You are on page 1of 23

This article was downloaded by: [Selcuk Universitesi]

On: 03 January 2015, At: 08:06


Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer
House, 37-41 Mortimer Street, London W1T 3JH, UK

Journal of Statistics and Management Systems


Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/tsms20

A novel approach to find optimal transshipment rules


in a two echelon distribution system
a
Junne-Ning Hwang
a
Department of Business Management , National United University , 1, Lienda, Miaoli ,
36003 , Taiwan
Published online: 14 Jun 2013.

To cite this article: Junne-Ning Hwang (2010) A novel approach to find optimal transshipment rules in a
two echelon distribution system, Journal of Statistics and Management Systems, 13:6, 1261-1282, DOI:
10.1080/09720510.2010.10701533

To link to this article: http://dx.doi.org/10.1080/09720510.2010.10701533

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of
the Content. Any opinions and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied
upon and should be independently verified with primary sources of information. Taylor and Francis shall
not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other
liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or
arising out of the use of the Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
A novel approach to find optimal transshipment rules in a two echelon
distribution system
Junne-Ning Hwang ∗
Department of Business Management
National United University
1, Lienda, Miaoli, Taiwan 36003
Abstract
Lateral transshipment is an effective means for improving customer service and
reducing total system costs, especially when the distribution center is acting as a coordinator
as well as permitting lateral transshipment in the system. In this study we propose nine
transshipment rules and evaluate their performances under emergent and complete trans-
shipment environments. We evaluate these transshipment rules through the two methods of
simulation and experiment design, and verify whether these transshipment rules perform
the same under the condition of a distribution system consisting of different numbers of
retailers. The research result shows that: (1) The transshipment rules that perform the best
Downloaded by [Selcuk Universitesi] at 08:06 03 January 2015

are designed by considering the ability of retailers in terms of the time needed to transship.
These rules are followed by those which are designed by considering the predicted ability of
retailers in the end of a certain period. The transshipment rules which are designed according
to the past ability certain period. The transshipment rules which are designed according to
the past ability of retailers perform the worst. (2) No matter the system consists of 5, 10,
or 20 retailers, the transshipment rules which are designed based on the ability of retailers
at the time to transship always perform the best. This is valuable information in designing
transshipment system.

Keywords and phrases : Two-echelon distribution system, transshipment rules, experiment design.

1. Introduction
The distribution center plays an important role in the supply chain
system in which transshipment is permitted. Many methods could be used
to improve the effects of risk pooling in a distribution system, we use
transshipment in this study. Research that assumes one customer demand
within a period makes it easier to obtain the optimal solution, but this
limits its application in real life.
∗ E-mail: jnhwang@nuu.edu.tw
——————————–
Journal of Statistics & Management Systems
Vol. 13 (2010), No. 6, pp. 1261–1282
°c Taru Publications
1262 J. N. HWANG

Gross [6], Das [4], Jonsson and Silver [8], and Diks and de Kok
[5] have presented models, where preventive transshipment takes place
before the realization of the entire ordering cycle’s demand, with the
purpose of achieving a better distribution of available inventory among
the stocking locations. Emergent transshipment models for repairable
items and/or low demand, typically followed a Poisson distribution,
including those of Cohen et al. [2], Lee [9], Axsater [1], Dada [3], and
Sherbrooke [13].
Hoadley and Heyman [7] examined a two-echelon inventory model
that included transshipments between higher echelon retail warehouses.
Their model did not allow for stochastic transshipment lead-times. Lee
[10] examined the impact of emergent transshipments and found that high
service levels could be maintained by using transshipments. He did not
examine various transshipment rules of when to make transshipments.
Axsater [1], and Sherbrooke [13] modeled emergent lateral transshipments
Downloaded by [Selcuk Universitesi] at 08:06 03 January 2015

to determine the impact on system performance and found that significant


backorder reductions were possible.
In Tagaras [14], a two-echelon distribution system consisting of two
retailers and one depot and an order-up-to-level policy were considered,
in which transshipment between the retailers was allowed. The depot
was assumed to have an infinite capacity, and the replenishment lead-
time was assumed negligible. This model was characterized as complete
pooling in that if there was an economic incentive to transship one item,
the maximum amount would be sent. Four transshipment policies were
proposed. Two of them were designed based on net inventory, and the
other two policies were designed based on inventory position.
Tagaras [14] has studied the performance of an inventory system with
one central warehouse and multiple retailer outlets, collaborating in the
case of an imminent shortage by moving inventory between themselves.
He concluded that the benefits of risk pooling through transshipment are
substantial and increase with the number of pooled locations. He devised
two transshipment policies:
(1) A random policy
(A) When a location faces a shortage and the other two have available
on-hand inventory, the source location is chosen randomly and
transships as much as needed to completely eliminate the shortage.
If its inventory is not sufficient, then the other location also sends
the quantity required to eliminate the remaining shortage.
A NOVEL APPROACH TO FIND OPTIMAL TRANSSHIPMENT RULES 1263

(B) When two locations face shortages and the other one has available
on-hand inventory, the receiving location is chosen randomly and
the sending location transships first to it as much as needed to com-
pletely eliminate the shortage. If there is still remaining inventory
at the sending location, then the other also receives the quantity
required to eliminate its shortage.

