Electronic copy available at: http://ssrn.com/abstract=663661
Published in International Journal of Forecasting, 3 (1987), 355-376.
Forecasting Methods for Marketing:
Review of Empirical Research1
J. Scott Armstrong
The Wharton School, University of Pennsylvania
Roderick J. Brodie
University of Canterbury, New Zealand
Shelby H. McIntyre
Santa Clara University, CA
Abstract: This paper reviews the empirical research on forecasting in marketing. In addition, it presents results
from some small scale surveys. We offer a framework for discussing forecasts in the area of marketing,
and then review the literature in light of that framework. Particular emphasis is given to a pragmatic
interpretation of the literature and findings. Suggestions are made on what research is needed.
Keywords: Bootstrapping, Econometric models, Expert systems, Intentions, Judgment, Market forecasting, Market
Casual observation indicates that forecasting is important to marketing practitioners. This is supported by
Dalrymple (1987) in his survey of 850 U.S. companies. Of the 134 companies that responded, 98.6% prepare
formal forecasts when they prepare formal marketing plans. This is in concert with the 93% of companies in an
earlier survey who reported that sales forecasting was “one of the most critical,” or a “very important” aspect of
their company’s success [Dalrymple (1975)]. Cerullo and Avila (1975), in a survey of the “Fortune 500 Largest
Industrial Companies,” found that 98% of the respondents thought that forecasting should be taught at business
schools. Also, sales forecasting was the most common of the nine activities in a survey of 353 marketing directors in
British textile firms [Jobber, Hooley and Sanderson (1985)].
However, practitioner interest in forecasting is not matched by academic attention in the marketing discipline.
A content analysis of 53 textbooks on marketing management and marketing research revealed that forecasting was
mentioned on less than one percent of the pages. (The sample included all relevant books that could be found at the
University of Pennsylvania library). Perhaps this is because forecasting is a new area for research. Or perhaps it is
because forecasting has not been explicitly addressed, though it is implicit in academic work. Another possibility is
that forecasting is more practical than theoretical and, as such, is looked down upon by some academics.
This difference between academic and practitioner interest was also found in a 1986 survey of the members of
the International Institute of Forecasters (IIF). Among practitioners, marketing people represented the largest group
of respondents out of seven categories (27% of 115 practitioners). In contrast, they were next to last in size among
academic groups (10% of the 87 academic respondents). This difference in apparent interest is significant at
1 This paper was intended to be published simultaneously in the Singapore Marketing Review. Due to differences in
the production schedules, this version in the International Journal of Forecasting appears later. This allowed time
for significant revisions. The Gunning Fog index for this paper is about 14. Acknowledgements: Rob Colgrove,
Douglas Dalrymple, Jehoshua Eliashberg, Robert Fildes, Michael Geurts, Kristiaan Helsen, Janet Kass, Peter
Leeflang, Essam Mahmoud, Spyros Makridakis, Stephen Schnaars, and Alan Singer provided helpful comments on
Electronic copy available at: http://ssrn.com/abstract=663661
p < 0.01. (An alternative hypothesis is that marketing academics respond to questionnaires relatively less frequently
than do other academicians.)
Forecasting is an important area for research. As evidence of this, consider the fact that papers in the Journal
of Forecasting ranked seventh among the 84 journals in management, business, and planning, with respect to the
citation impact factor for the years 1982 and 1983 (as reported in Journal Citation Reports, 1984, 1985). The
work done to date, much of it by academics outside of marketing, has had important implications for marketing
decision makers, as will be shown.
Makridakis and Wheelwright (1977) described various forecasting techniques and tried to show how they
were relevant to marketing. Our aims are a bit different. We hope to (1) summarize the more important empirical
research that relates to the choice and application of forecasting methods in marketing, (2) show how the research
from this special issue links with the existing research on forecasting, and (3) provide an agenda for future research.
The paper is organized to provide a framework for forecasting and then to examine research contributions within
2. A Framework for Marketing Forecasting
There are many factors to consider when forecasting for marketing decision-making. For instance, one could
(1) What forecasts are needed (e.g., sales, market share, competitive reactions, marketing costs)?
(2) What situation exists (e.g., stage of the product life cycle, state of the economy, degree of regulation in
(3) What forecast horizon is appropriate (e.g., current status, short-, medium-, or long-range)?
(4) What data are relevant and available?
(5) With what frequency must the forecast be prepared?
(6) Who will prepare the forecast, and how much time and resources will be committed to the task?
(7) Who will use the forecast, and in what manner?
(8) What process is to be used (e.g., how are the data to be gathered and analyzed, and how is the forecast to
(9) When is the forecast needed?
(10) What uncertainty measures are needed?
Certainly there are other relevant factors, but these serve to illustrate the complexity involved. The list implies
thousands of different circumstances. For example, one could examine a judgmental forecast of the long-range
expected market sales for a new product with few data, where the intent is to assess the effect on stockholders. In
such a situation, one might then consider various forecasting methods.
Although a comprehensive review of the literature was attempted, space does not allow discussion of the
research relevant to all of these aspects of forecasting. Instead, we have generated a framework for discussion and
hope to highlight an important subset of research contributions. This paper is centered around the first item in the list
of factors: “what forecasts are needed.” These needs are listed as “A Framework for Marketing Forecasts” in Exhibit
1. The arrows indicate the normal relationships among the factors in the analysis. One would start with the tasks at
the top and move down the page. The discussion below examines the boxes in that sequence.
Exhibit 1. A framework for marketing forecasts.
Can one make better forecasts of the organization’s external environment (the economy, population, social
trends, technological change, or government legislation)? If yes, will this significantly improve the ability to make
better market forecasts? The answers to these questions seem to be “probably,” and “only in certain situations,”
Substantial resources have been devoted to obtaining better forecasts of environmental factors such as
population, weather, resources, income, technology, and government controls. Ascher’s (1978) historical review of
forecast accuracy in many of these areas suggested that large errors are common and that, in practice, little
improvement has been achieved over the past few decades.
Improved environmental forecasts should lead to more accurate market forecasts. Surprisingly, research in this
area indicates that highly accurate environmental forecasts are not required. Consider the following findings:
(1) Measurement error in the causal variables (e.g., the environmental inputs to a market forecasting model) had
little impact on the accuracy of an econometric model in the few studies done on this topic [Denton and Kuiper
(1965), Denton and Oksanen (1972) and McDonald (1975)].
(2) Conditional econometric forecasts (those made with actual data on the causal variables) generally have not
been found to be more accurate than unconditional forecasts (where the causal variables themselves must be
forecasted). In fact, of 18 studies to date, only three have shown conditional forecasts to be more accurate, five
showed no difference, and ten showed them to be less accurate [Armstrong (1985, pp. 241-242), Rosenstone (1983)
and four studies from Fildes (1985)]. This inferiority of conditional forecasts is statistically significant (p < 0.05
using the sign test). A possible explanation for these strange findings is that the comparisons may not be fair. The
unconditional forecasts often include subjective revisions that may reduce the error in estimating the starting value;
such adjustments were not typically used for the conditional forecasts. However, the same perplexing result shows
up in the carefully done study by Dhrymes and Peristiani (1987). Other explanations are possible. Nevertheless, one
would not argue for less accurate forecasts of the causal variables.
The above results are useful. They imply that for many situations (typically those involving short-term market
forecasts) it is not fruitful to devote a large portion of the budget to obtaining highly accurate environmental
forecasts. These conclusions probably would not apply to situations involving large changes (e.g., long-range
forecasts). Additional research is needed in this area.
Still, much of the research has been valuable, particularly with respect to econometric methods. Econometric
methods have been widely used for environmental forecasts. To date, most econometric researchers have devoted
their efforts to short-term forecasting, an area that has yielded unimpressive or contradictory results [Armstrong
(1985), Fildes (1985)]. Econometric methods would be expected to be more useful for long-range forecasting
because the changes in the causal variables are not swamped by random error, as in the short run. In fact,
econometric methods are more accurate. Armstrong (1985, chapter 15) reported seven empirical comparisons of
methods used in long-range forecasting. In all comparisons econometric methods were more accurate than
extrapolations. Also, there was a 3 to 0 advantage for econometric versus subjective forecasts. Fildes (1985) located
20 studies on long-range forecasting; he coded them as 15 showing econometric to be more accurate, 3 ties, and 2
showing econometric to be less accurate than other methods.
