Omega 52 (2015) 201–212

Contents lists available at ScienceDirect

Omega
journal homepage: [Link]/locate/omega

Centralization versus decentralization: Risk pooling, risk

diversification, and supply chain disruptions$
Amanda J. Schmitt a,n, Siyuan Anthony Sun b, Lawrence V. Snyder c, Zuo-Jun Max Shen b
a
Center for Transportation and Logistics, Massachusetts Institute of Technology, Cambridge, MA, USA
b
Dept. of Industrial Engineering and Operations Research, University of California, Berkeley, CA, USA
c
Dept. of Industrial and Systems Engineering, Lehigh University, Bethlehem, PA, USA

art ic l e i nf o a b s t r a c t

Article history: We investigate optimal system design in a multi-location system in which supply is subject to
Received 18 November 2012 disruptions. We examine the expected costs and cost variances of the system in both a centralized
Accepted 17 June 2014 and a decentralized inventory system. We show that, when demand is deterministic and supply may be
This manuscript was processed by Associate
disrupted, using a decentralized inventory design reduces cost variance through the risk diversification
Editor Chandra
effect, and therefore a decentralized inventory system is optimal. This is in contrast to the classical result
Available online 24 June 2014
that when supply is deterministic and demand is stochastic, centralization is optimal due to the risk-
Keywords: pooling effect. When both supply may be disrupted and demand is stochastic, we demonstrate that a
Inventory control risk-averse firm should typically choose a decentralized inventory system design.
Risk
& 2014 Elsevier Ltd. All rights reserved.
Supply chain disruptions

1. Introduction cost but (as we prove) not the cost variance, the risk-diversification
effect reduces the variance of cost but not the expected cost.
As supply chains expand globally, supply risk increases. Classi- We prove that the risk diversification effect occurs in systems
cal inventory models have generally focused on demand uncer- with supply disruptions. We also consider systems with both
tainty and established best practices to mitigate demand risk. supply and demand uncertainty, in which both risk pooling and
However, supply risk can have very different impacts on the risk diversification have some impact, and numerically examine
optimal inventory management policies and can even reverse the tradeoff between the two. The risk mitigated through risk
what is known about best practices for system design. diversification is disruption risk or supply risk, whereas the risk
In this paper, we focus on the impact of supply uncertainty on a mitigated through risk pooling is demand risk. We employ a risk-
multi-location system and compare two policies: centralization averse objective to minimize both risk sources and determine
(stocking inventory only at a central warehouse) and decentralization which effect dominates the system and drives the choice for
(stocking inventory at multiple warehouses). Our analysis is a special optimal inventory system design.
case of One-Warehouse Multiple-Retailer (OWMR) system analysis; Specifically, comparing centralized and decentralized inventory
while most research on the OWMR model allows inventory to be policies, we contribute the following:
held at both echelons, we allow inventory to be held at only one
echelon in order to consider two opposing effects that can occur: risk
The exact relationship between optimal costs and inventory
pooling and risk diversification. The risk pooling effect occurs when levels when demand is deterministic and supply is subject to
inventory is held at a central location, which allows the demand disruption.
variance at each retailer to be combined, resulting in a lower
The exact relationship between optimal cost variances when:
expected cost [14]. The risk diversification effect occurs when inven- ○ Demand is deterministic and supply is subject to disruption.
tory is held at a decentralized set of locations, which allows the ○ Demand is stochastic and supply is deterministic.
impact of each disruption to be reduced, resulting in a lower cost
Formulations of the expected cost and cost variance when
variance [25]. Whereas the risk-pooling effect reduces the expected demand is stochastic and supply is subject to disruption.

Evidence that decentralization is usually optimal under risk-
averse objectives.
$
This manuscript was processed by Associate Editor Chandra.
n
Corresponding author.
E-mail addresses: amandaschmitt@[Link] (A.J. Schmitt),
The remainder of the paper is organized as follows. In Section 2
sysun@[Link] (S.A. Sun), [Link]@[Link] (L.V. Snyder), we review the relevant literature. In Section 3 we analyze the risk-
shen@[Link] (Z.-J. Shen). diversification effect in a multi-location system with deterministic

[Link]
0305-0483/& 2014 Elsevier Ltd. All rights reserved.
202 A.J. Schmitt et al. / Omega 52 (2015) 201–212

demand and disrupted supply. We consider stochastic demand inventory policies. These papers provide a foundation for our
and deterministic supply in Section 4. In Section 5 we consider analysis, but our application to the OWMR model provides new
both demand uncertainty and supply disruption and again com- insights on the impact of supply disruptions in complex systems.
pare inventory strategies using a risk-averse objective to choose Regarding the management of supply risk, Aydin et al. [4] point
the optimal inventory design. We summarize our conclusions in out that decentralization, a common approach to mitigate supply
Section 6. Proofs for all propositions and theorems are given in the risk, creates a misalignment of incentives between suppliers and
Appendix. buyers, competition among suppliers, competition among buyers,
and asymmetric information among the supply chain parties,
while Ellis et al. [13] discuss the implementation of supply
2. Literature review disruption risk research. Interested readers are referred to Babich
et al. [5], Yang et al. [35,36] and Tang and Kouvelis [28] for a closer
Supply chain risk management has been widely studied ever discussion of competition versus diversification. Decentralization
since the concept of uncertainty was introduced into inventory also provides flexibility for decision makers at the strategic level.
theory. Uncertainty in supply chains is usually classified as either For example, Yu et al. [38] consides the effects of supply disrup-
demand uncertainty or supply uncertainty. A simple model with tions on the decision of whether to single- or dual-source. Wang
demand uncertainty is the newsvendor problem, which deter- et al. [33] compare dual sourcing and process improvement to
mines the optimal order quantity or inventory level to minimize mitigate supply risk. Sawik [19,20] and Qi [18] discuss supplier
the expected cost under stochastic demand in a single period for a selection when the supply is subject to disruptions.
single location. Eppen [14] extends the newsvendor problem to a Decentralization may be deployed not only occur among
multiple-location model and shows that under demand uncer- suppliers, but also in the inventory systems themselves. Snyder
tainty, a centralized inventory strategy provides risk-pooling and Shen [25] use simulation to study multiple complex inventory
benefits and reduces expected cost versus a decentralized strategy. systems, including the OWMR system with supply uncertainty
Demand pooling is now a familiar idea in operations management with inventory at a single echelon. Their simulation results show
and serves as a major instrument to protect against demand that, under supply disruptions, expected costs are equal for
uncertainty. Corbett and Rajaram [10] generalize Eppen's work to centralized and decentralized systems, but the variance of the
the case of non-normal, dependent demand and show that the cost is higher in centralized systems. They call this the risk-
magnitude of the risk-pooling effect increases when the demands diversification effect and suggest that it occurs because a disruption
are less positively dependent. Berman et al. [7] analyze the benefit in a centralized system affects every retailer and causes more
of inventory pooling in a multi-location newsvendor framework drastic cost variability. They conclude that risk diversification
and show that the absolute benefit of risk pooling increases with increases the appeal of inventory decentralization in a system
demand variability. with disruptions. In the context of multi-location inventory
There are several ways to implement demand pooling, includ- management, decentralization or diversification is leveraged to
ing transshipments, postponement, and product substitution. Yang achieve lower cost variance instead of the expected cost, as in, e.g.,
and Schrage [34] explore the conditions that cause risk pooling to Schmitt and Snyder [21] and Tomlin [29].
increase inventory under the setting of product substitution. In this paper we consider supply and demand uncertainty in a
Paterson et al. [17] review the literature on transshipments as an multiple-demand-point system where inventory may be held at a
analogue of inventory pooling. Alptekinoglu et al. [1] propose a centralized warehouse to mitigate the demand risk or at multiple
model of inventory pooling to meet differentiated service levels warehouses to mitigate the supply risk. We assume that inventory
for multiple customers. In general, these forms of pooling leverage may be held at only a single echelon in order to draw clear
centralization so as to diminish the impact of demand uncertainty conclusions regarding centralization versus decentralization. We
on supply chain performance, which can be thought of as a explore the implications of the risk diversification effect by devel-
generalization of the risk-pooling effect studied by Eppen [14] oping an analytical model for the expected cost and cost variance of
and by this paper. On the other hand, our work deviates from this a multi-location system subject to supply uncertainty. We analyti-
stream of literature by considering supply uncertainty as well. cally prove the presence of risk diversification in this system,
Supply uncertainty has been considered in several settings, discuss its impact, and examine the system under uncertainty in
including newsvendor (e.g., [11,30]) and EOQ (e.g., [6,16,23]) both supply and demand. When demand is deterministic and
systems. The two most commonly considered forms of supply supply is subject to disruptions, we determine the optimal inven-
uncertainty are supply disruptions (in which supply is halted tory levels and costs. For that case and the case in which demand is
entirely for a stochastic amount of time) and yield uncertainty stochastic and supply is deterministic, we quantify the cost var-
(in which the quantity delivered from the supplier is random). iance. We combine supply disruptions and stochastic demand in a
Chopra et al. [9] and Schmitt and Snyder [21] consider systems subsequent model and formulate the expected costs and cost
that have both supply disruptions and yield uncertainty. The variances. To consider the effect of cost variances in the decision
literature on single-echelon (newsvendor) systems with both of making process, we adopt a risk-averse objective to incorporate the
these types of supply uncertainty is extensive, and we omit cost variance with the expected cost at the same time.
an exhaustive review here. The reader is referred to Atan and In the latter part of this paper, we consider risk-averse objectives
Snyder [2] and Snyder et al. [27] for reviews of the literature on for inventory optimization. Risk-aversion, a topic which is gaining
inventory models with supply disruptions and Yano and Lee [37] momentum in the operations literature, has been considered in
for a discussion of the literature on single-echelon systems newsvendor models to mitigate demand uncertainty. For example,
with yield uncertainty. Multi-period models with stochastic Eeckhoudt et al. [12] show that order quantities decrease with
demand and supply have been considered by Schmitt et al. [22] increasing risk-aversion. Van Meighem [32] considers resource diver-
and Tomlin [29], among others, and we rely on several of their sification in newsvendor models with risk-averse objectives, advocat-
results in this paper. Schmitt et al. develop a closed-form approx- ing diversifying resource availability to protect against risk. Tomlin and
imate solution for the optimal base-stock level in the face of Wang [31] consider a single-period newsvendor setting with supply
disruptions and stochastic demand. Tomlin investigates multiple disruptions; they model loss-averse and conditional value-at-risk
strategies for coping with disruptions, including acceptance (doing (CVaR) objectives when deciding between single- and dual-sourcing
nothing proactively), sourcing (using multiple suppliers), and and between dedicated and flexible resource availability. Chen
 A.J. Schmitt et al. / Omega 52 (2015) 201–212 203

