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

net/publication/235261785

Measuring the Bullwhip Effect in the Supply Chain

Article  in  Supply Chain Management · May 2000

DOI: 10.1108/13598540010319993

CITATIONS READS

194 1,946

2 authors, including:

Jan C Fransoo
Kühne Logistics University
165 PUBLICATIONS   3,063 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Supply risk management View project

Supply Chain Planning and Inventory Management View project

All content following this page was uploaded by Jan C Fransoo on 29 July 2014.

The user has requested enhancement of the downloaded file.

 Research paper Introduction

Measuring the bullwhip The ``bullwhip'' effect (Lee et al., 1997a,

1997b) refers to increasing variability of
effect in the supply demand further upstream in the supply chain
chain (Forrester, 1961). For example, HolmstroÈm
(1997) reports a supply chain where
variability, as measured by the standard
Jan C. Fransoo and deviation of weekly demand relative to
Marc J.F. Wouters average weekly demand, increases from 9 to
29 for two different product groups going
from consumer demand to plant supply.
Providing the supplier upstream with EPOS
(electronic point of sale) data can reduce this
bullwhip effect. Such information cuts short
all kinds of information distortions in a supply
The authors chain, which often lead to a bullwhip effect.
Jan C. Fransoo is Assistant Professor in the Department
Other improvements that can reduce the
of Technology Management, Technische Universiteit
bullwhip effect include the reduction of lead
Eindhoven, The Netherlands.
times, revising reorder procedures, limiting
Marc J.F. Wouters is Associate Professor in the
price fluctuations, and the integration of
Department of Technology Management, Technische
planning and performance measurement (Lee
Universiteit Eindhoven, The Netherlands.
et al., 1997a 1997b; Towill, 1996; Lee and
Billington, 1992).
This paper was motivated by a project
Keywords
(named SNEL) we conducted in two chains
Supply-chain management, Food, Data processing of convenience foods, namely salads and
ready-made pasteurized meals. In both
Abstract chains, the products are manufactured and
Increased demand variability in supply chains (the stored at a single manufacturing site, then
bullwhip effect) has been discussed in the literature. The cross-docked at a third-party logistics service
practical measurement of this effect, however, entails provider, stored at a regional DC of a
some problems that have not received much attention in supermarket chain and then transported to
the literature and that have to do with the aggregation of the retail franchisees. The objective of the
data, incompleteness of data, the isolation of demand project was to reduce the bullwhip effect by
data for defined supply chains that are part of a greater supplying EPOS data to the parties upstream
supply web. This paper discusses these conceptual in the supply chain. Exchanging EPOS data is
measurement problems and discusses experiences in considered to be an important enabler for
dealing with some of these problems in an industrial improving operations in food supply chains.
project. Also presents empirical results of measurements (This is discussed in the literature on ECR in
of the bullwhip effect in two supply chains. particular. See, for example, the special issue
of Supply Chain Management, Vol. 2 No. 3,
and the report by Kurt Salmon Associates,
Electronic access
1993.) However, actually measuring the
The current issue and full text archive of this journal is bullwhip effect and using EPOS data in
available at supply chains proved to be problematic.
http://www.emerald-library.com Many practical issues are caused by
limitations of information systems, but there
are also three conceptual measurement issues
that we encountered.
First, we experienced that exactly the same
basic data can lead to different measurements
of the bullwhip effect, dependent on the
sequence of aggregating these data in the
Supply Chain Management: An International Journal
Volume 5 . Number 2 . 2000 . pp. 78±89 analysis. Without understanding this and, if
# MCB University Press . ISSN 1359-8546 required, defining a common approach,
78
 Measuring the bullwhip effect in the supply chain Supply Chain Management: An International Journal
Jan C. Fransoo and Marc J.F. Wouters Volume 5 . Number 2 . 2000 . 78±89

