Int. J.

Production Economics 127 (2010) 103–111

Contents lists available at ScienceDirect

Int. J. Production Economics

journal homepage: www.elsevier.com/locate/ijpe

Forecasting and stock control: A study in a wholesaling context

A.A. Syntetos a,c,n, M.Z. Babai b,d, J. Davies c, D. Stephenson c
a
Centre for OR and Applied Statistics, Salford Business School, University of Salford, Maxwell Building, Greater Manchester M5 4WT, UK
b
BEM Bordeaux Management School, France
c
Centre for Operations Management, Management Science and Statistics, Salford Business School, University of Salford, UK
d
Princess Fatimah Alnijris’s Research Chair for AMT, Industrial Engineering Department, King Saud University, Saudi Arabia

a r t i c l e in fo abstract

Article history: Wholesalers add value to the products they deal with by essentially bringing them closer to the end
Received 12 December 2008 consumers. In that respect, the effective control of stock levels becomes an important measure of
Accepted 1 May 2010 operational performance especially in the context of achieving high customer service levels. In this
Available online 8 May 2010
paper, we address issues pertinent to forecasting and inventory management in a wholesaling
Keywords: environment and discuss the recommendations proposed in such a context in a case study organization.
Forecasting Our findings demonstrate the considerable scope that exists for improving current practices and offers
Inventory insights into possible managerial issues.
Demand categorisation & 2010 Elsevier B.V. All rights reserved.
Wholesaling

1. Introduction the service availability to consumers and reducing the overall

Wholesalers add value to the products they deal with by
essentially bringing them closer to the end consumers or
industrial clients. In that respect, availability becomes a crucial
operational performance measure whilst at the same time
inventory related costs constitute the main investment in this
environment. Consequently, effective stock control represents the
main opportunity for attaining high customer service levels and
generating considerable cost benefits.
The academic literature that deals with inventory manage-
ment related issues in the wholesaling context is rather extensive.
The relevant literature extends back to the 1950s and has
considered many topics ranging from the strategic level decision
making (such as network design and distributional strategies) to
more ‘tactical’ issues such as demand forecasting, inventory level
settings in the warehouses or stores and the replenishment
coordination between them. Nevertheless, despite the huge
literature dealing with this problem, very few papers are actually
considering empirical solution implementation and case studies
that may also offer qualitative insight are lacking.
In this paper, we explore forecasting and stock control related
opportunities for increasing service levels and reducing costs in a
wholesaling context. We analyse pertinent issues through a case
study of a UK-based wholesaler of engineering supplies. This
research work has been launched along with the initiation of a
project in the organisation under concern that aimed at increasing
stock levels. Our work has been conducted in collaboration with a
dedicated working group in that company. We will show, through
this case study, that although a rather ‘basic’ (as compared to the
theoretical advancements in this area) inventory control solution
has been proposed, the projected organisational savings are
substantial. This result is not surprising, but rather it has been
expected since practical in-house inventory management related
applications lag considerably behind the corresponding theore-
tical propositions in this area. In that respect, our work can be
seen as a contribution towards bridging the gap between the
theory and practice of OR in the area of inventory management.
The remainder of this paper is organised as follows: in Section
2, the case study organisation is presented along with a detailed
discussion of the pre-project processes governing stock control. In
Section 3, the relevant literature is overviewed for the purpose of
informing, where possible, our investigation. In Section 4, we
present the development of the proposed forecasting and stock
control solution followed, in Section 5, by details related to the
experimental structure of our investigation. In Section 6, the
empirical performance of the proposed solution is evaluated
against that of the system currently in place. The insights and
implications of our work are discussed in Section 7, followed, in
Section 8, by the conclusions of our paper and the consideration of
some next steps of research.
2. Case study organisation

n
Corresponding author. Tel.: + 44 0 161 295 5804; fax: + 44 0 161 295 5556. Valves Instruments Plus Ltd. (VIP) is a wholesaler of engineer-
E-mail address: a.syntetos@salford.ac.uk (A.A. Syntetos). ing supplies stocking a wide range of valves and ancillary

