Sie sind auf Seite 1von 13

Joint Cost Allocation for Multiple Lots

Author(s): Bala V. Balachandran and Ram T. S. Ramakrishnan


Source: Management Science, Vol. 42, No. 2 (Feb., 1996), pp. 247-258
Published by: INFORMS
Stable URL: http://www.jstor.org/stable/2633004
Accessed: 13/12/2008 09:20

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=informs.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the
scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that
promotes the discovery and use of these resources. For more information about JSTOR, please contact support@jstor.org.

INFORMS is collaborating with JSTOR to digitize, preserve and extend access to Management Science.

http://www.jstor.org
Joint Cost Allocation for Multiple Lots

Bala V. Balachandran * Ram T. S. Ramakrishnan


NorthwesternUniversity, Evanston,Illinois 60201
Universityof Illinois at Chicago,Chicago,Illinois 60607

W Az e consider the joint cost allocation problem that arises when several lots or resources are
available to serve different products or divisions. We provide a two-phase model,
wherein the first phase the optimal set of lots to be acquired is chosen and given the optimal
set, and the products using each acquired lot is also determined. In the second phase, a stable
full cost allocation method is developed that will not induce the divisions to form coalitions to
reduce the allocated joint costs. Utilizing the optimal dual solution of the lot selection phase, we
provide a joint cost allocation mechanism based on the concept of propensity to contribute and
show that this allocation is also stable. If in the first phase there is a dual gap, then we show
that there is no cost allocation in the core. A numerical illustration is provided.
(Cost Allocation;Stability;CoreSolutions;Dual Methods)

1. Introduction relevant for economic decisions, at least in the short run.


The allocation of the joint costs of common resources However, managers in both public and private sectors
across the divisions or products that need and use recognize the need for cost allocation in product costing,
those resources is practiced widely in industry. This inventory valuation, and in the measurement of the de-
practice is under greater scrutiny now, as firms are in- centralized unit's income. Modem cost accounting in-
creasingly recognizing the value of decentralization. novations have emphasized the need for full cost allo-
By creating autonomous units with delegated respon- cation when advocating activity based costing and man-
sibility toward personnel, administration, sales, and agement. (See Turney 1990 and Cooper 1990 for activity
distribution, companies are able to react quickly to based allocation and its impact on the accuracy of prod-
growing competitive global markets for inputs and uct costs.) In the public sector, especially in defense re-
outputs. This trend has been extensively documented, lated industries, major cost recoveries are heavily de-
and the prototype of the future corporation is a shell pendent on the allocated joint costs and transfer prices.
type corporate center with no unallocated costs. (See Thomas (1974) questions the usefulness of such alloca-
Dumaine 1992, Business Week 1992 for experiences of tions for decision making and emphasizes that alloca-
Johnson & Johnson, IBM, Asea Brown Baveri, and Du tions may only be "useful in reaching understandings
Pont.) whereby individuals or organizations agree to distri-
Decentralization requires full allocation of common butions of resources." Following Thomas (1974), sev-
costs. If the corporate center does not fully allocate costs, eral researchers have examined different "mutually sat-
then the operating divisions will impute lower costs to isfactory allocations" based on particular economic
the common factor inputs and consume the unallocated principles or behavioral axioms. A few have used
resources excessively. (See Fuchsberg 1992 for the in- mathematical programming approaches and duality
creased personnel costs arising due to decentralization theory for optimal allocations, as seen in Kaplan and
and underallocation.) Atkinson (1986) documents that Thompson (1971) and Kaplan and Welam (1974). An-
over 70%of the firms allocate such common costs. Econ- other group, (for instance, Moriarity 1975, 1976; Loud-
omists (Thomas 1974, Stigler 1966) maintain a view that erback 1976; and Jensen 1977) defined certain sets of
allocation of joint or common costs is arbitrary and ir- axioms or objectives that allocation models should
0025-1909/96/4202/0247$01.25
Copyright X) 1996, Institute for Operations Research
and the Management Sciences MANAGEMENT SCIENCE/Vol. 42, No. 2, February 1996 247
BALACHANDRAN AND RAMAKRISHNAN
Joint Cost Allocationfor Multiple Lots

