Forecasting by Extrapolation: Conclusions from Twenty-five Years of Research

Sophisticated extrapolation techniques have had a negligible payoff for accuracy in forecasting. As a result, major changes are proposed for the allocation of the funds for future research on extrapolation. Meanwhile, simple methods and the combination of forecasts are recommended. Comments Postprint version. Published in Interfaces, Volume 14, Issue 6, November 1984, pages 52-66. Publisher URL: http://www.aaai.org/AITopics/html/interfaces.html The author asserts his right to include this material in ScholarlyCommons@Penn. This journal article is available at ScholarlyCommons: http://repository.upenn.edu/marketing_papers/77 Published in Interfaces, 14 (Nov.-Dec. 1984), 52-66, with commentaries and reply. Forecasting by Extrapolation: Conclusions from 25 Years of Research J. Scott Armstrong Wharton School, University of Pennsylvania Sophisticated extrapolation techniques have had a negligible payoff for accuracy in forecasting. As a result, major changes are proposed for the allocation of the funds forfuture research on extrapolation. Meanwhile, simple methods and the combination of forecasts are recommended. Have advances in extrapolation methods helped to make short-range forecasts better now than in 1960? I am defining extrapolation as methods that rely solely on historical data from the series to be forecast. No other information is used. This class of methods is widely used in forecasting, especially for inventory control, process control, and in situations where other relevant data are not available. I describe a forecasting procedure that was used in 1960 and then present evidence from research published over the last quarter of a century. Short-Range Forecasting in 1960 As an industrial engineer at Eastman Kodak in the early 1960s, I examined the short-range forecasting system for color print orders from customers. These forecasts were needed to schedule part-time workers and to control inventories. The procedure that had been used for many years prior to 1960 had the following characteristics: − Weekly historical data on each type of order were collected on the basis of billing records. In general, these data were thought to be accurate. Outliers were adjusted or removed. − Graphs were prepared for the more important items. − The forecasts were then prepared judgmentally by a man who had been doing this job for many years. Existing literature implied that better forecasts could be obtained by using objective methods. Accordingly, I developed and tested a model that used exponential smoothing of deseasonalized data. The deseasonalizing also included adjustments for trading days and holidays. Search procedures were then used to find the most appropriate smoothing factors for the average and the additive trend. The procedures were based primarily on Brown [1959] and Shiskin [1958]. Historical comparisons showed that the exponential smoothing model was superior to the judgmental method for almost all items. Side-by-side comparisons over the next six months provided additional evidence on the superiority of the exponential smoothing model. Given what we now know, would it have been possible to improve upon this extrapolation model, which had been developed with methods described in publications prior to 1960?

[1]  E. S. Gardner,et al.  FORECASTING WITH EXPONENTIAL SMOOTHING: SOME GUIDELINES FOR MODEL SELECTION , 1980 .

[2]  J. Durbin,et al.  Local trend estimation and seasonal adjustment of economic and social time series (with discussion) , 1982 .

[3]  R. L. Winkler,et al.  Averages of Forecasts: Some Empirical Results , 1983 .

[4]  R. Lucas Econometric policy evaluation: A critique , 1976 .

[5]  Gene K. Groff Empirical Comparison of Models for Short Range Forecasting , 1973 .

[6]  D. Smith Adaptive Response for Exponential Smoothing: Comparative System Analysis , 1974 .

[7]  Robert Fildes,et al.  The accuracy of extrapolation methods; an automatic box–jenkins package sift , 1984 .

[8]  J. Hajnal,et al.  THE PROSPECTS FOR POPULATION FORECASTS , 1955 .

[9]  Robert J. Genetski,et al.  Long-Range Forecasting: From Crystal Ball to Computer , 1981 .

[10]  C. Chatfield,et al.  Box‐Jenkins Seasonal Forecasting: Problems in a Case‐Study , 1973 .

[11]  H. Dorn,et al.  Pitfalls in Population Forecasts and Projections , 1950 .

[12]  Chris Chatfield,et al.  The Holt-Winters Forecasting Procedure , 1978 .

[13]  Marcus O'Connor,et al.  An examination of the accuracy of judgmental extrapolation of time series , 1985 .

[14]  Jesse E. Raine SELF ADAPTIVE FORECASTING RECONSIDERED , 1971 .

[15]  Edward J. Lusk,et al.  The Accuracy of Alternative Extrapolation Models: Analysis of a Forecasting Competition Through Open Peer Review , 1983 .

[16]  J. M. Bates,et al.  The Combination of Forecasts , 1969 .

[17]  Robert L. Winkler,et al.  The accuracy of extrapolation (time series) methods: Results of a forecasting competition , 1982 .

