On The Value of Mix Flexibility and Dual Sourcing in Unreliable Newsvendor Networks - Copy in Quantification Folder

MANUFACTURING & SERVICE
OPERATIONS MANAGEMENT informs ®
Vol. 7, No. 1, Winter 2005, pp. 37–57 doi 10.1287/msom.1040.0063

issn 1523-4614 eissn 1526-5498 05 0701 0037 © 2005 INFORMS
On the Value of Mix Flexibility and Dual Sourcing in

Unreliable Newsvendor Networks
Brian Tomlin, Yimin Wang
Kenan-Flagler Business School, University of North Carolina at Chapel Hill, Chapel Hill, North Carolina 27599-3490
{brian_tomlin@unc.edu, yimin_wang@unc.edu}
W e connect the mix-flexibility and dual-sourcing literatures by studying unreliable supply chains that pro-
duce multiple products. We consider a firm that can invest in product-dedicated resources and totally
flexible resources. Product demands are uncertain at the time of resource investment, and the products can dif-
fer in their contribution margins. Resource investments can fail, and the firm may choose to invest in multiple
resources for a given product to mitigate such failures.
In comparing a single-source dedicated strategy with a single-source flexible strategy, we refine the common
intuition that a flexible strategy is strictly preferred to a dedicated strategy when the dedicated resources are
costlier than the flexible resource. We prove that this intuition is correct if the firm is risk neutral or if the
resource investments are perfectly reliable. The intuition can be wrong, however, if both of these conditions
fail to hold, because there is a resource-aggregation disadvantage to the flexible strategy that can dominate
the demand pooling and contribution-margin benefits of the flexible strategy when resource investments are
unreliable and the firm is risk averse.
We investigate the influence that resource attributes, firm attributes, and product-portfolio attributes have on
the attractiveness of various supply-chain structures that differ in their levels of mix flexibility and diversifica-
tion, and we investigate the influence these attributes have on the optimal resource investments within a given
supply-chain structure. Our results indicate that the appropriate levels of diversification and flexibility are very
sensitive to the resource costs and reliabilities, the firm’s downside risk tolerance, the number of products, the
product demand correlations and the spread in product contribution margins.
Key words: reliability; flexibility; dual sourcing; loss aversion; risk
History: Received: June 15, 2004; accepted: December 7, 2004. This paper was with the authors 1 month for
2 revisions.
1. Introduction offers a demand-pooling benefit and a contribution-

It is well established, both in the literature and margin option benefit (Van Mieghem 1998). There is,
in practice, that resource flexibility is advantageous however, a resource-aggregation disadvantage to the
for firms that sell multiple products with uncertain flexible strategy that to the best of our knowledge
demand, and that dual sourcing is advantageous for has been ignored in the mix-flexibility literature. A
firms that face uncertainty in supply. Research to date, single failure in the flexible strategy leaves the firm
however, has studied these supply-chain strategies in with no productive resource, whereas with the dedi-
isolation: The mix-flexibility literature has assumed cated strategy all resources must fail for the firm to
perfectly reliable supply, and the dual-sourcing liter- be in similar straits. Supply uncertainty, by which we
ature has focused on single-product problems. mean that the realized resource investment may dif-
Consider a firm selling multiple products that differ from the target investment, should therefore influ-
fer in their contribution margins (sales price less vari- ence the firm’s preference for either a dedicated or
able costs) and have uncertain demands. The firm flexible strategy. The firm, of course, is not limited to
might invest in product-dedicated resources only, or, these two strategies, and it may want to consider dual
alternatively, it might invest in one single flexible sourcing for one or more products to protect itself
resource that can produce all products. The dedicated against supply uncertainties. The goal of this paper
resources might be cheaper but the flexible strategy is to simultaneously study mix flexibility and dual
37
Tomlin and Wang: On the Value of Mix Flexibility and Dual Sourcing in Unreliable Newsvendor Networks
38 Manufacturing & Service Operations Management 7(1), pp. 37–57, © 2005 INFORMS
Figure 1 Four Network Structures
Single-source dedicated Single-source flexible Dual-source dedicated Dual-source flexible

(SD) (SF) (DD) (DF)
Resource Product
1 1 1
1
2 2 2
N N N
sourcing to provide insight into effective supply-chain differ. In the presence of supply uncertainty, there
design in the presence of supply and demand uncer- is a resource-aggregation disadvantage to a flexible
tainties. resource. In general, the desirability of any of the four
Figure 1 illustrates four canonical network struc- networks will be influenced by resource investment
tures for a firm that sells N products. Circles represent costs, resource reliabilities, product contribution mar-
products, squares represent resources, and arcs repre- gins, demand correlations, and the firm’s attitudes
sent the ability of a resource to fulfill demand for a toward risk.
product. SD represents a single-source dedicated net- Mix flexibility, whereby a resource has the ability
work, SF represents a single-source flexible network, to produce multiple products, has been investigated
DD represents a dual-source dedicated network, and in the operations literature as a design strategy
DF represents a dual source flexible network. We note for firms that sell multiple products with uncer-
that these structures represent possible sourcing deci- tain demand. Hereafter, we will simply use the term
sions rather than actual decisions, and so a firm with a flexibility rather than mix flexibility. Such literature
DD or a DF network may choose to single source one has primarily focused on single-period (newsvendor
or more products, even though it could dual source type) investments in dedicated and totally flexible
them. resources, and that is the focus of this paper. We refer
The dual-sourcing networks (DD and DF) offer the reader to Jordan and Graves (1995), Graves and
diversification benefits that are advantageous in the Tomlin (2003), and Muriel et al. (2004) for treatments
presence of unreliable resource investments. The flex- of partial flexibility. For single-period investments in
ible strategies (SF and DF) offer demand-pooling and dedicated and totally flexible resources, Van Mieghem
contribution-margin benefits that are advantageous in (2004a) establishes that component commonality and
the presence of demand uncertainty. The demand- resource flexibility are distinctions without a differ-
pooling benefit arises only if demands are not per- ence; the problems can be shown to be mathemati-
fectly positively correlated. The contribution-margin cally equivalent. As such, we use the term resource
benefit arises only if product contribution margins with the understanding that the resource in question
Manufacturing & Service Operations Management 7(1), pp. 37–57, © 2005 INFORMS 39
might be inventory or capacity. The resource may be strategies have been investigated in the context of
produced in house or may be provided by an outside random yield (Gerchak and Parlar 1990, Parlar and
supplier. Wang 1993, Anupindi and Akella 1993, Agrawal and
The archetypal total-flexibility model (e.g., Fine and Nahmias 1997, Swaminathan and Shanthikumar 1999,
Freund 1990) is a single-period, N product, uncertain- Dada et al. 2003, Tomlin 2004b), random disruptions
demand model in which a risk-neutral firm can invest (Parlar and Perry 1996, Gürler and Parlar 1997, Tom-
in N dedicated resources and one totally flexible lin 2004a), and credit risk (Babich et al. 2004), but all
resource. Investment costs are linear and there are no these papers assume a single product, so mix flexibil-
fixed costs. Gupta et al. (1992), Li and Tirupati (1994, ity is not relevant. We note that Tomlin (2004a) inves-
1995, 1997), and Van Mieghem (1998, 2004b) all inves- tigates the value of volume flexibility in unreliable
tigate variations on this theme. The paper of most supply chains.
direct relevance is Van Mieghem (1998). In that model, The rest of the paper is organized as follows.
the firm sells two products that differ in contribu- Section 2 introduces the general supply chain model.
tion margins. This difference in contribution margins In §3 we consider the SD and SF networks. In §4 we
makes flexibility valuable even if demands for both consider the DF and DD networks, and compare them
products are perfectly positively correlated, a result to the single-source networks. Conclusions and direc-
that contradicted the prevailing intuition. Our work tions for future research are presented in §5. Proofs of
relaxes two implicit assumptions (reliable investments all results can be found in Appendix A.
and risk neutrality) of these models. There is a bur-
geoning literature on non-risk-neutral decision mak- 2. The Model
ers in the single-product newsvendor context, e.g., We present a general model and identify each of the
Eeckhoudt et al. (1995), Agrawal and Seshadri (2000), supply networks (SD, SF, DD, and DF) as instances of
Schweitzer and Cachon (2000), Caldentey and Haugh the general model. There are N products n = 1 N .
(2004), and Chen et al. (2003). As far as we are aware, The marginal contribution margin for product n is pn .
the only mix-flexibility paper (other than this one) to We use the notational convention that p1 ≥ p2 ≥ · · · ≥
relax the risk-neutrality assumption is Van Mieghem pN . Let p = p1 pN . All vectors are assumed to
(2004b). That paper and our paper can be seen as be column vectors, and denotes the transpose oper-
complementary, in that the research questions being ator. The firm can invest in nonnegative levels (Kj )
addressed differ, as do the treatments of non-risk- of J different resources labeled j = 1 J . Let T be
neutral decision makers. Van Mieghem (2004b) inves- the N × J technology matrix with tnj = 1 indicat-
tigates how risk aversion influences the flexibility ing that resource j can produce product n. Demand
investment levels in perfectly reliable newsvendor
X = X 1 X
N is uncertain, with a joint density
networks by using concave-increasing utility func- fX x1 xN at the time of the investment decision.
tions (to investigate the directional influence of risk The demand-correlation matrix is denoted X with
aversion) and a mean-variance approach (to investi- element mn being the correlation coefficient for prod-
gate the magnitude of the influence). In contrast, we ucts m and n. The marginal density for X n is fX xn
n
investigate flexibility and dual sourcing in unreliable and the cumulative distribution function is FXn xn .
newsvendor networks and, in doing so, allow for non- Realizations of demand are denoted x = x1 xN .
risk-neutral firms by considering both loss aversion Resource investments are unreliable and the real-
(Kahneman and Tversky 1979) and the Conditional ized level K r for resource j is stochastically propor-
j
Value-at-Risk (CVaR) measure (Rockafellar and Urya- tional to the invested level Kj , i.e., K r = Y j Kj . In
j
sev 2000, 2002). particular, we assume a Bernoulli yield model in
Unreliable, single-product, single-resource prob- which Y j = 1 with probability j and Y j = 0 with prob-
lems have been widely studied in the yield and ability 1 − j . We refer to j as the reliability of resource
disruption literatures. In contrast, unreliable supply j. There is often a Bernoulli nature to the supply pro-
chains (multiple resources or multiple products, or cess (e.g., Anupindi and Akella 1993, Parlar et al. 1995,
both) have received less attention. Dual-sourcing Swaminathan and Shanthikumar 1999, Dada et al.
2003, Tomlin 2004b). In the case of inventory, Bernoulli a nonnegative investment vector K =K1 KJ to
supply processes arise due to batch failures, accep- maximize some objective function V K, where V K
tance sampling, supply-chain disruptions, or sup- depends on the firm’s terminal wealth. We consider
plier delays (if the delivery occurs too late to serve three different types of firms: a risk-neutral firm,
demand). Unless otherwise stated, we assume that the a loss-averse firm, and a firm concerned about down-
j are independent. Let
Y Y = Y 1 Y
J and let a real- side risk, each represented by a distinct form of the
ization be denoted by y = y1 yJ . objective function.
Investment costs for an unreliable resource can The risk-neutral objective function is given by
depend on both the investment (or ordered) level and K, so the risk-neutral invest-
VRN K = w0 + EX Y W
the realized (or delivered) level. For each resource j, ment problem is
we assume that the firm pays j cj per unit ordered K
VRN K∗ = w0 + max EX Y W (5)
and an additional 1 − j cj per unit delivered. We K≥0
refer to cj as the marginal total cost and j as the A loss-averse decision maker (Kahneman and Tversky
marginal committed cost. Equivalently, we can think 1979) attributes more significance to losses than to
of the firm paying cj per unit ordered but receiving a gains. Schweitzer and Cachon (2000) study loss aver-
rebate of 1 − j cj for each unit not delivered; in that sion in a classic single-product newsvendor setting
context 1−j is the failure rebate. Let c = c1 cJ , by using a piecewise-linear model that is a spe-
= 1 J , cj yj = cj j + 1 − j yj , and cy = cial case of Tversky and Kahneman’s (1992) two-part
c1 y1 cJ yJ . power-function model. We assume the same piece-
The firm’s investment problem can be formulated + K −
wise linear model here; VLA K = w0 + EX Y W
as a two-stage stochastic program. In the second
W K where W K = max!W K 0", W − K =
− +
stage, after demands and investments have been real- K 0" and ≥ 1. Increasing loss aversion
max!−W
ized, the firm allocates production to maximize the
is associated with an increasing . We note that
contribution.
VLA K = VRN K at = 1. The loss-averse investment
rK x y = max p s (1) problem is
sn qnj ≥0
+ K − W
VLA K∗ = w0 + max EX Y W − K (6)
s.t. sn ≤ xn n = 1 N (2) K≥0
J
For the terminal wealth distribution associated with
sn ≤ tnj qnj n = 1 N (3) a resource-investment vector K, the CVaR, denoted
j=1
VCVaR# K, is the mean of the left #-tail of the wealth