Table 1 Recently, a leading gasket manufacturer found it prof-


itable to procure multiple lots for their gasket products,
IncrementalProcessing Costs
where different product lines use different raw materi-
Lot
Lot# j Cost 1 2 ... n als (steel, rubber, cork, etc.). They found it worthwhile
to pile up such raw material inventory lots "to take ad-
1 JC1 /11 /21 I. l Inl vantage of economies of scale from different vendors"
2 JC2 /12 /22 I. 2 ... In2 so that the total costs were minimized. They were faced
with allocation of joint ordering costs of such multiple
j JCy lj 12j IiJ ... Inj
lots to their end products servicing original equipment
m- 1 JCm_1 /1,m-1 /2,m-1 ... li,m-1 ... In,m-i
manufacturers such as automobiles and agricultural
equipment. The firm had to select the optimal way of
m 0 /1m Im ... Iim Inm procuring the lots and had to decide on the allocation
of the ordering costs to the end products.
This paper considers the two problems-the selection
possess and provided models that satisfy their sets of of optimal lot set and allocation of common costs-si-
objectives. multaneously. The first problem is solved through a
Several models of allocation have been developed re- mathematical programming approach and several pa-
cently, based on game theoretic concepts such as the pers in the management science literature have mod-
core or Shapley value. These cost allocation schemes elled this problem and developed efficient algorithms.
have the stability property, such that the divisions shar- Crow (1972) considered an assignment method for a
ing the joint or common resources have no incentive to multi-mission system in a defense project with priori-
leave the common resource acquisition structure. Such ties, which was later extended by Jensen (1978) for sen-
allocation models are found in Jensen (1977), Hamlen sitivity of priorities via a parametric analysis of the lin-
et al., (1977, 1980) Balachandran and Ramakrishnan ear programming solution. Since linear programs can
(1981), Roth and Verrechia (1979), Gangolly (1981), and yield fractional solutions and optimal solutions can not
Hughes and Scheiner (1980), to name a few. All these vary continuously when integer values are required for
models are applicable only to the situation where a sin- variables, these methods are deficient. Further, stability
gle lot or resource is to be acquired for a group of di- among participants as defined through the game-
visions or products. Very little analysis has been carried theoretic core property may not exist in such allocations.
out for the case with multiple common resources. Many Sets of participants may have incentives to leave and
divisions may use the whole set of resources or lots, but form separate cost pools to acquire resources for only
all the divisions need not use the same common re- that set.
source. An alternate stream of current research has fo- This paper is aimed to reduce the gap that exists in
cused on the existence of a nontrivial role for allocations the multiple resource literature-between optimization
in resolving incentive conflicts in a multiple agent set- and cost allocation-and provides an allocation scheme
ting (Demski et al. 1988, Zimmerman 1979, Balachan- which has all the desirable stability properties enunci-
dran et al., 1988, Rajan 1992). These papers use hidden ated by Moriarity (1975). Secondly, we provide a model
action models in a principal-agent framework and show that eliminates the fractional outcome that can arise in
that some allocation of fixed common costs could re- the models of Jensen (1978) or Crow (1972). Our model
duce collusion problems. In this paper the common formulation and primal solution assignment generates
costs are not fixed and need to be solved first before an optimal integer solution. We use the algorithms pro-
allocation. Further in those papers incentive problems posed by Erlenkotter (1978) for the fixed charge prob-
arise due to effort aversion and private information. In lem which uses the optimal dual solutions. Thirdly, our
this paper the only incentive issue considered is the mo- model formulation provides the solution to both the
tivation of the unsatisfied divisions to form a separate problem of ex-ante lot selection from the existing set of
resource acquisition pool. feasible scarce lots and the problem of ex-post cost al-

248 MANAGEMENTSCIENCE/Vol. 42, No. 2, February 1996


BALACHANDRAN AND RAMAKRISHNAN
Joint Cost Allocationfor Multiple Lots

location, once the optimally selected lots and the divi- more than the incremental processing cost for at least
sions sharing the selected resources are known. Thus one resource. Similar to the single lot case, if a lot is
this allocation scheme does not influence the decision purchased for the production of a specific product, then
process in the first stage. Fourth, we use the "propensity the entire required amount of that product can be pro-
to contribute" concept developed by Balachandran and duced by further processing the specified lot.
Ramakrishnan (BR) (1981) and the optimal dual solu- Let the number of lots for these resources or raw ma-
tion of the ex-ante lot selection problem to obtain a terials to be m - 1, and let j be a typical lot so that j = 1,
unique cost allocation. As the dual solution has an eco- ... ., m - 1. For the purposes of the model, let the exter-
nomically meaningful interpretation, the ex-post allo- nal market also be a lot corresponding to the decision
cation procedure that is developed lies in the core. This of "buying directly from outside or open market." The
guarantees stability, so that the overall total profit is external market will be labelled with index m, as the last
maximized without sacrificing either the divisional in- lot. Thus there are m lots, where the first m - 1 lots
centives or their independence. Finally we relate the correspond to the resources and the last one corre-
dual gap that may arise in the lot selection phase to the sponds to "market." The firm consists of n different di-
subsidy that will be needed to ensure that the cost al- visions or products.
location is in the core. Let i be the index for a typical product or division so
that i1, .. ., n. Let
JC1:represent the joint cost of the resource or lot j;
2. Multilot Model Development (j= if . .. , m - 1),
For expositional ease, we adopt the same notation as in
Iij:be the incremental cost of further processing for
the papers of Balachandran and Ramakrishnan (1981), product i (i = 1, . . ., n) using lotj (j = 1,.. ., m - 1)
Moriarity (1975), or Louderback (1976) with one exten-
Iim:represent the cost of buying product i directly
sion. An index j is used as a subscript for the lot number, from the external market.
since all these authors consider only a single lot. Since there is no cost for the lot labelled m (market),
In this paper we consider multiple resources, lots, or
JCmwill be set equal to zero. The matrix of both joint
systems' where each one of these lots can be purchased and incremental costs, as given in Table 1, describes the
for a known cost. These lots are used by a group of variables.
divisions, economic entities, or subsidiaries so that fur-
ther processing yields the final required products (or 3. The Two-Phase Model
satisfies a mission as in Crow's (1972) problem). So The joint cost allocation problem for the multiple lot
there are two types of costs: joint cost of the lots and case will be analyzed as a two-phase model consisting
further processing costs for each product. If a product of an ex-ante decision problem and an ex-post allocation
cannot be obtained from any specific lot, we define the problem, given the decisions of the first phase. The first
corresponding processing cost as a large number (infin- phase involves the production or lot selection decision,
ity). The firm also has the option of buying the input while the second phase is concerned with joint cost al-
for that product directly from the market. This direct location ex-post, given the production decision of the
purchase cost is generally higher than the incremental first phase. The production decision is obtained by the
processing cost with some lot, before any allocation of formulation of a "fixed charge" optimization model
the lot cost. If it is not, we eliminate such products from similar to a fixed charge transportation problem so that
consideration since the raw material for that product the total corporate costs are minimized. Then utilizing
could be purchased directly. Thus, without any loss of the dual solution associated with this optimal solution
generality we assume that the direct market price is and the core theory presented in the earlier literature,2

'We will use the word "lot" in a generic sense. This model could be 2 For a full discussion of core and game-theory related definitions and
used for other situations as well, such as systems, resources, support illustrations see Hamlen et al. (1977, 1980), Jensen (1977), and Bala-
centers, and common overhead facilities. chandran and Ramakrishnan (1981).

