Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

International Journal of Forecasting 3 (1987) 379-391 379

North-Holland

SALES FORECASTING PRACTICES *


Results from a United States Survey

Douglas J. DALRYMPLE
Indiana Uniuersity, Bloomington, IN 47405, USA

Abstract. The paper presents the results of a survey designed to discover how business firms
prepare sales forecasts, what methods they prefer, and the accuracy of their predictions.
The survey showed that subjective, extrapolation and naive techniques are widely used by
American business firms in various forecasting situations. Also, some business firms are
reducing forecasting errors by making greater use of computers and seasonal adjust-
ments.

Accuracy, Business survey of forecasting methods, Computer forecasting.

1. Introduction

This study was designed to learn more about the use of sales forecasting techniques in business
organizations. It expands and updates an earlier survey [Dalrymple (1975)]. The present study also
examines trends in forecasting errors and the use of computers. The goal was to identify factors that
were related to forecast accuracy. Other objectives of the survey were to find out whether firms
prepare forecasts for alternative environments, combine forecasts, use confidence intervals, and keep
records of forecasting costs. The results of the study are compared with previous forecasting surveys
conducted by Mentzer and Cox (1984), Sparkes and McHugh (1984), Hanke (1984), Rush and Page
(1979), Rothe (1978), Wheelwright and Clarke (1976), Cerullo and Avila (1975), PoKemper and
Bailey (1970) and Sales Management (1967).

2. Methodology

Respondents for this study, sponsored by the Indiana University School of Business, were
obtained by mailing a 3 page questionnaire to executives in 860 U.S. companies in all 50 states. ’ The
cover letter, typed on the School’s letterhead, was addressed to the Marketing/Forecasting Manager
in order to encourage responses from people who were familiar with sales forecasting activities. The
questionnaires were sent by first class mail, and printed business reply return envelopes were
enclosed. Two mailings of the questionnaire were sent, a month apart, to remind people to return the
survey.

* The Gunning Fog Index for this Paper is about 13.


’ A copy of the questionnaire is available from the author.

0169-2070/87/$3.50 0 1987, Elsevier Science Publishers B.V. (North-Holland)


380 D.J. Dalrynple / S&s forecasting pructices

Only 134 usable questionnaires were returned, a 16% response rate. This was considerably below
the 35% return rate obtained in the 1975 study. Questionnaires were returned from more than 28
states. Although the response rate was low, confidential information about the identities of the
respondents suggested that the sample was representative. * Also, a wide cross-section of firms was
obtained, with the largest number of responses coming from manufacturing firms and fewer from
distribution, retailing, insurance, consumer services, and banking sectors. In addition, returns were
received from firms that varied widely in size. The typical firm that responded in 1983 was larger
than the typical respondent firm in the 1975 survey. More medium and large companies replied to
the 1983 survey. Average sales reported by the 1975 respondents were about $149 million a year
compared to about $300 million in the 1983 survey.

3. Results

The survey findings are presented in the order that an analyst might use them to prepare a
forecast. Analysts often start by setting some parameters to guide the forecasting task and then select
forecasting methods appropriate to the job.

3.1. Preparing the forecast

Strategic planning has become popular in American business as a guide to building profitable
product portfolios. For this approach to be effective, management needs to look at the long run
impact of alternative approaches and to be prepared with contingency plans. Thus, we expected that
respondent firms would be interested in preparing alternative forecasts for different strategies,
environments, and capabilities. However, the results of the survey indicated that companies provided
little support for this activity (table 1). Only 19% of the firms frequently prepared forecasts for
alternative strategies. Fewer than 10% viewed preparation of forecasting alternatives as a customary
procedure. These results are particularly significant since it is possible that firms most interested in
forecasting responded to the survey, while the nonrespondents may be less concerned with preparing
alternative forecasts. This may suggest that strategic planning has not greatly influenced sales
forecasting procedures.
The forecasting literature often suggests that analysts should present a range of forecasts to
management to show the boundaries of expected forecasting errors. However, this did not appear to
be a standard practice among the firms in the current survey (table 2). Seventy-seven percent of the
respondents said they either did not provide confidence intervals or did so only occasionally. Thus,

Table 1
Reported preparation of forecasts for alternative strategies, environments, and capabilities.