(2) Risk balancing policy

The rationale behind this policy is that the determination of trans-


shipment quantity should also take into account the risk of stock out in at
least the following period. The risk should be balanced between the two
senders or the two receivers. The transshipped quantity must be one that
equalizes the probability of a stock out in the following period.
An additional transshipment policy called a priority policy was also
Downloaded by [Selcuk Universitesi] at 08:06 03 January 2015

examined in Tagaras [15]. It dictates that when a location faces a shortage


and the other two locations have available on-hand inventory, the first
sender is the location with the maximum on-hand stock. This retailer
transships as much as is needed to completely eliminate the shortage. If its
inventory is not sufficient, then the other location also sends the quantity
required to eliminate the remaining shortage. Similarly, when there are
two locations that face shortages and the other location has available on-
hand inventory, the first receiver is the location with the larger shortage.
The sender transships first to it as much as needed to completely eliminate
the shortage. If there is still remaining inventory at the sender, then the
other location also receives the quantity required to eliminate its shortage.
Hwang [11] proposed a model to generalize the customer demand of
a distribution system by allowing multiple customer demands within one
period, he compared the difference between single demand and multiple
demands within one period and verified the main factors that affect a
distribution system as well as the ways in which they affect the distri-
bution system. Furthermore, Hwang proposed a concept of “Integrated
purchasing and rationed distributing” to construct a distribution system
and showed its prominent characteristics when customer service level and
system cost are both considered.
Transshipment is one way to reduce shortages, especially when one
retailer faces a shortage while another retailer still has inventory on-
hand. Having studied research related to the distribution system and
lateral transshipment, Hwang [12] has gathered some factors that are
1264 J. N. HWANG

discussed and analyzed more frequently. He drew a cause-effect diagram


to illustrate the relationships among the factors, and deduced the major
factors in studying a distribution system in which lateral transshipment
is permitted. Only a few of these factors are considered in classifying the
distribution systems. They are: (1) the nature of transshipment (includ-
ing emergent and preventive transshipment); (2) the transshipment rule
(which is used to determine transshipment priority, transshipment quan-
tity, and the time to stop transshipment). As the nature of transshipment is
determined, it is the transshipment rule that affects system performance.
In some research, the nature of transshipment and the transshipment rules
as a whole is called transshipment policy.
Hwang [12] also pointed out that many factors can be considered in
designing transshipment rules, such as cost, customer service level, the
nature of transshipment, and so on. The design of the transshipment rules
must based on certain purposes, and this limits the factors that could be
Downloaded by [Selcuk Universitesi] at 08:06 03 January 2015

considered in designing transshipment rules. For example, the purpose of


emergent transshipment is to reduce the total cost of the system, while the
purpose of preventive transshipment is to improve the customer service
level.
This paper is organized as follows. In section 2, we describe the
behavior of the system briefly. In section 3, we introduce the simulation
environment of this study. The content of the transshipment rules is
described in section 4. In section 5, we explain the methods we used
to evaluate the performance of the transshipment rules. In section 6, we
evaluate and distinguish transshipment rules by numerical examples, and
make some conclusions. Finally, we suggest directions for future research.

2. System description

An inventory distribution system consisting of one distribution cen-


ter and some retailers is considered. Only retailers face external demands.
Customer demands are independent of each other. The demands the
retailers face are independent of each other, too. When a retailer fails to
satisfy a customer demand due to a shortage of on-hand stock, the order
is backordered and will be filled in future periods. It is obvious that the
retailers consider the inability to meet the demand undesirable because
of the high penalty costs accompanied with backordering. At the end
of every review period, retailers place orders at the distribution center
to bring up their inventory position. These orders arrive after a positive
A NOVEL APPROACH TO FIND OPTIMAL TRANSSHIPMENT RULES 1265

deterministic lead-time of l period, and we assume all retailers have the


same replenishment lead-time.
In addition to the retailers, the distribution center also uses a periodic
review ordering policy to replenish the stock therein, so as to raise the
retailers’ inventory levels. In this paper, it is assumed that the durations of
the review period at the distribution center and at the retailers are equal,
and the reviews are almost synchronized. After receiving the orders from
suppliers, the distribution center has to decide how to allocate inventories
to retailers. We assume that the supplier’s capacity is infinite, so the
distribution center will never be backordered.
We operate the distribution system with the concept of “Integrated
purchasing and rationed distributing”, which was proposed and de-
scribed very clearly by Hwang [11].

3. The model
Downloaded by [Selcuk Universitesi] at 08:06 03 January 2015

There is only one distribution center in this study. In order to verify


whether the priority of transshipment is consistent when the system is
constructed with a different number of retailers, we consider 5, 10, and 20
retailers in the system. The decision of what time to transship and what
quantity can be transshipped is constrained to emergent and complete
transshipment. The system structure is drawn in Figure 1.

Figure 1
The distribution system
In Figure 1, the dotted lines are information flows. The retailers
relay the information about orders which consist of base quantity and
1266 J. N. HWANG

risk quantity to the distribution center. The distribution center then


releases purchasing orders to the supplier after orders from retailers are
integrated. The thick lines mean material flows. The inventory is sent from
the supplier to the distribution center, and the distribution center then
distributes them to retailers. The base quantity is sent with as much as is
required, and the risk quantity is distributed according to the proportions
that are calculated by using the parameters. The thin lines among retailers
indicate that lateral transshipments are allowed. Hwang’s study [11]
clearly describes how this model operates.
There are many factors affect the performance of a distribution sys-
tem, we keep some of these factors as constants to simplify the complexity
of system and reduce the number needed in a simulation. Customer
demand distributions that are faced by retailers are not identical, so we
keep the mean values of customer demand distributions that are faced by
a retailer in all simulation examples the same. The standard deviations of
Downloaded by [Selcuk Universitesi] at 08:06 03 January 2015

customer demand distributions are designed as control parameters. We


use variance coefficients instead of standard deviations for convenience.
The negative customer demand could not be ignored as the standard
deviation is greater than 0.3. In spite of customer demands are controlled
as a positive number by the program, the variance coefficients that we
adopt are still not more than 0.3. We select 0.1, 0.2 and 0.3 as the three
levels of customer demand variation.
Three types of costs are considered in this study. They are the holding
cost, the transshipment cost, and the penalty cost. We assume the cost
structure is identical in every simulation example. Finally, the default
service level of every retailer remains unchanged throughout the study.
The fixed conditions are listed in the next two tables.