MANAGEMENTSCIENCE/Vol. 42, No. 2, February 1996 249


BALACHANDRAN AND RAMAKRISHNAN
Joint Cost Allocationfor Multiple Lots

an allocation scheme is developed. This scheme has all With the costs defined in Table 1, the firm's mixed-
the advantages that Moriarity (1975) proposed, plus the integer program (IP) is given by:
property of economic justification implied by the dual- m n m
ity theory of optimization literature3 and the fairness IP: Minimize ZIp= I JCju; + I I IijXi1, (1)
and stability properties of the core theory. j=1 i=1 j=1

The matrix of costs presented in Table 1 identifies two


sets of costs. One is concerned with the fixed cost JCj, Subject to xij = 1, for i =1, ..., n, (2)
j=l
which is incurred if lot j is bought to be used by any set
of products. Iijrepresents the incremental cost of further uj - Xi}2 O0,for i =1...,n and j ,... ,(3)
processing. The problem of optimally selecting (1) the
lots to be bought and (2) the products assigned to use xij:2?O, fori=1,...,n and j=1,...,m, (4)
the chosen lot for further processing, is similar to the u1 Oorl, ..,m. (5)
forj=1,
fixed charge transportationproblem discussed in the op-
erations research literature.This model has been formu- The objective function (1) is the sum of the joint costs
lated for (1) "facility location" decisions and (2) further of resources or lots as given by the first term and the
"transportation"decisions given the facilities that are to incremental processing costs as given by the second
be opened from the decision in step 1. However, Erlen- term. The constraint set given by (2) implies that each
kotter (1978) has developed an efficient algorithm (a com- product i should be produced from some lot j or j = 1,
puter code and performance results are also included in ... . m - 1 or bought from outside (j = m). Constraint
the paper) which uniformly has achieved good compu- sets (3) and (4) require that xijcan be zero, but if it is to
tational results. We will make use of this algorithm and be one, then the associated uj must also be one. (A lot
show how the dual solution of this model has a meaning- cannot be used for production unless it is bought.) Con-
ful economic property for joint cost allocation. straint set (5) is obvious since we either buy or reject lot
j. However, in constraint set (4) we do not require xijto
Phase I: The Optimal Resource Selection Problem
be 0 or 1 but impose only nonnegativity. This relaxation
The "Fixed Charge" Multilot Problem.
is acceptable as the optimal solution will have xij equal
The firm's problem is to optimally select lots that should
to O or 1, because product i will be produced fully from
be bought, from the available set of lots, and to opti-
the open lot (uj = 1) with the least incremental cost.
mally assign the divisions or products to the selected
If, in the integer program (IP) above given by (1) to
lots. This can be formulated as a mixed-integer pro-
(5) we replace constraint set (5) by
gramming problem called the "fixed charge" multilot
problem. We use the expanded cost matrix (including 0 < uj < 1 forj = 1, ...,m, (5')
the market lot option as the source m) and a JC vector
of all joint costs as in column one of Table 1. Let uj be a and keep the other constraint sets (1) to (4) as they are,
decision variable for the purchase or rejection of avail- then the resultant program is a linear program denoted
able lot j: Let by (LP). This problem (LP) is called the "linear relaxa-
tion" of the integer program (IP). Exploiting the special
I1,if resource/lot j is purchased, structure in the constraints, Erlenkotter (1978) used a
l 0, if resource/ lot j is not purchased, "dual ascent" algorithm for the associated LP to obtain
= 1, if producti is made using resourcej, tight lower bounds for the IP. The dual of this LP is then
xii 1 used to obtain an integer solution for IP. The dual so-
otewie
lutions lead to a good candidate primal solutions,
thereby giving an upper bound. Since Erlenkotter (1978)
has discussed the solution procedure to solve Phase I,
3 Dual approaches were used in allocation applications by Kaplan and
Thompson (1971) and Kaplan and Welam (1974) for linear and non- our aim here is to show that the dual method immedi-
linear programs. The dual approach for integer, fixed charge trans- ately gives a solution to Phase II, thereby yielding a so-
portation problems is fully discussed by Erlenkotter (1978). lution to the joint cost allocation problem.

250 MANAGEMENTSCIENCE/Vol. 42, No. 2, February 1996


BALACHANDRAN AND RAMAKRISHNAN
Joint Cost Allocationfor Multiple Lots

Phase II: The Joint Cost Allocation the objective function ZDP (since no wi1'sare present in
Suppose a solution to Phase I, solving the ex-ante de- ZDP). Thus we have the "Condensed Dual" (CD) given
cision problem of buying lots at minimal cost, has been below, which has eliminated the m x n constraints given
obtained, i.e., the uj's are optimally solved, for the IP in (8).
given by (1)-(5). We can then use the one-dimensional n
"single lot" methods considered for joint cost allocation CD: Maximize ZDP= V, (6)
by Jensen (1977), Hamlen et al., (1977, 1980), and Bal- i=l

