See discussions, stats, and author profiles for this publication at: [Link]

net/publication/235321988

Sales forecasting for strategic resource planning

Article in International Journal of Operations & Production Management · September 2002

DOI: 10.1108/01443570210440519

CITATIONS READS

58 3,110

2 authors:

John G. Wacker Rhonda Lummus

Arizona State University Indiana University Bloomington
43 PUBLICATIONS 2,784 CITATIONS 25 PUBLICATIONS 3,060 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by John G. Wacker on 01 December 2014.

The user has requested enhancement of the downloaded file.

 T h e cu rre n t iss u e a n d fu ll tex t a rch iv e o f th is jo u rn a l is a v a ila b le a t
[Link] [Link]/[Link]

IJOPM
22,9
Sales forecasting for strategic
resource planning
John G. Wacker
1014 Iowa State University, Arizona State University,
Tempe, Arizona, USA and
Rhonda R. Lummus
Iowa State University, College of Business, Ames, Iowa, USA
Keywords Strategic evaluation, Strategic planning, Forecasting, Resource management
Abstract The purpose of this article is twofold. First, the article examines how managers can
make more effective use of sales forecasts for strategic resource allocation decisions. Second, the
article identifies those research issues in forecasting that must be addressed to better understand
the managerial side of forecasting. Managers can improve resource planning by understanding the
limitations of forecasts. These limitations are exemplified through several strategic forecasting
paradoxes that managers must recognize. The paradoxes suggested here are: first, the most
important managerial decisions a company can make are based on the least accurate forecasts;
second, the most useful forecast information for resource planning is the least accurate; and, third,
the organizations that need the most accurate forecast have the largest forecast error. By
recognizing these paradoxes managers can devote their attention to improving the use and
implementation of the forecast for better resource decisions. At the same time, future research
should focus on broadening the understanding of the role of forecasts in strategic decision making.

Introduction
This article focuses on the link between forecasts and specific resource
allocation decisions. The importance of the tie between forecasts and resource
decisions cannot be overstated for manufacturing organizations. Without the
link between forecasts and resource allocation, it would be impossible to
acquire resources to make on-time deliveries to customers. Also, without a sales
forecast there could be no long-term production planning of future sales to
allow for sufficient capacity and labor needs. In short, having product available
to meet sales would not be possible without a sales forecast. Strategically, the
success of manufacturing organizations is tied to the effectiveness of the link
between the forecast and the resource allocation plan.
The purpose of this article is to give strategic managers practical
suggestions for better understanding the limitations of forecast methodology.
Once managers understand the limitations of the forecast methods, they can
then identify those strategic resource allocation procedures that are most
helpful to reduce the impact of forecast error. By understanding these methods,
both managers and forecast experts can reduce the impact of forecast errors on
resource allocation decisions.
A secondary purpose of the paper is to identify needed research on the
International Journal of Operations &
Production Management,
strategic impacts of forecasting on managerial decision making. Little research
Vol. 22 No. 9, 2002, pp. 1014-1031.
# MCB UP Limited, 0144-3577
in forecasting has been done to aid in understanding the managerial side of
DOI 10.1108/01443570210440519 forecasting. Much of the focus has been on forecast methods, sources of
forecast error, and reducing forecast error. There is much potential for the Sales forecasting
improved use of forecasting, but little evidence of where to begin to better align
forecasters and strategic managers. This paper will begin by examining
previous forecasting research. A brief review of basic forecast theory is
included along with literature on the impact of forecast error on strategic
decisions. A discussion of the sources of forecast error and the costs associated
with forecast error follows. Forecasts alone have little or no value, but what is 1015
important is how they are used in making managerial decisions. This study
describes three basic forecast paradoxes that impact strategic resource decision
making. Suggestions are made for reducing the impact of forecast error on
strategic decisions. The paper concludes with suggestions for future research
to aid in understanding the managerial side of forecasting.