Forecasts for Not Prepared Prepared Usually No


alternative prepared occasionally frequently prepared answer

Strategies 39.1% 33.1% 18.8% 9.0% 0.0%


Environments 45.1 32.3 14.3 5.3 3.0
Capabilities 41.4 36.8 12.8 6.0 3.0

* The identities of a substantial portion of the sample were revealed to the author when respondents wrote and called to
obtain copies of the results.
D. J. Dalymple / Sales forecasiing pracrrces 381

Table 2
Use of upper and lower confidence intervals when presenting sales forecasts

Activity Percentage of firms


Intervals not used 48.1%
Intervals occasionally used 29.3
Intervals frequently used 11.3
Intervals usually used 9.8
No answer 1.5

standard procedure seems to be to provide point estimates for sales forecasts rather than confidence
intervals.
Another area of interest was the frequency of preparation of forecasts for market planning and
production scheduling (table 3). Sales forecasts for market planning are most often prepared monthly
or quarterly, with annual forecasts being made least frequently. Production forecasting is done more
often than sales forecasting; the largest number of firms prepare production forecasts monthly or
weekly.

3.2. Methods used

Respondent firms were asked to indicate their usage of 15 forecasting techniques (table 4).
Descriptions of each of the techniques were included in the questionnaire to help reduce respondent
classification errors. The two most popular forecasting techniques in the 1983 survey were sales force
composite and jury of executive opinion. These two subjective methods were the most popular in the
1975 study except that their positions were reversed. Intentions-to-buy surveys ranked third among
subjective methods and placed seventh overall (table 4). The proportion of firms using this technique
(16.4% in 1983 versus 15% in 1975) and the seventh place ranking were virtually the same as in the
earlier study. The results agree with findings reported by Mentzer and Cox (1984) and earlier studies
by Rush and Page (1979) Rothe (1978) Wheelwright and Clarke (1976) Cerullo and Avila (1975)
and PoKemper and Bailey (1970). The continued reliance on subjective techniques suggests that
many managers have yet to be convinced that other methods are better. The respondents’ preferences
for subjective approaches is supported by a study by Lawrence, Edmundson and O’Connor (1985)
that concluded that judgmental forecasting was as accurate as statistical techniques for a sample of
111 time series.
The most popular extrapolation forecasting method was the naive approach (table 4). Thirty
percent of the respondents to the survey questionnaire said they regularly used the naive method,
making it the third most common forecasting technique reported. This technique does not appear in
many surveys of forecasting procedures because most people do not believe it is accurate and do not
believe it is widely used. The current research study shows that not only do firms use the naive

Table 3
Frequency of sales forecast preparation

Activity Weekly Monthly Quarterly Semi- Annual l-5 No


annual Years answer
Market planning 4.5% 38.1% 29.9% 9.7% 16.4% 0.7% 0.7%
Production schedule 27.6 33.6 12.7 0.7 5.2 _ 20.1
382 D.J. Dalrymple / Sales forecastlngpructices

Table 4
Sales forecasting methods used by respondents.

Methods % Firms who use % Firms who use % Firms no longer used
regularly occasionally
Subjective

Sales force composite 44.8% 17.2% 13.4%


Jury of executive opinion 37.3 22.4 8.2
Intention to buy survey 16.4 10.4 18.7
Industry survey 14.9 20.9 17.9

Extrupoiatmn

Naive 30.6 20.1 9.0


Moving average 20.9 10.4 15.7
% Rate of change 19.4 13.4 14.2
Leading indicators 18.7 17.2 11.2
Unit rate of change 15.7 9.7 18.7
Exponential smoothing 11.2 11.9 19.4
Line extension 6.0 13.4 20.9