Surveys have also proven useful in forecasting environmental trends. For a summary of some of the more
successful consumer surveys (e.g., the Conference Board Survey), see Linden (1982) and Curtin (1982).
When forecasting environmental factors that are related to market prices, it is important to remember what might
be called Adam Smith’s Rule for Forecasters: “Forecasters cannot beat the market.” (Some people refer to this as the
rule of efficient markets.) In other words, when an active market of buyers and sellers is at work (such as in stocks,
bonds, money, commodities, land, and football), forecasters have not had much success at finding methods that can
improve upon the market’s forecast of prices. This rule assumes that the forecaster lacks inside information, so the
market price is a reflection of available information. Thus, the market forecasts prices as effectively as can any
existing forecasting method [for an example, see Brandt and Bessler (1983)]. To our knowledge, Adam Smith’s
Rule for Forecasters holds up rather well. This is one area where further research is not a high priority.
Organizations should have a system for scanning the environment to be sure that they do not overlook variables
that may have a large impact on their market. These variables can be tracked in a firm’s marketing information
system. Periodic brainstorming with a heterogeneous group of experts should be sufficient to identify which
variables to track. The key is to identify the important variables and the direction of their effects. Once identified,
only crude estimates of the coefficients of these variables are typically sufficient in order to obtain useful forecasts.
When large shocks are encountered, more sophisticated approaches may be useful. Leone (1987)
examined “intervention analysis” to assess the impact of an environmental shock such as the American
Dental Association’s endorsement of Crest toothpaste in 1965. His analysis showed that this model was
more accurate than several naive models in forecasting market share.
Research on market (industry) forecasting has produced a number of useful conclusions. This research
enables substantially more accurate forecasts today than 25 years ago. This section describes some of that
One of the major sources of error in forecasting, especially in short range forecasting, is estimation of
the current status. For example, in trying to forecast beer sales for next year, it is important to know the
current level of beer sales. Here are some conclusions:
(1) Since the most recent information is the most useful [e.g., Ash and Smyth (1973), Pashigian
(1964)], it follows that more frequent updating can improve forecasting accuracy. Computer information
systems allow for inexpensive and frequent updating of the current sales estimates from detailed data
bases. This means that the size of the aggregate market and the various submarkets can be estimated more
quickly and with greater precision. For instance, in the packaged goods industries, marketers can use
weekly scanner data at the Standard Metropolitan Statistical Area level delivered within a week, whereas
a few years ago they used two-month aggregates delivered six weeks after the fact and covering only 32
(2) Structured judgmental methods apparently improve the estimate of current status [Armstrong
(1985, p. 237)]. For an example, see Best’s (1974) use of the Delphi technique to estimate current demand
for the Oregon Business Review.
(3) Econometric methods for the estimation of the current status produce modest improvements in
accuracy in comparison to using only trade and production estimates [see Armstrong (1985, p. 236)]. An
interesting issue is how much expertise is needed for judgmental forecasting. Surprisingly, research to
date [Armstrong (1985, pp. 91-96)] indicates that high expertise in the subject area is not important for
judgmental forecasts of change. It is, however, important for assessing current levels. An important
conclusion, then, is not to spend heavily to obtain the best experts in the field to forecast change. But one
should avoid people who clearly have no expertise.
Extensive research over the last two decades has examined biases that occur in judgmental forecasting.
[For summaries, see Kahneman, Slovic and Tversky (1982) and Hogarth and Makridakis (1981).] Among
these biases are optimism, conservatism, anchoring, and an overemphasis on easily available data. While
some sources of bias have been identified, little knowledge exists as to how these biases affect marketing
forecasts. For example, how much bias might one expect for a new product forecast? Tyebjee (1987)
identifies several sources of this bias, and Tull (1967) provides evidence on the amount of such bias.
These findings might be used to adjust new product sales forecasts.
One of the most important findings about judgmental forecasting is that there is safety in numbers. It
is useful to combine judgmental forecasts. For this, only five to 20 experts are needed [Hogarth (1978)
and Libby and Blashfield (1978)]. But how should these forecasts be combined? Advice to date, based on
limited research, has favored the use of the median rather than the mean [Agnew (1985), Larroche and
In the previous section, it was noted that causal methods provide more accurate long-range
environmental forecasts. This applies also to market forecasts. These causal models need not be complex.
According to the empirical research, the accuracy of the forecast is not highly sensitive to the estimates of
the relationships [Armstrong (1985, pp. 225-232) reviews this evidence]. For example, the accuracy of a
forecast of automobile sales is not highly sensitive to estimates of the income elasticity. However, it is
important to identify the key variables and the directions of their relationships with the forecasted variables. Crude
estimates can then be made of the size of the relationship. The estimates of the magnitudes are likely to be important
for very large changes in the causal variables but, as yet, there is little evidence to support this viewpoint. In spite of
the failures of prior work, much current research is being devoted to the search for better estimation procedures.
Yokum and Wildt (1987) report some improvements due to sophisticated estimation procedures in rapidly
changing and complex markets.
Econometric models are especially useful for markets undergoing large changes. The evidence on this [from
Armstrong (1985) and Fildes (1985)] was described in the section on environmental forecasts. However, there is
need for further research on the conditions under which econometric models are useful in market forecasting.
An interesting consideration is whether marketing principles and concepts, such as the product life cycle (PLC),
enable better judgmental or econometric forecasts. Kotler (1984, p. 358) says: “A firm that is considering a new
product should forecast the PLC shape based on factors that influence the length of each stage.” But evidence to
support this statement is difficult to find [Rink and Swan (1979)]. Research using diffusion models may produce
some evidence in this area. For a discussion of the state of the art, see Mahajan and Wind (1986).
Company plans typically require the cooperation of many people. For example, if the organization decides to
implement a given marketing strategy, will it be able to carry out the plan? Often the organization fails to do what it
intends to do because of a lack of resources, misunderstanding, opposition by key stakeholders, or a lack of
commitment to the plan by key people. The need to forecast organizational behavior is frequently overlooked. Better
forecasting here might lead to more realistic plans and to plans that are easier to implement.
Surveys of key stakeholders may help to assess the likelihood that a given strategy can be implemented.
Because those who are not committed to a plan may be reluctant to admit it, projective questions may be useful
when asking about intentions.
The actions of a member in the organization are likely to depend on what others in the group do. For this
reason, group depth interviews may be helpful in assessing the commitment to implementation.
Sometimes people are unable to predict their own actions because they lack sufficient experience. For
example, how would someone behave if asked to implement a proposed strategy that requires learning a new job?
Simulations of this new situation, using techniques such as work samples or role playing, may be useful.
Actions by those outside the firm also affect decision-making. Role playing is well-suited to forecasting how
people will respond to these pressures. For example, accurate forecasts have been obtained by role playing (vs.
unaided opinions) in forecasting how a company would respond to government pressure to remove a product from
the market because of safety considerations [Armstrong (1987)].
In summary, little research has been aimed at forecasting company actions. These suggestions for projective
tests, group depth interviews, work samples, and role playing are based mostly on speculation.
Competitors may alter their actions in response to changes in the environment, actions by other firms, or
dissatisfaction with their own past performance. To what degree is it possible to forecast the actions of
competitors? The development of a successful marketing strategy often depends upon having a good forecast
of the actions and reactions by competitors. Most of the research on market share forecasting has assumed that
these actions are known [e.g., Brodie and de Kluyver (1987)]. After forecasting competitive actions and
reactions, one must then try to forecast the impact of these actions upon market share. This latter task has been
the primary concern of the research published to date. We think the problem of forecasting competitive action
deserves more attention than it has received to date.
A variety of techniques can be used to forecast competitive actions. These are summarized in Exhibit 2.
This exhibit also indicates the frequency of usage based upon those respondents who identified themselves as
practitioners (from various fields) in the survey of IIF members. There were 59 usable responses. Some
respondents indicated that their organization used more than one technique. Expert opinion was the most
frequently used method (85% of firms), followed by extrapolation with 58%. At the low end of the scale, only
7% of the firms reported using role playing.
Little evidence exists on the predictive validity of the methods in Exhibit 2. One way to assess relative
accuracy is simply to ask experts. We surveyed marketing experts at the European Marketing Association
Conference in Helsinki in June 1986. We also surveyed forecasting experts at the International Symposium on
Forecasting in Paris in June 1986. The results, in Exhibit 2, show much diversity of opinions. The role playing
approach, which was in last place among marketing experts, was the second most frequently recommended
approach by the forecasting experts.