(1) Cost structure


Cost Holding cost Transshipment cost Penalty cost
Value 1 2 8

(2) Mean customer demands and default service levels of retailers


Retailer 1 2 3 4 5
Mean demand 200 250 180 210 230
Default Service Level 85% 85% 90% 90% 95%
(Note. When the number of retailers that make up the distribution system is 10
or 20, the system is twofold or fourfold of this table)
A NOVEL APPROACH TO FIND OPTIMAL TRANSSHIPMENT RULES 1267

There are nine scenarios and four control variables (each with three
levels) to be considered in this study (see Table 1). The control variables
designed for the nine scenarios are all the same and for each at three
levels. We adopt type L9 experiment design allocation (see Table 2), so
there are 9 combinations to be executed in every scenario. Considering 9
scenarios, 9 experiment designs, and 9 transshipment rules, we get 729
simulation combinations. Each combination is simulated 150 times and
each time has 300 periods. We obtain one set of observation values for
each combination by averaging 150 times results, The simulation software
we use is eM Plant, which is a famous commercial software.
The transshipment mechanism is triggered when there is a retailer
that can’t satisfy its customer’s need. While there is on-hand inventory in
the system, the on-hand inventory will be transshipped to eliminate the
shortage or to improve customer service level. Figure 2 shows the logic of
transshipment. As described previously, this study is only concerned with
Downloaded by [Selcuk Universitesi] at 08:06 03 January 2015

emergent and complete transshipments. The sender will supply as much


quantity as possible, so the quantity that a retailer provides is:

Min [sender on-hand inventory, the shortage quantity faced by receiver]

In such a situation, the only thing that the transshipment mechanism


has to do is to determine a retailer to transship.
We propose nine transshipment rules to determine which retailer
should transship the inventory. Since the arrival of customer demand
is assumed to obey to Poisson distribution, the condition of two or
more retailers facing shortages simultaneously would not happen. As
the transshipment rule is selected, its logic will help to select a retailer
to transship the inventory. If the shortage still exists after transshipment
and there is on-hand inventory in the system, the transshipment rule
will determine the next retailer to transship inventory. The transshipment
process keeps going and stops when the inventory on-hand in the system
is exhausted or the shortage is completely eliminated.

4. Transshipment rules
4.1 The design of transshipment rule
The transshipment quantity in an emergent and complete transship-
ment environment is described in the previous section. Next we explain
the transshipment rules that determine which retailer should transship.
The transshipment logic is shown as Figure 2.
1268 J. N. HWANG

Table 1
Simulation scenarios and combinations
Nine simulation scenarios Experiment design & simulation combinations
# of retailer Customer demand Control variable Transshipment rule
type
5 1. Variable-multiple Four control variables 81 sets each of which
demands obeys each with 3 levels. contains 9 transshipment
Poisson (λi ) L9 experiment design is rules. 729 simulation
adopted (see Table 2). combinations are
2. Fixed-multiple
9 scenarios combine 9 obtained
demands
experiment designs, for a
3. Single demand
total of 81 sets
10 4. Variable-multiple
demands obeys
Poisson (λi )
5. Fixed-multiple
demands
6. Single demand
Downloaded by [Selcuk Universitesi] at 08:06 03 January 2015

20 7. Variable-multiple
demands obeys
Poisson(λi )
8. Fixed-multiple
demands
9. Single demand

Table 2
Experiment design

Control variables and levels


Simulation Arrival rage Demand variance Inventory distribution Lead
design ( λi ) (c.v.) ratio ( pi ) time
G1 5 0.1 Zi × σi 1
G2 5 0.2 σi2 2
G3 5 0.3 Ui + Zi × σi 3
G4 15 0.1 σi2 3
G5 15 0.2 Ui + Zi × σi 1
G6 15 0.3 Zi × σi 2
G7 25 0.1 Ui + Zi × σi 2
G8 25 0.2 Zi × σi 3
G9 25 0.3 σi2 1

( Zi : Safety factor of retailer; under default service level; σi :The standard deviation of
customer demand that retailer faced; Ui : The mean customer demand that retailer; faced)
A NOVEL APPROACH TO FIND OPTIMAL TRANSSHIPMENT RULES 1269
Downloaded by [Selcuk Universitesi] at 08:06 03 January 2015

Figure 2
The transshipment logic

The purpose of transshipment rules is to improve system perfor-


mance or to decrease system cost. This study assumes the cost structure
of all retailers are the same, so how to reduce the shortage is the key
point in reducing system cost or improving system performance. Based on
this truth, the transshipment rules are designed by considering the ability
of retailers. The retailer with the best ability should transship inventory
first. There are many indices that could be used to represent the ability of
retailers. We propose nine indices for designing transshipment rules. The
notations used in transshipment are listed below, and Table 3 shows the
content and formulas for each transshipment rule.