different people in one project may easily information and the delay of transferring
choose different ways. It also raised the physical products through the supply chain
question: what is the most suitable way of (lead times). We also refer to Van Aken
aggregating the demand information? In our (1978) for an early empirical study of
experience this depends on how that industrial dynamic control problems.
information will be used if provided. We will According to Lee et al. (1997b), the term
list the various ways in which the aggregation ``bullwhip effect'' was first used by Procter &
sequence can be conducted and identify Gamble, when they experienced extensive
which sequence of aggregation to use under demand amplifications for their diaper
which circumstances. Second, in order to product ``Pampers''. Lee et al. (1997a, 1997b)
assess the possible impact of transferring describe the bullwhip effect as the result of
EPOS data, we needed to filter out the part of information distortion in a supply chain,
the bullwhip effect that was actually caused by where companies upstream do not have
a lack of undistorted information at various information on actual consumer demand.
echelons in the supply chain. This proved to Consequently, their ordering decisions are
be very difficult in the supply chain we based on the incoming orders from the next
studied and we have reasons to think that this downstream company. This may lead to
is true for most projects in which a detailed amplified order variability: demand coming in
measurement of the bullwhip effect is needed. from a downstream company has a lower
Third, any supply chain is part of a greater variability than demand to an upstream
supply web. The divergent supply chain company. Figure 1 depicts the orders received
under study cannot be completely isolated (line) for a particular salad at a distribution
from the remainder of the web. One of the center in the SNEL project and the orders
difficulties that we encountered when placed (bars) by the DC at the manufacturer
measuring the bullwhip effect is that the size for the same salad. It clearly shows an
of the flow of goods at any upstream party is increase in variability between the two order
much larger than the size at any of the patterns. A measure for this bullwhip effect is
downstream parties. This is due to the fact the variability of upstream demand ±
that only a subset of the web of, for example, measured by the standard deviation of
the manufacturer is considered in this supply demand relative to mean demand ± divided by
chain. However, a full analysis of this feature the variability of downstream demand. A
needs wider research attention. value for this measure greater than one
The difficulties associated with the indicates amplified order variability.
measurement of the bullwhip effect have been Lee et al. (1997a, 1997b) have identified
motivated initially by experiences in industrial four major causes of the bullwhip effect:
practice. In the next section, we will position (1) Demand forecast updating. Links in the
the bullwhip effect, and measure it, in the supply chain base the expectations about
literature. Then, as described above, we will future demand on orders they receive
address the various measurement challenges from the succeeding link. An increase in
in section 3 of the paper. Section 4 is about orders leads to higher demand forecasts,
our experiences in dealing with these which is transferred to the next link by
measurement challenges in a project in increased order quantities. That link also
industry. The last section contains the sees an increase in demand, updates its
conclusions of this paper, as well as some forecasts and distorts information for the
suggestions for further research. subsequent link. It works in the reverse
way when end customer demand
decreases. A possible solution for this is to
Measuring bullwhip effects make data on consumer demand directly
available to companies further upstream
The first research to extensively study the in the supply chain. Further, a single
amplification of demand information in a source of forecasting can be determined
supply chain was reported by Forrester in his for the entire supply chain.
seminal book Industrial Dynamics (Forrester, (2) Order batching. Demands come in,
1961). He basically reduces the causes of this depleting inventories but the company
demand amplification to two types of delay, may not immediately place an order with
namely the delay of transferring demand its supplier. It often accumulates
79
 Measuring the bullwhip effect in the supply chain Supply Chain Management: An International Journal
Jan C. Fransoo and Marc J.F. Wouters Volume 5 . Number 2 . 2000 . 78±89

Figure 1 Example of increased demand variability

demands before issuing an order, for There are many conceptual issues related to
instance because of fixed order costs or measuring the bullwhip effect that are hardly
distribution efficiency. Consider a discussed in the literature. First, different
company that receives daily orders but choices can be made regarding the way data
places orders with its suppliers once a are aggregated. Neither current literature nor
week. Variability of orders placed with the business practice is clear about this matter.
suppliers is higher than the demands the For example, in the paper by HolmstroÈm
company itself faces. In addition to (1997) it is not indicated at which level of
making consumer demand data available detail the measurements have been carried
throughout the chain, reducing batch out. Sometimes the issue is that available data
sizes and increasing order frequencies are too aggregated. For example, at retail
assist in decreasing this effect. Reducing outlets, point-of-sale information is often
batch sizes can be enabled by using EDI aggregated directly (before being transferred
to reduce administrative ordering costs to a database) into daily or weekly demand.
and using third-party logistics service This may hamper an analysis for which hourly
providers to make less-than-full demand is needed.
truckloads economically viable. Second, measuring the total bullwhip effect
(3) Price fluctuations. Because of promotions does not tell which of the different causes
and trade deals, the price of a product contributes most and which solutions are
fluctuates, which increases variability of most relevant. For example, to assess the
demand. When the price of a product is possible benefits of exchanging demand
low, a customer buys in bigger quantities information, it is important to be able to
than needed. When the price returns to measure which part of the bullwhip effect is
normal, the customer buys less than due to incomplete demand information in a
needed to deplete its inventory. particular supply chain. Benefits for
Stabilizing prices and decreasing the transferring EPOS data must be clear before a
number of promotions is a way of system is set up to make this possible. In
reducing this effect. many cases a significant investment in
(4) Rationing and shortage gaming. When information systems is needed to collect and
product demand exceeds supply, a process the data. Furthermore, EPOS data is
supplier needs to ration its product to very valuable marketing information, that
customers. Knowing that, customers may retailers may not want to make available to
order more than they really need. Later, the manufacturer. By measuring the bullwhip
when there are no shortages, orders effect at various echelons in the supply chain,
disappear. Introducing rationing methods we may be able to find out at which echelon
based on past sales rather than on orders sharing information is particularly useful. By
placed takes away the incentive for specifying at which level of aggregation the
customers to inflate order sizes. information is useful, the various companies
80
 Measuring the bullwhip effect in the supply chain Supply Chain Management: An International Journal
Jan C. Fransoo and Marc J.F. Wouters Volume 5 . Number 2 . 2000 . 78±89