0925-5273/$ - see front matter & 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.ijpe.2010.05.001
104 A.A. Syntetos et al. / Int. J. Production Economics 127 (2010) 103–111
products and selling primarily to the construction sector. The
company was founded in 1985 and it is situated just outside the
Manchester city centre boundaries. Its customers throughout the
UK range from small to medium size companies to government
funded bodies, universities and major engineering contractors.
The company sells an extensive range of stock items which are
primarily stored in the warehouse ready for dispatch; it also
accepts ‘special’ orders which often involve highly complex
equipment sourced from various suppliers. The company’s supply
base is quite vast, given the wide range of items available in its
catalogue. Currently, supplies are primarily sourced from within
the UK; however, increasing numbers of items are sourced from
Italy and Asia. This is more efficient from an economic standpoint,
but the lead times can exceed in some cases 20 weeks rendering
an effective forecasting and stock control strategy an immediate
necessity.
2.1. Inventory management
Before our project was launched, the stock control process was
managed manually (by the stock control officer) based, essen-
tially, on a periodic re-order point method. The inventory
positions (that comprised only stock on hand, without consider-
ing planned receipts) were reviewed weekly (every Friday); if
stock on hand was below an arbitrarily specified re-order point,
an order was placed for a quantity that represented the average
demand over a certain number of previous weeks. The pre-
specified re-order quantities were suggested years ago by the
Managing Director and never updated since then. This means that
recently introduced codes are not even associated with a
suggested re-order level rendering the ordering process even
more arbitrary in nature. In addition to the weekly re-ordering
system, if stock is urgently needed for a particular item to
facilitate an order, the corresponding transaction would occur on
the day the problem was experienced. The process involved the
use of a paper-based report generated by the data system utilised
by the company, called Opera. Opera also facilitates the company’s
administrative and financial procedures and it supports the whole
information related infrastructure. In terms of stock ordering, data
on all transactions carried out against each item is available.
Similarly, additional information, like current stock levels,
quantity on order, etc., may also become available, if necessary,
in order to inform the ad hoc decisions made by the stock control
officer when re-ordering items.
The details related to each item need to be checked against the
system, as the re-order report contains various inaccuracies, such
as obsolete (no longer traded) items appearing on the report.
(Another software solution exists in the company that could
potentially offer reports that would be more useful in assisting
the stock control officer in the re-ordering process. Such software
though has not been utilised due to the lack of time available to
personnel.) A purchase order is then raised within Opera for the
necessary goods. Item prices are automatically entered by Opera;
the stock officer just enters the quantity required and date.
In the past, staff felt reasonably comfortable with the ordering
processes in place; in the presence of ‘unusual’ problems they
resorted to internal communications in order to resolve them.
That was feasible given the long service of particular individuals
within the organisation and their acquaintance with the im-
perfections of the system. However, as the company recently
started expanding with new staff being employed, the lack of
transparency in the system became far more obvious and the
need for increased levels of automation was exemplified. In
addition, the lack of any degree of science behind that process
facilitated our demonstration, at the conceptual level, of the
potential benefits that a well-informed system would offer to the
company.
3. Literature overview
The academic literature that deals with inventory manage-
ment related issues in the wholesaling context is rather extensive.
The relevant literature has considered many topics ranging from
the strategic level decision making (such as network design and
distributional strategies) to more ‘tactical’ issues, such as demand
forecasting, inventory level settings in the warehouses or stores
and the replenishment coordination between them (e.g. Das and
Tyagi, 1999; Germain and Groge, 1999). Nevertheless, and as
discussed in the previous section of our paper, the case study
organisation serves mainly the construction industry rather than
specific retail outlets. As such, theoretical advancements that
relate to the typical configuration of the relevant systems (i.e. one
wholesaler—many retailers) do not necessarily inform our
empirical investigation and consequently they are not explicitly
considered in this section.
3.1. Demand forecasting and stock control
Forecasting in a supply chain context has attracted a
considerable amount of academic research. This is true both for
fast moving items and, more recently, also for slow movers or
intermittent demand Stock Keeping Units (SKUs) that are
characterised by infrequent demand arrivals (see, for example,
Altay et al., 2008; Porras and Dekker, 2008; Syntetos et al.,
2009a,b; Teunter and Sani, 2009). Such studies are very relevant
for the purposes of our investigation since the great majority of
the SKUs in the case study organisation are intermittent in nature.
Most work on intermittent demand forecasting is based on
Croston’s (1972) influential article, which for many years was
neglected but in the last five years has seen 40 citations (Fildes
et al., 2008). The method was claimed to be unbiased, but despite
its theoretical superiority modest benefits were recorded in the
literature when it was compared with simpler forecasting
techniques, such as Simple Exponential Smoothing, SES (Will-
emain et al., 1994). Some empirical evidence even suggested
losses in performance (Sani and Kingsman, 1997). Syntetos and
Boylan (2001) showed Croston’s method to be biased and they
subsequently proposed an approximately unbiased modification
to Croston’s estimator (Syntetos and Boylan, 2005). This method is
known as Syntetos–Boylan Approximation (SBA) and it is further
discussed in Section 4.3). SBA has been independently assessed on
18,750 SKUs from the Royal Air Force (RAF, UK) and found to
outperform Croston’s estimator (Eaves and Kingsman, 2004).
Independent evidence in support of this method’s utility has also
been provided by Gutierrez et al. (2008). One other modification
proposed in the literature by Levén and Segerstedt (2004) was
found to be even more biased than the original Croston’s method
(Boylan and Syntetos, 2007) and therefore cannot be recom-
mended for practical applications. For a comprehensive review of
advancements in the area of supply chain forecasting in general,
the reader is referred to Fildes et al. (2008).
Some recent research work in this area has also been focusing
on new forecasting approaches that deal with demand aggrega-
tion and the use of combined forecasts, especially in the presence
of demand intermittence, demand seasonality and trend (e.g.
Chen and Boylan, 2007; Viswanathan et al., 2008). This need for
forecasting methods that consider product aggregation has
emerged from the growing assortments of products and the
shorter product life cycles in the wholesaling environment.
Moreover, this may be also due to demand data showing too
 A.A. Syntetos et al. / Int. J. Production Economics 127 (2010) 103–111
high variation and/or other issues related to constructing reliable
forecast models at the individual item level.
Typical stock control approaches discussed in the literature in the
wholesaling context are relying upon traditional well-known meth-
ods such as the min–max (s, S) policy (e.g. Schneider and Rinks, 1991).
Moreover, our experience indicates that in practical situations all
managers too often resort to very basic inventory rules, such as the
re-order point methods where the control parameters are set
arbitrarily. This is also true for the case study organisation.
Pearson (2006) looked at the newspaper wholesaler problem
and investigated the relationship between the two main targets
associated with the ‘optimal’ supply of newspapers (mean
number of unsold newspapers/overage and probability of no
stock-out/underage) and the forecasting method employed to
predict demand as well as their joint impact on the system’s
performance through a case study. By considering the time series
for newspaper demand over a 31-week period, he showed that the
employment of a prediction method in conjunction with a new
technique for adapting targets (that subsequently renders the
prediction method capable of satisfying both targets) resulted in
considerable improvements as compared to the previous situation
according to which a forecasting method would not simulta-
neously achieve low overage and low underage.
The relevance of performance measurement in the wholesaling
environment was also discussed by Ritchie and Kingsman (1985).
They investigated a wholesaling stock system based on a periodic
order-up-to-level policy with weekly reviews and replenish-
ments. The authors studied the implications of various measures
(such as return on sales, return on stock investment, stock turn,
etc.) for setting stock levels, and they showed that some measures
are very sensitive to the level of sales.
4. Solution development