N
distribution. The percentile # ∈ 0 1 is a param-
qnj ≤ yj Kj j = 1 J (4)
n=1
eter that reflects the firm’s taste for downside risk.
At # = 1, the firm is risk neutral, and the com-
where qnj denotes the production of product n by
plete wealth distribution is considered in the objec-
resource j, s = s1 sN denotes the sales of prod-
tive. For # < 1, the firm maximizes the mean of
ucts 1 N , constraint (2) ensures that sales do not
the wealth distribution falling below a specified per-
exceed realized demand, constraint (3) ensures that
centile level #. Increasing concern with downside risk
sales do not exceed production, and constraint (4)
is associated with a decreasing percentile #. In recent
ensures that resource usage does not exceed the real-
years, the CVaR measure has gained popularity as a
ized level.
risk measure in the finance literature, e.g., Rockafellar
Let w0 be the firm’s initial wealth and let W K be
and Uryasev (2000, 2002), Acerbi (2002), Acerbi and
the random gain or loss (denoted by positive or nega-
Tasche (2002), and Szegö (2002). Using Theorem 10 of
tive numbers, respectively) achieved by investment K.
Rockafellar and Uryasev (2002),
The firm’s realized profit on investment K is given by
wK = −cy K +rK x y. The firm’s random termi- VCVaR# K
K. In the first stage, before
nal wealth is then w0 + W 1 K−v0"
= w0 +max v + EXY min!W (7)
demands and yields are realized, the firm chooses v #
i.e., for a given investment vector K, VCVaR# K can the other in the sense that it is preferred for all pos-
be found by solving a maximization problem. We can sible parameters. To gain insight into what drives a
therefore write the CVaR investment problem as firm’s network preference, we restrict attention in this
section to the special case where the following three
VCVaR# K∗
assumptions all hold.
1
= w0 + max v + EXY min!W K−v0" (8) 1. The resource reliabilities are identical for all
K≥0v # resources, i.e., 1 = 2 = · · · = N = N +1 = , but the
VCVaR# K is jointly concave over K ≥ 0 and v; yield random variables are still independent across
see Rockafellar and Uryasev (2000, 2002). It is this resources.
joint concavity property, in addition to the fact that 2. The marginal committed costs are identical for
VCVaR# K is a coherent measure of risk as defined in all resources, i.e., 1 = 2 = · · · = N = N +1 = .
Artzner et al. (1999), that has given rise to its popu- 3. The marginal total costs are identical for the ded-
larity. The optimal investment is independent of the icated resources, i.e., c1 = c2 = · · · = cN = c.
initial wealth for all three objective functions, and we We now introduce two definitions, the second of
therefore assume that w0 = 0 without loss of generality. which will be a key metric in much of the analysis.
In closing, we note that each of the four supply Definition 1. The indifference cost cNI +1 is the value
networks (SD, SF, DD, and DF) can be obtained as of the flexible resource’s marginal total cost at which
a special case of the general model presented above. the firm is indifferent between the SD and SF net-
For example, if N = 2, then works; that is, cNI +1 is the value of cN +1 such that
V SF ∗ = V SD ∗ .
1 0 1
SD & T = ( SF & T = ( Definition 2. The flexibility premium + is the rela-
0 1 1 tive difference in the marginal total costs at which the

1 1 0 0 1 0 1 firm is indifferent between the SD and SF networks,
DD & T = ( DF & T = i.e., + = cNI +1 − c/c.
0 0 1 1 0 1 1
The flexibility premium + is a useful measure of the
3. Single-Source Networks: value of flexibility; the firm prefers the SF network as
long as cN +1 ≤ 1 + +c. Put another way, + > 0 implies
The Flexibility Premium
that the firm is willing to pay a higher price (rela-
In this section, we focus on the single-source networks
tive to the dedicated resources) for flexibility, whereas
SD and SF. There are N resources in the SD network,
+ < 0 implies that the firm requires a lower price for
and we label these n = 1 N with resource n dedi-
flexibility to be preferred.
cated to product n. There is a single (flexible) resource
We note that although we assume that resource
in SF, and we label this N + 1. We focus on the coun-
reliabilities, marginal committed costs, and dedicated
terbalancing effects of demand pooling and resource
marginal total costs are identical in this section, many
aggregation by assuming p1 = p2 = · · · = pN to elim-
of the following equations ((9)–(24)) can be extended
inate the contribution-option benefit. By eliminating
to the nonidentical case in a straightforward manner.
this contribution-option benefit, we can model the SF
investment problem as a single-product problem with 3.1. Risk-Neutral Firm
demand X N +1 = X 1 + X
2 + · · · X
N . We denote the den-
This SF investment problem is an extension of the
sity and cumulative distribution of total demand by classic single-product newsvendor model to allow for
fXN +1 xN +1 and FXN +1 xN +1 , respectively. Bernoulli investment failures and the specified invest-
Let V SD ∗ and V SF ∗ denote the optimal objective ment costs. The objective function is
values for the SD and SF networks. SF is strictly
SF
preferred if V SF ∗ > V SD ∗ , and is weakly preferred VRN KN +1 = − + 1 − cN +1 KN +1
K
if V SF ∗ ≥ V SD ∗ . Hereafter, the term preferred should N +1
+ p xN +1 fXN +1 xN +1 dxN +1
be understood to mean weakly preferred. Clearly the 0

firm’s preference will depend on the resource costs
and the reliabilities, so neither network will dominate + KN +1 1 − FXN +1 KN +1 (9)
It is relatively straightforward to show that Figure 3 Flexibility Premium for Risk-Neutral Objective with p/c = 5
18.0%
+ 1 − cN +1
KN∗ +1 RN = FX−1 1 − (10)
N +1
p
λ = 1.0
and
Flexibility premium, ∆
17.5%
∗ λ = 0.8
KN +1
SF ∗
VRN = p xN +1 fXN +1 xN +1 dxN +1 (11) λ = 0.6
0
17.0%
λ = 0.4
The SD investment problem can be modeled as N
independent single-product problems, so λ = 0.2
16.5%
∗ −1 + 1 − c λ=0
Kn RN = F Xn 1 −
p
n = 1 N (12) 16.0%
0.75 0.80 0.85 0.90 0.95 1
and Reliability, θ
N
Kn∗
SD ∗
VRN = p xn fXn xn dxn (13)
n=1 0
both figures; that is, SF is always preferred if cN +1 ≤ c.
For the case of N = 2 (i.e., two products), the In fact, as the following proposition shows, this obser-
vation is true in general.
risk-neutral flexibility premium +RN is plotted as a
function of both and in Figures 2 and 3 for inde- Proposition 1. For any demand random vector X
pendent uniformly distributed U 0 1 demands. (i) the risk-neutral flexibility premium +RN is nonnegative
We see that +RN can be increasing or decreasing in for all 0 ≤ ≤ 1, (ii) 0 ≤ ≤ c/p −1 − c ⇒ +RN = 0,
both and , depending on the magnitude of p/c. and (iii) +RN = 0 if X = 1, i.e., pairwise perfect positive
Observe that +RN is constant for = 0 in both cases. correlation for all products.
This observation is true in general, because the opti-
This proposition tells us that the SF network is pre-
mal resource investment is independent of the relia-
∗ −1
ferred to the SD network for all reliabilities and for
bility (Kn RN = FXn 1 − c/p) if investment failures are all demand distributions in the case of a risk-neutral
fully rebated, i.e., = 0. Observe that the flexibility
firm facing equal resource reliabilities and costs. This
premium is nonnegative for all combinations in
result was not obvious a priori (at least to the authors),
because the demand-pooling benefit of SF might be
Figure 2 Flexibility Premium for Risk-Neutral Objective with p/c = 2 outweighed by the resource-aggregation disadvan-
18% tage. In fact, even when there is no demand-pooling
benefit (i.e., X = 1), the firm is still indifferent between
λ=0 SF and SD. Recall that the contribution-margin ben-
17%
λ = 0.2 efit does not exist because p1 = p2 = · · · = pN . The
λ = 0.4
only explanation is that resource aggregation does
16% not exclusively penalize SF. Resource aggregation is
λ = 0.6 a disadvantage for SF from a downside perspective
15%
because the probability of multiple resources failing in
λ = 0.8
SD is less than that of the single resource failing in SF,
but an advantage from an upside perspective because
λ = 1.0
14% the probability of the single resource succeeding in
SF is higher than the probability of multiple resources
13%
succeeding in SD. For a risk-neutral firm, the upside
0.75 0.80 0.85 0.90 0.95 1 resource-disaggregation advantage, coupled with the
Reliability, θ demand-pooling benefit of the SF network, dominates
the downside resource-disaggregation disadvantage, The optimal flexible investment K3∗ is therefore given
resulting in SF being preferred to SD. Allowing for by
asymmetric contribution margins only increases the
SF ∗
c3 c3 K3
advantage of the SF network. SF ∗
FX3 K3 LA + − 1 F
LA
Risk neutrality implies a utility function that has a p X3 p