Exhibit 2. Methods to forecast competitive actions.
of experts (%)
(n = 59)
(n = 15)
(n = 18)
Expert opinion (experts who know about the situation
Extrapolation (statistical analysis of analogous situations)
Intentions (ask the competitor)
Experimentation (trying the strategy on a small scale)
Game theory (formal use of game theory)
Role playing (formal acting out of the interactions involved)
One obvious area of interest with respect to forecasting competitive action is that of competitive bidding.
Edelman (1965) reported substantial improvements for RCA’s bidding when a model was used instead of
unaided judgment. The model, which used managers’ judgments as inputs, was superior on six of seven bid
submissions. Replications or extensions of this research have been difficult to find.
Research on conflict situations suggests that role playing provides substantially more accurate forecasts
than can be obtained from expert opinion [Armstrong (1987)]. This evidence consists of comparisons between
unaided opinions and role playing in seven situations. Predictions from a low cost role-playing procedure were
correct on 70% of the occasions, while opinions were correct on only 20%. However, none of these situations
involved conflicts among competitors.
Game theory offers another way to forecast the actions of competitors. Eliashberg et al. (1986) compared
game theory and group decision theory in predicting the price selected in a simulated bargaining situation.
Game theory did well in this comparison. Hopefully, further work will test game theory versus current
predictive methods in realistic situations.
Actions of Suppliers, Distributors, Government, and Other Stakeholders
It is often important to forecast the reactions of suppliers, distributors, and government in order to
develop a successful marketing strategy. On occasion, one might also need to forecast the actions of other
interest groups, such as “concerned minorities.” A range of techniques similar to those for forecasting
competitors’ actions appears useful, but little research investigates the relative accuracy of these techniques.
As with forecasting competitors’ actions, different techniques may suit different situations.
In an attempt to forecast the decisions by supermarkets, Montgomery (1975) developed a model of a
supermarket buying committee. Predictions were made about the shelving of a new product. The model, based
on information such as advertising for the product, reputation of the supplier, margin, product novelty, and
retail price, provided reasonable predictions for a hold-out sample.
In Armstrong (1987), role playing was used to forecast relations between suppliers and distributors. In
the role play, Philco (called Ace Company in the role play), a producer of home appliances, was trying to
improve its share of a depressed market. Philco had developed a plan to sell appliances in supermarkets using a
cash register tape discount plan. Secrecy was important because Philco wanted to be first to use this strategy.
Implementation of such a plan depended upon the supermarket managers. Would the plan be acceptable to
them? In this case, a simple role playing procedure produced substantially more accurate forecasts of the
supermarket managers’ responses (8 of 10 groups were correct) than did unaided opinions (1 of 34 groups was
correct). In the actual situation, the supermarket managers did accept the plan proposed by Philco.
(Incidentally, the change in distribution channels led to substantial losses for Philco.) The superior accuracy of
role playing relative to opinions seems to be due to its ability to provide a more realistic portrayal of the
Marketing practitioners and researchers have devoted considerable effort to assessing how marketing
variables such as product, price, place, and promotion affect market share. This assessment is often based on
the market share theorem [Bell, Keeney and Little (1975)]. The theorem states that market share is
proportional to the marketing effort of the brand divided by the sum of marketing effort of all the brands in the
market. The market share theorem gives rise to a number of alternative model specifications [see Brodie and de
Kluyver (1987)]. However, until recently, studies using this theorem have tended to describe applications
without testing forecast validity.
Various approaches to predicting market share are examined below. These include judgment, expert
systems, and econometric models.
Judgment: In practice, market share forecasts have been arrived at judgmentally, so it is useful to
examine ways to improve this process. Formal processing of judgmental forecasts has been attempted with
respect to the allocation of the effort of salespeople across accounts [Lodish (1971)] and also to improve
profits from better advertising budget levels [Little (1970)]. A field experiment by Fudge and Lodish (1977)
identified an 8.1% improvement in sales for an experimental group of 10 salespeople as compared to a control
group of 10 salespeople who did not use the formal or “bootstrapping” model (named CALLPLAN) to do their
forecasting and thereby plan their calling strategy.
However, in general, the value of these models has been hard to demonstrate. For instance, Chakravarti,
Mitchell and Staelin (1979) found that managers in a game simulation did not benefit from an advertising
model (ADBUDG). The managers’ intuitive judgment would have provided better forecasts than the model,
and an econometric approach would have been best of all [Chakravarti, Mitchell and Staelin (1980)]. McIntyre
(1981b) used a setting that did not involve carryover effects, and did not deal with determining the budget level,
but only with the allocation of a fixed budget across activities. In that experiment, the model (a modification of
CALLPLAN) helped users in comparison with a non-using control group. Apparently the value of judgment-based
models depends upon the situation and the question being addressed. The key issue is probably how well the
decomposition of judgments in the model matches the knowledge and abilities of the manager (user).
Decomposition seems especially appropriate where uncertainty about the forecasted variable is high [Armstrong,
Denniston and Gordon (1975)].
The forecasting of large changes in strategy, as with the introduction of new products, introduces an additional
dimension: it is often necessary to forecast without relevant historical data. Intentions surveys are widely used here.
For an early application, see Payne’s (1966) validation of the use of intentions to forecast sales of a new type of
Expert Systems: In general, an expert system involves converting a judgmental process to an objective process.
One approach is to ask the forecaster to describe the rules that she uses, a process referred to here as “direct
bootstrapping.” For example, one might ask: “How would market share change if you increased the quality control
for your product so that consumer complaints would be cut in half? ‘What would happen to market share if you
added feature x, while your competitors did nothing?”
Another approach to expert systems is to infer the rules statistically. Typically, this is done by regressing
various marketing strategies against the market share as predicted by experts, referred to here as “indirect
Bootstrapping has been widely applied in marketing [see Huber (1975a and 1975b) for early reviews on this
topic]. Occasionally it has been used with experts, but typically it is consumer intentions that are modeled. Over
1,000 marketing applications had been made by indirect bootstrapping of consumer intentions by the early 1980s
[Cattin and Wittink (1982)]. These applications have been done under the umbrella term “conjoint analysis.” Cattin
and Wittink lament the lack of validation studies in marketing. Most of the validation has been done by
Bootstrapping is nearly always more accurate than judgment [Camerer (1981)], though the gains are usually
small. This conclusion is true for both direct and indirect bootstrapping. One explanation for this is that the model
applies the judge’s rules in a more consistent manner.
Leigh, MacKay and Summers (1984) and Neslin (1981) compared direct and indirect bootstrapping of
consumer preferences. Minor differences were found in accuracy, the Leigh, MacKay and Summers’ study favoring
the direct approach, while Neslin’s favored the indirect. If the decision makers have little awareness of their process,
indirect bootstrapping is preferred. Typically, however, the choice between direct and indirect bootstrapping will be
based on relative costs.
One would expect that bootstrapping of consumers and experts would each provide useful information.
Research on the relative accuracy of each approach would be useful. The bootstrapping of marketing experts is
typically less expensive and faster.
In our opinion, bootstrapping (including related approaches such as expert systems and conjoint analysis) is
one of the more important advances for forecasting in marketing over the past quarter century. In comparison to
unaided judgments, it allows for systematic, inexpensive, rapid, and accurate forecasts of alternative marketing
strategies. It also enables the decision makers to examine simultaneously the effects of different strategies (e.g.,
price, product design, sales efforts, and advertising) on market share.
Econometric methods: When one has access to historical data, econometric models are relevant for predicting
market share. This has been an area where substantial efforts have been made in marketing. This research has
produced generalizations about the effectiveness and relative importance of price, advertising and other elements of
the marketing mix [Aaker and Carman (1982), Leone and Schultz (1980) and Simon and Arndt (1980)]. The
research has also led to the development of methods to analyze competitive response [Clarke (1973) and Hanssens
(1980)]. However, little research has been done on predictive validity. Brodie and de Kluyver (1987) found only
four studies that evaluated the ex ante predictive ability of econometric models. While the econometric models
showed no overall superiority, it must be remembered that these studies examined only short-range forecasts.