may find that a particular level of aggregated same as the demand going into echelon l,
data may be exchanged to dampen the since echelon l may have customers outside
bullwhip effect and to improve operations, the specified supply chain. See Figure 2.
while detailed EPOS data are not exchanged We measure the bullwhip effect at a
if retailers want to protect that for marketing particular (set of) echelon(s) in the supply
purposes. chain as the quotient of the coefficient of
Third, different choices can be made variation of demand generated by this (set of)
regarding disaggregation of data, in particular echelon(s) and the coefficient of variation of
because any company in the supply chain to demand received by this echelon:
be analyzed may itself be part of other supply cout
chains. For instance, a manufacturer of !ˆ
cin
peanut butter may supply grocery chain A,
where a project is started to reduce the where
bullwhip effect. Variations in demand at the …Dout …t; t ‡ T ††
manufacturer are however influenced by the cout ˆ …1†
…Dout …t; t ‡ T ††
other customers of the peanut butter
manufacturer. How can aggregate and
information be disaggregated to obtain insight …Din …t; t ‡ T ††
into a particular supply chain, namely the cin ˆ …2†
…Din …t; t ‡ T ††
chain with grocery chain A?
Dout …t; t ‡ T † and Din …t; t ‡ T † are the
demands during the time interval …t; t ‡ T †
Measurement issues and will be written as Dout and Din in the
remainder of this paper.
In this section, we discuss three issues in
measuring the bullwhip effect: first, the Aggregation
sequence of aggregation of demand data; A straightforward measurement of the total
second, filtering out the various causes of the bullwhip effect requires data regarding all
bullwhip effect; and finally, the inconsistency demands and orders for a specific supply
in demand basis due to the fact that a studied chain at the required level of detail. The level
chain is always part of a greater web. of detail that is required depends on how
We will first introduce some terminology. undistorted demand data will be used by the
We consider a supply chain as part of a web of different companies involved. For now, let us
independent units (companies) (Fransoo et suppose demand information is available at
al., 1999). Considering a supply chain as a the following level: at each echelon, for each
subset of a web implies that each unit in the outlet, for each product, for each day.
supply chain under consideration may have There are several ways to aggregate demand
relations with some or multiple other units or
data for calculating the bullwhip effect. The
supply chains. A supply chain consists of
key difference between various ways of
several subsequent echelons. An echelon is
aggregation is the level of aggregation of
one level in the supply chain. An echelon may
demand data Din and Dout at which the
consist of several parallel outlets, e.g. several
standard deviation of demand is determined.
distribution centers may together form an
If the standard deviation of demand is
echelon ``DC''; several stores may together
determined for a demand series at the product
form an echelon ``retail shops''.
In the discussion below, we will consider a level, for each outlet of an echelon, the
supply chain, consisting of L echelons, coefficient of variation is likely to be higher
identified by index l, with (l =0 being the most than if it were determined for a demand series
upstream echelon. Each echelon consists of at the product level, but aggregated for the
Ml outlets, indicated by index ml Further, we entire echelon, not distinguishing between
distinguish between demand coming from the demand data at each outlet. Based on this, we
next downstream echelon (Dinl ) and demand can distinguish four levels of aggregation at
going out to the next upstream echelon which the standard deviation of demand can
(Doutl ). Demand is usually effectuated by be determined, assuming there are P products
placing orders. Note that the sum of demands and M outlets in the supply chain under
going out of echelon l+1 is not necessarily the consideration.
81
 Measuring the bullwhip effect in the supply chain Supply Chain Management: An International Journal
Jan C. Fransoo and Marc J.F. Wouters Volume 5 . Number 2 . 2000 . 78±89

