

or

Util

ities

Com

mun

icat

ion

Fin

ance

/Ban

ks

Law

s/R

egul

atio

n

Value AddedRaw Material

21 21

22

A catastrophe disrupts a supply chain by affecting one or more of its components.

Components “internal” to an organization, such as equipment and facilities, may remain

unaffected by an external catastrophe (a catastrophe that does not affect the premises of the

organization). “External” components, i.e., supplies, transportation, labor, utilities,

communication, laws/regulations, and finance/banks, are typically affected by catastrophes

that happen outside of an organization. In contrast, an internal catastrophe, such as fire and

industrial accidents, may affect internal components only. In Figure 2.8, internal components

are represented inside the boxes denoting various stages of a supply chain. External

components are shown by arrows.

In Table 2.3, we identify the severity and possibility of effects on components

resulting from various classes of catastrophes. The severity and possibility correspond to

“how much” and “how likely” a catastrophe might affect a component. Table 3 can be used

to help assess the vulnerability of various components in the event of a specific catastrophe.

It is important for a company to focus on mitigating catastrophes that have a high possibility

and severity of affecting critical components. For example, Table 2.3 shows that terrorist

attacks on infrastructure, cyber terrorism, and infrastructure destruction by a natural

catastrophe have a high possibility as well as severity of disrupting communication media.

Thus, a call center may benefit by adopting migration strategies that alleviate the effects of

these three types of catastrophes.

The threat posed by a catastrophe depends on company specific factors such as

industry, geographic location, political situation, culture, location of suppliers and customers,

economy, and the crises preparedness. Depending on these factors, the entries in Table 2.3

23

may vary for different companies. Our table depicts a typical manufacturing oriented supply

chain.

Besides severity and possibility of effects on components, various other useful

classification schemes can be devised. “Preparation time”, i.e., the time between the forecast

or awareness and the occurrence of a catastrophe could be important in mitigation planning.

A hurricane can be forecast, to a certain degree of accuracy, days in advance. However,

earthquakes do not give preparation time. The “length of disruption” is another important

factor that is critical for supply chains. Sheffi (2002) classified disruptions from catastrophes

based on supply chain failure modes. He considered disruptions in supply, transportation,

facilities, communications, demand, and freight breaches as possible modes of failure.

Table 2.3. Severity and possibility of effects of a catastrophe.

Classification of Catastrophes Examples

Transportation Utilities Commu-nication

Suppliers Custo-mers

Labor Laws Finance

Seve

rity

Poss

ibili

ty

Seve

rity

Poss

ibili

ty

Seve

rity

Poss

ibili

ty

Seve

rity

Poss

ibili

ty

Seve

rity

Poss

ibili

ty

Seve

rity

Poss

ibili

ty

Seve

rity

Poss

ibili

ty

Seve

rity

Poss

ibili

ty

Ter

rori

st

Attack on Infrastructure Power and communication services

M M H H H H L M L M L L L L L L

Public services, i.e., hotels, banks, and distribution services

L L L L L L L L M M M M L L L L

Violence, Mass Killing, Ethnic Killing

Bombing public placesM M L L L L L L H L H L M L L L

Biological, Chemical, and Nuclear Terrorism

Sarin gas, anthraxH H H M H M H H H H H H H H H H

Hoax or Propaganda Intended to Terrorize

BombingH L H M L L M M H L H L M L M L

Political Assassination H L L L L L L L L L H M H H L LSabotage of Transportation Media

Airplane bombing/hijacking, pirates, train derailment, bombing airplane/rail terminals

H H L L L L H L H L M L M L L L

Cyber Terrorism Computer viruses H L H H H H H L H L L L H L L LWar Gulf war H L L L L L H L H L L L H M L L

Nat

ural

Infrastructure Destruction Earthquakes, hurricanes, floods H H H H H H H L H L H H H L H MTransportation Disruption

Dust storms, blizzards, storms, snow, hail, rain, avalanches, winds, erosion, landslide

H H M L L L H L H L M L L L L L

Health Hazard Epidemic, famine L L H L L L H M H M H H H L L LExtreme Weather Cold wave, extreme temperature H L M L L L M L M L H L L M L LNatural Fires Eruption, volcano, forest fires H L H L M L H L H L M L L L L L

Acc

i-de

nt

Industrial Accidents Gas leakage M H M L M L H L H L H H M H L L

Transport Accidents Train derailment, airplane crash H H M L L L H L H L M L L L L L

Non

-te

rror

ist

Strikes Workers strikes L L L L L L L L L L H H H L L LPolitical strikes H H H L H L H M H M H H H L L L

Environmental Changes in government spending, lifestyle, manufacturing technology.

L L H L M L H L H L L L H L M L

24

2.5 Catastrophe Mitigation

Considering potential losses from disruptions, mitigation strategies could be critical for

survival of a company. Despite the risks, at times, 95% of Fortune 500 companies are not

equipped to manage a disruption that the company has not experienced before (Mitroff and

Alpaslan 2003).

Planning for robustness requires identification of critical components that a company

is excessively dependent on, or those that, if disrupted, can have a severe impact on supply

chain performance. Intel performs “what if” drills to identify the components and activities

that are excessively vulnerable (Lund 2002). Monahan et al. (2003) identifies the following

five characteristics of a vulnerable component: a bottleneck that other processes depend on, a

high degree of concentration of information flow, single or scarcity of suppliers, limited

alternatives, association with high risk geographic areas, and insecure access to important

infrastructure. In this section, we identify mitigation strategies that can help make supply

chains robust. Sections 2.5.1 through 2.5.3 identify mitigating strategies. In Section 2.5.5, we

discuss cost-benefit tradeoffs in implementing mitigating strategies.

2.5.1 Proactive Strategies

Feasibility and cost permitting, an organization should choose strategies that make supply

chains unaffected by any or many catastrophe(s). This section identifies proactive strategies,

which are decisions/plans/actions that are aimed towards reducing the vulnerability and

probability of disruptions. These strategies help reduce the number of exposure points, as

defined in Section 4. They can also reduce the effects of a catastrophe on a supply chain.

Locate facilities at “safe” locations. The frequency and type of catastrophes vary across

geographical regions. Asia is more prone to earthquakes than Europe (Jones 1981). Most

25

hurricanes occur between the latitudes of 30 degrees North and South (Alexander 1993).

Terrorist attacks occur more in some countries than others. Risks of disruption can be

considerably reduced by choosing locations that are less susceptible to catastrophes. Besides

geographic location and country, other factors such as ownership, symbolic importance, type

of construction, neighbors, government, community, and economic situation affect the

probability of a catastrophe. After the 1993 World Trade Center bombings, Morgan Stanley

realized the risks and moved its offices to several locations outside the towers. The strategic

decision paid returns on September 11, 2001.

Choose robust suppliers and transportation media. A supplier that is well-prepared to

cope with catastrophes can reduce the vulnerability of the entire chain. Often transportation is

the most vulnerable part of a supply chain. Selecting a transport company that has the ability

to handle disruptions can provide stability during catastrophes.

Establish secure communication links. Reliable and robust communication links can help

control and coordinate operations of a dispersed supply chain. With decentralized and global

supply chains, the need and benefits of communication links are significant.

Enforce security. Security can help prevent some intentional man-made catastrophes.

Information security can prevent cyber attacks by hackers, computer viruses, and

unauthorized access to communication media.

Efficient human resource management. Understanding employees can have huge returns.

For example, background information can prevent the hiring of workers that have a criminal

background. Moreover, such information can be used for assigning critical and sensitive

responsibilities. Coutu (2002) emphasized worker resilience as a determining factor in their

26

performance during catastrophes. Finally, a close understanding and relationship with

workers can help avoid strikes and production stoppages.

2.5.2 Advance Warning Strategies

An advance warning or forecast of a catastrophe can provide a company valuable preparation

time to align its capabilities to minimize disruption effects. Besides better catastrophe

mitigation ability, foresight can provide strategic advantages. In 2000, by watching their

supplier processes, Nokia increased its market share by 4%. Philips, a chip manufacturer,

suffered a fire in its Mexico plant. Nokia, anticipating the potential disruption, responded fast

to contact Philips to use its alternate facilities to meet Nokia’s demand. Ericsson was late. All

available capacity of Philips was taken by Nokia (Chopra and Sodhi 2004).

Enhance visibility and coordination in a supply chain. Organizations in a supply chain are

vulnerable to catastrophes that can affect any stage of a supply chain. Vertical coordination

can help prevent a catastrophe from disrupting multiple stages. Sharing information can help

companies anticipate a brewing problem at a supplier or customer. Horizontal coordination

can also allow companies (even competitors) to forecast disruptions such as a change in law,

shifting customer preferences, and changes in technology. Vendors, such as I2, Manugistics,

ViewVelocity, and Celarix offer specialized software to enable extensive visibility across a

supply chain.

Increase transportation visibility. Prompt information about transportation disruptions can

allow managers to prepare supply chains in a better way, thereby avoiding disruptions at their

facilities. For example, transportation vehicles can be rerouted through alternate routes,

orders at other suppliers for which transportation routes are undisrupted can be increased, and

orders can be expedited.

27

Monitor weather forecasts. Toyota uses WeatherData Inc., a weather forecasting company,

to monitor weather conditions for all of its 330 suppliers and transportation routes. Because

of lack of forecasting and planning, in 1999 a snow storm disrupted production at Ford Motor

Company. Toyota’s plants were uninterrupted (Murphy 1999).

Act according to terrorist threat level. The homeland security advisory system announces

the terrorist threat level using five colors. For supply chains, an increase in threat level

increases delays at entry ports because of higher custom or border checks. It may also require

companies to follow strict rules, which can disrupt normal operations.

Monitor trends. Trends such as changes in customer preference, laws and regulations, and

technology can create disruptions. These disruptions can result in loss of market, increased

taxes, and increased competition. In most cases trends occur slowly, and provide time for

organizations to adjust. Other changes such as laws and regulation can happen suddenly.

2.5.3 Coping Strategies

Coping strategies provide a supply chain with the ability to mitigate the effects of a

disruption. These strategies are built on flexibility and redundancy in components, which

provide options that can allow a company to offset the losses in a part of a supply chain by

gains from available alternatives (options).

Maintain multiple manufacturing facilities with flexible and/or redundant resources.

Having multiple facilities in different geographical and political regions can reduce the

probability of simultaneous disruptions at multiple locations. Redundant or flexible resources

at these facilities can provide disruption mitigation ability. Volkswagen manufactures cars in

multiple countries, such as the U.S., Brazil, Mexico, and Germany. These are also its major

markets. These facilities have both the flexibility to produce different models as well as

28

excess capacity to meet demand fluctuations. Disruptions at a specific plant can be

compensated for by increasing production at other locations using resources that are

otherwise redundant (Simchi-Levi et al. 2001).

Jordan and Graves (1995) show the benefits of having multiple flexible plants.

Considering uncertain (normally distributed) product demand, they show that “limited

flexibility (i.e., each plant builds only a few products), configured in the right way, yields

most of the benefits of total flexibility (i.e., each plant builds all products).” Although the

paper limited the analysis and results to uncertain demand, it is intuitive that ‘limited’

flexibility should also be beneficial under various disruptions scenarios.

Carry extra inventory. Excess inventory can mitigate disruptions without affecting normal

supply chain operations. Lack of inventory at a supplier can result in a shortage of supplies at

organizations down the line in a supply chain. In addition to mitigating disruptions, extra

inventory provides the advantage of helping to meet day to day demand fluctuations.

Alternate sourcing arrangement. Nonavailability of an alternate supplier can considerably

increase the risk of disruption. During a disruption at a supplier, other un-disrupted suppliers

should have the capability and capacity to increase their output to meet production share

from the disrupted suppliers. Li and Fung, a Hong Kong-based garment manufacturer,

reserves manufacturing capacities at multiple suppliers. This strategy ensures the availability

of flexible capacity when their customers such as Gap, Disney, and Gymboree order various

different designs and quantities (Lee and Wolfe 2003).

Flexible transportation. Supply chains should have flexibility in using alternatives such as

air, ground, and sea transportation. Alternate transportation is more important for global

companies. Options such as alternative routes and expedited service are also important. After

29

September 11, Continental Teves, an auto industry supplier, avoided disruptions because of

air restrictions by using a contingency relationship with transport firms to supplement air

transport by land (Martha and Vratimos 2002).

Maintain redundant critical components. It is advantageous to have backup for critical

components that can be maintained with limited investment. Auto companies maintain power

generators that can run the plants. Intel maintains a redundant communication system.

Standardize various processes. A product with a standardized and well-documented

process can be easily processed at different facilities and by different workers. If production

at a certain plant (producing non-standard products) is disrupted, other non-disrupted plants

may not be able to substitute its production.

Redesign products for component and process commonality to pool risks. Inventory

pooling by designing products with common components across multiple products allows a

limited or the same set of facilities to satisfy demands for multiple products. Postponement,

mass customization, and centralized inventory management are other techniques that take

advantage of risk pooling by component commonality.

Influence customer choice. The ability of a company to motivate customers to buy what

they want to sell is important, both during normal operations and catastrophes. During the

Taiwan earthquake in 1999, Dell was able to steer customers to buy computer configurations

that Dell could make from the available components, by giving them either a free or cheap

upgrade.

Insure against various risks. Buying insurance for various components and types of

catastrophes is one option open to companies. Various supply chain components can be

30

insured against natural calamities, accidents, and theft. Some coverage includes loss of

assets, profits, extra costs, and expenses because of physical loss or damage to property.

2.5.4 Cost/Benefit Trade-offs of Mitigation Strategies

Investment in mitigation planning and implementation consumes resources. The challenge is

to find economically viable ways to reduce vulnerability. It is not clear as to which strategies

can reduce risks without hurting efficiency. During the quality movement of the 80s, it was

recognized that increased quality pays benefits (Lee and Wolfe 2003). Similarly, we argue

that improving robustness can also increase efficiency. In this section we identify the

potential benefits of mitigation strategies.

Reduction in lead time and lead time variability. Secure transport mediums can help

reduce the time needed at customs and checkpoints, while reliable suppliers can reduce the

need for inspection, accounting, and bookkeeping. Traceability of transport mediums can

also increase the predictability of delivery time, thus reducing the associated uncertainty and

variability.

Better inventory management. Increased visibility of supplier operations and transport

mediums can reduce the uncertainty in supplies. This along with reduction in lead time can

reduce the amount of safety stock needed. This can also help match demand and supply.

Efficient production planning and forecasting. Better inventory decisions and reliable

information about exact customer demand can increase the efficiency of production planning

and forecasting. Flexibility and redundancy can provide options for smoothing variations in

production because of demand fluctuations.

Reduction in the bullwhip effect. Increased information sharing and coordination can

reduce bullwhip effects (Lee et al. 1997). Reduction in lead times also reduces bullwhip

31

(Chen et al. 2000). A robust supplier can be expected to have higher reliability in fulfilling

orders at the right time and quantity. This implies a lesser probability of rationing demand

among customers, which in turn reduces bullwhip.

Increase in customer service. Companies can win customers by making products available

during catastrophes, when competitors may not be able to reach customers. A better

coordination between supplier, manufacturer, and retailer can also help in understanding and

meeting the expectations and choices of customers.

Better demand management. Coordination and visibility between supply chain

organizations can provide critical information for demand management. For example, price

and promotion decisions can be made based on availability of supplies and customer demand.

Companies such as Dell and Amazon dynamically change price depending on the supplies

and customer demand.

Modular products also help in demand management as such products use common

modules that assemble to form final products. Using modularity and postponement,

companies can reduce the number of components required to make a variety of final

products. Fewer components can reduce the effect of compounded service levels. Thus the

firm can achieve a high service level with limited inventory of components.

2.5.5 Selection of Mitigation Strategies

Mitigation planning requires identification of vulnerability causing practices and the

catastrophes that can affect a company and then choosing strategies to mitigate them. The

choice of appropriate mitigation strategies depends on various factors such as location,

market, culture, operations, suppliers, product and process characteristics, ownership, and

manufacturing type, to name a few. Such a complex dependence may make it difficult to find

32

the strategies that best suit a company. In this section we develop two tables that can be a

useful aid for managers in achieving this aim.

A specific mitigating strategy can help to address many catastrophes, while a

catastrophe can be mitigated by choosing from different strategies. Different companies

usually require a different set of strategies. In Table 2.4, we present the effectiveness

tendencies of various strategies against different catastrophes as high, medium, low, and very

low. Depending on the types of catastrophes a company faces, managers can use Table 2.4 to

help choose strategies that best fit their needs. For example, environmental catastrophes such

as changes in government regulations, customer preferences, and technological changes can

be addressed by using the strategies of building an ability to influence customer demand and

proactively monitoring trends and changes in the environment.

For the vulnerability-causing practices used by a company, managers can use Table

2.5 to choose suitable mitigating strategies. For example, a vulnerability resulting from

outsourcing can be managed by selecting suppliers located at safe locations, monitoring for

natural calamities, and using flexible transportation. There can be correlations between

vulnerability causes and the catastrophe types that affect companies. For example, many U.S.

companies are vulnerable because of outsourcing manufacturing to Asia (Section 2.3). We

also found that Asia has the fastest increasing rate of terrorist attacks (Section 2.2). Thus

companies outsourcing operations to Asia should choose mitigation strategies (from Tables

2.4 and 2.5), which are most effective for outsourcing and terrorist catastrophes.

Investment in mitigation planning needs justification by appropriate cost/benefit

analysis, which requires an estimation of losses and benefits in numerical quantities. In many

cases this is not easy to calculate.

33

Another difficulty is a lack of understanding of potential long-term effects. For

example, to mitigate the risk of disruptions, companies can increase inventory. This can be a

short-sighted strategy. Increased inventory may not be an economical alternative as it may

result in quality problems from waste of resources, higher rejection rates, and poor

management. Such a strategy may make a company worse off than a catastrophe. Therefore,

more research is needed in identifying the potential long term deleterious effects of various

strategies. Section 2.6 identifies such research issues.

2.6 Research Problems in Disruptions Management

This section describes important research areas on managing disruptions in supply chains.

We also provide references to existing research efforts in some of these areas.

Develop disruptions databases. A catastrophe database can aid in identifying the possibility

and severity of disruptions for a company located in a specific region. Complex operations

and dependencies between various businesses make it difficult to identify the catastrophes

that might affect a company most severely. The Inter-university Consortium for Political and

Social Research and the Center for Research on the Epidemiology of Disasters maintain

databases for natural catastrophes and accidents. The National Counterterrorism Center

maintains a database of all terrorist events worldwide. These databases do not explicitly

identify affected regions and industries. Information from such databases could be modeled

or used to provide useful results for industry risk mitigation.

34

Table 2.4. Interrelationship of strategies and catastrophe types.

Natural Accidents

Non-terrorist Terrorist

Cat

astr

oph

es

Infr

astr

uctu

re

Des

truc

tion

Tra

nspo

rtat

ion

Dis

rupt

ion

Hea

lth

Haz

ard

Ext

rem

e W

eath

er

Nat

ural

Fir

es

Indu

stri

al A

ccid

ents

Tra

nspo

rt A

ccid

ents

Str

ikes

Env

iron

men

tal

Att

ack

on

Infr

astr

uctu

re

Vio

lenc

e, M

ass

Kil

ling

,

Bio

logi

cal,

Che

mic

al,

Nuc

lear

Ter

rori

sm

Hoa

x or

Pro

paga

nda

Pol

itic

al A

ssas

sina

tion

Sab

otag

e of

T

rans

port

atio

n M

edia

Cyb

er T

erro

rism

War

Strategies

Exa

mp

les

Ear

thqu

akes

Hur

rica

nes,

flo

ods

Sto

rms,

win

ds,

land

slid

e, s

now

, ra

in, a

vala

nche

s

Epi

dem

ic, f

amin

eC

old

wav

e,

extr

eme

tem

pera

ture

Eru

ptio

n, v

olca

no,

fore

st f

ires

Gas

leak

age

Tra

in d

erai

lmen

t, ai

rpla

ne c

rash

Wor

kers

str

ikes

, po

liti

cal s

trik

esC

hang

es in

go

vern

men

t

spen

ding

, lif

esty

le,

te

chno

logy

Pow

er,

com

mun

icat

ion,

an

d pu

blic

ser

vice

s

Bom

bing

pub

lic

plac

es

Sar

in g

as, a

nthr

ax

Bom

bing

Air

plan

e bo

mbi

ng/h

ijac

king

, p

irat

es, t

rain

de

rail

men

t

Com

pute

r vi

ruse

s

Gul

f w

ar

Pro

acti

ve

Locate facilities at safe locations H H H L M H M L H L H M H L L L L H

Choose suppliers located at safe locations H H H L L M M L M VL H M H L L L L H

Choose robust suppliers H H H - H H H H L L H M L L L H L M

Choose robust transportation M M H - L M H H L - M M H L M H L L

Establish secure communication links H H - - - M L L - - M L H L L VL H -

Enforce security - - - - - - H M H - H H M H L H L -

Efficient human resource management - - - - - - H H H - H H M H - H H -

Ad

van

ce

War

nin

g

Enhance visibility and coordination M M H M M H H H M L L L H L M H L

Increase transportation visibility M M H L M H H L - H L L H L H H L

Monitor and react to weather forecasts - H H H H H - - - - - - - - - - - -

Act according to terrorist threat level - - - - - - - - - - - - - H - - - -

Monitor trends: customer preferences, regulations, and technology

- - - - - - - - - H - - - - - - - -

34

35

Table 2.4 continued. Interrelationship of strategies and catastrophe types.

Cop

ing

Maintain multiple manufacturing facilities with flexible and/or redundant resources

M M H L H M H H M - H L M L M H M M

Carry extra inventory M M H - H H H H H - H H L H M H L L

Secure alternate suppliers H H H - M H H H H - H H H M H H L H

Choose flexible transportation options M M H - VL M H H L - M M H L M H - L

Maintain redundant critical components M M - - L L L L VL - H L L L L - M L

Standardize/simplify processes M M M H M M M H M M H H L M M L M

Product design for product commonality, controlled architecture, and postponement

H H H M H H H H M H H M L H H L H

Influence customer choice H H H H H H H H H H H H H H H H H H

Insurance against various risks H H H VL H H H H H - H L H - VL M M L

High (H), Medium (M), Low (L), or Very Low (VL).

35

36

Strategies

Vulnerability Breeding Practices

Globali-zation

Decentral-ization

Outsou-rcing

Reduction in Number of Suppliers

JIT Practices

Supply Chain

ComplexityLegisla

-tion

Pro

acti

ve

Locate facilities at safe locations M M M L H M H

Choose suppliers located at safe locations H H H H H M H

Choose robust suppliers H M M H M - -

Choose robust transportation M M M M H H H

Establish secure communication links L L L - VL - M

Enforce security - - - - - - L

Efficient human resource management - - - - - - M

Ad

van

ce

War

nin

g

Enhance visibility and coordination H H L M H L -

Increase transportation visibility M L M M H L H

Monitor and react to weather forecasts L M H H H H -

Act according to terrorist threat level L L L - M - H

Monitor trends: customer preferences, regulations, and technology - - - - - - H

Cop

ing

Maintain multiple manufacturing facilities with flexible and/or redundant resources M M M M L - -

Carry extra inventory M M M H H M L

Secure alternate suppliers M M M H H M H

Choose flexible transportation options H H H H H VL L

Maintain redundant critical components - - - - H - -

Standardize/simplify processes H H M L L H -

Product Design for product commonality, controlled architecture, and postponement M L L L L - L

Influence customer choice - - - - - - -

Insurance against various risks M M M M M L -

Table 2.5. Interrelationship of strategies and vulnerability breeding factors.

36

37

Develop models to understand and estimate the impact of various catastrophes. A

catastrophe affects businesses differently, causing disruptions to varying extents. This makes

it difficult to predict the consequences of some catastrophes. For example, before the

outbreak of SARS in 2003, it was very difficult to predict its effects on the airline and

electronics industries worldwide. Empirical studies, using statistics of past disruptions, can

help identify the catastrophes that can affect a business.

It is also difficult to identify the possibility and impact of less-familiar disasters such

as a dirty bomb. One way to understand the effect of such an event is by identifying similar

catastrophes that have occurred in the past. For example, learning from the effects of the

Sarin gas attack in a Tokyo subway could help in assessing the disruptions caused by a

biological attack. ‘What if’ analysis can be used to predict the effects of a catastrophe. The

economic impact of disasters such as avian flu (Allen 2006), dirty bomb (Rosoff and von

Winterfeldt 2007), hurricanes (von Winterfeldt 2006), and earthquakes can be studied by

simulating such events (Allen 2006).

Establish supply chain risk measures. Catastrophe mitigation strategies aim to reduce the

effects from disruptions but can require significant capital investments. The justification for

investment in mitigation planning would benefit from a measurement of risks. For decades,

finance researchers have investigated various risk measures for capital investments. Risk

measurement from the perspective of supply chain disruptions is relatively new.

Supply chains could benefit by developing metrics that can estimate the risks in

aspects such as supply, operations, inventory, transportation, and location. Applequist (2000)

developed risk metrics for supply chain projects. Zsidisin et al. (2004) identify factors used

by purchasing organizations to assess risk. Cavinato (2004) defined risk from the perspective

38

of a logistics firm. Sodhi (2005) suggests operational measures to estimate demand and

supply risks.

Identify disruptions indicators. Such indicators could be an abnormal variation in supplies

or demand, change in fuel prices, disruptions in related industries, political changes, and

technological developments. An indicator of potential supply shortage is an increase in

Homeland Security’s threat level from yellow to orange, which often results in extended

delays for components and supplies entering the U.S. Once identified, a company can align

its strategies and resources, in advance, to mitigate disruptions.

Supply chain vulnerability causes. In Section 3, we identified various causes of supply

chain vulnerability. Peck (2005) identifies drivers of vulnerability in a supply chain.

Vulnerabilities for a specific company may depend on various factors such as the industry,

location, operating strategies, suppliers, customers, political situation, and government

policies. Identification of industry specific vulnerabilities could be helpful for managers in

making robust decisions.

Evaluate the effectiveness of prevalent supply chain practices under disruptions. As

argued in Section 2.4, present supply chain practices are sometimes motivated by cost

efficiency alone. For example, to save production costs, a U.S. based pharmaceutical

manufacturer moved its manufacturing operations to India. During an Indian port strike in

2003, this company had to fly their product to the U.S. for over a month. The strike cost them

more in increased transportation costs in one month than three years of production cost

savings (Mentzor 2008).

Strategies such as lean production, outsourcing, decentralization, expediting, and sole

supplier need to be analyzed considering the possibilities of catastrophes. Also the role of

39

supply chain coordination while considering disruptions should be investigated. Current

supply chain literature addresses coordination while considering variations in demand and

price. Prevalent coordination mechanisms do not hedge a supply chain stage against

disruptions. The effects of a catastrophe are not symmetrical in a supply chain. For example,

a relatively small disruption at a manufacturer may not affect a customer’s operations as the

lead time between these stages may provide a cushion to absorb the disruptions. However, a

disruption at a customer’s facility is more likely to affect the manufacturer because of

increased bullwhip (Schmitt et al. 2007).

Identify learning from past disruptions. Experience from past disruptions can teach

important lessons, which can be used to make strategic decisions. A compilation of such

strategies for many industries and catastrophes can help managers in choosing mitigation

strategies appropriate for their companies. Kleindorfer and Saad (2005) studied the chemical

industry, Sinha et al. (2004) examined the aerospace industry, and Johnson (2001) identified

mitigation strategies in the toy industry.

Develop operating policies that are effective under disruptions. Much research identifies

optimal inventory, location, transportation, production, and procurement policies while

assuming stationarity. Unexpected events are sometimes not considered. In general,

uncertainty increases the complexity of these analyses, making it difficult to identify

effective policies. Considering today’s environment where both the number and the severity

of disruptions are increasing, it is important to revisit supply chain operating policies and

make changes to improve efficiency under disruptions. Snyder (2006) modeled a traditional

facility location model with disruptions. Some researchers have investigated inventory

policies under disruptions (Chao 1987; Arreola-Risa and DeCroix 1998)

40

Develop mitigation strategies. Resilient supply chains can be created by careful selection

and implementation of mitigation strategies. Some research has identified advantages of

using multiple suppliers rather than a sole supplier. These papers compare two suppliers that

have different risk levels and production costs (Chopra et al. 2007 and Tomlin 2006).

Hallikas et al. (2004) provided a risk management framework for supplier networks. Zsidisin

and Ellram (2003) used survey results to suggest strategies to manage supply risks.

Redundant and flexible resources can be used to mitigate disruptions. The amount

and kinds of resources, however, can vary depending on many factors including industry,

location of facilities and markets, customer preferences, and laws, to name a few. Identifying

the amount and kinds of these resources is difficult. Sheffi and Rice (2005) provide

recommendations to increase flexibility to mitigate disruptions. Zsidisin et al. (2005) discuss

the effective use of business continuity planning to manage supply risk.