We believe that the research on econometric models in marketing has focused on areas that are unlikely to
show returns. For example, much attention has been given to developing theoretically pleasing functional forms
such as the logit and probit [e.g., see Naert and Weverbergh (1985)]. Experience in fields other than marketing has
shown that the functional form is seldom important to the accuracy of forecasts. This finding has also been obtained
in marketing studies [e.g., Shocker and Srinivasan (1979), Reibstein and Traver (1982), Brodie and de Kluyver
(1984), Dalrymple and Haines (1970), and Ghosh, Neslin and Shoemaker (1984)], although Wilton and Pessemier
(1981) found that the functional form did affect accuracy in predictions of the purchase of an electric automobile.
The functional form is most likely to be important when large changes occur in the dependent variable.
The lackluster performance of econometric methods may result from their being used in situations involving
small changes. Econometric methods are most relevant for assessing the impact of large changes in strategy.
Theoretical support for this position is provided by Hagerty (1987) and empirical evidence is provided by Parker and
Srinivasan (1976). For example, site selection models should lend themselves to econometric modeling because of
the large differences in market response to the potential sites.
The preceding discussion of market share has emphasized the analysis of data. Significant gains have also
been made in the availability and accuracy of data [e.g., Neslin (1974)]. Data are becoming more timely, more
detailed, and more accurate. Better data on the variable to be forecasted are of particular importance. Interestingly,
however, “more” and “better” information (beyond a certain minimal level) on causal variables has typically
produced only modest improvements in forecast accuracy. This result pertains primarily to judgmental methods,
though similar findings have been obtained with econometric and extrapolation methods [Armstrong (1985)].
Other areas where research would be relevant for market share forecasting relate to more specific marketing
decision areas. Some key research questions are:
(1) How can one select the best salespeople? Research from personnel psychology has produced promising
results recently. Of particular interest is the use of “job samples.” Prospective applicants are trained on
key elements of the job for an hour or so, and then are judged by how quickly they master these elements
[Reilly and Chao (1982)].
(2) How should salespeople be assigned to regions? Forecasting of regional sales can help with these
decisions. As early as 1939, Wellman (1939) and Weld (1939) suggested the use of econometric models
to forecast regional sales. Armstrong (1970) found supporting evidence, but little research has been done
in recent years to validate this approach. However, the data have improved immensely in this area.
(3) How should one forecast sales when the salesperson’s compensation is based on performance relative to
the sales forecast?
(4) In the area of distribution, what is the best way to make forecasts for site selection? Although firms obtain
extensive data, they often process these data in traditional group meetings in order to forecast the
profitability of a given site. Substantial investments are made on the basis of these forecasts, so
improvements in the accuracy of forecasting would be valuable. One would expect such techniques as
judgmental decomposition, Delphi and consensus-seeking to be important here. Quantitative methods are
also relevant. For example, Armstrong and Andress (1970) found that segmentation (using AID) provided
more accurate forecasts then did regression analysis for the volume of gasoline sold at service stations.
The state of the art in site selection is reviewed in Craig, Ghosh and McLafferty (1984).
(5) In the area of pricing, questions might be addressed such as how would a specified negotiation strategy
affect the final price?
If one approaches the forecasting problem as outlined in exhibit 1, it is necessary only to multiply the industry
sales and the market share forecasts to calculate a company sales forecast. Often, however, attempts are made to
forecast sales directly. An interesting research question, then, is under what conditions is it useful to go through all
of the steps of exhibit 1 instead of forecasting sales directly? [See Naert and Leeflang, (1978), Chapter 8), for a
discussion of this issue.] Dalrymple’s (1987) survey shows that this direct approach to sales forecasting is widely
used. The direct approach seems most appropriate for short-range sales forecasting in situations where one is not
concerned about assessing the effects of alternative strategies.
Opinions studies are widely used in sales forecasting. In particular, opinions are often obtained from the sales
force [Wotruba and Thurlow (1976)]. It is important to learn more about how to pose the questions, how to
aggregate the responses, and how then to adjust for biases in the responses. Staelin and Turner (1972) demonstrated
how bias can be isolated from other errors. In their analysis of forecasts from a sales force, the bias was responsible
for up to 60% of the forecast error.
Some improvements in accuracy might be obtained by using bootstrapping of the experts who make sales
forecasts. This was done successfully for catalog sales forecasting by Michael (1971). Intentions surveys are
commonly used for short-range forecasting. The validation research supports this practice. Continuous
improvements have been made on the methodology for intentions surveys over the past half century. Perry (1979)
showed how intentions surveys have become more accurate in forecasting political elections in the U.S. Much has
been learned about how to reduce sampling error, non-response bias, and response bias. Current work is directed
toward how to word the question properly in order to reduce response bias. Kalton and Schuman (1982) review the
research in this area. Warshaw (1980) shows how to improve measures of purchase intentions. Kalwani and Silk
(1982) provide evidence of bias in purchase intentions. They address the bothersome question of how intentions are
related to behavior (e.g., what percentage of those who report a 60% probability of purchase actually purchase in the
given time period).
New product forecasting is a particularly important area for marketers, especially in view of the large
investments and the likelihood of large forecasting errors. Tull (1967), in a survey of new product introductions by
companies, found a median absolute error of 26% and a median optimistic bias of about 22%. Conjoint analysis and
expert systems are useful here, as are consumer intentions surveys. However, expert surveys are probably the most
commonly used method for new product forecasting. They are subject to numerous biases, especially for situations
where, as in new product sales, the growth rate is expected to be exponential [Wagenaar and Timmers (1979)].
Diffusion models, based on estimating trial and repeat functions, have become widespread in commercial use,
particularly in the packaged goods industry. ASSESSOR, BASES, NEWS and LITMUS are four such models.
These models are used both for pre-test-market and later for after-test-market to estimate ultimate market share.
Over 1500 pre-post validation assessments have been performed with these models and the results appear to be
impressive, though their validity may be questioned because the assessments were performed by the organizations
that are selling the services. Also, little is known about how the models compare with one another using comparable
test situations, or how well they compare with judgmental forecasts by management. Wind, Mahajan, and Cardozo
(1981), Urban and Katz (1983), Meade (1984), Assmus (1984) and Shocker and Hall (1986) reviewed the state of
the art. After doing so, they lamented the lack of systematic and unbiased forecast validation studies.
Some recent validation work has been done on diffusion models. Rao (1987) tested the forecasting ability of
six diffusion models as compared with five extrapolation models for sales of dishwashers, room air conditioners,
clothes driers (1949-1961), and color TVs (1963-1970). Forecasts for each of the models were prepared for one-,
two- and three-years ahead. Surprisingly, the diffusion models were less accurate as a group than the extrapolation
models. For example, over all product groups and all forecast horizons, a simple linear trend had a MAPS of 18.5.
Contrast this with the diffusion models where the MAPS ranged from 42.9 to 81.0. Unfortunately, the forecasts
were all made from a single point in time, 1961.
A substantial amount of research on extrapolation methods, beginning with Winters (1960), leads to the
conclusion that, beyond a modest level, additional sophistication is detrimental - it does not improve accuracy, but it
does increase costs and reduce understanding. For evidence on this, see Makridakis et al. (1982), Armstrong (1985),
and additional evidence from Schnaars (1984) for manufacturers’ sales, and Geurts and Kelly (1986) for retail sales.
These results were contrary to expectations. For example, Dancer and Gray (1977) found no gain in accuracy when
using adaptive parameters to extrapolate sales of liquor products. There are three important areas, however, where
added complexity improves accuracy. First, seasonal adjustments are important. Second, it helps to use a trend and
to dampen it [Gardner and McKenzie (1985)]. Third, combine forecasts from different extrapolation methods
[Makridakis and Winkler (1983)].
It is difficult to imagine a product that does not have a seasonal factor. As expected, then, seasonal factors have
proven to be of great value in marketing forecasting. Dalrymple’s (1987) survey results are consistent with the
hypothesis that the use of seasonal factors improves forecasts: the mean reported forecast error for those companies
using seasonal factors was 8.4% vs. 11.7% for those not using seasonal factors. Perhaps seasonal factors should also
be dampened, but no direct tests have yet been made.
Interestingly, some controversy exists as to whether mechanical extrapolations will do better than judgmental
extrapolations. A study by Lawrence et al. (1985) concluded in favor of judgmental or “eyeball” extrapolations, but
Carbone and Gorr (1985) and Mabert (1976) concluded the opposite. Of course, mechanical extrapolation methods
are less expensive when a large number of forecasts must be made, such as for inventory control.
Another issue is whether to combine subjective and quantitative approaches to extrapolation, and, if so, how?