[18]  Michael D. Geurts,et al.  Comparing the Box-Jenkins Approach with the Exponentially Smoothed Forecasting Model Application to Hawaii Tourists , 1975 .

[19]  R. Douglas Martin,et al.  ROBUST METHODS FOR TIME SERIES , 1981 .

[20]  George E. P. Box,et al.  Intervention Analysis with Applications to Economic and Environmental Problems , 1975 .

[21]  Johannes Ledolter,et al.  A Recursive Kalman Filter Forecasting Approach , 1983 .

[22]  R. Rosenthal,et al.  Statistical versus traditional procedures for summarizing research findings. , 1980, Psychological bulletin.

[23]  V. Mabert STATISTICAL VERSUS SALES FORCE–EXECUTIVE OPINION SHORT RANGE FORECASTS: A TIME SERIES ANALYSIS CASE STUDY , 1976 .

[24]  Gary Fromm,et al.  THE NBER/NSF Model Comparison Seminar: An Analysis of Results , 1976 .

[25]  Everette S. Gardner,et al.  Exponential smoothing: The state of the art , 1985 .

[26]  W. Ruland,et al.  On the Choice of Simple Extrapolative Model Forecasts of Annual Earnings , 1980 .

[27]  E. Adam INDIVIDUAL ITEM FORECASTING MODEL EVALUATION , 1973 .

[28]  E. Elton,et al.  Earnings Estimates and the Accuracy of Expectational Data , 1972 .

[29]  Robert Carbone,et al.  Comparing for Different Time Series Methods the Value of Technical Expertise Individualized Analysis, and Judgmental Adjustment , 1983 .

[30]  R. Kirby A Comparison of Short and Medium Range Statistical Forecasting Methods , 1966 .

[31]  Michèle Hibon,et al.  Accuracy of Forecasting: An Empirical Investigation , 1979 .

[32]  Jeffrey E. Jarrett,et al.  Note-Revising Forecasts of Accounting Earnings: A Comparison with the Box-Jenkins Method , 1983 .

[33]  Steinar Ekern Adaptive Exponential Smoothing Revisited , 1981 .

[34]  Gwilym M. Jenkins,et al.  Time series analysis, forecasting and control , 1972 .

[35]  Peter R. Winters,et al.  Forecasting Sales by Exponentially Weighted Moving Averages , 1960 .

[36]  Everette S. Gardner,et al.  Evolutionary operation of the exponential smoothing parameter: Revisited , 1983 .

[37]  Derek W. Bunn The Synthesis of Predictive Models in Marketing Research , 1979 .

[38]  J. Armstrong Forecasting with Econometric Methods: Folklore Versus Fact , 1978 .

[39]  J. Armstrong Monetary Incentives in Mail Surveys , 1975 .

[40]  Everett E. Adam,et al.  AN EMPIRICAL EVALUATION OF INDIVIDUAL ITEM FORECASTING MODELS , 1981 .

[41]  J. Shiskin,et al.  Decomposition of Economic Time Series: Business fluctuations are broken down into seasonal, cyclical, and irregular factors by computers. , 1958, Science.

[42]  C. Christ,et al.  Judging the Performance of Econometric Models of the U.S. Economy , 1975 .

[43]  C. Gray,et al.  AN EMPIRICAL EVALUATION OF CONSTANT AND ADAPTIVE COMPUTER FORECASTING MODELS FOR INVENTORY CONTROL , 1977 .

[44]  Vincent A. Mabert,et al.  FORECAST MODIFICATION BASED UPON RESIDUAL ANALYSIS: A CASE STUDY OF CHECK VOLUME ESTIMATION , 1978 .

[45]  Victor Zarnowitz,et al.  An Analysis of Annual and Multiperiod Quarterly Forecasts of Aggregate Income, Output, and the Price Level , 1979 .

[46]  Werner G. Frank,et al.  A Study of the Predictive Significance of Two Income Measures , 1969 .

[47]  C. Granger,et al.  Experience with Forecasting Univariate Time Series and the Combination of Forecasts , 1974 .

[48]  L. Brown,et al.  The Superiority of Analyst Forecasts as Measures of Expectations: Evidence from Earnings , 1978 .

[49]  R. Brown Statistical forecasting for inventory control , 1960 .

[50]  Ws Albrecht,et al.  TIME-SERIES PROPERTIES OF ANNUAL EARNINGS , 1977 .

[51]  M. J. Morris Forecasting the Sunspot Cycle , 1977 .

[52]  J. Ledolter,et al.  Parsimony and Its Importance in Time Series Forecasting , 1981 .

[53]  Michael Lawrence An exploration of some practical issues in the use of quantitative forecasting models , 1983 .

[54]  T. E. Johnson,et al.  Effectiveness of Earnings Per Share Forecasts , 1974 .