In order to investigate the effectiveness of current practices in
the company with respect to inventory management, empirical
demand data was obviously required. As expected, data collection
ended up being a rather complicated process due to the lack of
integration of the relevant information and the lack of any
previous attempts to collate data that collectively could be used
for stock control purposes. Our objective was to retrieve a
complete historical picture of how the system had been perform-
ing for as long a period as possible. Such a picture would comprise
weekly information on demands received, orders placed and
orders received, stock on hand and backordered as well as
supportive information in terms of initial stocks, unit cost details,
etc. The recommended solution could then be empirically
assessed in a dynamic fashion, i.e. we would be able to evaluate
how the new solution would have performed if it had been used
in this situation.
In this section, the data collection process is first briefly
described. Subsequently, issues related to classification of the
SKUs are discussed followed by our proposed intervention in
terms of forecasting and stock control modelling.
4.1. Data collection
As previously discussed, the company’s administrative and
financial procedures as well as the whole information related
infrastructure are controlled through Opera. The following informa-
tion was extracted from the Opera system via the XRL software.
 Demand and sales histories (volumes of sales, returns).
 Timing of new codes being introduced and sales since their
introduction.
105
 Historical ordered quantities to the suppliers (quantities and
timing).
 Actual weekly quantities in stock and initial quantities;
information retrieval was feasible only from January 2005
onwards.
 Costs and prices for all SKUs.
 Actual lead times (that were expressed in working days). This
information was also retrieved in conjunction with estimation
from purchasing staff.
A series of meetings took place to confirm the validity of this
information. Lead times constitute an approximate average of the
time difference between placing orders and receiving them. Their
expression in working days and the lack of compatibility with the
weekly time periods considered for our analysis introduced also a
new problem that of converting the relevant time periods into
weekly ‘buckets’. A programme was written in Visual Basic to
allow this conversion and the validity of the final outcomes was
checked with the purchasing manager of the company. The data
collection process resulted in 2156 demand histories accompa-
nied by all information needed for a stock control analysis.
At this point it is important to note that there is a number of
items managed by the company, but were not considered in our
investigation. Such items relate to: (i) SKUs that are never held in
stock, but which are on sale. If an order is placed for such items,
they are purchased directly from the suppliers on demand. Such a
procedure reflects short lead times and beneficial contractual
agreements with the relevant suppliers; (ii) the company imports
certain items in bulk getting special prices for doing so, given the
large quantities that such items are being ordered. It was decided
that these items would continue to be ordered in this fashion
reflecting economies-of-scale considerations.
4.2. Demand classification
The above discussed data collection process resulted in a data
set of 2156 SKUs with accompanying information about the unit
costs, the selling price per SKU as well as the actual replenishment
lead times. Weekly demand information was made available for
the period Jan. 2005–Sep. 2007. Demand characteristics varied
considerably amongst the SKUs with some being particularly slow
moving, having only one or two demands recorded over the entire
history. Some SKUs had not been moved at all.
The first stage of our analysis consisted of the classification of
the SKUs for the purpose of developing our understanding
regarding the relevant data and finding appropriate ways of
communicating our ideas with the management. No such analysis
had ever been performed before and no formal records existed of,
say which are the best sellers in the organisation and what are
their contributions to the total turnover. Correspondingly, no
attempts had ever been made to identify obsolete items or SKUs
that are very slowly moving.
Several classification rules have been proposed in the aca-
demic literature to facilitate managerial decision making. It is
common for organizations to classify their SKUs, assigning higher
service-level targets to some segments than others and treating
the different categories in distinct ways as far as forecasting and
stock control are concerned. In particular, ABC (Pareto) type
classifications are often used in practical applications. A Pareto
report lists all SKUs in descending/ascending order, by either
demand frequency, demand volumes, demand values or by
demand profit (Silver et al., 1998) and relevant categories (A, B,
C, etc.) are specified. Rules based on demand value (sales
volumes  selling prices) are perhaps the ones most commonly
encountered.
106 A.A. Syntetos et al. / Int. J. Production Economics 127 (2010) 103–111
For the purpose of this project, a Pareto classification analysis
has been conducted based on the contribution to profit (sales
volumes  net profit, where the latter is calculated as the selling
price minus the cost price). Management felt more comfortable
with the idea of identifying contribution to net profit as apposed
to gross sales for the purpose of identifying the best selling
products. Moreover, it is important to note that frequency based
rules that have often been recommended for use in the context of
slow moving or intermittent demands were also not perceived as
informative by the management. The results are presented in
Table 1; the classes were constructed as to indicate percentage
contribution to profit in increments of 20%.
The results were very surprising to the company; 20% of all
profit comes from just 11 items. There are only 233 items (about
11% of the entire stock base) generating 80% of the net profit.
There is a tremendous number of SKUs contributing literally
nothing to the profit of the organisation. Please also note that in
the majority of cases, such SKUs were not even important in terms
of complementing orders for a range of items. The analysis,
despite its simplicity, was particularly effective in initiating
managerial action: (i) first, it was decided to look closely at the
best sellers on a regular basis and introduce periodic updates of
the Pareto scheme; (ii) the analysis triggered an immediate
response with respect to ‘clearing’ the stock base and discarding
(in the majority of cases at a discounted price) obsolete items.
That aggressive write-off strategy was perceived as one of the
most important outcomes of our project.
Preliminary empirical analysis indicated a wide range of
underlying demand patterns, most of which were intermittent
in nature. Although management was convinced of the value of
more advanced classification schemes that capture demand
characteristics (e.g. Syntetos et al., 2005; Boylan et al., 2007)
and their potential value in facilitating a more effective approach
to forecasting and stock control (in terms of treating in a different
way the resulting categories), the relevant implementation was
decided to be excluded at that initial stage of intervention with
Table 1
SKUs classification with respect to profit.
Classes Number of Cumulative Cumulative
SKUs number of SKUs percentage of SKUs (%) profit (%)
1 11 11 0.51
2 36 47 2.18
3 62 109 5.06
4 124 233 10.81
5 1923 2156 100.00
50
45
40
Frequency (%)
35
30
25
20
15
10
5
0
[1, 2) [2, 3)
the system. This issue is further discussed in Section 7 of the
paper where the natural next steps of research are outlined.
4.3. Forecasting methods
The stock base of the company consists of very few ‘fast’ and
many intermittent SKUs, the latter being associated with sporadic
arrivals of demands, which are interspersed by some (many) time
period where no demand occurs. The distribution of the average
inter-demand intervals across all the SKUs considered in the
sample is indicated in Fig. 1. Inter-demand intervals equal to 1
mean that demand occurs in every single period (week). As the
classification of the SKUs based on their demand pattern and the
subsequent selection of different estimators was not feasible at
this stage, we considered the implementation of estimators across
the entire stock base. In particular two methods were evaluated:
Simple Exponential Smoothing (SES) and an estimator specifically
developed for intermittent demand patterns, the Syntetos–Boylan
Approximation (SBA) (Syntetos and Boylan, 2005).
As discussed in the previous section, the SBA is a bias
deduction modification to Croston’s method and it has been
shown to outperform this estimator both theoretically and
empirically. Given the non-intermittent nature of demand for
some of the SKUs, we opted for the inclusion of SES as well since
this constitutes a typical choice in industrial forecasting applica-
tions. Our decision for experimentation with these methods
reflected a careful preliminary analysis of the demand data set as
well as three main considerations: (i) ease of application and
communication to management; (ii) robustness and (iii) statis-
tical properties of the methods under concern.
When SES is used, the estimate of the demand level Ft made at
the end of period t 1 for the demand in period t is as follows:
Ft ¼ Ft1 þ aðDt1 Ft1 Þ ð1Þ
where Dt  1 is the actual demand in period t  1.
When the SBA method is used, the forecast is given by
 