Impact of government policy decisions. Since 911, the Homeland Security Department has

implemented measures such as Customs-Trade Partnership Against Terrorism (C-TPAT),

Container Security Initiative, and the 24 hour manifest. These measures impact supply

chains, since material or supplies can be delayed at borders. There is a need to study the

effect of such measures on supply chains. Bakshi and Gans (2007) performed an economic

analysis of C-TPAT to understand the motivations of firms to be part of it. Tomlin and

Snyder (2007) studied the value and impact of the threat advisory system to suggest that

supply chain operating policies should be based on the level of threat.

Identify supply chain structures resilient to disruptions. Supply chain designs vary across

industry and sometimes across companies. Supply chains have been characterized along

several dimensions including industry type, competitive landscape, bargaining power of

41

suppliers, bargaining power of customers, degree of modularization, degree of product

postponement, number of new product launches, number of competitors supplying similar

products, ease of switching, and demand predictability. Are there certain factors that make a

supply chain more robust? There is a need for models to find combinations of the above

factors that lead to a resilient supply chain. Santoso et al. (2005) provides an algorithmic

approach to design supply chains considering uncertainty. They provide methodology to

compare alternate supply chain designs to identify designs that have the least uncertainty.

Such methods could be helpful in choosing robust mitigation strategies.

2.7 Conclusions

This research stresses the need for higher awareness of catastrophes and their effects on

supply chain operations. We compile catastrophe and economic data from many sources to

reveal information of managerial and academic importance. This thesis contributes to the

literature by statistically verifying the increase in both the number and $ value of

catastrophes that impact U.S. businesses. We also reveal the regional locations and spread of

various types of catastrophes. The information can be used by companies to make better

planning decisions. We contribute to the understanding of supply chain risk management by

identifying vulnerability-causing factors, which when affected by a business decision can

impact the vulnerability of a supply chain. These factors can be used to compare possible

decisions based on the vulnerability that they may cause, and thus help make better and

robust decisions.

We overview a comprehensive set of mitigating strategies that can help reduce the

risks of disruptions. The strategies presented can be used for proactively managing,

developing advance warnings, and coping from catastrophes. We reveal potential benefits of

42

following these strategies during normal times. These can help justify investment in

catastrophe mitigation. Businesses should focus on identifying strategies best suited for their

companies. We develop several tables that can be practical tools to help managers to identify

the vulnerabilities of their supply chains. The tables can also be used to identify mitigation

strategies that best suit their companies for a given set of possible disruptions.

We hope that this study and the research areas identified can provide impetus for future work

on understanding, developing, and quantitatively analyzing risk mitigating strategies. While

we focus here on qualitative strategies, future research should extend the scope to quantify

the costs and benefits for businesses.

43

CHAPTER 3

MODELING AND ANALYZING ECONOMIC CONSEQUENCES OF SUPPLY

CHAIN DISRUPTIONS

3.1 Introduction

Resilience to disruptions is a critical issue in supply chain management. Disruptions can have

many sources and affect many supply chain activities. The causes range from natural (e.g.,

Hurricane Katrina, SARS pandemic) to accidental (Minneapolis I-35W bridge collapse,

credit collapse of 2009) to intentional (World Trade Center attacks). While individual

disruptions have a low probability of occurrence, there is a good chance overall of something

big and unexpected to happen. The consequences of disruptions in manufacturing,

transportation, electric power, and telecommunications can be substantial and long lasting,

with rippling effects felt throughout multiple business sectors. An Accenture study (Beverly

and Rodysill, 2007), which polled 151 supply chain executives in large U.S. companies,

indicates that 73% of the firms experienced disruptions in the past five years. Of those, it

took 36% more than one month to recover; and another 32% between a week and a month.

Disruptions are costly as well. Hendricks and Singhal (2003) found that following a

disruption, firms on average experience a 107% decrease in operating income, 7% lower

sales growth, and 11% higher costs. The firms also suffered 33-40% lower stock returns over

a three-year period, and share-price volatility rose by 13.5% in the year after the disruption.

In later work, Hendricks and Singhal followed with additional evidence on financial

44

performance (2005b), and market performance (2005a, 2009). Many others have reported

costly consequences of disruptions (e.g., Shaw, 1994; Murphy, 1999; Latour, 2001;

Chapman, Christopher, Juttner, and Peck, 2002; Helferich and Cook, 2002; Martha and

Subbakrishna, 2002; Rice and Caniato, 2003; Monahan, Laudicina, and Attis, 2003; Peck and

Juttner, 2003; Zwirn, 2003; Ross, 2003; O’Malley, 2003; Chopra and Sodhi, 2004; Cavinato,

2004; Sullivan, 2006).

Many firms lack clear contingency plans and well-defined roles concerning

disruptions. Mitroff and Alpaslan (2003) analyzed crisis readiness of Fortune 500 companies

over the past two decades. They found 95% of them not prepared for an unfamiliar disruptive

event. Hillman and Sirkisoon (2006) and Hillman and Keltz (2007) provide further evidence

of poor preparation for disruptions.

Sandia National Laboratory has responded to such contemporary concerns by

developing a macro-economic simulation model of industrial activity, and our role in the

model development was to assist in characterizing supply chain behavior. The chapter is

organized as follows. We review prior art in operations management and economics. Then,

we formulate a set of key research questions, whose answers we glean from case studies and

simulation experiments. We propose a four echelon assembly structure as a baseline in

further research by Sandia and others. The results we report as well as subsequent ones by

Sandia (Appendix A) focus on issues of response and recovery rather than prevention. Our

simulation results confirm that disruptions may have long-lasting, rippling, costly

consequences within a supply chain, and that expediting efforts may hinder not help system

recovery. We also find that a system cost function can be quite ill-behaved, even in the

45

absence of disruptions and expediting, and that conventional solution approaches, which

assume convex or unimodal behavior, may be inappropriate for real-world supply chains.

3.2 Literature Review

Sandia approached us with an intriguing question: How to model the macroeconomic impact

of a major disruptive event to supply chains. Empirical findings by others and case studies

we conducted indicate that specific high impact disruptions are improbable with distributions

hard to quantify. We also found evidence that commonly used mitigations by firms included

inventory buffering and expediting.

There is a rich and growing body of empirical and analytical work on disruption

management. Craighead, Blackhurst, Rungtusanatham, and Handfield (2007) and Snyder and

Shen (2009) provide thoughtful literature reviews. Craighead et al. (2007) cite as examples,

recent work on supply chain risks (Chopra and Sodhi, 2004), vulnerability (Svensson, 2000),

resilience (Sheffi and Rice, 2005), and business continuity (Zsidisin, Ragatz, and Melnyk,

2005). They also address a related issue, supply chain severity. Kleindorfer and Saad (2005)

and Stecke and Kumar (2009) provide qualitative frameworks for types of vulnerabilities and

mitigation methods. Sheffi (2007) covers managerial implications of disruptions, and

approaches such as buffering, redundancy and agility as important means to achieve

resilience.

Most analytical work on disruptions relies on single-echelon models, with

exponential or geometrical failure rates and durations, and inventory buffering as mitigation.

For example, Berk and Arreola-Risa (1994) considered an inventory system governed by an

EOQ policy, Parlor and Perry (1995), a (q, r) policy, and Arreola-Risa and DeCroix (1998),

an (s,S) policy.

46

More recent literature assumes stochastically recurrent disruptions, order-up-to

policies in a single echelon maintaining inventory, and in most cases, shortages backordered,

not lost. These papers address lead time disruptions due to border closures (Lewis et al.

2008), threat level information and evolving risk (Tomlin and Snyder, 2009), supplier

selection and reliability (Tomlin, 2006), product mix and supply diversification (Tomlin and

Wang, 2005), facility location (Berman et al. 2007, Church and Scaparra 2007, Scaparra and

Church 2008, Snyder and Daskin 2005, Snyder et al. 2006, Qi and Shen 2007).

Snyder and Shen (2009) observed in a review of the earlier work that “Disruption

models are generally much less tractable than their deterministic-supply counterparts and

require numerical optimization since closed-form solutions are rarely available.” They also

offered simulation as an alternative to analytical study “to gain insights using realistic models

rather than to find optimal solutions to exact but vastly simplified models.” They explore a

series of experiments entailing various supply chain configurations, all employing a single

inventory echelon and a periodic review (s, S) policy, to study questions of order size, order

frequency and inventory placement in the presence of disruptions. They find that total

inventory, order and backorder cost is fairly well behaved, and they employ line search to

find reasonably cost effective inventory parameters in which to make comparisons.

Wu and Chen (2009) apply a two-echelon optimal control system to model oil

industry behavior. One echelon represents an individual firm, and the other, an aggregation

of the industry. Their model has echelon decision variables of price and quantity. They

assume no lead times, unimodal total cost of inventory and operation, and unlimited

production capacity. Shortages are not permitted since equilibrium prices clear the markets.

They consider two types of disruptions to supply and demand – a single shock, and random

47

Brownian shocks. Among other conclusions, they find that price and inventory peak after a

single shock and dissipate slowly, with delayed propagation to the other echelon.

Disruptions represent exogenous shocks to a system, while expediting offers

endogenous means to cover shortages that may arise either from disruptions or normal

operating conditions. In an HP supply chain, Beyer and Ward (2002) reported over 65% of

orders expedited. Other examples from diverse industries include Amazon (Kelleher, 2003),

Caterpiller (Rao, Scheller-Wolf and Tayur, 2001), and Nintendo (Souder, 2004). Case studies

we will summarize shortly also indicate that shortages of relatively inexpensive electronic

parts motivate the frequent use of expediting to avoid cancelation or delay of substantial

revenue for value added assemblies.

While expediting appears rational, the effects of expediting across a supply chain are

not well understood. Research on expediting has focused on finding optimal ordering policies

at a single inventory echelon (ex. Fakuda, 1964; Whittmore and Saunders, 1977; Groenevelt

and Rudi, 2003; Huggins and Olsen, 2005; and references therein), although progress has

been hampered, in part, because of difficulties in analyzing crossover orders not arriving in

the sequence they were placed (Robinson et al. 2001; Hadley and Whitin, 1963; Nahmias,

1979; Zipkin, 1986). For analytical tractability some researchers avoid crossover by

restricting regular and expedited lead times to differ by one time unit or with instantaneous

expedited delivery (Fakuda, 1964; Lawson and Porteus, 2000; Huggins and Olsen, 2005).

Others apply heuristic policies that approximate optimality under generalized lead times

(Veeraraghavan and Wolf, 2008).

A worthy contribution to the disruption and expediting literature is the consideration

of well known bullwhip effects (Chen and Lee, 2009). With dependencies in material,

48

finance and information, variation in demand and supply at one echelon offers potential to

induce and amplify at others within and among firms, and disrupt the whole chain (Lee et al.

1997). Bullwhip effects have been attributed to both erratic human behavior (Sterman, 1989;

Croson and Donohue, 2006) and rational decision making (Lee et. al., 1997). Chen et al.

(2001b) and Chen and Lee (2009) provide analytical treatment of rational bullwhip effects.

Much of the focus has been on a single or serial two echelon supply chain with a single mode

of ordering, i.e., no expediting (Towill, 1991; Wickner et al. 1991).

Some recent research on bullwhip effects has been experimental in nature. Wikner et

al. (1991) studies the impact of various parameters related to structure, information and lead

time on a three-echelon supply chain. Sterman (1989) presents experimental findings on the

traditional beer game (four echelon system). The game is replicated 48 times, with averages

compared using t-tests. Croson and Donohue (2006) replicate a four echelon system eleven

times, with averages presented. Regression with p-tests shows that participants undervalue

supply as compared with on hand inventory and demand. Chatfield et al. (2004) analyze

results of a four echelon system over 30 replications. The system is simulated for 700

periods, with first 200 for warm up. Analysis of variance is used to test the impact of

information sharing on demand amplification. They offer the following rationale for their

methodology, “Simulation modeling of supply chains can provide both realism and utility …

by accounting for the natural variations that occur in the various processes within the supply

chain, and that could not be captured analytically.” Simulation has been used as a powerful

alternative to analytics in analyzing complicated supply chains effects in other real word

settings (e.g., Bowersox and Closs, 1996; Closs et al. 1998).

49

3.3 Bridging Macroeconomic and Operational Decision-making

We considered more complex supply chain structures, order interactions and mitigation

methodology than we found in prior literature, in part, because of a different research

mission. In addition to contributing to theory and practice in operations management, another

objective was to guide the development by Sandia National Laboratories of a macroeconomic

model of regional industrial behavior. We were more interested in exploring elements of

reality than tractability. Rather than present explanatory findings on a tightly defined

problem, we explore inter-related research questions in a series of controlled experiments.

With a sense of urgency, a group of economists within Sandia embarked on a project

to develop a large scale simulation to assess the economic impact of disruptions in critical

infrastructure on U.S. manufacturing firms and their supply chains. They wished to study

vulnerabilities in complex, realistic operating environments, and gain insights about

mitigation strategies. The need for managerial relevance implied fresh methodology capable

of relaxing traditional economic assumptions of aggregation, substitutability and well

behaved performance of supply and demand. Aggregation may be inappropriate because it

tends to cancel effects of variability and associated amplification at other echelons. While

fitting as mitigation for disruptions caused by natural disasters and accidents, substitutability

may not be an option in the event of intentional terrorist acts. For example, Sheffi (2006)

suggested that the U.S. Government may respond to an event such as uncovering a dirty

bomb at one seaport of entry by closing all seaports for a lengthy period.

Using enormous computing power, Sandia began to develop an agent-based model

capable of simulating the discrete events of millions of entwined enterprises within regional

supply chains, and trace the corresponding economic behavior (Appendix A). They chose the

50

Pacific Northwest as an initial test site, and we were one of three groups to assist them. The

others were: Argone National Laboratories to understand power generation and distribution,

and Lucent Technologies to model the telecommunications.

Our role was to help Sandia span boundaries in this interdisciplinary effort by

conducting exploratory research into multi-echelon inventory systems. We focused on the

ordering systems across firms in supply chains, which link the procurement, production,

distribution, and transportation activities.

Our field and simulation research summarized here assisted Sandia’s development by

characterizing basic agent firm structures and inter-relationships, defining model parameters,

identifying key performance drivers, hypothesizing critical aspects of supply chain behavior,

and validating non-linear search methodology. We posited seven research questions:

1) What is a reasonable baseline for the supply chain structure?

2) What inventory and disruption logic reflect realistic conditions in a supply chain?

3) Which performance metrics drive the economics in the supply chain?

4) What happens to supply chain system performance in the presence of a disruption?

5) How does expediting affect system performance?

6) How well-behaved is system performance -- should analytics or heuristics be applied

in system ordering?

7) How do genetic and line search compare in terms of solution effectiveness and

efficiency?

By addressing these key modeling issues, we provide insights to Sandia in construction of

their macro-economic model, and offer guidance for future operations management research

on supply chain management.

51

3.4 Key Modeling Issues

We are interested in a supply chain structure that offers enough complexity to reflect a

realistic supply chain composition, yet small enough to control the environment in our

experiments as well as Sandia’s model. To address the key modeling issues, we considered

the supply chain literature, conducted case studies of three firms in the Pacific Northwest,

Sandia’s initial region of application, and conducted a series of simulation experiments.

3.4.1 Key Issue 1: A Reasonable Baseline for the Supply Chain Structure

Most surveys on disruptions involved large firms. In support of Sandia’s effort, however, we

targeted small and medium sized electronics manufacturing firms in the Northwest, related

supply and demand activities elsewhere in the U.S. and overseas, and corresponding

transportation connections. The case firms we selected employ between 250 and 500 workers

in a predominantly small-firm regional economy. Of manufacturing companies in the

Northwest, over 98% employ less than 500 employees (U.S. Census Bureau, 2008). While

small in size individually, these firms drive much of the region’s economic growth in

employment, profit, and demand for non-durables and services. For example, the

manufacturing sector in the Washington State economy represents 20 percent of overall state

output, and is forecast as the fastest growing sector over the next 10 years (REMI, 2007).

Small firms also tend to be more vulnerable economically to disruptions than their

larger counterparts. Because of limited cash reserves and working capital, small firms are

more likely to fail in the event of disruptions. Many do not have the resources to prepare for

prevention, response, and recovery from major disruptions. For example, the Institute for

Business and Home Safety (2005) found that after Hurricane Floyd, 30% of impacted small

firms never re-opened, and another 20% closed after two years.

52

Economic behavior in the Northwest covers many industries and activities, but for

our part, we focused on electronics companies for three reasons. First, electronics are

important to the functioning of our society. Logic chips can be found in products ranging

from automobiles to electric toothbrushes.

Second, electronics assembly is susceptible to disruption because of the complexity of

assembly and component supply. We found in three case studies that electronics assembly

requires from 70 to 700 components to make one product type, and a shortage of any

component delays the completion and sale of the product. Even under stable conditions,

extremely high buffer levels are needed to prevent such delays. For example, if a company

maintains a 99% service level (amount supplied/amount demanded) for each component, the

probability of having, say, 700 components available at any point of time is 700 .99 , roughly

equal to 0.009%.

Third, electronics supply chains involve global, multinational interests that broaden

the exposure to disruption. We found in the case studies that most electronic components are

internationally sourced. Additionally, some of the electronic assemblies are embedded into

larger systems made by such customers as Boeing and Honeywell, who in turn, export their

products.

Electronics companies in the Northwest make products ranging from consumer

appliances to devices used by original equipment manufacturers (OEMs). To gain insights

into supply chain behavior in that industry, we interviewed personnel from three

representative electronics firms and some of their suppliers (Appendix C). We disguise

identities of the three because their management considers certain information proprietary

53

and security-sensitive. We label the three as firms INT, ABC, and XYZ, and summarize the

cases in Table 3.1 with details in Appendix A.

All three supply chains support a minimum of four echelons, and at least one that

requires component assembly. Figure 3.1 shows our rudimentary supply chain. Echelon 1

represents assembly of components A and B, and storage of the corresponding finished goods

awaiting demand from local and out-of-town customers, which might include OEM

manufacturers, distributors, retailers and ultimate consumers. At Echelon 2, Stages 2A and

2B represent transportation and storage of the respective components. In Stages 3A and 3B,

suppliers (often local) transport the parts, and either stores them as finished components, or

first adds value prior to storage. Stages 4A and 4B represent activities of two out-of town

distributors that order, purchase and store the parts for delivery to Echelon 3. The figure also

displays lead time parameters -- lead times (LT) under normal operating conditions and

expedited lead times (ELT). These times are based on observations in the case firms and

precedent in the literature. We discuss the model logic in the next section and details of the

parameters in Appendix B.

We cannot generalize beyond electronics, but this four echelon structure seems

reasonable as a baseline for a manufacturing supply chain. Among supporters of a minimum

of four echelons, Juneja and Rajamani (2003) cite an electronics supply chain with assembly

that includes Selectron (supplier), Matsushita (manufacturer), Panasonic (distributor), and

Best Buy (retail customer). Across industries, most aforementioned simulation research on

bullwhip effects covers four echelons. Additionally, Closs et al. (1998) applied a four

echelon assembly structure in simulation experiments to demonstrate that information

sharing contributes to better customer service. To motivate the use of agent based modeling,

54

Swaminathan et al. (1997) propose multiple echelons and assembly. We felt comfortable as

well including assembly of at least two components, e.g., when observing objects within

sight anywhere, whether at home, in an office, in a vehicle or in a factory, one would likely

see products that involve assembly of at least two components.

INT ABC XYZ

Model

Supply chain span

Electronic Co. Finished Product

60 models with over 2500 configurations

200 different finished product types

Three product groups, each that includes a lot of customization

Electronic Co. Primary customers

Manufacturers, distributors, and retailers

Automotive, utility, military, and aerospace industries

Aerospace OEMs

Electronic Co. Sole Sourced Components

Approx. 10% Approx. 20% Approx. 20%

Operation Type Assemble-to-order Make-to-order Primarily make-to-order

Electronic Co. Operations Model

Lean JIT system using visual controls and Kanban for internal processes. MRP ordering for externally sourced components.

Toyota style JIT for internal processes. MRP ordering for externally sourced components.

MRP ordering internally and externally.

Management issues

High product variety, high inventory and overhead costs, and delivery problems

Expectations for high customer service

Product cost and delivery performance

Risk causes

Recent disruptions

Power failure, transportation accident, sole source disruption

West coast port lockoutTexas port closure, Land-line service failure

Notable consequences of disruptions

Electric power vulnerability. Loss of sole-source supplier may introduce delays of up-to two years.

Better equipped than the other two because of lesser number of components and SKUs. No telecom backups or contingency plans.

Vulnerable to power and telecommunications failure. No backup generator or contingency plans

Mitigation planning

Buffering. Expediting. Basic preventive measures such as fencing, guards, and lighting.

Buffering. Expediting. Alternate sources and routing. Backup generators.

Buffering. Expediting. Alternate sources for most components. Geographically dispersed locations.

Telecommunication/power failure, transportation accidents, and sole sourcing

Risk

man

agem

ent

Table 1 -- Case Summaries

Ope

ratin

g Po

licie

s

Global spread with overseas transport, and distribution centers located elsewhere in the US, suppliers primarily local, local electronics manufacturer, customers ranging from local to overseas.

A 4-echelon supply chain with assembly is sufficiently representative.

Supp

ly C

hain

Table 3.1. Case summaries.

55

3.4.2 Key Issue 2: Inventory and Disruption Logic

We addressed one aspect of Sandia’s simulation model, the multi-echelon supply-chain

ordering system, which regulates the goods flows and inventories. While there are many

aspects of the Sandia model not covered, the ordering system serves as a fundamental means

of linking activities between companies in a supply chain.

Personal interviews revealed that operating managers of the case firms and their

suppliers do not share point-of-sale data within the respective supply chains. With buffering,

frequent rescheduling and repeated expediting, the managers believed their processes

Figure 3.1. Prototypical supply chain.

56

encountered the full effects of demand amplification. Most were familiar with features of the

Beer Game. There was a strong consensus among these managers that the most realistic way

to reflect the effects of disruptions is through local planning, i.e., deriving echelon-by-

echelon forecasts and buffer parameters without the benefit of shared point-of-sale data. We

recognize however, the existence of progressive supply chains that plan more holistically

(e.g., Brown, Schmitt, Schonberger, and Dennis, 2004; Ferdows, Lewis, and Machuca, 2004;

Li, Shaw, Sikora, Tan, and Yang, 2006).

We assume stationary, autocorrelated demand at Echelon 1 (see Figure 3.1). Each

echelon/stage, including Echelon 1, observes only the demand it receives from its immediate

stage customer without knowledge of the underlying demand distribution of final customers.

Using historical demand observed from the immediate customer, we forecast at every stage

the mean and variance of demand using single exponential smoothing (Snyder, Koehler,

Hyndman, and Ord, 2004). Exponential smoothing is a popular method used by many

companies (Gardner, 1985; Makridakis and Hibon, 2000).

The demand and variance estimates are applied, along with a service-level parameter,

in a single-stage order-up-to formula to calculate the replenishment order quantity.

Replenishment orders at one echelon/stage shown in Figure 3.1 become the demand at the

preceding stage. Inventory balance equations link each stage in a periodic (daily) review

system. Each stage follows the FIFO logic each day:

i. Launch a replenishment order if necessary using an order-up-to system

ii. Withdraw this day’s demand from available inventory to initiate shipment to the

customer

iii. Receive goods from the previous stage into inventory

iv. Update the inventory or backorder quantity (where backorders are permitted).

57

If demand at the succeeding stage is more than the available inventory, we assume a partial

shipment. The rest is lost at Echelon 1, and backordered at Echelons 2, 3, and 4. Other

approaches to shortages are applied as well in practice (Sodhi, 2005.) At Echelon 2, the

assembly stage, inventory of the two component types is maintained and orders for each are

placed if warranted. The quantity of an assembly order cannot exceed the minimum available

inventory of the two components.

The parameters we use in demand generation, forecasting and inventory control are

presented and justified in Appendix B. The order logic assumes periodic review with no

setup or order cost, infinite production rates, fixed lead times, and i.i.d. demands (Zipkin,

2004 and Nahmias, 1993). While optimally is by no means guaranteed with echelon by

echelon order-up-to policies in our system of stationary demand and lost sales at Echelon 1,

this logic is the most robust and suitable of those available. Such a policy is common in

practice and has been studied extensively in the literature. Nahmias (1993) and Axsater

(2000) provide details about order-up-to systems, which afford optimality for base-stock

policies under certain assumptions about the supply chain structure, shortages, order cost,

and demand distributions. Recent papers have found base-stock policies optimal over a

variety of unimodal risk-neutral and risk-averse objectives in single-item, single stage

systems with multi-period finite horizons and no order cost (e.g., Marinez-de-Albeniz and

Simchi-Levi, 2006; Chen et al., 2007; Huh et al., 2009).

Sandia wished to isolate the effects of a supply chain disruption on regional

economics. Consequently, we chose to induce a generic shock (time delay) to represent many

types of disruptions that occur in practice. After an initialization period of 1000 days, we

induce a 20-day disruption, and compare performance thereafter with the base case (without

58

expediting and disruptions). We consider disruptions at each end of our supply chain:

Echelon 1 (the electronics firm’s storage prior to shipment to customers), and Echelon

4/Stage A (supply by a parts distributor or component fabricator). During each disruption, the

facility at the affected echelon receives shipments or orders in-route before the start of the

disruption, but stops all other operations. It cannot place orders, produce orders, or make

shipments, and it turns away all demand during the disruption.

Management within all three case supply chains indicated that “time of a disruption”

becomes the critical factor, and expressed concerns about response and recovery. Prior

disruptions included: fire, weather disasters, worker strikes, raw material availability, power

failures, telecommunication failures, supply shortages, transportation breakdowns, and

machine breakdowns. Loss of a sole-source supplier introduced delays of up to two years to

find and procure alternate materials (as in the case of INT – Table 3.1). A disruption in

transportation of commodity components, with replenishment by sea, might involve as much

as a 90-day delay (as for Supplier 5 in case companies – Appendix D). Loss of electric power

or telecommunications would quickly stop activities in every firm in the area during the

length of the disruption, although essential telecommunication transactions might be handled

by cellular telephone, if the networks are not overloaded. These expressed concerns reflect

perceptions of the “new reality” as well as prior experience with disruptions such as port

lockouts, power outages, and loss of telecommunications. Management of the case firms

agreed that basic preventative measures such as fencing, guards, lighting and planning

against prior disruptions are insufficient, and that contingency planning deserves to be

elevated in importance within their organizations.

59

Firms within all three case supply chains frequently use expediting to avoid

shortages. Depending on the echelon, the percentage of orders expedited ranges from 5-20%,

with a premium cost per unit ranging from 10-50%. Bradley (1997) provides additional

motivation for expediting within the electronics industry. Expediting is typically

accomplished with faster transportation or production adjustments such as overtime,

additional shifts, part-time help, alternate routing, and responsive outsourcing. Following

prior practice and research, we apply two triggers to expedite lead times as orders are

launched, and experimented with a range of parameters for each type of trigger to achieve an

expediting frequency we observed in the case studies. The first is activated when the quantity

throughout the stage pipeline (on-order plus on-hand inventory) falls below the expected lead

time demand (Fakuda, 1960; Whitmore and Saunders, 1977; Groenevelt and Rudi, 2003;

Veeraraghavan and Wolf, 2008). The second trigger is activated when the on-hand inventory

falls below a demand-based target (Appendix B). As an example, Beyer and Ward (2002)

observed that Hewlett Packard’s supply network applies this second type of trigger to

expedite via air transport.

3.4.3 Key Issue 3: Performance Metrics

Three operational metrics drive important marginal economic effects in the case supply

chains. The first is the service level experienced by customers of electronics firms at the final

supply chain echelon. Firms at intermediate echelons typically have long-term relationships

with customers that call for backordering any items shorted (Sodhi, 2005). At the final

echelon, customers such as retail consumers and some manufacturers may have choices, and

shortages may be lost to the firm (INT and ABC). Others, e.g., some OEM manufacturers,

60

may have contractual arrangements that might instead permit backorders with the electronics

firm (INT and XYZ).

Shortages are undesirable anywhere in the supply chain, but countermeasures such

inventory positioning and expediting enable some orders to catch up. If shortages reach the

final supply chain echelon, however, the financial stakes may be exceedingly high. For

example, late delivery of avionics to Boeing Commercial, an OEM customer, may in turn

cause late delivery of an aircraft. This would result in loss of interest on delayed revenue

receipts of hundreds of millions of dollars, diminished revenue arising from contractual

penalties, and loss of goodwill with the airlines. In any event, the opportunity costs of

shortages at the final echelon can be severe in cases of either lost sales or backorders, with

loss of future business at stake.