Quantitative

Multiple regression analysis 12.7 9.0 20.9


Econometric models 11.9 9.0 19.4
Simple regression analysis 6.0 13.2 20.1
Box-Jenkins 3.7 5.2 26.9

method, but that it lags only the two most traditional methods among the 134 responding firms.
Also, the naive method had a rejection percentage of only 9% (table 4) suggesting that managers
were satisfied with the technique. The acceptance of the naive method by managers is supported by a
number of research reports [Makridakis and Hibon (1979), Makridakis et al. (1982) McLaughlin
(1983) and Schnaars (1984)]. The evidence in these studies suggests that the naive method provides
results that are as accurate as those from more complicated methods.
The second most popular extrapolation technique in the survey was the moving average (table 4).
Twenty-one percent of the firms reported regular usage of the moving average in 1983 as compared
with 24% in 1975. The moving average has maintained its popularity over time and ranked fourth in
both surveys. Note that almost twice as many firms reported they used the moving average (20.9%) as
those that reported the use of exponential smoothing (11.2%). This result is rather startling because
empirical research has shown that the adaptive nature of the exponential smoothing method
generally produces more accurate results than the moving average technique. For example, Makridakis
et al. (1982) reported average 18 period ahead forecasting errors of 19.6% for moving average
forecasts compared to only 16.9% for simple exponential smoothing, based on 1001 time series. The
survey results showed a preference for the moving average technique; this preference may be based
on familiarity rather than on ability to perform. The use of exponential smoothing in this study
(11.2%) was similar to that reported by Rothe (1978) of 14%.
The trend projection method was divided into three separate procedures in the 1983 survey:
percentage rate of change, unit rate of change, and line extension. This diluted the percentages of
firms reporting usage, but it allowed comparisons of the relative importance of the three different
projective methods. The percentage rate of change method proved to be the most popular projective
technique; its regular use was reported by 19% of the firms (table 4). Unit rate of change was next,
D.J. Dalynple / Sales jorecasrqpractices 383

with 16%, and the line extension method was mentioned by only 6% of the firms. The overall
popularity of the projective technique was revealed by an analysis showing that 25% of the firms
regularly used at least one of the three projective techniques. On this basis, the projective technique
would rank fourth in usage among all the procedures in the study.
Regression analysis proved to be the most popular quantitative forecasting procedure (table 4).
This technique was divided into simple time series regression and multiple regression for purposes of
comparison. The expectation was that simple regression analysis would be used by a larger number
of firms than the more complicated multiple regression analysis procedures. However, the reverse
proved to be true; multiple regression analysis was mentioned by 13% of the firms, whereas simple
regression was reported by only 6%. On a combined basis, the use of regression analysis exceeded the
use of exponential smoothing and approached that of the moving average. The popularity of multiple
regression analysis may be explained by the growing availability of computers and demonstrated by
preference for this method by large corporations. A survey by Rothe (1978) reported that 6% of the
respondents used regression analysis in forecasting.
Econometric models were used on a regular basis by about 12% of the responding firms, a figure
which exceeded that for the use of exponential smoothing (table 4). The preference for econometric
models was 4 percent higher than the 8% of firms that reported using simulation models in the 1975
survey. These results are similar to those of Mentzer and Cox (1984) who found that 9% to 10% of
the firms responding to their survey used simulation models when forecasting from three months to
over two years ahead.
Box-Jenkins was used regularly by 3.7% of the firms responding in 1983 (table 4). The survey also
revealed that this method had been dropped by 27% of those who returned questionnaires. These
results agree with Mentzer and Cox (1984) who found that Box-Jenkins was used by 5% to 6% of
their respondents and that many were dissatisfied with the technique. The lower rate of utilization of
Box-Jenkins may be due to its heavy demands for historical data and forecaster time.
Respondents were asked to indicate their utilization of 15 different forecasting methods in this
survey. The firms could check as many or as few of the methods as they found useful. A tabulation of
the responses showed that the average firm employed 2.7 forecasting techniques on a regular basis.
This suggests that most firms do not rely on a single forecasting method, but prefer to compare
forecasts from several techniques or use different techniques in different situations