Moriarty and Adams (1984) discuss the use of such combinations. Mathews and Diamantopoulos (1986), in a study
of short-term quantitative sales forecasts for 281 products, found that subjective revisions led to improved accuracy;
however, these results were based on only one company and one starting point, and the results were mixed.
Conversely, Adams (1986) found substantial reductions in error when quantitative adjustments were made to
judgmental forecasts; these reductions averaged 30% for three of the five series he examined.
The combination of forecasts from alternative methods is useful. Baker et al. (1980), in a study of the effects of
offshore nuclear plants on visits to recreational beaches, showed how alternative methods can yield forecasts that
differ substantially from one another. Sewall (1981) provided evidence that a combination of expert opinion and
consumer intentions was more accurate than either one alone in forecasting mail order sales of women’s clothing.
Lawrence et al. (1986) concluded that combined judgmental and statistical forecasts are more accurate than either
Finally, Schnaars (1986) has begun a promising line of research by examining various rules to select which
extrapolation model is most appropriate for sales forecasting. Rules that involve dampening the trend and combining
alternative forecasts led to substantial improvement in accuracy in comparison with the rule “pick the model that
provides the best fit to the historical sales.”
The forecasted level of marketing costs can affect the marketing plan. For example, costs may be so high
as to render a plan unprofitable.
Costs depend upon the environment, the expected level of sales, and the actions taken by the company.
How can these costs be forecasted? A logical starting point is the use of extrapolation. Typically, unit costs
decrease, but at a decreasing rate. Thus, a learning curve is often appropriate. The learning curve, a concept
that originated in educational psychology was adopted by industrial engineering in the early 1900s. It is in
widespread use today. Its use in marketing dates to the early 1970s. In marketing it is referred to as the
experience curve and has been extended to incorporate fixed costs [Day and Montgomery (1983)]. Despite the
popularity of the concept, its direct application as a forecasting tool is limited to industries characterized by
high growth, high levels of value added, continuous-process manufacturing, and capital-intensive industries
[Aaker (1984, p. 184)]. Under other circumstances the concept needs to be used with caution, as factors other
than experience may have a more important impact on costs.
Large changes in costs will typically be forecasted by judgment, such as engineering estimates, or by the
use of econometric models. For example, costs for servicing different types of accounts or different
geographical regions lend themselves well to econometric forecasting. Given the availability of relevant
historical data, econometric models are especially relevant for very large changes in costs, such as those
created by strikes, government edicts, or shortages.
Results: Impact on Stakeholders
Forecasts from each area in exhibit 1 can be used to examine how each of the stakeholders will be
affected. In fact, it may be useful to start the forecasting process with the analysis of stakeholders because this
indicates how the forecasts are to be used. For example, we might want to forecast whether a proposed plan
will benefit the consumers or the possible impact it might have on the local community.
3. Assessing Uncertainty
Estimates of uncertainty of the forecasts are needed in areas such as finding new retail sites, deciding
how much inventory to stock at wholesale and retail levels, or deciding whether a new product venture is too
risky. Good estimates of uncertainty, by providing better information on which to base inventories, can help to
reduce overstocks as well as stock-outs. The reduction in stock-outs, in turn, yields better data about demand;
this can help to produce better forecasts in the future.
With good estimates of forecast distributions, decision-makers facing non-symmetric loss functions can
improve (even optimize) their actions in light of the expressed uncertainty [McIntyre (1981a)]. For example, if
a manufacturer suffers $1 of expense per unit if he overproduces vs. $2 of expense (or “regret”) if he
underproduces actual demand, then he should produce more than expected demand. But how much more? The
forecast distribution information allows an optimal answer to be calculated for this question.
Much work has been done on judgmental estimates of uncertainty. One of the key findings is that judges
are typically overconfident. Fischoff and MacGregor (1982) found that 95% confidence ranges that are
estimated judgmentally typically fail to include the true value. This bias occurs even when subjects are warned
in advance about the overconfidence phenomenon. Nevertheless, judgmental expressions of uncertainty have
been found to be useful.
One way to assess uncertainty has been to examine the agreement among judgmental forecasts. For
example, Ashton (1985), in a study of forecasts of annual advertising sales for Time magazine, found that the
agreement among the individual judgmental forecasts was a useful proxy for accuracy.
Williams and Goodman (1971) provided one of the earliest studies on uncertainty with extrapolation
methods. In their study of short-range telephone forecasts, the statistically estimated 95% confidence intervals
contained the true value for 90% of the forecasts. This represents rather good calibration. But, typically,
measures of fit provide a poor guide to uncertainty in the forecast. Makridakis et al. (1987) show that
statistically estimated confidence intervals are misleading even when based on the one-period-ahead forecast.
These authors present tables demonstrating the extent of this overconfidence for different forecasting methods
and different forecasting horizons.
Probably the best way to assess uncertainty is to follow the track record of a given forecasting method in
actual use. The survey by Dalrymple (1987) found that 54% of the 134 surveyed firms maintained records of
forecasting accuracy on a cumulative basis. In our survey of members of the IIF, 62% of the 21 marketing
practitioners said that their organizations maintained a track record. Lacking such a record, one should try to
simulate the forecasting situation. These approaches should overcome the problems noted in Makridakis et al.
One way to assess uncertainty in a forecast is to try to develop a model of the uncertainty. This was tried
by More and Little (1980) in assessing uncertainty for new products. They examined such factors as buyer risk
and distribution difficulty.
Even if we could provide good assessments of uncertainty, it is not clear that they would be accepted or
used by the decision makers. Rush and Page (1979) found a decreasing use of measures of uncertainty for
metals forecasts from 22% during the period 1910-1939 to only 8% during 1940-1964. Tull’s (1967) survey
noted that only 4 of 16 (i.e., 25%) respondent companies provided confidence intervals with their forecasts. In
our 1986 survey of members of the IIF, 24% of the 21 marketing practitioners claimed that their firms did not
provide confidence intervals when forecasts were presented, while 19% said that they “usually did so.”
Dalrymple (1987) found that 48% did not use confidence intervals, and only 10% “usually” used them. In
summary, confidence intervals are not widely used.
It seems worthwhile to build confidence measures into the forecasting packages used by marketers.
Moriarty (1985) suggests operational procedures for incorporating information on mean error and bias from
4. The Use of Forecasts\
Surveys provide information about which forecasting methods are used by practitioners [Dalrymple
(1987)] and the criteria used to evaluate them [Mentzer and Cox (1984, p. 33-34)]. However, it is also
important to consider how the methods and criteria impact marketing decisions.
Decision-makers often ignore forecasts that are surprising or unpleasant. Griffith and Wellman (1979), in
a study of forecasts for hospitals beds in six Michigan hospitals, found that the forecasts were ignored if they
conflicted with management’s prior beliefs. Decision makers may be subject to various biases that lead them to
revise the forecasts. Tyebjee (1987), for example, discusses the advocacy and illusion of control biases in new
Perhaps managers are reluctant to accept forecasts because they do not understand the methods used to
obtain them. In our survey of the members of the IIF, 37% of the 19 marketing practitioners who responded to
this item said that the users of forecasts in their organization seldom understood the forecasting methods that
were used. Of the other respondents, 37% said the users “often” understood, and 26% said that they “almost
The problem cannot be overcome simply by better communication of the forecasts. For example,
deSanctis (1984) reviewed a substantial number of empirical studies and was unable to find evidence that
computer graphics led to better decision-making. Wagenaar et al. (1985) found that elaborate TV weather
forecasts did not communicate more effectively than did the much simpler radio reports.
If managers do not ignore forecasts, they may revise them. Knowing this, the preparers may also bias
their forecasts. Again, referring to our survey of members of the IIF, 44% of the 21 marketing practitioners
who responded said that the forecasts were subject to “politics.” In other words, the forecast is often
manipulated for such reasons as motivation or payment.
Some useful research has been done on how to gain acceptance of forecasts. Of particular interest is the
research on scenarios. These studies, almost all by psychologists, yield suggestions on how to make scenarios
seem more likely and thus, to gain attention and acceptance. For example, you could link events with plausible
causal explanations [Tversky and Kahneman (1983)], or you could ask the decision-makers to project
themselves into the situation and to predict how they would act [Sherman (1980), Anderson (1983)].
Applications of these findings to marketing situations would be helpful. Schnaars and Topol (1987) test some of
the propositions about scenarios for short-range sales forecasts; interestingly, scenarios that do not make
unexpected outcomes seem less surprising in their study.
