See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/223678491

A Global Supply Chain Model With Transfer Pricing and Transportation

Cost Allocation

Article  in  European Journal of Operational Research · February 2001

DOI: 10.1016/S0377-2217(99)00431-2

CITATIONS READS
315 5,712

2 authors:

Carlos J. Vidal Marc Goetschalckx

Universidad del Valle (Colombia) Georgia Institute of Technology
47 PUBLICATIONS   2,044 CITATIONS    79 PUBLICATIONS   5,785 CITATIONS   

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Carlos J. Vidal on 25 April 2016.

The user has requested enhancement of the downloaded file.

 European Journal of Operational Research 129 (2001) 134±158
www.elsevier.com/locate/dsw

Theory and Methodology

A global supply chain model with transfer pricing and

transportation cost allocation
Carlos J. Vidal a, Marc Goetschalckx b,*

a
Universidad del Valle, Cali, Colombia
b
School of Industrial and Systems Engineering, 765 Ferst Drive, Georgia Institute of Technology, Atlanta, GA 30332-0205, USA
Received 1 April 1998; accepted 1 October 1999

Abstract

We present a model for the optimization of a global supply that maximizes the after tax pro®ts of a multinational
corporation and that includes transfer prices and the allocation of transportation costs as explicit decision variables.
The resulting mathematical formulation is a non-convex optimization problem with a linear objective function, a set of
linear constraints, and a set of bilinear constraints. We develop a heuristic solution algorithm that applies successive
linear programming based on the reformulation and the relaxation of the original problem. Our computational ex-
periments investigate the impact of using di€erent starting points. The algorithm produces feasible solutions with very
small gaps between the solutions and their upper bound (UB). Ó 2001 Elsevier Science B.V. All rights reserved.

Keywords: Distribution; Modeling; Production; Global supply chain optimization; Transfer pricing

1. Introduction

The design and management of global supply chains are nowadays one of the most active research topics
in global logistics. Vidal and Goetschalckx (1997) identify several lacking features and opportunities for
research in the methodology for the strategic and tactical design of global logistics systems. Much of the
research ignores relevant international factors such as transportation mode selection, the allocation of
transportation cost among subsidiaries, the inclusion of inventory costs as part of the decision problem, the
explicit inclusion of suppliers, and the nonlinear e€ects of international taxation. Additionally, many global
supply chain models assume that transfer prices are ®xed and given.
A transfer price is the price that a selling department, division or subsidiary of a company charges for a
product or service supplied to a buying department, division, or subsidiary of the same ®rm (Abdallah,

*
Corresponding author. Tel.: +1-404-894-2300/2317; fax: +1-404-894-2301.
E-mail address: marc.goetschalckx@isye.gatech.edu (M. Goetschalckx).

0377-2217/01/$ - see front matter Ó 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 3 7 7 - 2 2 1 7 ( 9 9 ) 0 0 4 3 1 - 2
 C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158 135

1989). Transfer Pricing (TP) is one of the most controversial topics for multinational companies (MNCs).
According to O'Connor (1997), ``TP is the most important international tax issue facing multinationals
today, and is expected to remain so for the near future.'' In his survey of over 200 MNCs, 80% of them
identi®ed TP as the number one issue they have to face. Most researchers on global logistics have con-
sidered TP a typical accounting problem rather than an important decision opportunity that signi®cantly
a€ects the design and management of a global supply chain. In general, when a logistics analyst attempts to
determine the optimal ¯ows of products among facilities, the price of a product is almost always considered
a given parameter. However, this is not the case in real global logistics systems since management can
determine the transfer price with some degree of ¯exibility within given limits.
Undoubtedly, the TP problem is much more than an accounting problem. TP policies have signi®cant
e€ects on performance evaluation and motivation of subsidiary managers. Abdallah (1989) remarks that
making TP decisions for MNCs is an important, complex, ¯exible, and complicated task because it a€ects
other major functions of the ®rm such as marketing, production, location, transportation, and ®nance.
The impact of TP policies on taxable income, duties, and management performance is signi®cant. Ac-
cording to Nieckels (1976), small changes in transfer prices may lead to signi®cant di€erences in the after-
tax pro®t of a company. On the other hand, the arbitrary manipulation of transfer prices (as presented by
Cohen et al., 1989) is currently under careful observation by tax authorities and is strictly penalized. De-
spite these limitations and based on current regulations, in most of the cases, companies have some range of
values for their transfer prices. According to Stitt (1995), the Organization for the Economic Cooperation
and Development (OECD) de®nes a range of an acceptable transfer price rather than a single `correct'
transfer price. Abdallah (1989) comments on the diculty of using market prices as the selected TP policy
in an international environment, mainly because of the diversity of economic environments and trans-
portation costs. Moreover, there will always exist a tradeo€ between the low transfer price desired by the
buyer division, and the high transfer price desired by the seller division.
Several researchers have addressed the TP problem as an integral component of the optimization of a
global supply chain. Nieckels (1976) presents a nonlinear mathematical model to determine optimal
transfer prices and resource allocation in a multinational textile ®rm. His formulation includes transfer
prices as decision variables and a linear objective function for maximizing the global net income after taxes.
One limitation of this model is that it assumes that the company has a central distribution center from
which all products are distributed to the sales subsidiaries. A second shortcoming of this model is that it
does not include Bill of Materials (BOM) constraints, although these are signi®cant to the real system.
Instead, the problem is formulated as a raw material based model using some transformations of products
into raw materials. Finally, the author does not address the decision of allocating transportation costs since
they are always charged to the destination subsidiary.
Nieckel's solution approach begins with a heuristic procedure that assigns initial values to the transfer
prices, equal to either their lower or their upper bound (UB). The remaining program is a linear program
that is solved for the optimal ¯ows. Given these ¯ows, the set of transfer prices becomes variable again. To
®nd the new set of transfer prices, a systematic heuristic procedure based on the sign of the derivative of the
objective function with respect to the transfer prices is then applied. The solution method iterates between
the optimization of the linear program and the heuristic procedure to change transfer prices until no further
increase in the objective function is possible. According to the author, when the solution procedure stops, a
local optimum is found. Although mentioned for further research, Nieckels does not attempt to calculate an
UB on the solution to the problem in order to test the performance of his heuristic.
Cohen et al. (1989) present a preliminary formulation of a normative model that is a dynamic, nonlinear
mixed integer programming formulation. The model is nonlinear due to the inclusion of transfer prices and
decision variables to allocate overhead costs to plants. The objective function considers the maximization of
the ®rm's after tax pro®t. Transfer prices are treated as a markup for each product, applied on the pro-
duction cost plus the shipping cost, which includes tari€s. This constitutes a shortcoming of this model since
136 C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158

in most cases the tari€s depend on the value of the transferred products and in the paper they are con-
sidered given as a part of the shipping cost per unit. The solution approach is a hierarchical process that
iterates between solving mixed integer programs to determine optimal ¯ows and supplier contracts, and
optimal markups. This process is repeated until a local optimum has been reached. Since the markups do
not have an upper bound, the resulting solution may generate a strong reduction of taxes that would not be
acceptable to tax authorities.
No computational experience is presented in this paper, but the authors state that some variants of the
model have been successfully developed by them and/or other researchers. One of these is the work by
Cohen and Lee (1989), who describe a single-period multicommodity model that analyzes resource de-
ployment decisions faced by a personal computer manufacturer. It is important to note that this is the only
model that considers the decision of allocating transportation costs to either the origin or the destination by
formulating binary variables. However, in the presented solution process, this decision is made externally
and the corresponding binary variables are ®xed in the model.
In one of the most comprehensive models to date, Arntzen et al. (1995) present a multi-period, multi-
commodity mixed integer program to optimize the global supply chain at the Digital Equipment Corpo-
ration. The objective function considers the minimization of variable production costs, inventory costs,
shipping costs, ®xed production and production `style' costs, minus the savings from duty drawbacks and
duty relieves. All these terms are weighted by a factor a. The objective function also contains production
time and transportation time terms, weighted by a factor …1 ÿ a†. It is not clear how the model manages two
di€erent units of measure in the objective function beyond a user assigned weight. The TP decision problem
is not included in the formulation. In addition, taxes are only indirectly considered as a part of variable
production costs. The authors report on the use of `nontraditional methods', such as elastic constraints,
row factorization, cascaded problem solution, and constraint-branching enumeration. However, it is not
clear from the paper exactly which solution method has been applied.
More recently, Canel and Khumawala (1997) propose a mixed-integer single-product model for the
optimization of a global supply chain. The authors include TP decisions in their analysis, but they ®x the
transfer prices to either their lower or UB before solving the model. Additionally, transportation costs are
allocated to the destination subsidiary. As a consequence, this model does not include the TP problem and
the allocation of transportation costs as part of the decision process.
The explicit inclusion of global suppliers is ignored in most of the models described above. Furthermore,
inventory costs and their impact on the decision of selecting a transportation mode, which is very important
in an international scenario, are not considered in most of them.
This paper presents a tactical global supply chain model that extends previous research by simulta-
neously considering TP, transportation cost allocation, inventory costs, and their impact on the selection of
international transportation modes. Additionally, we provide an integrated and ecient procedure to
obtain near-optimal solutions and their UB to realistic cases. In Section 2 we present some general con-
siderations about TP. In Section 3 we describe the basic characteristics of the model, and in Section 4 we
illustrate the solution approach. Section 5 contains some computational experiments, and ®nally, in Section
6 we present some conclusions and insights for further research.