constant marginal return to wealth. Resource aggrega- + 1 − c3
=1− (18)
tion influences the distribution of the terminal wealth, p
so it is natural to ask whether alternate wealth pref-
and the resulting objective value is
erences would change the result that SF is preferred

K SF ∗
to SD. We now address this question by considering SF ∗ 3 LA
the loss-averse and CVaR objective functions. VLA = p x3 fX3 x3 dx3
0
SF ∗
c3 /pK3

LA
3.2. Loss-Averse Firm + − 1 x3 fX3 x3 dx3 (19)
0
We assume that N = 2 in this section, but the results
for the SF network extend in an obvious fashion to We note that Equations (14) to (19) extend directly to
the case where there are more than two products. the case where N > 2, with N + 1 replacing 3. We also
Recall that, using our convention, the total demand note that Equation (18) collapses to the risk-neutral
3 and the flexible resource is labeled K3
is labeled X optimal investment (10) when = 1.
SF ∗ SF ∗
when N = 2 (because N + 1 = 3). Closed-form solutions for K3 LA and VLA will not
The SF investment problem is an extension of exist in general. We have, however, been able to
the loss-averse single-product newsvendor model to 1 and X
obtain closed-form solutions for the case of X 2
allow for Bernoulli investment failures and the speci- having independent uniform distributions.
fied investment costs. The objective function is Proposition 2. Let X 1 and X
2 have independent
SF 3 K3 U 0 1 distributions. If
VLA K3 = EX3 Y3 W
3 K3 W
+ − 1EX3 Y3 W 3 K3 < 0 (14) p/2 − − 1c3 /p3 ≤ + 1 − c3
3 K3 is the random profit realized by an then

where W
investment of K3 . Therefore, SF ∗ 21 − + 1 − c3 /3 p
K3 LA =
SF 1 + − 1c3 /p3
VLA K3 =− + 1 − c3 K3

K
3 Otherwise,
+ p x3 fX3 x3 dx3 + K3 1 − FX3 K3

3
0

SF ∗ 1 c3
K3 LA = −1 + 1 − − 1 − 1
+ −1 −1 − c3 K3 + −c3 K3 FX3 c3 K3 /p 2 p

1/2
+ 1 − c3
c3 K3 /p · −2 +
+p x3 fX3 x3 dx3 (15) p
0

3 −1
1 c
and the first and second derivatives are · − 1 3 − 1
2 p
SF
dVLA K3
= p1 − FX3 K3 − − 1c3 FX3 c3 K3 /p We now turn to the SD network. Let wn xn yn Kn
dK3
be the realized profit generated by product n from
− + 1 − c3 (16)
an investment of Kn in dedicated resource n, and let
SF

n Kn be the random profit. Then
W
d 2 VLA K3
= −pfX3 K3 − − 1c3
dK32

wn xn yn Kn = −+1−yn cKn
cK c3
· fX3 3 3 ≤ 0 (17) +pmin!xn yn Kn " (20)
p p
The random total profit for the SD network is Figure 4 Flexibility Premium for Loss-Averse Objective with p/c = 2
W SD K1 K2 = W
1 K1 + W
2 K2 . The probability of a and = 02
loss depends on the realization of X 1 , X
2 , Y
1 , and Y
2 . 20%
Because losses and gains are weighted differently, the
loss-averse SD network cannot be decomposed into 15%
two single-product loss-averse problems as was pos- θ = 1.00

10%
sible in the risk-neutral case. The loss-averse objective
function is 5%
θ = 0.95
SD
VLA SD K1 K2 + − 1
K1 K2 = EX Y W θ = 0.90
0%
SD K1 K2 W
· EX Y W SD K1 K2 < 0 θ = 0.85
Flexibility premium negative,
(21) –5% i.e., flexible resource needs
θ = 0.80
to be cheaper to break even
where –10% θ = 0.75

2
SD K1 K2 =
EXY W −+1−cKn +pLXn Kn –15%
1 2 3 4 5 6 7 8 9 10
n=1 Loss-aversion coefficient, β
+Kn 1−FXn Kn (22)
z
LXn z = xn fXn xn dxn (23) and
0 
3

 1 p 1 p
SD K1 K2 W
EXY W SD K1 K2 < 0 
 2A = + 1−1+ 2

 2 c −1 c




2
= 1−2 G00 K1 K2 +1−G10 K1 K2 1 p
p ≥ 2c ⇒ 2B =
+1−G01 K1 K2 + 2 G11 K1 K2 (24) 
 2 −1 c






 2 p
We note that the Gy1 y2 K1 K2 expressions can be 
 · +1−+ −11− −
found in Appendix D and that yn denotes the real- c
ization (success or failure) of resource n in the Figures 4 and 5 illustrate the dependence of the
Gy1 y2 K1 K2 expressions. Closed-form solutions loss-averse flexibility premium +LA on the reliability
SD ∗ SD ∗
for Kn LA and VLA do not exist in general. and the loss-aversion coefficient .
Proposition 3. Let X 1 and X2 have independent Observe that +LA can be negative (especially as the
U 0 1 distributions. Then reliability decreases or loss aversion increases). This
means that the SD network can be strictly preferred
SD ∗ SD ∗
K1 LA = K 2 LA = −2A + 2A2 − 2B to the SF network even when the flexible resource
costs the same or less than the dedicated resources; a
where


2 result that cannot occur in the risk-neutral case (see

 c c Proposition 1). In the case of loss aversion, the down-

2 = −1 4 −4 −1 2


A
p p side resource-aggregation disadvantage of the flexi-




2 ble strategy is amplified by the higher weight placed

 1 c

 2
− 1− 1+ 2
− on losses, with the result that SD can outperform SF.



 2 p

 Comparing Figures 4 and 5, we see that moving from


3
2 3 c c −1 = 02 to = 08 makes SD preferable over a larger
p < 2c ⇒ · −1 +4 −5

 2 p p set of pairings, because the consequences of



 failure are larger when the marginal committed cost

 c

 2B = − +1− + −11− 2
increases. We note that +LA is not necessarily decreas-

 p

 ing everywhere in ; we observed +LA to be increas-



3



 3 c c −1 ing and then decreasing for instances with higher p/c
 · −1 2 +4 −5 values.
2 p p
Figure 5 Flexibility Premium for Loss-Averse Objective with p/c = 2 (that it be nondecreasing in wealth). Clearly, U1 con-
and = 08 tains all concave increasing utility functions; the com-
20% mon model for risk aversion. In fact U1 contains all
(locally and globally) risk-averse or risk-seeking utility
10% θ = 1.00
functions as long as u w ≥ 0 everywhere.
It is commonly accepted intuition that a flexible
0%
strategy is preferable to a dedicated strategy if the
θ = 0.95
–10% investment costs are equal. This intuition manifests
itself in the literature in the assumption that the unit
– 20% θ = 0.90 cost for the flexible resource is higher than for dedi-
Flexibility premium negative, cated resources. Using Propositions 1 and 4, we can
i.e., flexible resource needs θ = 0.85
– 30%
to be cheaper to break even
establish when the common intuition is in fact valid.
θ = 0.80 Remark 1. For the case of identical resources (mar-
– 40%
ginal total costs, marginal committed costs, and reli-
θ = 0.75
– 50% abilities), the SF network is preferred to the SD net-
1 2 3 4 5 6 7 8 9 10
work if either the resource investments are perfectly
Loss-aversion coefficient, β
reliable (i.e., = 1) or the firm is risk neutral. If neither
condition holds, then the SD network can be strictly
The insight that the flexible resource may have to preferred to the SF network.
be cheaper than the dedicated resources for the firm Thus, the common intuition is valid if either the risk
to prefer SF to SD does not hinge on the choice of neutrality or the perfect reliability assumption holds,
a loss-averse objective function. Qualitatively similar but can be incorrect if neither condition holds.
results can be shown to hold under the CVaR objec-
tive; that is the flexibility premium can be negative.
4. Dual-Source Networks: DD and DF
Specific propositions and results for the CVaR objec-
We now consider the DD and DF networks. We
tive can be found in Appendix B. choose to call these dual-source networks because the
firm can invest in dual resources for any given prod-
3.3. Flexibility Premium in
uct. We do not, however, force the firm to invest in
Perfectly Reliable Supply Chains
all available resources, so a single-source strategy for
The reader may have noticed that the loss-averse flex-
one or more products may be optimal.
ibility premium +LA is nonnegative everywhere for
= 1 in Figures 4 and 5. A similar phenomenon is 4.1. Two-Product DF Network
also observed for the CVaR objective. Such numeri- We label the resources as follows: resource 1 is ded-
cal results suggest that the flexibility premium + is icated to product 1, resource 2 is dedicated to prod-
always nonnegative for = 1. This suggestion is con- uct 2, and resource 3 is totally flexible. This DF
firmed by the following proposition. Define U1 as network is an extension of Van Mieghem (1998), here-
the set of utility functions that are nondecreasing in after referred to as VM98, in that we allow for unre-
wealth, i.e., more is (weakly) better. liable resource investments (with costs that are lin-
ear in both the target investment and the realized
Proposition 4. Let the firm have an initial wealth of
investment) and non-risk-neutral objective functions.
w0 . Let
X have any joint distribution. Let 1 = 2 = · · · =
The model collapses to that of VM98 if all reliabilities
N = N +1 = 1. Then (i) + ≥ 0 for all utility functions
equal one and the objective value is VRN K. VM98
u1 ∈ U1 , (ii) in particular + ≥ 0 for the loss-averse objective
restricts attention to c3 > max!c1 c2 ", because other-
function and the CVaR objective function.
wise the flexible resource clearly dominates at least
Proposition 4 is a quite general result because it one of the dedicated resources. We consider all c3 ≥ 0,
makes no assumptions on the demand distribution because in our more general model investing in the
and only very mild assumptions on the utility function dedicated resources may be optimal even if c3 ≤
min!c1 c2 ". There are eight possible structures to the prices, marginal total costs, marginal committed-costs,
DF ∗
optimal solution (corresponding to different combina- reliabilities, and failure rebates on VRN and KjDF ∗
RN .
tions of positive resource investments), and each of We first extend Propositions 3 and 4 of VM98 to
the eight structures can be optimal depending on the characterize the directional influence of marginal total
DF ∗
model parameters. costs and prices on VRN and KjDF ∗
RN when resource in-
Let us consider the risk-neutral problem. For any vestments are unreliable.
feasible investment vector K =K1 K2 K3 , the partial DF ∗
derivatives with respect to Kj are given by Proposition 5. For a risk-neutral firm, (i) VRN is a
nonincreasing convex function of the marginal total-cost
DF
6VRN K vector c and the marginal committed-cost vector , and a
= −c1 1 +1−1 1
6K1 nondecreasing convex function of the contribution-margin
+1 p1 1−3 P 81 +3 P 82 vector p( (ii) if the marginal demand density fXn x, n =
1 2, is either log-concave or log-convex, then the direc-
+3 p2 1−2 P 83 +2 P 84 DF ∗
tional sensitivity of the optimal investment vector KRN
DF
6VRN K with respect to p, c and is given by the following matri-
= −c2 2 +1−2 2
6K2 DF ∗
ces (e.g., 6K1
RN /6c3 ≥ 0)
+2 p2 1−3 P 85