Forecast literature
Understanding forecast theory
A prerequisite for the effective use of forecasting is an understanding of
forecast theory. The foundation of forecast theory is based on whether or not a
forecast is reliable to managers. This point is put succinctly by Fischhoff
(1994), who states that there are different levels of ``pedigree’’ or reliability of
numerical forecasts. The four levels of pedigree for numerical forecasts are:
(1) the highest level – established theory that offers theoretical explanation
from causal analyses;
(2) the second level – theoretical modeling that has causal relationships
hypothesized;
(3) the third level – the computational model that approximates causality;
and
(4) the lowest level – the statistical processing model that has no causality
offered to definitions/assertions (pure guesses).
The believability of the forecast is related to the strength of the theory
underlying the forecast. The forecast must include an explanation of how and
why it was constructed as well as an estimate of the accuracy of its predictions,
if it is expected to be more accurate. As a result, the significance of all business
forecasting depends upon forecast theory. For this research, the definition of
theory is that it answers the traditional questions of inquiry. Theory must
include the following four parts:
(1) definitions (answers ``who’’ and ``what’’ questions);
(2) domains (answers ``when’’ and ``where’’ the theory applies);
(3) explanations (answers ``how’’ and ``why’’ questions); and
(4) predictions (answers what ``should,’’ ``could,’’ and ``would’’ happen)
(Bunge, 1967; Wacker, 1998).
IJOPM From a forecasting theory perspective, times series models generally do not
22,9 answer how and why sales increase or decrease. These models use the patterns
in past data that are assumed to have unstated but stable causal relationships.
For example, December’s sales in retail stores are usually higher than other
monthly average sales, but the explanation of why they are higher requires
understanding the psychological needs of the consumer during the holiday
1016 seasons. These psychological needs are not mathematically introduced in the
time series models. Consequently, the model is not based on forecasting theory,
but rather on time series patterns of sales behavior. To state that a time series
forecast is built on forecasting theory elevates the forecast importance above
what is warranted by the definition of theory.
Causal models (econometric, regression, input-output analyses, etc.) have
been traditionally considered to be formal theory-based models (Chalmbers
et al., 1971). The causal models fulfill the requirements of theory since the
definitions, domains, relationships, and predictions are stated as to how and
why they apply to the forecast. These models have another difficulty in that
they assume there is causality between the explanatory variables and sales (i.e.
sales of windows are related to the number of housing starts – the explanatory
variable). In addition, causal models are not easy to implement as they often
rely on leading indicators that rarely have good explanatory power. To
overcome this drawback, many causal models rely on concurrent indicators
that usually have higher explanatory power; but the explanatory variables also
need forecasts which results in two statistical estimates: one for the sales
prediction model and one for the prediction of the explanatory variables. (This
concept sometimes is called the concept of ``second order guilt’’ since the overall
model predicts well, if the explanatory variables are accurately predicted.)
Consequently, the inaccurate forecast is not caused by poor statistical fit with
the model but with the prediction of the explanatory variables used in the
model.
In summary, the two most often used quantitative forecasting methods
therefore have strengths and weaknesses and neither provide accurate
forecasts. As a well-known forecasting academic observes, ``The most
important lesson we have learned in the field of forecasting over the last two
decades is that models which best fit available data (such models will be, in
everyday language, the equivalent of explaining as well as possible after the
fact, what has already happened in the past) are not necessarily the most
accurate ones in predicting beyond this data’’ (Makridakis, 1996). One of the
important lessons in forecasting is that the forecast is only as good as the
theory that it is based upon. The theory of forecasting is also the foundation for
understanding the sources of forecast errors.

Forecast error
The impact of forecast error on strategic decisions has been considered an
important issue since the earliest writings on production planning and control
(Holt et al., 1955). More recently, studies have found that forecast inaccuracy
has cost impacts on organizations (Biggs and Campion, 1982; Ritzman and Sales forecasting
King, 1993). The impact is severe because forecasts are a major driver for all
strategic resource allocation plans. When forecasts are inadequate, planning
becomes haphazard, there are frequent costly revisions to the plan and
inventory increases (Vollman et al., 1992). Forecast inaccuracy presents
challenges for resource managers, as they are responsible for alleviating the
adverse affects of forecast error. 1017
However, not all the empirical research supports the conclusion that high
forecast errors cause higher costs and loss of competitiveness. Some empirical
results suggest that forecast error is not a significant factor impacting
manufacturing goals (Wacker and Hanson, 1997). Other studies suggest that
forecast errors do not necessarily increase costs to the organization (Lee and
Adam, 1986; Biggs and Campion, 1982). This suggests that well-run
organizations learn to compete despite rapidly changing forecasts and plans. In
fact, competitive organizations do not depend on accurate forecasts for
competitive advantage, but learn to cope with forecast error and still be
competitive.
Interest in the costs associated with forecast inaccuracy has motivated
research aimed at reducing forecast error. This research has focused in two
basic directions. One direction has been the effect of various quantitative
forecasting methods on numerical forecast error. There is a rich literature on
the effect forecasting methods have on forecast error (see Armstrong, 1984).
The second direction is on how forecasts are used for managerial planning. In
this direction, the literature is much more sparse. The literature does not offer
an in-depth understanding of the role of forecasts in resource allocation
planning or how managers can reduce the impact of forecast errors on resource
plans. The purpose of this study is to fill that void by offering managerial
insights into how the forecast is used in making resource allocations decisions.
These insights can aid managers in reducing the financial impact of forecast
errors on their strategic resource plans. Before beginning a discussion of
forecast use in resource planning decisions, it is first necessary to describe the
traditional limitations of forecasting.
Traditional forecasting shortcomings include:
ambiguity – not being able to clearly state the events and their
likelihood of occurrence;
irrelevance – not offering forecasts that address the clients’ needs;
immodesty – not recognizing the limits of the forecast; and
impoverishment – not placing the forecast in broader decision-making
context (Fischhoff, 1994).
In other words, the forecast must be explicit, have a stated purpose, include the
forecast assumptions, and state its planned use for decision making. These
shortcomings highlight some of the practical managerial concerns about
forecasting. This study addresses the irrelevance and impoverishment
IJOPM shortcomings through focusing on how the forecast is used for resource
22,9 decisions. Overcoming these shortcomings can reduce the cost of forecast
errors on resource allocation decisions. Because of the importance of the
forecast error on resource allocation decisions, managers constantly search to
understand ``how’’ forecasts are created; and ``why’’ some forecasts are ``accurate
enough’’ to make resource decisions. The underlying rationale used by
1018 managers to relate forecasts to resource allocations will be examined in this
study.