2. Basic considerations on transfer pricing

The determination of transfer prices is one of the most dicult problems that MNCs have to face. The
problem is to ®nd adequate transfer prices so that the global corporate goals are satis®ed and the per-
formance measures are fair for all the ®rm's subsidiaries and divisions (Eccles, 1985). According to
O'Connor (1997), the problem originates from the con¯icts between the general goals of the global cor-
poration and the internal goals of its subsidiaries, and from the constraints imposed by the legal envi-
 C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158 137

ronment involving taxes and tari€s. The utilization of the arm's length principle, which establishes the use
of transfer prices as if the involved parties did not belong to the same corporation, is widely accepted.
Accordingly, Mueller et al. (1997) state that most current TP policies are based on either external market
prices or internal costs. Both have the advantage of being acceptable to tax authorities. The discussion of
the methods used by MNCs to determine their transfer prices is beyond the scope of this paper. For a
comprehensive explanation, see Belkaoui (1985), Eccles (1985), Abdallah (1989), and/or O'Connor (1997).
MNCs seek diverse objectives when setting their transfer prices. O'Connor (1997) presents the results of
several surveys concerning the key goals that MNCs try to achieve through their TP policies. In general, the
results of these surveys are not consistent since some of them give more importance to some objectives,
while others disregard the same goals. In summary, these objectives are to:
· Satisfy tax and other legal requirements;
· Produce pro®t maximization by minimizing worldwide taxes and duties;
· Produce a fair framework for performance evaluation and motivation of all related parties;
· Move funds internationally;
· Minimize exchange rate risk;
· Avoid exchange controls and quotas; and
· Increase pro®t share from joint ventures.
One of the most controversial objectives of TP policies is pro®t maximization by minimizing worldwide
taxes and duties. O'Connor (1997) states that although many companies argue that tax minimization is not
their main goal, in reality many ®rms appear to design TP policies to take advantage of di€erential tax
structures among countries. In spite of the current regulations imposed by most countries intended to avoid
the arbitrary manipulation of transfer prices, there exist signi®cant opportunities for pro®t maximization
that are not illegal. Very likely, the ¯exibility is decreasing each day, but the impact of small changes in
transfer prices on the global pro®t of a company may be substantial. Therefore, designing TP policies to
minimize taxes legally should not be ignored by MNCs that prefer to design policies consistent with the
corporate strategy and organizational structure of the company. Probably the only way to analyze such an
impact is using mathematical models like the one described in this paper.

2.1. A basic single-product example with two subsidiaries

This example illustrates the fundamental ideas for the further development of the model. Fig. 1 shows
the system under consideration. Subsidiary A, the producer and seller, located in country A, manufactures
a single product that sells to subsidiary B, located in a di€erent country B.
The following notation will be used in this example and in the next section (for the sake of simplicity, we
always use capital letters for parameters and lower case letters for decision variables):

Parameters
C transportation cost ($/unit)
D import duty charged in country B based on the value of the transferred product ˆ 12%
FA ®xed costs (other expenses) at subsidiary A ˆ $20,000/unit of time
FB ®xed costs (other expenses) at subsidiary B ˆ $120,000/unit of time
M market price of the product in country B ˆ $20/unit
Tl lower bound (LB) on the transfer price ˆ $11/unit
Tu UB on the transfer price ˆ $13/unit
TAXA corporate tax rate in country A ˆ 34%
TAXB Corporate tax rate in country B ˆ 50%
138 C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158

V variable costs for producing the product in subsidiary A ˆ $7/unit

XMAX maximum volume (or demand) of the product ˆ 20,000 units/unit of time
Variables
p the proportion of transportation costs allocated to subsidiary A ‰0 6 p 6 1Š
t the transfer price of the product between the two subsidiaries [$/unit]
x the ¯ow of the product between the two subsidiaries [units/unit of time]

Table 1 presents typical calculations to determine the net income after tax (NIAT) for each of the
subsidiaries, and for the global corporation as a whole. Abdallah (1989) and O'Connor (1997) present
similar tables for analyzing the impact of the change of a transfer price on the NIAT of a company. Note
that when the transfer price is increased by $1/unit to $12/unit, the NIAT increases. However, an additional
increase in the transfer price of $1/unit to $13/unit reduces the NIAT since subsidiary B loses money.
Evidently, there exists an optimal transfer price between $11/unit and $13/unit.
Fig. 2 illustrates the corporate tax rate function we consider for all countries in this paper. No tax credit
for losses is included in the initial analysis. However, the model can handle tax credits if they are assumed to
occur in the same time period under analysis. When the corporate tax rate function is more complex, that is,
when it has more `steps' and breakpoint values, additional analysis is necessary. However, for large
companies the function considered here is realistic and valid in most countries.
To derive the general function of the global NIAT in this example, we should consider four disjoint
regions for the net income before tax (NIBT): both subsidiaries make a pro®t, only subsidiary A makes a
loss, only subsidiary B makes a loss, and both subsidiaries make a loss. The way to obtain these expressions
is to constrain NIBT for both subsidiaries. The general expression obtained for the NIAT of the global
corporation is shown below.

Global NIAT
8
>
> 0:1tx ‡ 5:38x ÿ 73; 200 if …t ÿ 7†x ÿ 20; 000 P 0 and …20 ÿ 1:12t†x ÿ 120; 000 P 0;
>
>
>
< ÿ0:46tx ‡ 15:38x ÿ 133; 200 if …t ÿ 7†x ÿ 20; 000 P 0 and …20 ÿ 1:12t†x ÿ 120; 000 < 0;
ˆ
>
> 0:44tx ‡ 3x ÿ 80; 000 if …t ÿ 7†x ÿ 20; 000 < 0 and …20 ÿ 1:12t†x ÿ 120; 000 P 0;
>
>
>
:
ÿ0:12tx ‡ 13x ÿ 140; 000 if …t ÿ 7†x ÿ 20; 000 < 0 and …20 ÿ 1:12t†x ÿ 120; 000 < 0:

Fig. 3 shows the global NIAT vs. the transfer price t for di€erent values of ¯ow x. Obviously, for the NIAT
to be bounded, there must exist an UB on the ¯ow x from A to B. This bound is determined by a limited
capacity of one or the two subsidiaries, by a limited demand, and/or by other ¯ow constraints imposed on

Fig. 1. The basic single-product case with two subsidiaries.