Figure 2 Demand information at different echelons in a supply chain

(1) product/outlet (!1 ): this is the most differences may results in the final number, as
detailed analysis, determining the is illustrated in Table I. Demand data in that
standard deviation for all available table are random and serve to illustrate the
demand series, resulting in P  M different measurements and the different
standard deviations, and P  M bullwhip results that can be obtained.
measurements. . In measurement 1 the bullwhip effects are
(2) product (!2 ): demand per product is calculated for four product/outlet
aggregated over the outlets and this combinations and the overall bullwhip
indicates the variability in demand of a effect (!1 of 4,495 is the average,
product at the entire echelon, not weighted by average demand.
distinguishing between individual outlets, . In measurement 2 the effects are
and also assuming pooling between the calculated for two products, after
outlets. This results in P bullwhip aggregating product demand across the
measurements. two outlets. The overall bullwhip effect
(3) outlet (!3 ): aggregated over the products, (!2 of 3,642 is the average of the two
this indicates the variability in demand of measurements, weighted by average
an outlet, not distinguishing between demand.
individual products. This requires that . In measurement 3 the effects are
the product demands be added up, e.g. calculated for the two outlets, after
by using some kind of weighing factor. aggregating demand per outlet across the
This results in M bullwhip two products. The overall effect (!3 of
measurements. 4,636 is the average of the two
(4) echelon (!4 ): aggregated over the outlets measurements, weighted by demand per
and products, the variability of total outlet.
demand at the echelon can be . In measurement 4 demand is first
determined. Different product demands aggregated across both products and
can be added up using a weighing factor. outlets and the bullwhip effect of total
This results in 1 bullwhip measurement. demand (!4 is calculated and is 4,712.
Depending on the aggregation level, at which Table I demonstrates that from the same
the standard deviation is determined, many demand data, different bullwhip effects can
different bullwhip measurements result. If be calculated, depending on the way the
this needs to be added up into a single demand data are aggregated.
number, the overall bullwhip effect, a The main consideration for choosing a
(weighted) average of the individual specific type of aggregation is the particular
measurement can be taken. Considerable problem that is caused by demand
82
 Measuring the bullwhip effect in the supply chain Supply Chain Management: An International Journal
Jan C. Fransoo and Marc J.F. Wouters Volume 5 . Number 2 . 2000 . 78±89

Table I Different measurements of the bullwhip effect, based on different aggregation levels
Day
1 2 3 4 5  = !
Measurement 1: product/
outlet
Outlet A:
Product 1
Din 5 4 6 4 6 5 0,200
Dout 2 6 8 3 6 5 0,490 2,449
Product 2
Din 4 6 5 5 5 5 0,141
Dout 1 9 2 6 7 5 0,678 4,796
Outlet B:
Product 1
Din 14 15 14 13 14 14 0,051
Dout 12 13 10 18 17 14 0,242 4,796
Product 2
Din 15 16 14 15 15 15 0,047
Dout 16 17 11 19 12 15 0,226 4,796
Average: 4,495 (!1 )
Measurement 2: product
Product 1
Din 19 19 20 17 20 19 0,064
Dout 14 19 18 21 23 19 0,178 2,769
Product 2
Din 19 22 19 20 20 20 0,061
Dout 17 26 13 25 19 20 0,274 4,472
Average: 3,642 (!2 )
Measurement 3: outlet
Outlet A:
Din 9 10 11 9 11 10 0,100
Dout 3 15 10 9 13 10 0,458 4,583
Outlet B:
Din 29 31 28 28 29 29 0,042
Dout 28 30 21 37 29 29 0,197 4,655
Average: 4,636 (!3 )
Measurement 4: echelon
Din 38 41 39 37 40 39 0,041
Dout 31 45 31 46 42 39 0,191
Average: 4,712 (!4 )

fluctuations, in a particular supply chain demand per supermarket. This is an example

under investigation. The measurement should of !3 . As another example, suppose that a DC
give information about the causes of demand experiences the problem of having to discard
fluctuations. For example, suppose that too many of some perishable products. High
supermarkets experience difficulties in safety stocks are needed in the DC, because
managing the flow of incoming goods because orders from supermarkets vary greatly. Here
volumes of different deliveries fluctuate we are interested in amplification of demand
greatly. The relevant aggregation for this variability per product by all supermarkets to
problem is total volume per supermarket of which the DC supplies. This is an example of
orders placed upstream to the DC and the !2 . As a third example, imagine that a
measurement should explain to what extent supplier has problems producing all orders
there is an amplification of demand received from a particular DC, because total
fluctuation compared to total consumer demand varies greatly from week to week. We
83
 Measuring the bullwhip effect in the supply chain Supply Chain Management: An International Journal
Jan C. Fransoo and Marc J.F. Wouters Volume 5 . Number 2 . 2000 . 78±89