et al. [8] consider multi-period inventory models with risk-aversion, occur at the central warehouse in the centralized system or at any of
modeling both replenishment and pricing decisions. Tomlin [29] the multiple warehouses in the decentralized system. Disruptions
also considers a multi-period setting and employs a mean–variance occur independently at all locations in the system. Examples of
approach to consider risk in a two-supplier system in which one disruptions of this nature could include disruption of an upstream
supplier is subject to disruptions and the other is perfectly reliable supplier, failure of inbound transportation capabilities (e.g., a snow-
but more expensive. We show that under a risk-averse objective storm or flood that shuts down rail or major roadways), transporta-
function, the benefits of risk diversification typically surpass those of tion labor issues (e.g., a port strike), shipment quality issues (e.g., a
risk pooling and therefore decentralization is the optimal network bacterial contamination), or other major issues that interrupt
configuration. material flow to a single location. In our model, there is no back-
up source of material available (i.e., no cross-shipments or alternate
sources) when a disruption occurs at a facility. If a disruption occurs
3. Risk diversification effect with supply disruptions at the central warehouse, then all downstream locations feel any
resulting shortage equally. In both systems, disruptions “pause” the
In this section, we examine the impact of supply disruptions on flow of supply to the disrupted stage, but any inventory at that stage
both the centralized and the decentralized multi-location system. may still be used during the disruption; therefore, disruptions affect
Before we present the models, we describe our assumptions and a stage's supply-receiving function but not its demand-receiving
notation. function.
We model the disruptions as a random process governed by the
3.1. Assumptions and notation pmf πi and cdf F(i), where i is the number of consecutive periods
during which a given stage has been disrupted, as is typical in the
We consider a two-echelon inventory system where demand supply disruption literature (e.g., [15,29]). Therefore πi is the
occurs according to the same distribution across multiple loca- probability that a given stage has been disrupted for precisely i
tions. We assume that all locations follow a base-stock policy, and periods ði Z 0Þ and F(i) is the probability that it has been disrupted
we compare policies for stocking inventory at the central ware- for i periods or fewer. Disruptions may be governed by a Markov
house only (centralized system) or at multiple warehouses (decen- chain or a more general process. The locations have identical and
tralized system). We assume that the fixed ordering costs are zero independent disruption processes that are stationary over time.
for all the following analysis. This enables us to place orders in After modeling the disruptions using this random process, we can
every period regardless of any fixed order cost for a multi-period evaluate the expected costs and cost variances for both the
system. This is not unreasonable in internal distribution settings. centralized and decentralized multi-location systems. We use an
We discuss what impact this might have in our conclusions. additional subscript s to denote the disrupted-supply model
We also assume zero lead times since deterministic lead times discussed in this section.
would not impact optimal base-stock levels or order quantities,
but would simply require that orders be placed exactly the lead 3.2. Mean and variance of optimal cost
time quantity in advance.
A holding cost of h per unit per period is incurred at the For a base-stock inventory policy at a single warehouse, the
warehouse in the centralized system or at each of the warehouses costs in a given period depend on the state of the system, defined
in the decentralized system. The holding costs are evaluated after as the number of consecutive periods for which it has been
the demands are realized. Unmet demands are backordered, and a disrupted. The expected cost and optimal base-stock level for such
stockout penalty of p per unit per period is incurred in both a system are given by (see e.g., [26])
systems. 1
The parameters, decision variables (base-stock levels at each Es ½C ¼ ∑ π i ½hðSs  ði þ 1Þ dÞ þ þ pðði þ1Þd  Ss Þ þ  ð1Þ
location), and performance measures for the system are summar- i¼0

ized in Table 1. Note that since SnD is defined as the optimal base-
Ssn ¼ jd; where j is the smallest integer such that Fðj  1Þ Z p=ðp þ hÞ
stock level for an individual warehouse in the decentralized
system, the total inventory for that system is nSnD . ð2Þ
In both the centralized and decentralized systems, disruptions Schmitt et al. [22] discuss these results, demonstrating that the
occur randomly at the inventory location(s); that is, disruptions optimal base-stock level is an increasing step function of the
newsvendor fractile and the disruption probability. As the cost of
Table 1 disruptions increases, the optimal solution increases by discrete
Notation. jumps to the next whole period's worth of demand.
Incorporating these results in our analysis of the centralized
Notation Definition
and decentralized multi-location systems, we present the relation-
n Number of demand sets or locations; number of decentralized ships between the inventory levels, expected costs, and cost
warehouses variances for the various systems in the following theorem.
p Penalty cost per period at each facility
h Holding cost per period at each facility Theorem 1. In the systems subject to supply disruptions, the optimal
base-stock levels and performance measures in the centralized,
Sn Optimal base-stock level for an individual facility
decentralized, and single-facility setting are related as follows:
SnD Optimal base-stock level for each warehouse in the decentralized
system
SnC Optimal base-stock level for the warehouse in the centralized system 1. SsCn ¼ nSsDn ¼ nSsn :
2. Es ½C nC  ¼ Es ½C nD  ¼ nEs ½C n .
E½C Expected cost for an individual facility 3. V s ½C nC  ¼ nV s ½C nD  ¼ n2 V s ½C n :
E½C D  Expected cost for the decentralized system
E½C C  Expected cost for the centralized system
V ½C Variance of the cost for an individual facility
Proof. See Appendix.
V ½C D  Variance of the cost for the decentralized system
The theorem states that the total system inventory levels and
V ½C C  Variance of the cost for the centralized system
expected costs are equal in the centralized and decentralized
204 A.J. Schmitt et al. / Omega 52 (2015) 201–212