achandran and Ramakrishnan (1981) (BR) to solve the n

Phase II problem. We use the rule that "each lot's cost Subject to I MaxIO,vi - Ii>)c JCj,
i=l
must be entirely paid for by the products using that lot."
Sets of products such that all members of a given set j= 1,...,m. (10)
use only one source (lot) can then be formed. Then the
Here wijhas been set equal to Max(O, vi - Iij>.Note that
cost allocation methods of the one lot problem, i.e., the
this program CD has only n variables since we elimi-
one we propose here, or the Shapley Value can be used
nated m x n variables wij. Further, it has only m con-
on a set of decomposed single lot problems. Our prop-
osition here is that the solution method used in Phase I straints, since m x n constraints given by (8) have been
dropped. Good solutions for DP are relatively easier to
offers an economically justifiable allocation mechanism.
find via CD, even though (10) is piece-wise linear in v,.
We show here that the ex-post allocation mechanism
based on the optimal decisions of Phase I lies in the core.
Further, the two problems have been integrated into one 4. Solution to the Problem
model, thus linking the two objectives simultaneously. From any dual optimal solution vi, we can form a pri-
mal integer solution through:
The Dual Method
n
We now present a solution method used for solving IP.
The "fastest" solution technique has been by dual based uj = O, if , MaxIO, vi - Iij < JCj, (11)
i=l
methods. The dual for LP (1)-(4) will be formulated
n
with the relaxed constraint set (5') instead of (5). Let vi, uj = 1, if , maxIO, vi - Ii) = JCj. (12)
i = 1, . . ., n be the dual variables of constraint set (2), i=l
and wij, i = 1, .,. . n; j = 1, . . ., m, be the dual variables
for constraint set (3). Then the dual program for the LP Let J* -j Iuj = 11. Then, set (13)
is I 1, if Iij = minkeJ Iij, and
n
l0 otherwise.
DP: Maximize ZDP= V,, (6)
(14)
n We can also form the following complementary slack-
Subject to Xwij c? JCj, forj = 1, .., m, (7) ness conditions connecting the dual and the primal pro-
i=l
grams:
iv'-wij c Ii>, fori=1,...,nandj=1, ..., m, (8) (uj - xi) maxIO, vi - Iij = 0,
wij 2 O, fori =,...,nandj = 1, ...,m. (9) i = 1, ..., n; j=1,...,m, (15)
Since the objective function contains only vi's and be- n
cause of the special structure in constraints (7) and (8), uj1[c1 - Y maxIO, vi-Iij) =,j=1,...,m. (16)
it is easy and computationally efficient to solve the dual
program, (DP). We can reduce the wij's to their lowest Equation (16) will be always satisfied if the solutions of
possible (nonnegative) values without violating con- (11) to (14) are used. Equation (15) will not be satisfied
straints (8) and (9) to achieve the maximum value for if more than one j E J*has vi > Iij,for any i. There will

MANAGEMENT SCIENCE/Vol. 42, No. 2, February 1996 251


BALACHANDRAN AND RAMAKRISHNAN
Joint Cost Allocationfor Multiple Lots

be a dual gap as ZDP < ZIp,and ?7 discusses this situ- This is so, since the optimal primal objective value
ation. Until ?7 we will assume that there is no dual gap. (which is source costs plus the incremental costs) is
equal to the optimal dual objective value, 1n=, v*, by the
Dual Program Solution Methods
duality theory.
Though the dual program can be solved by regular
The economic justification is that v* is the cost of hav-
mathematical programming methods, as the problem
ing constraint i in (2) of LP, which is =1xij = 1. The
has a special structure, we can use a simple, easy-to-
constraint means that we have to make product i fully.
understand process to obtain the optimal solution. Er-
So product i must incur that cost. All the properties enu-
lenkotter (1978) has used the dual ascent and dual ad-
merated by Moriarity (1975) are satisfied. Let J*be the
justment procedures. For full details the reader is re-
set of open sources. Let j(i) be such that
ferred to Erlenkotter (1978). Here we shall describe only
the dual ascent procedure as it helps us in the solution
4ij(i)= Min Iij =si. (17)
to the allocation problem and also since we make use jeJ
of it in the example.
The sum of the allocated joint costs in each group using
The Dual Ascent Procedure. As the objective func- the same source is the cost of that source,
tion contains only vi's, they must be increased to their
maximum values subject to constraint (10). The vi's can E [Vi Si,] = JCk,
i;j(i)=k
be initiated to the lowest value of Iijin each column of
Table 1. The left-hand side of (10) will be zero as no Iij from the dual constraint (10) by definition.
is > vi. The vi's will then be sequentially increased to The most important property of the allocation we
the next higher value in each column so that constraint want to consider is the question of stability. If the allo-
(10) will not be violated, or to the highest value possible. cation scheme offers no incentives for the divisions to
We cycle through the columns until none of the vi's can break out and form subgroups or coalitions then our
be increased. This is the dual ascent procedure. The dual scheme will be stable. This is precisely the game-
adjustment procedure further improves the solution or theoretic concept of core. Let N denote the index set of
alters it to reduce the complementary slackness viola- j = 1, ... , n. An allocationiv is in the coreif the total cost
tions of (15). allocated to any coalition S C N is less than or equal to
Erlenkotter (1978) and Morris (1977) have demon- the costs that coalition S will incur if it goes alone. We
strated that in this type of problem, almost in all cases can prove that our dual based allocation is in the core.
the LP solution itself is an integer solution. From the
theory of duality, at the optimum, Z DPof the dual pro- THEOREM1. Cost allocationsbasedon any dual optimal
gram is equal4 to Z LP.But if the primal solution for LP solution v* are in the core.
turns out to be integers in uj, Z DP= Z I in (1). Since we
PROOF. See the appendix.
have solved the DP given by (6) and (10) we have v*,
Thus we see that dual solution based allocation
the optimal solution to the dual program. The dual so-
schemes are economically defensible, and are not sub-
lution v* facilitates the joint cost allocation problem as
ject to the criticisms of sterility and arbitrariness in al-
one scheme we can use is:
locations discussed by Thomas (1974). However, there
Rule: v* is the cost allocated to product i. is a problem, as the dual solution to the problem may
not be unique. This occurs because the primal solution
The total costs will be fully allocated with this scheme. can be highly degenerate with many basic variables tak-
ing zero values. There may be whole space of optimal
dual solutions. To find an unique solution that avoids
4In this section, we will assume that the dual optimal solution leads
to a primal feasible integer solution. If this is not true, then the game "arbitrariness"for the allocation problem, we can make
theoretic properties that we derive for the allocations may not hold. use of the "propensity to contribute" factor suggested
This situation is shown for some cases in ?7. by Balachandran and Ramakrishnan (1981).