Another area of concern is whether the decision maker can use the forecasts properly, given that the
forecast is accepted as valid. Prior research has shown that subjects have difficulties in revising their estimate
when they receive new information [Lyon and Slovic (1976)]. Thus, decision makers may not revise their
forecasts properly when a quantitative forecast differs from their initial forecast.
Computer software may have a significant impact on the use of forecasting. For example, might decision
makers have more confidence in a forecast that comes from a computer? Or, might there be a blind reliance on
complex forecasting models?
Significant gains have been made in forecasting for marketing in the past quarter century. Advances have
occurred in the development of qualitative methods such as Delphi, role playing, intentions and opinions
surveys, and bootstrapping. They have also occurred for quantitative methods such as extrapolation and
econometrics. The challenge now is to build up experience in applying these methods so generalizations can be
made about which methods are most appropriate in the different areas where forecasts are needed in marketing.
The evidence and research needs discussed in this paper are now summarized.
Environment and Market Forecasting
For these areas there is empirical evidence to support the use of econometric methods rather than
subjective methods for long-range forecasts. There is also some evidence to show that the accuracy of forecasts
of environmental variables is not as important as is identifying the key variables to include in the market
forecasting model. More research is required to investigate this latter issue.
For short-term forecasting and the estimation of the current size (status) of markets some empirical
evidence supports the combining of judgment with extrapolation and econometric methods. However, further
research is required to clarify under what conditions econometric models are useful and when and how to
combine these results with expert judgment. Special attention should be given to the increasing availability of
detailed data bases such as those derived from store scanner systems. Finally, for predicting prices, it is
important to be aware of the case of “efficient” markets. Here Adam Smith’s Rule for Forecasters holds; that
is, “Forecasters cannot beat the market.”
Actions by Company, Competitors, Suppliers, Distributors, and Community
There is little published research on forecasting in any of these areas, although suggestions can be
made about which methods to use. Research needs to be aimed at discovering which methods are
appropriate in which situations.
For forecasting company actions, the most appropriate methods are qualitative methods such as
projective tests, group depth interviews, work samples, and role playing.
For forecasting competitors’ actions, the methods suggested range from objective methods such as
extrapolation, experimentation, and game theory, to more subjective methods such as the use of expert
opinion, intentions surveys, and role playing. A similar set of candidate “methods” appears appropriate
for forecasting the actions of suppliers, distributors, and the various publics that might affect the
marketing activity of the organizations.
While considerable research effort has been directed toward modeling and analyzing the factors that
affect market share and sales, far less has been directed at examining the predictive validity and
forecasting accuracy of the models. The limited evidence that is available fails to show the superiority of
econometric methods over extrapolation for short-term market share forecasting. Further research is
necessary to examine conditions where econometric methods are likely to perform well. For example, it is
expected that these methods seem more relevant for forecasting the effects of large changes in the
marketing mix variables. When historical data are not available, or where they are expensive to obtain,
bootstrapping (including expert systems and conjoint analysis) appears to have considerable advantages
over unaided judgment. While these methods have been used extensively by academics and practitioners
in other areas of marketing, their performance in forecasting market share has yet to be evaluated.
For sales forecasting, there is considerable evidence to support the use of less sophisticated
extrapolation methods for short-term forecasting. However, controversy exists about whether
extrapolations are better than judgment or whether it is better to combine extrapolation with judgment. An
area that has received considerable recent attention has been forecasting new product sales using growth
curves. Here again there has been a lack of validation work.
This area has received little attention. Of importance is the question of when it is appropriate to use
extrapolation or judgmental methods. For discrete changes in costs, the research question is whether to
rely on current information and judgment or to also use econometric analyses of past costs.
Assessing Uncertainty and the Use of Forecasts
These areas have also received little attention in forecasting research and, more specifically,
Exhibit 3 summarizes directions for further research. In the areas of environmental, market, market share,
and sales forecasting, the need is for extensions of existing research. In the other areas, basic research is
Exhibit 3. Summary of research issues.
• Why aren’t conditional forecasts more accurate than unconditional ones in short-range
• What are acceptable levels of forecast errors for environmental inputs to market forecasting
• What is the best way to forecast the change in size and structure of markets?
• Under what conditions does expert judgment help to forecast change?
• How can biases in judgmental forecasting be corrected?
• Do marketing theories lead to better forecasts?
• What is the best way to assess uncertainty?
• Under what conditions do econometric models improve accuracy?
• What are the relative abilities of projective tests, group depth interviews, work samples, and
role playing to forecast behavior of key stakeholders in the organization?
• How to forecast resistance to a company’s actions?
Competitor’s Actions • What is the best way to forecast competitors’ actions (e.g., expert opinion, game theory,
analogies, or role playing)?
Actions by Suppliers,
• What is the best way to forecast the actions of these groups? (e.g., can role playing forecast
conflicts among channel members?)
Market Share • How can judgmental biases be controlled or assessed?
• To what extent do improved data lead to more accurate forecasts?
• Under what conditions is bootstrapping (expert systems) most useful?
• What is the most effective way to use bootstrapping?
• Under what conditions are econometric models useful?
• How can improved forecasts lead to better decisions? (e.g., How can one forecast success
of sales personnel? What is the best way to make forecasts for site selection?)
• What is the best way to forecast market share and sales by region?
Sales Forecasting • How can biases in judgmental forecasts by sales and marketing personnel be reduced?
• How can bias in consumer intentions surveys be reduced?
• When and how should one dampen the trend factor?
• When and how should one dampen the seasonal factor?
• How should one combine judgment and extrapolation?
Costs • Do econometric models provide improvement over judgment?
• Do econometric models help to predict large changes?
• When is it appropriate to use experience curves?
• Should “safety factors” be used to adjust for overconfidence?
• What methods are most effective in estimating confidence intervals in forecasting?
• What guidelines will help to effectively use forecasts in decision-making?
• How can scenarios be used to gain commitment?
• Can computer based expert systems help managers to use forecasts more
Underlying the research topics outlined in exhibit 3 is the need for objective validation research. Advocacy
research is expected to be inefficient; researchers tend to find what they hope to find. While the method of multiple
hypotheses is preferred, it makes a difference how we pose the research questions. Rather than trying to disconfirm
hypotheses, it seems easier for researchers to state the problem in terms of which hypothesis (or method) in a set of
reasonable hypotheses is most useful; that is, try to find supporting evidence for each hypothesis [Tweeny, Doherty
and Mynatt (1982)]. Another useful procedure is to ask under what conditions a hypothesis (or method) is most
Clearly, the quantity of research has been growing. As a crude measure, consider the references cited in this
paper. Only two citations were for work done prior to 1960. The average number of papers cited per year for the
1960, 1970s, and 1980s was 0.6, 3.7 and 10.8, respectively. In our opinion, the quality of research on forecasting in
marketing has also been improving.
Aaker, D.A., 1984, Strategic market management (Wiley, New York).
Aaker, D.A. and J. M. Carman, 1982, Are you over advertising? Journal of Advertising Research 22, 57-70.
Adams, Arthur J., 1986, Procedures for revising management judgments forecasts, Academic of Marketing Science
Agnew, Carson E., 1985, Bayesian consensus forecasts of macroeconomic variables, Journal of Forecasting 4, 363-
Anderson, Craig A., 1983, Imagination and expectation: The effect of imagining behavioral scripts on personal
intentions, Journal of Personality and Social Psychology 45, 293-305.
Armstrong, J. Scott, 1970, An application of econometric models to international marketing, Journal of Marketing
Research 7, 190-198.
Armstrong, J. Scott, 1985, Long-range forecasting: From crystal ball to computer (2nd ed.) (John Wiley, New
Armstrong, J. Scott, 1987, Forecasting methods for conflict situations, in: George Wright, ed., Judgmental
forecasting (John Wiley, Chichester) 157-176.
Armstrong, J. Scott and J. Andress, 1970, Exploratory analysis of marketing data: Trees vs. regression, Journal of
Marketing Research 7, 487-492.
Armstrong, J. Scott, W. B. Denniston and M. M. Gordon, 1975, The use of the decomposition principle in making
judgments, Organizational Behavior and Human Performance 14, 257-263.
Ascher, William, 1978, Forecasting: An appraisal for policy makers and planners (Johns Hopkins University Press,
Ash, J. C. K. and D. J. Smyth, 1973, Forecasting the United Kingdom economy. (Saxon House, Farborough).