d Z^ t
Ft ¼ 1 ð2Þ
2 T^ t
with T^ t ¼ T^ t1 þ dðTt T^ t1 Þ and Z^ t ¼ Z^ t1 þ lðZt Z^ t1 Þ being the
Cumulative estimates of the demand interval and size, respectively, updated
by using SES at the end of the demand occurring periods only. If
no demand occurs the relevant estimates remain the same.
20 In both cases, the variance of the forecast error is estimated
40
60
through the smoothed Mean Squared Error (MSE) approach, so
80 that
100
MSEt ¼ bðDt1 Ft1 Þ2 þ ð1bÞMSEt1 ð3Þ
Distribution of the demand intervals
[3, 4) [4, 5) [5, 6) [6, 7) [7, 8) [8, 9) [9, 10) [10, ¥)
Demand Interval (in weeks)
Fig. 1. Demand interval distribution.
 A.A. Syntetos et al. / Int. J. Production Economics 127 (2010) 103–111 107

The parameters a, b, d, l are smoothing constants. With analogue of the NBD and ‘‘although not having a priori support [in
regards to the first three parameters, low smoothing constants are terms of an explicit underlying mechanism, such as that
typically recommended in the literature and used by practitioners characterising compound distributions], the gamma is related to
in an intermittent demand context. In our empirical investigation, a distribution, which has its own theoretical justification, Boylan
these constants are set to 0.05 (please refer to Section 5.2). The (1997, p. 168)’’. The gamma covers a wide range of distribution
smoothing constant value to be used for the smoothed MSE shapes, it is defined for non-negative values only and it is
procedure is fixed to 0.25; preliminary analysis, not reported here, generally mathematically tractable in its inventory control
indicated that these values perform overall very well. applications (Burgin and Wild, 1967; Burgin, 1975; Johnston,
1980). Nevertheless, we recognize that if it is assumed that
demand is discrete, then the gamma can be only an approxima-
4.4. Stock control modelling tion to the distribution of demand.
In order to compute the ordering quantity Q, we use the EOQ
For stock control purposes, we proposed the application of the formula
periodic re-order point (T, r, Q) policy, with the review period rffiffiffiffiffiffiffiffiffiffiffiffi
T¼1. (For ease of presentation we will be simply referring to this 2AmD
Q¼ ð5Þ
policy as (r, Q) for the remainder of the paper.). Periodic order-up- h
to level policies that, according to our opinion, would be far more where A is the ordering cost, h the inventory holding charge and
suitable in such an environment were not further considered. This mD the mean demand over the within-sample initialisation period.
is because management opted for a policy that it would resemble For more details on the experimental structure of our investiga-
to the greatest possible extent current practices utilised in the tion and the within and out-of-sample arrangements, please refer
organisation. To recap, currently stock is being controlled by to the next section of the paper.
essentially a periodic re-order point policy albeit with a judge-
mental (arbitrary) setting of the re-order levels and order
quantities and manual operation (please refer also to Section 2). 5. Empirical investigation
Let us recall that under the periodic (r, Q) policy, time is
treated as a discrete variable. The system is controlled at the end 5.1. Empirical data
of every review period, and if the inventory position falls below
the re-order point r, a quantity Q is ordered. The quantity ordered
As previously discussed, the database consists of the individual
is received after the replenishment lead-time L.1 The re-order
demand histories of 2156 SKUs covering 142 consecutive periods
point is re-computed at the end of every period, whereas the
(weeks) from January 2005 to September 2007. Not all series were
ordering quantity is computed once and for all for each SKU.
considered for experimentation purposes. A considerable number
The parameters of the (r, Q) policy are computed by
of SKUs were found to be obsolete without having moved at all in
considering a Cycle Service Level (CSL) rather than a cost
the last three years, since January 2005. In addition, 315 SKUs
constraint in accordance with the company’s priorities and the
were associated with a considerable initial stock that resulted in
strong competition present in the market they operate in. The CSL
no ordering at all during the demand history. These SKUs were
is the probability of no stock-out during a cycle (the cycle is the
also excluded from our empirical investigation. (Very high initial
time period between two successive demands). The re-order point
stocks constitute also the case for many other SKUs distorting
computed for period t (at the end of period t  1) is denoted by rt
somehow the empirical results of our investigation. This issue is
and is given by
discussed further in the next section.) The screening process
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rt ¼ ðL þ1ÞFt þ F1 ðCSLÞ MSEt ðL þ 1Þ ð4Þ resulted in 1109 SKUs being considered for our simulation
purposes. Descriptive statistics related to the demand data are
where Fð:Þ is the cumulative distribution of the demand per given in Tables 2 and 3. The lead time (in days) distribution for all
period. 1109 SKUs is then presented in Fig. 2. At this point we should note
Naturally, an assumption of the probability distribution of the that some of the SKUs were ‘introduced’ at some point after
demand per period is needed in order to compute the inventory January 2005 and consequently the starting period used for
control policy parameters. For the purposes of our work, demand
was assumed to be Gamma distributed. The Gamma distribution Table 2
has been much used in the inventory control literature due to its Demand data descriptive statistics (across SKUs, to the 3rd decimal place).
high flexibility regarding the variance to mean ratio as well as
empirical evidence in its support for both smooth and inter- 1109 SKUs Demand sizes Demand intervals Demand per period
mittent demands.
Mean Variance Mean Variance Mean Variance
With regards to the latter demand patterns it is important to
mention the following. Since such series are constructed from Min. 1.000 0.000 1.000 0.000 0.007 0.007
constituent elements, inter-demand intervals or correspondingly 25% 1.636 0.500 4.516 10.403 0.071 0.159
demand arrivals and demand sizes, compound distributions may Median 3.152 4.917 9.500 64.436 0.246 1.094
75% 6.750 43.375 16.800 220.500 0.964 10.097
be theoretically justified for use in such a context of application. Max. 170.700 17,056.333 102.000 8580.500 78.785 9788.828
In particular, the Negative Binomial Distribution (NBD) constitu-
tes a very natural choice. When event arrivals are assumed to be
Poisson distributed and the order size is not fixed but follows a
logarithmic distribution, total demand is then negative binomially
Table 3
distributed over time. The gamma distribution is the continuous Lead time and unit cost information.