252 MANAGEMENTSCIENCE/Vol. 42, No. 2, February 1996


BALACHANDRAN AND RAMAKRISHNAN
JointCost Allocationfor Multiple Lots

5. Allocation Procedures b. cannot be greaterthan b', as division i can always use


An alternate set of allocation methods have been pro- another open source to get the next best incrementalcost.
posed in accounting literature without using the core This is analogous to using I,r (the alternatesource cost) in
concept. These methods produce unique allocations and Louderback's1976 method, for the one-lot problem.
are based on the concept of imputations (see Shubik Also, if biincreased excessively for product i, it might
1991, p. 141, for a summary). The core may not exist in be optimal for the division to consider buying an un-
many problems (see ?7 for some cases), but imputations used source after paying the fixed cost of that source.
always exist as they guarantee only that no individual The dual procedure can be used to check this. Source j
division has an incentive to use another source in any will be bought if the vi's change such that the slack on
proposed allocation. But there are many imputation so- constraint (10), viz.,
lutions, and several authors in management accounting n

literature (e.g., Hamlen et al. 1977, 1980; Jensen 1977) iC1 - max(O, vi - Ii),
i=l
have suggested the use of alternate minimum cost as a
basis for allocation. becomes zero. We can first use the base value si's for
For each product let bi be the alternate cost that it will vi's and obtain the slack on each unused source. Thus,
incur if it does not use the source assigned to it based for each unopened sourcej i J*,let,
on the minimum independent cost. Let si denote the in- qij= Ij + JCj- S max(O, Sk - Ikjl. (19)
cremental cost Iij incurred by product i in the optimal k*i

solution as defined in (17). For product i, si is the base Division i will suggest opening source j if the allocated
value for allocated costs, and it also has to take a share cost is increased to qij.We have to find the minimum of
of the fixed cost of the source it is using in the final all such values over all j * J (analogous to finding the
allocation. Let the share of the source costs be ri so that second-best among open sources), so that
the final allocation ai is si + ri. As enunciated in Bala-
14= Min(qij). (20)
chandran and Ramakrishnan (1981) for the single lot j0r
case, the propensity to contribute (PTC) is then defined
The maximum alternate cost bi must be the minimum
by:
of 14and bq so that product i will not switch to some
PTCi other open source or go out and try to open a new
source. Then biQmin(I4, bq). Let H(i) be the set of prod-
(Propensityto) ucts using the source of product i, denoted by j(i). So
contributeJ H(i) = (k:j(k) = j(i)). Within groups we can use the
.bi Si (18) PTC's derived from bi using (18) as the weighting fac-
tors so that the final allocation is:
_ ( Least cost by \ (Incremental\
alternate methodsJ V cost a=
bi- si
Hbk - S) + Si.
JC1(i) (21)
-k C=HU) ( k SO)
We can use the PTCs as the weighting factors within
We can easily show5 that lkeH(i) (bk - Sk) 2 JCj(i).So from
each group. The difficulty in applying this method in
(21), ai c bi. As bi = min(I4, bq), bi c bqand bi C 14. The
the multiproduct case is that the least cost by alternate
methods is difficult to characterize. The main purpose
of this section is to propose a definition of this cost. Let For any dual solution, vi, as there is no dual gap, from (15), vi
j(i) be the lot assigned to product i as defined in (17) b?. The primal solution assignment (11) imply vi < Iij + JCj
and J* be the set of open sources. Let bq be the incre- -Xk*i maxi0, Vk - Ikjl for every unopened source j. From (20), bK= Iij
mental processing cost using the second-best open + JCj - -k*-imaxi0, Sk - IkjI for some unopened source j. V1k 2 Sk, the
lowest cost among open sources as otherwise Vk can be increased with-
source for product i, i.e.,
out violating any dual constraints. This implies vi _ bi and v,i bi.
b= Min (IIi. From (12), for the open sources the dual constraint is binding and Sk
jEJ*,j*j(i) - Ikj(i) for all k i H(i). So XkEH(,) (Vk - Sk) = JCj(i) < XkEH(i) (bk - Sk).