would want to find out whether there is For each echelon, it is important to analyze
amplification of demand variability at the whether Dinl and Doutl are comparable.
level of total volume, so the relevant measure . Doutl may consist of orders resulting from
is !4 . more demand than just the demand of a
Underlying each of these measurements is defined chain Dinl . For example, a supply
the question of aggregation across time: chain for one type of retail outlet is
demand per hour, per day, per week, etc. This investigated and Dinl consists of demand
is the variable T in Equations 1 and 2. The from only these retail outlets. However,
heading of Table I indicates demand per day, the DC delivers the same products to
but, of course, the previous discussion of several other types of retail outlet as well
aggregation across products and/or outlets and places orders to suppliers …Doutl † to
would have been the same for demand per satisfy total demand. Consequently, the
hour, per week or any other period. Again, the bullwhip measured using the available Din
appropriate aggregation across time depends and Dout may contain amplifications
on the problem investigated. If, for example, caused by the chain that is not under
the supplier has daily production planning, consideration, or alternatively may be less
then demand amplification per day is than expected because of pooling effects.
relevant. But if the supplier plans production . Information on Doutl or Dinl may be
for a whole week (e.g. if production cycles incomplete. For example, demand data
exist), then a measurement at the week-level from only some retail outlets are available
is relevant. In that situation, demand could be (a sample of Dinl , rather than Dinl as a
highly erratic on a daily basis, but that does whole), while demand data of the DC are
not affect the supplier if they look at weekly completely available (complete Doutl ).
demand. While extrapolation of a sample of data
When measuring bullwhip effects in real regarding average demand may be
supply chains, the use of demand data should possible, it is conceptually less likely to be
be discussed to define the best way to gather possible regarding the standard deviation
and analyze data. This is particularly relevant, of demand, since unforeseen pooling
when different people are involved who, effects may occur.
without explicit consideration of these issues, For the supply chain, the relationship
may find themselves having inconsistent data between Dinl and Doutl‡1 of two subsequent
and analyses. echelons needs to be considered. Following
demand throughout the chain is difficult, if
Incomplete data and isolating supply Dinl consists of more demand than just the
chain demand defined chain (based on Doutl‡1 ) as in Figure
In many supply chains, demand data are not 2, or if Doutl‡1 concerns more than just the
available at the level of detail discussed in the defined chain while Dinl is precisely the
previous section. Instead, data are sometimes defined chain.
incomplete or only available at a highly
aggregated product or time level. For Understanding the different causes
example: daily data are needed per product, In the previous sections, we discussed the
while weekly demand data per product group measurement of the total bullwhip effect.
are available. Also, isolating a specific supply Measurement should provide us with insights
chain can be impossible, because each as to specific causes of the bullwhip effect.
company is actually part of a supply web and Singling out the specific causes as identified
has both suppliers and customers outside a by Lee et al. (1997a, 1997b), helps us to
specific supply chain that is investigated and assess benefits that may result from specific
where the bullwhip effect is measured. Since improvement measures to be taken, such as
data are incomplete in many cases, we need to the exchange of EPOS data. Specifying which
consider how to measure the bullwhip effect part of the total bullwhip effect can be
in these instances. Measuring the bullwhip attributed to a specific cause can be
effect with incomplete data creates two problematic because of:
concerns: consistency between Din and Dout . Unclear ordering policies, meaning that
for each echelon and consistency between order batching takes place but rules for
different echelons of one chain. this are unclear so rearranging data to
84
 Measuring the bullwhip effect in the supply chain Supply Chain Management: An International Journal
Jan C. Fransoo and Marc J.F. Wouters Volume 5 . Number 2 . 2000 . 78±89

account for the effect would not be organization has seven DCs in The
possible. Netherlands; one of these DCs was part of the
. No data on real demand. Separating out SNEL project. For the meals, this DC keeps
the effect of price fluctuations and of an inventory of several days. Delivery
shortage gaming would require some data frequencies to individual retail franchisees
on real demand compared to sales. vary between one and seven times per week
However, such real demand data will (over 60 per cent of the franchisees have
often not be available. deliveries two-three times per week). The
. No data on shortages. The comparison of retail outlets are independently owned
real demand with sales would be relevant franchise supermarkets. The retail
for moments when shortages occur and organization supplies about 500 retail
prices change. Such information is not franchisees with these ready-made meals.
always available, especially data on The second chain supplies salads. The
shortages and delivery performance are producer has an inventory of several days and
often not recorded systematically. products are coded with a shelf life of about
30 days directly after production. Deliveries
to the retail DC are three times per week (and
SNEL project sometimes more frequently) using the same
third-party logistics company as the meal
In this paper, we present empirical results producer. If there are full truckloads,
based on the industrial project SNEL. There transportation is direct from the producer to
are two kinds of results: first our experiences the DC, otherwise transport is consolidated
in dealing with some of the measurement overnight at the logistics service provider.
issues discussed above in a project in industry,
and second results of measurements of the Measuring the bullwhip effect in the
bullwhip effect in two supply chains. SNEL project
We will first introduce the SNEL project. In the SNEL project, the objective was to
Four companies, several retail stores, measure the bullwhip effect on the basis of
academic staff members of Eindhoven daily demand, because consumer demand can
University of Technology, and five students fluctuate heavily within the week and ordering
were involved. The set-up was similar to and delivering between the links occurred
earlier studies, like the SLIM project several times per week. Measuring weekly
(Wouters and Van Donselaar, 2000; Kopczak demand would cancel out much of the true
and Fransoo, 1999). The objective of this bullwhip effect experienced by the
nine-month project was to investigate how companies. Daily demand throughout the
chain-wide costs could be reduced and/or supply chain was analyzed during a period
service to consumers could be improved by from March 23 until June 5, 1998. This
streamlining processes and better period was chosen because of the expected
coordinating operations, in other words: changes in weather that cause fluctuations in
supply chain management. The project consumer demand (especially for salads) and
focussed on salads and chilled ready meals, we would be able to map how the supply
which are perishable goods that stay fresh for chain responded to such variability. In the
several weeks. salads supply chain, data were aggregated
The two supply chains are depicted in across products and outlets before calculating
Figure 3. In the first chain, the supplier the bullwhip effect (!4 ), while in the meals
produces ready-made pasteurized meals. The supply chain data were aggregated across
supplier has an internal shelf life of three outlets and the measurements per product
weeks due to excellent conditioning rooms. were averaged across products (!3 ). This was
Just before the products are delivered to the because total volume was relevant for
customers, they are coded with a shelf life of planning salads production, while volume per
an additional 21 days. Products are delivered product was relevant for planning the
to each of the retailers' DCs three times a production of meals.
week using a third-party logistics company. Since we were interested in the effect of
Deliveries are never full truckloads and independent demand forecasts throughout
therefore the logistics company consolidates the chain (and the lack of actual demand and
through its DC overnight. The retailer EPOS information), we needed to separate
85
 Measuring the bullwhip effect in the supply chain Supply Chain Management: An International Journal
Jan C. Fransoo and Marc J.F. Wouters Volume 5 . Number 2 . 2000 . 78±89