λi : The expect number of customer arrivals faced by retailer i in a period


ui : The expected customer demand quantity faced by retailer i
1270 J. N. HWANG

CD i : The quantity that the customer has demanded of retailer i at the time
transshipment takes place
Num Avail i : The number for which retailer; can still satisfy customer
demand at the time transshipment takes place
Num Doneratio: The ratio of customer demand number for retailer i ; it is
the number that the customer has demand of retailer i divided by the
mean customer demand number of retailer i in a period
Num Left: The number for which retailer i can still satisfy customer
demand at the end of a period
NUM i : The number that the customer has demanded of retailer i at the
time transshipment takes place
PQ i : The on-hand inventory level of retailer i at the time transshipment
takes place
O Left i : The expected on-hand inventory of retailer; at the end of a period
Downloaded by [Selcuk Universitesi] at 08:06 03 January 2015

Q Ratio1 i : The ratio of the quantity that the customer has demanded
of retailer i divided by the mean demand quantity of retailer i in a
period
Q Ratio2 i : The ratio of the quantity that the customer has demand of
retailer i divided by the maximum inventory level that satisfies the
default service level of retailer i
Q Ratio3 i : The ratio of the quantity that the customer has demanded of
retailer i divided by the risk quantity of retailer i in a period
R max i : The maximum inventory level that satisfies the default service
level of retailer i
Time Left i : The ratio of left time of retailer i , it is the on-hand inventory
quantity of retailer i divided by the mean customer demand quantity
of retailer i in a period

Some formulas seemed unreasonable. For example, it is illogical for


rule2 and rule4 if (λi − NUMi ) are negative. In fact, no matter whether
the value of (λi − NUMi ) is positive or not, these indices are only
used to judge the priority of transshipment rules, not be used for other
applications.
4.2 Transshipment rules evaluation
We use two methods to evaluate the performances of transshipment
rules. We call them the Relative method and Ratio method. The meaning
and steps of these methods are described as follows:
A NOVEL APPROACH TO FIND OPTIMAL TRANSSHIPMENT RULES 1271
Table 3
Contents and formulas of transshipment rules

Rule Rule content Formula

Rule 1 The retailer with the max PQ PQi = On-hand inventor of retaileri on
transships inventory first the time to transship
Rule 2 The retailer with the max Q Lefti = PQi − ui × (λi − NUMi )
Q Left transships first
CDi
Rule 3 The retailer with the max Q Ratioi = ui × λi
Q Ratio transships first
PQi
Rule 4 The retailer with the max Num Left i = ui − (λi − NUMi )
Num Left transships first
PQi
Rule 5 The retailer with the max Num Avail i = ui
Num Avail on transships first
Numi
Rule 6 The retailer with the max Num Doneratioi = λi
Downloaded by [Selcuk Universitesi] at 08:06 03 January 2015

Num Doneratio transships


first
PQi
Rule 7 The retailer with the max Time lefti = λi × ui
Time Left transships first
CDi CDi
Rule 8 The retailer with the max Q Ratio21 = R max1 = R meani + Zi ∗ R std1
Q Ratio2 transships first
CDi
Rule 9 The retailer with the max Q Ratio3i = Zi ∗ R stdi
Q Ratio3 transships first

4.2.1 The relative method


The Relative method distinguishes the priority of transshipment
rules by comparing their performance under the same conditions. The
performance is measured by which level of default service is reached. The
Relative method is implemented in two ways: (1) Comparing which level
of default service is reached by retailers and distribution center under
9 scenarios (see Table 1, column 2) (for simplicity, we call it Relative!
hereinafter); (2) Comparing which level of default service level is reached
by every transshipment rule under 81 sets of scenarios and experiment
design combinations (see Table 1, column 3) (for simplicity, we call it
Relative2 hereinafter). Advanced calculations proceed as the number is
figured out. The steps of the Relative method are listed below.
Step 1: Calculate which level of default service is reached
(1) For Relative1
1272 J. N. HWANG

1.1 Calculate the number for the default service level specified
by retailers and distribution center that is reached under
different rules. The’actual service levels are measured by
the delivered number and delivered quantity.
1.2 Sum the numbers of default service level reached accord-
ing to the same transshipment rule.
(2) For Relative2
Calculate the number of the default service level reached by the
system under 9 transshipment rules and 81 sets of simulation
sets (see Table 1, column 3).
Step 2: Ranking transshipment rules in the order of the number for which
the default service level is reached.
Step 3: Give each transshipment rule a position weight according to the
position at which it is located.
Downloaded by [Selcuk Universitesi] at 08:06 03 January 2015

A transshipment rule should be weighted more if it performs


better. In calculating the position weight of every transshipment
rule, we give the rule that performs the best point 9 (for there
are nine positions for nine transshipment rules). The next is given
point 8, point 7, and so on. If there are two or more rules that
perform the same, the position weights of these rules are the
same, too, and the positions weights are the average value of these
position weights. When a transshipment rule is assigned a position
weight, we call it the position equivalence.
Step 4: Calculate the integrated priority of transshipment rules
The priority of transshipment rules in specific situations can b’e
observed in step 3, but we can’t have a clear overall view of the
transshipment rules. In this step, the position equivalences are
added according to the number of retailers in the system and the
transshipment rules. Then the priority of transshipment rules is
obvious under a system which consists of a different number of
retailers.

