Aurora Borealis
1 August 2013 CFA Institute Journal Review

A New Approach to Predicting Analyst Forecast Errors: Do Investors Overweight Analyst Forecasts? (Digest Summary)

  1. Victoria Rati

The author explores sell-side analysts’ forecasts of company earnings and compares them with unbiased forecasts of company earnings derived from company characteristics. He finds there are predictable errors in the analyst forecasts but that investors fail to fully take them into account, relying more heavily on less-accurate analyst forecasts. These valuation errors can be successfully exploited.

What’s Inside?

The author summarizes a number of research projects that document the existence of errors or biases in analysts’ earnings forecasts and then explores how much weight investors place on such forecasts, which are known to be flawed, and what the ramifications are of their reliance.

The author calculates unbiased estimates of future company earnings using individual company characteristics and compares these characteristic forecasts with analyst forecasts. He finds that analysts are slow to incorporate into their forecasts new facts that can be gleaned from a firm’s characteristics. The analyst forecasts thus include predictable errors.

Despite these predictable errors, investors seem to rely more heavily on analyst forecasts and systematically underweight the more-objective characteristic forecasts. As a result, it is possible to build an investment strategy to exploit these observations. The author builds a long–short strategy that consistently earns abnormal returns of 5.8% a year in out-of-sample testing.

How Is This Research Useful to Practitioners?

When the characteristic forecast for an individual company exceeds the consensus analyst forecast, the company’s realized earnings also exceed the consensus analyst forecast. When a difference exists between the characteristic and the analyst forecast for a company’s earnings, the analyst forecast gets revised over time (i.e., in the lead-up to the earnings announcement) in the direction of the characteristic forecast. Thus, the gap between the two forecasts shrinks.

When the characteristic forecast exceeds the analyst forecast for a company, analysts are more likely to revise their investment recommendation upward. Characteristic forecasts have a significantly positive relationship with future returns, and analyst forecasts are negatively related to future returns. The characteristic forecast error is insignificantly different from zero (i.e., the characteristic forecast is an unbiased and solid predictor of the earnings result).

The author’s findings suggest that characteristic forecasts are better predictors of company earnings results than analyst forecasts. Thus, it is possible to build an investment strategy to exploit these observations. A long–short strategy using these findings consistently earns abnormal returns of 5.8% a year in out-of-sample testing, and the results are highly statistically significant. For firms in which the share price is more highly sensitive to earnings announcements, the strategy return increases to 9.4% a year.

When the strategy focuses on small firms, firms with historically disappointing earnings, or firms with low transparency (poor information environments), the characteristic forecast is an even stronger predictor of future realized returns.

These findings will be of interest to equity portfolio management practitioners, including chief investment officers, heads of equity, portfolio managers, and analysts (both qualitative or fundamental and quantitative investors). They are useful in understanding and evaluating analyst forecasts. Also, an equity strategy that uses the characteristic forecasts and strips out the analyst errors can offer enhanced portfolio returns.

How Did the Author Conduct This Research?

The author creates unbiased forecasts of future company earnings and compares them with analyst forecasts using a new approach. To create the unbiased forecasts, he uses the current characteristics of firms. According to the author, this cross-sectional estimation approach is new in this field of research. It is different from the traditional time-series regression analysis used in analyst forecast research. Traditional regression analysis focuses on fitting statistically derived equations to past analyst forecast errors and then using that relationship to forecast future errors. The regression analysis, however, has estimation biases that are eliminated in the estimation approach.

To derive the characteristic forecasts of future earnings, the author uses the latest characteristics of a firm. He uses the same firm characteristics that Fama and French documented (Journal of Business 2000): lagged earnings, book values, accruals, asset growth, dividend, and price. From the latest characteristics data available (historical data), coefficients for each characteristic are estimated and used to forecast future earnings. Data are sourced from the Compustat database.

Analyst forecasts are from the Institutional Brokers’ Estimates System (IBES) database. Data from the Compustat and IBES databases are combined with monthly return data from CRSP. The final sample consists of 51,591 data points (firm-years) from 1980 to 2009.

Abstractor’s Viewpoint

Given the importance to equity portfolio management of accurately forecasting stock returns, the author’s findings regarding the predictable errors within analyst forecasts is highly relevant. Even though the existence of analyst forecast biases has been well documented, the predictability of these biases is very interesting because it seems to have been ignored by the market. As a result of the discrepancies between the characteristic forecasts and analyst forecasts, the author finds that it is possible to build a long–short equity strategy (before transaction costs) using the unbiased characteristic forecasts to successfully exploit this differential. Given the annual transactions, trading costs are not judged to be material.

We’re using cookies, but you can turn them off in Privacy Settings.  Otherwise, you are agreeing to our use of cookies.  Accepting cookies does not mean that we are collecting personal data. Learn more in our Privacy Policy.