Intermittent demand categorization and forecasting
Submitting Institution
Buckinghamshire New UniversityUnit of Assessment
Business and Management StudiesSummary Impact Type
EconomicResearch Subject Area(s)
Mathematical Sciences: Statistics
Economics: Applied Economics, Econometrics
Summary of the impact
Our research team has developed new approaches to classifying demand
series as `intermittent'
and `lumpy', and devised new variants of the standard Croston's method for
intermittent demand
forecasting, which improve forecast accuracy and stock performance. These
approaches have
impacted the forecasting software of Syncron and Manugistics, through the
team's consultancy
advice and knowledge transfer. Subsequently, this impact has extended to
Syncron International
and JDA Software, which took over Manugistics. These companies'
forecasting software packages
have a combined client base turnover of over £200 billion per annum, and
their clients benefit from
substantial inventory savings from the new approaches adopted.
Underpinning research
This research originated in joint work by Roy Johnston (Warwick
University) and John Boylan
(Bucks New University). The research was undertaken in 1993 and 1994,
during which time John
Boylan was employed by Bucks as a full-time Lecturer. This early research,
based on simulation
modelling, identified conditions under which Croston's method is more
accurate than Single
Exponential Smoothing (1). A break-point for the average interval between
demands was found,
above which Croston's method is more accurate. An investigation was also
conducted on the effect
of other variables, such as variances and distributions. However, no
break-points were identified
which took these factors into account. Whilst this research did not form
the basis of
recommendations for Syncron or Manugistics, it did lay the groundwork for
later research which
was adopted by the companies.
The research was extended at Bucks New University by Aris Syntetos,
supervised and mentored
by John Boylan, between 1999 and 2001. During this time, Bucks employed
Aris Syntetos as a full-
time Lecturer and John Boylan as Head of Research. The first finding of
this research was that
Croston's method is biased (2); this was proven by mathematical analysis.
The second finding,
based on a Taylor series argument, was that a simple adaptation to
Croston's method would result
in approximately unbiased forecasts (3). Later, this adaptation has become
known as the Syntetos-
Boylan Approximation. The third finding, based on a mathematical analysis
of Mean Squared
Errors, was to extend the original categorization scheme (1) by including
a measure of the
`lumpiness' of the demand data (4).
As part of this research, the findings were tested on real data. This
confirmed the bias of Croston's
method and the reduction in bias by using the Syntetos-Boylan
Approximation. It also validated the
categorization rules, which had been derived on theoretical grounds.
Finally, it showed that these
forecasting benefits translated into inventory reductions in a re-order
interval context (5). However,
financial evaluations of this benefit were not possible because of the
absence of unit cost data.
The next phase of research was to evaluate the financial effect of the
new approaches on real
company data at Syncron (UK) Ltd. This research was part of a Teaching
Company Scheme (TCS)
by Syncron and Bucks New University. The research was undertaken by George
Karakostas,
supervised by John Boylan. George Karakostas was employed by the
University as a TCS
Associate, on a full-time basis, between 2001 and 2003, while John Boylan
continued to be
employed as Head of Research.
The research empirically demonstrated the robustness of the
categorization rules on real data (6).
It also showed that marked reductions in inventories could be achieved
from the application of the
Syntetos-Boylan Approximation. This finding was consistent with
independent empirical research
conducted at Lancaster University (Eaves and Kingsman (2004) Forecasting
for the ordering and
stock-holding of spare parts. Journal of the Operational Research Society,
55, 431-437). The
Bucks' research team found that savings of over 11.7% of inventory value
were attainable, with a
slight undershoot of the Customer Service Level.
References to the research
The research was disseminated in the following peer-reviewed journal
articles:
1. Johnston FR, Boylan JE (1996) Forecasting for items with intermittent
demand. Journal of the
Operational Research Society, 47, 113-121.
2. Syntetos AA, Boylan JE (2001) On the bias of intermittent demand
estimates. International
Journal of Production Economics, 71, 457-466.
3. Syntetos AA, Boylan JE (2005) The accuracy of intermittent demand
estimates. International
Journal of Forecasting, 21, 303-314.