4.2.2 The ratio method

The Ratio method identifies the priority of transshipment rules by


the ratio which is calculated from “actual service level/period cost” (the
ratio of effect-cost). The reason why we propose this method is that the
system cost is not considered in the Relative method, and this may result
A NOVEL APPROACH TO FIND OPTIMAL TRANSSHIPMENT RULES 1273

from a retailer’s higher service level contributing to a higher cost. To solve


such a problem, we propose the Ratio method in which the system cost is
considered. This method also improves the imperfections in evaluating.
The steps of the Ratio method are listed as:
Step 1. Calculate the ratio of effect-cost
1. The observation set contains the system cost, the actual service
level measured by the delivered number, and the actual service level
measured by the delivered quantity. Two ratios of effect-cost can be
derived from these data:

(A) The ratio of effect-cost measured by delivered number

The average delivered number


= .
System cost

(B) The ratio of effect-cost measured by delivered quantity


Downloaded by [Selcuk Universitesi] at 08:06 03 January 2015

The average delivered quantity


= .
System cost

Step 2. Normalize the ratios of service level-cost


The simulation combinations are not all the same in terms of the
number of retailers and in terms of customer arrival rate, while these two
affect service level and period cost very much (Hwang, 2003). In order to
compare transshipment rules on the same basis, we group 729 simulation
combinations into 81 sets (see Table 1, column 3) and normalize these
ratios by dividing the max ratio value into the set it belongs to.
Step 3. Average the normalized ratios for the same transshipment rule under
specific conditions and compare their priority
The 729 simulation combinations could be classified by scenarios,
customer demand types, parameters, or the combinations of them. Various
meanings in management could be obtained as the transshipment rules are
compared under different conditions.

5. Numeric results
Owing to the great amount of simulation and intermediate data, we
omit the original simulation results and most of the intermediate data.
Only final results after analysis are presented in context.
1274 J. N. HWANG

5.1 Evaluating the transshipment rules by the Relative method


Step 1: Calculate the number at which default service level is reached
Every retailer as well as distribution center has a default service level
in this research. The actual service levels are measured by the ratio of
average shortage number or by the ratio of the average shortage quantity.
From simulation data, we can calculate the number at which the service
level is reached, no matter whether it is measured by shortage number or
by shortage quantity.
Because of the complex process in manipulating the simulation data,
we use a position weight instead of the number at which the service
level is reached in the Relative2 method, and the position weight is
obtained directly by ranking the service level of each rule when processing
simulation data.
Step 2: Rank transshipment rules in descending order according to the number
Downloaded by [Selcuk Universitesi] at 08:06 03 January 2015

at which service level is reached


The transshipment rules can be ranked by the number at which
service level is reached. We represent these data with graphics, in which
the rules are listed from left to right and arranged from better to worse,
and the arrows mean the number decreases between rules. In Table 5, the
transshipment rules on the same bottom line mean the rules at which the
default service level is reached by them are the same, so they are given the
same position equivalence.
Step 3: Give transshipment rules position equivalences
The position equivalences of all transshipment rules are given by
the order shown in Table 5, and the position equivalences of the nine
transshipment rules under various situations are listed in Table 6.
It can be seen in Table 6 that transshipment rules perform a little
differently in different situations. Some of them perform better (with
higher total equivalences), for example, rule 1, rule 5 and rule 7, and some
rules always perform worse (with lower total equivalences), for example,
rule 3 and rule 6. It is difficult to have a global view of transshipment rules
from Table 6.
Step 4: Sum position equivalences according to the number of retailers in
the system and transshipment rules
In order to have a global view about transshipment rules, in this step
we add the position equivalences according to the number of retailers
A NOVEL APPROACH TO FIND OPTIMAL TRANSSHIPMENT RULES 1275
Table 5
The priority of transshipment rules represented by graphics
# of Measurement Customer Priority of transshipment rules
retailers Method Measured by demand type Better → Worse
5 Relative 1 Delivered F 1, 2, 4, 5, 7, 9 → 8 → 3, 6
Number V 5 → 1, 2, 4, 7 → 3, 6, 8, 9
S 1, 2, 5, 7 → 4 → 3, 6, 8, 9
Delivered F 2, 4, 5, 7, 9 → 1 → 3, 8 → 6
Quantity V 5 → 1, 2, 4 → 3, 6, 7, 8, 9
S 2 → 1, 4, 7 → 5 → 3, 8 → 9, 6
Relative 2 Delivered F 1 → 5 → 7 → 2 → 4 → 9 → 3 → 8, 6
Number V 5 → 1, 7 → 4 → 2 → 9 → 8 → 6 → 3
S 5→7→2→1→4→3→8→9→6
Delivered F 5→7→1→6→8→9→3→2→4
Quantity V 2 → 3 → 8 → 1, 9 → 7 → 4 → 5 → 6
S 1, 8 → 5 → 2 → 4 → 3 → 9 → 7, 10
Downloaded by [Selcuk Universitesi] at 08:06 03 January 2015

10 Relative 1 Delivered F 1, 2, 3, 4, 5, 7, 8, 9 → 7
Number V 1, 5, 7 → 3 → 2 → 4 → 6 → 9 → 8
S 1, 5, 7 → 2, 4 → 6 → 3, 8, 9
Delivered F 4 → 2, 7, 9 → 1, 3, 5, 8 → 6
Quantity V 7 → 1, 3, 5 → 4 → 2 → 6 → 9 → 8
S 2 → 4 → 1, 7→ 5 → 3, 8 → 9 → 7
Relative 2 Delivered F 7→5→1→4→2→6→3→9→8
Number V 5 → 1, 7 → 4 → 2 → 6 → 9 → 3 → 8
S 1→7→5→2→4→3→8→6→9
Delivered F 3 → 5 → 8 → 7 → 2 → 1, 9 → 4 → 6
Quantity V 5→7→1→3→9→2→4→8→6
S 1 → 7 → 5 → 2,4 → 3 → 8 → 6 → 9
Relative 1 Delivered F 1 → 2, 5 → 3, 4, 6, 7, 8, 9
Number V 1, 2, 5, 7 → 4 → 6 → 8 → 3, 9
20 S 1 → 2, 4, 5, 7 → 6 → 3, 8, 9
Delivered F 5 → 2, 4, 7 → 1 → 9 → 8 → 3, 6
Quantity V 1, 2, 3, 4, 5, 6, 7, 8, 9
S 4, 5, 7 → 1, 2 → 6 → 8 → 3, 9
Relative 2 Delivered F 1→7→4→5→2→3→9→6→8
Number V 5→7→1→2→4→6→8→3→9
S 1 → 7 → 5 → 4 → 2→ 6 → 3 → 8 → 9
Delivered F 2 → 4 → 1, 7 → 5 → 3 → 8 → 9 → 6
Quantity V 7→5→1→2→4→8→9→3→6
S 7→2→1→5→4→6→3→8→9
F: Fixed-multiple customer demand; V: Variable-multiple customer demand;
S: Single customer demand
1276 J. N. HWANG