Ashton, Alison H., 1985, Does consensus imply accuracy in accounting studies of decision making? Accounting
Review 60, 173-185.
Assmus, Geri, 1984, New product forecasting, Journal of Forecasting 3, 121-138.
Baker, Earl J., S. G. West, D. J. Moss and J. M. Weyant, 1980, Impact of offshore nuclear power plants: Forecasting
visits to nearby beaches, Environment and Behavior 12, 367-407.
Bell, D. E., R. L. Keeney and J. D. C. Little, 1975, A market share theorem, Journal of Marketing Research 12, 136-
Best, Roger J., 1974, an experiment in Delphi estimation in marketing decision making, Journal of Marketing
Research 11, 448-452.
Brandt, Jon A. and D. A. Bessler, 1983, Price forecasting and evaluation: An application in agriculture, Journal of
Forecasting 2, 237-248.
Brodie, Roderick J. and C. A. de Kluyver, 1984, Attraction versus linear and multiplicative market share models,
Journal of Marketing Research 21, 196-201.
Brodie, Roderick J. and C. A. de Kluyver, 1987, A comparison of the short-term forecasting accuracy of
econometric and naive extrapolation models of market share, International Journal of Forecasting 3, 423-437.
Camerer, Colin, 1981, General conditions for the success of bootstrapping models, Organizational Behavior and
Human Performance 27, 411-422.
Carbone, Robert and W. Gorr, 1985, Accuracy of judgmental forecasting of time series, Decision Sciences 16,
Cattin, Philippe and D. R. Wittink, 1982. Commercial use of conjoint analysis: A survey, Journal of Marketing
Cerullo, Michael, J. and A. Avila, 1975, Sales forecasting practices: A survey, Managerial Planning 24, 33-39.
Chakravarti, D., A. Mitchell and R. Staetin, 1979, Judgment-based marketing decision models: An experimental
investigation of the decision calculus approach, Management Science 25, 251-263.
Chakravarti, D., A. Mitchell and R. Staelin, 1980, A procedure for parameterizing decision calculus models of
dynamic market response, in: R. Leone, ed., Proceedings of the Second ORSA/TIMS Special Interest
Conference on Market Measurement and Analysis (Institute of Management Sciences, Providence, RI) 135-
Clarke, D. G., 1973, Sales advertising cross-elasticities and advertising competition, Journal of Marketing Research
Craig, C. Samuel, A. Ghosh and S. McLafferty, 1984, Models of the retail location process: A review, Journal of
Retailing 60, 5-36.
Curtin, Richard T., 1982, Indicators of consumer behavior: The University of Michigan surveys of consumers,
Public Opinion Quarterly 46, 340-352.
Dalrymple, Douglas J., 1975, Sales forecasting: Methods and accuracy, Business Horizons 18, 69-73.
Dalrymple, Douglas J., 1987, Sales forecasting practices: Results from a U.S. survey, International Journal of
Forecasting 3, 379-391.
Dalrymple, Douglas J. and G. H. Haines, Jr., 1970, A study of the predictive ability of market period demand-supply
relations for a firm selling fashion products, Applied Economics 1, 277-285.
Dancer, Robert and C. Gray, 1977, An empirical evaluation of constant and adaptive computer forecasting models
for inventory control, Decision Sciences 8, 228-238.
Day, George S. and D. B. Montgomery, 1983, Diagnosing the experience curve, Journal of Marketing 47, 44-58.
Denton, Frank T. and J. Kuiper, 1965, The effect of measurement errors on parameter estimates and forecasts: A
case study based on the Canadian preliminary national accounts, Review of Economics and Statistics 47, 198-
Denton, Frank T. and E. H. Oksanen, 1972, A multi-country analysis of the effects of data revisions on an economic
model, Journal of the American Statistical Association 67, 286-291.
deSanctis, Geraldine, 1984, Computer graphics as decision aids: Directions for research, Decision Sciences 15, 463-
Dhrymes, Phoebus, and S.C. Peristiani, 1987, A comparison of the forecasting performance of WEFA and ARIMA
time series models, International Journal of Forecasting (forthcoming).
Edelman, F., 1965, Art and science of competitive bidding, Harvard Business Review 43 (4), July-August, 53-66.
Eliashberg, J., S. A. Latour and A. Rangaswamy, 1986, Assessing the predictive accuracy of two utility-based
theories in a marketing channel negotiation context, Journal of Marketing Research 23, 101-110.
Fildes, Robert, 1985, Quantitative forecasting – the state of the art: Econometric models, Journal of the Operational
Research Society 36, 549-580.
Fischoff, Baruch and D. MacGregor, 1982, Subjective confidence in forecasts, Journal of Forecasting 1, 155-172.
Fudge, W. D. and L. M. Lodish, 1977, Evaluation of the effectiveness of a salesman's planning system by field
experimentation, Interfaces 8, 97-106.
Gardner, Everette S., Jr. and E. McKenzie, 1985, Forecasting trends in time series, Management Science 31, 1237-
Geurts, Michael D. and J. P. Kelly, 1986, Forecasting retail sales using alternative models, International Journal of
Forecasting 2, 261-272.
Ghosh, Avijit, S. Neslin and R. Shoemaker, 1984, A comparison of market share models and estimation procedures,
Journal of Marketing Research 21, 202-210.
Griffith, John R. and B. T. Wellman, 1979, Forecasting bed needs and recommending facilities plans for community
hospitals: A review of past performance, Medical Care 17, 292-303.
Hagerty, Michael R., 1987, Conditions under which econometric models will outperform naive models,
International Journal of Forecasting 3, 457-460.
Hanssens, D. M., 1980, Market response, competitive behavior, and time series analysis, Journal of Marketing
Research 17, 470-485.
Hogarth, Robin M., 1978, A note on aggregating opinions, Organizational Behavior and Human Performance 21,
Hogarth, Robin M. and S. Makridakis, 1981, Forecasting and planning: An evaluation, Management Science 27,
Huber, Joel, 1975a, Bootstrapping of data and decisions, Journal of Consumer Research 2, 229-234.
Huber, Joel, 1975b, Predicting preferences on experimental bundles of attributes: A comparison of models, Journal
of Marketing Research 12, 290-299.
Jobber, D., G. Hooley and S. Sanderson, 1985, Marketing in a hostile environment: The British textile industry,
Industrial Marketing Management 14, 35-41.
Kahneman, Daniel, P. Slovic and A. Tversky, 1982, Judgment under Uncertainty: Heuristics and Biases.
Cambridge: Cambridge University Press.
Kalton, Graham and H. Schuman, 1982. The effect of the question on survey responses: A review, Journal of the
Royal Statistical Society: Series A 145, Pt. 1, 42-73.
Kalwani, Manohar and A. J. Silk, 1982, On the reliability and predictive validity of purchase intention measures,
Marketing Science 1, 243-286.
Kotler, Philip, 1984, Marketing Management. Englewood Cliffs, NJ: Prentice-Hall.
Larreche, Jean-Claude and R. Moinpour, 1983, Managerial judgment in marketing: The concept of expertise,
Journal of Marketing Research 20, 110-121.
Lawrence, Michael J., R. H. Edmundson and M. J. O'Conner, 1985, An examination of the accuracy of judgmental
extrapolation of time series, International Journal of Forecasting 1, 25-35.
Lawrence, Michael J., R. H. Edmundson and M. J. O'Conner, 1986, The accuracy of combining judgmental and
statistical forecasts, Management Science 32, 1521-1532.
Leigh, Thomas W., D. B. MacKay and J. O. Summers, 1984, Reliability and validity of conjoint analysis and self-
explicated weights: A comparison, Journal of Marketing Research 21, 456-462.
Leone, Robert P., 1987, Forecasting the effect of an environmental change on market performance: An intervention
time-series approach, International Journal of Forecasting 3, 463-478.
Leone, R. P. and R. L. Schultz, 1980, A study of marketing generalizations, Journal of Marketing 44, 10-18.
Libby, Robert and R. K. Blashfield, 1978, Performance of a composite as a function of the number of judges,
Organizational Behavior and Human Performance 21, 121-129.
Linden, Fabian, 1982, The consumer as forecaster, Public Opinion Quarterly 46, 88-95.
Little, J. D. C., 1970, Models and managers: The concept of a decision calculus, Management Science: Applications
Lodish, L. M., 1971, CALLPLAN: An interactive salesman's call planning system, Management Science 18, 25-40.