4. Syntetos AA, Boylan JE, Croston JD (2005) On the categorization of
demand patterns.
Journal of the Operational Research Society, 56, 495-503.
5. Syntetos AA, Boylan JE (2006) On the stock-control performance of
intermittent demand
estimates. International Journal of Production Economics, 103, 36-47.
6. Boylan JE, Syntetos AA, Karakostas GC (2008) Classification for
forecasting and stock-
control: a case-study. Journal of the Operational Research Society, 59,
473-481.
All of the above references are in the public domain.
Evidence of the quality of the underpinning research is summarized below:
• Articles (1), (2), (3) and (4) have all been cited more than 100 times,
according to Google
Scholar. They were all referenced in the widely-cited review article by
Gardner (2006):
Exponential smoothing: the state of the art — Part II, International
Journal of Forecasting, 22,
637-666.
• Articles (1) and (3) were also cited in the major review article by De
Gooijer and Hyndman
(2006): 25 years of time-series forecasting research, International
Journal of Forecasting, 22,
443-473.
• Article (5) has been cited 70 times (Google Scholar, 1 October 2013),
including citations by
researchers at the universities of Brescia, Lancaster, Lulea, Richmond,
Saint-Etienne,
Stanford, Takming, Texas, Thessaloniki, Tilburg, and Valencia.
• Article (6) has been cited over 50 times (Google Scholar, 1 October
2013). The article is based
on research undertaken as part of a Teaching Company Scheme (TCS), now
known as a
Knowledge Transfer Partnership. The project received a good grade from the
TCS Central
Office. (Letter from EH Robson, TCS Director, 9 January 2004).
Details of the impact
Contribution and Impact on Syncron
In autumn 2001, Bucks New University and Syncron UK commenced a two-year
Teaching
Company Scheme. The aim was to enhance the capability of Syncron's demand
forecasting and
inventory management software.
The research team provided a review and critique of Syncron's Demand
Forecasting and stock
Replenishment Planning system (2). This identified the following issues,
amongst others: limited
capacity to store demand histories, inflexibility of hard-coded tables to
categorize demand, and
limited capability to forecast intermittent demand items. The team
proposed and evaluated new
methods for intermittent demand forecasting, based on the Syntetos-Boylan
Approximation and
variations of that approach (2, 3). Error evaluations showed marked
reductions in Mean Squared
Errors, of approximately 25%, when employing the new methods instead of
the existing method
(3). The research also showed that the old company categorization approach
was inappropriate,
and that greater forecast accuracy could be achieved with a new
categorization method proposed
by the University.
Syncron adopted the University's recommended approaches and these have
remained in effect
from the beginning of 2008 to the end of 2013 (4). On Syncron's web-page
for its Demand
Forecasting Software, the company highlights "Our best-in-class dynamic
demand forecasting for
stock with frequent, highly sporadic or intermittent demand". It then
headlines "Demand
Forecasting Software: Automatic Demand Classification". The enhancement to
Croston's method
is also cited: "Special forecasting techniques such as a refined Croston
methodology are used to
handle sporadic and intermittent demand" (4).
Contribution and Impact on Manugistics and JDA
In 2002, John Boylan was commissioned by Manugistics to provide
consultancy advice on the
development of a systematic demand classification and forecasting system.
This had arisen from
debate within the company about the relative merits of a formal
classification system and a "pick-best" methodology, based on minimizing model-fit error. Boylan (5)
recommended a formal
classification system, arguing that it could offer a more accurate and
sophisticated approach if
appropriate categorization rules were used.
The recommendation of a formal demand classification approach was
communicated by report (5)
and by a video-conference between Boylan and Manugistics' board members in
the US. This
recommendation was accepted by the company and implemented in their
software. After
Manugistics had been taken over by JDA, the approach was also implemented
as part of JDA's
Global Inventory Management software, and has remained the basis for the
JDA Demand
Classification package from 2008 to the end of 2013 (6). Boylan (5) also
recommended, based on
his categorization research with Aris Syntetos, that the company should
distinguish between
`erratic' and `lumpy' demand series. These categories remain at the heart
of the JDA Demand
Classification software (6), and the JDA Demand Management package.