Table 6
Integrated equivalences
# of Measurement Customer Transshipment Rules
retailers Method Measured by demand type Rule1 Rule2 Rule3 Rule4 Rule5 Rule6 Rule7 Rule8 Rule9

5 Relative 1 Delivered F 6.5 6.5 1.5 6.5 6.5 1.5 6.5 3 6.5
Number V 6.5 6.5 2.5 6.5 9 2.5 6.5 2.5 2.5
S 7.5 7.5 2.5 5 7.5 2.5 7.5 2.5 2.5
Delivered F 4 7 2.5 7 7 1 7 2.5 7
Quantity V 7 7 3 7 9 3 3 3 3
S 7 9 3.5 7 5 1.5 7 3.5 1.5
Relative 2 Delivered F 9 6 3 5 8 1.5 7 1.5 4
Number V 7.5 5 1 6 9 2 7.5 3 4
S 6 7 4 5 9 1 8 3 2
Delivered F 7 2 3 1 9 6 8 5 4
Quantity V 5.5 9 8 3 2 1 4 7 5.5
S 8.5 6 4 5 7 1.5 8.5 3 1.5
Relative 1 Delivered F 5.5 5.5 5.5 5.5 5.5 1 5.5 5.5 5.5
Downloaded by [Selcuk Universitesi] at 08:06 03 January 2015

Number V 8 5 6 4 8 3 8 12 2
S 8 5.5 2 5.5 8 4 8 2 2
10 Delivered F 3.5 7 3.5 9 3.5 1 7 3.5 7
Quantity V 7 4 7 5 7 3 9 1 2
S 6.5 9 3.5 8 5 1 6.5 3.5 2
Relative 2 Delivered F 7 5 3 6 8 4 9 1 2
Number V 7.5 5 2 6 8 4 7.5 1 3
S 9 6 4 5 8 2 7 3 1
Delivered F 3.5 5 9 2 8 1 6 7 3.5
Quantity V 7 4 6 3 9 1 8 2 5
S 9 5.5 4 5.5 7 2 8 3 1
20 Relative 1 Delivered F 9 7.5 3.5 3.5 7.5 3.5 3.5 3.5 3.5
Number V 7 7 1.5 5 7 4 7 3 1.5
S 9 6.5 2 6.5 6.5 4 6.5 2 2
Delivered F 5 7 1.5 7 9 1.5 7 3 4
Quantity V 5 5 5 5 5 5 5 5 5
S 5.5 5.5 1.5 8 8 4 8 3 1.5
Relative 2 Delivered F 9 5 4 7 6 2 8 1 3
Number V 7 6 2 5 9 4 8 3 1
S 9 5 3 6 7 4 8 2 1
Delivered F 6.5 9 4 8 5 1 6.5 3 2
Quantity V 7 6 2 5 8 1 9 4 3
S 7 8 3 5 6 4 9 2 1

in the system and transshipment rules and show it in Table 7. The


performance of every transshipment rule can be observed clearly from
Table 7 under a different number of retailers in the system, rule 1, rule 5
and rule 7 perform the best, followed by rule 2 and rule 4. The others
perform the worst.
A NOVEL APPROACH TO FIND OPTIMAL TRANSSHIPMENT RULES 1277
Table 7
Sum of equivalence under the relative method

# of Rules and the sum of equivalence


retailers Rule 1 Rule 2 Rule 3 Rule 4 Rule 5 Rule 6 Rule 7 Rule 8 Rule 9

5 82 78.5 38.5 64 88 25 80.5 39.5 44


10 81.5 66.5 55.5 64.5 85 27 89.5 44.5 36
20 86 77.5 33 71 84 38 85.5 34.5 28.5

5.2 Evaluate the transshipment rules by the Ratio method

Step 1. Calculate the ratios of effect-cost


There are 729 simulation combinations (see Table 4) in this study,
and each of them obtains a set of observations after simulation. The
observations contain the system cost, the actual service level measured by
the delivered number and the delivered quantity in a certain period, and
Downloaded by [Selcuk Universitesi] at 08:06 03 January 2015

so on. By using these three values, we can calculate the ratios of effect-cost
that are measured by the delivered number and by the delivered quantity.
Step 2. Normalize the ratios of effect-cost
The 729 effect-cost ratios can be classified into 81 groups (see column
1, 2 and 3 in Table 4), each with 9 transshipment rules. For each group, the
effect-cost ratios in the same group are divided by their maximum ratio.
Step 3. Calculate the average of the normalized value for the same transshipment
rule under certain conditions
Once the normalized values are figured out, the performances of
transshipment rules can be compared according to scenarios and parame-
ters for meaningful analyses.
Only the normalized values that were averaged according to the
number of retailers and parameters are listed in this study (see Table 8).
The normalized values that were averaged according to the experiment
design are excluded owing to the large amount of data. In Table 8, the
symbols F, V and S indicate three types of customer demand; N means
the performance is measured by the delivered number in a certain period;
and Q means the performance is measured by the delivered quantity in a
certain period. It can obviously be seen that the priority of transshipment
rules are represented by the average of normalization values under
different parameters combination. Here we describe the values in the total
average column only.
Downloaded by [Selcuk Universitesi] at 08:06 03 January 2015