Sources of forecast errors

Recent studies suggest that systematically measuring forecast error improves
forecast accuracy (Wacker and Sprague, 1995; 1998). In addition, forecasters
must understand the sources of forecast error to understand why the error
occurred. Forecast error results from one or more specific reasons (how and why
it occurred), which indicates all forecast error analyses are based on theory.
Forecast errors are traditionally caused either by internal sources, external
sources, or the chosen model (Chalmbers et al., 1971). The internal sources
include the four Ps of marketing: price (the price of the item), product (product
quality), place (distribution system), and promotion (advertising, sales
incentives, and other promotions). The external sources include the general
economic conditions (unemployment, income, etc.) and competitors (the four Ps
of your competitors). The chosen model source is self-explanatory.
After a significant forecast error occurs, organizations investigate the
forecast error source by first evaluating the internal sources as they are easier
to find, and then the external sources. Investigating why the sales forecast from
marketing was not accurate is generally much easier than analyzing
competitors or the national economy. If the internal analysis does not prove
fruitful, external sources are investigated to determine if the fundamental
underlying demand pattern has changed. The model is the most difficult place
to find errors since the model may not perform well in one period, but may
perform well overall. Plossl (1973) suggested that the forecast model not be
adjusted each period to take into consideration forecast error, but rather if the
error continues, determine the error source and establish procedures to
ameliorate forecast errors from that source. His concern was that firms
continually compensate for forecast error and exacerbate the size of the error.
Organizations can improve judgement in forecasting by seeking out the
sources of forecast error and modifying the forecast as deemed necessary
rather than spending significant effort and cost trying to find the ``right’’
forecast model. The investigation of the cause of forecast error facilitates
learning, which leads to improving the managerial judgment that is always
present in the forecast.

Relating forecast accuracy to cost reduction

The key reason managers insist on an accurate forecast is that they believe it
reduces costs. Research shows that in many companies, 10 percent or more of
net gross profit is lost because forecast inaccuracy causes overages and Sales forecasting
shortages in inventory (see for example Ritzman and King, 1993). Additionally,
the inaccurate forecast causes production replanning that creates purchasing,
financing, and scheduling difficulties. Managers make resource decisions
based on the forecast, only to find that in the increasingly competitive world,
product configurations change rapidly. These changes cause the product
specific forecast to be inaccurate and require changing resource allocations 1019
between products. Specific time horizons exist for resource commitments and
when changes are made within those horizons, costly misallocation of
resources can result. This misallocation occurs when the ``drop dead’’ date for a
specific resource decision passes without a change in the resource allocation. A
change after the date increases costs, results in missed sales opportunities, and
decreases managerial performance. Managers tend to believe the resource
reallocation decision was missed due to forecast inaccuracy. Consequently,
they believe their lost opportunities are the result of ``poor’’ forecasting. These
experiences may explain why less than 50 percent of managers are satisfied
with quantitative forecasting methods (Sanders and Mancroft, 1994).
These experiences lead managers to insist on more accurate forecasting
because they recognize the cost implications of changing forecasts. This
insistence provides interesting insights into strategic decision making. Managers
insist on more accurate forecasts because they use them to make strategic
resource plans for labor, materials and equipment. Forecast errors force the
organization to replan all the resources. The more difficult it is to replan, the
more important an accurate forecast is. In other words, the more inflexible the
resource plan, the more important forecast accuracy is. Many times managers
rely on forecast accuracy to overcome the managerial inflexibility of their
planning procedures. An accurate forecast is important, but if managers
understand the limitations of forecast accuracy they can plan their resources
effectively through tying resource decisions to specific forecasts.

Relating forecasts to managerial decisions

Time frames and resource decisions
Planning and forecasting are closely inter-related and ``Strategic planners
should probably spend more time underselling the importance of forecasting the
future and less time building up false expectations concerning forecasts with
techniques which, at best, have a mixed record’’ (Naylor, 1983). The importance
of forecast accuracy for strategic planning is usually oversold since firms often
cannot differentiate between the level of detail needed for different resource
decisions. The level of detail includes determining the length of one forecast
time period (i.e. weeks, months, quarters, etc.). Forecasts are usually broken
down by time periods so they can be used for budgeting purposes or because
sales fluctuate radically between periods (Makridakis, 1996). Budgets are based
on marketing plans that become the basis for specific resource plans. A more
effective approach to strategic resource planning is to examine how resource
needs change if different marketing plans are implemented.
IJOPM Many times organizations create forecasts independent of resource
22,9 allocation decisions. This occurs when the final forecast is developed in one
functional area and then distributed throughout the organization. Top
management approves this forecast, but it is not integrated into resource plans
until later. Those who create the forecast should consider how the forecast is to
be used by other functions to reduce the adverse effects of forecast inaccuracy.
1020 One method to reduce the forecast’s adverse effects is to associate specific time
frames with specific resource decisions. Each organization has a different
chronological time period for making resource decisions and the forecast and
planning horizons should relate to these decision time periods to match
decision making with forecasting.