MANAGEMENTSCIENCE/Vol. 42, No. 2, February 1996 253


BALACHANDRAN AND RAMAKRISHNAN
Joint Cost Allocationfor Multiple Lots

Table2 the lowest independent costs for each product. Table 3


shows the dual solution for the first three iterations:
IncrementalProcessing Costs Ij
Producti
The dual ascent procedure is complete. The currentso-
Lot
Lot # Cost lution is also optimal. Let J*= (2, 3, 4) and assign products
j JCj 1 2 3 4 5 6 7 1, 2, 4, and 7 to source 2, products 3 and 5 to source 3, and
product 6 to source 4 (open market). Since among open
1 50 60 180 100 00 100 20 105 sources J*,only one source is available through the condi-
2 60 55 110 00 80 100 30 100
tions vij - Iij> 0, for all i, the complementarityslackness
3 70 65 co 50 100 40 40 0

4 0 80 160 120 140 50 25 170 conditions (15) and (16) are satisfied. Hence we have a pri-
mal optimal solution. The minimum independent costs are

si = 55 110 50 80 40 25 100 for i = 1,...,7.


allocated cost is less than the independent cost of the sec- Allocation of Costs
ond-best open source for each i, and the minimum cost The maximum allocations that can be charged to each
incurred to open a new source. This PTC based allocation division i based on open set (2, 3, 4) are:
scheme a is an alternative to the core allocation. To sum-
marize, the essential differences are:(1) the core allocation V = 65 160 120 100 50 30 170 for i = 1, ...,7.
may not exist in some situations (see ?7) while a always
Using the si's we can calculate qil, the maximum slack
exists, (2) the core allocation may not be unique while a is
of the only closed source, j = 1, as:
unique, and (3) core satisfies group rationalitywhile a sat-
isfies only individual rationality. qil = Ij + 50 - I maxI0, Sk - Ikl = bK.
k*i
Illustration of the Allocations
The following example illustrates the dual method and The maximum allocations that can be charged to each
the allocation procedure. Consider the data in Table 2 product based on closed sets are:
associated with symbols of Table 1 for three lots and = 105 225 145 oo 145 70 150 for i = 1, ...,7.
bK
seven products.
Source 4 corresponds to the alternative of directly Note that it is optimal only for division 7 to go for
buying from the outside market. A cost of ooimplies that source 1 if source 2 is not available by offering to pay
the product concerned cannot be served by the resource 45 for the joint costs. Division 7 will be willing to pay
mentioned in the row. Further, the underlined costs are the rest of the joint cost 5 as it benefits by going to

Table3

Slack = JCj- Dualsolution v',


max10,Sk -Ikji
Iteration k*i 1 2 3 4 5 6 7

1 50 55 110 50 80 40 20 100
60
70 '-' = 455
2 45 60 160 100 85 50 25 100
0
10 v i= 580
3 35 60 160 110 85 50 25 100
0
0 ,i= 590

254 MANAGEMENTSCIENCE/Vol. 42, No. 2, February 1996


BALACHANDRAN AND RAMAKRISHNAN
Joint Cost Allocationfor Multiple Lots

source 1. bi, the maximum charge is the minimum of Table4


b7 and b,, i.e.,

bi = 65 160 120 100 50 30 150 for i = 1, ...,7. Lot# Lot Cost


i Jjc 1 2 3
Hence the propensity to contribute PTCi = bi - si
1 100 50 50 0
PTCi = 10 50 20 50 for i = 1, 2, 4, 7 2 100 50 0 50
3 100 0 50 50
PTCi = 70 10 fori = 3,5 4 210 40 40 40
PTC6= 5
For source 2 we have I PTCi = 130 and r, = 10 60
= 4.6 and so on, to get ri = (4.6, 23.1, 9.2, 23.1) for i equitable full cost allocation rule is to allocate 110 to
= 1, 2, 4, 7. For source 3 we have L PTCi = 80 and ri each product. But this allocation is not in the core, as any
= (61.3, 8.7) for i = 3, 5 and r6 = 0. So the final allo- two divisions can acquire an appropriate source and
service themselves for a total cost of 200 while they are
cated costs are
charged 220 in the joint cost allocation procedure. It is
ai = si + ri = (59.6, 133.1, 111.3, 89.2, 48.7, 25, 123.1) clear that there is no cost allocation that is in the core
for this example.
for 1, .. .,7.
We will analyze the existence of core solutions in' a
specialized setting where all the products are identical.
7. Dual gap and the existence of In that case, a source's cost structure depends only on
core solutions its capacity expressed in the number of products that it
In ?4 we assumed that the optimal dual solution leads can serve. Let s (1 c s c n) denote the capacity and
to an integer primal solution that attains the same ob- JC(s) denote the joint cost of a source that can serve a
jective value. From the theory of duality, the optimal maximum of s products at an incremental cost of I(s)
dual objective value will equal the optimal objective for each product that it serves. Without loss of gener-
value of only for the unconstrained primal program ality we will assume super additivity, i.e., for 1 c k c s,
(not restricted only to integer solutions). While in many
iC(s) + sI(s) c [JC(k)+ kI(k)I
applications the dual gap may not arise (see Tables II-
VI in Erlenkotter 1978), the interesting question is + [JC(s - k) + (s - k)I(s - k)]. (22)
whether we can obtain stablefull cost allocation schemes
if there is a dual gap. In this section, we will first illus- If super additivity is not satisfied, then the source of a
trate this situation with an example and derive a general given size s never needs to be considered and can be
result connecting the dual gap and the existence of core replaced by two sources of capacity k and s - k with
allocation solutions. Consider the following example in lower total costs. Note that the cost structure described
Table 4. in Table 4 satisfies super additivity. The Iij matrix of
It is easy to see that the optimal dual solution is vi Table 1 can be completed as follows: for sources of size
= 100 for i = 1,2,3, and the optimal dual objective value s, the independent cost Iij is I(s) for s products and
is L vi = 300. The constraints (10) are binding for j = 1, equals m for the rest of n - s products. With this cost
2,3, but not for] =4. The corresponding primal solution structure we can show the following.
is uj =forj = 1, 2, 3 and ai =only if Iij< oofor j = 1,
THEOREM 2. Forthecaseof identicalproducts,a corefull
2, 3. Note that the primal objective value Ej JCj1u costallocationruleexistsif andonlyif thereis no dualgap.
+ Ei Ej Iijaijalso equals 300. But the optimal primal in-
tegersolution is to open and use only source 4 for all the The intuition behind the theorem can be seen by ob-
products. The primal objective value is 330, and there is serving that, as the cost function is super additive, the
a dual gap. As all the products are identical, the only primal optimal solution is to choose the largest capacity