System expediting, the second metric, offers obvious advantages in preventing

shortages. Nevertheless, expediting introduces significant premiums for transportation and

production that drains profit margins throughout the supply chain.

The third metric, system inventory, also drains profit margins in the supply chain.

Well-positioned inventory, however, provides a safety net against disruptions, and can

decrease expediting and shortages. All three case firms used ERP, but none had visibility of

total supply chain inventory. We chose total supply chain inventory as a metric because

without such visibility, important system inventory costs might be overlooked.

3.4.4 Key Issue 4: System Performance with a Disruption

To further guide experimentation by Sandia and other researchers under more rigorous

operating circumstances, we conducted simulation experiments using the aforementioned

model structure, and adapted the experimental design accordingly to address Key Issues 4-7.

61

We observe the performance on a day-by-day basis over 100-day time blocks,

replicate the experiments 100 times, and compare performance between the disrupted and

base cases. A period of 100 days covers five months in a 240-working day year.

It is important to ensure that our initiation period is long enough to remove transient

effects, and the run length effectively captures steady-state performance. Each simulation

was run for 2000 days. The first 1000 establish steady state conditions, and we observe

performance over the remaining time. Pilot experiments suggested that our choice of

initiation period satisfies Welch’s conditions outlined in Law and Kelton (2000).

We distinguish issues of independence of a performance measure within and across

replications by drawing on Chapter 9 of Law and Kelton (2000). We do not claim or expect

independence within replications, either in demand or supply. Demand is by definition

autocorrelated over time, and inventory levels and replenishment orders are clearly linked

from one period to the next. Indeed, we wish to induce bullwhip effects over time within

each replication to reflect the effects observed in practice. Across replications, we applied

terminating sampling and chose independent random number seeds to initiate each sample

replication.

Cost structures varied substantially by firm and item in our case studies. We

addressed this complication by analyzing one set of weights in Key Issue 6 to demonstrate an

example that counters conventional research assumptions. For more generality in 7, we

consider a design with a range of costs. With 4 and 5, we apply MANOVA across

performance metrics, and focus on statistical inferences consistent among them as reported

over replications.

62

For Key Issue 4), our MANOVA design has two fixed factors: time block and

location of disruption. We do not enable the expediting option as mitigation, and observe the

service-level and system-inventory metrics as dependent variables. We found that values

from each metric are drawn from different distributions, according to Pillai's Trace, Wilks'

Lambda, Hotelling's Trace, and Roy's Largest Root statistics at the .01 level (Arbuckle,

2005). Results for the two metrics are summarized in Figure 3.2. Each point in the graphs

depicts the mean value over an indicated five-month time block.

Waller-Duncan multiple-range post-hoc tests disclose that for both disrupted

locations, service level is significantly different between disrupted and base-case states over

the first three time blocks (15 months). In the final eight time blocks, there is no significant

difference among the means and zero. All of these results hold at the .01 and .05 levels.

System inventory is significantly different over the first six time blocks (30 months).

In the final five time blocks, there is no significant difference among the means and zero.

Clearly, the effects of disruptions on the two performance metrics last for a long time. There

are severe decreases in service level for more than a year. Additional system inventory over

the base case exceeds ten weeks of demand for more than two years after a disruption at

Echelon 1.

In analyses comparing disruptions at Echelons 1 and 4, the MANOVA statistics at the

.01 level indicate significant deterioration in service level over the first five months after a

disruption at Echelon 1, as well as increased system inventory (more than twice the amount)

in months six through 30.

63

a. Each point represents a mean value over a time block. Waller-Duncan multiple range tests indicate that points not circled are significantly different from one another; those circled indicate no significant difference between one another and zero. These results hold at the .01 and .05 levels.

Figure 3.2. Performance effects of a disruption.a

6363

64

We observe a strong bullwhip influence, especially for a disruption at Echelon 1,

where order variability is amplified backwards in the supply chain. With the disruption at

Echelon 4 rather than 1, final assembly orders have a chance to recover over the longer

cumulative lead times; the derived order-up-to levels were more consistent; and the

associated system inventory levels were lower on average. This is in contrast to findings by

Wu and Chen (2009) who found that regardless of source in a two stage system, larger

fluctuations occur closer to the disruption and dampen when propagating away. This can be

attributed to the differences in our model of more stages, the presence of lead times, the lack

of perfect information flows, and the corresponding presence of bullwhip effects.

Our experimentation sheds light on the relevance of Key Question 4), and highlights

the importance of studying the effects of disruptions at different stages in the supply chain.

This led us to recommend that Sandia consider a variety of disruptions at various stages and

between stages.

3.4.5 Key Issue 5: System Performance under Expediting

We compare performance results of the base case (without expediting and disruptions) with

an expediting case (no disruptions). If the expediting option is enabled, we permit expediting

at all echelons, with order crossover a possibility. Figure 3.1 shows values of the LTs and

ELTs we chose in the experiments with details in Appendix B.

To compare performance between the base case and expediting, we use all three

performance metrics (final-echelon service level, system inventory, and system expediting)

as dependent variables using MANOVA. Values from the three metrics are drawn from

different distributions, according to the aforementioned statistics at the .01 level.

65

We also observe significant performance differences at the .01 level between the base

and expediting cases for each metric. While significant, however, the relative performance

differences raise issues about the value of expediting. Shortages improve with expediting by

only 0.18 units per day on average demand of 111 units, while total system inventory

increases by an average of 787 units per day (almost eight days of average demand). System

inventory increases so much because expediting increases the variability in order quantity

and frequency, and this variability is amplified downstream in the supply chain.

Expediting, while considered necessary (Bradley, 1997; Cohen et al., 2003), is

usually expensive (Arslan, Ayhan, and Olsen, 2001; Groenevelt and Rudi, 2003). Beyer and

Ward (2002) observed that expediting by air in HP’s supply chain costs as much as five times

more than standard shipment by sea. In the case firms, we found that expediting was quite

expensive as well.

According to the personnel we interviewed, expediting offers a first line of defense

against variability deemed irregular as well as that considered part of normal business

conditions. Our experimental findings raise questions about the value of this practice as a

mitigation approach of choice. We are not the first to raise the issue. Even before bullwhip

effects were understood, the practice of expediting was challenged because of nervousness it

may induce in MRP systems (e.g., citations in Schmitt, 1984). Regardless of the evidence, we

do not expect firms to discontinue the practice of expediting. It would be difficult to convince

managers of an electronics firm not to expedite one component when the remaining 700

needed for assembly and sale of a finished product are available. However, practitioner

behavior might be influenced by future research that finds merit in the application of certain

types of expediting at specific supply chain stages, or between stages.

66

3.4.6 Key Issue 6: System Performance and Applicability of Analytics, Heuristics

Traditionally, researchers offering analytics or heuristics have been careful in claiming utility

only within their system assumptions. Base-stock policies have been shown optimal for

stationary stochastic demand in simple supply chains (Axsater, 2000; Porteus, 2002).

Without showing optimality, some have used order-up-to and other well known policies in

simple supply chains in the presence of supply or demand disruptions (Chao, 1987; Gupta,

1996; Arreola-Risa and DeCroix, 1998; Snyder, 2006, Lewis et al., 2008). Yet even in single-

stage stationary systems, researchers have not found tractability when shortages are lost (e.g.,

Karlin and Scarf, 1958; Janakiraman and Roundy, 2004; Janakiraman, Seshadri, and

Shantihikumar, 2006), and others have proposed heuristic solutions (Nahmias, 1979;

Donselaar et al. 1996; Metters 1997; Ketzenberg et al. 2000).

In this section, we observe non-unimodal total cost behavior within a range of stage

order-up-to levels under conditions of no expediting or disruption. To make a case, we

choose a holding cost of $1/unit/day, a backorder cost of $2/unit/day (at Echelons 2, 3, and

4), a lost-sales cost of $3/unit (at Echelon 1), and a stage production/transportation cost of

$1/unit. We present results for this case, and in the next section, report summary statistics

over a variety of costs.

Figure 3.3 shows plots of the simulation results of total cost over various stage order-

up-to levels only for the first of the replications. The two graphs to the left focus solely on

behavior at Echelon 1, while the ones to the right show interactions between Echelon 1 and

Echelon 2.

The graphs on the left show total cost values for discrete order-up-to levels at Echelon

1, while allowing the other three echelons to derive order-up-to levels from stage demand

67

forecasts. The top left graph displays total cost performance versus Echelon 1 order-up-to

levels over the range [1, 2000]. Clearly, the local optima vary considerably in value, e.g., one

yields a total cost 6.95 times larger than the lowest observed total cost value, given the

starting point.

The bottom left graph depicts a finer grain relationship over the range [1000, 1530]. It

highlights the striking volatility of the cost function.

The graphs on the right of Figure 3.3 show the interactive behavior between the first

two echelons, depicting total supply chain cost while varying Echelon 1 order-up-to levels

for prescribed Echelon 2 levels. In this set of experiments, Echelons 3 and 4 derive order-up-

to levels from stage demand forecasts. The course and fine grain representations at the top

and bottom right, respectively, generalize the previous observations solely about Echelon 1.

Clearly, computationally-efficient iterative line searches or analytic searches (based on

assumptions of well-behaved first and second order functions) would derive very poor

solutions in this operating scenario, depending on the starting point and method efficiency.

There are multiple ways to demonstrate our concerns about ill-structured performance

behavior and inappropriateness of exact approaches, which include deriving violations of

Kuhn-Tucker conditions and finding derivatives with the wrong sign. We chose another way,

the presentation of this counter example, one where exact approaches would yield a total cost

almost seven times larger than optimum. Existence of such a counter example, of any

magnitude, is sufficient to obviate generality of exact analytical approaches. In the next Key

Question, we consider the efficacy of a unimodal search method over a wide variety of cost

structures.

Figure 3.3. Shape of total cost function.b,c

Total Supply Chain Cost vs Echelon 1 Order-up-to level over the Quantity Range [1, 2000]

0

10000

20000

30000

40000

50000

60000

70000

80000

1

107

213

319

425

531

637

743

849

955

1061

1167

1273

1379

1485

1591

1697

1803

1909

Order-up-to at Echelon 1

To

tal

Su

pp

ly C

hain

Co

st

Total Supply Chain Cost vs Echelon 1 Order-up-to levelover the Quantity Range [1000, 1530]

9500105001150012500135001450015500165001750018500

1000

1028

1056

1084

1112

1140

1168

1196

1224

1252

1280

1308

1336

1364

1392

1420

1448

1476

1504

Order-up-to at Echelon 1

To

tal

Su

pp

ly C

hain

Co

st

Total Supply Chain Cost vs Echelon 1 Order-up-to level over the Quantity Range [1, 3000], at various Echleon 2 Order-up-to Levels

0

10000

20000

30000

40000

50000

60000

1

15

1

30

1

45

1

60

1

75

1

90

1

10

51

12

01

13

51

15

01

16

51

18

01

19

51

21

01

22

51

24

01

25

51

27

01

28

51

Order-up-to at Echelon 1

To

tal S

up

ply

Ch

ain

Co

st

Total Supply Chain Cost vs Echelon 1 Order-up-to level over the Quantity Range [940, 1357], at various Echleon 2 Order-up-to Levels

7800

17800

27800

37800

47800

94

0

96

1

98

2

10

03

10

24

10

45

10

66

10

87

11

08

11

29

11

50

11

71

11

92

12

13

12

34

12

55

12

76

12

97

13

18

13

39

Order-up-to at Echelon 1

To

tal S

up

ply

Ch

ain

Co

st

Ech. 2 Level = 2500

Ech. 2 Level = 3000

Ech. 2 Level = 3500

Ech. 2 Level = 4000

Ech. 2 Level = 2500

Ech. 2 Level = 3000

Ech. 2 Level = 3500

Ech. 2 Level = 4000

b. Total Supply Chain Cost while Varying Echelon 1 Order-up-to Level; Three Other Echelons DeriveOrder-up-to Levels using Demand Forecasts

c. Total Supply Chain Cost while Varying Echelon 1 Order-up-to Level at prescribed Order-up-to Levelsat Echelon 2; Other Two Echelons Derive Order-up-to Levels using Demand Forecasts

68

69

3.4.7 Key Issue 7: Genetic Search versus Line Search

We next examine the modality of the objective function surface under a variety of cost

settings. Sandia programmed a general genetic search (GA) function to “jump over” local

optima within a wide range of parameter values. We conducted experiments to examine

Sandia’s GA approach to search for cost-effective order-up-to quantities at Echelon 1, while

allowing demand forecasts to guide decision making at other Echelons. The GA utility

represents candidate solutions as binary numbers. Crossover and mutation operators are used

to overcome local minima. For details, see Goldberg (1989).

We consider line search (LS) as a basis for procedural contrast and to examine

performance modality. LS ensures optimality only when the objective function is unimodal.

In our problem setting, a LS approach that increments upwards from an order quantity of 0 to

find a local optima may result in a significantly higher than optimal cost. To facilitate a

reasonably fair comparison with GA, our adaptation of LS begins with an order-up-to

quantity equal to the mean lead time demand at Echelon 1 and searches each way in the

neighborhood (upward first) until we encounter local optima. In pilot experiments this

adaptation resulted in solutions much better than starting at 0.

Table 3.2. Various costs at two levels each.

Back Ordering

Cost

Cost of Lost Sales

Carrying Cost

Expediting Cost

Low 4 6 2 3High 8 12 4 5

70

To represent a wide variety of operating conditions, we selected a range of cost

parameters as presented in Table 3.2. The parameters are based on findings in literature and

case studies. Realistic costs depend on the type of product, industry, and supply chain, among

other factors. Cohen et al. (2003) observed that with short life cycles and obsolescence in the

semiconductor industry, the cost of losing a sale is about twice the cost of backlogging, while

the holding cost is three times the cost of backlogging. We follow these ratios for setting the

cost parameters. By contrast, Faaland et al. (2004) experimented with lost-sales costs in a

single-echelon ranging from 62 to 500 times the periodic inventory holding cost, yielding

shortage values much higher than the ones we chose. Their parameters were based on a

cross-industry survey by Boer and Jeter (1993). Our choice of relatively low shortage costs

reported here is intended to explore benefits of GA over LS under modest conditions. We

observed in pilot experiments that larger shortage costs relative to inventory carrying costs

further exaggerated differences between GA and LS.

Table 3.3 shows the cost savings under various cost scenarios. The percentage

improvements represent averages over 100 replications (observed over periods 1000-2000)

for each of the 16 cost combinations. The superiority of GA over LS at Echelon 1 holds at

the 0.005 level using student-T tests. GA as compared with LS yields overall cost savings

greater than 16%, and in one of the cost setting by more than 30%.

Genetic Search yielded significantly better costs at the expense of computation time.

On a laptop, the time per run for GA was about 14 hours, and our adaptation of LS averaged

about 10 minutes. While the computation time for GA may be acceptable for Sandia with its

fast complex of computers, we recognize an opportunity in future work to adapt standard GA

utilities to special properties of this problem setting.

71

Table 3.3. Results for Line and Genetic Search under various cost combinations.

Back Ordering

Cost

Cost of Lost Sales

Carrying Cost

Expediting Cost

Line Search

LS

Genetic Search

GA

MeanPercentage

Cost Savings

Low Low Low Low 551.01 472.38 14.27Low Low Low High 606.78 493.61 18.65Low Low High Low 808.55 685.79 15.18Low High Low Low 621.22 496.91 20.01High Low Low Low 690.61 516.98 25.14High High Low Low 764.22 544.26 28.78High Low High Low 996.62 922.66 7.42High Low Low High 742.80 524.13 29.44High High High Low 1028.71 927.69 9.82High High Low High 797.62 551.08 30.91High Low High High 994.61 922.56 7.24Low High High High 974.21 896.36 7.99Low High High Low 873.29 812.41 6.97Low High Low High 668.32 510.28 23.65Low Low High High 852.02 833.00 2.23High High High High 1071.11 941.23 12.13

3.5 Conclusions and Future Directions

Disruptions can have many sources covering the gamut from natural to accidental to

intentional. Regardless of cause, disruptions can have long-lasting, widespread, and costly

effects on supply chains. We describe aspects of a stream of research to assess the economic

impact of supply chain disruptions. The overarching research mission of our sponsor, Sandia

National Laboratories, is to develop a large-scale simulation model that depicts regional and

national economic behavior to a disruptive event. Simulation is intended to augment existing

analytical and statistical models whose utility may depend on the validity of inherent

simplifying assumptions. Such optimization models routinely assume aggregation of demand

and supply data, substitutability of supply options, independent and steady-state behavior of

72

underlying stochastic distributions, optimization over well-behaved objective functions, and

simple supply chain structures (Chen, Sim, Simchi-Levi, and Sun, 2007 and citations).

Sandia’s concern was that with such high stakes, entities within the U.S. government and

private sectors cannot afford to wait for researchers to overcome the substantial challenges of

overcoming simplifying assumptions in the optimization models.

Sandia’s agent-based simulation has the capability to incorporate supply chain

networks with a million or more firms and supporting infrastructure. The first project for the

system has been to address the Pacific Northwest region. We concentrated on one stage of

this effort on how to model business activity of small and medium sized firms within supply

chains. It is anticipated that these insights will be useful in other research as well.

We conducted case studies of three electronics firms in the region, and drew from the

cases to offer a fundamental set of design requirements, performance drivers, and important

research questions. Sandia may not need to represent entire supply chains, but if electronics

assembly is representative, their model should include at least:

(i) Four echelons per industry,

(ii) An echelon with assembly,

(iii) Bullwhip effects facilitated by a multi-echelon, multi-stage inventory

system with local planning,

(iv) Shortages along the supply chain in the form of backorders and lost sales,

(v) Capability to expedite at all stages,

(vi) Service level at the final stage, system expediting and system inventory as

performance metrics, and

(vii) Disruptions in the form of time delays at various stages in the supply

chain.

To guide Sandia’s experimentation, we followed with research questions, whose answers

were driven by simulation results. One important finding was that the system cost function

73

can be quite ill-behaved in a four-echelon supply chain, even in the absence of disruptions

and expediting. We reveal a weakness of analytical optimization approaches in the present

setting by providing one example where these methods failed to obtain satisfactory solutions,

along with supportive statistical evidence over a wide range of operating conditions.

Analytical methods that assume unimodal behavior may be inappropriate for real-world

supply chains.

Another finding confirms that disruptions may have long-lasting, rippling, and costly

consequences within the supply chain structure presently considered. Another is that standard

industry practices of setting order parameters locally and using expediting as mitigation seem

to exacerbate these undesirable effects. In addition, our cases, experiments and subsequent

observations by Sandia (Appendix A) support Craighead et al. (2007), who suggest

simulation analysis and field study as means to verify propositions that the severity of a

disruption is related to time, its location, the structure of the supply chain, and types of

mitigation. We believe that additional field study across industries is warranted.

We address disruptions at each end of a four echelon supply chain, but we suggest

that an expanded study may find value in investigating disruptions elsewhere. In support of

this, Sandia’s model has capabilities to explore more fully the hypothesis that mitigation

efforts should focus on specific areas of network criticality, i.e., to explore disruptions at any

supply chain stage as well as in infrastructure shared by many firms within a region, such as

transportation hubs, electric power, and telecommunications. Their model also offers more

structural flexibility, e.g., parts distributors may have multiple customers, which may enable

demand aggregation and dampening of the demand amplification elsewhere in the supply

chain.

74

Our experimental results suggest caution, however, regarding mitigation efforts and

resiliency. Information sharing offers clear advantages in a supply chain, and authors have

addressed this issue (e.g., Milgrom and Roberts, 1988; Lee et al., 2000; Chatfield, Kim,

Harrison, and Hayya, 2004; Sodhi, 2005). As a caveat, our experiments disclose that

information must be discounted considerably to control bullwhip effects in moving from the

source of uncertainty in a supply chain. For instance, we derived a very low forecasting

weight of 0.01 on the most recent local information for the firms in the last echelon, Echelon

4. While we support the notion that increased information and flexibility to react are

generally desirable, we caution that it is possible to overreact. A disruptive event creates a

critical watershed. The issue is how much weight to place on the news of such an event and

what to do about it.

Furthermore, with the irregular cost objective surface documented by our results, we

believe future efforts should be directed towards developing efficient and effective search

methods to find order parameters yielding local optima close to global cost values. Further

research is needed to explore the efficacy of hybrid GA approaches as well as other

metaheuristics (Corne et al., 1999; Kimbrough et al., 2002; Glover and Kochenberger, 2003;

Rego, 2005). A brute force approach would be to increase the generation count of GA to

further improve solution quality, but this would likely exacerbate the formidable computation

problem. We think a better alternative would be to take advantage of problem structure and

domain knowledge (Holland, 1992; Gen and Cheng, 2000). One approach would be to assign

the lead time demand as one of the starting solutions for GA, since the operators of crossover

and mutation in standard utilities do not allow a ‘directed’ intelligent search. This might help

solution effectiveness and efficiency as it did with LS in our experiments.

75

Our findings suggest that consideration of lost sales, multiple echelons, and assembly

perturb stationary behavior that might otherwise be found in less complex systems, and that

disruptions and expediting exacerbate the problem. Further research is warranted on dynamic

order-up-to policies with parameters updated periodically, whether the adjustments are made

using adaptive search over demand history, or cost-based search directly on order-up-to

parameters. This premise is further supported by simulation experiments conducted by Ross

et al. (2008) who found a time-varying order-up-to policy more effective than a static policy

in terms of the total costs of holding, ordering and lost sales. They considered a single

product at a single stage in the presence of cyclical demand and recurrent disruptions. From a

practical perspective, case firms such as those we studied, which employ periodic review

time-phased order point systems, may be able to incorporate such dynamic ordering policies.

Additional insights may also result from future simulation research that relaxes some

of our simplifying assumptions. Sandia has already extended our model to include

price/demand elasticity functions and a diverse customer base for agent firms. A broader

spectrum of costs and revenues can be advantageously investigated within an experimental

design to identify the conditions under which mitigation approaches are effective against

certain disruptions. Our experiments embraced both normal and expedited activity lead

times, but did not consider capacity interactions that might result from finite replenishment,

setups and specific capacity adjustments such as overtime, additional shifts, part-time help,

alternate routing, and sub-contracting. Non-stationary demand, stochastic lead times,

stochastic failure times, and lead time/demand elasticity represent other realistic extensions.

However, prior work as well as ours suggests that variability in quantities and lead times,

whether from normal operations, disruptions, or expediting, is amplified in supply chains. It

76

is possible, although unlikely, that future extensions such as those we suggest would

contribute to improved problem tractability. The practical value of our contribution is

affirmed in the following feedback from Sandia [2007]: “[This work] demonstrated the

importance of careful design in modeling the realities of supply chain behavior, and provided

strong motivation for further simulation development and experimentation.”

77

CHAPTER 4

ADAPTIVE SEARCH METHODS FOR ORDERING DECISIONS IN MULTI-

STAGE SUPPLY CHAINS

4.1 Introduction

To remain competitive, today’s marketplace requires that supply chain managers make

efficient decisions. These decisions can be made using sophisticated mathematical modeling

and solution techniques. Advancements have been made in mathematical modeling; however,

some have not had the desired impact on decision-making in supply chains (Lourenco, 2005).

Moreover, many analytically tractable models suffer from restrictive assumptions (Kochel

and Nielander, 2005). The difficulty is partly due to the complexity of real-world supply

chain decisions. Other factors such as supply chain disruptions also add to the difficulties in

decision-making.

Supply chain management decisions include strategic, operational, and tactical.

Ganeshan et al. (1999) further classify operational decisions into inventory management and

control; production, planning, and scheduling; information sharing, coordination and

monitoring; and operations. Our work deals with inventory management, which is a critical

part of supply chain management as it impacts all business functions and can account for 20-

40% of the total item cost (Schroeder, 2008; Ballou, 1992). We further focus on periodic

review order-up-to ordering policies. These policies are popular in practice and are shown to

be optimal for single-echelon, independent items, and stationary stochastic demand (Axsater,

78

2004). Despite their popularity, the optimality of these policies under relatively simple but

practical situations such as lost sales and expediting is not known (Karlin and Scarf, 1958;

Janakiraman, Seshadri, and Shanthikumar, 2006; Veeraraghavan and Wolf, 2008).

Lack of optimal policies has motivated use of simulation and heuristics in research

and practice. Simulation has been useful because of its ability to handle complex problems

involving stochastic variables and non-linear functions (Amodeo et al. 2009). Simulation is

primarily used for evaluation of inventory policies and is not an optimization tool (Daniel

and Rajendran, 2005). We embed metaheuristic optimization toots in our simulations to

search for efficient solutions.

We propose and test practically implementable adaptive search methods in a multi-

stage supply chain that includes assembly, lost sales, and expediting. The metaheuristic

methods proposed can accommodate a wide range of real-life supply chain constraints. To

our knowledge, this is the first study that uses adaptive search methods for supply chain

ordering decisions. We also consider the possibility of lost sales, which increases the

practical relevance of our model. Besides order-up-to levels, our decision variables include

percentage of orders expedited and two expediting triggers. We believe ours is the first such

effort in using metaheuristics for expediting decision variables.

We propose a novel objective function, for both local and global supply chain

planning. Our objective function involves holding costs, backorder costs, lost sales costs, and

expediting costs, and resembles the exponential smoothing method. Depending on the

volatility of demand and possible disruptions, the ‘smoothing parameters’ can be adjusted,

making the objective function well-suited for steady as well as disrupted supply chains.

79

Many inventory models, including the pioneering work of Clark and Scarf (1960),

assume stationary demand and find a static order-up-to policy; however, the stationary

assumptions may not be valid in practice. Market trends; seasonal demand and production;

variable material, labor, and equipment availability; and disruptions are some of many causes

that can violate the stationary assumption. Therefore, a dynamic inventory policy may be

more appropriate. Our design of search methods and the objective function allow us to

‘optimize’ under a dynamic situation; i.e., decision variables are time variant and are

determined each period.

Metaheuristic methods have been shown to perform better when designed with

problem-specific properties (Holland, 1992; Chen et al., 2004; Gen and Cheng, 2000). We

reveal useful properties of objective function and feasible solution space. These properties

are then used to make several problem-specific adaptations to increase search efficiency.

Specifically, candidate list strategy, intensification, reference set, and a flexible aspiration

criterion are introduced to increase the solution quality and efficiency of adaptive search.

Some aspects of scatter search are also incorporated. We also compare the performance of

adaptive search with genetic search and Fibonacci search. Using cost parameters from

Schmitt et al. (2009) we show the superiority of adaptive search over other methods.

Both stationary and non stationary demand scenarios are considered. We effectively

use Adaptive search to find dynamic ordering policies under a seasonal and disrupted

demand. The results are shown to be superior to best static policies obtained assuming

complete future demand information.

The remainder of the chapter is organized as follows: We review related literature in

Section 4.3; we describe the supply chain model and parameters in Section 4.4; Section 4.5

80

includes a description of metaheuristic methods; in Section 4.6, we present the results;

Section 4.7 lists extensions, and Section 4.8 concludes the chapter.

4.2 Literature Review

Since the pioneering work of Clark and Scarf (1960), inventory management in supply chains

has received much attention. See Axsater (1993) or Federgruen (1993) for a thorough

literature review. Clark and Scarf modeled a periodic review serial supply chain with fixed

lead time between echelons and a cost function that includes inventory holding, backorder,

and per item ordering cost. Assuming complete backorder, they showed that an order-up-to

policy is optimal. Federgruen and Zipkin (1984) extend the results to infinite horizon and

show that calculating order-up-to values is simpler in infinite horizon. Chen and Zheng

(1994) provide optimality results for continuous review policies. Chen and Song (2001)

consider a state-dependent demand function and show that a state-dependent order-up-to

policy is optimal for long run average cost function. Muharremoglu and Tsitsiklis (2001)

simplify the optimality proofs of order-up-to policies and establish the optimality in

stochastic lead time and state-dependent demand. Parker and Kapuschiski (2004) add

capacity constraints. They show that a modified order-up-to policy is optimal when a lower

echelon has a smaller capacity.

Clark and Scarf (1962) generalize the problem by including a fixed ordering cost.

They show that the optimal policy is complex, and even if the optimal policy is identified it

would be difficult to implement, making it unattractive in practice. Moreover, under general

conditions, a simple base stock policy may not be optimal or computation of parameters may

become challenging (Swaminathan and Tayur, 2003). Optimal policy under lost sales, even

with deterministic demands, is an open problem (Swaminathan and Tayur, 2003). Optimal

81

policies under generalized expediting are also not known (Veeraraghavan and Wolf, 2008).

Most analytical research uses dynamic programming with recursive algorithms to find

optimal policies. However, developing such algorithms is mathematically and

computationally difficult (Clark, 1960; Shang and Song, 2003). Many supply chain inventory

decisions are based not on algorithms but rule of thumb (Lourenco, 2005).