Figure 3 Two supply chains in the SNEL project

the effect of promotions out of the data set. In effects out of the data. We will discuss the
the period under consideration, there was one difficulties we encountered by eliminating
significant promotion in the salad chain. It each of these causes.
was easily possible to take the effect of the First, filtering the bullwhip effect due to
promotions out of the demand data of the order batching proved to be hard. In some
DCs, since this was a clear spike in demand echelons of the supply chains studied, the
on a single day. However, at the ordering policy was limited by a maximum
manufacturer, that effect of the promotion number of deliveries per week. This means
was much more spread out, so taking out this that during each week, there were one or two
effect was not possible over there. days without any delivery, resulting in zero
In each of the echelons in both chains, we demand. Taking these days out of the
needed to identify the level of aggregation at measurement is the simplest option, but it still
which to measure the demand and order means that the day before and the day after
patterns. In order to properly measure the non-delivery, a relatively high demand exists.
bullwhip effect caused by a single stage, the Further, one may question whether taking
denominator of the division ± i.e. the size of days out is possible if the sample size is much
the average demand ± needs to be the same. larger than the seven franchisees that were
This means that the mean of Doutl must equal studied in the SNEL project. Simply taking
the mean of Dinl at every echelon in the supply the zeroes out of the database does not
chain. For instance, a detailed analysis of suffice, since for each zero, it needs to be
consumer demand at the franchisees was only determined whether this was a zero due to
conducted at a limited number (seven) of non-ordering or a zero due to ordering not
franchisees. To determine the bullwhip effect being possible. Since the number of orders
caused by the franchisees, we only included from the DC to the producer is much less, the
the coefficient of variation in demand based effect of the delivery and ordering policy
on the orders from this subset of franchisees, could be taken out of this echelon. Further
even though complete and detailed ordering batching occurred because a minimum order
information was available for all franchisees. size existed, e.g. in the meals chain, meals
If this is not done, then a possible pooling could only be ordered in multiples of four.
effect is already included in the bullwhip For a single franchisee, this is several days of
measurement and the measurement is demand. It was impossible to filter this effect
inappropriate. out of the data.
Next to selecting the same database at each Second, a bullwhip effect caused by price
of the echelons in each supply chain, the fluctuations rarely occurred in the supply
specific bullwhip cause that we were chains we studied. Based on interviews
interested in, namely the lack of a unitary conducted at the participating companies, we
forecast, had to be filtered out of the total concluded that this was mainly caused by the
bullwhip effect. In order to do this, we short shelf life of the product (especially for
considered each of the other three bullwhip meals) and a consequently increased risk for
causes and attempted to eliminate their the buyer of ordering based on price
86
 Measuring the bullwhip effect in the supply chain Supply Chain Management: An International Journal
Jan C. Fransoo and Marc J.F. Wouters Volume 5 . Number 2 . 2000 . 78±89