Table 8
1278

Average of normalized values of rules under different situations

Rules F-N F-Q V-N V-Q S-N S-Q


5 10 20 5 10 20 5 10 20 5 10 20 5 10 20 5 10 20
1 0.948 0.981 0.989 0.928 0.981 0.983 0.990 0.982 0.991 0.976 0.975 0.986 0.960 0.994 0.978 0.967 0.993 0.978
2 0.965 0.921 0.992 0.948 0.923 0.988 0.937 0.929 0.960 0.925 0.923 0.949 0.969 0.963 0.980 0.981 0.965 0.977
3 0.895 0.800 0.951 0.898 0.823 0.951 0.864 0.818 0.833 0.890 0.851 0.859 0.595 0.578 0.702 0.607 0.612 0.707
4 0.954 0.924 0.989 0.936 0.925 0.983 0.947 0.934 0.955 0.930 0.926 0.943 0.956 0.948 0.972 0.949 0.947 0.969
5 0.954 0.982 0.986 0.934 0.982 0.980 0.999 0.999 0.995 0.983 0.991 0.988 0.955 0.975 0.978 0.960 0.973 0.974
6 0.713 0.777 0.914 0.663 0.772 0.894 0.870 0.832 0.844 0.895 0.856 0.862 0.493 0.544 0.750 0.480 0.565 0.750
7 0.956 0.989 0.987 0.937 0.989 0.981 0.994 0.983 0.996 0.980 0.975 0.991 0.969 0.982 0.979 0.965 0.981 0.975
8 0.890 0.812 0.949 0.892 0.838 0.950 0.864 0.805 0.838 0.891 0.848 0.867 0.592 0.571 0.699 0.592 0.603 0.703
9 0.899 0.815 0.945 0.899 0.830 0.940 0.876 0.818 0.838 0.900 0.861 0.866 0.527 0.524 0.671 0.520 0.549 0.672
Rules Average of F Average of V Average of S Average of N Average of Q Total average
5 10 20 5 10 20 5 10 20 5 10 20 5 10 20 5 10 20
1 0.938 0.981 0.986 0.983 0.979 0.988 0.963 0.993 0.978 0.966 0.986 0.986 0.957 0.983 0.982 0.961 0.984 0.984
2 0.956 0.922 0.990 0.931 0.926 0.954 0.975 0.964 0.979 0.957 0.938 0.977 0.951 0.937 0.971 0.954 0.937 0.974
3 0.896 0.812 0.951 0.877 0.834 0.846 0.601 0.595 0.704 0.785 0.732 0.829 0.798 0.762 0.839 0.791 0.747 0.834
4 0.945 0.925 0.986 0.939 0.930 0.949 0.953 0.947 0.970 0.952 0.936 0.972 0.939 0.933 0.965 0.945 0.934 0.968
5 0.944 0.982 0.983 0.991 0.995 0.991 0.957 0.974 0.976 0.969 0.985 0.987 0.959 0.982 0.980 1.964 0.984 0.983
6 0.688 0.775 0.904 0.883 0.844 0.853 0.487 0.554 0.750 0.692 0.718 0.836 0.680 0.731 0.835 0.686 0.724 0.836
7 0.946 0.989 0.984 0.987 0.979 0.993 0.967 0.981 0.977 0.973 0.985 0.987 0.961 0.981 0.982 0.967 0.983 0.985
8 0.891 0.874 0.949 0.877 0.874 0.853 0.592 0.578 0.701 0.782 0.769 0.829 0.791 0.782 0.840 0.787 0.776 0.834
9 0.899 0.859 0.9-42 0.888 0.859 0.852 0.523 0.532 0.671 0.768 0.748 0.818 0.773 0.752 0.826 0.770 0.750 0.822
J. N. HWANG
A NOVEL APPROACH TO FIND OPTIMAL TRANSSHIPMENT RULES 1279

Note the values of the three columns in total average. There exist
some interesting phenomena. We can find that some transshipment rules
always perform better, for instance, rule 1, rule 5, and rule 7. The next are
rule 2 and rule 4. The values of the other transshipment rules are always
the smaller. The phenomenon seems similar to the one in Relative method,
although the performances of transshipment rules are a little different in
the same group.

6. Conclusion
In order to compare the differences between the two evaluating
methods, we put the evaluation results of these two methods together
in Table 9 and rank them by priority. It can be seen from Table 9 that
although the ranking order of the nine transshipment rules are not all the
same under the three environments, the variance is very small. Rule 1,
rule 5 and rule 7 always appear in the first three orders; rule 2 and rule 4
Downloaded by [Selcuk Universitesi] at 08:06 03 January 2015

always appear in the order of 4 and 5. The order variance of the other
transshipment rules is much greater, but they always rise and fall from
order 6 to order 9.
Another phenomenon can be seen from Table 9: that is, the relation-
ship between the principles of rules design and the performances of rules.
The rules in the first three order (Group 1) are designed by considering
the ability of the retailers at the time transshipment takes place; the
transshipment rules in the fourth and fifth order (Group 2) are designed by
considering the predicted ability of retailers at the end of a certain period.
Finally, the transshipment rules located in the last 4 orders (Group 3) are
designed by considering the known ability of retailers, that is, the past
ability of retailers.
There is one thing that must be declared. We only considered the
logic that could be used to rank orders in designing transshipment rules;
they were not classified in advance. After a large number of systematic
simulations and measurement of the results by the two methods, we
find the performances of transshipment rules are related to which kind
of retailer ability is considered in designing the transshipment rules (see
Table 10). The transshipment rules which are designed by considering the
ability of retailers at the time the transshipment takes place performed the
best, followed by those which are designed by considering the predicted
ability of retailers at the end of a certain period. Finally, the transshipment
rules which are designed by considering the past ability of retailers
performed the worst.
Downloaded by [Selcuk Universitesi] at 08:06 03 January 2015

