Important Errata for

J. Scott Armstrong Long-Range Forecasting, 2nd Edition, 1985.
Return to Table of Contents

Chapter

Page

Paragraph

Text Change

1

��� 9

first

�my interpretations were correct�

3

44

last

�The subscription rates of those in the scenario exercise were about double ��

5

78

3

Makridakis and Wheelwright 1984: should read Wheelwright, Stephen C. and Makridakis, Spyros

6

102

last

�forecast the thickness of a thick piece of paper�

109

2

Rosenthal and Jacobsen (1968)

114

3

Armstrong & Lusk (1987)

7

155

4

Henrichs

159

3

�� an assumption of stability of the transition matrix is made when Markov chains are used.�

162

1

winsorizing� [not windsorizing]

170

1

Schnaars and Bavuso [1986]

174

2

�� the moving average did not have a trend factor.� [no comma after average]

179

4

�� indicators. Simple smoothing ��

184

1

Geurts [not Geruts]

8

196

next to last

�� to choose variables that are�

226

5

�� I surveyed a group of leading econometricians [Armstrong, 1978].�

243

1

for the simple regression.� [not single]

9

263

4

�� and then describes how the system changes over time.�

10

285

2

Delete �It is also found in the large scale econometric models.�

Exh. 10-5

X4 = years of seniority in the current job

12

322

3

This chapter� instead of �Chapter 12�

327

1

�Both models were much more accurate than extrapolations during this period.�

13

344

Theil Table

Root Mean Square Errors� for heading of columns 2-5

350

4

�R2 should not be used for the calibration sample. Instead you should use (the adjusted R2.�

352

Equation in 9

360

Exh. 13-9

Academics (n=62) and Practitioners (n=61)

14

384

6

�search for disconfirming evidence ��

15

403

last

�A survey on econometric forecasting was conducted in late 1975 (Armstrong 1978).�

405

2

�The results in the following table �� [no comma after results]

410

3

�� assumed that the 1960 party vote would be the same as the 1956 party vote�

412

1

�the econometric model � was more accurate than the three extrapolation methods�

17

445

1

Armstrong & Lusk (1987)

Appendices

459

2

�Suppose that data are available on the population of a region ��

(The growth term is squared because there are two 10-year intervals from 1960-1980.)

495

1

Schnaars and Bavuso [1986]

References

517

Armstrong 1978: (LRF 226, 403, 433)

518

Armstrong, Denniston, and Gordon 1975: Organizational Behavior and Human Performance; (LRF58) [delete 181]

520

Bennion 1952: (LRF 199) [not 129]

Best 1974: (LRF 118, 119, 141)

529

Crow and Noel 1965: 1150 Silverado Street

547

Kaplan et al. 1950: (LRF 120-121, 136)

555

Mintz 1969: no parentheses around 875-881

557

Nelson 1972: American Economic Review, 52

564

Schneidman 1971: �Perturbation and Lethality ��

565

Sigall, H., Aronson, E., and van Hoose, T. 1970

574

Zajonc 1976: (LRF Page not available)

Updated Bibliography

578

Ackoff 1983: vol. 20

579

Ahlburg 1984: �A mechanical adjustment ��

581

Armstrong 1979: (LRF 437, 444) [not 437-444)

583

Armstrong 1984a: vol. 14 (Nov.-Dec.)

584

Add: Armstrong, J. Scott and Lusk, Edward J [1987], �Return Postage in Mail Surveys: A Meta-Analysis,� Public Opinion Quarterly, vol 51 (2), 233-248.

585

Ascher 1978: (LRF 47, 687)

591

Camerer 1981: �General conditions for the success of bootstrapping models,�

592

Carbone and Gorr 1985: Decisions Sciences, vol. 16, pp. 153-160 [replaces in press]

595

Collins 1976: �The segmented econometric models were more accurate ��

596

Replace Dalrymple 1985 reference for working paper: Dalrymple, Douglas J. [1987], �Sales forecasting practices: results from a 1983 U.S. survey,� International Journal of Forecasting, vol. 3, pp. 379-391.

599

Dielman 1985: Journal of Forecasting [not International Journal of Forecasting]

606

Fischer 1982: vol. 29

609

Gardner 1985b: Journal of Forecasting, vol. 4, pp. 1-28.

610

Glass 1976: (LRF 444)

Glass et al. 1981: (LRF 445)

620

Lawrence et al. 1985: International Journal of Forecasting, vol. 1, pp. 25-35.

623

Makridakis et al. 1982: (LRF348 �) [not 454]

624

Makridakis and Wheelwright 1984: should read Wheelwright, Stephen C. and Makridakis, Spyros

627

McWhorter et al. 1987 continuation, line 4: �� from 1950.I to 1970.IV�

628

Mitchell and Klimoski 1982 continuation, line 4: � � used the same subsample, but it started ��

630

Morris 1981: �Will such procedures improve our ability to forecast? (My guess is �No.�) That is a good topic for further research.

Mosteller and Tukey 1977: (LRF 24, 196, 341)

631

Murphy and Brown 1984: (LRF 142, 272, 424)

637

Reid et al. 1981: Journal of Advertising [not Journal of Advertising Research]

641

Schnaars and Bavuso 1986 [not 1985]: �Extrapolation models on very short-term forecasts,� Journal of Business Research, vol. 14, pp. 27-36.

643

Shrader et al. 1984: �to 14 comparisons favoring formal planning, 4 ties, and 2 favoring informal planning.�

645

Stewart and Glantz 1985: vol 7, no. 2, pp. 159-183.

646

Sudman and Bradburn 1982 [not 1983]

648

Wagennar et al. 1985: Ergonomics, vol. 28, pp. 756-772.