fluctuations. In addition, the producer in the demand variability slightly by 1.26 when
salads chain, who induced promotions several placing orders with the supplier. The people
times during the observed period, requires with the supplier who manage finished goods
advanced ordering for promotions, and thus inventory and plan production amplify
these could be easily eliminated from the demand variability by 1.75 when making the
data. production plan. The numbers for the salads
Third, shortage gaming did occur and this supply chain have the same meaning.
was a major problem in the salads chain. The total bullwhip effect is the coefficient of
Especially during sudden increases in variation of the production plan, divided by
temperature, as we were able to observe in the the coefficient of variation of consumer
study, retail franchisees tend to inflate demand. Under specific conditions, this is the
demand in order to make sure that they product of the measured effect at each
receive sufficient products. Since we assumed echelon. Suppose Echelon 3 is the retail
that joint forecasting and transfer of EPOS franchisee, Echelon 2 is the distribution
information would assist in preventing this, center, and Echelon 1 is production, then
we did not eliminate this effect from the data. cout1 cout2 cout3 cout1
ˆ
Should we have wanted to do this, it would cin1 cin2 cin3 cin3 ;
have been a difficult task and would have to
have been based on detailed interviewing of provided there is consistency between Dinl
the decision makers in the supply chain. and Doutl‡1 so cinl ˆ coutl‡1 . However, this
In summary, we managed to rule out part of consistency can only be guaranteed in a
the causes that were related to issues where completely isolated chain. Therefore, we
transferring direct EPOS information would argue that the multiplication is only consistent
not be of help. However, we did encounter in a very small number of cases in practice.
difficulties in measuring this, in particular in Note there is a considerable difference
filtering the effect of order batching and between the two supply chains. The bullwhip
shortage gaming. effects in the salads chain are larger than in
the meals chain (except for production). The
The size of the bullwhip effect in the main reason is that if there is a sudden
SNEL project increase in temperature, people tend to go out
The description of the measurement process and barbecue, using lots of salad. Since the
shows that, given data availability and the exact level of sales is very hard to forecast, the
structure of the two supply chains, the franchisees tend to inflate their orders. If the
measurement of bullwhip effects and its orders do come through and are delivered, the
different causes required some creative franchisees are suddenly stuck with very high
solutions. Being a disregarded issue in the inventory levels of fairly perishable products,
ECR literature, the measurement of the and they react by deflating orders as much as
bullwhip effect cannot be considered as a possible: the birth of the bullwhip effect.
straightforward measurement of means and It appears to be beneficial to transfer EPOS
standard deviations of demand. Results of information due to the size of the bullwhip
measuring the bullwhip effects for both effect, although the benefits in the salads
supply chains in the SNEL project are chain would be much higher. Since the
presented in Table II. In the supply chain for products in both chains have a limited shelf
meals, the retail franchisees amplify demand life, the benefits would be in decreased out-
variability by 1.67 when placing orders with of-stock situations and decreased
the DC. The planners at the DC amplify downpricing for products that are close to the
end of their shelf life. Further, and this was
especially true in the salads chain, a more
Table II Measurements of bullwhip effects per echelon stable production plan would lead to a much
in each supply chain more efficient use of available capacity. If it is
Bullwhip effect (!) beneficial at the chain level to transfer
Echelon Meals Salads information, the next question is how this
Production 1.75 1.23 would affect each of the individual parties in
Distribution center 1.26 2.73 the chain (Fransoo et al., 1999).
Retail franchisee 1.67 2.09 Since the retail franchisees are independent
franchisees, it is very unlikely that a mere
87
 Measuring the bullwhip effect in the supply chain Supply Chain Management: An International Journal
Jan C. Fransoo and Marc J.F. Wouters Volume 5 . Number 2 . 2000 . 78±89