 C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158 139

Table 1
Net income after tax for the basic example
Detail Subsidiary A Subsidiary B Global corporation
(Transfer price t ˆ $11/unit)
Sales
20,000 @ $11/unit 220,000
20,000 @ $20/unit 400,000 400,000
Variable costs ($7/unit) 140,000 140,000
Procurement costs 220,000
Import duties (12%) 26,400 26,400
Fixed costs 20,000 120,000 140,000
Net Income Before Tax (NIBT) 60,000 33,600 93,600
Taxes (34% and 50%) 20,400 16,800 37,200
Net Income After Tax …NIAT † 39,600 16,800 56,400

(Transfer price t ˆ $12/unit)

Sales
20,000 @ $12/unit 240,000
20,000 @ $20/unit 400,000 400,000
Variable costs ($7/unit) 140,000 140,000
Procurement costs 240,000
Import duties (12%) 28,800 28,800
Fixed costs 20,000 120,000 140,000
Net Income Before Tax (NIBT) 80,000 11,200 91,200
Taxes (34% and 50%) 27,200 5,600 32,800
Net Income After Tax …NIAT † 52,800 5,600 58,400

(Transfer price t ˆ $13/unit)

Sales
20,000 @ $13/unit 260,000
20,000 @ $20/unit 400,000 400,000
Variable costs ($7/unit) 140,000 140,000
Procurement costs 260,000
Import duties (12%) 31,200 31,200
Fixed costs 20,000 120,000 140,000
Net Income Before Tax (NIBT) 100,000 ÿ11,200 88,800
Taxes (34% and 50%) 34,000 0 34,000
Net Income After Tax …NIAT † 66,000 ÿ11,200 54,800

Fig. 2. Tax rate function.

140 C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158

Fig. 3. Global net income after tax function.

the system. Notice that the optimal solution to the problem depends on both, the ¯ow and the bounds on
the transfer price: It may be equal to the LB, or to the UB, or may lie in between them.
Since the NIAT depends on the ¯ow and on the transfer price, it is not practical to develop a model
based on the functions shown in Fig. 3. A more general approach that considers all the disjoint regions,
similar to the method presented by Nieckels (1976) and Cohen et al. (1989), is the following. We de®ne two
new variables:

A ˆ Net income before tax …subsidiary A† ‰$=unit of timeŠ;

B ˆ Net income before tax …subsidiary B† ‰$=unit of timeŠ:

The NIBT may take a negative (loss), zero, or a positive (pro®t) value. Therefore, A and B are free vari-
ables, that can be substituted by:

A ˆ A‡ ÿ A ÿ ; where A‡ P 0 and Aÿ P 0;
B ˆ B ‡ ÿ Bÿ ; where B‡ P 0 and Bÿ P 0:

Note that the NIAT of the global corporation is the sum of the remainder after taxes of the NIBT of all
subsidiaries, minus the sum of the losses of all subsidiaries, provided …A‡ ; Aÿ † and …B‡ ; Bÿ † are not positive
at the same time. For the model developed here, this is true and will be shown below. Therefore, we can
formulate the following optimization problem:

…P1† Maximize NIAT ˆ 0:66A‡ ÿ Aÿ ‡ 0:5B‡ ÿ Bÿ

 C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158 141

subject to

tx ÿ 7x ÿ A‡ ‡ Aÿ ˆ 20; 000;
ÿ 1:12tx ‡ 20x ÿ B‡ ‡ Bÿ ˆ 120; 000;
11 6 t 6 13;
0 6 x 6 20; 000;
A‡ ; Aÿ ; B‡ ; Bÿ P 0:

It is very important to note that the pairs of variables …A‡ ; Aÿ † and …B‡ ; Bÿ † cannot be positive at the same
time in any optimal solution. To show this, suppose there exists an optimal solution such that A‡ ˆ k1 > 0,
and Aÿ ˆ k2 > 0. There are three possible cases: (a) Suppose k1 > k2 . Since there is only one constraint in
which A‡ and Aÿ appear, it is possible to rede®ne a new feasible solution with A‡ ˆ k1 ÿ k2 > 0, and
Aÿ ˆ 0, and all other variables the same as before. It is easy to observe that the new value of the objective
function is greater than the original one as long as there is a non-negative tax rate for the subsidiary.
Therefore, the original solution cannot be optimal, which is a contradiction. For the case (b) k1 ˆ k2 and (c)
k1 < k2 , the proof is similar, by de®ning new feasible solutions with A‡ ˆ 0; Aÿ ˆ 0, and with
A‡ ˆ 0; Aÿ ˆ k2 ÿ k1 > 0, respectively. Hence, this property avoids the contradiction of having a pro®t and
a loss in a subsidiary at the same time.
Problem (P1) can be linearized by de®ning a new variable y ˆ tx, which represents the amount in dollars
transferred from subsidiary A to subsidiary B through the ¯ow x, using the transfer price t. With this
substitution, we obtain the following problem:
…P2† Maximize NIAT ˆ 0:66A‡ ÿ Aÿ ‡ 0:5B‡ ÿ Bÿ

subject to

y ÿ 7x ÿ A‡ ‡ Aÿ ˆ 20; 000;
ÿ 1:12y ‡ 20x ÿ B‡ ‡ Bÿ ˆ 120; 000;
11x 6 y 6 13x;
0 6 x 6 20; 000;
A‡ ; Aÿ ; B‡ ; Bÿ P 0:

Evidently an optimal solution to Problem (P2) will also be optimal to Problem (P1). This is so since
given optimal values y  and x , we can always ®nd an optimal transfer price t ˆ y  =x , provided x > 0. In
the general case x may be equal to zero, but the transfer price here is meaningless since there is no ¯ow
from A to B, and it will not be calculated by the algorithm developed below. The unique optimal solution to
Problems (P1) and (P2) is the following (see Fig. 3):

NIAT  ˆ $59; 400=unit of time;

y  ˆ $250; 000=unit of time;
x ˆ 20; 000=units=unit of time;
t ˆ y  =x ˆ $12:5=unit;
A‡ ˆ $90; 000=unit of time
…all other variables are equal to zero†:
142 C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158

2.2. Transportation cost allocation

In this paper, the transfer price ± called `basic' transfer price ± is considered to be independent of the
transportation cost. The basic transfer price depends on the technique being used. For example, a cost-plus
method is appropriate to determine the set of transfer prices of components between internal suppliers and
manufacturing plants. Market prices can be adequate for the set of transfer prices of ®nished products
shipped from manufacturing plants to distribution centers (DCs). In any case, the transportation cost al-
location can be modeled independently of the basic transfer price. The transportation cost allocation re-
¯ects the terms of the transaction, which are more precisely de®ned as INCOTERMS. Some of these are,
for example, free on board (FOB), free alongside ship (FAS), and cost, insurance and freight (CIF).
The transportation cost allocation may be de®ned as p ˆ the proportion of transportation costs allo-
cated to the origin (subsidiary A for the basic example), where 0 6 p 6 1. If p ˆ 0, then the destination
subsidiary pays for all the transportation costs. On the other hand, if p ˆ 1, then the origin subsidiary
assumes all the transportation costs. It is possible to obtain unique optimal solutions with 0 < p < 1, and
therefore it is appropriate to model p as a continuous variable. As an illustration and following the same
notation as in the basic example, Table 2 shows general expressions for the NIAT for the case when duties
are paid on the CIF value, assuming both subsidiaries are pro®table.