+3 1−1 P 86 +P 87 ≥0 ≥0 ≥0
DF ∗
:p KRN =
≥0 ≥0 ≥0
+1 P 84 +P 88 +P 89
DF
6VRN K  
≤0 ≤0 ≥0
= −c3 3 +1−3 3
6K3  
+3 p1 1−1 P 810 +1 P 82
DF ∗
:c KRN =
 ≤0 ≤0 ≥0 

≥0 ≥0 ≤0
+3 p2 1−1 1−2 P 811 +1 1−2
 
·P 83 +P 812 +1−1 2 P 86 ≤0 ≤0 ≥0
 
+1 2 P 84 +P 89 (25) DF ∗
: KRN =
 ≤0 ≤0 ≥0 

where each of the 8k , k = 1 12, is a demand- ≥0 ≥0 ≤0
space region, and the union of the 8k cover the
demand space. We note that because of the Bernoulli Actual expressions for the sensitivities are available
failures, the problem does not lend itself to a strict on request, but are very involved and not particu-
partitioning (disjoint 8k ) of the demand space as in larly insightful. Proposition 5 proves that the perfect-
Figure 1 of VM98. Expressions for the P 8k can be reliability results of VM98 for the c and p vectors
found in Appendix C. VRN DF
K can be shown to be still hold for an unreliable network. As discussed in
jointly concave in K1 K2 K3 . Let VRN DF ∗
denote the VM98, the presence of the flexible resource means that
DF∗
optimal objective value and Kj RN the optimal invest- the optimal investment level for the resource dedi-
ment level for resource j = 1 2 3. An interior solution cated to product n is influenced by the resource ded-
K1DF ∗ DF ∗ DF ∗
RN > 0 K2 RN > 0 K3 RN > 0 is optimal if and only
icated to product 3 − n, n = 1 2. We also see that
if the three derivatives in (25) all equal zero for some the marginal total costs and committed costs have the
K1 > 0 K2 > 0 K3 > 0. If such a solution exists, it same directional influence. VM98 does not address
is unique. If such a solution does not exist, then one marginal committed costs because such costs are irrel-
or more of the Kj∗ RN must equal zero and so at least evant in a perfectly reliable network. We note that the
one of the products is single sourced. Closed-form assumption of log-concavity or log-convexity for the
solutions for K1 ∗ ∗ ∗
RN K2 RN K3 RN do not exist for the marginal demand densities is a mild one that is met
VM98 model and so will not exist for our model. We by the Uniform, Normal, Weibull, Gamma, Pareto,
can, however, characterize the directional influence of and Logistic distributions, as well as by many others.
DF ∗
Proposition 6. For a risk-neutral firm, (i) VRN is a tive value and KjDD∗ RN the optimal investment level for
nondecreasing convex function of the reliability vector ; resource j = 1 4. AA93 proved that there was no
(ii) if the marginal demand densities fXn x are uniform, closed-form solution for the optimal investment levels
then the directional sensitivity of the optimal investment in their model, and so there will not be a closed-form
DF ∗
vector KRN with respect to is given by the following solution to our more general model. We can, how-
matrix:   ever, still say something about the directional influ-
≥0 ≥0 ≤0
ence of the costs, reliabilities, and prices on these opti-
 
DF ∗
: KRN =
 ≥0 ≥0 ≤0 
 mal values.
DD ∗
≤0 ≤0 ≥0 Proposition 7. For a risk-neutral firm, (i) VRN is a
As one would expect, the firm benefits from nonincreasing convex function of the marginal total-cost
increasing resource reliabilities. We again see a cross- vector c and the marginal committed-cost vector , and a
dependence for the dedicated resources because of the nondecreasing convex function of the contribution-margin
presence of the flexible resource. An increase in the vector p and the reliability vector ; (ii) the directional
DD ∗
reliability of a dedicated resource j = 1 2 increases sensitivity of the optimal investment vector KRN with
the investment in that resource j. This decreases the respect to p, c, , and is given by the following matrices
investment in the flexible resource, and as a conse-
quence increases the investment in the other dedi- ≥0 ≥0 =0 =0
DD ∗
:p KRN =
cated resource 3 − j. ≥0 ≥0 =0 =0
 
4.2. Two-Product DD Network ≤0 ≥0 =0 =0
We now consider the DD network in which there  
 ≥0 ≤0 =0 =0 
are four dedicated resources (j = 1 2 dedicated to  
DD ∗
:c KRN =



product 1 and j = 3 4 dedicated to product 2). There  =0 =0 ≤0 ≥0 
are 16 possible structures to the optimal solution  
(corresponding to different combinations of positive =0 =0 ≥0 ≤0
resource investments); each of the structures can be  
≤0 ≥0 =0 =0
optimal depending on the model parameters.  
 ≥0 ≤0 =0 =0 
This model is an extension of the single-period ver-  
sion of Model 1 in Anupindi and Akella (1993), here-
DD ∗
: KRN =



2 = 0 (i.e., single  =0 =0 ≤0 ≥0 
after referred to as AA93. Setting X  
product), 1 = 2 = 0 and V K = VRN K recovers =0 =0 ≥0 ≤0
the AA93 model. For any feasible investment vec-  
≥0 ≤0 =0 =0
tor K =K1 K2 K3 K4 , the partial derivatives (for the  
risk-neutral objective function) with respect to Kj are  ≤0 ≥0 =0 =0 
 
given by
DF ∗
: KRN =



 =0 =0 ≥0 ≤0 
DD
6VRN K  
= −cj j + 1 − j j
6Kj =0 =0 ≤0 ≥0
+ j 3−j 1 − FX1 K1 + K2
For the DF network, the investment level for dedi-
+ 1−3−j FX1 Kj j = 1 2 (26) cated resource for product n = 1 2 was influenced by
DD
6VRN K the costs and reliabilities of the dedicated resource for
= −cj j + 1 − j j
6Kj product 3 − n, because of the coupling effect of the
+ j 7−j 1 − FX2 K3 + K4 flexible resource. This proposition tells us that in the
DD network, investment levels for a resource dedi-
+ 1−7−j FX2 Kj j = 3 4 (27) cated to product n = 1 2 are influenced by the other
DD
VRN K can be shown to be jointly concave in dedicated resource for product n, but not by the ded-
DD ∗
K1 K2 K3 K4 . Let VRN denote the optimal objec- icated resources for product 3 − n.
4.3. Numerical Studies was fixed at p2 = 10. The relative contribution mar-
Sections 4.1 and 4.2 established the directional influ- gin (p1 /p2 ) was varied from 10 to 12 in increments
ence of costs, reliabilities, and prices on the optimal of 005, giving us a total of five contribution-margin
profits and absolute investment levels within a given ratios. In §§4.1 and 4.2, we analytically characterized
network structure, but they did not speak to either the the influence of a change in the reliability of a sin-
influence of model parameters on the relative invest- gle resource, so we chose to investigate the influence
ment levels within a network structure or the relative of changes in the overall network reliability in the
attractiveness of the different network structures. numerical studies. To that end, we assume that j =
We address such questions through two numerical for all resources (resource failures are still indepen-
studies. For the case of discrete demand distributions dent) and vary from 02 to 10 in increments of 01,
(i.e., when probabilities are characterized by scenarios giving us a total of nine supply-chain reliabilities. =
rather than densities), the firm’s investment problem 01 was not used because it can be shown that all
can be formulated as a scenario-based stochastic lin- resources will have zero investment at this reliability
ear program. Formulations for the general model (for level. We chose nine different objective functions: risk
each of the three objective functions) are presented neutral, loss aversion ( = 2 3 4 5), and CVaR (# =
in Appendix E. We used this approach in the two 095 09 085 08). All other model parameters were
numerical studies. held constant across problem instances because they
The first study was designed to address two types were not the focus of the study. The marginal com-
of questions. First, for a given network structure, we mitted cost was fixed at = 02 for all resources. The
want to understand how the investment mix (i.e., per- marginal total costs were fixed at cj = 5 for all dedi-
centage invested in each resource) is influenced by cated resources and at cj = 6 for all flexible resources.
demand correlation, contribution-margin difference, A full factorial set of tests was run for each network,
reliability, and investment criterion. These were cho- so a total of 7 × 5 × 9 × 9 = 2835 instances was solved
sen because they are key drivers of the attractiveness for each of the four networks. The optimal solution
of flexibility and dual sourcing. Second, we want to and objective value for each problem instance was
understand the relative value of the various network stored in a database that is available from the authors
structures. For example, consider a firm that currently on request.
operates an SD network. How much benefit does it The investment-mix question is particularly rele-
get by moving to a dual-sourcing network? Which vant for the DF network. For the risk-neutral objec-
dual-sourcing network is preferred, and under what tive, Tables 1 to 3 present the percentage invested in
circumstances? We fixed the number of products at the flexible resource,
N = 2 in this study. The second study was designed DF ∗
K3
DF ∗ RN
to investigate the relative performance of the four net- %K3 RN = DF ∗ DF ∗ DF ∗

