How Important Is Past Analyst Forecast Accuracy?

The Wall Street Journal rates analysts on the basis of past earnings forecast accuracy. These analyst ratings are important to practitioners who believe that past accuracy portends future accuracy. An alternative way to assess the likelihood of “more” or “less” accurate forecasts in the future is to model the analyst characteristics related to the accuracy of individual analysts' earnings forecasts. No evidence yet exists, however, as to whether an analyst characteristics model is better than a past accuracy model for distinguishing more accurate from less accurate earnings forecasters. I show that a simple model of past accuracy performs as well for this purpose as a more complex model based on analyst characteristics. The findings are robust to annual and quarterly forecasts and pertain to estimation and prediction tests. The evidence suggests that practitioners' focus on past accuracy is not misplaced: It is as important as five analyst characteristics combined. Practitioners often rely on the accuracy of analysts' past earnings forecasts to predict future forecast accuracy. A number of sources provide ratings or rankings of past accuracy. For example, the Institutional Investor All-America Research Team rankings and the StarMine SmartEstimate are based partly on past accuracy, and the ratings published annually in the Wall Street Journal are based entirely on past accuracy. An alternative way to assess the likelihood of “more” or “less” accurate forecasts in the future is to model the analyst characteristics related to the accuracy of analysts' earnings forecasts. No evidence yet exists, however, as to whether an analyst characteristics model is better than a past accuracy model for distinguishing more accurate from less accurate earnings forecasters. I addressed this issue in the study reported here.Consistent with Wall Street wisdom that past accuracy portends future accuracy, several researchers have shown that past accuracy is significantly positively correlated with current accuracy. Other academic researchers have identified numerous intuitively appealing analyst characteristics related to earnings forecast accuracy, including the analyst's company-specific experience, the analyst's general experience, the number of companies the analyst follows, the number of industries the analyst follows, and the size of the analyst's brokerage house. To investigate this issue, I used (1) a two-factor PASTACC (past accuracy) model consisting of forecast age and past accuracy and (2) a six-factor ANCHAR (analyst characteristics) model consisting of forecast age and the five analyst characteristics. I explored how these two models perform relative to each other.I used both quarterly and annual data and measured performance in two ways—estimation and prediction. The data are from the Thomson Financial I/B/E/S U.S. Detail file of annual and quarterly analyst earnings forecasts for the 13-year period 1986–1998; I used the last forecast made by each analyst prior to the earnings' announcement.I assessed the estimation performance of the models by comparing their adjusted R2s. The model with the greater explanatory power I considered the better model for estimation purposes. I used 12 years of data, for 1987–1998, based on the parameter estimates of each prior year to obtain predictive results. Using the estimated values of the dependent variable, I identified the two extreme deciles representing analysts expected to be the “most accurate” and “least accurate” earnings forecasters. I determined the mean actual values of the dependent variable in the extreme deciles to ascertain these same analysts' actual forecast accuracy, and I evaluated comparative predictive performance by determining which model was better at identifying actual analyst performance.I show that the PASTACC model performs as well as the ANCHAR model for both quarterly and annual data and in both the estimation and prediction tests. Therefore, the response to the query “how important is past analyst earnings forecast accuracy?” is that, when combined with forecast age, past accuracy is as important as the five analyst characteristics of forecast accuracy I examined.My evidence suggests that the Street's focus on past accuracy is not misplaced. Practitioners who wish to create weighted consensus estimates by using detailed analyst forecasts that are more accurate than a simple (unweighted) consensus can do just as well by using a two-factor model encompassing forecast age and past accuracy as they can using a six-factor model composed of forecast age and the five intuitively appealing analyst characteristics I examined.Foreknowledge of analyst forecast accuracy is valuable. Researchers have shown that “knowing” how likely the accuracy of a forecast of an upcoming quarterly earnings number is and buying stocks on the basis of whether a weighted consensus estimate exceeds a simple consensus estimate can be profitable. Of course, no model allows an investor or manager to “know” the future, but my evidence suggests that one is likely to make as much money using a simple model based on forecast age and past accuracy as one would by using a complex model based on forecast age and five analyst characteristics.