The Ombudsman: Research on Forecasting: A Quarter-Century Review, 1960-1984

Before 1960, little empirical research was done on forecasting methods. Since then, the literature has grown rapidly, especially in the area of judgmental forecasting. This research supports and adds to the forecasting guidelines proposed before 1960, such as the value of combining forecasts. New findings have led to significant gains in our ability to forecast and to help people to use forecasts. What have we reamed about forecasting over the past quarter century? Does recent research provide guidance for making more accurate forecasts, obtaining better assessments of uncertainty, or gaining acceptance of our forecasts? Comments Postprint version. Published in Interfaces, Volume 16, Issue 1, January 1986, pages 89-109. Publisher URL: http://www.aaai.org/AITopics/html/interfaces.html This journal article is available at ScholarlyCommons: http://repository.upenn.edu/marketing_papers/75 (Reprinted with permission from Interfaces (1986)16:1 (January-February), 89-103) Research on Forecasting: A Quarter-Century Review, 1960-1984 J. Scott Armstrong The Wharton School of the University of Pennsylvania Before 1960, little empirical research was done on forecasting methods. Since then, the literature has grown rapidly, especially in the area of judgmental forecasting. This research supports and adds to the forecasting guidelines proposed before 1960, such as the value of combining forecasts. New findings have led to significant gains in our ability to forecast and to help people to use forecasts. What have we reamed about forecasting over the past quarter century? Does recent research provide guidance for making more accurate forecasts, obtaining better assessments of uncertainty, or gaining acceptance of our forecasts? I will first describe forecasting principles that were believed to be the most advanced in 1960. Following that, I will examine the evidence produced since 1960. Forecasting Principles in 1960 Forecasts serve many needs in organizations and are employed for both short-range and long-range planning. They help in making decisions on production, personnel, finance, and marketing. Figure 1 shows the types of forecasts that are needed in addressing these decisions. For example, forecasts of competitive actions can help to assess a proposed strategy. The same needs for forecasts existed in 1960. I drew upon the pre-1960 literature and upon my experience (I was just starting my life as a forecaster in 1960) to de scribe what methods would have been used in 1960. From this, I developed 1960-style guidelines, six for making forecasts, one for estimating uncertainty in the forecasts, and one for gaining acceptance of the forecasts: (1) Decomposition should be used whenever feasible. Complex problems should be broken into a series of subproblems, each of which is to be solved, and the results then synthesized. Of particular importance are decompositions to separate the forecasting of company sales into industry sales and market share, estimate current status separately from forecasting change, and provide forecasts separately for each consumer market. Decomposition can be used with judgment, econometric or extrapolation methods. Its use in extrapolation (whereby average, trend, and seasonal components are examined) is especially popular. (2) Extrapolation should be used as one of the forecasting methods whenever the data allow. Exponential smoothing of deseasonalized data using a trend estimate will produce adequate results. (3) Obtain opinions from experts in the topic area, and do this in highly structured ways. Use these opinions to estimate current status, make forecasts of change, forecast in situations where data on the variable of interest (for example, the sales variable) are absent or of poor quality, and forecast the effects of actions by the firm when considering the reactions of major competitors, but do not use expert opinion to adjust the forecasts produced by objective methods. (4) Intentions surveys (for example, consumers' intentions) should be used to forecast short-term behavior for important events.