K1 RN + K2 RN + K3 RN
works as the number of products increases.
DF ∗
We now describe the design and discuss the results where the numbers presented are medians for %K3 RN
for the first study. The demand distribution for each taken across the study instances. For example, there
product was characterized by 200 demand scenarios.
The demand scenarios were drawn randomly from Table 1 Influence of Correlation on Percent
Investment in the Flexible Resource
a bivariate normal distribution. Because the demand in the DF Network
variance was not a focus of this study, we fixed the
%K3DFRN∗
demand mean and standard deviation to be 100 and
30 respectively, for both products. We varied the cor- −100 38
relation coefficient () from −1 to 1 in increments of −069 38
−038 35
1/3, giving us a total of seven demand distributions −004 33
that varied in their correlation coefficient. The actual 031 31
correlations were −100, −069, −038, −004, 031, 066 30
100 29
066, and 100. The contribution margin for product 2
Table 2 Influence of Relative Contribution Table 4 Influence of Correlation on Percent of Instances in Which DF
Margin on Percent Investment in the Is Better, Worse, or Equal to DD
Flexible Resource in the DF Network DF ∗ DD ∗ DF ∗ DD ∗ DF ∗ DD ∗
VRN > VRN VRN < VRN VRN = VRN
DF ∗
p1 /p2 %K 3 RN
−100 8222 1778 000
−069 7778 2222 000
1.00 30
−038 7111 2889 000
1.05 32
−004 6444 3556 000
1.10 33
031 4444 5556 000
1.15 34
066 1778 8222 000
1.20 36
100 000 8889 1111
were 315 risk-neutral instances for the DF network,

for a dedicated resource (because c1 = c2 = 5, c3 = 6).
so there were a total of 45 instances for each of the
At low reliabilities, this higher cost outweighs any
seven correlations. As can be seen, the median per-
flexibility benefits provided by the flexible resource
centage invested in the flexible resource increased in
because the benefits only arise in the unlikely event
the relative contribution margin and decreased in the
that the resource investment succeeds. As reliabil-
demand correlation. Even for perfectly positively cor-
ity increases, there is initially a dramatic increase in
related demands, the median investment in the flex- DF ∗
%K3 RN because the flexibility and diversification ben-
ible resource was 29%. This number is driven by
efits of the flexible resource become significant, but
two factors. First, as shown by VM98, an asymmet-
as the network reliability continues to improve there
ric contribution margin can make the flexible resource DF ∗
is a significant decrease in %K3 RN as the diversifica-
attractive even if = 1. Second, the flexible resource
tion benefit becomes less important (and nonexistent
offers a diversification benefit in unreliable networks.
at = 1). While the numbers in Tables 1 to 3 are pre-
For those instances with perfectly positively corre-
sented for the risk-neutral objective, the same obser-
lated demand and identical contribution margins (i.e.,
vations for the influence of correlation, relative mar-
no flexibility benefit), the average flexible investment
gin, and reliabilities held for the loss-averse and CVaR
was 0% when = 1 (i.e., perfectly reliable), but was
objective functions.
18% when = 08. This demonstrates the importance
We compared the optimal objective value across the
of the diversification benefit that the flexible resource
networks for each problem instance. Given the study
provides in the DF network when investments are
design, the DF network can never underperform the
unreliable. Reliability itself has a somewhat complex
DF ∗ SD or SF networks, but can underperform the DD net-
influence on %K3 RN . For a given supply-chain reli- work. The DF network outperformed the SD network
ability, the expected marginal investment cost cj + DF ∗ SD ∗
(i.e., VRN > VRN ) in 93.65% of the 315 risk-neutral
1 − is 20% higher for the flexible resource than
instances and the performance was equivalent in the
other 6.35% of the 315 instances. The median expected-
profit improvement was 14.63% over the 93.65% of
Table 3 Influence of Reliability on Percent DF ∗ SD ∗
Investment in the Flexible Resource cases for which VRN > VRN , indicating that a firm
in the DF Network can gain significant advantage by moving from an SD
%K3DFRN∗
Table 5 Influence of Relative Margin on Percent of Instances in Which
0.2 22 DF Is Better, Worse, or Equal to DD
0.3 38
DF ∗ DD ∗ DF ∗ DD ∗ DF ∗ DD ∗
0.4 38 p1 /p2 VRN > VRN VRN < VRN VRN = VRN
0.5 37
0.6 35 1.00 3810 6032 159
0.7 32 1.05 4444 5397 159
0.8 27 1.10 5079 4762 159
0.9 23 1.15 5714 4127 159
1.0 17 1.20 6508 3333 159
Table 6 Influence of Network Reliability on Percent of Instances in is somewhat indifferent between these two networks,
Which DF Is Better, Worse, or Equal to DD but this is not the case. As Tables 4 to 7 demon-
DF ∗
VRN DD ∗
> VRN DF ∗
VRN DD ∗
< VRN DF ∗
VRN DD ∗
= VRN strate, the network preference is highly driven by the
demand correlation, the relative contribution margin,
0.2 000 10000 000
0.3 857 9143 000 the resource reliability, and the investment criterion.
0.4 4286 5714 000 The DF network is much more attractive at lower
0.5 5714 4286 000 demand correlations, higher relative margins, and
0.6 6286 3714 000
0.7 6857 3143 000 higher reliabilities, whereas the DD network is much
0.8 6571 3429 000 more attractive at higher demand correlations, lower
0.9 6857 3143 000 relative margins, and lower reliabilities. The attrac-
1.0 8571 000 1429
tiveness of the DF network increases as either the
demand correlation decreases or the relative contri-
bution margin increases because the demand-pooling
network to a DF network. The DF network offers both
and contribution-margin benefits of flexibility are
flexibility and diversification benefits, so the question
higher in such circumstances. The DF network is less
arises as to which type of benefit is really driving the
attractive at low reliabilities because the extra level of
superior performance. We investigated this question
diversification provided by the DD network is very
by creating a network with the same diversification
beneficial if resources are very unreliable. The firm’s
benefit but no flexibility benefit. This was done by
investment criterion is also a key driver of network
creating a special DD network in which the failures
preference. As the firm becomes more loss averse,
of resources 2 and 3 were perfectly positively corre-
it prefers the DD network in a higher percentage of
lated (i.e., if one failed, so did the other) and having
instances.
their marginal total costs be identical to the flexible
In the second study, we investigated the influence
resource cost (i.e., c2 = c3 = 6). This network outper-
of the number of products on the relative perfor-
formed the SD network in 66.67% of the instances, and
mance of the four networks. This was done by fix-
in those instances the median expected profit improve-
ing all other parameters and varying the number of
ment was 3.48%. Comparing these results with those
products from two to five. Demand for each prod-
for the DF network, we see that, whereas the diver-
uct was assumed to be independent (as the influence
sification benefit of the DF network is significant, the
of correlation was established above) and was char-
flexibility benefit is more significant.
acterized by 200 demand scenarios randomly drawn
Although the DF network provides some diversi-
from a normal distribution with mean and stan-
fication, the (independent failure) DD network pro-
dard deviation of 100 and 30, respectively. Product
vides more diversification but no flexibility. For the
contribution margins were assumed to be identical
risk-neutral objective, the DF network outperformed
DF ∗ DD ∗ (as the influence of contribution-margin differences
the DD network (VRN > VRN ) in 51.11% of the
DF ∗ DD ∗ was established above) and equal to 10. Resource
instances, underperformed (VRN < VRN ) in 47.30%
of the instances, and performed equally in 1.59% of
the instances. Such data might suggest that a firm
Table 8 Influence of the Number of Products on the Relative Difference
in Profit Between the DF and DD Networks
Table 7 Influence of Loss Aversion on Percent of Instances in Which N =2 N =3 N =4 N =5