Difficulties with analytics have motivated a part of research to focus on heuristics or

approximate solution procedures for finding order-up-to values (Chen, 1999). To overcome

difficulties with order-up-to policies, Chen (1999) and Cachon (1995) implemented (R, NQ)

policies, where R is the reorder point and Q is the base order quantity. N is the minimum

integer required to increase inventory beyond R. This policy is proposed as an approximation

to (s, S) policy. By converting an N-stage serial system to a 2N-stage problem, Shang and

Song (2003) provide an easily implementable myopic heuristic for determining order-up-to

quantities. Under restricted assumptions on demand, horizon, and cost function, they show

their heuristic to be close to optimal. Under capacity constraints, de Kok (1989) proposed a

modified base stock policy by using the inventory position at the start of a period which

equals S − X, where X is the waiting time in a D/G/1 queue. He also provided a heuristic for

computing order-up-to quantity. Bollapragada and Morton (1999) consider an ordering

policy problem with setup cost and provide an effective myopic heuristic by approximating

the non-stationary problem with a stationary problem and solving for the stationary problem.

Using an application of the Hewlett-Packard supply chain, Lee and Billington (1993) propose

a simple search heuristic that looks for service level targets. Other heuristic methods can be

found in Hausman and Peterson (1972), Heath and Jackson (1994), Donselaar et al. (1996),

82

Graves et al. (1998), Eynan and Kropp (1998), Rajagopalan and Swaminathan (2001),

Roundy and Muckstadt (2000), Lu et al. (2006), Hurley et al. (2006), and Levi et al. (2008).

Simulation, because of its ability to manage real-world complexity, has been used as

a powerful alternative to analytics. In the domain of supply chain inventory management,

simulation has been used to study the performance under various inventory policies (Towill

et al. 1992), supply chain configurations (Souza et al. 2000), Web-based operations (Beamon

and Chen, 2001), operating policies (Holweg and Bicheno, 2002), demand uncertainty (Closs

et al, 1998), information sharing (Strader et al. 1998; Li et al. 2005), and market conditions

(Ye and Farley, 2005). Chatfield et al (2004) study the effect of lead time and information on

bullwhip effect.

Simulation is not an optimization tool and is primarily used for performance

evaluation and not problem solving. To address this shortcoming, simulation has been

hybridized with search methods creating simulation-based optimization tools. These tools

take advantage of the flexibility and versatility of simulation and the optimization power of

search heuristics (Gosavi, 2003). Glasserman and Tayur (1994, 1995) find the base stock

using a simulation-based optimization procedure, where infinitesimal perturbation analysis

was used to develop an efficient solution methodology. Rao et al. (2000) analyzed various

supply chain configurations. Besides dynamic programming, they used simulation-based

optimization to find inventory levels for various configurations in Caterpillar’s construction

equipment supply chain. Ettl et al. (2000) used a conjugate gradient method as a search

routine to find base-stock levels for each echelon with the objective of minimizing the supply

chain inventory while maintaining required service levels.

83

Metaheuristics has been used in many supply chain applications; however, their use

for ordering decisions is relatively new. Daniel and Rajendran (2005) provide a Genetic

Algorithm (GA) based simulation method to find periodic review order-up-to values in a

four-echelon serial supply chain. The cost objective consists of holding and shortage costs

across all echelons. The performance of GA is compared with solutions from random search

and complete enumeration. The solutions from GA are shown to be near optimal. Both

stochastic and static lead times are modeled. Final customer demand at echelon 1 is drawn

from a uniform distribution. While assuming all shortages to be backlogged, they test six

different cost combinations of shortages and holding. GA is used to search through a time

horizon of 1200 periods, repeated 30 times. Daniel and Rajendran (2006) extend the model to

include advanced Genetic Algorithm techniques of gen-wise crossover and random keys

presentation of chromosomes. Furthermore, they consider separate objectives for holding

costs and shortage costs and develop a GA to find non-dominated Pareto solutions.

Kochel and Nielander (2005) modeled a continuous review, base stock, five-echelon

serial supply chain with the option of transshipment between the echelons. The modeled

parameters include cost of holding, backorder, and shipment. Various combinations of

transshipment costs were studied. They also discuss the possibility of using the method for

non-serial supply chains. The tests include a comparison of centralized and decentralized

supply chains. GA was used as the metaheuristic method. Amodeco et al. (2009) model a

periodic review, two-stage supply chain that includes three suppliers and one manufacturer.

The manufacturer assembles three components from the suppliers to produce one final

product. A dual objective of customer service level and inventory cost is considered. The

lead times and costs are fixed, while demand at the manufacturer is exponentially distributed.

84

All unmet demand is backlogged. They report the usefulness and success of GA in finding

the base stock levels.

We contribute to research in simulation-based optimization for ordering decisions and

in general in inventory management. Our work differs from earlier research in, i) model; ii)

supply chain operational parameters; iii) dynamic policies; iv) objective function; and v)

metaheuristic method used. We borrow the supply chain model and parameters from our

earlier work in Schmitt et al. (2009). The model is based on literature findings and case

studies of electronics companies. It includes four echelons, including an assembly stage. To

explore a more realistic supply chain, we consider expediting at all stages, which is common

in practice to avoid shortages (Schmitt et al. 2009). We also consider the possibility of

shortages at the lowest echelon.

We also propose a time variant dynamic ordering policy. Much of inventory literature

is focused on static policies, partly because of modeling and analytical difficulties in

treatment of dynamic policies. However, prior research has shown the superiority of dynamic

policies. Donselaar et al. (1996) compare the performance of a static optimal base stock

policy with a dynamic myopic heuristic base stock policy and show empirically that the latter

significantly outperforms the stationary base stock policy. Ross et al. (2008) while modeling

supply chain disruptions, found instances where a time-varying dynamic order-up-to policy is

more effective than a static policy in terms of the total costs of holding, ordering, and lost

sales. From a practical perspective, when considering real-world scenarios where stationary

assumption may not hold, dynamic ordering policies should outperform static policies. (Note

that static policies are a subset of dynamic policies.)

85

Our design of objective function captures variations in environmental and supply

chain parameters over time. The objective function is, in essence, similar to an exponential

smoothing function, which is an effective forecasting method and is popular in practice

(Snyder et al., 2004; Gardner, 1985; Makridakis and Hibon, 2000). Our objective function

has the required properties to react and adapt to supply chain variations, making it suitable

for use under supply chain disruptions, seasonality, and trends.

Adaptive search has dramatically changed our ability to solve a host of problems in

applied science, business, and engineering (Glover and Laguna, 2008). Its adaptive memory

is well-suited to facilitate solutions to complex problems. We implement some improvements

in a Tabu search and compare the results with GA and Fibonacci searches. We also

successfully use Adaptive search method under seasonal and disrupted demand.

4.3 The Supply Chain Model

Our model is designed to reflect the importance and benefits of using Adaptive search

methods in a supply chain that represents realistic operations. Using electronics industry data

and findings from literature, Schmitt et al. (2007) developed a similar supply chain structure

to model operations in multi-echelon supply chains.

4.3.1 Supply Chain Structure and Activities

Figure 3.1 (See previous chapter) shows the model supply chain with four echelons. Echelon

1 has only one stage, corresponding to delivery of the finished goods to customers. Echelons

2, 3, and 4 each include two stages, A and B, which refer to activities associated with two

components (A and B) that go into the finished product. The stages of echelons 2, 3, and 4

are referred to as stages 2A, 2B; 3A, 3B; 4A and 4B, respectively.

86

Raw materials are processed and transported through the four echelons to be sold to

customers. Echelon 1 includes the processes of a retailer who sells the finished product to

customers. Echelon 2 assembles two component parts (A and B) from stages 3A and 3B to

form the final product, which is then ready to be shipped to echelon 1. If either component A

or B is short, the assembly order cannot be launched. Echelon 3 consists of two distribution

centers, stages 3A and 3B. Echelon 4 consists of two manufacturers at stages 4A and 4B,

who acquire raw materials from respective suppliers to manufacture components A and B.

These two components are then supplied to stages 3A and 3B, respectively.

4.3.2 Supply Chain Parameters and the Inventory Logic

Subscript 1i represents the lowest supply chain echelon, i.e., the retailer. Goods flow from

a higher echelon to a lower echelon. Note that echelons 2, 3, and 4 have two stages each. To

avoid ambiguity between stages and echelons, we describe the model for stages 1, 2, 3, and

4, by dropping the component notation of A and B. Echelons 2, 3, and 4 can represent either

of the two components, A and B. The equations and logic elaborated for i=1, 2, 3 and 4 can

be used for the echelons i=1, 2A, 3A, 4A or i=1, 2B, 3B, 4B. Unless specified otherwise, all

descriptions and equations below are for echelon i.

Notation.

Indices

i Supply chain echelon, }4,3,2,1{}4,3,2,1{ BBBorAAAi

t Time period in days

87

Decision variables

tiY Order-up-to quantity at time t at echelon i

tie Fraction of an order expedited at time t from echelon i

tiOTh On-hand inventory threshold at time t for echelon i

tiPTh Pipeline inventory threshold at time t for echelon i

Parameters

1 Mean demand per period at echelon 1

AR(1) coefficient of demand at echelon 1

Independent and identically distributed normal random variable, ),0( N

eiL Expedited lead time in days at echelon i

iL Regular lead time in days at echelon i

i Fraction of shortages backordered at echelon i

Discount factor

h Holding cost per unit per period

c Backordering cost per unit per period

g Cost of lost sales per unit

r Cost of regular order per unit

f Cost of expediting per unit

Other variables

tD1 Final customer demand in period t at echelon 1

tiTh Binary variable,

;

,

0

1

otherwise

OThIorOThIIfTh

ti

ti

ti

tit

i

88

tis Shortage at time t at echelon i

tib Number of units backordered at time t at echelon i

til Lost demand at time t at echelon i

tiS Number of units shipped at time t from echelon i

tiRS Number of units in a regular order at time t from echelon i

tiES Number of expedited units at time t from echelon i

tiI On hand inventory at time t at echelon i

tiQ Quantity-on-order (pipeline inventory + backorders) at time t at echelon i

tiq Quantity ordered at time t at echelon i

tLiq ,ˆ Estimated mean lead time demand at time t at echelon i

tLi

, Estimated variance of lead time demand at time t at echelon i

tiC Total discounted objective function at time t for echelon i

tZ A set of decision variables at time t

Final customer demand at echelon 1: Realistic supply chain models account for uncertainty

in demand (Hadley and Whitin, 1967 and Porteus, 2004). Demand uncertainty is faced by the

retailer and is amplified at the upper echelons of the supply chain (Kahn, 1987; Chen et al.,

2000; and Lee et al., 1997b). Variability in order quantities at each echelon, which becomes

the demand at preceding echelons, drives demand amplification. See Lee et al. (1997a) for

various causes and industry examples of demand amplification.

89

To model customer demand at echelon 1, an autoregressive process with a lag of one

period, AR(1), is used. Autoregressive models have been successfully applied to model

demand of many products such as electricity, water, groceries, and electronics (Weigend and

Gershenfeld, 1993). The customer demand observed in period t at echelon 1 is of the

form 1111tt DD , where 01 is the mean demand, is the correlation coefficient,

and is an independent and identically distributed normal random variable, ),0( N . To

avoid negative orders, demand is truncated at 0. For analytical convenience, researchers

sometimes assume negative demand as returns (Kahn, 1987; Lee et al., 1997; Chen et al.,

2000).

Backorders, lost sales, and partial demand: If the on-hand inventory is insufficient to meet

the demand, partial demand is satisfied. Of the unsatisfied demand, a fraction is

backordered. The other 1 fraction is lost sales.

Expedited orders in response to shortages: To overcome and avoid shortages, echelons can

expedite orders. An expedited order can be produced and transported in eiL periods. Regular

orders take iL periods, where iei LL .

Following prior practice and research, we apply two triggers (thresholds) to expedite

lead times as orders are launched. Orders are expedited when the on-hand inventory falls

below an on-hand inventory threshold, tiOTh . Since this threshold triggers expediting when

on-hand inventory is low, it can help overcome shortages (Beyer and Ward, 2002). To avoid

shortages, pipeline inventory should be considered when making expediting decisions

(Fakuda, 1960; Groenevelt and Rudi, 2003; Whitmore and Saunders, 1977). When pipeline

90

inventory falls below a second threshold of tiPTh , orders are expedited. Pipeline inventory is

the sum of in-transit and on-hand inventory.

An expedited order is costlier than a regular order. Therefore, firms may expedite

partial orders (Arslan, Ayhan, and Olsen, 2001). If either of the expediting thresholds is met

at time t, only a fraction tie of the order placed in period t is expedited, while the remaining

1- tie is produced and shipped by the regular method.

Sequence of events: For each period, echelon i receives into inventory goods from the

previous echelon (i+1), launches a new order, withdraws from inventory this period’s order

by its immediate customer (i-1), initiated delivery, and updates the backorder quantity. To

elaborate, the sequence of events for echelon i are as follows. At the start of time period t,

echelon i receives into inventory regular shipped order iLtiRS , that was shipped Li periods

ago and expedited order eiLt

iES , shipped eiL periods ago from echelon i+1. This results in

updating the inventory level tiI and the quantity-on-order t

iQ .

,1 eii Lt

iLt

iti

ti ESRSII

1ti

ti QQ .

eii Lt

iLt

i ESRS

The echelon then receives an order (demand) of tiq 1 from its immediate lower echelon.

Depending on the inventory and backorders at echelon i, shipments to echelon i-1 are

determined.

)},,min(,0max{ 11ti

ti

ti

ti qbIS

where tib represents the backorders. Echelon i also places an order of t

iq to its immediate

upper echelon, which ships tiS units to echelon i. The regular shipment is t

iRS ti

ti

ti SThe )1( ,

91

while tiES t

iti

ti SThe units are expedited. t

iTh is a binary variable, which is 1 when either of

the two expediting thresholds are met.

The shortage is calculated as }.,0max{ 11 t

iti

ti

ti Iqbs

Of this shortage, a fraction is

backordered )( ti

ti sb and the rest is lost ))1(( t

iiti sl . The inventory and quantity-on-

order are then updated to compensate for shipments. ti

ti

ti SII 1

1

and ti

ti

ti SQQ 1 .

4.3.3 Ordering Policies

Each echelon/stage uses a periodic review order-up-to policy to determine the replenishment

order quantity. Zipkin (2004), Nahmias (1993), and Axsater (2000) discuss order-up-to

systems, which afford optimality for base-stock policies under certain assumptions on

backorders, order cost, and demand distributions. However, the optimality of such policies

under lost sales in not known (Jankiraman et al. 2006).

At the start of each period, the inventory position is inspected and ordering decisions

are made. The order quantity is calculated by 11

1

ti

ti

ti

ti qYYq , where t

iY is the order-up-

to quantity in period t. Traditionally, tiY is determined using a demand forecast.

tLi

tLi

ti zqY ,,ˆ , where tL

iq ,ˆ and tLi

, are an estimate of mean and standard deviation of the

lead time demand. z is a constant chosen to meet a desired service level. Our supply chain

model includes assembly of only two components. Lowering the service level, z, may help

imitate the effects from assembly of many components. Therefore, the findings from our

study are not limited to 2-component assembly systems.

.

,

,0

,1

otherwise

PThIorOThIIfTh

ti

ti

ti

tit

i

92

Final customer demand information is not shared across the supply chain in our

model. Each echelon/stage, including echelon 1, observes only the demand it receives from

its immediate echelon customer without knowledge of the underlying demand distribution of

the final customers.

4.3.4 Objective Function and Decision Variables

In order to determine the order quantity in time period t, echelon i determines the following

four decision variable values:

i) Order-up-to quantity, tiY , which can then be used to determine the order

quantity, tiq .

ii) The fraction of tiq that should be expedited, t

ie .

iii) The pipeline inventory threshold, tiPTh .

iv) The on-hand inventory threshold, tiOTh .

Traditional inventory literature typically considers objective functions that include the

cost of holding and backorder. See Axsater (2000) and Zipkin (2004) for a review. In

practice, however, lost sales and expediting costs are incurred. To make our model realistic

we include these two costs along with holding and backorder costs. The objective is an

arithmetic sum of the above costs discounted across periods. At echelon i, during time period

t, the following objective is minimized,

.)1()( 1

,,, tiiii

t Z

ti

ti

ti

ti

ti

ti

PThOTheYZ

ti CfESrRSglcbhICMin

The cost parameters associated with inventory, backordering, lost sales, regular

orders, and expediting, are h, c, g, r, f , respectively. At time period t, we determine tZ , a set

93

of four decision variable values, which when used to determine order quantities from period

1 through t, minimize tiC . Every period, the cost function is minimized to determine decision

variable values, which are used to determine the expedited and regular order quantities. We

model a decentralized supply chain which involves independent decision making at every

echelon. Echelon i makes an ordering decision by minimizing the objective function tiC

using the historical demand information from the immediate lower echelon.

The construction of objective function allows us to find dynamic, time variant

decision variables. Since the objective function places a higher weight on current period, as

compared to past periods, it is suitable for non-stationary supply chains. The objective

function and solution space react to random and non-random (i.e., trend or seasonal) changes

in demand, which then allows the search routine find decision variable that are cost effective

in a dynamic environment.

4.4 Description of Metaheuristic Search Methods

Modern products are a complex assembly of many components (often hundreds), which

require companies to be coupled. The complex interactions of material and information

between companies can make it difficult to derive optimal ordering policies. Analytical

studies approach such problems by approximating parameters such as demand, lead time,

ordering quantity, and ordering time (Axsater, 2000). However, the functions that can

realistically represent the objectives of managing a supply chain may not have properties that

make them tractable for analytical methods. As we show in section 4.5, the objective

functions considered in this chapter are not well-behaved. Therefore, we employ

metaheuristic search to find decision variable values. To provide a test of the efficacy of our

94

resulting method, we conduct a computational study comparing it to a procedure based on

Fibonacci search and also to a Genetic Search.

4.4.1 Fibonacci Search

In Fibonacci search, the local minimum of a function is found by evaluating the function

many times over possible values of a variable. The number of iterations is based on the range

of the variable and the tolerance (accuracy) of the solution required. If Fn represents the nth

term in a Fibonacci sequence and the required tolerance is , then the required number of

steps (evaluations) are n-2 if (b0 – a0) / Fn < , where b0 and a0 are the upper and lower limits

of the variable.

Fibonacci search has some major limitations as it guarantees reaching optimality only

if the objective function is unimodal on [b0 , a0], i.e., there exists a unique u[b0 , a0] such

that the objective is decreasing on [b0 , u] and increasing on [u , a0]. In the problem under

consideration, as seen in section 5, the objective function is not unimodal. Moreover,

Fibonacci search is applicable only to a single variable optimization. Our objective function

requires finding multiple variables simultaneously. Although primitive, the simplicity and

speed (run time) of Fibonacci provide motivation for its use in various applications. We use

Fibonacci search to compare with other Adaptive search methods for supply chain

applications.

4.4.2 Genetic Search

Genetic search is a widely used metaheuristic method based on a metaphor that

represents a solution as a chromosome. These chromosomes are altered using genetic

95

methods of crossover and mutation to form new chromosomes and subsequent generations.

Details of Genetic search (GS) can be found in Goldberg (1989).

We applied a basic GS procedure to our problem. A chromosome is represented by a

vector of binary digits B = [B1, B2, B3, B4], whose components Bi, i = 1, …, 4 refer to the

decision variables 1) order-up-to values, (2) expediting ratio, (3) on-hand inventory

threshold, and (4) pipeline inventory threshold, respectively. More precisely, the vector B is

given a binary representation by expressing each component Bi as a binary vector itself,

where the number ni of binary digits in each Bi depends on the desired precision of the

decision variable. The mapping between a binary chromosome and an integer/decimal

decision variable is performed by the customary mathematical representation by reference to

binary components yik, i Ki = {0, …, ni} (hence the number of these variables is actually ni

+ 1), to give Bi = ∑(2kyik: k Ki).

Each generation contains P chromosomes. With a probability pc, a chromosome is

selected for crossover. The location of crossover is randomly selected. A single point

crossover is performed that results in two offspring from two parents. With probability pm, a

binary digit of a chromosome is flipped (from 0 to 1 or from 1 to 0). The offspring are added

to the population pool, increasing the size of the population beyond P. The binary

chromosomes are mapped to decimal/integer decision variable values. These values are used

to calculate the order quantity (regular and expedited). The resulting costs are recorded.

The costs define the fitness of a chromosome. Using an elitist strategy, chromosomes

are sorted in increasing order of cost. The best P chromosomes are selected to become the

next generation population. The procedure is repeated until a pre-set limit on the number of

generations is reached.

96

4.4.3 Tabu Search Strategies for Supply Chain Applications

Tabu search (TS) is a metaheuristic based on the use of adaptive memory. The simplest form

of this memory stores the attributes of recently selected moves and exploits this information

by forbidding new moves that would cause the resulting solution to contain attributes of

solutions recently visited. Forbidden attributes (and hence moves) are called tabu, and the

number of iterations that an attribute remains tabu is called its tabu tenure. The simplest

versions of TS apply the same tabu tenure to each attribute. In the present context, we may

conveniently allow the decision variables and the values assigned to them to constitute the

attributes of interest.

The memory embodied in tabu search has several functions. The simplest, but highly

useful, function is to prevent the method from becoming trapped in local optima by allowing

non-improving moves to be made provided they are not tabu. Additional flexibility is

provided by means of aspiration criteria which, if satisfied, permit the tabu status of a move

to be overridden, and hence allow certain “sufficiently attractive” moves to be selected in

spite of being tabu. Tabu search has been successfully used in a wide variety of applications

ranging from scheduling, TSP, network design, architectural design, image recognition,

among others. See Glover and Laguna (1997) for a general coverage of TS strategies and

applications, and see Stecke (2005) for some recent supplementary application examples.

Although a supply chain application using tabu search is reported in Glover and Laguna

(1997) to find optimal distribution strategies, to date TS has not been used to optimize

ordering decisions in a complex supply chain.

A description of our application of tabu search to minimize total cost in a multi-stage

supply chain is now given. For illustration and notational convenience, when appropriate,

97

superscripts t and i are dropped. We also slightly change the notation used to describe the

Genetic Search approach since we no longer refer to binary representations. Let x = (x1, x2,

…, xn) denote a solution vector, where in the present supply chain context n = 4. Our TS

method proceeds from a current solution x = xo to a next solution x' N(xo), where N(xo)

denotes the neighborhood of xo. More specifically, TS requires x' to be an element of an

associated neighborhood N'(xo) = (N(xo) – T) A. equivalently, N'(xo) = N(xo) – (T – A)),

where T denotes the current set of tabu solutions and A identifies those solutions in N(xo) that

satisfy the aspiration criteria. Such criteria typically include default conditions to assure that

N'(xo) is not empty, in cases where the tabu set T would otherwise contain all of N(xo).

Details of the aspiration criteria we employ with our method are provided in Section 5.

We select the particular solution x N(xo) that yields the best objective function

evaluation; that is, x = arg min(C(x'): x' N'(xo). When x is thus selected, attributes of xo

that are changed by the move that produces x are labeled tabu, thus preventing a move that

revisits xo until the tabu tenure attached to these attributes expires. Thus, implicitly, this

causes the tabu set T to be enlarged by adding a set T(xo) consisting of xo and all solutions

that share the attributes of xo that are labeled tabu, In practice, it is not necessary to identify

T(xo) explicitly, since it is sufficient merely to forbid moves that would reinstate tabu

attributes. (The operation determining tabu status in the present context will likewise be

clarified in Section 5.)

For simplicity of description, however, we may summarize a simple form of TS by

the following pseudo code.

98

Tabu Implementation Pseudo Code

Step 1. Choose an initial solution xo, and identify the best solution found so far as x*= xo. Let Iteration=0, and T = and A =.

Step 2. Set Iteration=Iteration+1 and generate the set of solutions

N'(xo) = (N(xo) – T) A.

Step 3. Choose the best solution x N'(xo) identified by

x = arg min(C(x'): x' N'(xo)).

Step 4: Identify the tabu attributes associated with the move from xo to x, thus

implicitly identifying the set T(xo) and updating the tabu set T := T(xo)(T –

E), where E is the subset of T consisting of solutions whose tabu tenure has

expired. (T – E likewise is known implicitly simply by reference to attributes

that remain tabu after eliminating those whose tenure expires.).

Step 4. If C(x) < C(x*) then set x* = x. Update the aspiration set A by reference to

the quality of x*.

Step 5. If Iteration has reached a selected cut-off value, then stop. Else go to Step 2.

We remark that the choice of x as the best solution in the current neighborhood N'(xo)

is often modified, particularly in the case of large neighborhoods, by employing a candidate

list strategy that identifies x as the best solution from a strategically generated subset of

N'(xo). Our method for the current supply chain model employs a simple instance of such a

strategy, as noted below.

Definition of neighborhood: We select an elementary neighborhood N(xo) that constitutes the

set of solutions obtained by changing (increasing or decreasing) exactly one of the

components xjo of the vector xo. Thus a vector x' N(xo) is generated by setting xr' = xr

o ± Δr

for a chosen index r, and xj' = xjo for j ≠ r, where Δr is a positive multiple of a chosen fixed

increment δr. We have chosen δ1 = 1 (for order-up-to values), δ2 = 0.01 (for the expediting

99

ratio), and δ3 = δ4 = 1 (for the two expediting thresholds). This construction is based on the

intuition that for appropriate (sufficiently small) δj values, the objective function will often be

monotonic over some region of positive multiples of these values.

Candidate list strategy and intensification: The search is carried out in two steps. The first

step is a coarse search, which searches the neighborhood of a solution using a coarse

increment/decrement. In a coarse search, a neighborhood is obtained by replacing the values

δj by larger values δj+, where δ1

+ = 10, δ2+ = 0.05, and δ3

+ = δ4+ = 10. Such a strategy allows

an accelerated search through a wide range of decision variable values. The best solutions

obtained from a coarse search become candidates for starting solutions for a finer search that

employs the fundamental δj values. This rudimentary candidate list strategy not only saves

time in its coarse-grid phase, but also ensures that the search during the finer-grid phase is

intensified in regions around candidate solutions obtained by coarse search.

Diversification by reference sets: At each time period, multiple starting solutions are used to

ensure diversification of a search space. A diverse set of preferred candidates for initial

solutions are generated using a Tabu reference set, RefSet.

RefSet begins empty and during the first few periods the starting solutions used to launch the

search is chosen randomly. Each starting solution is added to RefSet as it is selected. After a

specified number of periods, the starting solutions are instead selected from among the

current candidate solutions by choosing the one that maximizes the minimum distance from

all elements of RefSet. The distance D(xo, x) between a candidate starting solution xo and a

solution x RefSet is defined to be

D(xo, x) = ∑((xjo – xj): j J)

100

where J = {1, 2, …, n} is the index set for the components of x (and where in the present

case, n = 4).

A pseudo code to for generating candidate solutions and starting solutions can be

outlined as follows, where F denotes a current set of feasible decision vectors obtained by

sampling, and Eo denotes an exclusion set consisting of all of the best decision vectors x*

identified in previous periods.

Algorithm to Find Candidate Solutions (adapted from Glover, 2006)

Step 1. Begin with RefSet to consist of the randomly generated starting solutions for

the first s periods (for s small, e.g., s = 10)

Step 2. Select xo F – Eo by the rule:

xo = arg max min (D(xo, x): x RefSetEo)

Let RefSet.= (RefSet – x#) xo, where x# denotes the oldest solution in RefSet.

Step 3. Repeat Step 2 until a specified number of starting solutions are obtained.

The definition of D(xo,x) is motivated by the goal of inducing the new initial solution xo to lie

in a region that is unexplored by previous iterations. The number of data points considered is

limited by dynamically updating the members of RefSet, by adding the most recent, solution

xo and deleting the oldest x#. The exclusion set Eo is used to diversify the search and to focus

on regions away from known local optima.

In addition to generating starting solutions in the manner described, we also use the

most recent best solutions x* as starting points in subsequent time periods. Note that in most

periods, especially in the absence of disruptions, the best solutions will tend to have the

property whereby the values of goods flowing from one period to the next will not vary

greatly across the periods.

101

Handling Tabu status: To keep track of the Tabu status of an order-up-to value, a two

dimensional matrix, the Tabu status matrix, TSM(j, z), j J, z Z, is used, where Z

represents the set of values that can be assigned to the variables xj of x, TSM thus consists of

4 rows (one for each index in J) and a number of columns equal to |Z|.

The matrix stores the iteration count up to which a decision variable and value

combination is Tabu, i.e., forbidden. For example, if from iteration Iter to 1Iter , the order-

up-to value moves from 1Y to 2Y , IterYTSM )1,1( . This implies a Tabu tenure of

periods, i.e., order-up-to values may not move back to 1Y until iteration Iter . In general,

solutions closer to 1Y are expected to have a performance similar to 1Y . Therefore, Tabu

tenure is assigned, although shorter than Iter , to cells in the neighborhood.