MANAGEMENTSCIENCE/Vol. 42, No. 2, February 1996 255


BALACHANDRAN AND RAMAKRISHNAN
Joint Cost Allocationfor Multiple Lots

source of size n. The total cost will be JC(n) + nI(n). costs, etc., outlined in the introduction, may be compro-
The full cost allocation to each product is vi4 = JC(n)/ mised.
n + I(n). Is this allocation a feasible solution in the dual
program? If for sources of size s, v I(s), then con-
straints (10) of the dual program will not be binding. If
8. Conclusion
We have provided a two-phase model which simulta-
for sources of size s, vi > I(s), then constraint (10) is: neously handles the acquisition-assignment decision of
multiple lots (ex-ante) in Phase I and the joint cost al-
s(JC(n) + I(n) - I(s)) c JC(s). location (ex-post), given the optimal selection of lots in
n
Phase I. Utilizing the optimal dual solutions of the lot
Combining these two cases, we can see that there will selection problem, a joint cost allocation mechanism
be no dual gap if and only if based on the concept of propensity to contribute is pro-
vided. This allocation is shown to be in core if there is
JC(n) Cs no dual gap. For the case of identical products, we also
)+I(n) )+ I(s) V s, 1 c s c n. (23)
n s show that if there is a dual gap, then there is no core
But note that this is also the condition for a coalition of full cost allocation rule.
size s not to break away from the whole group and ac- The existing literatureessentially has considered differ-
quire a source of size s only for the coalition. The allo- ent cost allocation schemes for a single lot case. Here we
cated cost to the coalition s(JC(n)In + I(n)) should be consider multiple lots and prove stability, a concern that
less than JC(s) + sI(s), the cost for the coalition alone. does not arise with one lot. Further,these cost allocation
Since (23) implies that allocation v* is in the core, the schemes are done on an ex-post basis, while here we dis-
theorem is proved. cuss ex-antedecisions as well. Severalcriteriafor evaluating
For the situation with symmetrical products, the es- cost allocation schemes are discussed in the literature,but
sential condition is (23), which says that the product the emphasis has been on (1) rationalityand fairness and
costs computed using the full costing method must be (2) neutralitywith respect to decision making (Hamlen et
decreasing as the capacity increases. If this is true, then al. 1977, p. 616). However, most of these models consider
there will be no dual gap, and there are cost allocations the product/purchase decision independent of the cost al-
that are in the core. If there is a dual gap and if the firm location problem. Our model permits the simultaneous
wants to preserve stability, then the firm must subsidize analysis of ex-ante producdiondecision and ex-post allo-
joint resource acquisition. The minimum subsidy cation mechanisms. Further,our allocations satisfy all the
needed is the dual gap. criteriaenunciatedearlierand those of Moriarity(1975).We
Various economic rationale for allocating common combine the richness of duality theory, the stability of the
costs (relating to pricing for usage to reduce congestion core, the incentive compatibilityrequirementsof game the-
costs, etc.,) are provided in Banker et al. (1988), Miller ory, and the propensity to contributeaspects appealing to
and Buckman (1987). Cohen and Loeb (1990) show that fairness.Finally a numericalillustrationis provided, based
firms facing price competition will bid as if they are on the algorithm presented in the paper.6
planning to allocate fixed costs completely ex-ante. 6
The authors would like to thank the referee and the editor for their
These papers assume that one agent, i.e., the central
valuable comments.
headquarters, can decide on the allocation scheme and
ignore the motivation of the individual divisions to ac-
Appendix
cept the cost allocation. The results of this section com-
plement the result of those papers as we show that a THEOREM1. Allocationsbasedon any dual solution are in the core.
subsidy (equal to the dual gap) will be required if tidy
PROOF. For any allocation v to be in the core, no coalition should
allocation methods are to be used. But once the firm have any incentives to break away and incur lower total costs. Let S
starts subsidizing common resource acquisition, then all denote any subset of products and let c(S) be the minimum cost that
the advantages of decentralization, accurate product the coalition will incur, if it acquires a separate set of resources. Sup-