DF Is Better, Worse, or Equal to DD
0.3 −1287 −903 −588 −368
VLADF ∗ > VLADD ∗ VLADF ∗ < VLADD ∗ VLADF ∗ = VLADD ∗ 0.4 −456 045 408 667
0.5 −143 390 746 1012
1 5111 4730 159 0.6 −024 512 861 1104
2 4146 5679 174 0.7 017 522 851 1078
3 3750 6071 179 0.8 057 494 774 988
4 3500 6321 179 0.9 137 480 713 879
5 3383 6429 188 1.0 248 514 679 838
Table 9 Drivers of a Firm’s Preference for a Flexible or Dedicated Strategy
Element Attribute Influence on preference Reason
Product portfolio Demand correlations Preference for DF increases as demands become Negative correlation increases the demand-pooling
more negatively correlated. benefit of the flexible resource in the DF
network.
Contribution margins Preference for DF increases as the spread in Wider margin range increases the
contribution margins increases. contribution-margin option benefit (VM98) of
the flexible resource in the DF network.
Number of products Preference for DF increases as the number of The demand-pooling benefit of the flexible resource
products increases. increases as the number of products increases.
Resources Reliabilities Preference for DF decreases as resource Higher probability of resource failures increases
investments become less reliable. the diversification benefits of the DD network.
Firm Risk tolerance Preference for DF decreases as firm becomes In an unreliable network, the extra diversification
more concerned about downside risk. provided by the DD network lowers the firm’s
downside risk.
costs were assumed to be the same as in the above even if dedicated resources are more expensive than
study. We chose a risk-neutral objective (as the influ- a flexible resource.
ence of non-risk-neutral objectives was established We provided analytical results for the directional
above) and solved the investment problem for each influence of prices, marginal costs (total and commit-
of the four network structures for network reliabilities ted), and reliabilities on the optimal expected profit
= 03 10 and number of products N = 2 3 4 5. and resource levels for both the DD and DF networks.
We then calculated the relative network performance In contrast to the DD network, optimal dedicated-
as and N vary. The relative performance for the DF resource levels in the DF network are dependent on
DF ∗ DD ∗ DD ∗
and DD networks (100 × VRN − VRN /VRN ) can resources that are dedicated to other products, a result
be found in Table 8. For any given value of , the that suggests that flexible supply chains may not lend
relative performance of the DF network improves as themselves to decentralized design as easily as do
the number of products increases. The reason for this product-dedicated supply chains.
is that the demand-pooling benefit of the flexibility A comprehensive numerical study was undertaken
resource increases with the number of products. to investigate how the attributes of three key supply-
chain elements—namely, product portfolio, resources,
5. Conclusions and the firm—influence the desirability of a given
In this paper, we bridged the mix-flexibility and design strategy. As one would expect, the desirabil-
dual-sourcing literatures by studying four canonical ity of a dual-sourcing network (DD versus SD or
supply-chain design strategies. Comparing the SD DF versus SF) increases as supply-chain reliability
and SF networks, we identified the critical roles that decreases. The story is more nuanced when comparing
risk tolerance and resource reliabilities play in the rel- single-sourcing networks (SD versus SF) or dual-
ative attractiveness of the two networks. We refined sourcing networks (DD versus DF). Table 9 summa-
the prevailing intuition that an SF-type network is rizes the key results for such comparisons. We note
preferable to an SD-type network if a flexible resource that, whereas Table 9 is framed in terms of dual-
is no more costly than a dedicated resource. In par- sourcing networks, a similar story holds for single-
ticular, we proved that the intuition is valid if either sourcing networks.
the resource investments are perfectly reliable or the We conclude by identifying two dimensions not
firm is risk neutral, but the intuition can be wrong considered in this paper. First, a firm’s enthusiasm for
if neither condition holds. All things being equal resource diversification might be dampened by scale
(resource costs and reliabilities), a dedicated strategy economies in resource investments or by coordination
can actually be strictly better than a flexible strategy. costs in dealing with multiple suppliers for the
In fact, the dedicated strategy can be strictly better same product. Second, supply-chain design may be a
n = 1 + N a1n X 1 + N b1n and so

decentralized rather than a centralized endeavor, and Furthermore, Nn=1 X n=2 n=2
this raises the question of how to coordinate resource-
N

N
N
N
investment decisions. We hope that future research ES; Xn = 1 + 1 −
a1n ES; X b1n = n
ES; X
n=1 n=2 n=2 n=1
will further refine the insights provided in this paper
by addressing these and other considerations. Therefore, +RN = 0.
Proof of Proposition 2. If X 1 and X 2 are independent
and identically distributed (i.i.d.) U 0 1, then X 1 + X
3 = X 2
Acknowledgments
has a triangular distribution over 0 2 and so FX3 x = x2 /2
The authors would like to thank the senior editor and the
for 0 ≤ x ≤ 1 and FX3 x = 2x − x2 /2 − 1 for 1 < x ≤ 2. Proof
two referees for their comments and suggestions, which sig-
follows from application of Equation (18).
nificantly improved this paper.
Proof of Proposition 3. Both resources are identical,
SD ∗ SD ∗
so an optimal solution will have K1 LA = K2 LA . We there-
Appendix A. Proofs
SF ∗ SD ∗ fore can restrict attention to investments of the form K1 =
Proof of Proposition 1. (i) + ≥ 0 ⇔ VRN − VRN ≥ 0 at
K2 = K, and K ∗ must be less than or equal to one because
cN +1 = c. Using Equations (10), (11), (12), and (13), we then
the demand is U 0 1. Using Equations (21)–(24) and the
have
Gy1 y2 K1 K2 equations in Appendix D, one can derive the
FX−1 ; following expressions for VLA SD
K K. If p > 2c, then
N +1
+≥0 ⇔ xN +1 fXN +1 xN +1 dxN +1
0 SD
VLA KK
N Kn∗

− xn fXn xn dxn ≥ 0 (A-1) K2

n=1 0
= 2 p −c+1−K −p − −1 1−2 cK
2
F −1 ;
where ; = 1 − + 1 − c/p. Now, 0 X xf x dx = 1−c1−K2 2 2 c 3 3
+ + K (A-2)
−;ES; X, where ES; X is the ;-expected shortfall for a 2p 3p3
continuous random variable X as defined in Acerbi and and if p ≤ 2c, then
Tasche (2002, hereafter AT02). Note that ; ∈ 0 1. By sub-
SD
stituting this into (A-1), rearranging the terms, and using VLA KK
N +1 = N X
the fact that X

n=1 n , we then have K2

= 2 p −c+1−K −p − −1 1−2 cK

N
N 2
+RN ≥ 0 ⇔ ; n − ES;
ES; X n
X ≥ 0

2
1−c1−K 5c p 2c 3
n=1 n=1 + + 2 − − 3 K3
2p 6 4 3p
Now ES; X is subadditive (see AT02), i.e.,

p 2c 2

N N − 2c − − K2 (A-3)
2 p
ES; n ≤
X n
ES; X
SD
n=1 n=1 VLA K K is a concave (cubic) function in K (for both p/c
and so +RN ≥ 0. regions) and so the first-order condition is sufficient for
(ii) It is straightforward to show that the flexible-only optimality. For both p/c regions the first-order condition
resource level and the dedicated-only resource levels will is quadratic with one single nonnegative root. This root
be positive if and only if > c/p − 1 − c. Therefore, is the optimal investment level given in the proposition
statement.
SF ∗
≤ c/p − 1 − c ⇒ VRN SD ∗
− VRN = 0 Proof of Proposition 4. (i) Let KSD ∗ = K1∗ KN∗ be
the optimal SD investment for some u1 ∈ U1 . Consider the
SD
(iii) 1n = 1 for n = 2 N . Therefore, for each for n = following feasible SF investment: KN +1 = Nn=1 Kn∗ . Let W
2 N we have Xn = a1n X 1 + b1n for some constants a1n and W be the profit random variables for the SD and SF
SF
and b1n with a1n > 0. Using the positive-homogeneity and network using these strategies, with distributions denoted
translation-invariance properties of the expected shortfall by FW SD w and FW SF w. For any demand realization x =
measure, we have ES; aX + b = aES; X − b; see AT02, x1 xN , the SD and SF networks will have the follow-
Proposition 3.1(iii) and (iv). We then have ing terminal wealths:

N
N
N
N
n = ES; X
ES; X 1 + 1 − b1n
a1n ES; X w SD KSD ∗ = w0 + p min!xn Kn∗ " − c Kn∗ (A-4)
n=1 n=2 n=1 n=1

N

N
N
N
N
N
= 1+ 1 −
a1n ES; X b1n w SF Kn∗ = w0 + p min xn Kn∗ − c Kn∗ (A-5)
n=2 n=2 n=1 n=1 n=1 n=1
so where

N h11 = −1 p1 3 I4 K1 K2 K3 +1−3 I4 K1 K2 0
SF
w Kn∗ SD
−w K SD∗

n=1 +I6 K1 K2 0+p2 2 3 I2 K1 K2 K3

N
N
N
+I5 K1 K2 K3 +1−2 3 I2 K1 0K3
= p min xn Kn∗ − min!xn Kn∗ " ≥ 0 (A-6)
n=1 n=1 n=1
+I5 K1 0K3 −I4 K1 K2 K3
SF first-order stochasti-
Therefore, FW SF w ≤ FW SD w, so W +p1 −p2 3 I6 K1 K2 K3
SD
cally dominates W . Therefore,
h12 = h21 = −1 2 3 I2 K1 K2 K3

N
h13 = h31 = −1 3 p1 I4 K1 K2 K3 +p2 2 I2 K1 K2 K3
EuSD
1 K
SD ∗
≤ E uSF
1 Kn∗
n=1
+1−2 I2 K1 0K3 −I4 K1 K2 K3
for all utility functions u1 ∈ U1 (see Equation (4) in Levy
+p1 −p2 I6 K1 K2 K3
1992) and so

h22 =−2 p2 1 3 I1 K1 K2 K3 +I2 K1 K2 K3 +I3 K1 K2 K3
EuSD
1 K
SD ∗
≤ EuSF ∗
1 KN +1
+1−1 3 I2 0K2 K3 +I3 0K2 K3
Therefore + ≥ 0.
+1−3 I1 K1 K2 0+I3 K1 K2 0
(ii) The loss-averse objective function VLA K is identical
to an expected utility objective function where the utility h23 = h32 = −2 3 p2 1 I1 K1 K2 K3 +I2 K1 K2 K3
function is piecewise-linear increasing (with the breakpoint
+1−1 I2 0K2 K3
at w0 ). Such a utility function is in the U1 set, so the loss-

averse result follows directly from Proposition (4). W SF first- h33 = −3 p1 1 I4 K1 K2 K3 +1−1 I4 0K2 K3
SD SF
order stochastically dominates W , so W second-order
stochastically dominates W SD . Recalling that CVaR is the +p2 1 2 I1 K1 K2 K3 +I2 K1 K2 K3
average value of the profit falling below the #-percentile +1 1−2 I1 K1 0K3 +I2 K1 0K3
level, we can use Theorem 3 in Levy (1992) to state that
−I4 K1 K2 K3 +1−1 2 I2 0K2 K3