Resource decision time periods

The immediate, short, medium, long, and very long time frame forecasts should
be tied to specific resource decisions. The immediate time frame forecast should
be the basis for material changes (shortages and overages) and schedule
changes (overtime or under-time). These decisions are reactions to
specific changes in the demand forecast. The short decision time frame is used
to make changes in the number of workers. These employment decisions
(downsizing or upsizing the organization) require longer time frames than
material and scheduling changes. The medium decision time frame is used for
making equipment changes (purchase or disposal). The decision to bring in
new equipment is based on meeting competitive objectives. The long decision
time frame is used for making changes in the plant/facility. These decisions
focus on the firm’s long-term competitive advantage and the development of
new markets. The very long decision time frame is used for substantial
changes in technology. Except in high velocity environments (see Mansfield,
1988), the very long decision time frame is not formally forecast in most
organizations since the benefits of technological forecasting may not offset the
difficulty and the cost of the forecast. For this reason, this study does not
include a discussion of the very long time frame.
Each of the time frames is important for a certain decision, and each decision
is linked to specific resources. The decision time periods specify how far into
the future the organization must forecast to cover all resource allocation
implications. The forecast should predict far enough into the future to provide
information on the new plant/facility decision. Without a forecast, the new
plant/facility decision is made by default and not by deliberate decision. The
chronological time for each decision time frame is different for each company.
For example, the time frame required for forecasting the need for a new
McDonalds or other fast food outlet is generally about six to eight months from
the initial study to opening the facility. At the other extreme of the
chronological time spectrum, is an electric company’s decision to build a
nuclear power plant where the long-term decision time frame may be as long as
10 to 15 years. All other time frames follow in sequence from the long-term time
frame, although they may overlap. Equipment comes in before labor is hired,
labor is hired before the schedule is set, and materials are ordered after the Sales forecasting
labor force is in place. In both the fast food outlet and the nuclear power plant,
the medium, short, and immediate time frames are all sequentially shorter time
periods from long to immediate.
In all industries, the forecasting period is set by resource decisions that have
different chronological time periods depending on the organization. Managers
must focus on the resource allocation decision that dictates the chronological 1021
time period that must be forecast. All forecasts should be tied to the decisions for
which they provide information. As a general rule, forecasting beyond the long-
term time frame is usually not worthwhile. Whatever the length of the forecast
horizon, there are other challenges for firms using forecasts for resource
decisions. One of these challenges is the effect of forecast error on strategic plans.

Forecast paradox
Forecast error and time frames
Table I shows the results of two of the most often cited studies on forecast
accuracy. A product group forecast is a wide grouping across many different
products. A product line forecast is more specific, with products classified
together based on the marketing strategy or production process. The product
forecast is for the individual stock-keeping unit. Reading across the row for
product group, the forecast error increases as the forecast goes farther into
future. The same holds true for product lines and individual products. The least
accurate of all forecasts is the long-term forecast (> two years). This forecast is
used for new facility strategic decisions that have long been considered the
most important financial decisions an organization can make (see for example
Hayes et al., 1984).
The result of long-term forecast error is an inherent paradox for all
organizations:
Paradox 1. The most important strategic decisions a company makes are
based on the least accurate information.

Wide product groups and forecast error

Referring again to Table I and reading down the columns from wide product
groups to individual product forecasts, it becomes apparent that the more

Aggregation level < Three months < Two years > Two years

Product group 10 (10) 15 (10) 20 (15)

Product line 11 (12) 16 (12) 20 (19)
Product 16 (16) 21 (12) 26 (27)
Notes: All numbers are mean absolute percent errors (MAPE). Numbers in parentheses are
for British managers Table I.
Forecast errors and
Sources: Mentzer and Cox (1984); Fildes and Beard (1992) levels of aggregation
IJOPM narrow the product grouping, the less accurate the forecast. The product
22,9 specific forecast is used for planning the specific resources (materials and
labor) required to build a particular product. Organizations make specific
resource decisions based on the product’s specifications. However, the more
product-specific the forecast is, the less accurate it is. Consequently, there is a
second inherent paradox:
1022
Paradox 2. Forecast information that is most useful for resource planning is
the least accurate.

Higher cost, longer lead time products and forecast error

Long-term product-specific forecasts are the least accurate, as shown above,
and in addition, forecasts for certain products may be less accurate than those
for others. The differences may relate to the product durability differences
between industries. Table II gives the coefficient of variation of demand of the
Standard Industrial Classifications (SIC) for durable and non-durable goods
for a ten-year period. At one end of the durability spectrum is the housing and
capital equipment markets, at the other end are items such as food and
clothing. Table II illustrates that generally durable products have a larger
coefficient of variation than non-durable products. Assuming that the more
variability there is in demand the more difficult it is to forecast, then durable

Coefficient of
SIC Code variation (%)

SIC 512 Drugs, drug proprietors and druggists sundries 32.08

SIC 504 Professional and commercial equipment and supplies 30.79
SIC 506 Electrical goods 27.26
SIC 511 Paper and paper products 21.99
SIC 509 Miscellaneous durable goods 20.04
SIC 516 Chemical and allied products 19.83
SIC 503 Lumber and other construction material 19.27
SIC 507 Hardware, plumbing and heating equipment, and supplies 19.27
SIC 519 Miscellaneous non-durable goods 16.68
SIC 513 Apparel, piece goods and notions 16.50
SIC 508 Machinery, equipment, and supplies 14.94
SIC 505 Metals and minerals, except petroleum 14.36
SIC 502 Furniture and home furnishings 14.01
SIC 515 Farm-product raw materials 13.65
SIC 518 Beer, wine, and distilled alcoholic beverages 13.50
SIC 501 Motor vehicles and automotive parts and supplies 11.71
SIC 517 Petroleum and petroleum products 11.49
Table II. SIC 514 Groceries and related products 10.21
The coefficient of Overall durable versus non-durable goods
variation for various SIC 50 Durable goods – total 18.63
goods by standard SIC 51 Non-durable goods – total 12.42
industrial classification
(SIC) 1988-1997 Data source: Survey of current business
goods are more difficult to forecast. An underlying rationale for these Sales forecasting
differences stems from the notion that the more perishable an item is, the
easier it is to forecast.
The forecast variability for durable products has economic justifications.
The more durable a good is (the longer it lasts), the higher the cost of the item.
When a product is durable, customers can postpone their purchasing decision
to a time that is more advantageous. This postponement is usually to a time 1023
when the customer has more disposable income (usually during upswings in
the economy). Consequently, forecasting highly durable goods is more difficult
since these goods are susceptible to external sources of error such as economic
changes. Durable goods also often have durable components and durable
components may have long lead times. Organizations that produce durable
goods that have higher costs and longer lead times need to use the most
accurate forecast. This difference between products leads to paradox 3:
Paradox 3. The organizations that need the most accurate forecast have the
largest forecast error.