256 MANAGEMENTSCIENCE/Vol. 42, No. 2, February 1996


BALACHANDRAN AND RAMAKRISHNAN
Joint Cost Allocationfor Multiple Lots

pose that when coalition S is alone it acquires the set of resources Cooper, R., "Does Your Company Need a New Cost System," in B.
denotedby J. Then, Brinker (Ed.), Emerging Practices in Cost Management, Warren,
Gorham & Lambert, Boston and New York, 1990.
c(S) = E MinIIij+ X JCj. (24) Crow, R. T., "An Approach to the Allocation of Common Costs of
ies Ji-I jl
Multi-Mission Systems," Naval Res. Logistics Quarterly, 19, 3
Once the set of resources J is bought, each product will be produced (1972), 431-448.
from the source in J that has lowest incremental cost Iij.The first term Demski, J. S., D. Sappington, and P. T. Spiller, "Incentive Schemes with
in (24) is the total incremental processing cost and the second term is Multiple Agents and Bankruptcy Constraints," J. EconomicTheory,
the total source cost. Allocation v will be in the core only if 44 (1988), 156-167.
Dumaine, B., "Is Big Still Good?", Fortune,April 20, 1992, 50-60.
c(S)X v vi VSCN, (25) Erlenkotter, D., "A Dual-Based Procedure for Uncapacitated Facility
ieS
Location," Oper. Res., 26 (1978), 992-1099.
where the vi's are the allocated costs based on a dual allocation and Fuchsberg, G., "Decentralized Management Can Have Its Draw-
N is the set of all products. backs," WallStreetJ., December 9, 1992.
As the dual solutions satisfy constraints (1), we have Gangolly, J. S., "On Joint Cost Allocation: Independent Cost Propor-
tional Scheme (ICPS) and Its Properties," J. AccountingRes., 19, 2
Y, MaxI0, vi -Iij} JC; forj =1,. M.m (1981), 299-312.
ieN
Hamlen, S., W. Hamlen, and J. Tschirhart, "The Use of Core Theory
Adding these constraints for j E J, in Evaluating Joint Cost Allocation Schemes," Accounting Rev.,
July (1977), 616-627.
X X Max(O, vi - Iij jICj. (26) , and , "The Use of Generalized Shapley Allocation
iGN je
je1
in Joint Cost Allocation," Accounting Rev., April (1980), 269-
Let ](i) be defined such that, 287.
Hughes, J. S. and J. H. Scheiner, "Efficiency Properties of Mutually
= Min(IiiJ.
Ii,j-(i) (27) Satisfactory Cost Allocations," Accounting Rev., 55, 1 January
jel
(1980), 85-95.
Then from (26) we have, Jensen, D. L., "A Class of Mutually Satisfactory Allocations," Account-
ing Rev., October (1977), 842-856.
X Cjc2 Kaplan, R. S. and G. L. Thompson, "Overhead Allocation Via Mathe-
E (Max(O, i- + E_ MaxO0, i- Iij)
jel ieN jel,j*j(i) matical Programming Models," Accounting Rev., 46 (1971), 352-
364.
X Max(O, vii,- I,())I E Max(O, vi - I,j(j) since S C N
and U. Welam, "Overhead Allocation with Imperfect Markets
ieN ieS
and Nonlinear Technology," Accounting Rev., July (1974), 477-

ieS
vi - I Ii,j(j)
ieS
X vi - I Min(Iij from (27), 484.
ieS ies JeJ
Loehman, E. T. and A. B. Whinston, "An Axiomatic Approach to Cost
which proves that vi's are in the core from (24) and (25). C Allocation for Public Investment," Public Finance Quarterly,Oc-
tober (1976), 315-328.
References Louderback, J. G., "Another Approach to Allocating Joint
Atkinson, A. A., "Intra-Firm Cost and Resource Allocation: Theory Costs: A Comment," Accounting Rev., July (1976), 683-
and Practice," ResearchMonograph,Canadian Academic Account- 685.
ing Association, Toronto, Ontario, Canada, 1987. Miller, B. L. and A. G. Buckman, "Cost Allocation and Opportunity
Balachandran, B. V. and R. T. S. Ramakrishnan, "JointCost Allocation: Costs," ManagementSci., 33 (1987), 626-639.
A Unified Approach," AccountingRev., January (1981), 85-96. Moriarity, S., "Another Approach to Allocating Joint Costs," Account-
, L. Li, and R. Magee, "On the Allocation of Fixed and Variable ing Rev., October (1975), 791-795.
Costs from Service Departments," ContemporaryAccounting Res., , "Another Approach to Allocating Joint Costs: A Reply," Account-
Fall (1987), 166-185. ing Rev., July (1976), 686-687.
Banker, R. D., S. M. Datar, and S. Kekre, "Relevant Costs, Congestion Morris, J. G., "On the Extent to Which Certain Fixed-Charge Depot
and Stochasticity in Production Environments," J. Accountingand Location Problems Can Be Solved by LP," J. Oper.Res. Society,29,
Economics,10 (1988), 171-197. 1 (1977) 71-76.
Business Week, "A Big Company that Works," May 4, 1992, 123- Rajan, M. V., "Cost Allocation in Multi Agent Settings," Accounting
132. Rev., August (1992), 527-545.
Cohen, S. I. and M. Loeb, "Implicit Cost Allocation and Bid- Roth, A. and R. Verrechia, "The Shapley Value as Applied to Cost
ding for Contracts," Management Sci., 6, 9 (1990), 1133- Allocation: A Reinterpretation," J. AccountingRes., Spring (1979),
1138. 295-303.

MANAGEMENT SCIENCE/Vol. 42, No. 2, February 1996 257


BALACHANDRAN AND RAMAKRISHNAN
Joint Cost Allocationfor Multiple Lots

Shubik, M., "GameTheoryin the SocialSciences,"MIT Press, Cambridge, , "The Allocation Problem, Part Two," Studies in AccountingRes.,
MA, 1991. 9, American Accounting Association, 1974.
Stigler, G. S., "The Theory of Price," 3rd Edition, Macmillan, New Tumney,P. B., "Using Activity-Based Costing to Achieve Manufacturing
York, 1966. Excellence," in B. Brinker (Ed.), EmergingPracticesin Cost Manage-
Thomas, A. L., "The Allocation Problem in Financial Accounting The- ment, Warren, Gorham & Lambert,Boston and New York, 1990.
ory," Studies in Accounting Res., 3, American Accounting Associ- Zimmerman, J., "The Cost and Benefits of Cost Allocation," Accounting
ation, 1969. Rev., July (1979), 504-521.

Acceptedby BruceMiller;formerDepartmentalEditor;receivedFebruary1991. This paperhas beenwith the authors 19 monthsfor 1 revision.

258 MANAGEMENTSCIENCE/Vol. 42, No. 2, February 1996

Das könnte Ihnen auch gefallen