Minimum Maximum Average

1
The order of events in a period (week) is assumed to be as follows: a demand
occurs (may be zero), net inventory levels are determined and holding/backorder Lead time (days) 1 165 14.474
costs are incurred (for each unit on hand or backordered, respectively), an order Unit price (£) 0.01 1639.680 47.229
can be placed, and an order can arrive.
108 A.A. Syntetos et al. / Int. J. Production Economics 127 (2010) 103–111
0.12
0.1
0.08
Frequency
0.06
0.04
0.02
0
0 10 20 30 40
Fig. 2. Lead-time distribution.
simulation purposes as well as the summary statistics and the
performance evaluation related to these items refer to the
corresponding time intervals.
5.2. Experimental structure and simulation related details
We use the demand data set described above to empirically
assess the performance of the approach we proposed to the case
study organization, namely the classical re-order point (r, Q)
policy in conjunction with two possible estimators, against the
performance of the arbitrary stock control related practices
currently in use.
For the purposes of our empirical investigation, we split the
demand history available for each SKU into two parts. The 1st part
(i.e. within-sample) represents 40% of the total number of periods
and is used in order to initialise the estimates of level and
variance of demand. The 2nd part (out-of-sample) constitutes the
remaining 60% of the total number of periods and is used for the
out-of-sample generation of results and evaluation of perfor-
mance. Please note that the relative rather than absolute break-
down of the demand time series is due to the differing number of
periods available for the SKUs depending on their introduction in
the stock base of the organization. Moreover, and given that some
of the lead times are very long (maximum lead time ¼165
working days), the rule introduced ensures a reasonable within-
sample length for initialisation purposes. Optimisation of the
forecasting parameters (smoothing constants) has not been
considered at this stage of our investigation. As discussed
previously, the smoothing constant values used are the following:
a ¼ d ¼ l ¼0.05 and b ¼0.25 (reflecting the usual values of practical
usage in such contexts of application).
6. Implementation, empirical results and discussion
Since no precise information about various inventory related
charges was available, we resorted to discussions with the
company for the purpose of approximating such figures that
were essential for us in terms of the empirical investigation. This
exercise was also perceived as very useful by the individuals
participating in our meetings in terms of developing their
understanding on inventory related costs and the effect that they
may have on the bottom line of the organization. Regarding the
ordering costs, we decided to experiment with a figure that is
proportional to the unit cost. For the case study organization, the
ordering cost expressed in relative terms is between 50% and
400% of the SKU cost. As such, we experimented with the
following proportions: 50%, 75%, 100%, 200% and 400% of the unit
Lead-Time Distribution
50 60 70 80 90 100 110 120 130 140 150 160
Lead-Time (in days)
cost in order to analyze the effect on the results. At this point
it is important to note that the company operates with very
high margins (unit price–unit cost) to allow a profit to be
made given the high ordering related expenses. The distribution
of the ratio unit price/unit cost (%) across all SKUs is presented
in Fig. 3.
Regarding the inventory holding (h) and backordering (b)
charges, the ratio h/b ¼10% was thought to accurately represent
current business. The total inventory costs include the inventory
holding cost, the backlog cost and the ordering cost. Results are
generated by considering four target cycle service levels: 87%,
91%, 95% and 99%. For each of these targets we report (across all
SKUs): (i) the average achieved cycle service level, (ii) the average
inventory holding and backlog volumes, (iii) the average number
of orders placed and (iv) the total inventory cost. Results are
presented for both forecasting methods we experimented with:
SES (Table 4) and SBA (Table 5) for the case of the ordering cost
being 100% of the unit cost. (Please note that all other ordering
cost related structures lead to similar insights.)
The results indicate that the very high initial stocking levels
are carried forward affecting the outcome of our investigation.
The aggregate achieved service levels appear to be very high
(when compared against the target) as a result of this initial over-
stocking. Nevertheless, performance differences are still evident
and they did allow a comprehensive assessment of the compara-
tive merits of the new approach against that currently being used.
In particular, the new approach results in considerably lower
inventory costs whilst at the same time offering similar service
levels. The superior stock control performance of the new
approach is conclusive, being reflected on reduced inventories,
reduced backlogs as well as a lower number of orders being
placed.
The cost related benefits decrease with the target service level
and increase with the ordering fraction (percentage of the unit
cost). The aggregate service levels achieved when employing the
SBA are always less than the ones induced by SES and this is in
accordance with theoretical expectations. SES has long been
shown to be positively biased in an intermittent demand context
(Croston, 1972) with that bias obviously being translated in an
inflated service level. SBA in contrast is approximately unbiased,
with a little negative bias remaining (Syntetos and Boylan, 2005).
In the case study data both methods over-achieve the target
levels; the latter estimator though induces far lower inventory
related costs at an aggregate level and as such it is preferred for
practical implementation.
In summary, the company decided to adopt the proposed (r, Q)
policy, in conjunction with the SBA, operating under a CSL target
of 95%. (In this case, the cost related savings have been shown to
 A.A. Syntetos et al. / Int. J. Production Economics 127 (2010) 103–111 109