3.3. Effects of forecast horizons

Table 5 shows the relationship between the use of forecasting techniques and the length of the
forecast horizon for the current survey and Mentzer and Cox’s study. The percentage of, respondents
using the various methods tended to be higher in the Mentzer and Cox survey. This was partially due
to the fact that their study included a smaller number of techniques, thus allowing respondents to
concentrate their responses on fewer choices. Respondents to both surveys mentioned the largest
number of techniques for use in making medium-length forecasts. A smaller number of procedures
was mentioned for short-term forecasts and a relatively few techniques were reported for long-range
forecasting.
The most popular short term methods in both surveys included sales force composite, jury of
executive opinion, customer expectations, moving average, and projection techniques. These tech-
niques were also the more frequently used medium-length procedures. In the Mentzer and Cox study
regression analysis was one of the most frequently mentioned procedures for medium length
forecasts; the use of leading indicators was popular for medium-length forecasts in the current
survey.
384 D. J. Dahymple / Sales forecasting pructices

Table 5
Percentage of respondents using techniques for different forecast horizons

Methods Forecast period

Dalrymple Mentzer & Cox a Dalrymple Mentzer & Cox a Dahymple Mentzer & Cox a
short up to 3 months medium 3 mo.-2 yrs. over over 2 years
1-3 months 4 mo.-1 yr. 1 year
Subjectme

Sales force composite 23.1% 31% 34.3% 36% 5.2% 8%


Jury of executive opinion 18.7 31 29.1 42 6.1 38
Intention to buy survey 10.4 11.2 4.5
25 24 12
Industry survey 8.2 > 15.7 > 11.4 >

Extrapolation

Naive 34.3% _ 17.9% 0.7% _


Moving average 17.9 24 12.7 22 2.2 5
% rate of change 13.4 8.2
21 28 21
Unit rate of change 9.1 >
10.4 9.1 > 4.5 >
Exponential smoothing 9.1 24 9.0 17 6.7 6
Line extension 6.0 13 8.2 16 3.7 10
Leading indicators 3.1 _ 20.1 _ 1.6 _

Quantitative

Box-Jenkins 6.0% 5% 3.7% 6% 2.2% 2%


Multiple regression
analysis 5.2 11.9 ‘i 4.5
Simple regression 14 36 28
analysis 5.2 1.5 I 3.0 1
Econometric models 2.2 4 10.4 9 7.5 10

* Adapted from Mentzer and Cox (1984).

The number of firms reporting use of the above techniques for long range forecasting declined.
Jury of executive opinion held up fairly well in the Mentzer and Cox survey, but moving average,
exponential smoothing, regression analysis, and other procedures recorded sharp declines in usage.
Customer expectations, projective techniques, regression analysis, and econometric models seemed to
have had the highest degree of acceptance for long-range projections in both surveys.
The naive model was the most popular method for short-range forecasts in the current study; it
was used by 18% of the firms of medium-length forecasts. However, fewer than 1% of the
respondents mentioned the use of this approach for forecasts over one year. The failure of survey
respondents to use the naive model for long-term forecasting contrasts sharply with Schnaars’ (1984)
findings that showed the naive method can accurately forecast from l- to 5-years ahead. These
results suggest that the naive model deserves more attention for long range forecasting than it
received in this study.

3.4. Market and size effects

One factor that helps explain preferences for forecasting methods is the type of product sold by
the firm (table 6). Note that industrial firms have a much stronger preference (34%) than producers
D.J. Daltympie / Sales forecusring practices 385

Table 6
Regular usage of sales forecasting techniques by industrial and consumer firms

Methods Percent of Percent of


industrial firms consumer firms
Subjectrue

Sales force composite 33.9% 13.0%


Jury of executive opinion 25.4 19.6
Industry survey 6.8 8.7
Intentions to buy 6.8 4.3

Extrapolation

Naive 18.6 17.4


Leading indicators 16.9 2.2
Moving average 8.5 10.9
Unit rate of change 6.8 6.5
Percentage rate of change 5.1 15.2
Exponential smoothing 3.4 10.9
Line extension 1.7 6.5