3. The model

Fig. 4 illustrates the system under consideration. The suppliers are classi®ed into two groups: internal
suppliers and external suppliers. For the external suppliers there is no possible decision with respect to
transfer prices since they sell directly to the company using market prices. For the internal suppliers a set of
optimal transfer prices for the period under consideration is to be determined by the model. The manu-
facturing plants receive components and raw materials from the suppliers, and perform assembly opera-
tions to obtain ®nished products. Then, the plants send ®nished products to DCs using a second set of
transfer prices, which also is to be determined by the model. Finally, the ®nished products are distributed
worldwide from the DCs to retailers and/or ®nal customers.
The complete formulation of the model is very detailed and lengthy, since it requires the de®nition of sets
for all the logistics objects (suppliers, manufacturing plants, DCs, and customers), transportation channels
between the logistics objects, capacity constraints, ¯ow conservation constraints, and cost de®nitions.
Many of the constraints are common in supply chain models and therefore the complete model has not
been included in this paper but it is given in Vidal and Goetschalckx (1998). A verbal formulation, as-
sumptions, and constraints which are unique to this model are given next.
The assumptions of the proposed model can be divided into two classes. The ®rst class contains standard
supply chain modeling assumptions. The detailed list of these assumptions can be found in Vidal and
Goetschalckx (1998) and they are not repeated here. The second class of assumptions deals with transfer
prices and taxation. These assumptions of the model are the following:
· All internal suppliers, plants and DCs are considered to be subsidiaries of the US parent company and
are supposed to be actively involved in manufacturing, selling, shipping, and servicing activities. There-
fore, the Subpart F income of the Internal Revenue Code does not apply here, since all the income of the
subsidiaries is considered active income (for a more detailed description of these concepts, see Glautier
and Bassinger, 1987; Radebaugh and Gray, 1993; Bodner, 1997). As a consequence, the deferral principle
applies, that is, the income is not taxed to the parent company until the dividends are received by the
shareholders.
· Each internal supplier, plant, and distribution center is assumed to be taxed on their local-source income,
except for the parent company, which is subject to tax on its worldwide income. The company attempts to
Table 2
General expression for the NIAT when duties are paid on the CIF valuea
Detail Subsidiary A Subsidiary B Global corporation
Sales
x @ $t/unit tx
x @ $M/unit Mx Mx
Procurement costs tx
Transp. cost Cpx C…1 ÿ p†x Cx
Other variable costs Vx Vx
Import duties D…t ‡ C†x D…t ‡ C†x
Fixed costs FA FB FA ‡ FB
NIBT …t ÿ Cp ÿ V †x ÿ FA f‰M ÿ …1 ‡ D†CŠ ÿ …1 ‡ D†t ‡ Cpgx ÿFB ‰M ÿ C ÿ V ÿ D…t ‡ C†Šx ÿ…FA ‡ FB †
Taxes …TAXA †…t ÿ Cp ÿ V †x …TAXB †f‰M ÿ …1 ‡ D†CŠ ÿ …1 ‡ D†t ‡ Cpgx …TAXA †…t ÿ Cp ÿ V †x
ÿTAXA FA ÿTAXB FB ‡…TAXB †f‰M ÿ …1 ‡ D†CŠ ÿ …1 ‡ D†t ‡ Cpgx
ÿ…TAXA FA ‡ TAXB FB †
NIAT …1 ÿ TAXA †…t ÿ Cp ÿ V †x …1 ÿ TAXB †f‰M ÿ …1 ‡ D†CŠ ÿ …1 ‡ D†t ‡ Cpgx …1 ÿ TAXA †…t ÿ Cp ÿ V †x
ÿ…1 ÿ TAXA †FA ÿ…1 ÿ TAXB †FB ‡…1 ÿ TAXB †f‰M ÿ …1 ‡ D†CŠ ÿ…1 ‡ D†t ‡ Cpgx
ÿ‰…1 ÿ TAXA †FA ‡ …1 ÿ TAXB †FB Š
a
It is assumed in the above table that both subsidiaries make a pro®t.
C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158
143
144 C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158

Fig. 4. Global logistics system modeled.

maximize the total pro®t after tax in all the countries where it operates, but no income is remitted as a
dividend to the shareholders, and therefore the parent company is only taxed on its local income. Further
re®nements of the model may consider withholding taxes and the reception of dividends from subsidiaries.
· The structure of the organization and its degree of vertical integration allow for the centralization of de-
cisions about the transfer prices.
· Customer demand may or may not be completely satis®ed. This may generate positive slack demand
variables in an optimal solution, especially when the company wants to analyze the impact of the unex-
pected absence of key internal suppliers and/or the impact of producing and selling non-pro®table prod-
ucts.
· All transfer price variables have a lower and an upper bound, which re¯ect feasible markups for produc-
tion costs and a pro®t margin, or for possible discounts from market prices.
· As a requirement for the justi®cation of transfer prices to tax authorities, all the transfer prices from a
given origin and for a given component (or ®nished product) must be the same for all the destinations.
The transfer price, of course, does not include the transportation cost, which is modeled separately. This
can be thought of as the transfer price being determined based on production costs, and therefore there is
no reason to allow it to be di€erent for di€erent buyers. The allocation of the transportation cost to the
origin or destination is a decision variable that is considered independently of the transfer price. If trans-
fer prices are allowed to be di€erent, the problem is easier to solve as we show below.
· Import duties are paid by the importing country, based on the FOB value of the transferred products, or
based on the CIF value, as appropriate. No export duties are considered.

Verbal formulation of the model

MODEL P…x; t; v; p†
Maximize global after tax pro®t (given in dollars for the time period under analysis)
ˆ After tax pro®t of internal suppliers
+After tax pro®t plants
+After tax pro®t of DCs
Subject to Expressions for the NIBT of internal suppliers, plants, and DCs
Suppliers' capacity (internal and external suppliers)
Production capacity at plants
 C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158 145

Customer demand constraints

Bill of materials at plants and balance constraints at DCs
Minimum pro®t for internal suppliers, plants and DCs (optional)
Bounds on transfer prices and general bounds on decision variables

The after tax pro®t of suppliers, plants and distribution centers are included in the objective function
with the following generic form
X 
…1 ÿ TAXi †ibts‡ ÿ
i ÿ ibtsi :
i

Here TAXi is the corporate tax rate (%) of the home country of logistics object i; ibtsi is the net income
before tax of logistics object i, which can be a supplier, manufacturing plant, or distribution center. The
dimensions are ($/unit of time).
The net income variables are free variables, since the net income before tax may be negative, zero, or
positive. Therefore, each of these variables is replaced by the di€erence between a ``plus'' non-negative
variable (pro®t variable) and a ``minus'' non-negative variable (loss variable).
The net income before tax of logistics objects such as plants and DCs is computed similarly to the net
income before tax of DCs located in countries where duties are charged on the CIF value, which is given in
the following constraint:
X X X 1  X X X  1  
MPRICElp wklmp ÿ HANDCkp ‡ TRCWMklm Wp wklmp
l2C…k† m2T …k;l† p2P
El l2C…k† m2T …k;l† p2P
Ek
X X X VPkp H h  
p i
ÿ TTWMklm ‡ …CSF †SHIPFREQklm ‡ SSFWkp TTWMklm wklmp
l2C…k† m2T …k;l† p2P
Ek
X X X 1  

ÿ tppldcjp …1 ‡ DUTYjkp † ‡ …1 ÿ propwjkm ‡ DUTYjkp †TRCPWjkm Wp xjkmp
j2M m2T …j;k† p2P …j†
Ej
 
1
ÿ FIXDCk ˆ ibtwc‡ ÿ
k ÿ ibtwck ; k 2 W c:
Ek
The net income before tax of logistics objects is computed similarly to the net income before tax of DCs
located in countries where duties are charged on the FOB value, which is given by:
X X X 1  X X X  1  
MPRICElp wklmp ÿ HANDCkp ‡ TRCWMklm Wp wklmp
l2C…k† m2T …k;l† p2P
El l2C…k† m2T …k;l† p2P
Ek
 h
X X X VPkp H p i
ÿ TTWMklm ‡ …CSF †SHIPFREQklm ‡ SSFWkp TTWMklm wklmp
l2C…k† m2T …k;l† p2P
Ek
X X X 1  

ÿ tppldcjp …1 ‡ DUTYjkp † ‡ …1 ÿ propwjkm †TRCPWjkm Wp xjkmp
j2M m2T …j;k† p2P …j†
Ej
 
1
ÿ FIXDCk ˆ ibtwfk‡ ÿ ibtwfkÿ ; k 2 W f :
Ek

The minimum reasonable pro®t of logistics object k can have a LB dictated by external factors and this is
modeled by the following type of constraint:
ibtwc‡
k P MINPROn :
146 C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158

The upper and LBs on transfer prices are modeled by the following type of constraints:

TPSUPLlir 6 tpsuplir 6 TPSUPLuir ; i 2 S 0 ; r 2 R…i†;

l u
TPPLDCjp 6 tppldcjp 6 TPPLDCjp ; j 2 M; p 2 P …j†:

Bounds on the proportions for allocating transportation costs are modeled by the following type of con-
straints:

0 6 prospijm 6 1; i 2 S 0 ; j 2 M…i†; m 2 T …i; j†;

0 6 propwjkm 6 1; j 2 M; k 2 W ; m 2 T …j; k†:

For the sake of clarity, we present the sets, indexes, parameters, and decision variables next.

