comparing judgmental to extrapolative forecasts: it's time to ask why and when
TRANSCRIPT
Intrmarvxal Journal of Forecasting 4 (1985) 171-173
Sorrh-Holland 171
ED ITORWL
Comparing Judgmental to Extrapolative Forecasts: It’s Time to Ask Why and When
The forecasting literature has devoted considerable attention to evaluating the comparative
accuracy of judgmental versus extrapolative forecasts. Surprisingly, little attention has been paid to identifying circumstances in which judgmental forecasts are likely to be comparatively more accurate
than extrapolative forecasts. One approach to the issue was taken by Armstrong (1985. Ch. 15). who hypothesized that subjective methods are likely to be relatively more accurate than objective methods when small changes are expected in the environment and few objective data are available. This editorial takes a somewhat different approach. hypothesizin, 0 that certain environmental factors need
to be considered when comparing judgmental to extrapolative forecasts. Three factors are proposed: the influence that the forecaster exerts over the event being forecast, the self-selection bias of the forecaster. and the inside information of the forecaster. There is no presumption that these are the only (or necessarily the most important) factors that deserve the consideration of forecasters. The point here is simply that environmental factors are too often ignored by practitioners and academics when they assess the predictive validity of judgmental versus extrapolative forecasts.
Consider earnings forecasts made by a firm’s manager. These forecasts represent an extreme case
in the sense that the forecaster influences the outcome of the number being forecast, he possesses a self-selection bias, and he is privy to inside information. Regardin, 0 the first factor, the manager has control over a variety of income-impacting events, such as the deferral of sales. the build up of inventories. and the amount to he recorded as bad debt expense. When all else fails, the manager can even change the firm’s accounting principles in an effort to report an actual earnings number that conforms with the forecast. Regarding the second factor, managers generally do not make estimates of their firms’ future earnings numbers publicly available. It is reasonable to presume that when managers do make their forecasts publicly available, they are better able to ‘predict’ (manufacture?) earnings than when they do not reveal their estimates to the public. Regarding the third factor. it is evident that managers have inside information about their firm’s production, investment, and financing activities, and hence are privy to asymmetric information regarding their firms’ future reported earnings numbers.
Consider the weather forecasts of a weather forecaster. Weather forecasts represent the opposite
end of the forecast spectrum in the sense that weather forecasters cannot influence the weather, these forecasters do not have a self-selection bias, and they do not possess inside information about what the weather will be at any future time. The second factor desencs some clarification. Cnlike the manager who can decide to predict earnings when she perceives she has a comparative advantage in doing so. the weather forecaster cannot take this attitude ttitiard weather forecasts. Surely. she could improve her forecasts relative to extrapolative techniques by showing up for work when the future state of nature is known with near certainty, and staying home when there is considerable uncertainty regarding which future state of nature will prevail. Nevertheless, this (self-selection) option is unavailable to the weather forecaster who seeks continuous employment.
0169-2070/Mjf3.50 ‘7 19RR. Elsevirr Science Puhh\hers R.V. (North-Holland)
U’hilr the weather forecaster and the firm’s manager represent two opposite ends of the spectrum. there are many types of forecasters who lie somewhere in between these two extremes. Consider the security analyst who predicts a firm’s future earnings numbers. The security analyst has some influence over the reported earnings numbers, some self-selection bias. and some inside information.
Regarding the first factor, security analysts and their clients do not like to be surprised by unexpected reported earnings numbers, especially those that are Iower than predicted by the analyst. If the analyst and his clients are surprised. the analyst can sanction the firm’s managers by not recommending the firm’s financial securities to his clients. Knowing this. firms‘ managers may undertake actions to create earnings numbers that approximate the analysts’ expectations. As t0 the second factor. analysts do not follow all firms in all industries. Instead. they self-select certain firms (industries). presumably those of which they are most knowledgeable. 11loteover. many analysts will not make forecasts for a firm when it is subject to heightened uncertainty. such as an impending strike or a merger. Indeed. analysts have been known to request that their forecasts be withdrawn at times of heightened uncertainty. Regardin, 0 the third factor, analysts are constantly in touch with managers who provide them with inside information regarding their firms’ future earnings. In fact, analysts’ forecasts are often managers’ forecasts in disguise. Managers have incentives to let the
capital markets know what their firms* future earnings are going to be. and analysts have incentives to provide these forecasts to their clients [Ajinkya and Gift (19W)j.
Thus, it is likely that the comparative accuracy of judgmental versus extrapolative forecasts is greatest for the manager, least for the weather forecaster, and somewhere in between for the analyst. There exists some empirical evidence consistent with these hypotheses. Managers and analysts outperform extrapolative techniques in forecasting firms’ future earnings numbers 0ver short and long horizons, and managers earnings forecasts generally are more accurate than forecasts by analysts [Armstrong (19X3)]. On the other hand, while the weather forecaster has an advantage in predicting the probability of precipitation and trmpcraturcs over short (e.g.. 12-24 hours) lead times. her forecasts of these two factors arc not more accurate than extrapolative forecasts over long horizons. Moreover. the forecasts prepared by weather forecasters of cloud cover are less accurate than those of extrapolative forecasts for all lead times [Murphy and Brown (19S4)j.
Other forecasters also lie on the spectrum somewhere in between the manager and the weather forecaster. Those closer to the manager include the loan officer who must estimate the pr~~bab~lity of loan default and the mutual fund manager who must estimate the performance of individual stocks relative to that of the stock market in tht da, ’ * ooregate. Those closer to the weather forecaster include
the economist who estimates the U.S. balance of payments deficit, and the mutual fund manager who predicts changes in the stock market in Ihe aggregate.
In sum, judgmental forecasts are more likely to outperform extrapolative forecasts when the forecaster can exert influence over the event being forecast, is able to self-select those items to be forecast, and possesses inside information about the event under consideration. On the other hand, extrapolative forecasts are more likely to outperform judgmental forecasts when the forecaster cannot affect the outcome of the event being forecast. is not able to forego making forecasts when there exists considerable uncertainty over the outcome of the event being forecast. and does not possess inside information.
Researchers need to determine how tradeoffs among environmental factors impact on the comparative accuracy of judgmental versus extrapolative forecasts, and to estimate uhere different forecasters lie along the spectrum. Such research will enable both practitioners and rrscarchers to better understand oh! and W&H judgmental forecasts are likely to be more accurate than extrapola-
tive forecasts.
Elf1rorull 173
References
Alinky. B.B. and NJ. Gift. 1984. Corporate managers’ eammgs forecats and symmetrical adJustmats of market expectations. Journal of Accounting Research 22. 425-444.
Armstrong. J.S.. 1993. Relative accuracy of Judgemental and extrapolative methods in forecasting annual earnings. Journal of Forecasting 2. 437-447.
Armstrong, J.S.. 1985. Long range forecasting: From crystal ball to computer. ?d ed. (Wiley. New York).
Murphy. A.H. and B.C. Brown, 1984. A comparative evaluation of ObJectlve and subyxtive weather forecasts m the United States, Journal of Forecastmg 3, 369-393.
Lawrence D. Brown Alumni Professor of Accounting. SUNY at Buffalo. Buffalo. NY, USA