People Index

654

Ascher [not ASCHER]

654

Add Ayton, Peter, 585

662

Makridakis, Spyros, omit 454

664

Ohlin, Lloyd E., 354, 558 [not 221, 531]

667

Stengel, Casey, 653 [not 753]

669

Wheelwright, Steven C., add 649

Subject Index

672

Anticipations, 536 [not 537]

673

Backcasting, add 343-345

680

Pharaoh [not Pharoh]

681

Political forecasting, 636, 646 [not 636-646]

683

Suicides, 564 [not 562]

Theil�s U, 579 [not 578]

685

Winsorizing [not Windsorizing]

Return to Table of Contents

Taken from J. Scott Armstrong, 1985, Long-Range Forecasting, 2nd ed., p. 487.

Some people are impressed by a high R2. There are simple things that can be done to raise R2; other than that, they are of no value. The most important thing is not to use these rules (please don't), but to be aware that others use them.

  1. Discard outliers after you examine the regression results.
  2. Aggregate the data, especially when it reduces sample size significantly.
  3. Experiment by trying lots of variables.
  4. Try different functional forms.
  5. Use stepwise regression and retain all coefficients with t statistics greater than 1.0 (Haitovsky, 1969).
  6. Include a lot of variables in the final equation.
  7. Use R2 rather than .

These rules should yield R2 values of over 99% for time series data and about 90% for cross-sectional data. My advice is that you should not use R2 for time series data. The dangers outweigh any potential benefits.

The way that you structure the problem will affect the forecasting method you need.

Look first at the chart describing components of the sales forecast.

from J. Scott Armstrong (2001), "Introduction," in Principles of Forecasting, Norwell, MA: Kluwer Academic Publishers, p 5, reprinted courtesy of Kluwer Academic Publishers

Then consider the table below, describing possible methods for forecasting each component. The listed methods are likely to be most useful for the problem area stated. Particular attention should be given to the methods indicated in italics. Remember that combined forecasts typically lead to further improvements, so it is good to select more than one method.

Component

  Conditions   Methods for Forecasting
Environment   Small changes   expert forecasting extrapolation
    Large changes   rule-based forecasting; analogies; econometric methods
Market   Small changes   expert forecasting; intentions; extrapolation
    Large changes   analogies; rule-based forecasting; econometric methods
Company actions   Small changes   expert opinions; intentions; role-playing; analogies
    Large changes   expert forecasting; role-playing; analogies
Competitors' actions   Small changes   expert forecasting; role-playing; analogies; extrapolation
    Large changes   expert forecasting; role-playing; analogies
Actions by suppliers, distributors, and government   Small changes   expert forecasting; role-playing; analogies; extrapolation
    Large changes   expert forecasting; role-playing; analogies
Market share   Small changes   extrapolation; analogies
    Large changes   judgmental bootstrapping; rule-based forecasting; econometric methods
Costs   Small changes   extrapolation
    Large changes   expert forecasting; analogies; rule-based forecasting; econometric methods
Sales   Small changes   intentions; extrapolation
    Large changes   expert forecasting; judgmental bootstrapping; intentions; conjoint analysis; analogies; expert systems; econometric methods

For a further description of each area and the related research, see "Forecasting for Marketing" by Armstrong and Brodie.

Benchmarks are most useful if based on the specific situation. The ideal benchmark should be based on "how well does the current method forecast in this situation." However, this is not always easy to assess. In such cases, it may be useful to evaluate how accurately forecasters have been able to forecast in such situations (and how did they do that).

  1. Corporate Earnings Forecasting
  2. New Product Forecasting
  3. Sales Forecasting
  4. Employment Forecasting

1.Annual corporate earnings forecasts; one-year horizon

  1. Management forecasts were superior to professional analyst forecasts (the mean absolute percentage errors were 15.9 and 17.7, respectively, based on five studies using data from 1967--1974) and

  2. Judgmental forecasts (both management and analysts) were superior to extrapolation forecasts on 14 of 17 comparisons from 13 studies using data from 1964-1979 (the mean absolute percentage errors were 21.0 and 28.4 for judgment and extrapolation, respectively).

Source: J. Scott Armstrong (1983), “Relative Accuracy of Judgmental and Extrapolative Methods in Forecasting Annual Earnings,” Journal of Forecasting, 2, 437-447

2. New product forecasting

3. Sales forecasting

Mentzer & Cox (“Familiarity, application, and performance of sales forecasting techniques,” Journal of Forecasting, 3, 1984, 27-36) examined forecast errors for various levels in the product hierarchy and for different horizons:

Typical Errors for Sales Forecasts (Entries are MAPEs)
Level Forecast Horizon
Under 3 months 3 months to 2 years Over 2 years
Industry 8 11 15
Corporate 7 11 18
Product group 10 15 20
Product line 11 16 20
Product 16 20 26

Source: Mentzer and Cox’s survey results from 160 corporations are crude because most firms do not keep systematic records. Further, the study was ambiguous in its definitions of the time interval. “Under 3 months” probably refers to ‘monthly’ in most cases, but the length of time is not apparent for “Over 2 years.”

4. Employment forecasting

Mean Absolute Percentage Error in Forecasts of Engineering Employment, 54 Firms, 1957-1976
  Forecasting Horizon (years)
Industry 0.5 2 5 10
Aerospace 10.3 15.9 41.2 88.7
Electronics 4.6 12.4 15.4 26.5
Chemicals 3.2 5.7 17.3 22.0
Petroleum 2.8 5.5 13.1 9.4
Average 5.2 9.9 21.8 36.6

Source
: Brach, Peter and Edwin Mansfield (1982), "Firm's Forecasts of Engineering Employment," Management Science, 28, 156-160.

Subcategories