Reducing the adverse effect of forecast error on decisions

Forecasts at all levels of aggregation have been shown to have significant
errors (see Table I). This table also illustrates that the forecast error is usually
between 10-25 percent, as measured by mean absolute percent error. A logical
conclusion is that forecast error is inevitable and managers are left to contend
with the impact of the errors. Although forecast error increases for longer-term
forecasts and more specific products, there are ways organizations can reduce
the adverse effects of forecast error. To some degree, organizations cause
themselves unnecessary difficulties from the inevitable, less than perfect
forecasts. The following sections describe three suggestions for reducing the
effect of forecast error on decisions.

Relate specific resource decisions to specific time frames

An earlier discussion describes how specific forecasts for specific periods of
time impact resource decisions. These decisions are also affected by the three
paradoxes described above. The adverse effects of forecast error are most
severe for the most important decisions, for the most specific information, and
for certain types of costly goods. Managers try to minimize the impact of
forecast errors on resource decisions by following this general rule, ``Postpone
resource decisions as long as possible.’’ This rule affects both the forecast’s
time dimension and specifics on product definitions. From a time standpoint,
the rule implies that managers should not forecast farther into the future than
the resource decision requires. For the product definition dimension, managers
should not forecast a more specific product than is necessary to make the
resource decision.
For example, one company contacted by the authors purchased material by
specific item for two years into the future. Their rationale was: ``It is nice to
know what you are going to buy in the future so you can make strategic
IJOPM purchasing decisions (long-term contracts, supplier partnerships, etc.).’’
22,9 Unfortunately, the forecast error of 25 percent+ makes these forecasts
impractical since they give the supplier a false sense of confidence in the orders
for the specific components. Fortunately in this case, the organization was
allowed to change its order to the supplier since the supplier wanted to
establish a long-term relationship with the company. A better approach would
1024 have been to write a contract for a dollar amount and ``firm up’’ the specific
components much later. From a theoretical perspective, tying resource
decisions to specific time periods enables managers to delay decisions to a later
time period, thereby reducing the cost of the forecast error.
In the enterprise resource planning literature, time fences have been
suggested as a method to control forecast effects on resource decisions. Time
fences act as a trigger to ``firm up’’ the resource plan against the forecast and
serve as a mechanism to freeze the schedule to ensure no changes are made that
will impact specific resources. The immediate time period fence (product time
fence) specifies the time when material may be on order and schedule changes
should not occur. Schedule changes that occur inside the product time fence are
likely to cause material shortages or excess inventory. The short time period
fence (demand time fence) specifies the time where demand drives the capacity
requirements (usually labor) of the organization. Additional customer orders
placed inside the capacity time fence usually cause labor shortages or slack
time. The medium time fence (the planning time fence) extends to the point
where the company can still acquire new equipment to meet significant
changes in demand. Changes to the equipment plan inside the planning time
fence result in insufficient time for the company to receive and install new
equipment to meet demand. The long-term time fence (capacity time fence)
extends to the point where the company would have adequate time to acquire a
new plant/facility to meet demand. Changes made inside the capacity time
fence result in the inability to deliver product to the customer from that facility.
Each decision is tied to a specific time fence and these time fences can be
established using manufacturing planning software that will assure the user
that changes will not impact strategic decisions. In short, because the forecast
error is larger the longer the forecast time horizon, resource decisions should be
postponed to the last possible moment. Companies can reduce the impact of
forecast error by tying the resource decision to specific time fences since it uses
the ``best’’ information at the latest possible moment. Table III summarizes the
resource allocation decision, type of planning horizon, and the time fence.

Resource allocation decision Planning horizon Time fence name

Table III. Material Immediate time period Product time fence

Resource allocation Labor Short time period Demand time fence
decisions and their Equipment Medium time period Planning time fence
time fences Plant Long time period Capacity time fence
Generally, the longer the time fence the less specific the forecast should be, as Sales forecasting
resource decisions made based on long-term forecasts are not product-specific.
For example, plant size is more product-flexible than plant equipment; plant
equipment and labor are more flexible than purchased material. Managers can
make the accuracy of the forecast less important by setting time fences and
keeping the forecast (and the resources) as general as possible for as long as
possible. The time fences serve as ``trigger’’ mechanisms to ensure that the time 1025
period does not pass before the decision is made.
Managers can reduce the adverse effects of forecast accuracy by
understanding when to make resource decisions. Consequently, because
managers can reduce these adverse effects, empirical evidence suggests that
managers rate forecast flexibility, ease of use, and ease of implementation
almost as important as forecast accuracy in selecting a forecast method
(Yokum and Armstrong, 1995). Managers that understand the tie between the
resource decision and the forecast horizon can be more flexible in their resource
plans on long-term decisions. This flexibility facilitates better implementation
of forecasts and better strategic planning.