SD

N
VCVaR K1∗ KN∗ ≤ VCVaR
SF
Kn∗ +1−1 1−2 I2 00K3 −I2 0K2 K3
n=1
+p1 −p2 1 I6 K1 K2 K3 +1−1 I6 0K2 K3
Therefore,
where the line integrals Ik K1 K2 K3 k = 1 6 are given
SD
VCVaR K1∗ KN∗ ≤ VCVaR
SF
KN∗ +1 in Appendix C and are similar to the Ik expressions in the
proof of Proposition 1 in VM98. Because we have to general-
ize for the case of investment failures, we write the function
so + ≥ 0 for the CVaR objective function.
in terms of Ik K1 K2 K3 rather than simply Ik .
Proof of Proposition 5. Proof of this proposition fol-
Proof of the gradients of Kj∗ with respect to c, , and
lows the same structure as that of VM98 Propositions 3
DF p follows from use of the Implicit Function Theorem (IFT)
and 4. VRN K is nonincreasing convex in c = c1 c2 c3 on
DF ∗ for each of the eight possible optimal solution structures.
the convex set + +
3 for each K ∈ 3 . VRN c is therefore non- Application of the IFT requires detailed algebra that is avail-
increasing convex in c as maximization preserves convex-
DF
able from the authors on request. We note that the interior
ity. A similar argument holds for = 1 2 3 . VRN K is solution structure K1∗ > 0 K2∗ > 0 K3∗ > 0 is by far the most
nondecreasing convex in p = p1 p2 on the convex set p ≥ cumbersome, and the only one in which we needed the suf-
DF ∗
p2 ≥ 0 for each K ∈ + 3 . VRN p is therefore nondecreas- ficient condition of log-convexity or log-concavity for the
ing convex in p as maximization preserves convexity. The proof.
DF
Hessian matrix for VRN K is Proof of Proposition 6. Proof of this proposition fol-
DF
  lows the same structure as that of Proposition 5. VRN K is
h11 h12 h13 nondecreasing convex in = 1 2 3 on the convex set
  DF ∗
  0 ≤ j ≤ 1, j = 1 3, for each K ∈ +
3 . VRN K is therefore
 h21 h22 h23 
  nondecreasing convex in as maximization preserves con-
h31 h32 h33 vexity. Proof of the gradients of Kj∗ with respect to follows
from use of the IFT for each of the eight possible optimal where the Gy1 y2 K1 K2 v expressions can be found in
solution structures. Application of the IFT requires detailed Appendix D. Numerical results (available from the authors
algebra that is available from the authors on request. We note on request) prove that the flexibility premium +CVaR can be
that the interior solution structure K1∗ > 0 K2∗ > 0 K3∗ > 0 negative, so SD can be strictly preferable to SF even when
is by far the most cumbersome and the only one in which the flexible resource costs the same or less than the ded-
we used the sufficient condition of i.i.d. uniform demands icated resources, a result that echoes the loss-averse case.
for the proof. In the case of CVaR, the downside resource-disaggregation
DD
Proof of Proposition 7. VRN K is nonincreasing con-
benefit of the dedicated strategy is amplified by the fact that
vex in c (and in ) on the convex set + +
3 for each K ∈ 3 .
DD ∗ the left tail of the profit distribution is always factored in,
VRN K is therefore nonincreasing convex in c (and in )
as maximization preserves convexity. VRN DD
K is nondecreas- whereas the right tail is not, with the result that SD can out-
ing convex in p (and in ) on the convex set p ≥ p2 ≥ 0 perform SF.
DF ∗
(0 ≤ j ≤ 1, j = 1 3) for each K ∈ + 3 . VRN K is there-
fore nondecreasing convex in p (and in ) as maximiza- Appendix C. Expressions for P 8k and Line
DD
tion preserves convexity. VRN K is separable in K1 K2 Integrals Ik K1 K2 K3
and K3 K4 , i.e., the problem decomposes into two single-
product problems. The resources dedicated to product i are
thus independent of resources dedicated to product 3 − i, P 81 = fX x1 x2 dx2 dx1

K1 0
i = 1 2. Proof of the gradients of Kj∗ follows from use of
the IFT for each of the four possible optimal solution struc- P 82 = fX x1 x2 dx2 dx1
K1 +K3 0
tures for the single-product problem; the detailed algebra is K1 +K3
available from the authors on request. P 83 = fX x1 x2 dx2 dx1
K1 K1 +K3 −x1
Appendix B. The Flexibility Premium for the K1 +K3
CVaR Measure P 84 = fX x1 x2 dx2 dx1
K1 K1 +K2 +K3 −x1
Chen et al. (2003) prove that K ∗ = FX−1 #p − c/p − v is the
optimal investment for a perfectly reliable single-product P 85 = fX x1 x2 dx2 dx1
0 K2
newsvendor problem under a CVaR# objective, where v is
K3
the salvage value (assumed to be zero in our work). We
P 86 = fX x1 x2 dx2 dx1
extend this result to allow for Bernoulli investment failures. 0 K2 +K3 −x1
Recall that XN +1 = X
1 + X
2 + · · · X
N and cN +1 is the marginal
total cost for the flexible resource. P 87 = fX x1 x2 dx2 dx1
K3 K2

Proposition 8. For the SF network, the optimal investment
P 88 = fX x1 x2 dx2 dx1
level for the CVaR# objective is K1 +K3 K2

K1
∗ −1 # c3 1− cN +1 P 89 = fX x1 x2 dx2 dx1
K3CVaR# = FXN +1 1− + 1−−1 (B-1)
p p 0 K2 +K3

Proof. Available from the authors on request. P 810 = fX x1 x2 dx2 dx1
We note that at = 1 (B-1) gives the perfect-reliability K3 0
K3
investment of Chen et al. (2003) for the case of zero salvage
P 811 = fX x1 x2 dx2 dx1
value, and it also gives the risk-neutral optimal investment 0 K3 −x1
(10) at # = 1. K1
Using Equation (7), the SD investment problem can be P 812 = fX x1 x2 dx2 dx1
0 K3
formulated as K1
SD∗
VCVaR# I1 K1 K2 K3 = fX x1 K2 + K3 dx1
0
K1 +K3
1
= max v+ EXY min WSD K1 KN −v0 (B-2) I2 K1 K2 K3 = fX x1 K1 + K2 + K3 − x1 dx1
K1 ≥0KN ≥0
v
#
K1

As with loss aversion, the SD investment problem cannot be I3 K1 K2 K3 = fX x1 K2 dx1
K1 +K3
decomposed into N single-product problems. For the two- K2
product case, one can show that I4 K1 K2 K3 = fX K1 + K3 x2 dx2
0
SD K1 K2 −v0"
EXY min!W
I5 K1 K2 K3 = fX K1 x2 dx2
K2 +K3
= 1−1 1−2 G00 K1 K2 v+1 1−2 G10 K1 K2 v
+1−1 2 G01 K1 K2 v+1 2 G11 K1 K2 v (B-3) I6 K1 K2 K3 = fX K1 + K3 x2 dx2
K2

Appendix D. Gy1 y2 K1 K2 and Gy1 y2 K1 K2 v c K − p − c2 K2 + v c1 K1 + c2 K2 + v
+ p NX1 X2 1 1
Expressions p p
We present the expressions for the general case where

c K − p − c2 K2 + v c1 K1 + c2 K2 + v
the marginal committed costs () and marginal total costs + OX1 X2 1 1
p p
(c) can differ for resources 1 and 2. The Gy1 y2 K1 K2
− c1 K1 + c2 K2 + v
expressions found in Equation (24) in the loss-averse sec-

tion are given by Gy1 y2 K1 K2 = Gy1 y2 K1 K2 v = 0, where c K − p − c2 K2 + v c1 K1 + c2 K2 + v
· MX1 X2 1 1
Gy1 y2 K1 K2 v are given below. We present the more gen- p p
eral Gy1 y2 K1 K2 v expressions because they are used in p − c1 K1 − c2 K2 < v ≤ p − c2 K2 − c1 K1 ⇒
the CVaR analysis.
G11 K1 K2


 c K + c2 K2 + v c K + c2 K2 + v
 0 for v < −c1 1 K1 +c2 2 K2
 = p NX1 X2 1 1 − NX1 X2 K1 1 1
G00 K1 K2 v = −c1 1 K1 +c2 2 K2 +v p p



 c1 K1 + c2 K2 + v
for v ≥ −c1 1 K1 +c2 2 K2 + OX1 X2

p
 c1 1 K1 +c2 K2 +v


 X2pL c K + c2 K2 + v

 p − OX1 X2 K1 1 1 − c1 K1 + c2 K2 + v

 p



 −c K
2 2 +c 1 1 K1 +v


 c K + c2 K2 + v c K + c2 K2 + v

 c K +c K +v · MX1 X2 1 1 − MX1 X2 K1 1 1

 ·FX2 1 1 1 2 2 p p
G10 K1 K2 v = p

 c K − p − c1 K1 + v

 for v ≤ p −c 2 K2 −c1 1 K1 + pK1 − c1 K1 − c2 K2 − vFX2 2 2

 p




 pLX2 K2 +K2 1−FX2 x2

 c2 K2 − p − c1 K1 + v

 + pLX2 1 − FX1 K1

 −c2 K2 +c1 1 K1 +v p


for v > p −c2 K2 +c1 1 K1

v > max!p −c1 K1 −c2 K2 p −c2 K2 −c1 K1 " ⇒
 c 1 1 K1 +c2 K2 +v

 pLX2 G11 K1 K2

 p




 c1 1 K1 +c2 K2 +v c K −p −c2 K2 +v

 −c2 K2 +c1 1 K1 +vFX2 = pLX1 1 1 +pLX2 K2

 p p


G01 K1 K2 = for v ≤ p −c2 K2 −c1 1 K1 c K −p −c2 K2 +v

 +K2 1−FX2 K2 FX1 1 1

 p

 pLX2 K2 +K2 1−FX2 x2



 c K −p −c2 K2 +v

 −c2 K2 +c1 1 K1 +v −c1 K1 +c2 K2 +vFX1 1 1

 p
for v > p −c2 K2 +c1 1 K1

c1 K1 −p −c2 K2 +v c1 K1 +c2 K2 +v
v ≤ min!p −c1 K1 −c2 K2 p −c2 K2 −c1 K1 " ⇒ +p NX1 X2
p p

c K +c K +v c K +c K +v
G11 K1 K2 = p NX1 X2 1 1 2 2 −NX1 X2 K1 1 1 2 2
p p

c1 K1 +c2 K2 +v c1 K1 −p −c2 K2 +v c1 K1 +c2 K2 +v
+OX1 X2 +p OX1 X2
p p p

c K +c K +v
c1 K1 +c2 K2 +v −OX1 X2 K1 1 1 2 2 −c1 K1 +c2 K2
−c1 K1 +c2 K2 +vMX1 X2 p
p