systems, but the cost variance is n times greater for the centralized we evaluate the cost variances of the systems at optimality in
system, suggesting that the decentralized system is preferred for a Section 4.2. To the best of our knowledge, ours is the first study to
risk-averse decision maker. The reduction in the cost variance for a consider the cost variance at optimality in these systems.
decentralized system while incurring the same level of the
expected cost as in the centralized system is called the risk 4.1. Expected cost
diversification effect. The risk diversification effects occur when
the supply is subject to disruption and demand is deterministic. The expected cost and optimal base-stock level for a single
The intuition behind Theorem 1, i.e., the risk diversification warehouse subject to stochastic demand following normal dis-
effect, is as follows. Because the centralized system is just an tribution Nðμ; σ 2 Þ are well known as
aggregation of the demands from the decentralized system with !
the identical random disruption process, the total base-stock levels Sd  μ
Ed ½C ¼ hðSd  μÞ þσðp þ hÞΦ1 ð3Þ
at optimality should be the same for the two systems. That is to σ
say, the total inventory devoted to each of the n sets of customers  
is the same in both systems. Therefore each customer set faces the p
Sdn ¼ μ þ σΦ  1 ð4Þ
same number of stockout periods (effects of disruptions) on p þh
R1
average in both systems due to the independence and homoge- where Φ1 ðxÞ ¼ x ðv  xÞφðvÞ dv is the standard normal loss func-
neity of the supply disruption processes. By defining the service tion. The expected cost and optimal base-stock level here can be
level as the number of stockout periods on average, commonly directly applied to a single warehouse in the decentralized system.
known as the type-1 service level, another way to say the above is For the centralized system, the demand follows a normal distribu-
that identical disruption processes in both systems cause the tion Nðnμ; σ 2C Þ, where σ 2C ¼ nσ 2 . Therefore, for the centralized
customer set to experience the same service level. We state this system,
formally in the following Corollary. !
Sd  nμ
Corollary 2. In the multi-location system subject to supply disrup- Ed ½C C  ¼ hðSdC nμÞ þ σ C ðp þ hÞΦ1 C ð5Þ
σC
tions, if base-stock levels are set optimally, the distribution of
consecutive periods for which a given warehouse is stocked out is  
p
identical for the centralized and decentralized systems. SdCn ¼ nμ þ σ C Φ  1 ð6Þ
pþh
The same stockout distribution causes the expected holding Since the decentralized system functions as n single-warehouse
and stockout costs to be the same in both systems. However, systems, Ed ½C D  ¼ nEd ½C and SdDn ¼ Sdn . We have the following
disruptions are less frequent and more severe in the centralized theorem:
system than in the decentralized system due to the independence
of the disruptions in the decentralized system and the aggregation Theorem 3 (Eppen, 1979). The optimal expected costs and inventory
of the demands in the centralized system. Therefore the centra- levels for the centralized and decentralized multi-location systems
lized system incurs a greater cost variance. subject to stochastic demand are as follows:
The relationships presented in Theorem 1 demonstrate the risk pffiffiffi
Ed ½C nD  ¼ nEd ½C nC  ð7Þ
diversification effect in systems in which the supply is subject to
 
disruption. If demand is deterministic, the risk from uncertain pffiffiffi p
supply is mitigated in the decentralized system, as the lower nSdDn  SdCn ¼ ðn  nÞσΦ  1 ð8Þ
pþh
variance for that system demonstrates. Moreover, expected costs
for the centralized and decentralized systems are equal. Snyder Proof. Given by Eppen [14]. □
and Shen [25] demonstrate a similar result using simulation,
though their model differs slightly from ours in that they assume This theorem exhibits the benefits of centralization: by serving
that inventory at a given stage may not be used when that stage is the warehouses' demand from a centralized inventory site, we
disrupted. However, the risk pooling effect, which favors a cen- pool the risk from demand uncertainty. As a result, the optimal
pffiffiffi
tralized system, is observed when demand is stochastic and supply expected cost for the centralized system is n times smaller for
is deterministic. We present these results in the next section. the decentralized system, and less inventory is required. Next, we
evaluate the cost variance for both systems at optimality.

4. Risk pooling effect with stochastic demand 4.2. Cost variance

In this section, we present the risk pooling effect when the The following theorem presents the relationship between the
demand is stochastic but supply is deterministic. We consider the cost variances for the two systems.
classical model discussed by Eppen [14]. Following Eppen, con-
sider two alternative system configurations: the centralized sys- Theorem 4. At optimality, the cost variances for the centralized and
tem and the decentralized system. In both systems, the supply is decentralized multi-location systems subject to stochastic demand
deterministic while the demand is stochastic. The demand is are equal:
normally distributed with the same mean μ and variance σ2 at V d ½C nC  ¼ V d ½C nD  ð9Þ
each of the decentralized warehouses. Therefore, the demand at
the central warehouse in the centralized system has mean nμ and
Proof. See Appendix, Section A.3. □
variance nσ 2 . We employ a superscript d to denote this model with
stochastic demand. Optimal base-stock levels are the major con- This theorem claims that centralization reduces the expected
cerns for both the centralized system and the decentralized cost of the system while not affecting the cost variance compared
system. We therefore review the results concerning the expected to a decentralized multi-location system, which is referred to as
costs and their corresponding optimal base-stock levels presented the risk pooling effect. This suggests that centralization is optimal
by Eppen [14]. Since we are interested in not only the average under the presence of stochastic demand and deterministic
performance of the systems but also the stability of the systems, supply.
 A.J. Schmitt et al. / Omega 52 (2015) 201–212 205

We therefore obtain the following conclusions so far. Under than in the decentralized system as suggested by the risk pooling
stochastic demand and deterministic supply, centralization is effect. However, we demonstrate numerically in Section 5.3 that
optimal due to the risk pooling effect, which affects the expected the centralized system has a lower expected cost than the
cost only and has no impact on cost variance. In contrast, under decentralized system in most of the instances we test, which
supply disruptions and deterministic supply, decentralization is confirms that, when demand is stochastic, the risk pooling effect
optimal due to the risk diversification effect, which affects the cost usually holds (i.e., the centralized system has lower expected cost),
variance only and has no impact on the expected cost. A natural even when supply is subject to disruption.
question is, when both demand uncertainty and disruptions are
present, which system is optimal, i.e., which prevails: risk pooling 5.2. Cost variance
or risk diversification? We address this question in the next
section. We present the cost variance for arbitrary base-stock levels
(SbD in the decentralized system and SbC in the centralized system)
in the following proposition.
5. System with supply disruptions and stochastic demand
Proposition 6. When demand is normally distributed at the ware-
In this section we compare the cost means and variances for houses and supply is subject to disruptions, the cost variances for the
centralized and decentralized multi-location systems in which decentralized and centralized multi-location systems are
each warehouse is subject to stochastic demand and the whole " !!
1
2 2 SbD iμ
system is subject to disrupted supply. We assume that when V b ½C D  ¼ n ∑ π i  1 ðh SbD 2 2SbD iμ þiσ 2 þi2 μ2 Þ þ2iσ 2 ðp2  h ÞΦ2 pffi
i¼1 σ i
supply is not disrupted, the yield is deterministic. We use a !!!2 #
1 pffi Sb iμ
superscript b to denote this model since both supply and demand  ∑ π i  1 hðSbD iμÞ þσ iðp þhÞΦ1 D pffi ð13Þ
are stochastic. We formulate the expected costs and cost variances i¼1 σ i
in Sections 5.1 and 5.2, respectively. We perform numerical !!
analysis to compare the performance of the two systems in 1
2 2 SbC  niμ
V b ½C C  ¼ ∑ π i  1 h ðSbC 2  2SbC niμ þ niσ 2 þ n2 i2 μ2 Þ þ 2niσ 2 ðp2  h ÞΦ2 pffiffiffiffiffi
Section 5.3. Finally, in Section 5.4, we introduce a mean–variance i¼1 σ ni
objective that allows us to choose optimally between the two !!!2
1 pffiffiffiffiffi Sb  niμ
systems when risk aversion is accounted for. We summarize our  ∑ π i  1 hðSbC  niμÞ þ σ niðp þ hÞΦ1 C pffiffiffiffiffi ; ð14Þ
i¼1 σ ni
findings in Section 5.5.
where Φ1 ðÞ is the standard normal loss function and Φ2 ðÞ is the
5.1. Expected costs standard normal second-order loss function.