Agree on one forecast

A number of years ago, one of the authors was called on to improve the forecast
at a company. During the visit, the author asked the top-level managers for the
following year’s forecast. The president stated there would be a 15 percent
increase in sales for the next year. The vice-president of marketing stated his
staff rolled-up the sales quotas and determined the forecast. The vice-president
of manufacturing operations stated his managers used an exponential
smoothing method using last year’s production data to create a forecast, and
the controller stated that the operations research group (under his leadership)
had determined the sales using a statistical model. As might be expected, all
the forecasts were significantly different from each other. The president was
expecting a certain increase in sales, marketing was planning on another level
of sales, production was producing something other than what marketing
needed, and the pro forma statements for the next year were budgeting
resources based on another forecast.
It immediately became apparent that the forecasting method was not the
problem at all. The organization was producing one thing, it was selling
something else (resulting in stockouts for some items and excess inventory in
other items), and the firm was budgeting for other production levels. The
forecast model was not the problem, but the problem was the lack of a basic
agreement on which was the correct forecast, resulting in excess inventories in
some products and shortages in others. A simplistic interpretation of this
situation is that they had too many forecasts; however, each functional area
used the forecast for their own purposes. To some degree, each function felt
they were justified in developing their ``own’’ forecast. When each separate
functional area generates a forecast for a different purpose, the result is
IJOPM misallocated resources and poor performance on strategic goals (Hayes et al.,
22,9 1984).
Studies on how final forecasts are derived in an organization help provide
additional insight into the difficulties of developing organization-wide
forecasts. Plossl’s (1973) seminal article on ``Getting the most from forecasts’’
provides insight into how to improve the forecast. One of Plossl’s suggestions
1026 is that the forecast should be a group cooperative effort. Empirical research has
tested his suggestion and provides rather interesting results. Plossl’s assertion
that getting more functions involved would improve the forecast could not be
borne out by empirical tests. Studies by Wacker and Sprague (1995, 1998) did
not support Plossl’s assertion that the number of functions involved in the
process improved the forecast. Their results show that having many functions
involved in forecast development does not improve forecast accuracy. A
plausible explanation is that firms are not cautious in determining the type of
information that forecasters bring to the meeting. From a practical perspective,
all functions may not have access to market information, and as a result reduce
forecast accuracy by incorporating self-interests. The mere fact that everybody
is involved with the forecast does not assure improved accuracy.
An additional forecast development issue is the degree of involvement of top
level management. Plossl warns that the forecast should not be a ``hopes and
wishes’’ forecast which may occur if top management is involved. The ``hopes
and wishes’’ scenario results when top-level managers’ performance is
measured by higher sales. Plossl’s hypothesis is supported by empirical results
indicating top management involvement does not improve forecast accuracy
(Wacker and Sprague, 1995). The president may be responsible for the forecast,
but the president may not have the best market information and may bias the
forecast by making the forecast too ``rosy.’’ There is no empirical evidence to
suggest that involvement by top-level management results in improved
forecast accuracy.

Combine models with managerial judgmental in forecasting

Results from Sanders and Ritzman (1989) and the ``M’’ competition (Makridakis,
1989) lead to one major conclusion: simple forecasting models (moving
averages, Holt-Winters, trend analyses, etc.) perform as well as complex models
(econometric, input-output analyses, Box-Jenkins, Fourier analyses, spectral
analyses, etc.). Both the complex time series and causal models were less
accurate than simple models for two reasons. First, the complex time series
models (Box-Jenkins, Fourier analyses, and spectral analyses) assume that
sales are related to time periods. Second, the forecasts are limited by a lack of
theoretical basis as described earlier. Since the complex forecasting methods do
not improve forecast accuracy, it is important to use managerial judgement to
improve the accuracy.
The high cost of complex forecasting models suggests that managers expect
greatly improved accuracy and are relatively dissatisfied with the models when
they are not more accurate. Empirical studies on the level of satisfaction with
sales forecasts indicate that forecast users are generally more dissatisfied with Sales forecasting
complex forecast methods compared to simple forecasting techniques (Mentzer
and Kahn, 1995). The complex forecast models, therefore, have two major
disadvantages: they do not greatly increase forecast accuracy, and their high
cost leaves managers viewing these forecasts as a waste of money and leaving
them with ``disconfirmed expectations’’.
A second conclusion from empirical studies is that combining several simple 1027
techniques improves forecast accuracy, but little theoretical work has focused
on why this is true (Makridakis et al., 1982). Makridakis (1989) suggested
several hypotheses on why combining forecasts improves forecast accuracy.
These reasons include:
measurement errors always exist and combining makes them smaller;
unstable or changing patterns or relationships cause single models to be
unreliable, but combining several models improves their accuracy; and
models usually minimize past data errors that were poor predictors of
future sales.
Makridakis (1989) also offers suggestions to improve forecast accuracy
through combining forecasts:
use various forecasting techniques and use empirical findings to help
with formulation of the model,
use complementary methods such as simple exponential smoothing with
trend or simple exponential smoothing with judgmental forecasts.
Combining models with judgmental inputs gives forecasters the best
information on past data patterns. Judgement forecasts have also been shown
to improve forecast accuracy because humans may be better able to detect
patterns in time series data and to integrate outside information (Remus et al.,
1995). Thorough understanding of ``how and why’’ judgmental forecasting
improves the forecast requires a better understanding of the cognitive
processes involved (Goodwin and Wright, 1993). More field research is required
to understand the cognitive processes used in judgmental forecasting.