Distribution of the ratio: Unit price/Unit cost

12

10

Frequency (%)
8

0
25 5)

)
75 5)

00 0)

)
50 0)

)
00 0)

25 5)

)
75 5)

)
25 5)

50 0)

> )
0
[1 150

[2 225

[2 275

[3 350

[4 400

[4 475

00
50
[1 12

[1 17

[2 20

[2 25

[3 30

[3 32

[3 37

[4 42

[4 45

,5
0,

,
50

25

75

50

00

75
0
[1

Ratio: Unit price/Unit cost (%)

Fig. 3. Relationship between unit price and unit cost.

Table 4
Empirical results (ordering cost: 100% of unit cost)—SES.

1109 SKUs–VIP Holding (volume) Backlog (volume) No. of orders Total cost CSL

VIP approach 30.040 0.232 3.167 179.746 0.980

New approach
CSL¼ 87% 16.706 0.317 4.074 92.024 0.980
CSL¼ 91% 19.843 0.267 4.162 97.368 0.983
CSL¼ 95% 25.713 0.214 4.335 108.654 0.986
CSL¼ 99% 45.144 0.132 4.986 147.865 0.991

Table 5
Empirical results (ordering cost: 100% of unit cost)—SBA.

1109 SKUs–VIP Holding (volume) Backlog (volume) No. of orders Total cost CSL

VIP approach 30.040 0.232 3.167 179.746 0.980

New approach
CSL¼ 87% 15.354 0.391 3.843 84.227 0.975
CSL¼ 91% 18.076 0.333 3.867 85.224 0.979
CSL¼ 95% 23.524 0.257 3.941 88.977 0.984
CSL¼ 99% 44.559 0.138 4.698 128.648 0.991