p −c2 K2 −c1 K1 < v ≤ p −c1 K1 −c2 K2 ⇒ c K −p −c2 K2 +v c1 K1 +c2 K2 +v
· MX1 X2 1 1
G11 K1 K2 p p

c K +c K +v
c K −p −c2 K2 +v −MX1 X2 K1 1 1 2 2
= pLX1 1 1 +pLX2 K2 +K2 1−FX2 K2 p
p

c K −p −c1 K1 +v
c K −p −c2 K2 +v + pK1 −c1 K1 −c2 K2 −vFX2 2 2
·FX1 1 1 p
p

c K −p −c1 K1 +v
c K −p −c2 K2 +v +pLX2 2 2 1−FX1 K1
−c1 K1 +c2 K2 +vFX1 1 1 p
p
where subject to
w2
MX1 X2 w1 w2 = FX2 w2 − x1 fX1 x1 dx1 w mi+ −w mi− = w mi m = 1M i = 1I (E-6)
w1
w2 w mi+ w mi− ≥ 0 m = 1M i = 1I
NX1 X2 w1 w2 = x1 FX2 w2 − x1 fX1 x1 dx1
w1 and (E-1), (E-2), (E-3), (E-4), (E-5). (E-7)
w2
OX1 X2 w1 w2 = LX2 w2 − x1 fX1 x1 dx1 CVaR LP formulation:
w1

MX1 X2 w = MX1 X2 0 w 1 I M
max v + mi zmi
NX1 X2 w = NX1 X2 0 w # i=1 m=1
subject to
and
OX1 X2 w = OX1 X2 0 w zmi ≤ w mi − vmi m = 1 M i = 1 I (E-8)
zmi ≤ 0 m = 1 M i = 1 I
Appendix E. Linear Program Formulations for
Investment Problem and (E-1), (E-2), (E-3), (E-4), (E-5) (E-9)
There are a total of M = 2J possible yield scenarios. A yield
scenario m = 1 M is defined by a vector y1m yJm
where yjm ∈ !0 1", with yjm = 0 denoting failure. The prob-
ability of scenario m is denoted by m . Let there be I dif- References
ferent demand scenarios, each defined by a demand vector Acerbi, C. 2002. Spectral measures of risk: A coherent represen-
x1i xNi , with probability i , i = 1 I. If the demand tation of subjective risk aversion. J. Banking Finance 26(7)
scenarios are randomly generated from some joint distribu- 1505–1518.
tion X, then i = 1/I for i = 1 I. By assumption, yields Acerbi, C., D. Tasche. 2002. On the coherence of expected shortfall.
J. Banking Finance 26(7) 1487–1503.
are independent of demands and so we have a total of MI
Agrawal, N., S. Nahmias. 1997. Rationalization of the supplier base
yield-demand scenarios, each specified by an mi pair with
in the presence of yield uncertainty. Production Oper. Manage-
an associated probability of mi = m i . ment 6(3) 291–308.
Risk-neutral linear program (LP) formulation: Agrawal, V., S. Seshadri. 2000. Impact of uncertainty and risk aver-
sion on price and order quantity in the newsvendor problem.

I
M
Manufacturing Service Oper. Management 2(4) 410–423.
max mi w mi
i=1 m=1
Anupindi, R., R. Akella. 1993. Diversification under supply uncer-
tainty. Management Sci. 39(8) 944–963.
subject to Artzner, P., F. Delbaen, J.-M. Eber, D. Heath. 1999. Coherant mea-
sures of risk. Math. Finance 9(3) 203–228.

N J
Babich, V., A. N. Burnetas, P. H. Ritchken. 2004. Competition and
w mi = pn snmi − cj j +1−j yjm Kj diversification effects in supply chains with supplier default
n=1 j=1 risk. Working paper, Weatherhead School of Management,
m = 1M i = 1I (E-1) Case Western Reserve University, Cleveland, OH.
Caldentey, R., M. Haugh. 2004. Optimal control and hedging of
snmi ≤ xni n = 1N m = 1M i = 1I (E-2) operations in the presence of financial markets. Working paper,
J
Stern School of Business, New York University, New York.

snmi ≤ tnj qnjmi n = 1N m = 1M i = 1I (E-3) Chen, X., M. Sim, D. Simchi-Levi, P. Sun. 2003. Risk aversion in
j=1 inventory management. Working paper, Massachusetts Insti-
tute of Technology, Cambridge, MA.

N
Dada, M., N. C. Petruzzi, L. B. Schwarz. 2003. A newsvendor model
qnjmi ≤ yjm Kj j = 1J m = 1M i = 1I (E-4) with unreliable suppliers. Working paper, Krannert Graduate
n=1
School of Management, Purdue University, West Lafayette, IN.
Kj snmi qnjmi ≥ 0 n = 1N j = 1J Eeckhoudt, L., C. Gollier, H. Schlesinger. 1995. The risk-averse (and
prudent) newsboy. Management Sci. 41(5) 786–794.
m = 1M i = 1I (E-5)
Fine, C. H., R. M. Freund. 1990. Optimal investment in prod-
uct flexible manufacturing capacity. Management Sci. 36(4)
Loss-averse LP formulation: 449–466.
Gerchak, Y., M. Parlar. 1990. Yield randomness, cost trade-offs

I
M
and diversification in the EOQ model. Naval Res. Logist. 37(3)
max mi w mi + − w mi −
i=1 m=1
341–354.
Graves, S. C., B. T. Tomlin. 2003. Process flexibility in supply chains. model with Markovian supply availability. Internat. J. Produc-
Management Sci. 49(7) 907–919. tion Econom. 42(2) 131–136.
Gupta, D., Y. Gerchak, J. A. Buzacott. 1992. The optimal mix Rockafellar, R. T., U. S. Uryasev. 2000. Optimization of conditional
of flexible and dedicated manufacturing capacities: Hedging value-at-risk. J. Risk 2(3) 21–41.
against demand uncertainty. Internat. J. Production Econom. Rockafellar, R. T., U. S. Uryasev. 2002. Conditional value-at-risk for
28(3) 309–319. general loss distributions. J. Banking Finance 26(7) 1443–1471.
Gürler, Ü., M. Parlar. 1997. An inventory problem with two ran- Schweitzer, M. E., G. P. Cachon. 2000. Decision bias in the newsven-
domly available suppliers. Oper. Res. 45(6) 904–918. dor problem with a known demand distribution: Experimental
Jordan, W. C., S. C. Graves. 1995. Principles on the benefits of man- evidence. Management Sci. 46(3) 404–420.
ufacturing process flexibility. Management Sci. 41(4) 577–594. Swaminathan, J. M., J. G. Shanthikumar. 1999. Supplier diversifica-
Kahneman, D., A. Tversky. 1979. An analysis of decision under risk. tion: Effect of discrete demand. Oper. Res. Lett. 24(5) 213–221.
Econometrica 47(2) 263–292. Szegö, G. 2002. Measures of risk. J. Banking Finance 26 1253–1272.
Levy, H. 1992. Stochastic dominance and expected utility. Manage- Tomlin, B. T. 2004a. Coping with supply chain disruptions:
ment Sci. 38(4) 555–593. Dual sourcing, volume flexibility and contingency planning.
Li, S., D. Tirupati. 1994. Dynamic capacity expansion problem with Working paper, Kenan-Flagler Business School, University of
multiple products: Technology selection and timing of capacity North Carolina, Chapel Hill, NC.
additions. Oper. Res. 42(5) 958–976. Tomlin, B. T. 2004b. Supplier diversification: The impact of
Li, S., D. Tirupati. 1995. Technology choice with stochastic demands supplier-reliability forecasting on supplier selection and sourc-
and dynamic capacity allocation: A two-product analysis. ing quantities. Working paper, Kenan-Flagler Business School,
J. Oper. Management 12(3) 239–258. University of North Carolina, Chapel Hill, NC.
Li, S., D. Tirupati. 1997. Impact of product mix flexibility and allo- Tversky, A., D. Kahneman. 1992. Advances in prospect theory:
cation policies on technology. Comput. Oper. Res. 24(7) 611–626. Cumulative representation of uncertainty. J. Risk Uncertainty
Muriel, A. Somasundaram, Y. Zhang. 2004. Impact of partial 5(4) 297–323.
manufacturing flexibility on production variability. Working Van Mieghem, J. A. 1998. Investment strategies for flexible
paper, Mechanical and Industrial Engineering Department, resources. Management Sci. 44(8) 1071–1078.
University of Massachusetts, Amherst, MA. Van Mieghem, J. A. 2004a. Note: Commonality strategies: Value
Parlar, M., D. Perry. 1996. Inventory models of future supply uncer- drivers and equivalence with flexible capacity and inventory
tainty with single and multiple suppliers. Naval Res. Logist. substitution. Management Sci. 50(3) 419–424.
43(3) 191–210. Van Mieghem, J. A. 2004b. Risk-averse newsvendor networks:
Parlar, M., D. Wang. 1993. Diversification under yield randomness Resource sharing, substitution, and operational hedging. Work-
in inventory models. Eur. J. Oper. Res. 66(1) 52–64. ing paper, Kellogg School of Management, Northwestern
Parlar, M., Y. Wang, Y. Gerchak. 1995. A periodic review inventory University, Evanston, IL.

On The Value of Mix Flexibility and Dual Sourcing in Unreliable Newsvendor Networks - Copy in Quantification Folder

Cargado por

Información del documento

Descripción original:

Título original

Derechos de autor

Formatos disponibles

Compartir este documento

Compartir o incrustar documentos

Opciones para compartir

¿Le pareció útil este documento?

¿Este contenido es inapropiado?

Copyright:

Formatos disponibles

On The Value of Mix Flexibility and Dual Sourcing in Unreliable Newsvendor Networks - Copy in Quantification Folder

Cargado por

Copyright:

Formatos disponibles

MANUFACTURING & SERVICE

OPERATIONS MANAGEMENT informs ®

Vol. 7, No. 1, Winter 2005, pp. 37–57 doi 10.1287/msom.1040.0063