1280

Table 9
Comparison of results measured by two methods

# of retailers = 5 # of retailers = 10 # of retailers = 20

Priority Group Ratio Relative Ratio Relative Ratio Relative

Rule Ratio Rule Sum of Rule Ratio Rule Sum of Rule Ratio Rule Sum of
equivalences equivalences equivalences

1 7 0.967 5 88 1 0.984 7 89.5 7 0.985 1 86

2 1 5 0.964 1 82 5 0.984 5 85 1 0.984 7 85.5

3 1 0.961 7 80.5 7 0.983 1 81.5 5 0.983 5 84

4 2 0.954 2 78.5 2 0.937 2 66.5 2 0.974 2 75.5

5 4 0.945 4 64 4 0.934 4 64.5 4 0.968 4 71

6 3 0.791 9 44 8 0.776 3 55.5 6 0.836 6 38

7 8 0.787 8 39.5 9 0.752 8 44.5 8 0.834 8 34.5

8 9 0.771 6 38.5 3 0.747 9 36 3 0.834 3 33

9 6 0.686 3 25 6 0.724 6 27 9 0.822 9 28.5


J. N. HWANG
A NOVEL APPROACH TO FIND OPTIMAL TRANSSHIPMENT RULES 1281
Table 10
Classification of transshipment rules

Group priority Rules Basis of rules design

1 rule 1, rule 5, rule 7 The ability of retailers at the time that transship-
ment begins

2 rule 2, rule 4 The predicted ability of retailers at the end of a


period

3 rule 3, rule 6, rule 8, The ability of retailer performance


rule 9

The phenomenon of logical steady-going clustering is obtained


through the integration of many indices and two measurement methods,
so we classify the nine transshipment rules that we propose in this study
into three groups. The priority of the three groups is consistent in the
Downloaded by [Selcuk Universitesi] at 08:06 03 January 2015

distribution system which consists of a different number of retailers. It is


worthwhile in practical use and for future research.

7. Future research
Maybe transshipment is not indispensable in supply chain manage-
ment, but it is an effective means to improve customer service and reduce
total system cost when the distribution center acts as a coordinator as
well as lateral transshipment is permitted in the system. It is valuable
research to design more economic transshipment rules and to measure the
performance of transshipment rules by using more practical and specific
measuring methods. In this study, we only consider customer service level
and cost in measurement system performance, and the scope is limited
to emergent and complete transshipment. This study can be succeeded
by preventive and partial transshipments while the delayed time can be
considered as the criterion in performance measuring.

References
[1] S. Axsäter (1990), Modeling emergency lateral transshipments in
inventory systems, Management Science, Vol. 36, pp. 1329–1338.
[2] M. A. Cohen, P. R. Kleindorfer and H. L. Lee (1986), Optimal stocking
policies for low usage items in multi-echelon inventory systems,
Naval Research Logistics Quarterly, Vol. 33, pp. 17–38.
[3] M. Dada (1992), A two-echelon inventory system with priority
shipments, Management Science, Vol. 38, pp. 1140–1153.
1282 J. N. HWANG

[4] C. Das (1975), Supply and redistribution rule for two-location in-
ventory systems: one period analysis, Management Science, Vol. 21,
pp. 765–776.
[5] E. B. Diks and A. G. De Kok (1996), Controlling a divergent two-
echelon network with transshipments using the consistent appro-
priate share rationing policy, International Journal of Production and
Economics, Vol. 45, pp. 369–379.
[6] D. Gross (1963), Centralized inventory control in multi-location
supply systems, in Multistage Inventory Models and Techniques, H.E.
Scarf, D.M. Gilford and M.W. Shelly (editors), Stanford University
Press, Stanford, CA, pp. 47–84.
[7] H. Hoadey and D. P. Heyman (1977), A two-echelon inventory
model with purchases dispositions shipments, returns and trans-
shipments, Naval. Res. Legist. Quart., Vol. 24, pp. 1–19.
[8] H. Josson and E. A. Silver (1987), Analysis of a two-echelon in-
ventory control system with complete redistribution, Management
Downloaded by [Selcuk Universitesi] at 08:06 03 January 2015

Science, Vol. 33, pp. 215–217.


[9] H. L. Lee and C. Billington (1993), Material in decentralized supply
chains, Operations Research, Vol. 41, pp. 835–847.
[10] H. L. Lee (1987), A multi-echelon inventory model for repairable
items with emergency lateral transshipments, Management Science,
Vol. 33, pp. 1302–1316.
[11] J. N. Hwang and C. E. Lee (2003), A Study of Two-echelon distri-
bution system with multiple demand within one period, Industry
Forum, Vol. 4, pp. 65–96.
[12] J. N. Hwang and C. E. Lee (2005), Two-echelon distribution system
with multiple demands, Journal of Statistics & Management Systems,
Vol. 8 (2), pp.225–260.
[13] C. C. Sherbrooke (1992), Multi-echelon Inventory systems with
lateral supply, Naval Research Logistics, Vol. 39, pp. 29–40.
[14] G. Tagaras (1989), Effects of pooling on the optimization and service
levels of two-location inventory systems, IEEE Tran., Vol. 21, pp. 250–
257.
[15] G. Tagaras (1999), Polling in multi-location periodic inventory distri-
bution system, The international journal of Management Science, pp. 39–
59.

Received July, 2009

You might also like