Research contributions and suggestions for future research

The forecast literature to date does not offer a managerial understanding of the
role of forecasting in resource allocation planning and how managers can
reduce the impact of forecasting errors on resource planning. This article
provides practical suggestions for better understanding the limitations of
forecast methodology and how to better cope with the impact of forecast error.
The goal of managers is to reduce the financial impact of forecast errors. This
can be accomplished by overcoming two traditional forecasting shortcomings.
The forecasts must be relevant, meaning they have a stated purpose and
address the client’s needs. Second, the forecast must be placed in the broader
IJOPM decision-making context of how it will be used as part of a company’s strategic
22,9 planning process
Managers insist on accurate forecasts because of the costs associated with
the replanning of labor, material and equipment needs. This study has
suggested three basic forecasting paradoxes that can impact strategic resource
decision making. Paradox 1 states that the most important strategic decisions a
1028 company makes are based on the least accurate information. Paradox 2 says
forecast information that is most useful for resource planning is the least
accurate. Paradox 3 states that organizations that need the most accurate
forecast have the largest forecast error. The paradoxes arise from several
issues. Forecasts for wide product groups and for short-term purposes are the
most accurate, yet these forecasts do not provide the information needed for
strategic resource planning. Generally, forecasts for durable goods are less
accurate than forecasts for perishable goods. Firms that produce higher cost,
longer lead-time durable goods can expect larger forecast errors.
Although there is little that management can do to eliminate forecast errors,
organizations have several alternatives to reduce the adverse effect forecast
errors have on strategic resource planning. These alternatives include:
Managers can make better decisions by recognizing that forecasts are
used for specific resource decisions that have time specific frames. A
forecast should be no more detailed than the resource decision
requires. As much as possible, the forecast should be non-product
specific and should be tied to specific time fences for each resource.
Although forecasts must be agreed upon by all functional areas, there is
no empirical evidence to suggest that increasing the number of functions
(or top management) involved in the forecast development will improve
forecast accuracy.
There is no empirical evidence that increasing the statistical
sophistication of the forecast model improves forecasting accuracy;
however combining forecast methods improves forecast accuracy,
especially when the combination includes a judgmental forecasting
method.
These discussions of forecasting and how it is used to make resource allocation
decisions lead to several research questions which must be answered if
forecasters are to help managers improve their decisions. These questions
include:
How do the resource allocation decisions and their time frames vary by
industry? How does forecast accuracy affect resource allocations?
What strategic policies should organizations adopt to enable their
managers to react more effectively to forecast errors?
Are forecast performance measures available that relate the accuracy of
the forecast with the use of the forecast? How are forecast accuracy and
 its related forecast resource decision evaluated to determine the impact Sales forecasting
of the forecast error on the decision’s effectiveness?
How can communication be improved between forecast users and those
who prepare the forecasts? What policies should the forecast developers
use to ensure that the forecast is disseminated to all functions?
This study was designed to give strategic managers practical suggestions for 1029
better understanding the limitations of the forecast. Through understanding
these limitations, managers can plan their reactions to forecast errors to enable
better performance. These suggestions are based on the recognition that
forecasts are derived to serve specific resource decisions, and that these
decisions must be congruent with the company strategy. Future forecasting
research should focus on broadening the understanding of the role of forecasts
in strategic decision making.
Through a review of the forecasting literature and an understanding of the
implications of the forecasting paradoxes, a need for future forecasting
research has been identified. Research is required that further analyzes the
impact of forecast error on managerial decision making. Decision makers are
not interested in the accuracy of the forecast method, but rather the success of
the forecast in decision making. Organizations should focus on alternatives to
reduce the adverse effects of forecast errors. Future forecasting research must
focus on broadening the understanding of the role of forecasts in strategic
decision making.