be at least 40%.) Such a target reflected, according to manage-
ment, the best trade-off between costs and achieved service levels
taking also into account the high stocking levels inherent in the
current situation. At this initial stage of the intervention with the
system, the above discussed solution would apply to all SKUs in a
fully automated way with scope for managerial adjustments on
the top 11% of the best selling items and continuation of the
write-off strategy for those products that are obsolete. Periodic
review of the Pareto classification scheme would also ensure that
management retains an up-to-date summarized and comprehen-
sive picture of the entire stock base.
7. Insights and implications
We have recently undertaken a project (Syntetos et al.,
2009a,b) in collaboration with an international business machine
manufacturer for the purpose of improving their European spare
parts logistics operations. The actual intervention was mainly
associated with the re-configuration of the company’s SKU
classification system aiming at improved inventory management
practices. The main managerial implication of that study is very
similar to the one resulting from the project described in this
paper. Although a rather ‘basic’ (as compared to the theoretical
advancements in this area) inventory management solution was
proposed, the organisational benefits were indeed substantial.
Similar findings were reported by Cohen and Dunford (1986).
They considered a telecommunications company and examined
the procedures employed for forecasting the demand for
telephones. Such forecasts were required in order to control the
relevant inventories in the phone centres. Although this appeared
to be a straightforward exercise, data difficulties made it
impossible to use most of the potentially useful forecasting
models. The use of a very simple forecasting approach resulted in
an inventory reduction of 45%. Moreover, Alexander and Gift
(1985) analysed a naval stock control system in USA. Their
analysis that aimed at evaluating the stock control formulae of the
system under concern revealed many inherent operational
problems. Very simple analysis procedures led to the discovery
and evaluation of these problems. These procedures also aided in
explaining the problems and suggesting corrective measures to
management.
Results such as those discussed above are not particularly
surprising since practical inventory management related applica-
tions lag considerably behind the corresponding theoretical
propositions in this area. In that respect, our work can be seen
110 A.A. Syntetos et al. / Int. J. Production Economics 127 (2010) 103–111
as a contribution towards bridging the gap between OR theory
and practice.
At the moment there is an over-emphasis in the inventory
literature on theoretical developments. Although advancements
of the current state of knowledge are obviously absolutely
necessary, one should not lose sight of the fact that OR is all
about implementation and adding value to real-world practices.
Consequently, academia’s understanding of inventory manage-
ment can benefit greatly from insights obtained through case
studies. Case studies illuminate the actual problems managers
deal with and the difficulties/constraints involved in the solution
development process. Specificities related to the problem in hand
may obviously differ from one situation to the other and in that
respect generality (in a strict sense) may be questioned. However,
the conceptualisation, discussions and outcomes of such studies
(including ours) should provide practical insights to other
industrial organisations as well.
Based on the case analysis, the following strategic and
operational-level observations concerning factors that have an
impact on the performance of relevant inventory management
systems can be made:
 Inventory planning needs to be carefully linked to the
information infrastructure of the company under considera-
tion. Data retrieval and exploitation should constitute a major
issue of concern in real-world applications.
 Inventory modelling may provide many insights into the
relative value of several organisational improvement oppor-
tunities and successfully trigger strategic changes (see also
Cohen et al., 1990, 1999). For example, and as discussed in
Section 4.2, the write-off strategy adopted by the organisation,
in response to our initial evaluation of their system, was
perceived as one of the most important outcomes of our
project. Although the majority of items were discarded at a
reduced price offered to the customers, other possibilities were
considered or indeed utilised as well (e.g. returning the items
to the suppliers, scrapping, etc.). The very issue of selecting the
best possible write-off strategy and its linkage to organisa-
tional priorities and context has received very limited atten-
tion in the academic literature.
 Intermittent demand structures prevail in many wholesaling
contexts; given the difficulties that such patterns introduce to
the forecasting and stock control task further research in this
area is viewed as very important both from an academic and a
practitioner perspective.
 The estimation procedure selected for application purposes in
our work is one that has been developed for intermittent
demand items. However, the stock base of the company under
concern did include a few fast items as well. The company’s
requirement for the application of one single approach to
inventory management necessitates the consideration of the
trade-off between robustness (performance across an entire
stock base) and optimality. In real-world applications the former
is obviously far more important than the latter. That is to say, a
method/approach that may be optimal for one particular model
may be severely sub-optimal for another model. Syntetos et al.
(2006) argued that robustness of a method across a wide range
of possible underlying demand models is more important than
optimality under one particular model. The same is true with
regards to the performance of one single estimator across SKUs
with differing underlying demand structures (e.g. an entire
stock base). The requirement for one single approach to
forecasting and stock control is something that should be
expected from small and medium sizes organisations just
embarking into automated inventory management systems.
Given the frequency with which such a requirement is
encountered in practice and given the difficulties associated
with its effective operationalisation, further quantitative and
qualitative research into the issue of robustness of single systems
into a variety of underlying conditions would appear to be
merited.
8. Conclusions and way ahead
The area of inventory management in the wholesaling
environment has attracted considerable attention from the
academic community. However, case studies that consider a
solution development and implementation in a real application
are lacking. In this paper such a study has been performed on the
system of a UK-based wholesaling company dealing with
engineering supplies. The company serves mainly the construc-
tion industry, thus selling in large quantities without though
necessarily having to deal with specific retail outlets. The
company manages a wide range of products controlling their
inventories through ad hoc arbitrary procedures.
Following a major intervention with the company’s system for
the purpose of retrieving relevant data, a typical ABC classification
revealed the tremendous scope for improving the system through
an increased managerial attention to the best selling items and
introduction of a write-off policy for obsolete SKUs. Subsequently,
the performance of the classical (r, Q) policy in conjunction with
two possible forecasting methods was compared against current