Financial Benefit to Syncron's and JDA's Clients
Syncron's clients for its Global Inventory Management software include
the following companies
(7), listed with their annual revenues: Deutsche Bahn (39.3 billion Euro,
2012), JCB (£2.75 billion,
2011), Mazda (2.3 trillion Yen, 2011), Metso Minerals (3.5 billion Euro,
2012, Mining and
Construction), Renault Trucks (4.3 billion Euro, 2011) and Volvo
Construction Equipment (65 billion
SEK, 2011).
JDA Demand Management software clients include (8): Avon Products ($10.72
billion, 2012),
Bristol-Myers Squibb ($18.8 billion, 2009), Canadian Tire ($8.98 billion,
2010), Dell ($62.1 billion,
2012), Harley Davidson ($5.31 billion, 2011), Hyundai ($84 billion, 2012),
Kraft Foods ($18.34
billion, 2012), O2 (UK) (£2.97 billion, 2010), Renault (41.3 billion Euro,
2012), Swire Pacific, (HK$
36.29 billion, 2011), Toshiba Semiconductor ($12.36 billion, 2010), and
Vodafone (£46.4 billion,
2012).
The combined turnovers of these clients amount to approximately £240
billion, at current exchange
rates. Application of aggregate `Days of Inventory Outstanding', by
Industrial Sector (9), gives a
very approximate estimate of total inventories of £19 billion. JDA Demand
Management is also
used by the Defense Logistics Agency, which reported inventories of £13
billion in 2012 (10),
giving approximate total inventories of £32 billion. There is no reliable
data available on the
proportion of the companies' inventories which are slow-moving. However, a
modest assumption
would be at least one-third, giving an estimate of slow-moving stock of
£10 billion.
The most comprehensive analysis of the financial impact of enhanced
categorization methods and
of the application of the Syntetos-Boylan Approximation (SBA) was
conducted by Eaves and
Kingsman (2004): Forecasting for the Ordering and Stock-Holding of Spare
Parts, Journal of the
Operational Research Society, 55, 431-437. They concluded that enhanced
categorization (using a
similar but not identical approach to that recommended by Bucks) and
application of the SBA
method resulted in a saving of 13.6% of the total value of inventory, with
slightly smaller savings if
Croston's method were used instead of SBA. This is consistent with the
financial saving of 11.7%
identified in the TCS project with Syncron. If the clients of Syncron and
JDA have been able to
achieve similar savings, of at least 10% in inventory values, then this
equates to a saving of £1
billion of stock. Whilst it is accepted that this number is highly
approximate, it gives an indication of
the financial impact of the research described in this case-study. It is
difficult to estimate the
environmental benefit of the associated reduction of inventory
obsolescence, although this should be
considerable. Excessive slow-moving stock is most prone to obsolescence
and to the concomitant
waste of resources in making goods that are never used.
Sources to corroborate the impact
- Review and Critique of Syncron Demand Forecasting and Stock
Replenishment Planning
Software. GC Karakostas, 2002.
- Enhancement of Single-Level Forecasting Software Module. GC
Karakostas, 2002.
- Boylan JE, Syntetos AA, Karakostas GC (2008) Classification for
forecasting and stock-
control: a case-study. Journal of the Operational Research Society, 59,
473-481.
- Letter from Managing Director, Syncron UK Ltd, 19 November 2013.
- Development of a theoretically consistent, scientifically based and
systematic demand
forecasting system (Final Report for Manugistics), John Boylan, 2002.
- Email from Vice President, Product Management, JDA Software, 14
October 2013.
- Syncron Supply Chain Inventory Management. Inventory Management
Software.
www.syncron.com/en/Solutions/global-inventory-management
- JDA Demand Management Brochure www.jda.com/company/display-collateral/pID/488/
-
Supply
Chain News: Inventory Performance 2011 (Supply Chain Digest 22/7/11) www.scdigest.com/assets/FirstThoughts/11-07-22.php?cid=4759
- Defense Logistics Agency, Annual Financial Report, Fiscal Year 2012
(Unaudited).