Quantiiative

Econometric models 10.2 4.3


Multiple regression 10.2 4.3
Simple regression 5.1 2.2
Box-Jenkins

Number of firms 59 46

of consumer goods (13%) for the sales force composite method. These results agree with previous
research and suggest that the close relationship between industrial salespeople and their customers
encourages industrial firms to use their sales force to forecast. Industrial companies also showed a
17% preference for the use of leading indicators, as compared with only 2.2% of the consumer goods
producers who mentioned this approach (table 6). This finding implies that industrial firms employ
forecasters who have used more relevant leading indicators. Table 6 also indicates that industrial
firms are more likely to use sophisticated econometric models and multiple regression analysis
procedures than the consumer goods firms. Consumer goods manufacturers had few strong prefer-
ences for forecasting methods as compared with the industrial respondents, and also mentioned
fewer techniques.
When the forecasting methods used by small firms were compared with those of large companies,
some interesting differences were observed (table 7). Managers in small firms had stronger prefer-
ences for subjective and extrapolation procedures and executives in the large firms were more likely
to use quantitative methods such as Box-Jenkins, regression analysis, and econometric models.

3.5. Forecast accuracy

This survey asked a series of questions to determine recent trends in forecasting errors and to
identify variables that affect accuracy. Three resources that are sometimes used to help reduce errors
are computers, purchased forecasting software, and consultants. In the eight years since my first
survey, the availability of computers and forecasting software has increased dramatically. Sixty-four
386 D.J. Dalrymple / Sales forecasting practices

Table I
Regular use of sales forecasting techniques by small and large firms

Methods Percent of small firms Percent of large firms


$1 to $100 M more than $500 M
Suhjectir~e

Jury of executive opinion 37.0% 20.0%


Sales force composite 33.3 40.0
Industry survey 14.8 0.0

Extrupolatm

Naive 27.8 20.0


% rate of change 16.7 5.0

Box-Jenkins 0.0 10.0


Simple regression analysis 1.9 15.0
Multiple regression analysis 5.5 25.0
Econometric models 0.0 40.0

Number of firms 54 20

percent of the respondents now say they always or frequently use computers in sales forecasting
compared to 44% who gave this response in 1975 (table 8). These results are similar to the findings of
Cerullo and Avila (1975) who reported that 57% of a group of Fortune 500 firms used computers in
forecasting. This trend toward increased use is supported by Mentzer and Cox (1984); their study
showed that 16% of their respondents had recently adopted computer forecasting models.
The current survey showed that the use of consultants in sales forecasting was similar to that
observed in the 1975 survey (table 8). Cerullo and Avila (1975) also found that consultants were not
widely employed (4% of 56 companies).
Most of the computer programs used in sales forecasting were developed internally (table 9). Note
that the proportion of firms buying computer forecasting programs on the outside doubled from 10
to 21% since the last survey. A number of colleagues had suggested that a few programs dominate the
computer forecasting software market. The results of the survey showed the reverse, since it revealed
that 29 different programs were being used but only one program was mentioned by more than four
firms. 3

Table 8
Reported usage of computers and consultants in sales forecasting.

Factors Always Frequently Occasionally Never No answer

Computers (1975) 25.0% 19.0% 16.0% 33.0% 7.0%


Computers (1983) 39.1 24.8 20.3 15.8 _

Consultants (1975) 3.0 6.0 27.0 56.0 8.0


Consultants (1983) 3.8 7.5 31.6 54.1 3.0

’ A list of externally supplied forecasting software mentioned by respondents is available from the author on request
D.J. DalTmple / Sales forecasting practices 387

Table 9
Sources of computer programs used in sales forecasting

Source of program Percentage of firms


1975 1983
Developed internally 55% 71%
Purchased with software package 10 21
Provided by outside consultants 9 8
No answer 26 _

100% 100%

Table 10
Maintenance of forecasting success records.

Response Percent of firms

Records are kept on cumulative file for ready access 54.1%


Records are kept for one year only 26.3
No forecasting records are kept 14.3
No answer 5.3

Error estimates. Surprisingly, about 80% of the surveyed firms maintained records of forecast
accuracy on a regular basis (table 10). This finding made it easier to believe the extent of forecast
errors reported in table 11. To help assure that the errors shown in table 11 were consistent, an error
formula (% error = (Forecast-Actual)/Actual) was printed on the questionnaire and respondents
were asked for errors for specific time horizons and levels of aggregation. The results suggest that
mean absolute percentage errors (MAPE) increased with the length of the forecast horizon except
for quarterly forecasts (table 11). A typical one-year forecast had an error of 9.9% with a standard
deviation of 7.9%. The size of forecasting errors presented in table 11 corresponds with those recently
reported by Mentzer and Cox (1984).
When the one-year forecasting errors shown in table 11 were compared with the errors reported in
responses to earlier surveys, the results shown in table 12 were obtained. The average forecasting
errors for the current survey were somewhat higher than the errors reported in previous surveys.
There is considerable doubt that these findings represent a trend. A probable explanation is that the
changes reflect different samples of firms or perhaps improvements in the accuracy of the estimates.