At a single warehouse, a single period's demand is denoted D Proof. See Appendix.

and is distributed as Nðμ; σ 2 Þ, while the total demand in i periods is We have confirmed numerically that it is not always true that
distributed as Nðiμ; iσ 2 Þ. The cost for a single warehouse subject to V b ½C nC  Z V b ½C nD  (as the risk diversification effect would suggest).
normally distributed demand and disrupted supply is given by However, we demonstrate numerically in Section 5.3 that this
Schmitt et al. [22] as inequality holds for most of the instances we tested. Thus, when
Z b Z !
1 S 1 supply may be disrupted, the risk diversification effect usually
Eb ½C ¼ ∑ π i  1 h ðSb  idÞf i ðidÞ dd þ p ðid  Sb Þf i ðidÞ dd holds (i.e., the decentralized system has lower cost variance), even
b
i¼1 1 S
!! when demand is stochastic.
pffi b
1 S  iμ
1
b
¼ ∑ πi  1 hðS iμÞ þ σ iðp þ hÞΦ pffi ð10Þ
i¼1 σ i 5.3. Numerical study
b
Schmitt et al. [22] prove that E ½C is convex and argue that the
In this section, we perform a numerical study to compare the
optimal base-stock
pffi level cannot be found in closed form because
performance of the two systems when supply disruptions and
ΦððS  iμÞ=σ iÞ appears in the first-order condition for i ¼ 0;
stochastic demand are present simultaneously. More specifically,
1; …; 1.
we model the supply disruptions as a two-state, discrete-time
Using (10), we derive the expected costs for the two systems in
Markov process with disruption probability α and independent
the following Proposition.
recovery probability β. (That is, the probability of being disrupted
Proposition 5. When demand is normally distributed at the ware- [not disrupted] in the next period given that the system is not
houses and supply is subject to disruptions, the expected costs for the disrupted [disrupted] in the current period is α [β].)
centralized and decentralized multi-location systems are The steady-state probabilities for the disruption process are
!! given by [22]
pffi b
1 SD  iμ
1
b b
E ½C D  ¼ n ∑ π i  1 hðSD  iμÞ þ σ iðp þ hÞΦ pffi ð11Þ β αβ
i¼1 σ i π0 ¼ ; πi ¼ ð1  βÞi  1 ; i Z 1 ð15Þ
α þβ αþβ
!!
1 pffiffiffiffiffi Sb  inμ In Section 5.3.1, we vary the input parameters independently to
Eb ½C C  ¼ ∑ π i  1 hðSbC  inμÞ þσ inðp þhÞΦ1 C pffiffiffiffiffi ð12Þ determine the impact of each on the performance of the two
i¼1 σ in
systems. In Section 5.3.3 we examine the impact of the disruption
characteristics on the systems.
Proof. See Appendix.

Clearly, for the decentralized system, SbDn ¼ Sbn . For the centra- 5.3.1. Impact of parameters
lized system, we have confirmed numerically that SbCn a nSbn in We conducted a full-factorial experiment to determine the
general, as in the deterministic-demand model in Section 3. Unlike effect of the input parameters on the optimal base-stock levels and
that model, SbCn is neither consistently greater than nor consistently the cost mean and variance at optimality in the two multi-location
less than nSbn when demand is stochastic. As a result, the expected systems. In this section, we are interested in how the parameters
cost for the centralized system at optimality is not consistently less affect the performance metrics for a given set of customers.
206 A.J. Schmitt et al. / Omega 52 (2015) 201–212

Therefore, we look at a system that has 6 customer sets, i.e., n ¼6. Besides the overall results from the numerical study, we are
Without loss of generality, we fix μ to be 150 and h to be 1 and interested in how individual parameters affect the results. Fig. 2
then vary the single-location standard deviation, σ, and the stock- plots the relative difference in Eb ½C n , V b ½C n  and ρb ½C n  between the
out penalty, p, because the model is sensitive to changes in the decentralized and the centralized system for different parameters.
ratios σ/μ and z ¼ p=ðp þ hÞ (the newsvendor fractile) rather than to We found that the difference in expected cost is more sensitive to
changes in the individual parameters. In addition, we vary the changes in parameters than the difference in cost variance is. The
disruption and recovery parameters α and β. Table 2 lists the benefits of risk pooling/centralization are enhanced when σ
values tested for each parameter. The total number of parameter increases, z decreases, α decreases, or β increases. That is to say,
combinations is 12  6  6  16 ¼6912. the expected cost reduction from centralization is increased if high
For each combination of parameters, we optimized the uncertainty in demand is observed, backlogging incurs low pen-
expected cost functions (11) and (12) numerically to solve for alty cost, disruptions in supply do not occur frequently, or the
the optimal base-stock levels in both systems. For each instance, recovery rate is high. The intuition for these observations is as
we evaluate the total stock levels (nSbDn or SbCn ), the expected costs follows. The risk pooling effect is more pronounced when the
demand uncertainty is high, as is evident from the model in
at optimality Eb ½C n , the cost variances at optimality V b ½C n , and the
Section 4. When the recovery rate is high, or the supply disrup-
coefficient of variation (CV) of the cost at optimality, given by
qffiffiffiffiffiffiffiffiffiffiffiffiffiffi tions are not frequent, the supply suffers low uncertainty, as in the
ρb ðC n Þ ¼ V b ½C n =Eb ½C n . Table 3 summarizes the numerical results deterministic supply model in Section 4, and therefore risk pooling
for the 6912 instances we tested. The decentralized system has a benefits. When the penalty cost is low, the cost of depleting the
higher total stock level at optimality in 87% of the instances tested. stocks during disruptions is low, which has the same effect as low
This occurs because of the discrete nature of disruptions. If disruption rate, and therefore risk pooling benefits.
demand is deterministic, then the optimal base-stock level is We found the cost variance is, in general, lower for the
always an integer multiple of the demand (see Section 3.2), and decentralized system, confirming the risk diversification effect.
demand stochasticity perturbs these discrete base-stock levels Although the difference in cost variance is not sensitive to
only slightly. Since the total standard deviation of demand is parameter changes, their relative impacts on the system config-
smaller in the centralized system, the jumps are less smooth uration decision, which are reflected by the coefficients of varia-
(closer to the deterministic-demand solution) and may jump tion, are visible. The risk diversification effect is more pronounced
slightly above the optimal decentralized system levels. when σ decreases, z increases, α increases, or β decreases, which
The decentralized system has a higher expected cost at optim- are the opposite conditions for the risk pooling effects to be
ality for 99% of the instances tested, which confirms the presence visible. We can therefore use similar arguments as before to
of the risk-pooling effect in that centralization reduces expected interpret such results.
cost even with supply disruptions. In contrast to the higher
expected cost, the decentralized system has lower cost variance/ 5.3.2. Impact of number of demand locations
coefficient of variation in cost for more than 99% of the instances, We next present the impact of the number of demand loca-
which reveals the presence of the risk-diversification effect in that tions, n, on the differences between the centralized and decen-
decentralization reduces the cost variance even under demand tralized systems. Instead of varying n independently, we fix the
stochasticity. The histograms in Fig. 1 provide a distributional view total demand nμ to be 900 and the total demand standard
of the CV of the costs in decentralized and centralized systems. pffiffiffi
deviation nσ to be 60, and then vary the number of demand
The range of CV for the decentralized system is from 0 to 3.5 while locations n to obtain the demand distribution at individual facil-
the range of CV for the centralized system is from 0 to 9, indicating ities. This setting best mimics the situation in which the total
that the cost variance is much higher in the centralized system. demand in the market is stable and the demand at each individual
facility depends on the number of facilities built. We tested eight
scenarios consisting of the various combinations of the following:
Table 2
Parameter levels. disruption rate α is low/high, recovery rate β is low/high and
newsvendor fractile z is low/high. Fig. 3 summarizes the differ-
Variable Values ences in expected cost and cost variances for the two systems.
As is expected, an increase in the number of facilities amplifies
α (failure probability) 0.001, 0.005, 0.025, 0.01, 0.02, 0.05,
0.08, 0.1, 0.15, 0.2, 0.25, 0.3 the risk pooling effect as well as the risk diversification effect. The
risk pooling effect is most amplified when the disruption rate α is
β (recovery probability) 0.5, 0.75, 0.85, 0.90, 0.95, 0.99 low and the newsvendor fractile z (or the penalty cost p) is low. The
underlying intuition falls into the same argument as in Section 5.3.1.
σ (demand st. dev.) 5, 10, 15, 20, 25, 50
But the reduction in expected cost from risk pooling is smaller in scale
z ¼ p=ðp þ hÞ 0.50, 0.60, 0.65, 0.68, 0.71, 0.74, 0.77, 0.80, when the disruption rate is high or the penalty cost is high. In contrast,
(newsvendor fractile) 0.83, 0.86, 0.89, 0.92, 0.95, 0.98, 0.99, 0.995 the reduction in cost variance due to the risk diversification effect is
visible in all scenarios. As a result, as long as the supply disruptions
are frequent or the stockout penalty is high, risk-diversification is
more pronounced than risk-pooling. We therefore can conjecture that
Table 3
a risk-averse decision maker will be in favor of a decentralized system,
Decentralized system stocks more total inventory units, incurs higher expected
which is confirmed by the implementation of the mean–variance
total costs Eb ½C n  and lower cost variances V b ½C n , and therefore lower coefficient of
variation ρb ½C n , in most of the numerical experiments.
objective in Section 5.4.