Sets and indexes

C set of market zones, indexed by l
C…k† set of market zones that can be served by distribution center k 2 W
M set of manufacturing plants, indexed by j
P set of ®nished products, indexed by p
P …j† set of products that can be produced in manufacturing plant j; P …j†  P
T set of transportation modes, indexed by m
T …j; k† set of available transportation modes between plant j 2 M and distribution center
k 2 W ; T …j; k†  T
T …k; l† Set of available transportation modes between distribution center k 2 W and
market zone l 2 C…k†; T …k; l†  T
W set of DCs, indexed by k
Wc set of DCs located in countries where duties are charged on the CIF value of
imported products; W c  W
Wf set of DCs located in countries where duties are charged on the FOB value of
imported products; W f  W ; W c [ W f ˆ W ; W c \ W f ˆ ;
Parameters (Note: For convenience and ease of reading, all parameters are de®ned using capital letters)
DUTYjkp import duty rate on the value of ®nished product p 2 P …j†, shipped from plant
j 2 M to distribution center k 2 W
Ek exchange rate of country of distribution center k 2 W [monetary units of the
respective country/dollar]
FIXDCk ®xed cost of distribution center k 2 W [monetary units country of distribution
center k per unit of time]
H holding cost given in $/($. units of time) (units of time consistent with those of the
average transportation time parameters de®ned below) [in general, given in $/$. year]
HANDCkp handling cost of ®nished product p 2 P in distribution center k 2 W [monetary
units country of distribution center k/unit of p]
MINPROn minimum ``reasonable'' pro®t of subsidiary n (internal supplier, plant, or
distribution center) [monetary units of the corresponding country/unit of time]
(These parameters are to be used judiciously by management)
 C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158 147

MPRICElp market price or selling price of ®nished product p 2 P in market zone l 2 C

[monetary units country of market zone l/unit of p]
SHIPFREQijm frequency of shipments of raw materials from internal supplier i 2 S 0 to plant
j 2 M…i†, using transportation mode m 2 T …i; j† [units of time (in general, days)]
SHIPFREQklm frequency of shipments from distribution center k 2 W to customer zone l 2 C…k†,
using transportion mode m 2 T …k; l† [units of time (in general, days)]
SSFWkp safety stock factor of ®nished product p 2 P at distribution center k 2 W
l u
TPPLDCjp ; TPPLDCjp LB and UB on the transfer price of ®nished product p 2 P …j†, shipped from
plant j 2 M to any distribution center in W [monetary units country of plant j/
unit of p]
TPSUPLlir ; TPSUPLuir LB and UB on the transfer price of raw material r 2 R…i†, shipped from internal
supplier i 2 S' to any plant in M…i† [monetary units country of supplier i/units of r]
TRCPWjkm transportion cost per weight unit (not including duties) of ®nished products
shipped from plant j 2 M to distribution center k 2 W , using transportion mode
m 2 T …j; k† [monetary units country of plant j/weight unit]
TRCWMklm transportion cost per weight unit (not including duties) of ®nished products
shipped from distribution center k 2 W to market zone l 2 C…k†, using transpor-
tation mode m 2 T …k; l† [monetary units country of distribution center k/weight
unit]
TTPWjkm average transportation time from plant j 2 M to distribution center k 2 W , using
transportion mode m 2 T …j; k† [units of time]
TTWMklm average transportation time from distribution center k 2 W to market zone
l 2 C…k†, using transportation mode m 2 T …k; l† [units of time]
VPkp inventory value of ®nished product p 2 P , given in monetary units of the country of
distribution center k 2 W per unit of p
Wp weight of a unit ®nished product p 2 P [weight units/unit of p]

Decision variables (Note: For convenience and ease of reading, all decision variables are de®ned using small
letters.)
ibtwck net income before tax of distribution center k 2 W c [$/unit of time]
ibtwfk net income before tax of distribution center k 2 W f [$/unit of time]
Note: The above ®ve sets of variables are free variables, since the net income before tax may be
negative, zero, or positive. Therefore, each of these variables is replaced by the di€erence
between a plus non-negative variable (pro®t variable) and a minus non-negative variable (loss
variable). For example, the set of free variable ibtwcj …j 2 M c † generates the sets of
non-negative variables ibtwc‡ c ÿ c
j …j 2 M † and ibtwcj …j 2 M †].
propwjkm proportion of transportation costs of ®nished products shipped from plant j 2 M to
distribution center k 2 W , using transportation mode m 2 T …j; k†, allocated to plant j
prospijm proportion of transportion costs of raw materials shipped from internal supplier i 2 S 0 to
plant j 2 M…i†, using transportation mode m 2 T …i; j†, allocated to internal supplier i
tppldcjp transfer price of ®nished product p 2 P …j†,shipped from plant j 2 M to any distribution
center in W (not including transportation costs) [monetary units country of plant j/unit of p]
wklmp amount of ®nished product of p 2 P shipped from distribution center k 2 W to market zone
l 2 C…k†, using transportation mode m 2 T …k; l† [units of p/unit of time]
xjkmp amount of ®nished product p 2 P …j†, produced at plant j 2 M and shipped to distribution
center k 2 W , using transportation mode m 2 T …j; k† [units of p/unit of time]
148 C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158

4. Solution approaches

4.1. General structure of the problem

Clearly, Problem P…x; t; v; p† is a non-convex optimization problem with a linear objective function, a set
of linear constraints, and a set of bilinear equalities. The problem has the following general structure:

P…x; t; v; p† Maximize d T0 v

subject to

cTr x ‡ d Tr v ‡ xT Ar t ‡ xT B r p ˆ f r ; r ˆ 1; 2; . . . ; m;
Cx 6 b;
T l 6 t 6 T u;
0 6 p 6 1;
x P 0; t P 0; v P 0;

where Ar ; Br …r ˆ 1; 2; . . . ; m† and C are the matrices of coecients; b the right-hand side vector of ¯ow-
related constraints; cr …r ˆ 1; 2; . . . ; m†; d r …r ˆ 0; 1; 2; . . . ; m† the vectors of coecients; f r …r ˆ 1; 2; . . . ; m†
the ®xed costs at internal suppliers, plants, and DCs; p the vector of proportions to allocate transportation
costs; t vector of transfer prices; T l ; T u the LB and UBs on transfer prices; v vector of pro®t and loss
variables; x is the vector of material ¯ows.
Problem P…x; t; v; p† can be transformed into an inde®nite quadratic problem with bilinear constraints by
replacing some variables from the bilinear equalities in the objective function. Therefore, according to
Pardalos and Vavasis (1991), it is NP-hard.
Speci®c variations of this problem have been widely studied. For example, the classic bilinear problem
has a bilinear objective function with linear constraints represented by two separable polyhedra in x and y
(Konno, 1976; Gallo and Ulk  ucu, 1977; Vaish and Shetty, 1977; Sherali and Shetty, 1980; Al-Khayyal and
Falk, 1983; Judice and Faustino, 1991). The jointly constrained bilinear program is a variation of the latter,
in which the constraints are linear but cannot be separated into two disjoint polyhedra in x and y (Al-
Khayyal, 1990; Sherali and Alameddine, 1992). The more general case when the convexity of the feasible set
is relaxed is a more dicult problem to solve (Al-Khayyal, 1990). An example of this case is a general
bilinear problem having bilinear constraints, as shown in Al-Khayyal (1992). This problem has received less
attention in the literature, and has been more intensively studied for speci®c cases, namely the pooling
problem (Floudas and Visweswaran, 1990; Visweswaran and Floudas, 1990; Floudas and Aggarwal, 1990;
Visweswaran and Floudas, 1993), and some applications in farm management (Bloemhof-Ruwaard and
Hendrix, 1996). Other researchers have addressed the problem indirectly by analyzing more general
problems, such as polynomial programming problems (Sherali and Tuncbilek, 1992), general quardratic
programs (Al-Khayyal et al., 1995), and general constrained non-convex problems (Floudas and Vis-
weswaran, 1993; Ben-Tal et al., 1994; Androulakis et al., 1995).
To the best of our knowledge all the computational results reported in the above research correspond to
relatively small instances of the problems. However, in global supply chain models, we usually face medium
to large-scale optimization problems for which none of these global optimization approaches appear to
work satisfactorily. For this reason, and given the structure of the problem, an optimization-based heuristic
procedure using successive LP solutions has been developed. Its description is given next.
 C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158 149

4.2. A successive LP solution procedure

4.2.1. Substitution of proportion variables

It is important to note the proportion of transportation costs, represented by the variables
propwjkm and prospijm , are allowed to be di€erent from a speci®c origin to di€erent destinations, using a
given transportation mode. There exists considerable ¯exibility to de®ne the fraction of the transportation
cost paid by the subsidiary and the terms of each transaction can be di€erent. As a consequence, the bilinear
terms in x and p can be linearized by using the following substitutions:

X
prospijm Wr Sijmr ˆ zijm ; i 2 S 0 ; j 2 M…i†; m 2 T…i; j†; …1†
r2R…i†\R…j†

X
propwjkm Wp xjkmp ˆ zjkm ; j 2 M; k 2 W ; m 2 T…j; k†: …2†
p2P …j†

The rationale for the de®nition of these variables comes from the fact that the proportions to allocate
transportation costs do not depend on the speci®c component or ®nished product since these costs are
expressed per unit of weight. Using this substitution, we transform the problem into the following for-
mulation:

P…x; t; v; z† Maximize d T0 v

subject to

cTr x ‡ d Tr v ‡ xT Ar t ‡ gTr z ˆ f r ; r ˆ 1; 2; . . . ; m;
Cx 6 b;
T l 6 t 6 T u;
z ÿ Ex 6 0;
x P ; t P 0; v P 0; z P 0;

where the new set of constraints z ÿ Ex 6 0 corresponds to the transformed bounds on the proportions for
allocating transportation costs (see constraints 0 6 p 6 1 in Problem P…x; t; v; p††, namely:
X
0 6 zijm 6 Wr sijmr ; i 2 S 0 ; j 2 M…i†; m 2 T…i; j†; …3†
r2R…i†\R…j†

X
0 6 zjkm 6 Wp xjkmp ; j 2 M; k 2 W ; m 2 T…j; k†: …4†
p2P …j†

4.2.2. Rede®nition and substitution of TP variables

We can relax the limitation that all transfer prices must be the same from each origin to all destinations
for a given raw material (or ®nished product) by rede®ning the transfer price decision variables as follows:
150 C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158

tpsuplijr transfer price of raw material r 2 R…i†, shipped from internal supplier i 2 S 0 to plant j 2 M…i†
(not including transportation costs) [monetary units country of supplier i/unit of r],
tppldcjkp transfer price ®nished product p 2 P …j† shipped from plant j 2 M to distribution center k 2 W
(not including transportation costs) [monetary units country of plant j/unit of p].

By doing this, we are now able to make the following substitutions to linearize the remaining nonlinear
terms existing in the bilinear constraints of problem P…x; t; v; z†:
X
tpsuplijr sijmr ˆ yijr ; i 2 S 0 ; j 2 M…i†; r 2 R…i† \ R…j†; …5†
m2T …i;j†

X
tppldcjkp xjkmp ˆ yjkp ; j 2 M; k 2 W ; p 2 P…j†: …6†
m2T …j;k†

The rationale for the de®nition of these variables is similar to that of the substitution of the proportion
variables, because the transfer prices ± as de®ned in this paper ± do not depend on the transportation mode
since the transportation costs are considered separately. Therefore, the summation may be made over all
available transportation modes. It is important to observe that the new lower and UBs on the transfer
prices also have three indexes according to this rede®nition. Since we are interested in developing a new
problem that is equivalent to the original Problem P…x; t; v; p†, we need to rede®ne the bounds on transfer
prices as:

TPSUPLlijr ˆ TPSUPLlir ; i 2 S 0 ; r 2 R…i† \ R…j† 8j 2 M…i†;

TPSUPLuijr ˆ TPSUPLuir ; i 2 S 0 ; r 2 R…i† \ R…j† 8j 2 M…i†;

i l
TPPLDCjkp ˆ TPPLDCjp ; j 2 M; p 2 P …j† 8k 2 W ; …7†
u u
TPPLDCjkp ˆ TPPLDCjp ; j 2 M; p 2 P …j† 8k 2 W :

Moreover, for the problem P…x; y; v; z† to be equivalent to the original problem P…x; t; v; p†, we need to add
constraint of the form xT F q y ˆ 0; q ˆ 1; 2; . . . ; h, which ensure that all transfer prices from a given origin to
all destinations for a given raw material (or ®nished product) are equal. More precisely, these constraints
are the following:

yijn r yijn‡1 r
P ˆ P ;
m2T …i;jn † Sijn mr m2T …i;jn‡1 † Sijn‡1 mr

i 2 S0; r 2 R…i† \ R…j†; jn and jn‡1 2 M…i†; n ˆ 1; 2; . . . ; j M…i† j ÿ1;

…8†
yjkn p yjkn‡1 p
P ˆ P ;
m2T …j;kn † xjkn mp m2T …j;kn‡1 † xjkn‡1 mp

j 2 M; p 2 P …j†; kn and kn‡1 2 W ; n ˆ 1; 2; . . . ; j W j ÿ1

for all possible combinations of jn and jn‡1 , and kn and kn‡1 . These constraints are to be relaxed so that we
can obtain an UB on the original problem P…x; t; v; p†. Hence, we obtain the following problem:
 C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158 151

P…x; y; v; z† Maximize d T0 v

subject to

cTr x ‡ d Tr v ‡ eTr y ‡ gTr z ˆ f r ; r ˆ 1; 2; . . . ; m;

Cx 6 b;
Dl x 6 y 6 Du x;
z ÿ Ex 6 0;
xT F q y ˆ 0; q ˆ 1; 2; . . . ; h …constraints to be relaxed†
x P 0; y P 0; v P 0; z P 0:

The new constraints Dl x 6 y 6 Du x come from the original bounds on transfer prices in problem
P…x; t; v; z†, and are given by:
X X
TPSUPLlijr Sijmr 6 yijr 6 TPSUPLuijr Sijmr ; i 2 S 0 ; j 2 M…i†; r 2 R…i† \ R…j†;
m2T …i;j† m2T …i;j†
X X …9†
l u
TPPLDCjkp xjkmp 6 yjkp 6 TPPLDCjkp xjkmp ; j 2 M; k 2 W ; p 2 P…j†:
m2T …j;k† m2T …j;k†

The relaxation of problem P…x; y; v; z† provides an UB on the original problem P…x; t; v; p†. Let us call this
relaxed problem PR …x; y; v; z†. This UB will be utilized for testing the performance of the overall procedure,
whose description is given next.

4.2.3. The successive LP solution procedure

If we ®x the set of variables x in problem P…x; t; v; z†, then the problem becomes linear in t, and vice
versa. Thus, we can iterate by successively ®xing one set of variables and solving the remaining LP for the
other set. The process can be terminated when the change in the objective function value is negligible. Fig. 5
summarizes the overall procedure. We use the notation P…x; t; v; z j x† and P…x; t; v; z j t† to denote Problem
P…x; t; v; z† with ®xed ¯ows and with ®xed transfer prices, respectively. The diculty in solving the original
problem P…x; t; v; p† to optimality is due to its multi-extremality characteristic, that is, the existence of
multiple local optima. Consequently, the heuristic solution obtained by the procedure is highly dependent
on the starting point. In what follows, we test the performance of diverse starting points instances of the
problem.

5. Computational experience

The procedure described in the previous Section has been implemented using AMPL connected with
CPLEX. All the experiments have been done in an IBM RS6000 model 590 with 512 MB of RAM. We have
conducted extensive computational experiments using diverse starting points and two sets of instances of
di€erent sizes. All instances have been carefully generated to approximate real instances as much as pos-
sible. The main characteristics of these instances are shown in Table 3. In what follows, we describe each of
the starting points we have tested.
152 C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158

Fig. 5. A successive LP solution procedure for the TP problem.

Table 3
Main characteristics of the small and medium instances solved
Characteristics Small instances Medium instances
Total number of suppliers 11 50
Number of internal suppliers 3 12
Number of manufacturing plants 3 8
Number of distribution centers 8 10
Number of customer zones 20 80
Number of components 10 35
Number of ®nished products 5 12
Average transportation modes/arc 2.8 3.1
Number of decision variablesa 1544 10,100
Number of constraintsa 521 2926
a
The number of decision variables and constraints shown correspond to problem PR …x; y; v; z† after preprocessing.