')11,1( IterYTSM , ')11,1( IterYTSM , ")21,1( IterYTSM , and

")21,1( IterYTSM , with "' .

Also, for the order-up-to value to become 2Y , we must have first examined its

neighborhood to find 2Y to be the most preferred. So likewise, we wouldn’t want to move

very soon to 2Y or one of its nearby values. Therefore, point 2Y and its neighbors are also

added to the Tabu set, i.e., IterYTSM )2,1( , ')12,1( IterYTSM ,

')12,1( IterYTSM , ")22,1( IterYTSM , and ")22,1( IterYTSM .

The search algorithm starts with all entries of TSM as zero, which implies that all

decision variable values are acceptable. After each iteration the to and from cells of TSM are

updated. When of encountering a cell that already contains a Tabu tenure entry from previous

iterations, tenure with a larger value takes precedence.

Flexible aspiration criterion. Dimensioned similar to TSM, an aspiration matrix

),1(),6,1();,( ZZkZkAM stores in each cell the best (smallest) objective function value

102

achieved when this cell was reached in a previous move. Thus, upon making the move from

1Y to 2Y , the objective function value changes from )1(YC to )2(YC and

)}2,1(),2(min{)2,1( YAMYCYAM . This (minimum) value should also be entered in the

adjacent cells that have also been affected by the move, hence setting

)}12,1(),2(min{)12,1( YAMYCYAM and )}12,1(),2(min{)12,1( YAMYCYAM .

If a move creates an objective value smaller than the value in the AM cell that is being

moved to, the aspiration criterion will allow the move even if it is Tabu. This evidently

permits more flexibility than an approach that overrules Tabu status only if a new best

objective function value is achieved. Yet it also assures that no solution will be revisited.

4.5 Results

To represent a variety of operating conditions, we selected a range of cost parameters as

presented in Table 3.2 (See previous chapter). Although literature reports a wide variation in

the choice of cost parameters, we chose the parameters based on our case studies. For

example, Daniel and Rajendran (2005) used a holding cost between 1-8 and backordering

cost between 2-32. Our selection falls approximately in the range of the cost parameters

found in literature. See Schmitt et al. (2009) for a review of cost parameters. Relative costs

depend on the type of product, industry, and supply chain, among other factors. Cohen et al.

(2003) observed that with short life cycles and obsolescence in the semiconductor industry,

the cost of losing a sale is about twice the cost of backlogging, while the holding cost is one

third the cost of backlogging. By contrast, Faaland et al. (2004) experimented with lost-sales

costs in a single-echelon ranging from 62 to 500 times the periodic inventory holding cost,

yielding shortage values much higher than the ones we chose. Their parameters were based

on a cross-industry survey by Boer and Jeter (1993). Our motivation in the choice of

103

relatively low costs is to compare the metaheuristic methods under modest conditions found

in our case studies. We observed in pilot experiments that larger shortage costs relative to

inventory carrying costs further exaggerated differences between Tabu, Genetic, and line

search.

4.5.1 Implementation of Metaheuristics

We model a decentralized supply chain where echelons do not share demand or inventory

information. In every period, a echelon observes the demand information from its immediate

customer. Using the historical demand information, echelon i determines a set of four

decision variables, tiZ , that minimize the total cost function, t

iC . Demand originates with the

customer. Echelon 1 observes this demand and finds tZ1 , which is then communicated to

echelon 2. Echelon 2 then uses this new demand information—along with historical

demand—and finds tZ2 . The process is repeated for echelons 3 and 4. As a representative

echelon, we record and present the costs incurred at echelon 1. Echelon 1’s proximity to the

final customer may make it an important decision maker. We do not suspect the results and

conclusions in this chapter to be different if any other echelon was chosen for cost

comparison.

Fibonacci Search: As Fibonacci search is a type of gradient search, it can be used for

searching only one variable at a time. Therefore, we performed a sequential search on the

four decision variables. The search started with finding local minima for order-up-to at

echelon 1. Subsequently, expediting level, pipeline threshold, and on-hand threshold

inventory were searched. The same procedure was repeated for all other echelons.

104

Genetic Search: A Genetic search construction very similar to that described in Chapter 3 is

used. The binary chromosome length was increased to accommodate four decision variables.

A chromosome of 45 binary digits was selected. The four decision variables, order-up-to

level, fraction expedited, pipeline inventory threshold, and on-hand inventory threshold are

each represented by 15, 7, 9, and 11 binary digits, respectively. Such a representation limits

the lower and upper limits of the values on which these decision variables are searched. For

example, order-up-to level is searched between the values of 0 ( 02 -1) and 32767 ( 152 -1),

which are obtained by converting a binary chromosome to a decimal representation. The

number of binary digits chosen to represent each decision variable is large enough to cover

the global minima.

Specific experimental parameters were as follows: A population size of 100 was used.

After experimentation, crossover and mutation probabilities of 0.5 and 0.005, respectively,

were selected. These probabilities resulted in fastest convergence. An elitist selection

criterion was used. Each run involved 100 generations. The results presented in this chapter

are the average of 100 independent runs.

Adaptive Tabu Search: Tabu search is performed over four decision variables using the

neighborhood structure presented in Section 4.3. Each run starts with 20 randomly generated

solutions. After every run, 10 solutions are saved and used as a starting point for the next run.

Another 10 solutions are generated so as to cover the unsearched regions of feasible solution

space. (Note that in the absence of disruptions, the solutions may vary very little across

periods.) The second set of diversified solutions helps cover the regions that were previously

unsearched. (See Section 4.3 for the mechanism to generate these diversified solutions.) Each

105

Table 4.1. Cost comparison of various metaheuristic solution methods.

Backordering Cost

Cost of Lost Sales

Carrying Cost

Expediting Cost

Line Search

Genetic Search Fibonacci

Genetic Search Tabu Search

Decision Variable: Order-up-to quantity

Decision Variables: Order-up-to quantity, Expediting level, Pipeline inventory threshold, On-hand inventory threshold

Low Low Low Low 551.01 472.38 528.23 469.32 462.35

Low Low Low High 606.78 493.61 584.3 443.26 433.86

Low Low High Low 808.55 685.79 789.78 692.34 670.74

Low High Low Low 621.22 496.91 621.01 446.89 439.16

High Low Low Low 690.61 516.98 681.09 477.85 468.29

High High Low Low 764.22 544.26 761.28 512.39 479.59

High Low High Low 996.62 922.66 903.26 938.05 927.19

High Low Low High 742.80 524.13 741.21 508.16 493.12

High High High Low 1028.71 927.69 993.86 833.56 783.66

High High Low High 797.62 551.08 774.93 489.56 461.28

High Low High High 994.61 922.56 971.22 942.21 908.99

Low High High High 974.21 896.36 972.56 846.46 846.11

Low High High Low 873.29 812.41 871.85 779.68 765.29

Low High Low High 668.32 510.28 669.52 461.26 453.84

Low Low High High 852.02 833.00 832.12 736.27 729.54

High High High High 1071.11 941.23 1062.42 933.59 861.52

105

106

Table 4.2. Percentage cost comparison.

% Improvement

Fibonacci (Multiple Variables)vs. 2.17Line Search (Single Variable)Genetic Search (Multiple Variables)vs. 4.89Genetic Search(Single Variable)Tabu Search (Multiple Variables)vs. 7.84Genetic Search (Single Variable)Tabu Search (Multiple Variables)vs. 3.10Genetic Search (Multiple Variables)

Note: A single variable search involves searching order up-to values. Multiple variable searches are performed over four decision variables as discussed in this chapter.

initial solution’s neighborhood is searched for 100 iterations. The results shown in this

chapter are the average of 100 independent runs.

4.5.2 Cost Comparison between Search Methods under Static Demand

Table 4.1 shows the average costs for 100 independent runs under the three metaheuristics

methods used in this chapter. Each run covers 2000 periods. The first 1000 periods are used

to reach a steady state. Data is collected from period 1001 through period 2000. Variance

reduction is achieved through the use of 50 pairs of antithetic variables (Law and Kelton,

2000). The method of antithetic sampling helps in reducing sampling errors. The percentage

of cost savings by using various methods is presented in Table 4.2. For comparison purposes,

the tables also include the costs obtained in Chapter 3, where only order-up-to quantity was a

decision variable.

The costs in Table 4.1 clearly illustrate the importance of searching over multiple

variables. Table 4.2 summarizes the cost information in Table 4.1. The results show the

107

importance of optimizing expediting variables. When searched over four decision variables,

Genetic search resulted in an average savings of 4.89% when compared to search over only

order-up-to quantity. Even simple local search method of Fibonacci resulted in a cost savings

of over 2%. In the multi-variable optimization, Tabu search provides an added improvement

of 3.10% over Genetic search.

To test the consistency of methods we compared the variability of solutions. Table

4.3 presents the average cost variability for 100 runs.

We computed cost variability in 100 independent runs in each cost combinations.

Table 4.3 presents the average for 8 cost combinations. Besides lower cost, Tabu search also

reaches the best solutions consistently. A low variability from Tabu search shows

consistency of the method over Genetic search, which implies that Tabu search should be

expected to reach the best solutions in less number or runs. Also, if constrained by

computation power and time, Tabu search should be preferred.

Table 4.3. Consistency of metaheuristics methods.

Average Cost Variability

Fibonacci 24.27Genetic Search 12.18

Tabu Search 9.21

108

Table 4.4. Sample results.

Run

Quantity Shipped per Period

Lost Sales

Non-expedited Orders

Expedited Orders Backorders Inventory Cost

Fibonacci1 99.01 11.49 76.27 24.00 30.66 129.46 522.512 98.69 11.82 76.39 24.62 33.20 129.33 536.223 99.04 11.96 75.94 22.65 34.28 128.47 533.744 98.77 12.49 73.92 27.21 29.29 130.88 535.525 99.07 12.19 83.59 26.39 41.52 127.32 573.04

Genetic Search1 106.45 4.02 106.93 10.74 9.33 181.70 458.322 106.15 4.37 105.94 11.77 10.50 190.34 496.213 107.47 3.51 107.03 7.43 9.43 181.39 466.274 107.38 3.94 107.73 8.56 10.00 183.80 473.115 106.50 4.79 109.79 10.85 12.75 203.81 498.88

Tabu Search1 107.02 3.41 108.62 8.29 7.79 187.76 451.332 107.24 3.29 111.40 9.28 9.04 184.17 488.313 107.55 3.44 112.98 8.64 9.39 185.39 449.874 107.32 4.01 116.19 10.63 10.49 189.96 466.355 107.17 4.09 117.14 11.51 10.19 199.38 442.32

108

109

400

500

600

700

800

900

1000

1100

1200

1300

1400

0 10 20 30 40 50 60 70 80 90

Run

Cos

t

Tabu Search

Genetic Search

Figure 4.1. Sample convergence.

Figure 4.2. Average convergence in 100 runs.

110

Table 4.4 shows sample results from the runs. The table includes only the first five

runs (of 100 runs). It is clear that the service level achieved by Genetic search and Tabu

search exceeds that of Fibonacci search. Table 4.4 also provides average quantity shipped to

the final customer, average lost sales per period, expedited and non expedited orders placed

by echelon 1, and backorders and inventory at echelon 1.

Figures 4.1 and 4.2 shows a sample convergence and average convergence for Tabu

and Genetic search. We observed that Tabu search took significantly lesser time (about 30

minutes) as compared to Genetic search, which took over 12 hours on a laptop computer.

Genetic search converge faster in the first 0-10 iterations, which could be attributed to the

power of mutation operator in exploring new solution regions. Also noticeable from the

figures is the number of iterations needed to converge. In most instances Tabu search took

about 40 iterations to converge. In contrast besides taking significantly higher computer run

time, Genetic search exhibited convergence after 80 iterations. In 7 out of 1600 instances,

Genetic search resulted in superior solutions. This may show that provided enough

generations, Genetic search may narrow the cost gap. However, the computational burden

would also increase substantially.

4.6 Comparison between Static and Dynamic Policy

To outline the benefits of dynamic policy and the power of Adaptive search in searching

optimal dynamic policies we test the heuristics under a set of non-stationary demands. We

consider a seasonal and disrupted demand.

111

4.6.1 Adaptive Search Methods under Seasonal Demand

We model the effects of a seasonal demand with a sinusoidal demand function.

where SD(t) represent the seasonal demand at time period t and A the amplitude of seasonal

demand. The average demand function for 100 runs of a cost combination is shown in Figure

4.3. Since the modeled demand function has symmetric seasonality, the mean long term

demand does not change as a result of seasonality. In such a demand scenario one may

expect static policies to have a fair long term performance.

To understand the effects of seasonality we limit ourselves to a minimum set of

conditions. We allow only one season. The season starts in period 1601 and ends at period

1850, which could represent 250 working days in a year and one season. The modeled

amplitude is 50. Three different ordering strategies were considered.

Static Policy with Past Information (SPPI): SPPI is a static ordering strategy which uses past

demand information to estimate a static ordering policy. This policy is then applied to make

ordering decisions in the future. In our case, first 1000 periods are used to establish a steady

state. Using demands for periods 1001 though 1500, Tabu search determines static policy

parameters. This policy is then applied for periods 1501 though 2000 and costs are recorded.

SPPI is a Level Strategy where ordering policies are kept fixed despite seasonality.

Static Policy with Demand Information (SPDI): To estimate the best static strategy we

assume that the demand information is known in advance and determine the policy that

minimizes the total cost. In our case we assume that demands (including seasonality) for

112

periods 1501 through 2000 are known in advance. SPDI minimizes the costs for periods 1501

thought 2000. If the search routines reaches is able to reach the optimal solution, SPDI

should be the global minimum for all static policies.

Dynamic Policy (DP): Under a dynamic policy, each period ordering policies are determined

using the past information. No assumptions are made about future demand. Since the demand

is seasonal from period 1601, we expect the dynamic policy to chase the changes in demand

function, albeit at a slower rate because of discounting.

In all three cases we use Tabu search to optimize the ordering variables of order-up-to

quantity, proportion of orders expedited, and expediting triggers. In the case of SPPI and

SPDI a one-time search is used to find the policies. Under DP, the search is performed every

period as the demand information unfolds. Table 4.5 lists the results in terms of total costs for

the three methods. The results are summarized in Table 4.6. As expected SPDI, which

assumes prior demand information, result in a cost improvement of 1.4% over SPPI. The

improvement is not as large as we had expected. This could be attributed to the symmetric

seasonality which maintains average long term demand to a fixed level. To counter the

increase in demand in upcycle, the order-up-to quantity were marginally (1053 to 1061)

higher in SPDI as compared to SPPI. Any increase in ordering quantity in SPDI may reduce

the shortages in the upcycle but at the expense of carrying more inventories for the rest of the

periods.

DP results in improvements of 3.2 % over static policy of SPPI. The improvement

underlines the benefit of using dynamic policies. We also observed a mean cost improvement

of 1.8% in using DP over SDPI, which assumes complete demand information. Note that

SDPI is expected to the best static policy. We expect these percentage improvements to

113

increase with amplitude of seasonality. These results illustrate that Tabu search method

developed in this chapter can be successfully used for ordering decisions in nonstationary

conditions and are superior to static policies. Figure 4.4 shows a smoothed response in order-

up-to quantities at echelon 1. The order-up-to quantities change along with demand

seasonality. The discounting of information helps in smoothing the response. Vertical lines in

Figure 4.3 represent the periods of start, end, crest, and trough of the seasonal demand. The

order-up-to values do not exactly follow the seasonal demand. There is a delay in the

seasonality induced in order-up-to quantity because of discounting. Also the ordering patter

return to pre-season level within 25 periods of the end of seasonality, ensuring that the

inventories in the supply chain do not overreact to seasonal demand.

Figure 4.3. Seasonal demand function.

114

Table 4.5. Performance of ordering policies under seasonal demand.

Backordering Cost

Cost of Lost Sales

Carrying Cost

Expediting Cost

Static Demand Seasonal Demand

SPPI SPDI DPLow Low Low Low 462.35 499.51 492.35 482.21

Low Low Low High 433.86 457.23 453.20 452.34

Low Low High Low 670.74 699.13 682.13 677.93

Low High Low Low 439.16 453.28 449.65 442.40

High Low Low Low 468.29 503.16 499.72 480.29

High High Low Low 479.59 520.94 512.49 502.33

High Low High Low 927.19 992.46 977.18 965.36

High Low Low High 493.12 545.57 533.11 515.20

High High High Low 783.66 833.35 826.64 819.54

High High Low High 461.28 509.89 499.48 477.58

High Low High High 908.99 989.52 974.65 953.31

Low High High High 846.11 870.72 869.95 856.58

Low High High Low 765.29 779.65 775.46 772.39

Low High Low High 453.84 515.28 496.26 471.18

Low Low High High 729.54 777.83 768.24 748.23

High High High High 861.52 898.82 883.72 882.83

Table 4.6. Percentage improvements in using dynamic policies.

Percentage improvements

SPDI vs. SPPI 1.40DP vs. SPPI 3.2DP vs. SPDI 1.8

115

Figure 4.4. Order-up-to levels for echelon 1 under a seasonal demand.

4.6.2 Adaptive Search Methods under Disruptions

Disruptions violate the stationary conditions needed for optimality of static order-up-

to policies. Our design of objective function and adaptive search methods allows us to find

dynamic ordering policies that are intuitively superior to static policies under disruptions. As

discussed earlier, our objective function has similarities with exponential smoothing, where

the smoothing constant is set depending on variability. Similarly, the discount factor, , in

the objective function can be set depending on variability in the environment. For example,

under stationary and stable conditions, should be set at a low value, possibly in the range

of 0 - 0.3, which is also considered the preferred range for the smoothing constant in

exponential forecasting (Schroeder, 2008). During or just after a disruption, a high value of

may be warranted. A higher value would make the search methods more responsive to

information in current periods and give less weight to periods long past.

Our algorithm and objective function make our method ‘reactive’ to disruptions. The

methodology is designed to look at the past information, and adapt so as to make ordering

116

decisions that minimize the costs. The method should also be helpful in recovering from

disruptions. In general, proactive strategies are preferred over reactive strategies (Stecke and

Kumar, 2009). However, many disruptions may not be foreseen (ex., earthquakes, terrorist

attacks, accidents) and, for others, the ending time may not be certain (ex. loss of supplier,

political unrest, workers’ strike), making it difficult to forecast the future. In such instances, a

reactive strategy may be the best option, and our methodology could be helpful.

To test the performance of adaptive search under disruptions we device an

experimental setup similar to Section 4.6.3.1 with the demand function shown in Figure 4.5.

The figure represents a disruption in customer demand between the periods of 1601 and

1650. During disruptions, no customer demands are placed at echelon 1. A transportation

strike or a natural catastrophe could be responsible for such disruptions. We also assume that

demand during disruptions is not backlogged but is entirely lost. The costs from SPPI, SDPI,

and DP are presented in Table 4.7 and summarized in Table 4.8. The results show that

dynamic policies obtained by using Adaptive search methods outperform the best static

policy. The average cost advantage of DP over SPPI and SDPI are 4.77% and 0.83%,

respectively. We expect the ratios to be higher as the length or severity of disruption

increases. A response in order-up-to values at echelon 1 because of disruptions is shown in

Figure 4.6. As with the seasonal demand, under disrupted demand, order-up-to quantities

change as the disruption progresses and then return to normal after disruption ends.

117

Figure 4.5. Demand function with disruptions in final customer demand

Table 4.7. Performance of ordering policies under disruptions

Backordering Cost

Cost of Lost Sales

Carrying Cost

Expediting Cost Disrupted Demand

SPP SDI Dynamic

Low Low Low Low 499.51 473.16 471.25

Low Low Low High 457.23 453.98 447.97

Low Low High Low 699.13 649.20 650.18

Low High Low Low 453.28 427.29 424.87

High Low Low Low 503.16 475.56 468.06

High High Low Low 520.94 510.68 507.02

High Low High Low 992.46 942.31 941.40

High Low Low High 545.57 508.41 491.52

High High High Low 833.35 806.54 800.55

High High Low High 509.89 487.23 479.96

High Low High High 989.52 958.01 954.43

Low High High High 870.72 858.30 852.56

Low High High Low 779.65 760.62 751.93

Low High Low High 515.28 468.82 466.38

Low Low High High 777.83 751.27 742.44

High High High High 898.82 884.12 878.46

118

Table 4.8. Percentage improvements in using dynamic policies.

Percentage improvements

SPDI vs. SPPI 3.97DP vs. SPPI 4.77DP vs. SPDI 0.83

Figure 4.6. Order-up-to levels for echelon 1 under a disruption.

4.7 Extensions to a Global Supply Chain

Our model and proposed methodology is flexible enough to accommodate a variety of

problems, instances, and constraints. In this chapter we used local planning and decision

making; however the methodology and search methods developed can be applied to a

centralized supply chain where ordering decisions are made with a global objective.

Under global or centralized planning, a single decision-maker determines the ordering

policies for all six stages with the objective of minimizing the total supply chain cost.

Centralization of information and decision-making has been shown to create value in supply

chains (Cachon and Fisher, 2000). For the supply chain model in this chapter, a global

119

optimization problem will have 24 decision variables (four variables for each stage times six

stages) to be searched simultaneously. The objective function includes the sum of costs at all

six stages. At time period t, the following objective is minimized.

.)1()( 1

}6,...,1{,...,,, 621ttttt GZ

t

ii

ti

ti

ti

ti

ti

ZZZGZ

t GCfESrRSglcbhIGCMin

(GC represents the global cost and GZ represents the set of decision variables)

The objective is similar to the local objective presented in Section 4. It is easy to see

that the new objective does not require any significant changes in the Tabu search

methodology developed in Section 5. However, the dimensions of Tabu Status Matrix, TSM,

and Aspiration Matrix, AM, must be increased to accommodate the increase in number of

decision variables. The global problem also requires a significantly higher computation

power. For the local objective problem, Tabu search run time was significantly lower than

Genetic search run time. We expect this difference in run time to be many folds greater for

global problems, indicating the superiority of Tabu search over Genetic search.

4.8 Conclusions

Inventory management decisions are critical for efficient and effective operations in a supply

chain. We consider the problem of making ordering decisions in a periodic review multi-

stage supply chain. The problem under general but practical conditions is difficult to handle

analytically and computationally tedious (Clark and Scarf, 1960; Lee and Billington, 1993).

We propose an alternate methodology that involves using adaptive search methods to search

for optimal solutions. Besides order-up-to quantity, our decision variables were percentage of

orders expedited and two thresholds for expediting. We compared the solutions from

adaptive search with Genetic search and Fibonacci search. In over 99.5% of the problem

120

instances, Adaptive search resulted in solutions superior to Genetic search. Although we

cannot claim universal superiority of adaptive search over Genetic search, under the supply

chain structure modeled and the cost combinations which were derived from case studies, we

observed adaptive search outperforms Genetic search in both solution quality and run time.

To increase the efficiency and run time of adaptive search, we explored properties of

feasible solution space and implemented problem-specific adaptations; namely, candidate list

strategy, intensification, reference set, and a flexible aspiration criterion. Our objective

function has the ability to capture the environmental variations that may impact supply chain

operations. Using such an objective function, we found a time variant dynamic policy.

We also used the Adaptive search methods developed under seasonal and disrupted

demand scenarios. The results show that adaptive search methods can be an effective tool in

practical situations where customer demand could be nonstationary. Our model can be easily

adapted to help decision-making under a variety of constraints and supply chain

configurations. The flexibility of adaptive search methods and their problem-specific

design—along with the power of adaptive memory—could be a useful tool for making

supply chain ordering decisions in practice.

121

CHAPTER 5

CONCLUSIONS

In this dissertation, we addressed the problem of managing supply chains under disruptions.

We focussed on the need for higher awareness of catastrophes and their effects on supply

chain operations. To achieve this we compile catastrophe and economic data from many

sources to reveal information of managerial and academic importance and showed that there

is an increase in both the number and dollar value of catastrophes that impact U.S.

businesses. Our work can help managers make better plans to manage disruptions make

better planning decisions. Our work also contributes to supply chain literature by identifying

vulnerability-causing factors, which when affected by a business decision can impact the

vulnerability of a supply chain. To manage dsiruptions we propose and overview a

comprehensive set of mitigating strategies that can help reduce the risks of disruptions. We

reveal potential benefits of following these strategies during normal times. These can help

justify investment in catastrophe mitigation. We also identify 16 research problems in the

area of supply chain disruptions management. In this dissertation we address some of these

problems.

We describe aspects of a stream of research to assess the economic impact of supply

chain disruptions. The overarching research mission of our sponsor, Sandia National

Laboratories, is to develop a large-scale simulation model that depicts regional and national

economic behavior. Simulation is intended to augment existing analytical and statistical

models whose utility may depend on the validity of inherent simplifying assumptions. Such

122

optimization models routinely assume aggregation of demand and supply data,

substitutability of supply options, independent and steady-state behavior of underlying

stochastic distributions, optimization over well-behaved objective functions, and simple

supply chain structures. Sandia and Los Alamos National Laboratories formed NISAC to

address the prevention, response, and recovery from disruptive events. The concern was that

with such high stakes, entities within the U.S. government and private sectors cannot afford

to wait for researchers to overcome the substantial challenges of relaxing simplifying

assumptions in the optimization models. We anticipate that our guidelines will be useful in

other research as well. Our case studies of three electronics firms as surrogates for general

industrial behavior in the region, and drew from the cases to offer a fundamental set of

design requirements, performance drivers, and important research questions. Our findings led

to recommending a minimum set oc conditions that Sandia may need to model.

We formulated three Key Questions, to provide insight into the cost function and

consequences of common supply chain mitigation strategies. A primary finding was that the

system cost function can be quite ill-behaved in a four-echelon supply chain, even in the

absence of disruptions and expediting. We reveal a weakness of analytical optimization

approaches in the present setting by providing an example where these methods would fail to

obtain satisfactory solutions. Analytical methods that assume convex or unimodal behavior

may be inappropriate for real-world supply chains. Another finding confirms a conjecture

that disruptions may have long-lasting, rippling, and costly consequences within the supply

chain structure presently considered. A third finding is that standard industry practices of

setting order parameters locally and using expediting as mitigation seem to exacerbate these

undesirable effects.

123

Our experimental results suggest caution, however, regarding mitigation efforts and

resiliency. Information sharing offers clear advantages in a supply chain, and many authors

have addressed this issue (e.g., Milgrom and Roberts, 1988; Lee, 2000; Chatfield, Kim,

Harrison, and Hayya, 2004; Sodhi, 2005). As a caveat, our experiments disclose that

information must be discounted considerably to control bullwhip effects in moving

downstream in the supply chain, away from the source of uncertainty. We derived a very low

forecasting weight of 0.01 on the most recent local information for the firms in the last

echelon, Echelon 4. While we support the notion that increased information and flexibility to

react are generally desirable, we caution that it is possible to overreact. A disruptive event

creates a critical watershed. The issue is how much weight to place on the news of such an

event and what to do about it. Furthermore, with the irregular system objective surface

documented by our results, we believe future efforts should be directed towards developing

efficient and effective search methods to find parameters yielding local optima close to

global values.

In Chapter 4 we develop metaheuristic models for ordering decisions in supply

chains. We consider the problem of making ordering decisions in a periodic review multi-

stage supply chain. We propose an alternate methodology that involves using adaptive search

methods to search for optimal solutions. Besides order-up-to quantity, our decision variables

were percentage of orders expedited and two thresholds for expediting. We compared the

solutions from adaptive search with Genetic search and Fibonacci search. In over 99.5% of

the problem instances, Adaptive search resulted in solutions superior to Genetic search.

Although we cannot claim universal superiority of adaptive search over Genetic search,

under the supply chain structure modeled and the cost combinations which were derived

124

from case studies, we observed adaptive search outperforms Genetic search in both solution

quality and run time.

We also used the Adaptive search methods developed under seasonal and disrupted

demand scenarios. The results show that adaptive search methods can be an effective tool in

practical situations where customer demand could be nonstationary. Our model can be easily

adapted to help decision-making under a variety of constraints and supply chain

configurations. The flexibility of adaptive search methods and their problem-specific

design—along with the power of adaptive memory—could be a useful tool for making

supply chain ordering decisions in practice.

In summary, this dissertation contributes to supply chain management literature in

planning and operational inventory decisions under disruptions. We reveal the environmental

and inherent vulnerability-causing factors in supply chain that are contributing towards a

higher rate and dollar damage of disruptions. The strategies identified can be useful for

supply chain decisions. Our model and insights from Chapter 3 should also help develop

models that are close to real situations. Our alternative metaheutistic methods could be of

practical relevence as it can be used under various complex real life constraints.

125

APPENDIX A

THE SANDIA AGENT-BASED MODEL

Why the Northwest?