Decentralized Frequency (%)

5.3.3. Impact of disruption profile
Stocks more 6011 86.96 Two pairs of ðα; βÞ values may result in the same fraction of
Higher cost 6901 99.84 periods disrupted, defined as
Lower variance 6909 99.96 α
Lower CV 6909 99.96 1  π0 ¼ ; ð16Þ
αþβ
 A.J. Schmitt et al. / Omega 52 (2015) 201–212 207

Decentralized System Centralized System

1400 1400

1200 1200

1000 1000

Frequency
Frequency

800 800

600 600

400 400

200 200

0 0

0 2 4 6 8 0 2 4 6 8
Coefficient of Variation Coefficient of Variation

Fig. 1. Coefficient of variation ρb ½C n  distribution for (a) decentralized system and (b) centralized system.

1.4 1.4 1.4 1.4

1.2 1.2 1.2 1.2

1.0 1.0 1.0 1.0

0.8 0.8 0.8 0.8

0.6 0.6 0.6 0.6

0.4 0.4 0.4 0.4

0.2 0.2 0.2 0.2

0.0 0.0 0.0 0.0

5 10 15 20 25 50 0.5 0.65 0.74 0.8 0.86 0.95 0.995 0.001 0.01 0.025 0.08 0.15 0.25 0.5 0.75 0.85 0.9 0.95 0.99

−0.70 −0.70 −0.70 −0.70

−0.80 −0.80 −0.80 −0.80

−0.90 −0.90 −0.90 −0.90

5 10 15 20 25 50 0.5 0.65 0.74 0.8 0.86 0.95 0.995 0.001 0.01 0.025 0.08 0.15 0.25 0.5 0.75 0.85 0.9 0.95 0.99

−0.4 −0.4 −0.4 −0.4

−0.5 −0.5 −0.5 −0.5

−0.6 −0.6 −0.6 −0.6

−0.7 −0.7 −0.7 −0.7

−0.8 −0.8 −0.8 −0.8

5 10 15 20 25 50 0.5 0.65 0.74 0.8 0.86 0.95 0.995 0.001 0.01 0.025 0.08 0.15 0.25 0.5 0.75 0.85 0.9 0.95 0.99

Fig. 2. Relative difference in expected cost ðEb ½C D   Eb ½C C Þ=Eb ½C C , cost variance ðV b ½C D   V b ½C C Þ=V b ½C C , and coefficient of variation ðρb ½C D   ρb ½C C Þ=ρb ½C C  of decentralized
and centralized systems at optimality.

but still have very different characteristics. For example, if α and β effect of the disruption profile on the optimal base-stock levels and
are both close to 0, then disruptions are rare but long, while if they performance measures of the two systems.
are both close to 1, then disruptions are frequent but short. Snyder We generated twelve groups of ðα; βÞ pairs. Within each group,
and Tomlin [24] refer to the frequency/duration characteristics of the ðα; βÞ pairs have the same percentage down-time (16) but
a disruption as the disruption profile. In this section, we study the a different disruption profile. For each ðα; βÞ pair, we vary the
208 A.J. Schmitt et al. / Omega 52 (2015) 201–212

1.4 α=0.01,β=0.75,z=0.80 1.4 α=0.15,β=0.75,z=0.80

● α=0.01,β=0.75,z=0.99 ● α=0.15,β=0.75,z=0.99
1.2 α=0.01,β=0.99,z=0.80 1.2 α=0.15,β=0.99,z=0.80
Relative Cost Difference (%)

Relative Cost Difference (%)

α=0.01,β=0.99,z=0.99 α=0.15,β=0.99,z=0.99
1.0 1.0

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

● ● ● ● ●
0.0 ● 0.0 ● ● ● ● ● ●

2 3 4 5 6 7 8 9 2 3 4 5 6 7 8 9
Number of Demand Locations n Number of Demand Locations n

0.2 α=0.01,β=0.75,z=0.80 0.2 α=0.15,β=0.75,z=0.80

Relative Cost Variance Difference (%)

Relative Cost Variance Difference (%)

● α=0.01,β=0.75,z=0.99 ● α=0.15,β=0.75,z=0.99
0.0 α=0.01,β=0.99,z=0.80 0.0 α=0.15,β=0.99,z=0.80
α=0.01,β=0.99,z=0.99 α=0.15,β=0.99,z=0.99

−0.2 −0.2

−0.4 −0.4 ●
●
●
−0.6 −0.6
● ●
● ●
● ●
−0.8 ● −0.8
● ●

−1.0 −1.0

2 3 4 5 6 7 8 9 2 3 4 5 6 7 8 9
Number of Demand Locations n Number of Demand Locations n

Fig. 3. Risk pooling (expected cost reduction) and risk diversification (cost variance reduction) of centralized and decentralized systems under low/high disruption rate, low/
high recovery rate, and low/high stockout penalty.

the CV difference ðρ½C nD   ρ½C nC Þ=ρ½C nC , which captures the relative

0.20

impact of cost variance with respect to expected cost, for the

centralized and decentralized systems. The figure suggests that the
disruption profile has a significant impact on the performance
0.15

measures.
The difference in CV slightly increases when the percent
Disruption Rate α

down-time decreases. Moreover, the difference in CV increases

when the disruption rate increases for given percent down-
0.10

time. This implies that the risk-diversification effect is more

pronounced when the disruptions are frequent/short rather than
rare/long. These results confirm results by Snyder and Tomlin [24],
Tomlin [29], and others that suggest that rare/long disruptions
0.05

are more difficult to plan for than frequent/short ones. In other

words, when disruptions are more frequent, the risk pooling effect
is less pronounced and the risk diversification effect is more
pronounced.
0.00

0.05 0.10 0.15 0.20

% Down Time 5.4. Optimal risk-averse design
Fig. 4. Reduction in coefficient of variation via decentralization increases when
disruption rate α increases for different disruption profiles. We have shown that for the OWMR system with both supply
disruptions and stochastic demand, inventory decentralization
reduces the cost variance via the risk-diversification effect while
disruption rate α in eight levels if a valid corresponding β can be centralization reduces the expected cost via the risk-pooling effect.
found. In all instances, we set σ ¼25 and p ¼4 (newsvendor fractile It is natural to ask which effect is more pronounced, and therefore
z¼ 80%) and, as before, set n¼ 6, h ¼1 and μ¼150. Fig. 4 presents which system a risk-averse decision maker should prefer.
 A.J. Schmitt et al. / Omega 52 (2015) 201–212 209