References
Armstrong, J.S. (1984), ``Forecasting by extrapolation: conclusions from 25 years of research’’,
Interfaces, Vol. 4 No. 6, pp. 52-66.
Biggs, J.R. and Campion, W.M. (1982), ``The effect and cost of forecast error bias for multiple-
stage production-inventory systems’’, Decision Sciences, Vol. 13 No. 4, pp. 570-84.
Bunge, M. (1967), Scientific Research 1: the Search for System, Springer-Verlag, New York, NY.
Chalmbers, J.C., Mullick, S.K. and Smith, D.D. (1971), ``How to choose the right forecasting
technique’’, Harvard Business Review, July-August, pp. 45-74.
Fildes, R. and Beard, C. (1992), ``Forecasting systems for production and inventory control’’,
International Journal of Operations & Production Management, Vol. 12 No. 5, pp. 4-27.
Fischhoff, B. (1994), ``What forecasts (seem to) mean’’, International Journal of Forecasting,
Vol. 10, pp. 387-404.
Goodwin, P. and Wright, G. (1993), ``Improving judgmental time series forecasting: a review of
the guidance provided by research’’, International Journal of Forecasting, Vol. 9, pp. 147-61.
Hayes, R.H., Wheelwright, S.C. and Simon, H.A. (1984), Restoring Our Competitive Edge:
Competing through Manufacturing, John Wiley & Sons, New York, NY.
Holt, C.C., Modigliani, F. and Simon, H.A. (1955), ``A linear decision rule for production and
employment scheduling’’, Management Science, Vol. 2 No.1, pp. 10-30.
Lee, T.S. and Adam, E.E. Jr (1986), ``Forecast error evaluation in material requirements planning
(MRP) production and inventory systems’’, Management Science, Vol. 32 No. 9, pp. 1186-205.
IJOPM Makridakis, S., Anderson, N. H., Carbone, R., Fildes, M., Hibon, R., Lewdowski, J., Newton, E.,
Parzens, E. and Winkler, R. (1982), ``The accuracy of extrapolation (time series) methods:
22,9 results of a forecasting competition’’, Journal of Forecasting, No. 1, pp. 111-53.
Makridakis, S. (1989), ``Why combining works?’’, International Journal of Forecasting, Vol. 5,
pp. 601-3.
Makridakis, S. (1996), ``Forecasting: its role and value in planning and strategy’’, International
Journal of Forecasting, Vol. 12, pp. 513-37.
1030
Mansfield, E. (1988), ``The speed and cost of industrial innovation in Japan and United States’’,
Management Science, Vol. 34 No. 10, pp. 1157-68.
Mentzer, J.T. and Cox, J.E. Jr (1984), ``Familiarity, application, and performance of sales
forecasting techniques’’, Journal of Forecasting, Vol. 3, pp. 27-36.
Mentzer, J.T. and Kahn, K.B. (1995), ``Forecasting technique familiarity, satisfaction, usage, and
application’’, Journal of Forecasting, Vol. 14, pp. 465-76.
Naylor, T. (1983), ``Strategic planning and forecasting’’, Journal of Forecasting, Vol. 2, pp. 109-18.
Plossl, G. (1973), ``Getting the most from forecasts’’, Production and Inventory Management
Journal, First Quarter, pp. 1-15.
Remus, W., O’Connor, M. and Griggs, K. (1995), ``Does reliable information improve the accuracy
of judgmental forecasts?’’, International Journal of Forecasting, Vol. 11, pp. 285-93.
Ritzman, L.P. and King, B.E. (1993), ``The relative significance of forecast errors in multistage
manufacturing’’, Journal of Operations Management, Vol. 11, pp. 51-65.
Sanders, N.R. and Mancroft, K.B. (1994), ``Forecasting practices in US corporations: survey
results’’, Interfaces, Vol. 24 No. 2, pp. 92-100.
Sanders, N.R. and Ritzman, L.P. (1989), ``Some empirical findings on short-term forecasting:
technique complexity and combinations’’, Decision Sciences, Vol. 20 No. 3, pp. 635-40.
Vollman, T.E., Berry, W.L. and Whybark, D.C. (1992), Manufacturing Planning and Control
Systems, 3rd ed., Richard D. Irwin Corp, Homewood, IL.
Wacker, J.G. (1998), ``A definition of theory: implications for different theory-building research
methods in operations management’’, Journal of Operations Management, Vol. 16, pp. 361-85.
Wacker, J.G. and Hanson, M. (1997), ``Some practical advice to manufacturing managers:
Empirical results from the Global Manufacturing Research Group’’, Production and
Inventory Management Journal, Vol. 38 No. 3, pp. 64-72.
Wacker, J.G. and Sprague, L. (1995), ``The impact of institutional factors on forecast accuracy: a
manufacturing executive perspective’’, International Journal of Production Research,
Vol. 33 No. 11, pp. 2945-58
Wacker, J.G. and Sprague, L.G. (1998), ``Forecasting accuracy: comparing relative effectiveness of
practices between seven developed countries’’, Journal of Operations Management, Vol. 16,
Summer, pp. 271-90.
Yokum, T.J. and Armstrong, J.S. (1995), ``Beyond accuracy: comparison of criteria used to select
forecasting methods’’, International Journal of Forecasting, Vol. 11, pp. 591-7.

Further reading
Armstrong, J.S. (1983), ``Relative accuracy of judgmental and extrapolative methods in
forecasting annual earnings’’, Journal of Forecasting, Vol. 2, pp. 437-47.
Armstrong, J.S. (1989), ``Combining forecasts: the end of the beginning or the beginning of the
end’’, International Journal of Forecasting, Vol. 5, pp. 89-98.
Chatfield, C. (1988), ``Apples, oranges and mean square error’’, International Journal of
Forecasting, Vol. 4, pp. 515-18.
 Dalrymple, D.J (1987), ``Sales forecasting practices: results from a United States survey’’, Sales forecasting
International Journal of Forecasting, No. 3, pp. 379-91.
Makridakis, S. (1993), ``Accuracy measures: theoretical and practical concerns’’, International
Journal of Forecasting, Vol. 9, pp. 527-30.
Makridakis, S. (1993), ``The M2-competition: a real time judgmentally based forecasting study’’,
International Journal of Forecasting, Vol. 9, pp. 5-22.
O’Connor, M., Remus, W. and Griggs, K. (1983), ``Judgmental forecasting in times of change’’, 1031
International Journal of Forecasting, Vol. 9, pp. 163-72.
Thompson, R.A. (1990), ``An MSE statistic for comparing forecast accuracy across series’’,
International Journal of Forecasting, Vol. 6, pp. 219-27.

View publication stats