NISAC targeted the Northwestern U.S. as the first study site because it represents a

significant, but relatively isolated microcosm within the U.S. economy, and it has

international borders that are attractive for illegal crossings, important cultural traits, and

desirable targets for terrorists. Rough terrains in the Cascade Mountains, where Washington

and Idaho border Canada, make it attractive for individuals to cross illegally. The long

seacoast in the Puget Sound is vulnerable as well. The San Juan Islands provide good cover

in a sparsely patrolled environment with multiple entry routes into the United States. One

pre-9/11 attack was thwarted by an alert U.S. Customs Agent -- Ahmed Ressam was

apprehended on December 14, 1999 after taking a ferry from Victoria British Columbia to

Port Angeles, Washington. He was attempting to smuggle several hundred pounds of

explosives and detonators into the U.S.

Northwest culture and tradition also play a role in security issues. Residents in the

region have been traditionally sympathetic to various causes. The large metropolitan areas of

Seattle and Tacoma, with their large, diverse population, provide an environment that enables

illegals to blend.

Finally, the manufacturing, logistics, technology, and financial bases in the region are

representative of many other metro areas in the United States. Targets abound in the

Northwest, ranging from structures offering symbolic targets (the Seattle Space Needle) to

126

facilities and infrastructure critical to the operation of the economy. Kalid Mohamed,

admitted mastermind of 9/11, planned to follow by attacking a “Plaza Bank” in Seattle

(likely the Columbia Tower), along with three other targets in LA, Chicago, and NY (Baldor,

2007).

The ports of Seattle and Tacoma comprise the third largest load center in the U.S.

Approximately 3.5 million TEUs move through these ports annually, as well as oil and other

bulk commodities. In addition to supplying Washington, these ports provide a gateway for

cargo destined for Oregon, Northern California, Idaho, Montana, Nevada, and other western

states. Seattle and Tacoma ports also serve as the primary transshipment points for the

majority of cargo supplying Alaska. Furthermore, 75% of all imports pass via rail to Chicago

and the East Coast. Some have even referred to these Northwest ports as the “Port of

Chicago” because of the volume of shipments destined for Chicago (Seattle DOT, 2005).

The region also hosts major U.S. companies participating in and supporting

worldwide commerce. Boeing’s main base of commercial operations is in the Puget Sound,

with many suppliers nearby. Microsoft, Amazon, and other IT companies are headquartered

in the region. This is also the home of Weyerhaeuser and many of its lumber and

construction companies. Additionally, there are thousands of small and mid-sized

manufacturing firms in the Northwest. There is also a significant military presence with

Army, Air Force, and Navy bases.

The NISAC Agent-Based Laboratory for Economics™ (N-ABLE™) has been

developed by Sandia National Laboratories, through funding from DHS. Drawing from

theory in microeconomic, enterprise, and inventory literature, it has been designed to

simulate the impacts of facility disruptions, transportation disruptions, pandemics, and

127

hurricanes on the U.S. high-tech manufacturing, food manufacturing and distribution, and

chemicals sectors. Analysis of these simulations have brought new insight to DHS regarding

the ability of manufacturing sectors to respond in the short run to national-level disasters, and

the efficacy of private industry and homeland security policies.

Figure A.1 shows an economic agent, the fundamental modeling unit in N-ABLE.

(Swaminathan et al. (1997) describe another agent-based simulation that models a supply

chain.) Each N-ABLE agent is an enterprise of: buyers who input materials, producers of

finished products that use components, labor, capital equipment and infrastructure, on-site

inventories, and sellers of the finished products into segmented markets. Output 1 involves

the assembly of two materials, and Output 2 (e.g., a spare part) follows a serial process. An

agent firm can also depict distribution or transportation hub activity, e.g., where distributor

assembly represents assembling, packing, and shipping an order of SKUs to a customer. The

agent accounting logic captures revenues as well as variable costs of materials, storage, labor,

and overhead. Physical infrastructure includes electric power, telecommunications, and

transportation (Downes, Ehlen, Loose, Scholand, and Belasich, 2006).

N-ABLE’s architecture is agent-based and object-oriented for efficiency, portability,

scalability, and repeatability across types of computers and operating systems (Eidson and

Ehlen, 2005). This has enabled development on individual laptops, and large-scale testing on

Sandia’s Thunderbird supercomputer. Because of this computing power and model

architecture, the system can handle hour-by-hour discrete events for upwards of one million

U.S. manufacturing firms.

128

Figure A.1. The N-ABLE Enterprise Agent.

The first large-scale application of N-ABLE was to model approximately 14,500

small and medium sized firms (with SIC codes) in the Pacific Northwest and their

corresponding supply chains. The network of agent nodes and arcs followed the baselines of

at least four agents per industry, at least one assembly stage per industry, and order-up-to

inventory systems connecting agent inputs and outputs. The network representation was

substantially enhanced through macroeconomic analysis of regional product flows between

SIC codes. Insights from this analysis included:

By observing regionally the shortages that buyers experience in markets, as

supplier production and shipments are disrupted in a region, the initial disturbance

can “flash over” nationally along complex paths and over relatively short time

periods. This phenomenon seems to be driven by supply chain network

characteristics, markets, and transportation modes in the supply chain.

129

By tracing which buyers at which agent firms expedite their orders and for how

long, individual expediting can amplify or dampen bullwhip effects across the

region. This impacts the harmonic frequency (influencing duration) and amplitude

of order behavior, and the rate of recovery.

By tracing in-transit inventory, a disruption in a transportation hub caused firms to

substitute costly transport alternatives, longer transport routes, and amplified

ordering that drained available supply for prolonged periods.

Since undertaking the Pacific Northwest project, Sandia has expanded the

fundamental enterprise structure of the N-ABLE firm to embed more detailed transportation

networks, pricing functions, and cost parameters that capture the economic effects of, among

other metrics, the three performance drivers proposed here (service levels, system expediting,

and system inventory). Currently, Sandia is expanding N-ABLE’s infrastructure to more

realistically characterize behavior and to test policies within and across U.S. manufacturing

sectors. Examples include: integrating Oak Ridge National Laboratory’s inter-modal

transportation system into N-ABLE to observe transportation vulnerabilities in the U.S.

manufactured foods sector (Downes et al., 2005b), and integrating pipeline infrastructure to

uncover vulnerabilities in the U.S. chemicals sector (Downes et al., 2005a).

To illustrate the size, scope, and capability of analysis, Figure A.2 shows output from

one of the N-ABLE simulations currently being applied to the U.S. manufactured-food

supply chain. This application covers 200,000 domestic manufacturing firms, domestic

distributors, domestic retail establishments, foreign suppliers, and consumers of 50 food

commodities. Shipping occurs via highway, rail, and water-based transportation networks.

The figure shows Hurricane Katrina’s regional effects on the U.S. food distribution system.

130

Firms shown as colored dots produce or purchase food commodities over the rail and

highway networks shown in the figure as the thin gray lines. In an in-control state, producers

are shown in gray, and buyers in red. Producers and buyers in green depict a shortage state.

The simulation results verified that the hurricane caused shortages in complex ways. The

firms located directly in the hurricane’s path are green, but with the loss of transportation to

and from the region, there is also a varied array of firms in the East, Northeast, and Midwest

that also experience shortages.

Figure A.2. Example of Regional Shortages (in green) caused by N-ABLE Simulation of Hurricane Katrina.

With such a large number of buyers, sellers, intermediaries, and interconnections,

Sandia modelers believe that N-ABLE will offer an important tool in characterizing how

supply systems can adapt to a myriad of potential man-made and natural disruptions. For N-

131

ABLE to be effective in advocating change in policy and practice in the prevention, response,

and recovery from disruptions, the modelers will need to establish protocols to replicate and

statistically analyze the effects of disruptions and coutermeasures.

132

APPENDIX B

CASE STUDIES OF ELECTRONICS COMPANIES IN THE PACIFIC

NORTHWEST

We conducted personal interviews with personnel from three electronics firms and some of

their suppliers to gain insights into supply chain behavior. Each firm has a different SIC

code. We disguise their identities because their management considers certain information

proprietary and security-sensitive. We label the three as firms INT, ABC, and XYZ. The

supply chains for the three firms are represented in Figure B.1.

INT Corporation

INT designs and makes products that track inventory in-motion and at-rest. They sell to

manufacturers, warehouses, and retail firms. With a well-defined new product development

(NPD) process, INT commits cross-teams within all disciplines to the NPD process. INT also

runs textbook lean manufacturing with visual controls, Andon lights, Kanban system, and

reader boards that clearly state daily goals and hourly status. However, INT incurs relatively

high inventory and overhead costs, as well as delivery problems associated with maintaining

over 2,500 possible configurations within a diverse product line of 60 product models. This

complexity also exposes them to disruptions in power, transportation, and

telecommunications. Depending on the customer and product, INT may incur shortages

either in the form of backorders or stockouts.

133

Figure B.1. Case Study Supply Chain Structures.

134

Echelon 1 – Delivery to INT’s Customers

Three value added resellers comprise 55% of INT’s sales, while hundreds of other customers

represent the remaining share. After sitting in finished goods inventory an average of four

days, INT transports 40% of its products through distribution centers in Canada, Singapore,

Holland, Mexico, and Brazil. INT ships 2,000 orders (7,000 items) per week with five to 30

products per shipment. Domestic outbound transportation from the U.S. manufacturing plant

is via FedEx (75% ground, 25% air). Except for Canada and Mexico, 100% of the

international shipments are FedEx Air.

Echelon 2 – INT’s Assembly of the Electronic Products

INT has a genuine interest in providing high quality products to its customers. They have an

assemble-to-order manufacturing process featuring lean and just-in-time techniques. They

utilize one manufacturing shift per day and offer a seven to 10 day assembly lead time to

customers. Manufacturing lead time averages two days, 60% of which constitutes direct labor

touch time. 200 production orders are launched daily, and 10-300 purchased SKUs are used

for each product. There is additional lead time for component picking and staging. Only 1%

of purchased SKUs are shared in common between saleable products. This non-commonality

presents challenges to keep inventory levels low and service levels high as well as potential

vulnerability to disruptions in the supply chain.

Echelon 3 – INT’s Component Procurement

INT purchases components (5,000 SKUs per year) from 500 suppliers with business totaling

$275 million. Approximately 10% of all SKUs are sole-sourced. 650 to 700 material

135

shipments composed of 1,500 SKUs are received weekly, and the SKUs range from one to

three per shipment. 55% of purchases are direct imports; 90% of which are transported via air

(freight forwarders), and 10% by ship through ports. Inbound transportation comprises less

than 1% of sales. The majority of SKUs purchased by INT are made in Singapore, U.S.,

China, Malaysia, and Mexico (in order of volume).

We focused on two of INT’s major suppliers (Supplier 1 and Supplier 2) and two

others INT considers critical (Supplier 3 and Supplier 4). A common distributor (Supplier 5)

to INT, ABC, and XYZ is discussed later. INT’s order lead time from Supplier 1 and

Supplier 2 ranges from 45-90 days. INT transports the materials through Expediters

International using 15% expedited air (two to three days) and 85% regular air (three to five

days). INT holds 40-45 days of on-hand inventory from Supplier 1, and 70-80 days from

Supplier 2. If INT faced an immediate loss of Suppliers 1 or 2, it has more than ten other

suppliers with which to choose, but comprehensive supply might take two to four months.

The order lead time from Supplier 3 (a critical supplier) ranges from 45-90 days, and

all of its materials are transported via air (three to five days) through freight forwarders, in

which multiple carriers are used. INT holds 30-45 days of on-hand inventory from Supplier

3. INT’s order lead time from Supplier 4 (another critical supplier) ranges from 30-45 days.

They hold 60-70 days of on-hand inventory from Supplier 4. These suppliers transport 90%

of the materials via FedEx air (two days), and the rest, FedEx ground (five days).

Suppliers 3 and 4 are critical to our understanding of supply chain vulnerabilities.

With immediate removal of Supplier 3, one half of INT’s products would cease to be

manufactured for 30-45 days while it switches to an Asian supplier. If INT lost Supplier 4, it

would take an estimated two years for INT’s engineers to redesign certain components

136

common to its product line (patented by Supplier 4), and find a new suppler to build to

blueprint.

Echelon 4 – Delivery to INT’s Suppliers

This section describes the replenishment processes for two of INT’s major suppliers

(Supplier 1 and Supplier 2), and two of INT’s critical suppliers (Supplier 3 and Supplier 4).

Supplier 1 is located in Singapore and accounts for 16% of INT’s annual purchases. Its order

lead time from time of order placement to receipt ranges from 15-120 days and its on-hand

finished goods inventory ranges from seven to 10 days. Inbound transportation is

predominantly ground since Supplier 1 sources 100% of its materials from local Asian

countries.

Supplier 2 is located in Malaysia and accounts for 4% of INT’s annual purchases. Its

order lead time ranges from 1-130 days and on-hand finished goods inventory ranges from

seven to 10 days. The majority of Supplier 2’s materials are transported via ground

transportation from Thailand and China.

Supplier 3 is a sole-sourced French supplier that provides components used in half of

INT’s finished products. Supplier 3 sources brass, copper, and other metals with an average

order lead time of two months.

Supplier 4 is a sole-sourced supplier with sites in Mexico and Asia, and makes

common components used to manufacture numerous INT products. Products made in Asia

are 100% sourced from Asia, while those manufactured in Mexico are 60% sourced in

Mexico and 30% in Asia.

137

INT’s Contingency Planning

Over INT’s supply base, 10% of its components are sole-sourced and proprietary. If INT lost

a supplier for one of these components, it would take six to nine months on average to find

another supplier to design and manufacture the needed part. For a non-proprietary product,

INT could find a new supplier within two months. (According to Berry et al. (1994), U.S.

electronics OEMs sole-source 7% of their components, and single-source 30%. Sole sourcing

often helps to create long-term supplier relationships and streamline information flow that

might reduce demand amplification. However, this efficiency presents associated risks of

disruption.)

INT is highly dependant upon electricity and currently does not own a backup

generator to power its operations or mission critical systems. This vulnerability became

apparent when INT experienced manufacturing downtime of two days after a power box

located off property was destroyed in a car accident. $11 million in products were shipped

late to INT’s customers because they lacked a contingency plan. Additionally, if a mode of

shipment were suddenly removed, INT would expedite outbound products via air, and would

incur seven to ten times the cost over its current ground vessel costs.

ABC CORPORATION

ABC makes high-end instruments for diagnosing and monitoring the performance of

electrical equipment. ABC’s costs are considerably higher than their competitors, as is the

quality of their products. ABC’s shortages usually become lost sales.

ABC Corporation utilizes cutting-edge manufacturing processes modeled after the

Toyota Production System. This includes one-piece production and build-to-order

138

manufacturing. ABC averages 95% on-time order fulfillment with a five-day delivery

schedule (from order to shipment). ABC carries less than two weeks of on-hand inventory.

The majority of its products are used in automotive, utilities, hi-tech, military, and aerospace

industries.

Echelon 1 – Delivery to ABC’s Customers

ABC sells its 200 types of finished products to distributors accounting for 50% of sales, and

the other half to individual customers, including the U.S. Government. ABC offers customers

a one to five week lead time. ABC ships 50 to 150 orders per day with one to 100 items per

shipment. 40% of their sales are exports.

Echelon 2 – ABC’s Assembly of the Electronic Products

ABC manufactures 50% of their products in-house (the Product Group), and outsources the

rest to various turnkey operations in East Asia (the OSP Group). ABC uses 70 to 150 SKUs

to manufacture each product. Should a disruption occur, this relatively small number of

SKUs in a product might enable ABC to provide higher service levels and a correspondingly

lower risk of delay than INT and XYZ. Assembly orders are typically delayed, rather than

split, until sufficient backordered components arrive to complete assembly.

Echelon 3 – Delivery of Components to ABC

The in-house fabrication lead time for ABC’s Product Group is one to 10 days, with the

build-to-order process taking one to seven of those days. For the OSP Group, ABC’s

procurement lead time is one to 12 weeks (80% is two weeks or less). The material and

139

component on-hand inventory level for the Product Group and OSP Group averages one to

two weeks (critical inventory may reach four weeks). Inbound transportation into ABC’s

factory is roughly 80% ground and 20% air (with shippers: 55% UPS, 28% FedEx, and 17%

Roadway). ABC receives 40 to 50 material shipments per day with one to 20 component

SKUs per shipment. ABC buys many of their supplies on consignment, not paying for them

until they enter assembly. An example of this includes materials from Supplier 2 that

provides an on-site warehouse for ABC. Supplier 2 coordinates just-in-time all logistics for

warehousing and transporting components to ABC’s assembly processes. Although ABC

pays a premium for this service, it allows the JIT operation to run seamlessly with minimal

inventory.

Echelon 4 – Delivery to ABC’s Suppliers

Suppliers to the OSP Group include two domestic and 18 international suppliers, totaling

over $60 million per year in purchases. The OSP Group’s three largest suppliers in China and

Japan account for 63% of total supplies.

Unless stated otherwise, subsequent metrics relate to ABC’s in-house Product Group.

The Product Group’s suppliers also represent over $60 million per year in purchases, and are

located in the United States, Canada, Mexico, China, Japan, and Malaysia. The four largest

suppliers account for 43% of the total of this supply, and are located in New York, Arizona,

Canada, and China. Supplies to these four are, in turn, covered by 60% domestic vendors

located in NY and Arizona, 20% in Canada, and 20% in China. Inbound shipments for the

domestic suppliers arrive from U.S. seaports, Mexico, and Canada. Overall, the transport mix

is 30% air, 30% ship, and 40% truck.

140

Twenty percent of INT’s supply base is sole-sourced. An example of a critical sole-

sourced material includes proprietary custom component features assembled in California.

They, in turn, obtain silicon from the Far East and component systems from California. The

sole-source supplier then transports the custom features through distribution partners with a

stocking agreement. The distributors provide forecasts of demand and the sole-source

supplier manufactures to forecast with a buffer. If demand spikes from INT and other

customers, the sole-source supplier expedites the channel to satisfy the critical need and refill

the stock and buffer.

ABC’s Disruptions and Contingency Plans

When asked to discuss a disruption that had occurred in their business, ABC mentioned the

West Coast port lockout. As the lockout approached, ABC prepared with contingency plans

to use alternative sources and routing. They also established larger inventory buffers to cover

demand over longer lead times. Increased inventory, holding, and expediting costs were

incurred during the lockout, but were considered negligible relative to costs of possible

shortages.

ABC is highly dependent on electricity and telecommunications. They have backup

generators to run their mission critical IT system, including ERP, for one week should a

power disruption occur. They do not have telecommunications backup or contingency plans.

They believe these contingency plans must be more fully developed to decrease disruption

risks in critical infrastructure inherent in the post-911 business world.

141

XYZ CORPORATION

XYZ designs, develops, and manufactures systems for use by OEM aerospace

manufacturers. XYZ focuses on three product groups for use on 42 aircraft types, which

serve 450 customers in 49 countries. XYZ is more vertically integrated than most of its

competition. This enables rapid prototyping of new products, but presents challenges to keep

product costs down and maintain service levels.

Echelon 1 – Delivery to XYZ’s Customers

XYZ’s six largest customers comprise 51% of its business, with the largest two representing

one third of total business. XYZ quotes lead times of four to 20 weeks, and takes two to four

weeks to assemble its products. XYZ averages 200-250 orders from customers per week,

which translates into 355 shipments per week. 25% of the orders are exported, all via

airfreight. Domestically, nearly all deliveries are either customer pickup or ground delivery.

Customers determine the delivery service with typical options of UPS surface, BAX, and

FedEx ground/economy.

The price of XYZ’s products is insignificant relative to the value of its customers’

products. XYZ’s customers accept, but do not appreciate, backorders. Repeated poor delivery

performance would result in temporary or permanent loss of a customer in XYZ’s small

customer base.

Echelon 2 – XYZ’s Assembly of the Electronic Products

XYZ’s end products contain between 200 and 700 component SKUs. Of the assembly orders,

65% are made-to-order, and the rest forecasted spares. This mix results in varied finished

142

goods inventory levels: less than one week for make-to-order products, three to four weeks

on expensive spares, and up to three months on less expensive spares. XYZ operates with a

direct labor percentage of 11-13%, with 15-20% touch time of the total lead time. Assembly

orders are typically delayed until sufficient backordered components arrive to fully satisfy

demand requirements, although orders may be split for customer orders deemed critical

Echelon 3 – Delivery of Components to XYZ

XYZ draws from over 300 international and domestic component suppliers and distributors,

receiving approximately 55 material shipments per day with an average of two SKUs per

shipment. 20% of XYZ’s components are sole-sourced, and consequently represent a

significant supply risk. With 90% of their shipments coming via diverse ground delivery and

only 10% expedited by air, XYZ is reasonably buffered against risks in the transportation

system. Only seven to 10% of their component shipments are direct imports, though it is

unclear what percentage of this represents critical inventory. XYZ experiences a purchasing

lead time between six and eight weeks, including one to seven days of transportation. XYZ

averages 10 weeks of inventory to cover the lag between ordering and delivery.

Echelon 4 – Delivery to Suppliers of XYZ

Of the more than 300 international and domestic suppliers of XYZ, we chose three (Supplier

6, Supplier 7, and a common Supplier 5) for Echelon 4. Supplier 6 has a distribution hub in

Boston and Supplier 7 in Fort Worth. These distributors employ diversification strategies,

multiple transportation modes, and inventory holding policies to help ensure minimal risk to

their customers.

143

Supplier 6 procures materials from source companies in Israel, Japan, Europe, and

U.S., and makes its components in the Philippines, Malaysia, Mexico, and various other

foreign countries. Multiple source companies ensure redundancy and limit supplier risk at

this echelon. Once fabricated, the electronic components are shipped to Supplier 6’s central

U.S. distribution hub in Boston. The mode of shipment depends on the origin of the foreign

country and the nature shipments, but 85-90% of Supplier 6’s components are shipped to

Boston via air/truck combination, and 10-15% via seaport/truck combination. All domestic

shipments are transported via truck. Supplier 6 incurs a 20-50% premium for rush orders

Supplier 7 operates similarly to 6. A distribution hub in Fort Worth, Texas receives

components from abroad and maintains inventory that turns an average of six times per year.

Representatives at Supplier 7 estimated that in the event of a complete halt of deliveries, they

would run out of components for key customers within six weeks. To keep the likelihood of a

complete shutdown at a minimum, Supplier 7 geographically dispersed its own supply

companies by operating in Europe, Asia, the Middle East, South America, and North

America. Supplier 7 also diversifies its modes of shipment receipt; 25% of all foreign

shipments arrive via air (Dallas/Ft. Worth airport), 50% via ports in Texas and New Orleans,

Louisiana, and 25% via ground transport (usually shipments from Central America and

Mexico). All domestic shipments are handled by trucking shippers.

Despite this diversification strategy, Supplier 7 encountered problems with a Texas

port closure. This left representatives with a feeling that despite an ability to reroute, an

unannounced port closure of greater than one hundred days would be problematic

economically. Additionally, Supplier 7 represents a potential source of vulnerability for

XYZ’s critical inventory. A crisis would result in Supplier 7 emptying pipeline inventory of

144

critical parts within six weeks. With only two additional weeks of inventory on XYZ’s

shelves, XYZ would find itself in a materials shortage within two months.

XYZ’s Disruptions and Contingency Plans

Besides the economic risks born by the nature of the business, supply chain structure and

transportation dependencies, XYZ representatives also sense exposure in manufacturing and

testing to electrical and telecommunications disruptions. The company does not operate with

a backup generator, and has no contingency plans in the event of a prolonged blackout.

Additionally, nearly all customer and order records are stored electronically. Although a

blackout poses no real risk of data loss, the company admitted that data access could hamper

significantly the firm’s operations were a blackout to last longer than a few days.

Regarding telecommunications risk, the company is dependent upon phone, email, and

internet connections for order receipt and fulfillment, customer service, intra-organizational

communications, business research, and other important functions. Recently, XYZ

experienced loss of land-phone service for just under a week. During this time,

representatives were forced to use cellular phones, an issue that caused significant customer

service friction. Quoting company representatives “Were both electric power and

telecommunications to go down simultaneously, this company would stop dead in its tracks”.

COMMON SUPPLIER 5

Supplier 5 is a large U.S.-based components distributor that supplies electronics firms

worldwide. The following is from the perspective of Supplier 5, versus that of its customers

INT, ABC, and XYZ. Supplier 5 services 40,000 North American customers, 30,000 of

145

which are found in their Components Division. Supplier 5’s customer sales and outbound

transportation activities are voluminous and diverse, maintaining in the aggregate a relatively

stable delivery schedule of 100,000 shipments per week. 65% of shipments to major

customers are made via UPS, FedEx, and LTL ground, and 23% are shipped via two to three

day air (UPS, FedEx, and DHL). EMERY two to three day air is used to deliver shipments to

Canada. 12% of deliveries are made using FedEx overnight service. Currently, 25-30% of

shipments are expedited over their normal delivery time, which is similar to their pre-9/11

levels in the peak of 2000.

Supplier 5 has roughly 280 suppliers and receives 3500 shipments per week (2500

shipments/week for the Components Division – 71% of total in North America alone). 25%

of Supplier 5’s materials are imports, purchased from American multinational companies

with overseas manufacturing plants, rather than from foreign companies. The majority of

Supplier 5’s material shipments arrive via two to three day air, with the remainder arriving

via a geographically favorable California seaport. Supplier 5 does not choose its mode of

shipping. In its terms of procurement, ownership of goods is not taken until they reach the

U.S. port. Part delivery to Supplier 5’s manufacturing headquarters is usually one to two days

via ground transportation. Overall, Supplier 5 operates with a total order/transportation lead

time of two to three weeks for semiconductors, three to 12 weeks for non-commodity items,

and seven-week average for all others.

The characteristics of Supplier 5’s supply chain render it reasonably well protected

against transportation, supplier, and infrastructure risk. In the event that a mode of domestic

or international outbound shipment was removed, Supplier 5 would simply switch to another

ground carrier. Outbound transportation could be shifted to other carriers, while realizing

146

only a two to three day delay in lead time and a 10% increase in shipping costs. Supplier 5

maintains a system for managing significant disasters, where major customers would receive

first priority, other customers who provide stable product forecasts second priority, and so

forth.

With respect to its supply chain, less than 20% of the purchased SKUs are sole

sourced, and contingency plans are in place to handle disruptions for all but 8% of their

supplier’s goods. Since Supplier 5 employs engineers on staff, a planned contingency for a

loss of a sole-source supplier would be to redesign the part and outsource it. The current

expectation is that a disruption to Supplier 5’s supply chain would extend delivery lead times

by only two to three days, but result in a 30% increase in freight costs (non-contract

shipments would need to be auctioned). Supplier 5 maintains a disaster recovery team to

minimize the effects of a major disruption in their supply chain. In the event that a major

mode of shipment was removed, Supplier 5 is confident that it would be able to switch

modes quickly (for example, seaport to airport), relying on relationships with multiple

carriers (UPS, EMERY, DHL, FedEx). These carriers are anxious to service essential

customers such as Supplier 5 and would be able to absorb extra demand. For example, when

UPS workers threatened to strike, Supplier 5 began the process of coordinating alternative

carriers. Supplier 5’s disaster planning team has also developed contingency plans to deal

with commercial disruptions caused by the SARS virus.

Supplier 5 is reasonably well insulated against disruptions to critical civic

infrastructure such as electric power or telecommunications networks. Supplier 5 asserts that

their status as a global company protects them against major disruptions since their many

plants around the world can accommodate the loss of one plant. That being the case, Supplier

147

5 acknowledges their dependence on electrical power, and maintains backup capabilities to

counter a grid disruption. Its power company supplies electricity through two separate

electrical grids. Thus, a backup grid is in place to sustain operations.

Supplier 5 has a central IT site that handles all networking for each location

worldwide. They have an on-site diesel generator that can feed their IT systems for up to four

days. These systems are tested weekly to ensure properly working condition. Additionally,

they have a one-hour battery backup, and maintain backup data facilities off-site.

Similarly, all telecommunications lines have dual backup to avoid disruptions to their

data communications systems. If the phone lines are down, Supplier 5’s email system will

still be operational. Additionally, managers are instructed to use cell phones 100% of the

time internally, which is effective given the mobile nature of managers’ work. Otherwise,

Supplier 5 does not maintain a telecommunications contingency plan.

148

APPENDIX C

CASE INTERVIEW PROTOCOL

Electronics Firm

Customer Base/Distribution

1. Who are your major customers?

2. Approximately what proportion of your company’s revenue does each major customer

provide?

3. What are the major products your company makes for these customers?

If you know, what are the SIC/NAIC codes associated with these products?

4. How are products transported to each of these major customers?

What are the modes, format, and proportion of shipments (e.g., Express mail,

DHL, two day air, overnight delivery; 80% of shipments to major customer

x)?