practices. The findings were such that management was con-
vinced to move into a fully automated approach to stock control
(that comprises the order-point policy and the use of an
appropriate forecasting method, SBA) albeit with the possibility
of overriding the system if necessary. The company retained
previous policies with respect to some items that are associated
with economies-of-scale purchasing procedures and special
contractual agreements with specific suppliers.
The collaboration of the academic team with the company
under concern will continue as part of a knowledge transfer
programme subsidised by the UK government and a series of
further modifications have already been planned for the near
future. At the operational level, the possibility of eventually
introducing a more informative SKU classification scheme has
already been discussed. Such a scheme would allow a more
effective approach to inventory management by separating items
into various categories and treating them differently for forecast-
ing and stock control purposes. Moreover, performance measure-
ment, and the way this is captured through appropriate metrics,
constitutes one other important opportunity for advancing the
operational efficiency and effectiveness of the company. At the
strategic level, a complete back-to-back (customers to suppliers)
alignment of information has been proposed for linking demand
and supply in a more comprehensive manner.
Acknowledgements
The work described in this paper has been financially
supported by the company involved and the Department for
Business, Enterprise and Regulatory Reform, UK. The empirical
findings of the paper emerged from a Knowledge Transfer
Partnership (KTP 6202) between the company and the University
of Salford. More information on this project may be obtained at:
http://www.business.salford.ac.uk/research/ommss/projects/
Stock-Control/.
In addition, the work conducted by the first two authors (A.A.
Syntetos and M.Z. Babai) has also been supported by the Engineering
 A.A. Syntetos et al. / Int. J. Production Economics 127 (2010) 103–111
and Physical Sciences Research Council (EPSRC, UK) Grant no. EP/
D062942/1. More information on this project may be obtained at:
http://www.business.salford.ac.uk/research/ommss/projects/Fore
casting/.
References
Alexander, S.M., Gift, W.J., 1985. Discovering and correcting problems in a naval
stock control system. Interfaces 15, 41–47.
Altay, N., Rudisill, F., Litteral, L.A., 2008. Adapting Wright’s modification of Holt’s
method to forecasting intermittent demand. International Journal of Produc-
tion Economics 111, 389–408.
Boylan, J.E., 1997. The Centralisation of Inventory and the Modelling of Demand.
Unpublished PhD Thesis, University of Warwick, UK.
Boylan, J.E., Syntetos, A.A., 2007. The accuracy of a modified Croston procedure.
International Journal of Production Economics 107, 511–517.
Boylan, J.E., Syntetos, A.A., Karakostas, G.C., 2007. Classification for forecasting and
stock control: a case study. Journal of the Operational Research Society 59,
473–481.
Burgin, T.A., 1975. The gamma distribution and inventory control. Operational
Research Quarterly 26, 507–525.
Burgin, T.A., Wild, A.R., 1967. Stock control experience and usable theory.
Operational Research Quarterly 18, 35–52.
Chen, H., Boylan, J.E., 2007. Use of individual and group seasonal indices in sub-
aggregate demand forecasting. Journal of the Operational Research Society 58,
1660–1671.
Cohen, R., Dunford, F., 1986. Forecasting for inventory control: an example of when
’simple’ means ’better’. Interfaces 16, 95–99.
Cohen, M., Kamesam, P.V., Kleindorfer, P., Lee, H., Tekerian, A., 1990. Optimizer:
IBM’s multi-echelon inventory system for managing service logistics. Inter-
faces 20, 65–82.
Cohen, M., Zheng, Y.-S., Wang, Y., 1999. Identifying opportunities for improving
Teradyne’s service-parts logistics system. Interfaces 29, 1–18.
Croston, J., 1972. Forecasting and stock control for intermittent demand.
Operational Research Quarterly 23, 289–303.
Das, C., Tyagi, R., 1999. Manufacturer selection and price negotiation for
competitive wholesale distribution operations. International Journal of
Operations & Production Management 19, 977–992.
Eaves, A.H.C., Kingsman, B.G., 2004. Forecasting for the ordering and
stock-holding of spare parts. Journal of the Operational Research Society 55,
431–437.
Fildes, R., Nikolopoulos, K., Crone, S., Syntetos, A.A., 2008. Forecasting and
operational research: a review. Journal of the Operational Research Society
59, 1150–1172.
Germain, R., Groge, C., 1999. Wholesale operations and vendor evaluation. Journal
of Business Research 21, 119–129.
111
Gutierrez, R.S., Solis, A.O., Mukhopadhyay, S., 2008. Lumpy demand forecasting
using neural networks. International Journal of Production Economics 111,
409–420.
Johnston, F.R., 1980. An interactive stock control system with a strategic
management role. Journal of the Operational Research Society 31, 1069–1084.
Levén, E., Segerstedt, A., 2004. Inventory control with a modified Croston
procedure and Erlang distribution. International Journal of Production
Economics 90, 361–367.
Pearson, M., 2006. The application of prediction capability and adaptive target
control to the newspaper business. European Journal of Operational Research
168, 475–491.
Porras, E., Dekker, R., 2008. An inventory control system for spare parts at a
refinery: an empirical comparison of different re-order point methods.
European Journal of Operational Research 184, 101–132.
Ritchie, E., Kingsman, B.G., 1985. Setting stock levels for wholesaling: performance
measures and conflict of objectives between supplier and stockist. European
Journal of Operational Research 20, 17–24.
Sani, B., Kingsman, B., 1997. Selecting the best periodic inventory control and
demand forecasting methods for low demand items. Journal of the Operational
Research Society 48, 700–713.
Schneider, H., Rinks, D.B., 1991. Empirical study of a new procedure for allocating
safety stock in a wholesale inventory system. International Journal of
Production Economics 24, 181–189.
Silver, E.A., Pyke, D.F., Peterson, R., 1998. Inventory Management and Production
Planning and Scheduling 3rd ed. John Wiley & Sons, New York.
Syntetos, A.A., Boylan, J.E., 2001. On the bias of intermittent demand estimates.
International Journal of Production Economics 71, 457–466.
Syntetos, A.A., Boylan, J.E., 2005. The accuracy of intermittent demand estimates.
International Journal of Forecasting 21, 303–314.
Syntetos, A.A., Boylan, J.E., Croston, J.D., 2005. On the categorisation of demand
patterns. Journal of the Operational Research Society 56, 495–503.
Syntetos, A.A., Boylan, J.E., Croston, J.D., 2006. A note on the categorisation of
demand patterns. Journal of the Operational Research Society 57, 1257–1258.
Syntetos, A.A., Keyes, M., Babai, M.Z., 2009a. Demand categorisation in a European
spare parts logistics network. International Journal of Operations and
Production Management 29, 292–316.
Syntetos, A.A., Nikolopoulos, K., Boylan, J.E., Fildes, R., Goodwin, P., 2009b.
Integrating management judgement into intermittent demand forecasts.
International Journal of Production Economics 118, 72–81.
Teunter, R., Sani, B., 2009. On the bias of Croston’s method. European Journal of
Operational Research 194, 177–183.
Viswanathan, S., Widiarta, H., Piplani, R., 2008. Forecasting aggregate time series
with intermittent sub-aggregate components: top–down versus bottom–up
forecasting. IMA Journal of Management Mathematics 19, 275–287.
Willemain, T.R., Smart, C.N., Shockor, J.H., DeSautels, P.A., 1994. Forecasting
intermittent demand in manufacturing: a comparative evaluation of Croston’s
method. International Journal of Forecasting 10, 529–538.