Table 11
Mean absolute percentage errors by forecasting horizons

Forecasting horizon Mean error Standard Median


(MAPE) deviation errors
One month 9.5% 7.7% 7.8%
One quarter 9.3 5.9 8.2
One year 9.9 7.9 8.0
Three year 14.2 8.3 11.0
388 D.J. Dal~~~ple / Sales forewsrrngpracticeJ

Table 12
Trends in reported one year sales forecasting errors

Mean percent error Mean percent error Mean percent error


1961 S M Survey * 1915 survey 1983 survey

5% 6.9% 9.9%

“ Sales Munagrment (1967).

Fuctors related to accuracy. Data on the effect of computer usage were obtained by correlating
monthly forecasting errors with reported computer usage (table 13). The negative relation observed
between errors and use of computers (r = - 0.27) says that as computer usage increased, forecasting
errors declined. Forecasting errors also decreased in large firms, when data were seasonally adjusted
and when more persons were involved in the forecasting process.
The possibility that size of firm influenced the correlations between MAPE, computer use, and
other factors led us to examine the partial correlations, controlling for size (table 13). Observe that
the partial correlation for seasonal adjustments was unchanged and that the correlation between
computer use and MAPE declined only from -0.27 to -0.26. These results suggest that the
correlations presented in table 13 are not merely a result of size differences among respondents.
The relationship between monthly forecasting errors and four categories of computer usage took
this form:

Computer use Always Frequently Occasionally Never


Percentage errors 7.8% 8.7% 11.7% 14.2%

Note that mean forecasting errors increased from 7.8% for firms that always used the computer to
14.2% for those that never used the computer. These results seem to suggest that forecasters who used
computers and the associated quantitative forecasting techniques produced more accurate forecasts
than those who relied on more subjective approaches. However, it is possible that firms using
computers had more historical data and more predictable time series than the noncomputer users.
When forecasting errors were calculated for different sizes of firms, we obtained:

Size of firm $l-$lOOM $lOl-$500M $501-$lB More than $lB


Forecasting errors 11.2% 9.2 9.2% 5.9%

Small firms reported average monthly errors of 11.2% as compared to 5.9% for large firms. This
finding seems reasonable since large firms often operate in more stable and mature markets than the
volatile developing markets in which small firms usually operate.

Table 13
Correlations between reported monthly forecasting errors ( MAPE) and selected factors.

Factors Correlations Significance n Partial correlations with


with MAPE level MAPE controlling for size
Use of computers - 0.27 0.004 93 - 0.26
Size of firm -0.18 0.037 90
Use of seasonal adjustments -0.16 0.063 93 ~0.16
Number of people preparing
forecast -0.13 0.106 93 -0.11
D.J. Dalrymple / Sales forecasting practrce.r 389

Table 14
Seasonal data adjustments and monthly forecasting errors

Response Percentage Mean forecasting


of respondents errors

Forecasts are seasonally adjusted 39.1% 8.4%


Forecasts are sometimes seasonally adjusted 33.8 9.6
Forecasts are never seasonally adjusted 27.1 11.7