If delivery time is not guaranteed, what is the average and range of delivery

times for each mode?

Approximately how many shipments are made per day (week or month)?

Approximately how many product types are in a shipment?

What percentage of your shipments are exports? What modes of transportation

do you use for these exports?

149

If a mode of domestic or international shipment mode were suddenly

removed:

- How would you then ship goods to each of these customers?

- What additional delay would be incurred?

- What additional shipment cost (as a percentage over preferred method)

would be incurred?

- If faced with the problem of rationing a shortage of components, what

factors would you consider in deciding which customers would receive

priority?

Production

1. Would you classify your business as make to stock, make to order, assemble to order,

other?

2. For the products sold to the major customers identified earlier, what is the approximate

average and range of:

Quoted lead times for these products?

Manufacturing lead times for these products?

- What percentage of this lead time would you guess is actual direct labor

touch time?

- Purchased lead times for materials used to make these products?

3. Approximately how much finished goods inventory (in days – average and range) do you

maintain for these products?

150

4. Approximately how many production orders are launched per day (week or month) into

your factory?

5. What is the approximate average and range of direct labor percentage in a product's

selling price; raw material percentage; profit before taxes?

6. What is the approximate average and range of number of SKU’s for purchased material

for any given product?

Approximately what is the percentage of purchased SKUs that are shared in

common between saleable products?

Supply Base

1. Approximately what percentage of the purchased SKUs is sole sourced?

2. Approximately how many suppliers of materials do you have?

3. Who are your major suppliers?

Approximately what is the proportion of business for this supplier to total

supplier business?

4. Choose two critical purchased materials (e.g., ones that tend to be in short supply, are

indispensable, are sole sourced, have long lead times, etc.).

What are they?

Who supplies them?

How are they transported?

From the time you place a purchase order, what is the approximate average

and range of lead times for materials delivered by each of these major and

critical suppliers?

151

Approximately how much material inventory (in days – average and range) do

you keep on hand for materials provided by each of these major and critical

suppliers?

If you know, what is the SIC or NAIC code of each supplier or commodity

they supply?

How do these suppliers transport materials to you?

What are the modes, format, and proportion of shipments?

If delivery time is not guaranteed, what is the average and range of delivery

times for each mode?

5. Approximately how many material shipments do you receive per day (week or month)?

Approximately how many material SKUs are in a typical shipment?

What percentage of your shipments are direct imports?

What modes of transportation do you use for these imports?

If a mode of shipment (domestic or international) were suddenly removed,

how would each supplier then ship goods to your site?

If faced with the problem of losing one of these suppliers, what alternatives

would you have, which alternative would you choose, and why?

Who would you guess are major suppliers of materials that your supplier uses

to make or store the materials shipped to you?

How would you guess these materials are delivered to your supplier’s

supplier?

152

More on Possible Disruptions

1. Do you recall some disruption (even a small one) in your business?

What happened?

What were the short-term consequences, e.g., delays, additional costs?

How was the problem resolved?

2. Electric Power: In what general ways is your firm dependent on electric power?

What if there is a disruption in power; how would you respond?

Do you have emergency back up?

- Over what duration?

- Which critical items would be supplied with emergency power and which

would not?

3. Telecommunications: In what general ways is your firm dependent on

telecommunications?

What if there is a disruption in production due to telecommunications; how

would you respond?

Do you have emergency back up?

- Over what duration?

- Which critical items would be supplied with emergency power and which

would not?

4. Have we missed any important vulnerabilities with respect to critical infrastructure?

153

Component Supplier

Customer Base/Distribution

1. Approximately how many customers do you have?

Who are your major customers?

Approximately what proportion of your company’s revenue does each major

customer provide?

2. Approximately how much material inventory (in days – average and range) do you keep

on hand for a typical SKU?

3. Shipments

Approximately how many shipments do you make per day (week or month)?

- Approximately how many SKU’s are in a shipment?

In general, how are shipments made to these major customers?

- What are the modes, format, and proportion of shipments?

- What percentage of the shipments to customers is expedited over normal

delivery at this time?

- What was the percentage when business was at its peak; when was the

peak?

4. What percentage of your shipments is exported?

What modes of transportation do you use for these exports?

5. If a mode of domestic or international shipment mode were suddenly removed,

How would you then ship goods to each of these customers?

- What additional delay would be incurred?

154

- How would lead times be affected?

- What additional shipment cost (as a percentage over preferred method)

would be incurred?

6. If faced with the problem of rationing a shortage of components, what factors would you

consider in deciding which customers would receive priority?

Supply Base

7. Approximately how many suppliers of materials do you have?

Who are your major suppliers?

Approximately what is the proportion of business for this supplier to total

supplier business?

Choose two critical purchased materials (e.g., ones that tend to be in short

supply, are indispensable, are sole sourced, have long lead times, etc.).

- What are they?

- Who supplies them?

- How are they transported?

8. Approximately how many material shipments do you receive per day (week or month)?

Approximately how many material SKUs are in a typical shipment?

What percentage of the material shipments is expedited over normal delivery

at this time?

- What was the percentage when business was at its peak; when was the

peak?

What percentage of your shipments is directly imported?

155

- What modes of transportation do you use for these imports?

Approximately what percentage of the purchased SKUs is sole sourced?

If faced with the problem of losing one of these suppliers, what alternatives

would you have, which alternative would you choose, and why?

Choose one major overseas supplier that delivers components to you through

a seaport.

- How does this supplier transport materials to you? What are the modes,

format, and proportion of shipments (e.g., truck from China manufacturer

to Chinese port, carrier to the Port of Long Beach, from the Port of Long

Beach -- 80% of shipments via Express mail, and 20% of shipments

overnight delivery)?

- From the time you place a purchase order, what is the approximate

average and range of lead times for components delivered by this

supplier?

- If a mode of shipment (domestic or international) were suddenly

removed, how would the supplier then ship goods to your site?

How would lead times be affected?

What additional shipment cost (as a percentage over preferred method) would be incurred?

More on Possible Disruptions

1. Do you recall some disruption (even a small one) in your business?

What happened?

What were the short-term consequences, e.g., delays, additional costs?

156

How was the problem resolved?

2. Electric Power

In what general ways is your firm dependent on electric power?

What if there is a disruption in power; how would you respond?

Do you have emergency back up?

- Over what duration?

- Which critical items would be supplied with emergency power

and which would not?

3. Telecommunications

In what general ways is your firm dependent on telecommunications?

What if there is a disruption in production due to telecommunications; how

would you respond?

Do you have emergency back up?

- Over what duration?

- Which critical items would be supplied with emergency power and which

would not?

Have we missed any important vulnerabilities with respect to critical infrastructure?

157

BIBLIOGRAPHY

Abcede A., “Trust Remains a Barrier in Spotting Employee Theft,” National Petroleum News, Vol. 86, No. 8, pp. 30-33 (1994).

Albeniz de A. V. and Simchi-Levi, D., “A Portfolio Approach for Procurement Contracts,” Production and Operations Management, Vol. 14, No. 1, pp. 90-114 (2005).

Alexander, D., Natural Disasters, Taylor and Francis Books, Ltd., London, UK (1993).

Allen, B. S., “Planning for Disaster: M.I.T. Simulates a Major Supply Chain Disruption,” MIT Center for Transportation Logistics, Press Release (February 17, 2006).

American Institute of Certified Accountants. 2002. Statement on Auditing Standards No. 99, Consideration of Fraud in a Financial Statement Audit. AICPA: New York, NY (2002).

Amodeco, L., Prins, C., and Sanchez, D. R., “Comparison of Metaheuristic Approaches for Multi-Objective Simulation-Based Optimization,” In Supply Chain Inventory Management Source, Lecture Notes in Computer Science, No. 5484, pp. 798 – 807 (2009).

Annan, K., “An Increasing Vulnerability to Natural Disasters,” International Herald Tribune(September 10, 1999).

Applequist, G. E., Pekny, J. F., and Reklaitis, G. V., “Risk and Uncertainty in Managing Chemical Manufacturing Supply Chains,” Computers and Chemical Engineering, Vol. 24, No. 9, pp. 2211–2222 (2000).

Arbuckle, J. L., Amos 6.0 User's Guide, SPSS Inc, Chicago, IL (2005).

Arreola-Risa, A. and DeCroix, G. A., “Inventory Management under Random Supply Disruptions and Partial Backorders,” Naval Research Logistics, Vol. 45, No. 7, pp. 687-703 (1998).

Arslan, H., Ayhan, H., and Olsen, T. L., “Analytic Models for When and How to Expedite in Make-to-order Systems,” IIE Transactions, Vol. 33, No. 11, pp. 1019-1030 (2001).

Axsater, S., “Continuous Review Policies for Multi-Level Inventory Systems With Stochastic Demand,” In: Graves, S., Rinnooy Kan, A., Zipkin, P. (Eds.), Handbook in Operations Research and Management Science, Vol. 4, North-Holland, Amsterdam(1993).

Axsater, S., Inventory Control. Norwell, MA: Kluwer Academic Publishers (2000).

Baganha, M.P. and Cohen, M.A., “The Stabilizing Effect of Inventory in Supply Chains,” Operations Research, Vol. 46, No. 3, pp. 572–583 (1998).

Bakshi, N. and Gans, N., “Securing the Containerized Supply Chain: An Economic Analysis of C-TPAT,” Working Paper No. 2007-11-19, the Wharton School, University of Pennsylvania (2007).

158

Balakrishnan, K., Iyer, A., Seshadri, S., and Sheopuri, A., “Indian Auto Supply Chain at the Cross-Roads,” Working paper, Krannert School of Management, Purdue University, West Lafayette, IN (2007).

Baldor, L.C., “State Bank on 9/11 Terrorists’ hit List,” The Associated Press, Seattle Post-Intelligencer (March 15, 2007).

Ballou, R. H., Business Logistics Management. Prentice-Hall, Englewood Cliffs, NJ (1992).

Beamon, B.M., and Chen, V. C. P., “Performance Analysis of Conjoined Supply Chains,”International Journal of Production Research, Vol. 39, No. 14, pp. 3195–3218 (2001).

Beccerra, M. and Gupta, A., “Trust within Organization: Integrating the Trust Literature with Agency Theory and Transaction Cost Economics,” Public Administration Quarterly, Vol. 23, pp. 177-203 (1999).

Berk, E., and Arreola-Risa, A., “Note on Future Supply Uncertainty in EOQ Models,” Naval Research Logistics, Vol. 41, No. 1, pp. 129-132 (1994).

Berry, D., Towill, D. R., and Wadsley, N., “Supply Chain Management in the Electronics Products Industry,” International Journal of Physical Distribution and Logistics Management, Vol. 24, No. 10), pp. 20-32 (1994).

Beverly, R. and Rodysill, J., “Risky Business,” Outlook Journal, accessed August 14, 2007, available at http://www.accenture.com/Global/Research_and_Insights/Outlook/By_Issue/Y2007/RiskyBusiness.htm (2007).

Bews, N.and Rossouw, G., “A Role for Business Ethics in Facilitating Trustworthiness,” Journal of Business Ethics, Vol. 39, No. 4, pp. 377-390 (2002).

Beyer, D. and Ward, J., “Network Server Supply Chain at HP: A Case Study,” In J. Song, and and D. Yao (Eds.), Supply chain structures: Coordination, information and optimization. Norwell, MA: Kluwer Academic Publishers, Chapter 8 (2002).

Boer, G. and Jeter, D., “What’s New about Modern Manufacturing? Empirical Evidence on Manufacturing Cost Changes,” Journal of Management Accounting Research, Vol. 5, pp. 61-83 (1993).

Bollapragada, S. and Morton, T. E., “A Simple Heuristic for Computing Vonstationary (s, S) Policies,” Operations Research, Vol. 47, No. 4, pp. 576-584 (1999).

Bowersox, D. J., and Closs, D. J., Logistics Management: The Integrated Supply Chain Process, McGraw-Hill, New York (1996).

Bradley, J. R., “Managing Assets and Subcontracting Policies,” PhD. Dissertation, Stanford University, Stanford, CA (1997).

Brown, K. A., Schmitt, T. G., Schonberger, R. J., and Dennis, S., “QUADRANT Corporation Applies Lean/JIT Concepts in a Project Environment,” Interfaces, Vol. 34, No. 6 (2004).

Brown, R. B. Smoothing, Forecasting and Prediction, Englewood Cliffs, NJ: Prentice-Hall (1963).

159

Cachon, G., and Fisher, M., “Supply Chain Inventory Management and the Value of Shared Information,” Management Science, Vol. 46, No. 8, pp. 1032-1048 (2000).

Callioni, G., Montgros, X., Slagmulder, R., Wassenhove, L. V., and Wright, L., “Inventory Driven Costs,” Harvard Business Review, Vol. 83, No. 3, pp.135-141 (2005).

Caplan, D., “Internal Controls and Detection of Management Fraud,” Journal of Accounting Research, Vol. 37, No. 1, pp. 101-17 (1999).

Carlsson, D. and Rönnqvist, M., “Supply Chain Management in Forestry––Case Studies at Södra Cell AB,” European Journal of Operational Research, Vol. 163, No. 3, pp. 589-616 (2005).

Carr, S. M. and Karmarkar, “U. S. Competition in Multiechelon Assembly Supply Chains,” Management Science, Vol. 51, No. 1, pp. 45–59 (2005).

Cassola G., “The Trust Factor,” Managerial Auditing Journal, Vol. 8, pp. 25-28 (1993).

Caster, P., Massy, D., and Wright, A., “Research on the Nature, Characteristics, and Causes of Accounting Errors: The Need for a Multi-method Approach,” Journal of Accounting Literature, Vol. 19, pp. 60-92 (2000).

Cassola, G., “The Trust Factor,” Managerial Auditing Journal, Vol. 8, pp. 25-28 (1993).

Caster, P., Massy, D., and Wright, A., “Research on the Nature, Characteristics, and Causes of Accounting Errors: The Need for a Multi-Method Approach,” Journal of Accounting Literature, Vol. 19, pp. 60-92 (2000).

Cavinato, J. L., “Supply Chain Logistics Risks: From the Back Room To the Board Room. International Journal of Physical Distribution and Logistics Management, Vol. 34, No. 5, pp. 383-387 (2004).

Center for Research on the Epidemiology of Disasters, Université Catholique de Louvain, Ecole de Santé Publique, Brussels, Belgium, < http://www.cred.be/> (2007).

Chao, H. P., “Inventory Policy in the Presence of Market Disruptions,” Operations Research, Vol. 35, No. 2, pp. 274-281 (1987).

Chapman, P., Christopher, M., Juttner, U., and Peck, H., “Identifying and Managing Supply Chain Vulnerability,” Logistics and Transport Focus, Vol. 4, No. 4, pp. 1-5 (2002).

Chatfield, D., Kim, J., Harrison, T., and Hayya, J. “The Bullwhip Effect – Impact of Stochastic Lead Times, Information Quality, and Information Sharing: A Simulation Study,” Production and Operations Management, Vol. 13, No. 4, pp. 340-353 (2004).

Chaudhuri, A., Khan, S., Lakshmiratan, A., Py, A., and Shah, L., “Trust and Trustworthiness in a Sequential Bargaining Game,” Journal of Behavioral Decision Making, Vol. 16, pp. 311-340 (2003).

Chen, Y. F., Ryan, J. K., and Simchi-Levi, D., “The Impact of Exponential Smoothing Forecasts on the Bullwhip Effect,” Naval Research Logistics, Vol. 47, No. 4, pp. 271-286 (2000a).

Chen, F., Drezner, Z., Ryan, J. K., and Simchi-Levi, D., “Quantifying the Bullwhip Effect in a Simple Supply Chain: The Impact of Forecasting, Lead Times, and Information,” Management Science, Vol. 46, No. 3, pp. 436-443 (2000).

160

Chen, X., Sim, M., Simchi-Levi, D., and Sun, P., “Risk Aversion in Inventory Management,” Operations Research, forthcoming (2008).

Chen, Li., and Lee, H. L., “Information Sharing and Order Variability Control Under a Generalized Demand Model,” Management Science, Vol. 55, No. 5, pp. 781–797 (2009).

Chen, X., Sim, M., Simchi-Levi, D., and Sun, P., “Risk Aversion in Inventory Management,”Operations Research, Vol. 55, No. 5, pp. 828-842 (2007).

Chen, F., and Song, J. S., “Optimal Policies for Multi-Echelon Inventory Problems With Markov-Modulated Demand,” Operations Research, Vol. 49, No. 2, pp. 226-234(2001).

Chen, F., and Zheng, Y. S., “Lower Bounds for Multi-Echelon Stochastic Inventory Systems,” Management Science, Vol. 40, No. 11, pp. 1426–1443 (1994).

Chen, P., Yunsong, G., Lim, A., and Rodrigues, B., “Multiple Crossdocks with Inventory and Time Windows,” Computers and Operations Research, Vol. 33, No. 1, pp. 43-63 (2006).

Chen, F., “On (R, NQ) Policies in Serial Inventory Systems.” In Tayur, S., R. Ganeshan and M. J. Magazine (eds.). Quantitative Model for Supply Chain Management. Kluwer Academic Publishers, Boston, MA (1998).

Chopra, S., Mohan, U., and Reinhardt, U., "The Effect of Supply Chain Disruption on Sourcing Strategy," Working Paper, Kellogg School of Business, Northwestern University (2005).

Chopra, S. and Sodhi, M. S., “Managing Risk to Avoid Supply Chain Breakdown,” Sloan Management Review, Vol. 46, Issue 1, pp. 63-61 (2004).

Clark, A., and Scarf, H., “Optimal Policies for A Multi-Echelon Inventory Problem,”Management Science, Vol. 6, No. 4, pp. 475–490 (1960).

Clark, A., and Scarf, H., “Approximate Solutions to a Simple Multi-Echelon Inventory Problem,” In Arrow, K. J., and Scarf, H. (eds.). Studies in Applied Probability and Management Science, Stanford University Press, Stanford, CA (1962).

Closs, D. J., Roath, A. S., Goldsby, T. J., Eckert, J. A., and Swartz, S. M., “An Empirical Comparision and Response-Based Supply Chain Strategies,” International Journal ofLogistics Management, Vol. 9, No. 2, pp. 21-34 (2008).

Church, R. L., and Scaparra, M. P., “Protecting Critical Assets: The R-Interdiction Median Problem With Fortification,” Geographical Analysis, Vol. 39, No. 2, pp. 129-146 (2007).

Cohen, M., Ho, T., Ren, J., and Terwiesch, C., “Measuring Imputed Cost in the Semiconductor Equipment Supply Chain,” Management Science, Vol. 49, No. 12, pp. 1653-1670 (2003).

Colbert, J. and Bowen, P., “A Comparison of Internal Controls: COBIT, SAC, COSO, and SAS 55/78,” Available at

161

http://www.isaca.org/Content/ContentGroups/Bookstore6/Book_Reviews/A_Comparison_of_Internal_Controls_COBIT,_SAC,_COSO_and_SAS_55_78.htm (2004).

Coletti, A., Sedatole, K., Towry, K., “The Effect of Control Systems on Trust and Cooperation in Collaborative Environments,” The Accounting Review, Vol. 80, pp. 477-501 (2005).

Colquitt, J., Scott B., LePine J., “Trust, Trustworthiness, and Trust Propensity: A Meta-Analytic Test of Their Unique Relationships with Risk Taking and Job Performance,” Journal of Applied Psychology, Vol. 92, No. 4, pp. 909–927 (2007).

Corne, D., Dorigo, M., and Glover, F., New Ideas in Optimization. McGraw-Hill, New York, NY (1999).

Connell, J., Ferress, N., and Travaglione, T., “Engendering Trust in Manager-Subordinate Relationships: Predictors and Outcomes,” Personnel Review, Vol. 32, pp. 569-590 (2004).

Coutu, D. L., “How Resilience Works,” Harvard Business Review, Vol. 80, No. 5, pp. 46-55 (2002).

Craighead, C. W., Blackhurst, J., Rungtusanatham, M. J., and Handfield, R. B., “The Severity of Supply Chain Disruptions: Design Characteristics and Mitigation Capabilities,” Decision Sciences, Vol. 38, No. 1, pp. 131-156 (2007).

Croson, R., and Donohue, K., “Behavioral Causes of the Bullwhip Effect and the Observed Value of Inventory Information,” Management Science, Vol. 52, No. 3, pp. 323-336(2006).

Cunningham, L., “The Appeal and Limits of Internal Controls to Fight Fraud, Terrorism, Other Ills,” Journal of Corporation Law, Vol. 29, pp. 267-336 (2004).

Daniel, J. S. R., and Rajendran, C., “Heuristic Approaches To Determine Base-Stock Levels in a Serial Supply Chain With a Single Objective and With Multiple Objectives,”European Journal of Operational Research, No. 175, pp. 566–592 (2006).

Daniel, J. S. R., and Rajendran, C., “A Simulation-Based Genetic Algorithm for Inventory Optimization in a Serial Supply Chain,” International Transactions in Operational Research, Vol. 12, pp. 101–127 (2005).

Das, T. K. and Teng, B., “The Risk-Based View of Trust: A Conceptual Framework,” Journal of Business and Psychology, Vol. 19, pp. 85-116 (2004).

Dejonckheere, J., Disney, S. M., Lambrecht, M. R., and Towill, D. R., “The Impact of Information Enrichment on the Bullwhip Effect in Supply Chains: A Control Engineering Perspective,” European Journal of Operational Research, Vol. 153, No. 3, pp. 727-750 (2004).

Donselaar, K., Kok, T., and Rutten, W., “Two Replenishment Strategies for the Lost Sales Inventory Model: A Comparison,” International Journal of Production Economics, Vol. 46–47, No. 1, pp. 285–295 (1996).

Downes, P. S., Ehlen, M. A., Loose, V. W., Scholand, A. J., and Belasich, D. K., “Estimating the Economic Impacts of Infrastructure Disruptions to the U.S. Chlorine Supply

162

Chain: Simulations Using the NISAC Agent-Based Laboratory for Economics (N-ABLE™),” SAND2005-2031, Sandia National Laboratories (2005a).

Downes, P. S., Ehlen, M. A., Loose, V. W., Scholand, A. J., and Belasich, D. K., “An Agent Model of Agricultural Commodity Trade: Developing Financial Market Capability within the NISAC Agent-Based Laboratory for Economics (N-ABLE ),” Sandia National Laboratories (2005b).

Downes, P. S., Ehlen, M. A., Loose, V. W., Scholand, A. J., and Belasich, D. K., “Modeling in N-ABLE™, National Infrastructure Simulation and Analysis Center, SAND2006-6586, Sandia National Laboratories (2006).

de Kok T., “A Moment-Iteration Method for Approximating the Waiting Time Characteristics of the GI/G/1 queue,” Probability of Engineering and Information Sciences, Vol. 3, pp. 273-287 (1989).

Eidson, E. D. and Ehlen, M. A., “NISAC Agent-Based Laboratory for Economics (N-ABLE): Overview of Agent and Simulation Architectures,” SAND2005-0263, Sandia National Laboratories (2005).

Elangovan, A. and Shapiro, D., “Betrayal of Trust in Organizations,” The Academy of Management Review, Vol. 23, pp. 547-566 (1998).

Ettl, M., Feign, G.E., and Lin, G. Y., “A Supply Network Model With Base-Stock Control and Service Requirements,” Operations Research, Vol. 48, No. 2, pp. 216–232(2000).

Eynan A., and Kropp, D., “Periodic Review and Joint Replenishment in Stochastic Demand Environments,” IIE Transactions, Vol. 30, No. 11, pp. 1025-1033 (1998).

Faaland, B. H., Schmitt, T. G., and Arreola-Risa, A., “Economic lot scheduling with lost sales and setup times,” IIE Transactions, Vol.36, No. 7, pp. 629-640 (2004).

Faaland, B. H. and Schmitt, T. G., "Cost-Based Scheduling of Workers and Equipment in a Fabrication and Assembly Shop," Operations Research, Vol. 41, No. 2 (1993).

Federgruen, A., “Centralized Planning Models for Multiechelon Inventory Systems Under Uncertainty,” In: Graves, S., Rinnooy Kan, A., Zipkin, P. (Eds.). Handbook in Operations Research and Management Science, Vol. 4, Logistics of Production and Inventory. North-Holland, Amsterdam (1993).

Federgruen, A., and Zipkin, P., “Computational Issues in an Infinite Horizon Multi-Echelon Inventory Model,” Operations Research, Vol. 32, No. 4, pp. 818–836 (1984a).

Ferdows, K., Lewis, M. A., and Machuca, A. D., “Rapid-fire Fulfillment,” Harvard Business Review, Reprint R0411G (2004).

Fellingham, J. and Newman, P., “Strategic Considerations in Auditing,” The Accounting Review, Vol. 55, pp. 639-50 (1985).

Fellingham, J., Newman, P., and Patterson, E., “Sampling Information in Strategic Audit Settings,” Auditing: A Journal of Practice and Theory, Vol. 8, pp. 1-21 (1989).

Finkle, J., “Economic Development after Katrina: Lessons from National Post-Disaster Response and Recovery,” available at

163

<http://www.iedconline.org/Downloads/GulfCoast/Katrina_ Lessons.pdf> (October 25, 2005).

Fischbacher, U. and Stefani, U., “Strategic Errors and Audit Quality: An Experimental Investigation,” The Accounting Review, Vol. 82, No.3, pp. 679-705 (2007).

Floch, J. and Olson, C., “Trust is not an Internal Control,” The CPA Journal, Vol. 73, pp. 10-11 (2003).

Flynn, S., “Why America is Still an Easy Target,” Time, pp. 38-41, (July 26, 2004).

Fukuda, Y., “Optimal Policies for the Inventory Problem with Negotiable Lead Time,” Management Science, Vol. 10, No. 4, pp. 690-708 (1960).

Ganeshan, R., Jack, E., Magazine, M. J., and Stephens, P., “A Taxonomic Review of Supply Chain Management Research,” In Tayur, S., Ganeshan, R., Magazine, M.J. (eds.) Quantitative Models for Supply Chain Management. Kluwer Academic Publishers, Massachusetts, pp. 841–879 (1999).

Gardner, E. S., “X Exponential Smoothing: The State of the Art,” Journal of Forecasting,Vol. 4, pp. 1-28 (1960).

Gen, M., and Cheng, R., Genetic Algorithms and Engineering Optimization, John Wiley andSons, New York (2000).

Glasserman, P., and Tayur, S., “The Stability of a Capacitated, Multiechelon Production-Inventory System under a Base-Stock Policy,” Operations Research, Vol. 42, No 5, pp. 913–925 (1994).

Glasserman, P., and Tayur, S., “A Sensitivity Analysis for Base-Stock Levels in Multiechelon Production–Inventory Systems,” Management Science, Vol. 41, No. 2, pp. 263–281 (1995).

Glover, F., “Tabu Search- Part I,” ORSA Journal on Computing, Vol. 1, No. 3, pp. 190-206 (1989).

Glover, F., “Tabu Search- Part II,” ORSA Journal on Computing, Vol. 2, No. 1, pp. 4-32 (1990).

Glover, F., “Parametric Tabu Search for Mixed Integer Programs,” Computers and Operations Research, Vol. 33, No. 9, pp. 2449-2494 (2006).

Glover, F., “Future Paths for Integer Programming and Links to Artificial Intelligence,” Computers and Operations Research, Vol.13, No. 5, pp. 533-549, (1986).

Glover, F., A Template for Scatter Search and Path Relinking. In J.K. Hao, E. Lutton, E. Ronald, M. Schoenauer and D. Snyers (Eds.). Artificial Evolution, Lecture Notes in Computer Science, 1363, 3-54, Springer-Verlag (1998).

Glover, F. and M. Laguna, Tabu Search, Kluwer Academic Publishers, Boston (1997).

Goldberg, D. E., “Genetic Algorithms in Search, Optimization, and Machine Learning,” Addison-Wesley Pub. Co. (1989).

Gosavi, A., Simulation-Based Optimization: Parametric Optimization Techniques and Reinforcement Learning. Kluwer Academic Publishers, Boston, MA (2003).

164

Groenevelt, H. and Rudi, N., “A Base Stock Inventory Model with Possibility of Rushing Part of Order,” Working paper, Simon School of Business, University of Rochester, Rochester, NY (2003).

Gupta, D., “The (Q, R) Inventory System with an Unreliable Supplier,” INFOR, Vol. 34, No. 2, pp. 59-76 (1996).

Hadley, G. and Whitin, T. M., Analysis of inventory systems, Engelwood Cliffs, NJ: Prentice Hall (1963).

Hallikas, J., Karvonen, I., Pulkkinen, U., Virolainen, V. M., and Tuominen, M., “Risk Management Processes in Supplier Networks,” International Journal of Production Economics, Vol. 90, No. 1, pp. 47–58 (2004).