5.4.1. Mean–variance objective In order to investigate the conditions under which centralization
We employ the classical mean–variance approach, minimizing would be an optimal choice, we consider the case where the supply is
the following objective function: fairly stable. We set n¼6, μ¼ 150, σ¼ 25, α¼ 0.001, and β¼0.95,
ð1  κÞE½C þ κV½C ð17Þ resulting in a percent down-time of 1.05%. Fig. 5 plots the regions in
which each system is optimal for κ A ½0; 1 and p=ðp þ hÞ A ð0; 1Þ
where κ A ½0; 1. The larger κ is, the more risk-averse the decision (keeping h fixed to 1, i.e., p A ð0; 1Þ) and a zoom-in when the
maker is. Many other objectives for risk-averse decision making newsvendor fractile is greater than 0.5, i.e., p A ½1; 1Þ. The region in
have been proposed in the finance and operations literature. We which the centralized system is optimal is barely visible for the choice
chose to focus on the mean–variance objective primarily for of high service level/high stockout penalty. Clearly, the centralized
analytical tractability. Numerical studies using the Conditional system is optimal only for very risk-neutral decision makers operating
Value-at-Risk (CVaR) objective produced similar insights, but a under low service levels: for κ Z0:08 and p=ðp þ hÞ Z 0:5, the decen-
more formal theoretical analysis was precluded by the increased tralized system is almost always optimal. When κ¼0.5, i.e., when the
complexity of the objective function. decision maker places the same importance on the expected cost and
We optimized (17) numerically using a line-search procedure. the cost variation (standard deviation), centralization is optimal only if
This function is neither convex nor concave for all values of κ. the stockout penalty is under 0.2, i.e., centralization is optimal only if
There are values of κ and Sb such that (17) is convex and values for the service level is set to be very low. The intuition behind this result
which is not convex. By testing the function numerically, we found can be reasoned as follows. When stockouts are relatively expensive,
that for κ r 0:05, the range we used in our tests below, (17) is i.e., z is low, the impact of supply risk on the cost is small compared to
convex at nearly every value of Sb, with the exception of Sb  0. We the demand risk, and therefore the firm should centralize in order to
also found that the function is convex for all S Z μ, in the instances mitigate the demand risk through the risk pooling effect. When
we tested, and this range typically includes the optimal base-stock stockouts are relatively expensive, i.e., z is high, the variability
level. Moreover, the second derivative of (17) is positive at the dominates the objective, and the firm therefore should decentralize
solution found by our optimization procedure for every instance in order to mitigate the supply risk due to the risk diversification
we tested. Therefore, we are confident that our results represent effect.
the globally optimal solutions for each instance. The numerical studies suggest that the magnitude of the cost
variance dominates the expected cost for a risk-averse decision
5.4.2. Numerical study maker, as was implied in Fig. 1. Therefore, a risk-averse decision
One might argue that, because the variance term typically has a maker will always prefer a decentralized system configuration
higher order of magnitude than the expected cost term, the mean– when the uncertainty in supply reaches a certain level.
variance objective would artificially favor the decentralized system,
since it has a lower cost variance. Therefore, we chose to minimize the 5.5. Summary of numerical comparisons
pffiffiffiffiffiffiffiffiffi
mean-standard deviation objective ð1  κÞE½C þ κ V½C instead of the
mean–variance objective ð1  κÞE½C þ κV½C. (Our numerical tests, not For the multi-location system subject to both demand uncer-
given in detail here, confirm that the mean–variance objective gives tainty and supply disruptions:
similar results.) Our numerical study shows that even for very small
values of κ, the decentralized system is almost always optimal under
The risk-diversification effect exists, resulting in lower cost
pffiffiffiffiffiffiffiffiffi
the mean-standard deviation objective ð1 κÞE½C þ κ V½C. As long as variance in the decentralized system.
disruption rate α is high enough (exceeds 0.01) or the recovery rate β is ○ The risk-diversification effect is more pronounced when
no better than 0.80 (β¼1 corresponds to immediate response), the disruptions are more frequent, longer and more costly.
decentralized system is always optimal. This implies that if the firm is
The risk-pooling effect exists, resulting in lower expected cost
subject to either frequent supply disruption or slow disruption in the centralized system.
response, than decentralized system configuration should be preferred ○ An increase in the number of locations amplifies the risk-
for a risk-averse decision maker. pooling effect as well as the risk-diversification effect.

Penalty Cost p Penalty Cost p

0 0.2 0.7 1.5 4 Inf 1 1.5 2.3 4 9 Inf
0.10
0.8

0.08
Risk Aversion Level κ

Risk Aversion Level κ

0.6

0.06
Decentralized is Optimal Decentralized is Optimal
0.4
0.04

0.2
0.02
Centralized Centralized

0.0 0.00
0.0 0.2 0.4 0.6 0.8 1.0 0.5 0.6 0.7 0.8 0.9 1.0
Newsvendor Fractile z Newsvendor Fractile z
Fig. 5. Left: Optimal system configuration under 1.05% down time with various choices of newsvendor fractile z / stockout penalty p and risk aversion level κ. Right: A zoom-
in of the left figure for z Z 0:5.
210 A.J. Schmitt et al. / Omega 52 (2015) 201–212

○ The magnitude of the risk-pooling effect is less than that of would decrease the risk diversification benefits of decentraliza-
the risk-diversification effect unless the supply disruptions tion, whereas negatively correlated disruptions would increase
are rare or the stockout penalty is low. those benefits.

The number of demand locations affects the optimal solutions Another obvious extension to our work could address the
○ The risk-pooling effect does not always result in lower total assumption that inventory can only be held at one echelon of the
base-stock levels. supply chain, which we made to simplify our analysis and clarify
○ The risk-pooling effect is more pronounced when disrup- our conclusions. Of course, centralized and decentralized systems
tions are less frequent and shorter, or when demand is more are not the only choices available to a firm operating in a multi-
variable. location environment; the firm may also choose a hybrid system in

Disruption profiles affect the optimal solutions. which inventory is held at both echelons. Such a system may well
○ For the same percent down-time, more frequent disruptions provide a desirable balance between the risk-diversification and
result in more pronounced risk-diversification effect. risk-pooling effects. These systems are significantly harder to
○ Rare/long disruptions are more difficult to plan for than analyze (as the literature on multi-location system attests), but an
frequent/short ones through decentralization. investigation of these two competing effects in this context is an

For a risk-averse firm, the decentralized system is typically the important avenue for future research.
optimal inventory design.
○ This holds unless service level (newsvendor fractile), κ value,
or failure probabilities are very low. Acknowledgments

In conclusion, a risk-averse decision maker will prefer a This research was supported in part by National Science Founda-
decentralized system unless supply uncertainty is very low or tion Grants DGE-9972780, DMI-0522725, and DMI-0621433. This
the penalty cost is low, because the magnitude of risk-diver- support is gratefully acknowledged.
sification effect dominates the risk-pooling effect in most of the
cases.
Appendix A. Proofs

6. Conclusions A.1. Loss functions

In this paper, we consider a multi-location system with both Let X be a random variable with pdf f(x) and cdf F(x). The loss
supply and demand uncertainty. We investigate the risk- function for X is given by
diversification effect, in which the expected cost is the same in Z 1 Z 1
the centralized and decentralized systems but the cost variance is GðxÞ ¼ ðt  xÞf ðtÞ dt ¼ ð1  FðtÞÞ dt ð18Þ
smaller in the decentralized system. The intuition behind this x x

effect is that by distributing inventory at multiple sites, the impact and the second-order loss function is
of any one disruption is smaller, even though each site is still Z Z 1
affected by the same number of disruptions. The firm benefits 1 1
HðxÞ ¼ ðt  xÞ2 f ðtÞ dt ¼ GðtÞ dt: ð19Þ
from not putting all its eggs in one basket; although the same 2 x x