This survey revealed a significant negative correlation (-0.16) between the use of seasonal
adjustments and reported forecasting errors (table 13). Firms that made greater use of seasonable
adjustments reported a lower level of errors. The effect of seasonal data adjustments on forecasting
errors is shown in table 14. Thirty-nine percent of the responding firms said their forecasts were
seasonally adjusted. Another 34% reported that forecasts were sometimes adjusted and 27% indicated
they did not use any seasonal corrections. Firms that routinely adjusted their data reported average
errors of 8.4% compared with 11.7% for those that made no seasonal adjustments. This finding shows
that simple seasonal adjustments can reduce forecasting errors. The finding also agrees with
empirical research [Makridakis et al. (1982) and Makridakis and Hibon (1979)] that indicates
seasonal adjustments improve forecasting accuracy.
Research has shown that combining forecasts from two or more methods to obtain a single sales
forecast can result in fewer forecasting errors [Makridakis et al. (1982) and Makridakis and Winkler
(1983)]. Fewer than half of the firms responding to our survey used the procedure of combining
forecasts, as is shown in table 15. About 61% of the respondents either did not combine forecasts or
did so only occasionally. This contrasts with the results of PoKemper and Bailey (1970) who claim
that combining is common practice among business forecasters. When the responses shown in table
15 were correlated with mean monthly forecasting errors, no significant relationship was found.
These results suggest that combining forecasts may be restricted to situations where uncertainty is
high. The absence of a significant association between the use of combined forecasts and reported
forecasting errors deserves further research. Respondents were also asked whether they keep track of
expenditures on sales forecasting. Sixty-five percent of the firms said they did not and 26% replied
that they had partial records. Only 9% of all firms had detailed information on their expenditures for
forecasting. One interpretation of this finding is that business firms apparently treat forecasting as a
free good and that forecasting managers should buy more computers and hire more people. Another
possibility is that forecasting is treated so casually that it is not even given a formal budget.
Obviously, neither condition is particularly desirable.

Table 15
Frequency of combining sales forecasts by respondents

Response Percentage of firms


Forecasts usually combined 19.5
Forecasts frequently combined 18.8
Forecasts occasionally combined 29.3
Forecasts never combined 31.6
No answer 0.8

100.0%
D.J. Dalrymple / Sala forecasting practices

4. Discussion

The results of this survey reveal several contradictions that need to be explained. One issue is the
continued strong preference of respondents for subjective and extrapolation techniques despite the
growing availability of computers that make possible the efficient use of numerical forecasting
techniques. For example, Sparkes and McHugh (1984) found that the use of simple forecasting
methods was preferred by British executives and that few British companies were taking steps to
adopt more sophisticated techniques. Sparkes and McHugh were quite concerned that British
executives lacked knowledge of or interest in techniques that would benefit by computerization. Also,
Mentzer and Cox (1984) report that for U.S. forecasts up to two-years-ahead, the most popular
methods were jury of executive opinion and sales force composite. For three-month forecasts,
Mentzer and Cox show that analysis of customer expectations was the next most popular method,
while trend-line analysis ranked just below the moving average technique. The current study also
found that the sales force composite and jury of executive opinion were the most frequently used
methods among American forecasters, followed by the naive method. At the same time, this study
showed that firms that made the greatest use of the computer had the fewest forecasting errors.
The strong preferences for subjective and extrapolation techniques cannot be dismissed with the
comment that the methods are generally believed to be inaccurate and that managers are unaware of
more advanced techniques. Research by Lawrence, Edmundson and O’Connor (1985) has shown that
subjective methods can be as accurate as more complex statistical procedures. The current survey
indicated that business firms keep records of forecasting accuracy and that they are unlikely to
continue to use a method that fails to perform. A better explanation may be that business
organizations tend to favor methods that are familiar and easy to use. When Mentzer and Cox asked
their respondents what factors in addition to accuracy were considered, the most common response
was ease of use. The ease of use factor helps explain the observed preferences for simple techniques
and may also explain the low acceptance for Box-Jenkins procedures.
Another issue is that although a great many business enterprises are using simple forecasting
techniques, business schools continue to place heavy emphasis on the use of complex procedures in
student forecasting projects. For example, Hanke (1984) found that professors favored regression
analysis (62 mentions), Box-Jenkins (25 mentions), time series decomposition (25 mentions), and
exponential smoothing (16 mentions). Also, the majority of professors used some kind of computer
forecasting package or canned programs in their forecasting courses.
One explanation is that the business community still finds simple techniques appealing while
professors are now emphasizing the more complex methods that belong to the emerging computer
age. Also, it may be easier to teach routine numerical procedures than the more abstract subjective
approaches. Another possibility is that professors are not giving sufficient attention to explaining
how to use the more complicated numerical procedures and therefore such quantitative techniques
are forgotten by students after they graduate. Others might say that the use of numerical forecasting
models would expand if the computer programs were more user-friendly.