Handfield, R., Branhardt, R., and Powell, N., “Mapping the Automotive Textile Supply Chain: The Importance of Information Visibility,” Journal of Textile and Apparel, Technology and Management, Vol.3, No. 4, pp.1-19 (2004).

Hardin R., “Trust and Trustworthiness,” Russell Sage Foundation: New York, NY (2002).

Hausman, W. H., and Peterson, R., “Multiproduct Production Scheduling for Style Goods With Limited Capacity, Forecast Revisions and Terminal Delivery,” ManagementScience, Vol. 18, No. 7, pp. 370-383, (1972).

Heath, D. C., and Jackson, P. L., “Modeling the Evolution of Demand Forecasts With Application To Safety Stock Analysis in Production/Distribution Systems,” IIE Transactions, Vol. 26, No. 3, pp. 17-30 (1994).

A. Hertz, E. Taillard, D. de Werra, "A Tutorial on Tabu Search", Proceedings of Giornate di Lavoro AIRO'95, (Enterprise Systems: Management of Technological and Organizational Changes), pp. 13-24 (1995).

Helferich, O. K. and Cook, R. L., Securing the Supply Chain, Council of Logistics Management (2002).

Hendricks K. B. and Singhal, V., “The Effect of Supply Chain Glitches on Shareholder Value,” Journal of Operations Management, Vol. 21, No. 5, pp. 501-522 (2003).

Hendricks, K. B., and Singhal, V. R., “An Empirical Analysis of the Effect of Supply Chain Disruptions on Long-Run Stock Price Performance and Equity Risk of the Firm,”Production and Operations Management, Vol. 14, No. 1, pp. 35-52 (2005a).

Hendricks, K. B., and Singhal, V. R., “Association Between Supply Chain Glitches andOperating Performance,” Management Science, Vol. 51, No. 5, pp. 695-711 (2005b).

Hendricks, K. B., Singhal, V. R., “Demand-Supply Mismatches and Stock Market Reaction: Evidence from Excess Inventory Announcements,” Forthcoming in Manufacturing and Service Operations Management (2009).

Hershauer, J. C., Walsh, K. D., and Tommelein, I. D., “Issues in construction supply chains,” Proceeding of the European Operations Management Association 8th Annual Conference (EurOMA '01), 3-5 June, Bath, UK, 345-355 (2001).

Hillman, M., and Keltz, H., “Managing Risk in the Supply Chain—A Quantitative Study,”accessed May 2009, available at

165

http://www.amrresearch.com/Content/DisplayPDF.aspx?compURI=tcm:7-13798. (2007).

Hillman, M., and Sirkisoon, F., “Pandemic Readiness Study. Report,” AMR Research, accessed May 2009, available at http://www.amrresearch.com/research/reports/images/2006/0605AMR-A-19413.pdf(2006).

Holland, J. H., Adaptation in Natural and Artificial Systems. MIT Press, Cambridge, MA(1992).

HSC, “Human Security Report,” prepared by the Human Security Center, The University of British Columbia, Canada (2005).

Hwang, P. and Burgers, P., “Properties of Trust: An Analytical View,” Organizational Behavior and Human Decision Processes, Vol. 69, pp. 67-73. (1997).

Huggins, E. L., and Olsen, T. L., “Inventory Control With Generalized Expediting,” Working paper, School of Business Administration, Fort Lewis College (2005).

Huh, W. T., Janakiraman, G. Muckstadt, J. A., and Rusmevichientong, P., “Asymptotic Optimality of Order-up-to Policies in Lost Sales Inventory Systems,” Management Science, Vol. 55, No. 3, pp. 404-420 (2009).

Hurley, G., Jackson, P., Levi, R., Roundy, R. O., and Shmoys, D. B., New Policies for Stochatic Inventory Control Models - A Theoretical and Computational Study. Working paper. Sloan School of Management, MIT, Cambridge, MA (2006).

Institute for Business and Home Safety, accessed September 14, 2005, available at http://www.ibhs.org/publications/ (2005).

James, H., “The Trust Paradox: A Survey of Economic Inquiries into the Nature of Trust and Trustworthiness,” Journal of Economic Behavior and Organization, Vol. 47, pp. 291-307 (2002).

Janakiraman, G. and Roundy, R. O., “Lost-Sales Problems with Stochastic Lead Times: Convexity Results for Base-Stock Policies,” Operations Research, Vol. 52, No. 5, pp.795-803 (2004).

Janakiraman, G., Seshadri, S., and Shanthikumar, J. G., “A Comparison of the Optimal Costs of Two Canonical Inventory Systems,” Working paper, Stern School of Business, New York University (2006).

Johnson, M. E., “Learning from Toys: Lessons in Managing Supply Chain Risk from the Toy Industry,” California Management Review, Vol. 43, No. 3, pp. 106–126 (2001).

Jones, E. L., The European Miracle: Environments, Economies, and Geopolitics in the History of Europe and Asia, Cambridge University Press, Cambridge (1981).

Jordan, W. C. and Graves, S. C, "Principles on the Benefits of Manufacturing Process Flexibility," Management Science, Vol. 41, No. 4, pp. 577 – 594 (1995).

Juneja, M., and Rajamani, D., “Lecture Series: Consumer Electronics Supply Chain Management,” accessed December 2004, available at https://som.utdallas.edu/c4isn(2003).

166

Kahn, J. A. “Inventories and Volatility of Production,” American Economic Review, Vol. 77, No. 4, pp. 667-679 (1987).

Karlin, S., and Scarf., H., “Inventory Models of the Arrow-Harris-Marschak Type with Time Lag. in K. J. Arrow, S. Karlin, and H. E. Scarf (Eds.),” Studies in the mathematical theory of inventory and production, Stanford, CA: Stanford University Press, pp.155–178 (1958).

Kelleher, K., “Why FedEx is gaining ground,” Business 2.0, 56-57 (October, 2003).

Ketzenberg, M., Metters, R., and Vargas, V., “Inventory policy retail outlets,” Journal of Operations Management, Vol. 18, No. 3, pp. 303–316 (2000).

Kimbrough, S. O., Wu, D., and Zhong, F., “Computers Play the Beer Game: Can Artificial Agents Manage Supply Chains?” Decision Support Systems, Vol. 33, No. 3, pp. 323-333 (2002).

Kleindorfer, P. R. and Saad, G. H. “Managing Disruption Risks in Supply Chains,” Production and Operations Management, Vol. 14, No. 1, pp. 53-68 (2005).

Knights, D., Noble, F., Vurdbakia, T., and Willmott, H., “Chasing Shadows: Controls, Virtuality, and the Production of Trust,” Organization Studies, Vol. 22, pp. 333-337 (2001).

Latour, A., “Trial by Fire: A Blaze in Albuquerque Sets off Major Crisis for Cell-phones Giants” Wall Street Journal, p. A1 (January 29, 2001).

Law, A. M., and Kelton, W. D., Simulation Modeling and Analysis. McGraw-Hill, New York, NY (2000).

Lawson, D., and Porteus, E., “Multi-Stage Inventory Management With Expediting,”Operations Research, Vol. 48, No. 6, pp. 878-893 (2000).

Lee, H., “Can Technology Save the Day?” Internet Week (October 8, 2001).

Lee, H. L., Padmanabhan, V., and Whang, S., “Information Distortion in a Supply Chain: the Bullwhip Effect,” Management Science, Vol. 43, No. 4, pp. 546-558 (1997).

Lee H. L., Padmanabhan, V., and Whang, S., “The Bullwhip Effect in Supply Chains,” Sloan Management Review, Spring 1997, pp. 93-101 (1997b).

Lee, H. L. and Wolfe, M., “Supply Chain Security without Tears,” Supply Chain Management Review, pp. 12-18 (January-February 2003).

Lee, H. L., So, K. C., and Tang, C. S., “The Value of Information Sharing in a Two-Level Supply Chain,’ Management Science, Vol. 46, No. 5, pp. 626-643 (2000).

Lester, S. and Browser, H., “In the Eyes of the Beholder: The Relationship between Subordinates’ Felt Trustworthiness and Their Work Attitudes and Behaviors,” Journal of Leadership and Organizational Studies, Vol. 10, pp. 17-33 (2003).

Lewis, B. M., Erera, A. L., and White, C. C., An Inventory Control Model With Possible Border Disruptions. Working paper, School of Industrial and Systems Engineering, Georgia Institute of Technology (2008).

167

Li, J., Sikora, R., Shaw, M., and Tan, G.W., “A strategic analysis of inter-organizational information sharing,” Decision Support Systems, Vol. 42, No. 1, pp. 251-266 (2006).

Li, G., Wang, S., Yan, H., and Yu, G., “Information Transformation in a Supply Chain: A Simulation Study,” Computers and Operations Research, Vol. 32, No. 3, pp. 707–725 (2005).

Lieberman, M. B. and Asaba, S., “Inventory Reduction and Productivity Growth: A Comparison of Japanese and US Automotive Sectors,” Managerial and Decision Economics, Vol. 18. No. 2, pp. 73-85 (1997).

Loebbecke, J., Eining, M., and Willingham, J., “Auditor’s Experience with Material Irregularities: Frequency, Nature, and Detectability,” Auditing: A Journal of Practice and Theory, Vol. 9, pp. 1-28 (1988).

Lott, N. and Ross, T., “Tracking and Evaluating U.S. Billion Dollar Weather Disasters, 1980–2005,” Proceedings of the 86th American Meteorological Society Annual Meeting, Atlanta, GA (January 29- February 2, 2006).

Lund, S., “Securing the High Tech Supply Chain,” Supply Chain Response to Terrorism -Planning for the Unexpected, MIT Industrial Liaison Program, Cambridge MA (2002).

Makridakis, S. and Hibon, M., “The M3 Competition: Results, Conclusions and Implications,” International Journal of Forecasting, Vol. 16, No. 4, pp. 451-476 (2000).

Marinez-de-Albeniz, V. and Simchi-Levi, D., “Mean-Variance Trade-Offs in Supply Contracts,” Naval Research Logistics, Vol. 53, No. 7, pp. 603-616 (2006).

Martha, J., and Subbakrishna, S., “Targeting a Just-In-Case Supply Chain for the Inevitable Next Disaster,” Supply Chain Management Review, Vol.5, pp. 18-24 (2002).

Martha, J. and Vratimos, E., “Creating a Just-in-Case Supply Chain for the Inevitable Next Disaster,” Mercer Management Journal, No. 14, pp. 70-77 (November 2002).

Matsumara, E. and Tucker, R., “Fraud Detection: A Theoretical Foundation,” The Accounting Review, Vol. 67, pp. 753-82 (1992).

McKnight, D. and Chervany, N., “The Meanings of Trust,” MISRC Working Paper Series, University of Minnesota, Available at http://misrc.umn.edu/wpaper/wp96-04.htm (1996).

Mentzor, J. T., “Global Supply Chain Risk Management,” International Journal of Physical Distribution and Logistics Management, Vol. 38, No. 3, pp. 192-223 (2008).

Metters, R., “Production Planning with Stochastic Seasonal Demand and Capacitated Production,” IIE Transactions, Vol. 29, No. 11, pp. 1017–1029 (1997).

Metters, R., “Quantifying the Bullwhip Effect in Supply Chains,” Journal of Operations Management, Vol. 15, No. 2, pp. 89-100 (1998).

Milgrom, P. and Roberts, J., “Communication and Inventory as Substitutes in Organizing Production,” Scandinavian Journal of Economics, Vol. 90, No. 3, pp. 275-289 (1988).

168

Mitroff, I. I. and Alpaslan, M. C., “Preparing for Evil,” Harvard Business Review, Vol. 81, No. 4, pp. 109-115 (April 2003).

Monahan, S., Laudicina, P., and Attis, D., “Supply Chains in a Vulnerable, Volatile World,” Executive Agenda, Vol. 6, No. 3, pp. 5-16 (Third Quarter 2003).

Muharremoglu, A. and Tsitsiklis, J. N., “Dynamic Leadtime Management in Supply Chains,”Unpublished manuscript, Columbia Business School, Columbia University (2003).

Murphy, T., “JIT When ASAP Isn’t Good Enough,” Wards Auto World, Vol. 35, No. 5, pp. 67-73 (May 1999).

Nahmias, S., Production and Operations Analysis, 6th edition, McGraw-Hill, Boston MA (2009).

Nahmias, S., Production and operations analysis. Burr Ridge, IL: Irwin (1993).

O’Malley, S., “SARS Could Cost Economy $1bn,” The Australian (July 12, 2003).

Parlar, M., and Perry. D., “Analysis of a (Q; R; T) Inventory Policy With Deterministic and Random Yeilds When Future Supply is Uncertain,” European Journal of Operational Research, Vol. 84, No. 2, pp. 431-443 (1995).

Patterson A., “Strategic Sample Size Choice in Auditing,” Journal of Accounting Research, Autumn, pp. 272-93 (1993).

Peck, H., “Drivers of Supply Chain Vulnerability: An Integrated Framework,” International Journal of Physical Distribution and Logistics Management, Vol. 35, No. 4, pp. 210–232 (2005).

Peck, H. and Juttner, U., “Risk Management in the Supply-Chain,” Centre for Logistics and Supply Chain Management, Cranfield School of Management, Cranfield UK (2003).

Parker, R., and Kapuschiski, R., “Optimal Policies for a Capacitated Two-Echelon Inventory System, Operations Research, Vol. 52, No. 5, pp. 739–755 (2004).

Perrow, C., Normal Accidents: Living With High Risk Technologies, Princeton University Press, Princeton NJ (1984).

Pielke, R. A. and Landsea, C. W., “Normalized Hurricane Damages in the United States, 1925-1995,” Weather and Forecasting, Vol. 13, No. 3, pp. 621-631 (1998).

Porteus, E. L., Foundations of Stochastic Inventory Theory. Stanford, CA: Stanford University Press (2004).

Qi, L., and Shen, Z. M., “A Supply Chain Design Model with Unreliable Supply,” Naval Research Logistics, Vol. 54, No. 8, pp. 829-844 (2007).

Randall, T. and Ulrich, K., “Product Variety, Supply Chain Structure, and Firm Performance: Analysis of the U.S. Bicycle Industry,” Management Science, Vol. 47, No. 12, pp. 1588 – 1604 (2001).

Rajagopalan S., and Swaminathan, J. M., “A Coordinated Production Planning Model With Capacity Expansion and Inventory Management,” Management Science, Vol. 47, No. 11, pp. 1562-1580 (2001).

169

Rao, U., Scheller-Wolf, A., and Tayur, S., “Development of a Rapid-Response Supply Chain at Caterpillar,” Operations Research, Vol. 48, No. 2, pp. 189-204 (2000).

Reed, M., “Organization, Trust and Control: A Realist Analysis,” Organization Studies, Vol. 22, pp. 201-28 (2001).

Rego, C., RAMP: A New Metaheuristic Framework for Combinatorial Optimization. In: Metaheuristic Optimization via Memory and Evolution: Tabu Search and Scatter Search, C. Rego and B. Alidaee (Eds.), Kluwer Academic Publishers, 441-460 (2005).

REMI model (Regional Economic Modeling, Inc.), Amherst, MA (2007).

Rice, J. B. and Caniato, F., “Building a Secure and Resilient Supply Network,” Supply Chain Management Review, pp. 22-30 (September-October 2003).

Robinson, L. W., Bradley, J. R., and Thomas, L. J., “Consequences of Order Crossover under Order-Up-To Inventory Policies,” Manufacturing and Service Operations Management, Vol. 3, No. 3, pp. 175-188 (2001).

Rosoff, H. and von Winterfeldt, D., “A Risk Economic Analysis of Dirty Bomb Attacks on the Ports of Los Angeles and Long Beach,” Risk Analysis, Vol. 27, No. 3, pp. 533-546 (2007).

Ross, A. M., Rong, Y., and Snyder, L. V., “Supply Disruptions with Time-Dependent Parameters,” Computers and Operations Research, Vol. 35, No. 11, pp. 3504-3529(2008).

Ross, T., “Billion Dollar U.S. Weather Disasters 1980-2003,” National Climate Data Center, Asheville NC (October 2003).

Roundy, R. O., and Muckstadt, J. A., “Heuristic Computation of Periodic-Review Base Stock Inventory Policies,” Management Science, Vol. 46, No. 1, pp. 104–109 (2000).

Sadgrove, K., The Complete Guide to Business Risk Management, Grover Publishing, Aldershot, Hampshire, UK (2005).

Sandia National Laboratories Private conversation with N-ABLE Project Leader (2007).

Santoso, T., Ahmed, S., and Goetschalckx, M., “A Stochastic Programming Approach for Supply Chain Network Design Under Uncertainty,” European Journal of Operational Research, Vol. 167, pp. 96–115 (2005).

Scaparra, M. P., and Church, R. L., “A Bilevel Mixed-Integer Program for Critical Infrastructure Protection Planning Source,” Computers and Operations Research, Vol. 35, No. 6, pp. 1905-1923 (2008).

Schmitt, T., “Resolving uncertainty in manufacturing systems,” Journal of Operations Management, Vol. 4, No. 4, pp. 331-345 (1984).

Schmitt, T. G., and Faaland, B. H., “Scheduling Recurrent Construction,” Naval Research Logistics, Vol. 51, No. 8 (2004).

Schmitt, T. G., Kumar, S., Stecke, K. E., Ehlen., M. A., and Glover, F., “Modeling and Analyzing Economic Consequences of Supply Chain Disruptions,” Working Paper

170

No. SOM 2007102, School of Management, University of Texas at Dallas, Dallas TX (August 2007).

Schroeder, R. G., Operations Management Contemporary Concepts and Case studies. McGraw-Hill, New York, NY (1993).

Seattle DOT (Department of Transportation), Freight Mobility Strategic Action Plan, SDOT, Seattle Municipal Tower, Seattle, WA, p. 22 (2005).

Seyed-Mahmoud, A., “Improving Logistics Operations across the Food Industry Supply Chain,” International Journal of Contemporary Hospitality Management, Vol. 16, No. 4, pp. 263-268 (2004).

Shah, N., “Pharmaceutical Supply Chains: Key Issues and Strategies for Optimization,” Computers and Chemical Engineering, Vol. 28, pp. 929–941 (2004).

Shang, K. H., and Song, J. S., “Newsvendor Bounds and Heuristic for Optimal Inventory Policies in Serial Supply Chains,” Management Science, Vol. 49, No. 5, pp. 618–638(2003).

Shaw, H., “Major Tokyo Quake World Cost 1.2 Trillion, Study Says,” New York Times, September 20 (1994).

Sheffi, Y., “Supply Chain Security,” Supply Chain Response to Terrorism - Planning for the Unexpected, MIT Industrial Liaison Program, Cambridge MA (December 2002).

Sheffi, Y., “Supply Chain and Terrorism,” Ch. 8 in The Towers Lost and Beyond, Kausel, E., Editor, MIT Press, Billerica MA (2003).

Sheffi, Y., “The Resilient Enterprise,” Presented at the Conference on National Security, Natural Disasters, Logistics and Transportation: Assessing the Risks and the Responses, University of Rhode Island, RI, September 25-26 (2006).

Sheffi, Y., “The Resilient Enterprise: Overcoming Vulnerability for Competitive Advantage,” MIT Press, Cambridge, Massachusetts (2007).

Sheffi, Y. and Rice, J., “A Supply Chain View of the Resilient Enterprise,” MIT Sloan Management Review, Vol. 47, No. 1, pp. 41–48 (2005).

Shibano, T., “Assessing audit risk from errors and irregularities,” Journal of Accounting Research Supplement, pp. 110-40 (1990).

Simchi-Levi, D., Snyder, L., and Watson, M., “Supply Chain Responsiveness in the Wake of a Major Crisis,” iSource Magazine, http://www.isourceonline.com/ (October 2001).

Sinha, P., Whitman, L., and Malzahn, D., “Methodology to Mitigate Supplier Risk in an Aerospace Supply Chain,” Supply Chain Management: An International Journal, Vol. 9, No. 2, pp. 154–168 (2004).

Skinner, D. and Spira, L., “Trust and Control – A Symbiotic Relationship,” Corporate Governance 3, pp. 28-35 (2003).

Snyder, L. V., “Facility Location under Uncertainty: A Review,” IIE Transactions, Vol. 38, No. 7, pp. 537-554 (2006).

171

Snyder, L. V., “A Tight Approximation for a Continuous-Review Inventory Models with Supply Disruptions,” Working paper, Department of Industrial and Systems Engineering, Lehigh University, Bethlehem, PA (2006).

Snyder, L. V., and Daskin. M. S., “Reliability Models for Facility Location: The Expected Failure Cost Case,” Transportation Science, Vol. 39, No. 3, pp. 400-416 (2005).

Snyder, R. D., Koehler, A. B., Hyndman, R. J., and Ord, J. K., “Exponential smoothing models: Means and variances for lead-time demand,” European Journal of Operations Research. Vol. 158, No. 2, pp. 444-455 (2004).

Snyder, R. D., Koehler, A. B., and Ord, J. K., “Lead time demand for simple exponential smoothing: An adjustment factor for the standard deviation,” Journal of Operations Research Society, Vol. 50, No. 10, pp. 1079-1082 (1999).

Snyder, L. V., Scaparra, M. P., Daskin, M. S., and Church. R. L., Planning for Disruptions in Supply Chain Networks. In TutORials in Operations Research, Johnson, M. P., B. Norman, and N. Secomandi (eds.), Chapter 9, Baltimore: INFORMS, 234-257 (2006).

Snyder , L. V. and Shen, Z. J. M., Supply and Demand Uncertainty in Multi-Echelon Supply Chains. Working paper, Lehigh University (2009).

Sodhi, M. S., “Managing Demand Risk in Tactical Supply Chain Planning for a Global Consumer Electronics Company,” Production and Operations Management, Vol. 14, No. 1, pp. 69-79 (Spring 2005).

Souder, E., Retailers Rely More on Fast Deliveries. The Wall Street Journal (January 14, 2004).

Stauffer, D., “Risk: The Weak Link in your Supply Chain,” Harvard Management Update, Vol. 8, No. 3, pp. 3–5 (2003).

Stecke, K. E., “Using Mathematics to Solve Some Problems in Industry,” Informs Transactions on Education, Vol. 5, No. 2 (2005).

Stecke, K. E., and Kumar, S., Sources of Supply Chain Disruptions, Factors That Breed Vulnerability, and Mitigating Strategies,” Journal of Marketing Channels, Vol. 16, No. 3, pp. 193-226 (2009).

Sullivan, L., “Most Companies Fail to Plan for Supply-Chain Disruption: Study,” AMR Research, Accessed August 14, 2007, available at http://www.intelligententerprise.com/ showArticle.jhtml?articleID=187002838 (2006).

Svensson, G., “A Conceptual Framework for the Analysis of Vulnerability in Supply Chains,” International Journal of Physical Distribution and Logistics Management, Vol. 30, No. 9, pp. 731-749 (2000).

Swaminathan, J. M., Smith, S. F., and Sadeh, N. M., “Modeling the Dynamics of Supply Chains: A Multi-Agent Approach,” Decision Sciences, Vol. 29, No. 3, pp. 607-632 (1997).

172

Swinney, R. and Netessine, S., “Long-term Contracts under the Threat of Supplier Default,” Manufacturing and Services Operations Management, forthcoming (2008).

Therkelsen, D. and Fiebich, C., “The Supervisor: The Linchpin of Employee Relations,” Journal of Communication Management, Vol. 8, pp. 120-129 (2003).

Thomas, S., Kumar, S., Stecke, K., Ehlen, M., and Glover, F., “Modeling and Analyzing Economic Consequences of Supply Chain Disruptions,” working paper, University of Texas at Dallas (2007).

Tomlin, B., “On the Value of Mitigation and Contingency Strategies for Managing Supply Chain Disruption Risks,” Management Science, Vol. 52, No. 5, pp. 639–657 (2006).

Tomlin, B. T. and Snyder, L. V., “On the Value of a Threat Advisory System for Managing Supply Chain Disruptions,” Working Paper, Kenan-Flagler Business School, University of North Carolina-Chapel Hill (2007).

Tomlin, B. T., and Wang, Y., “On the Value of Mix Flexibility and Dual Sourcing in Unreliable Newsvendor Networks,” M&SOM, Vol. 7, No. 1, pp. 37-57 (2005).

Towill, D. R., “Supply Chain Dynamics,” International Journal of Computer Integrated Manufacturing, Vol. 4, No. 4, pp. 197-208 (1991).

Towill, D. R., Naim, M. M., and Winker, J., “Industrial Dynamics Simulation Models in the Design of Supply Chains,” International Journal of Physical Distribution and Logistics Mangement, Vol. 22, No. 5, pp. 3–13 (1992).

Turner, K., and Williams, G., “Modeling Complexity in the Automotive Industry Supply Chain,” Journal of Manufacturing Technology Management, Vol. 16, No. 4, pp. 447-458 (2005).

U.S. Census Bureau. County business patterns. Washington, D.C., accessed on July 25, available at www.census.gov/ (2009).

U.S. Department of State, Annual "Patterns of Global Terrorism" Reports for the Years 1991 through 2003, <http://www.state.gov/>

Using Mathematics to Solve Some Problems in Industry Kathryn E. Stecke, INFORMS Transaction on Education Volume 5, Number 2 (January 2005).

Veeraraghavan, S., and Scheller-Wolf, A., “Now Or Later: A Simple Policy for Effective Dual Sourcing in Capacitated Systems,” Operations Research, Vol. 56, No. 4, pp.850-864 (2008).

von Winterfeldt, D. “A Risk and Decision Analysis Framework to Protect New Orleans Against Another Katrina,” in R. J. Daniels, D. F. Kettl, and H. Kunreuther (Editors) on Risk and Disaster, Pennsylvania State University Press, PA (2006).

Weigend, A. S. and Gershenfeld, N. A., “Time Series Prediction: Forecasting the Future and Understanding the Past,” Addison-Wesley, Menlo Park, CA (1993).

Whitmore, A. S. and Saunders, A., “Optimal Inventory under Stochastic Demand with Two Supply Options,” SIAM Journal of Applied Mathematics, Vol. 32, No. 2, pp. 293-305 (1977).

173

Wicks, A., Berman, S., and Jones, T., “The Structure of Optimal Trust: Moral and Strategic Implications,” The Academy of Management Review, Vol. 24, pp. 99-116 (1999).

Wikner, J., Towill, D. R., and Naim, M., “Smoothing Supply Chain Dynamics,”International Journal of Production Economics, Vol. 22, No. 3, pp. 231-248 (1991).

Wilks, J. and Zimbelman, M., “Using Game Theory and Strategic Reasoning Concepts to Prevent and Detect Fraud” Vol.18, No. 3, pp. 173-185 (2004).

Wu, O. Q., and Chen, H., Optimal Control and Equilibrium Behavior of Production-Inventory Systems. Working paper, University of British Columbia (2009).

Ye, N. and Farley, T., “Information Sharing and Control in Homogenous and Hetergeneous Supply Networks Under Different Market Conditions,” International Journal of Modelling and Simulation, Vol. 26, No. 2, pp. 160-168 (2005).

Zipkin, P., “Stochastic Lead Times in Continuous-Time Inventory Models,” Naval Research Logistics Quarterly, Vol. 33, pp. 763–774 (1986).

Zipkin, Paul H., Foundations of Inventory Management, Boston: McGraw Hill, 2000.

Zsidisin, G. and Ellram, L., “An Agency Theory Investigation of Supply Risk Management. Journal of Supply Chain Management, Vol. 39, No. 3, pp. 15–29 (2003).

Zsidisin, G., Ellram, L., Carter, J., and Cavinato, J., “An Analysis of Supply Chain Assessment Techniques,” International Journal of Physical Distribution and Logistics Management, Vol. 34, No. 5, pp. 397–413 (2004).

Zsidisin, G., Melnyk, S., and Ragatz, G., “An Institutional Theory Perspective of Business Continuity Planning for Purchasing and Supply Management,” International Journal of Production Research, Vol. 43, No. 16, pp. 3401–3420 (2005).

Zsidisin, G. A., Ragatz, G. L., and Melnyk, S. A., “Supply Continuity Planning: Taking Control of Risk,” Inside Supply Management, Vol. 16, No. 1, pp. 21-26 (2005).

Zucker, G., “Production of Trust: Institutional Sources of Economic Structure,” 1840-1920. Research in Organizational Behavior, Vol. 8, pp. 53-111 (1987).

Zwirn, E., “Blackout? What Blackout,” < http://www.cfo.com> (August 18, 2003).

VITA

Sanjay Kumar was born to Lakshman Singh and Lakshmi Devi in Tezpur, in the Indian state

of Assam. After completing high school, Sanjay pursued education in Production and

Industrial Engineering. He developed an interst in teaching and research and after completing

a Masters in Industrial and Management Engineering. He then joined the University of Texas

at Dallas for a Ph.D. in Management Science with a concentration in Operations

Management.