transfer of the EPOS data would improve on Conclusions

the trade-off between downpricing and out-
of-stock. This is due to the fact that a chain- In this paper, we have discussed the
wide trade-off is likely to be very different measurement of the bullwhip effect. A correct
from a local trade-off made by the franchisee. measurement is an essential start to
This can be easily understood by the newsboy investigating problems caused by demand
model, in which a trade-off is made between amplification and to assess which measures
overage cost (cost of having ordered too can be taken to reduce this amplification. We
much) and underage cost (cost of having conclude that:
ordered too little). It is clear that, taking the
. A correct bullwhip measurement needs to
cost structure of the chain as a whole, the be well defined. The correct sequence of
ratio of overage and underage cost is smaller data aggregation needs to be agreed upon
in order to be able to compare data. As
than in the case of the cost structure for the
we have demonstrated in section 3,
individual franchisee. The overage cost for the
considerable differences in measurement
whole chain includes all purchased raw
can occur if a different sequence of
material and labor put into the product, while
aggregation data is applied. Aggregating
the overage cost for the franchisee
data is essential, since the inbound
additionally includes the profit margins and
demand at an echelon and the outbound
the fixed costs of the manufacturer and the
demand at an echelon need to have the
retail organization (DC). Thus, the retail
same quantity basis. Therefore, usually
franchisee is unlikely to change its ordering
on the inbound demand, aggregation
policy without the proper financial incentives
needs to take place. The sequence of data
from the retail organization.
aggregation needs to be based on the
A viable option of reducing the bullwhip is
specific problem that is under
by eliminating the amplification in variability
investigation. For instance, if the chain is
that is caused by the DC. For the salads
designed as a postponed chain, the
chain, the effect of the DC is considerable.
products at the final stage can be
This would mean that transferring EPOS data
aggregated immediately towards the
would be beneficial. Given the discussion
products before the push/pull barrier,
above, and thus assuming the retail franchisee whereas this cannot be done for a chain in
may not be willing to change his ordering which individual product distinctions are
policy if there is no financial incentive to do relevant all the way up to the
so, transferring the (aggregate) retail order manufacturer.
may be a better option to reduce the bullwhip . The measurement needs to be
effect than transferring detailed consumer determined for each echelon separately,
demand. Assuming the same financial such that the benefits of partial solutions
conditions in the meals chain, it may not be may be traded off against benefits of
beneficial at all to transfer any information integral solutions. Each of the echelons
about demand for meals to the manufacturer, may contribute to creating a bullwhip
since the DC hardly causes a bullwhip effect effect to a greater or smaller extent.
in the supply chain. Benefits of transferring Therefore, in order to make a proper
information are likely to be marginal and an trade-off, it is important to distinguish the
investment in information systems to make contribution of each of the echelons in
the data exchange possible would probably the supply chain.
not pay off. . The measurement needs to be filtered,
It pays off to find out the detailed which means that it should be analyzed
mechanics of the bullwhip effect in a specific which parts of the overall effect are the
supply chain. Filtering out the various causes result of the different causes. For
of the bullwhip effect at each of the echelons example, transferring EPOS data is
in the chain, and assuming other policies commonly advised, but is not relevant for
(such as pricing) would not be changed, can all causes. Filtering may hardly be
make the specific benefits of transferring possible in practice, as we have seen in
EPOS data clear and help decide whether an the SNEL project. Difficulties that we
investment in information systems would may come across include how to deal with
pay off. ordering policies.
88
 Measuring the bullwhip effect in the supply chain Supply Chain Management: An International Journal
Jan C. Fransoo and Marc J.F. Wouters Volume 5 . Number 2 . 2000 . 78±89

We have also discussed that demand data may HolmstroÈm, J. (1997), ``Product range management: a case
be incomplete and that isolating demand data study of supply chain operations in the European
grocery industry'', Supply Chain Management, Vol.
for a particular chain can be intractable, since
2 No. 3, pp. 107-15.
companies will usually have suppliers and Kopczak, L.R. and Fransoo, J.C. (1999), ``When industry
customers outside a defined chain under and academia collaborate'', Supply Chain
investigation. This is a relevant area for future Management Review, Vol. 2 No. 4, pp. 68-75.
research. Kurt Salmon Associates Management Consultants (1993),
Efficient consumer response ± Enhancing Consumer
We experienced in a practical study that the Value in the Grocery Industry, January.
theory of measurement of the bullwhip effect Lee, H. and Billington, C. (1992), ``Managing supply chain
in a practical setting has received very limited inventory: pitfalls and opportunities'', Sloan
attention so far. In this paper, we have made a Management Review, Vol. 33 No. 3, pp. 65-73.
Lee, H., Padmanabhan, V. and Whang, S. (1997a),
first start with a method to accurately
"Information distortion in a supply chain: the
document and define various ways of bullwhip effect", Management Science, Vol. 43 No.
measuring the bullwhip effect. 4, pp. 546-58.
Lee, H., Padmanabhan, V. and Whang, S. (1997b), ``The
bullwhip effect in supply chains'', Sloan
Management Review, Vol. 38 No. 3, pp. 93-102.
References Towill, D. (1996), ``Time compression and the supply chain
± a guided tour'', Supply Chain Management, Vol. 1
Forrester, J.W. (1961), Industrial Dynamics, MIT Press, No. 1, pp. 15-27.
Cambridge, MA. Wouters, M.J.F. and Van Donselaar, K.H. (2000), ``Design
Fransoo, J.C., Wouters, M.J.F. and De Kok, A.G. (1999), of operations management internships across
``Operational control in a multi-stage supply chain organizations ± learning OM by doing OM'',
with limited information exchange'', working paper Interfaces, in print.
TM/LBS/9909, Subdepartment of Operations Van Aken, J.E. (1978), ``On the control of complex
Planning and Control, Technische Universiteit industrial organizations'', PhD thesis, Eindhoven
Eindhoven, Eindhoven, The Netherlands. University of Technology, The Netherlands.

89

View publication stats