5. Conclusions

Our survey revealed that most firms do not prepare alternative sales forecasts for different
strategies, environments, or capabilities. Also, few companies construct confidence intervals around
their forecasts nor do they combine forecasts to improve efficiency. Typical respondents used several
different sales forecasting techniques; these tended to vary by time horizon, size of firm, and type of
product. Subjective forecasting techniques remain popular and the naive model was used by a
surprisingly large number of firms for short- and medium-range predictions.
D.J. Dulrymple / Sales forecasting practices 391

The current survey suggests that firms can improve forecasting accuracy by increasing their use of
computers and the numerical forecasting procedures that go with them. Research has shown that the
use of methods such as simple exponential smoothing can lower forecasting errors compared with the
moving average methods favored by many respondents to this study.
The survey also showed that firms that seasonally adjust their sales forecasts reported a lower level
of errors. These results have been confirmed by empirical research and provide strong support for
adopting this procedure. Seasonal adjustments appear to be the easiest and most cost-efficient way to
improve sales forecasting accuracy.

References

Armstrong, J.S. et al., 1983, The accuracy of alternative extrapolative models: Analysis of a forecasting competition through
open peer review, Journal of Forecasting 2, 259-311.
Dalrymple, Douglas J., 1975, Sales forecasting methods and accuracy, Business Horizons 18, 69-73.
Cerullo, Michael J. and Alfonso Avila, 1975, Sales forecasting practices: A survey, Managerial Planning 24. 33-39.
Hanke, John, 1984, Forecasting in business schools: A survey, Journal of Forecasting 3, 229-234.
Lawrence, Michael J., Robert H. Edmundson and Marcus J. O’Connor, 1985, An examination of the accuracy of Judgmental
extrapolation of time series, International Journal of Forecasting 1, 25-35.
Makridakis, Spyros, and Michele Hibon, 1979, Forecasting accuracy and its causes: An empirical investigation, The Journal of
the Royal Statistical Society, Series A, 142, Part 2, 97-125.
Makridakis, S. et al., 1982, The accuracy of extrapolation (time series) methods, Journal of Forecasting 1, 111-153.
Makridakis, Spyros, and Robert L. Winkler, 1983. Averages of forecasts: Some empirical results, Management Science 29.
987-996.
Mentzer, John T. and James E. Cox, Jr., 1984, Familiarity, application, and performance of sales forecasting techniques,
Journal of Forecasting 3, 27-36.
McLaughlin, Robert L., 1983, Forecasting models: Sophisticated or naive? Journal of Forecasting 2, 274-276.
PoKemper, Stanley J. and E. Bailey, 1970. Sales forecasting practices (The Conference Board, New York).
Rothe, James T., 1978, Effectiveness of sales forecasting methods, Industrial Marketing Management 7, 114-118.
Rush, Howard and William Page, 1979, Long term metals forecasting: The track record: 1967-1964, Futures 11, 321-337.
Sales Management, December 15, 1967, Sales forecasting: Is five percent error good enough? 37-42.
Schnaars, Steven P., 1984, Situational factors affecting forecast accuracy, Journal of Marketing Research 21, 290-297.
Sparkes, John R. and A.K. McHugh, 1984, Awareness and use of forecasting techniques in British industry, Journal of
Forecasting 3, 37-42.
Wheelwright, Steven C. and D.G. Clarke, 1976, Corporate forecasting: Promise and reality, Harvard Business Review 54,
40-42.

Biography: Douglas J. DALRYMPLE is Associate Professor of Marketing in the School of


Business at Indiana University. He received his D.B.A. from Michigan State University
and has taught at the University of California, Los Angeles, the Georgia Institute of
Technology, the University of San Diego, and the University of North Carolina,
Greensboro. His articles have appeared in Decision Sciences, Industrial Marketing
Management, Journal of Business Research and other journals. Professor Dalrymple is
the coauthor of Marketing Management: Strategies and Cases, 4th ed. and the author of
Sales Management: Concepts and Cases, 2nd ed.

You might also like