number of eggs may be destroyed, they are not all destroyed at Let Φ1 ðzÞ be the standard normal loss function and Φ2 ðzÞ as the
once. We proved that the risk-diversification effect occurs in standard normal second-order loss function.
multi-location systems with supply disruptions. The following lemma presents properties we use in the proofs
In contrast, the classical risk-pooling effect prevails under below. The proof of the lemma is omitted; it follows from well
demand uncertainty; under this effect, the expected cost is smaller known results concerning loss functions (e.g., [3,39]).
in the centralized system, and we prove that the variance is equal
in the two systems. We showed numerically that, when disrup- Lemma 7.
tions and demand uncertainty are both present, both effects occur
to a certain extent, but that in most cases the risk-diversification
effect strongly dominates the risk-pooling effect. Therefore, a 1. Let X have cdf F and loss function G. Then for any function θðxÞ,
decentralized system is optimal when disruptions and demand d
uncertainty are both present, except when the service level GðθðxÞÞ ¼  θ0 ðxÞ½1  FðθðxÞÞ: ð20Þ
dx
(newsboy fractile) is very low, the firm is very risk neutral, and/
or the system is very reliable. 2. If X is normally distributed with mean μ and variance σ2 and f(x) is
Natural extensions to this research can be made by relaxing the its pdf, then
assumptions in our models. For example, we assumed zero fixed  
Z 1
ordering costs. An argument could be made that this ignores an Sμ
ðx  SÞ2 f ðxÞ dx ¼ 2σ 2 Φ2 : ð21Þ
additional cost benefit from centralization; synchronizing orders S σ
could reduce the total fixed costs. However, as we have shown,
risk diversification would suggest that a risk-averse firm should
not synchronize orders if individual shipments may be disrupted A.2. Proof of Theorem 1, Section 3.2
at the supplier.
Another extension could relax our assumption of indepen- A.2.1. Optimal base-stock levels
dently distributed disruptions. Eppen [14] shows that demand SsDn ¼ Ssn since each warehouse acts as a single-facility system.
correlation affects the magnitude of the risk pooling effect; Moreover, from (2) we know that Ssn ¼ jd, where j is the smallest
positively correlated demand decreases the risk pooling benefits integer such that Fðj  1Þ Zp=ðp þ hÞ. Similarly, in the centralized
of centralization and negatively correlated demand increases system, the optimal base-stock level is SsCn ¼ jnd for the same j
them. A similar study could be conducted for supply uncertainty. (since the definition of j does not depend on the demand).
We would expect that positively correlated supply disruptions Therefore, SsCn ¼ nSsn .
 A.J. Schmitt et al. / Omega 52 (2015) 201–212 211

A.2.2. Expected costs at optimality  ½hðSdn  μÞ þ σðp þ hÞΦ1 ðXÞ2

In the decentralized system, the total expected cost is simply
¼ h ððSdn Þ2  2Sdn μ þ σ 2 þ μ2 Þ þ 2σ 2 ðp2  h ÞΦ2 ðXÞ
2 2
the sum of the single-facility costs, so Es ½C D  ¼ nEs ½C.
 ½h ððS Þ  2S μ þ μ Þ þ 2σhðp þ hÞðSdn  μÞΦ1 ðXÞ
dn 2 dn
2 2
In the centralized system, the optimal expected cost at the
centralized warehouse, using nSsn in place of SsCn in (1), is given by þ σ 2 ðp þ hÞ2 Φ1 ðXÞ2 
1
¼ h σ þ2σ 2 ðp2  h ÞΦ2 ðXÞ  2σhðp þhÞðSdn  μÞΦ1 ðXÞ
2 2 2
Es ½C nC  ¼ ∑ π i ½hðnSsn  ðiþ 1ÞndÞ þ þ pððiþ 1Þnd  nSsn Þ þ 
i¼1  σ 2 ðp þ hÞ2 Φ1 ðXÞ2
1
sn þ sn þ
¼ n ∑ π i ½hðS  ði þ 1Þ dÞ þ pðði þ 1Þd S Þ  2 2
¼ σ ½h þ 2ðp2  h ÞΦ2 ðXÞ  2hðp þ hÞXΦ1 ðXÞ  ðp þ hÞ2 Φ1 ðXÞ2 
2
i¼1
ð33Þ
¼ nE ½C  ¼ Es ½C nD 
s n
ð22Þ
Let Y equal the sum of the terms inside the brackets. Note that Y is
a function of X, not of σ or μ alone; therefore, Y is the same for an
A.2.3. Cost variances at optimality individual warehouse or for the centralized warehouse in the
For the decentralized system, since each warehouse operates centralized system.
independently, the variance of the cost is equal to the sum of the Since the decentralized system functions as n individual ware-
single-warehouse variances, or V s ½C D  ¼ nV s ½C. houses, we have
Now, for a single warehouse,
1 V d ½C nD  ¼ nV d ½C n : ð34Þ
s n 2 2 sn þ 2 2 sn þ 2
E ½ðC Þ  ¼ ∑ π i ½h ððS  ði þ 1Þ dÞ Þ þ p ðððiþ 1Þ d  S Þ Þ  ð23Þ
i¼1 Applying (33),

In the centralized system, using SsCn ¼ nS , sn

V d ½C nC  ¼ σ 2C Y ¼ nσ 2 Y ¼ nV d ½C n  ¼ V d ½C nD : ð35Þ
1
Es ½ðC nC Þ2  ¼ ∑ π i ½h ððnSsn  ðiþ 1ÞndÞ þ Þ2 þ p2 ððði þ1Þnd  nSsn Þ þ Þ2 
2 as desired. □
i¼1
1
¼ n ∑ π i ½h ððSsn  ðiþ 1ÞdÞ þ Þ2 þ p2 ððði þ 1Þd Ssn Þ þ Þ2 
2 2 A.4. Proof of Proposition 5, Section 5.1
i¼1
2 s
¼ n E ½ðC Þ  n 2
ð24Þ Eb ½C D  is simply equal to nEb ½C. Similarly, Eb ½C C  ¼ nEb ½C, but
pffiffiffi
with the pooled demand standard deviation equal to σ n and the
Thus, using (22) and (24), demand mean equal to nμ.
V s ½C nC  ¼ Es ½ðC nC Þ2   Es ½C nC 2 ¼ n2 Es ½ðC n Þ2   ðnEs ½C n Þ2 ¼ n2 ðEs ½ðC n Þ2  Es ½C n 2 Þ
A.5. Proof of Proposition 6, Section 5.2
¼ n2 V s ½C n  ð25Þ
Thus V s ½C nC  ¼ nV s ½C nD . □ Using (21) from Lemma 7, we get
Z Z !
1 Sb 1
b 2 2 b 2 2 b 2
A.3. Proof of Theorem 4, Section 4.2 E ½C  ¼ ∑ π i  1 h ðS  idÞ f i ðidÞ dd þ p ðid  S Þ f i ðidÞ dd
i¼1 1 Sb
1
 Z 1
2 2
At optimality, the centralized and decentralized base-stock ¼ ∑ πi  1 h ððSb Þ2  2Sb id þ i2 d Þf i ðidÞ dd
levels satisfy i¼1 1
Z 1 
  þðp2  h Þ
2
ðid  Sb Þ2 f i ðidÞ dd
SdDn  μ p Sdn  nμ
¼Φ ¼ C : ð26Þ Sb
σ p þh σC !!
1
2 2 Sb  iμ
¼ ∑ π i  1 h ððSb Þ2  2Sb iμþ iσ 2 þ i2 μ2 Þ þ 2iσ 2 ðp2  h ÞΦ2 pffi
Let i¼1 σ i
 