Lyon, Don and P. Slovic, 1976, Dominance of accuracy information and neglect of base rates in probability
estimation, Acta Psychologica 40, 287-298.
Mabert, Vincent A., 1976, Statistical versus sales force-executive opinion short range forecasts, Decision Sciences 7,
McDonald, John, 1975, An analysis of the significance of revisions to some quarterly U.K. national income time
series, Journal of the Royal Statistical Society: Series A 138, 242-256.
McIntyre, Shelby H., 1981a, Reducing over-reaction in contact markets, Industrial Marketing Management 10, 283-
McIntyre, Shelby H., 19816, An experimental study of the impact of judgment-based models, Management Science
Mahajan, Vijay and Y. Wind, 1986, Innovation Diffusion Models of New Product Acceptance. Cambridge, MA:
Mahajan, Vijay, E. Muller and S. Sharma, 1984, An empirical comparison of awareness forecasting models for new
product introduction, Marketing Science 3, 179-206.
Makridakis, Spyros and S. Wheelwright, 1977, Forecasting: Issues and challenges for marketing management,
Journal of Marketing 41 (October) 248b138.
Makridakis, Spyros and R. L. Winkler, 1983, Average of forecasts: Some empirical results, Management Science
Makridakis, Spyros, M. Hibon, E. Lusk and M. Belhadjali, 1987, Confidence intervals: An empirical investigation
of the series in the M-Competition, International Journal of Forecasting 3, 489-508.
Makridakis, Spyros et al., 1982, The accuracy of extrapolation (time series) methods: Results of a forecasting
competition, Journal of Forecasting 1, 111-153.
Mathews, Brian P. and A. Diamantopoulos, 1986, Managerial intervention in forecasting: An empirical investigation
of forecast manipulation, International Journal of Research in Marketing 3, 3-10.
Meade, Nigel, 1984, The use of growth curves in forecasting market development – A review and appraisal, Journal
of Forecasting 3, 429-451.
Mentzer, John T. and J. E. Cox, Jr., 1984, Familiarity, application, and performance of sales forecasting techniques,
Journal of Forecasting 3, 27-36.
Michael, George C., 1971, A computer simulation model for forecasting catalog sales, Journal of Marketing
Research 8, 224-229.
Montgomery, David B., 1975, New product distribution: An analysis of supermarket buyer decisions, Journal of
Marketing Research 12, 255-264.
More, Roger A. and B. Little, 1980, The application of discriminant analysis to the prediction of sales forecast
uncertainty in new product situations, Journal of the Operations Research Society 31, 71-77.
Moriarty, Mark M., 1985, Design features of forecasting systems involving management judgments, Journal of
Marketing Research 22, 353-364.
Moriarty, Mark M. and A. J. Adams, 1984, Management judgment forecasts, composite forecasting models, and
conditional efficiency, Journal of Marketing Research 21, 239-250.
Naert, Philippe A. and P. Leeflang, 1978, Building Implementable Marketing Models. Boston, MA: Martinus
Naert, Philippe A. and M. Weverbergh, 1985, Market share specification, estimation, and validation: Toward
reconciling seemingly divergent views, Journal of Marketing Research 22, 653-661.
Neslin, Scott A., 1981, Linking product features to perceptions: Self-stated versus statistically revealed importance
weights, Journal of Marketing Research 18, 80-86.
Nevin, John R., 1974, Laboratory experiments for estimating consumer demand: A validation study, Journal of
Marketing Research 11, 261-268.
Parker, Barnett R. and V. Srinivasan, 1976, A consumer preference approach to the planning of rural primary
health-care facilities, Operations Research 24, 991-1025.
Pashigian, B. Peter, 1964, The accuracy of Commerce - S.E.C. sales anticipations, Review of Economics and
Statistics 46, 398-405.
Payne, Donald E., 1966, Jet set, pseudo-store and new product testing, Journal of Marketing Research 3, 372-
Perry, Paul, 1979, Certain problems in election survey methodology, Public Opinion Quarterly 43, 312-325.
Rao, Sanjay-Kumar, 1985, An empirical comparison of sales forecasting models, Journal of Product
Innovation Management 4, 232-242.
Reibstein, David J. and P. A. Traver, 1982, Factors affecting coupon redemption rates, Journal of Marketing
Reilly, Richard R. and G. T. Chao, 1982, Validity and fairness of some alternative employee selection
procedures, Personnel Psychology 35, 1-62.
Rink, David and J.E. Swan, 1979, Product life cycle research: A literature review, Journal of Business
Research 7, 219-242.
Rosenstone, Stephen J., 1983, Forecasting Presidential Elections New Haven, CT: Yale University Press.
Rush, Howard and W. Page, 1979, Long-term metals forecasting: The track record: 1910-1964, Futures 11,
Schnaars, Steven P., 1984, Situational factors affecting forecast accuracy, Journal of Marketing Research 21,
Schnaars, Steven P., 1986, An evaluation of rules for selecting an extrapolation model on yearly sales
forecasts, Interfaces 16 (Nov-Dec.), 100-107.
Schnaars, Steven P. and M. T. Topol, 1987, An empirical test of multiple scenarios, International Journal of
Forecasting 3, 405-419.
Sewall, Murphy A., 1981, Relative information contributions of consumer purchase intentions and
management judgment as explanators of sales, Journal of Marketing Research 18, 249-253.
Sherman, Steven J., 1980, On the self-erasing nature of errors of prediction, Journal of Personality and Social
Psychology 39, 211-221.
Shocker, Allan D. and W. G. Hall, 1986, Pretest market models: A critical evaluation, Journal of Product
Innovation Management 3, 86-107.
Shocker, Allan D. and V. Srinivasan, 1979, Multiattribute approaches for product evaluation and generation: A
critical review, Journal of Marketing Research 16, 159-180.
Simon, J. L. and J. Arndt, 1980, The shape of the advertising response function, Journal of Advertising
Research 20, 11-28.
Staelin, Richard and R. E. Turner, 1972, Error in judgmental sales forecasts: Theory and results, Journal of
Marketing Research 10, 10-16.
Tull, Donald S., 1967, The relationships of actual and predicted sales and profits in new-product introductions,
Journal of Business 40, 233-250.
Tversky, Amos and D. Kahneman, 1983, External versus intuitive reasoning: The conjunction fallacy in
probability judgment, Psychological Review 90, 293-315.
Tweney, Ryan D., M E. Doherty and C. R. Mynatt, 1982, Rationality and disconfirmation: Further evidence,
Social Studies of Science 12, 345-441.
Tyebjee, Tyzoon T., 1987, Behavioral biases in new product forecasting, International Journal of Forecasting
Urban, Glen L. and G. M. Katz, 1983, Pre-test-market models: Validation and managerial implications,
Journal of Marketing Research 20, 221-234.
Wagenaar, Willem A., R. Schreuder and A. H. C. van der Heijden, 1985, Do TV pictures help people to
remember the weather forecast? Ergonomics 28, 756-772.
Wagenaar, Willem A. and H. Timmers, 1979, The pond-and-duckweed problem: Three experiments in the
misperception of exponential growth, Acta Psychologica 43, 239-251.
Warshaw, Paul R., 1980, Predicting purchase and other behaviors from general and contextually specific
intentions, Journal of Marketing Research 17, 26-33.
Weld, L. D. H., 1939, The value of the multiple correlation method in determining sales potentials, Journal of
Marketing 3, 389-393.
Wellman, H. R., 1939, The distribution of selling effort among geographical areas, Journal of Marketing 3,
Williams, W. H. and M .L. Goodman, 1971, A simple method for the construction of empirical confidence
limits for economic forecasts, Journal of the American Statistical Association 66, 752-754.
Wilton, Peter C. and E. A. Pessemier, 1981, Forecasting the ultimate acceptance of an innovation: The effects
of information, Journal of Consumer Research 8, 162-171.
Wind, Yoram, V. Mahajan and R. N. Cardozo, eds., 1981. New Product Forecasting. Lexington, MA:
Winters, Peter R., 1960, Forecasting sales by exponentially weighted moving averages, Management Science
Wotruba, Thomas R. and M. L. Thurlow, 1976, Sales force participation in quota setting and sales forecasting,
Journal of Consumer Research 8, 162-171.
Yokum, J. Thomas and A. R. Wildt, 1987, Forecasting sales response for multiple time horizons and
temporally aggregated data: A comparison of constant and stochastic coefficient models, International
Journal of Forecasting 3, 479-488.