5.1. Optimal ¯ows from the relaxed problem (Opt_Flows)

In this case we solve problem PR …x; y; v; z† and take the optimal set of ¯ows as the starting point of the
procedure. It is important to note that these ¯ows are feasible to the original problem P…x; t; v; z† and
therefore a good behavior is expected.
 C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158 153

5.2. Optimal transfer prices from the relaxed problem (Opt_TP)

Assuming that the optimal solution to problem PR …x; y; v; z† contains at least a set of transfer prices that
are not feasible, we use the following expressions to determine a suitable set of feasible transfer price to start
the process:
P
j2MMS…i†:r2RRM …j† ysuplijr
tpsuplir ˆ P P ; i 2 SINT ; r 2 RRS; …i†
j2MMS…i†m2TSM …i;j†:r2RRM…j† Sijmr
P …10†
k2w ypldcjkp
tppldcjp ˆ P P ; j 2 M; p 2 PPM…j†
k2W m2TMW …j;k† xjkmp

It is important to note that these expressions yield a weighted average of all di€erent transfer prices, so that
the total amount transferred from the given origin remains constant. If there is no ¯ow from some origins,
then we select the LB on the transfer price to start the process.

5.3. Heuristic rule to determine initial transfer prices (Heu_TP)

Reducing the expression for the NIAT of the global corporation in Table 2, we ®nd that one of its terms
is equal to ‰…1 ÿ TAXA† ÿ …1 ÿ TAXB†…1 ‡ D†Štx. This suggests that if the expression in the brackets is
greater than zero, then the transfer price should be set to its UB; otherwise, it should be set to its LB.
Nieckels (1976) presents the same heuristic to start his solution procedure. Of course, the optimal value
depends on other factors such as ¯ows, but this simple rule is worth implementing in the general procedure.
Speci®cally, we apply the following rules for setting the starting transfer prices in this case:
X  
For all i 2 SINT ; r 2 RRS…i† : if …1 ÿ TAXi † ÿ …1 ‡ DUTYijr †…1 ÿ TAXj † > 0;
j2MMS…i†:r2RRM …j†

then let tpsuplir ˆ TPSUPLuir ; otherwise let tpsuplir ˆ TPSUPLlir ;

X 
For all j 2 M; p 2 PPM…j† : if …1 ÿ TAXj † ÿ …1 ‡ DUTYjkp †…1 ÿ TAXk † > 0;
k2W
u l
then let tppldcjp ˆ TPPLDCjp ; otherwise let tppldcjp ˆ TPPLDCjp :

5.4. LBs, UBs, and middle point(LB_TP, UB_TP, and Mid_TP)

We have also examined the behavior of the procedure when setting all the transfer prices either to their
LB or to their UB. In addition, we allow the procedure to begin with all the transfer prices set to the middle
of the interval, that is, equal to (LB + UB)/2.

5.5. Setting di€erent initial ¯ows (Init_Flows ˆ hValuei)

The procedure can also be implemented with any value of the initial ¯ows even using infeasible ¯ows.
This is so since we eliminate all constraints exclusively concerning ¯ows when we solve the problem
P…x; t; v; z j x† (see Fig. 5). One case is when we set all the initial ¯ows to zero, but any initial value can be
used.
154 C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158

Table 4
Computational results for small and medium instances
Starting % Gap from the Solution No. of
point upper bound time (s) iterationsb
Small
instances No.a
1 Opt_Flows 1.882 4.51 2
Opt_TP 1.028 4.56 2
Heu_TP 1.889 13.83 11
LB_TP 1.889 13.66 11
UB_TP 1.743 5.71 3
Mid_TP 1.306 452.28 506
Init_Flows ˆ 0 1.889 13.24 11
Init_Flows ˆ 50,000 0.756 19.70 >17

2 Opt_Flows 0.000 3.46 1

Opt_TP 0.008 4.44 2
Heu_TP 0.000 3.33 1
LB_TP 0.000 3.33 1
UB_TP 0.036 4.40 2
Mid_TP 0.045 4.41 2
Init_Flows ˆ 0 0.000 4.23 2

3 Opt_Flows 0.183 7.33 5

Opt_TP 0.183 8.39 6
Heu_TP 0.381 6.06 4
LB_TP 0.381 6.06 4
UB_TP 0.129 29.35 27
Mid_TP 0.154 12.03 10
Init_Flows ˆ 0 0.381 6.96 5

4 Opt_Flows 1.032 23.21 20

Opt_TP 1.095 4.64 2
Heu_TP 1.328 42.64 38
LB_TP 1.432 32.64 29
UB_TP 1.134 26.13 23
Mid_TP 1.386 41.09 36
Init_Flows ˆ 0 1.432 32.73 29

5 Opt_Flows 1.101 308.73 269

Opt_TP 1.707 820.29 737
Heu_TP 24.746 437.18 336
LB_TP 24.746 440.87 336
UB_TP 13.571 197.72 174
Mid_TP 19.242 1,030.37 889
Init_Flows ˆ 0 24.764 360.87 312

Medium
instances No.a
1 Opt_Flows 0.022 80.84 2
Opt_TP 0.022 96.55 3
Heu_TP 0.022 100.32 3
LB_TP 0.022 572.25 28
UB_TP 0.024 3,054.75 160
Mid_TP 0.023 348.50 17
Init_Flows ˆ 0 0.022 572.69 29
 C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158 155

Table 4 (Continued)
Starting % Gap from the Solution No. of
point upper bound time (s) iterationsb
2 Opt_Flows 0.000 41.19 1
Opt_TP 0.000 40.53 1
Heu_TP 1.465 84.10 5
LB_TP 0.400 61.29 3
UB_TP 0.282 49.44 2
Mid_TP 0.470 46.69 2
Init_Flows ˆ 0 0.400 70.71 4

3 Opt_Flows 0.251 127.16 2

Opt_TP 0.400 423.10 13
Heu_TP 3.473 107.48 2
LB_TP 1.384 118.95 2
UB_TP 0.782 147.35 3
Mid_TP 0.614 140.41 3
Init_Flows ˆ 0 1.384 143.81 3

4 Opt_Flows 0.773 242.18 7

Opt_TP 0.525 117.22 2
Heu_TP 6.691 126.49 3
LB_TP 2.186 143.41 3
UB_TP 1.515 126.21 2
Mid_TP 1.965 239.12 7
Init_Flows ˆ 0 2.186 206.57 5

5 Opt_Flows 1.340 10,637.81 372

Opt_TP 0.504 21,836.41 793
Heu_TP 6.527 698.22 37
LB_TP 5.260 14,295.65 677
UB_TP 3.174 34,447.42 1,356
Mid_TP 4.829 26,312.53 1,177
Init_Flows ˆ 0 5.241 22,873.70 1,028
a
Instances of the same size have the same structure shown in Table 3, but they di€er by demand, tax rate, and other parameters. Small
and medium instances 5 are de®ned with unconstrained transfer prices (LB ˆ 0 and UB ˆ A large number).
b
The number of iterations represents the number of times the convergence criterion is tested (see Fig. 5).

The procedure converges for all the cases we have tested, even when it starts with infeasible ¯ows. The
main results are summarized in Table 4. Note that the quality of the local solution obtained with the
procedure is highly dependent on the starting point. For most of the instances we have solved, the best
solutions have been obtained when the starting point is either the optimal ¯ows (Opt_Flows) or the optimal
transfer prices (Opt_TP) from the relaxed problem. A noticeable exception occurs in the small instance No.
1 for Init_Flows ˆ 50,000. It is important to note the behavior of the procedure in the small and medium
instances No. 5. These instances were created with a large ¯exibility for selecting all transfer prices. Of
course, these instances are not likely to be found in reality due to strict regulations of tax authorities, but
they provide a benchmark for the behavior of the procedure since we expect the gap between the local
solution and the UB to be larger when the transfer prices are less constrained. In instances No. 5 the
starting point with Opt_Flows and with Opt_TP from the relaxed problem signi®cantly outperforms the
other starting points.
It is also important to observe that the procedure converges very fast, especially when starting with
Opt_Flows or with Opt_TP. Even when the number of iterations is larger, the typical behavior of the
156 C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158

Fig. 6. Objective value progress for small instance No. 3 (starting point ˆ UB_TP).

procedure is characterized by very good solutions early on in the process (Fig. 6). In other less frequent
cases, the convergence is slower, but we have not observed this performance when starting with Opt_Flows
or Opt_Tp. This allows us to obtain very good solutions very fast.
Finally, concerning the characteristics of the solutions, all instances in Table 4 have been solved using
inequalities …6† for the customer demand constraints. This allows us to identify non-pro®table products
and costly customers to be served. The behavior of the procedure when satisfying total demand (equality
constraints for customer demand) is similar, although the gaps between the local solutions and their cor-
responding UBs are slightly larger. Most of the solutions show the transportation cost being totally allo-
cated to the origin or to the destination, with very few proportions lying in the interval (0, 1).