p ð36Þ
XΦ ð27Þ
pþh
V b ½C ¼ Eb ½C 2   ðEb ½CÞ2
Using (21) from Lemma 7, !!
1
2 2 Sb  iμ
Z Sd Z 1 ¼ ∑ π i  1 h ððSb Þ2 2ðSb Þiμþ iσ 2 þ i2 μ2 Þ þ 2iσ 2 ðp2 h ÞΦ2 pffi
Ed ½C 2  ¼ h
2
ðSd  dÞ2 f ðdÞ dd þ p2 ðd  Sd Þ2 f ðdÞ dd ð28Þ i¼1 σ i
1 Sd
!!!2
Z Z 1 pffi Sb  iμ
2
1
d 2 d 2 2
1
d 2
 ∑ π i  1 hðSb iμÞ þ σ iðp þhÞΦ1 pffi ð37Þ
¼h ððS Þ  2S d þ d Þf ðdÞ dd þ ðp2  h Þ ðd  S Þ f ðdÞ dd i¼1 σ i
1 Sd
ð29Þ In the decentralized system, the total variance is just the sum of
! the individual-warehouse variances since each retailer operates
2 2 Sd  μ independently. In the centralized system, the total demand
¼ h ððSd Þ2 2Sd μþ σ 2 þ μ2 Þ þ 2σ 2 ðp2  h ÞΦ2 ð30Þ
σ seen by the centralized warehouse is distributed as Nðniμ; niσ 2 Þ.
Therefore,
At optimality, we have
1 
Ed ½ðC n Þ2  ¼ h ððSdn Þ2  2Sdn μþ σ 2 þ μ2 Þ þ 2σ 2 ðp2  h ÞΦ2 ðXÞ
2 2
ð31Þ 2
V b ½C C  ¼ ∑ π i  1 h ððSbC Þ2  2SbC niμ þniσ 2 þ n2 i2 μ2 Þ
i¼1
In addition, from (3), !!
2 SbC niμ
d n dn 1 þ 2niσ 2 ðp2  h ÞΦ2 pffiffiffiffiffi
E ½C  ¼ hðS  μÞ þ σðp þ hÞΦ ðXÞ: ð32Þ σ ni
Therefore, !!!2
pffiffiffiffiffi b
1 SC  niμ
1
 ∑ πi  1 hðSbC  niμÞ þ σ niðp þ hÞΦ p ffiffiffiffi
ffi
V d ½C n  ¼ Ed ½ðC n Þ2   Ed ½C n 2 i¼1 σ ni
2 dn 2 dn 2 2 2 2 2 2
¼ h ððS Þ  2S μ þ σ þμ Þ þ 2σ ðp  h ÞΦ ðXÞ ð38Þ
212 A.J. Schmitt et al. / Omega 52 (2015) 201–212

References [21] Schmitt AJ, Snyder LV. In finite-horizon models for inventory controlunder
yield uncertainty and disruptions. Computers and Operations Research
2012;39:850–62.
[1] Alptekinoglu A, Banerjee A, Paul A, Jain N. Inventory pooling to deliver
[22] Schmitt AJ, Snyder LV. Shen ZJM. Inventory systems with stochastic demand
differentiated service. Manufacturing & Service Operations Management
and supply: properties and approximations. European Journal of Operational
2013;15(1):33–44.
Research 2010;206(2):313–28.
[2] Atan Z, Snyder LV. Inventory strategie stomanage supply disruptions. In:
[23] Snyder LV. A tight approximation for acontinuous-review inventory model
Supply chain disruptions. Springer; 2012. p. 115–39.
[3] Axsäter S. Inventory control. 1st ed. Boston, MA: Kluwer Academic Publishers; with supplier disruptions. International Journal of Production Economics,
2000. forthcoming.
[4] Aydin G, Babich V, Beil DR, Yang ZB. Decentralized supply Risk management. [24] Snyder LV, Tomlin B. Inventory management with advanced warning of
In: Handbook of integrated risk management in globals Upply chains. New disruptions. Working paper, Bethlehem, PA: P.C. Rossin College of Engineering
York: John Wiley & Sons; 2012. p. 389–424. and Applied Sciences, Lehigh University; 2008.
[5] Babich V, Burnetas AN, Ritchken PH. Competition and diversification effects in [25] Lawrence VS, Zuo-Jun MS. Supply and demand uncertainty in multi-echelon
supply chains with supplier default risk. Manufactur- ing & Service Operations supply [Link] Paper. Lehigh University; 2006.
Management 2007;9(2):123–46. [26] Snyder LV, Max SZ-J. Fundamentals of supply chain theory. Hoboken, NJ: John
[6] Berk E, Arreola-Risa A. Note on “future supply uncertainty in EOQ models”. Wiley & Sons; 2011.
Naval Research Logistics (NRL) 1994;41(1):129–32. [27] Snyder LV, Atan Z, Peng P, Rong Y, Schmitt AJ, Sinsoysal B. OR/M Smodels for
[7] Berman O, Krass D, Mahdi TM. On the benefits of risk pooling in inventory supply chain disruptions: areview. SSRN eLibrary 2010.
management. Production and Operations Management 2011;20(1):57–71. [28] Tang SY, Kouvelis P. Supplier diversification strategies in the presence of yield
[8] Chen X, Sim M, Simchi-Levi D, Sun P. Risk aversionin inventory management. uncertainty and buyer competition. Manufacturing & Service Operations
Operations Research 2007;55(5):828–42. Management 2011;13(4):439–51.
[9] Chopra S, Reinhardt G, Mohan U. The importance of decoupling recurrent and [29] Tomlin B. On the value of mitigation and contingency strategies for managing
disruption risk sina supply chain. Naval Research Logistics 2007;54(5):544–55. supply chain disruption risks. Management Science 2006;52(5):639–57.
[10] Corbett C J, Rajaram K. Ageneralization of the inventory pooling effect to [30] Tomlin B. Disruption-management strategies for short life-cycleproducts.
nonnormal dependent demand. Manufacturing & Service Operations Manage- Naval Research Logistics 2009;56:318–47.
ment 2006;8(4):351–8. [31] Tomlin B, Wang Y. On the value of mix flexibility and dul sourcing in
[11] Dada M, Petruzzi N, Schwarz L. A newsvendor's procurement problem when unreliable news vendor networks. Manufacturing & Service Operations
suppliers are unreliable. Manufacturing & Service Operations Management Man- agement 2005;7(1):37–57.
2007;9(1):9–32. [32] Van Meighem JA. Risk mitigation in news vendor networks: resource diversi-
[12] Eeckhoudt L, Gollier C, Schlesinger H. The risk-averse (andprudent) newsboy.
fication, flexibility, sharing and hedging. Management Science 2007;53
Management Science 1995;41(5):786–94.
(8):1269–88.
[13] Ellis SC, Shockley J, Henry RM. Making sense of supply disruption risk
[33] Wang Y, Gilland W, Tomlin B. Mitigating supply risk: dual sourcing or process
research: a conceptual frame work grounded in enactment theory. Journal
improvement? Manufacturing & Service Operations Management 2010;12
of Supply Chain Management 2011;47(2):65–96.
[14] Eppen GD. Note: effects of centralization on expected costs in a multi-location (3):489–510.
news boy problem. Management Science 1979;25(5):498–501. [34] Yang H, Schrage L. Conditions that cause risk pooling to increase inventory.
[15] Hopp WJ, Liu Z. Protecting supply chain networks against catastrophic failures. European Journal of Operational Research 2009;192(3):837–51.
Working paper, Department of Industrial Engineering and Management [35] Yang Z, Aydin G, Babich V, Beil DR. Supply disruptions, asymmetric informa-
Science, Northwestern University, Evanston,IL; 2006. tion, and a backup production option. Management Science 2009;55(2):
[16] Parlar M, Berkin D. Future supply un certainty in EOQ models. Naval Research 192–209.
Logistics (NRL) 1991;38(1):107–21. [36] Yang BZ, Aydin G, Babich V, Beil DR. Using a dual- sourcing option in the
[17] Paterson C, Kiesmüller G, Teunter R, Glazebrook K. Inventory models with presence of asymmetric information about supplier reliability: competition vs.
lateral transshipments: areview. European Journal of Opera- tional Research diversification. Manufacturing & Service Opera- tions Management 2012;14
2011;210(2):125–36. (2):202–17.
[18] Qi L. Acontinuous - review inventory model with random disruptions at [37] Yano CA, Lee HL. Lot sizing with random yields: areview. Operations Research
the primary supplier. European Journal of Operational Research 2013;225(1): 1995;43(2):311–34.
59–74. [38] Yu H, Zeng A Z, Zhao L. Single or dual sourcing: decision- making in the
[19] Sawik T. Selection of supply portfolio under disruptionrisks. Omega 2011;39 presence of supply chain disruption risks. Omega 2009;37(4):788–800.
(2):194–208. [39] Zipkin PH. Foundations of inventory management. 1st ed. Boston, MA:
[20] Sawik T. Selection of resilient supply portfolio under disruption risks. Omega McGraw-Hill Higher Education; 2000.
2013;41(2):259–69.