6. Conclusions and further research

In this paper we have presented a global supply chain model that includes explicitly TP and transpor-
tation cost allocation as decision variables. The model also considers the selection of transportation modes
based on approximate expressions for the inventory costs generated by the utilization of each transpor-
tation mode. Since the resulting problem is NP-hard, we have developed a heuristic successive LP-solution
procedure and UBs to test its performance. Our computational experience shows satisfactory results with
 C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158 157

very close gaps between the local solutions found by the procedure and their corresponding UBs even for
instances presenting a high degree of ¯exibility in the transfer prices. In some cases the procedure provides
provably optimal solutions. Our ongoing research comprises the test of larger instances, and the application
of global optimization procedures to the original problem to obtain either optimal solutions or solutions
with a prescribed gap, at least for the small instances. This will provide additional benchmarks for the
developed procedure.
Further research should consider pro®t repatriation, withholding taxes, and cash ¯ows; the inclusion of
additional international features such as the rules of origin, duty drawbacks, and import quotas that may
exist in regional trading agreements; and the analysis of more complex corporate rate functions. The in-
clusion of location decisions, especially related to the opening of new DCs is also a fertile area for further
research.
Since the algorithm that we presented here can ®nd near-optimal solutions to medium instances in
acceptable computation time, the model described here can be useful in the optimization of global supply
chains. MNCs can apply these techniques to increase signi®cantly their global pro®ts after taxes.

Acknowledgements

This research was partially funded by the Instituto Colombiano para la Ciencia y la Tecnologia,
COLCIENCIAS; by the Universidad del Valle, Cali, Colombia; and by the Fulbright Commission, ad-
ministered by the Latin American Scholarship Program for the American Universities (LASPAU).

References

Abdallah, W.M., 1989. International Transfer Pricing Policies: Decision Making Guidelines for Multinational Companies. Quorum
Books, New York.
Al-Khayyal, F.A., Falk, J.E., 1983. Jointly constrained biconvex programming. Mathematics of Operations Research 8 (2), 273±286.
Al-Khayyal, F.A., 1990. Jointly constrained bilinear programs and related problems: An overview. Computers & Mathematics with
Applications 16 (11), 53±62.
Al-Khayyal, F.A., 1992. Generalized bilinear programming: Part I Models, applications and linear programming relaxation. European
Journal of Operational Research 60, 306±314.
Al-Khayyal, F.A., Larsen, C., Voorhis, T.V., 1995. A relaxation method for nonconvex quadratically constrained quadratic programs.
Journal of Global Optimization 6, 215±230.
Androulakis, I.P., Maranas, C.D., Floudas, C.A., 1995. aBB: A global optimization method for general constrained nonconvex
problems. Journal of Global Optimization 7, 337±363.
Arntzen, B.C., Brown, G.G., Harrison, T.P., Trafton, L.L., 1995. Global supply chain management at digital equipment corporation.
Interfaces 25 (1), 69±93.
Belkaoui, A., 1985. International accounting. Greenwood Press, Westport.
Ben-Tal, A., Eiger, G., Gershovitz, V., 1994. Global minimization by reducing the duality gap. Mathematical Programming 63, 193±212.
Bloemhof-Ruwaard, J.M., Hendrix, E.M.T., 1996. Generalized bilinear programming: An application in farm management. European
Journal of Operational Research 90, 102±114.
Bodner, P.M., 1997. International taxation. in: Frederick, D.S.C. (Ed.), International Accounting and Finance Handbook, second
edition, Wiley, New York.
Canel, C., Khumawala, B.M., 1997. Multi-period international facilities location: An algorithm and application. International Journal
of Production Research 35 (7), 1891±1910.
Cohen, M.A., Lee, H.L., 1989. Resource deployment analysis of global manufacturing and distribution networks. Journal of
Manufacturing Operations Management 2, 81±104.
Cohen, M.A., Fisher, M., Jaikumar, R., 1989. International manufacturing and distribution networks: A normative model framework.
In: Ferdows, K. (Ed.), Managing International Manufacturing. North-Holland, Amsterdam, pp. 67±93.
Eccles, R.G., 1985. The Transfer Pricing Problem: A Theory for Practice. Lexington Books, Lexington.
Floudas, C.A., Aggarwal, A., 1990. A decomposition strategy for global optimum search in the pooling problem. ORSA Journal of
Computing 2 (3), 225±235.
158 C.J. Vidal, M. Goetschalckx / European Journal of Operational Research 129 (2001) 134±158

Floudas, C.A., Visweswaran, V., 1990. A global optimization algorithm (GOP) for certain classes of nonconvex NLPs ± I Theory.
Computers and Chemical Engineering 14 (12), 1397±1417.
Floudas, C.A., Visweswaran, V., 1993. Primal-relaxed dual global optimization approach. Journal of Optimization Theory and
Applications 78 (2), 187±225.
 uc
Gallo, G., Ulk u, A., 1977. Bilinear programming: An exact algorithm. Mathematical Programming 12, 173±194.
Glautier, M.W.E., Bassinger, F.W., 1987. A Reference Guide to International Taxation: Pro®ting from Your International
Operations. Lexington Books, Lexington, MA.
J
udice, J.J., Faustino, A.M., 1991. A computational analysis of LCP methods for bilinear and concave quadratic programming.
Computers and Operations Research 18 (8), 645±654.
Konno, H., 1976. A cutting plane algorithm for solving bilinear programs. Mathematical Programming 11, 14±27.
Mueller, G.G., Gernon, H., Meek, G.K., 1997. Accounting: An International Perspective, fourth ed. Irwin Book Team, Chicago.
Nieckels, L., 1976. Transfer Pricing in Multinational Firms: A Heuristic Programming Approach and a Case Study. Wiley, New York.
O'Connor, W., 1997. International transfer pricing, in: Frederick, D.S.C. (Ed.), International Accounting and Finance Handbook,
second ed. Wiley, New York.
Pardalos, P.M., Vavasis, S.A., 1991. Quadratic programming with one negative eigenvalue is NP-hard. Journal of Global Optimization
21, 843±855.
Radebaugh, L.H., Gray, S.J., 1993. International Accounting and Multinational Enterprises, third ed. Wiley, New York.
Sherali, H.D., Shetty, C.M., 1980. A ®nitely convergent algorithm for bilinear programming problems using polar cuts and disjunctive
face cuts. Mathematical Programming 19, 14±31.
Sherali, H.D., Alameddine, A., 1992. A new reformulation-linearization technique for bilinear programming problems. Journal of
Global Optimization 2, 379±410.
Sherali, H.D., Tuncbilek, C.H., 1992. A global optimization algorithm for polynomial programming problems using a reformulation-
linearization technique. Journal of Global Optimization 2, 101±112.
Stitt, I., 1995. Transfer pricing for the next century. Accountancy 116, 92±94.
Vaish, H., Shetty, C.M., 1977. A cutting plane algorithm for the bilinear programming problem. Naval Research Logistics Quarterly
24, 83±94.
Vidal, C.J., Goetschalckx, M., 1997. Strategic production-distribution models: A critical review with emphasis on global supply chain
models. European Journal of Operational Research 98, 1±18.
Vidal, C.J., Goetschalckx, M., 1998. A heuristic for the design of global supply chains with transfer pricing and transportation cost
allocation. Research Report, School of Industrial and Systems Engineering, Georgia Institute of Technology, Atlanta, Georgia.
Visweswaran, V., Floudas, C.A., 1990. A global optimization Algorithm (GOP) for certain classes of nonconvex NLPs ± II Application
of theory and test problems. Computers and Chemical Engineering 14 (12), 1419±1434.
Visweswaran, V., Floudas, C.A., 1993. New properties and computational improvement of the (GOP) algorithm for problems with
quadratic objective functions and constraints. Journal of Global Optimization 3, 439±462.

View publication stats