Analyst Forecasting Ability

Analyst Forecasting Ability

A study by Elton, Gruber, and Gultekin demonstrated the value of accurate forecasts. 12 They found that if an investor was armed with perfect information about the growth of earnings that would occur, an investor could have generated significant abnormal positive returns.

12 Edwin Elton, Martin Gruber, and M. Gultekin, “Expectations and Share Prices,” Management Science (September 1981).

FINANCIAL STATEMENT ANALYSIS

Given the value of good forecasts, let’s look at how well analysts do in forecasting EPS.

There has been extensive research dating back to the late 1960s that has investigated how well analysts do in forecasting earnings. We’ll look at some recent evidence on the subject. Before we discuss the recent evi- dence, we must first define how to measure the forecast error.

A simple procedure is to look at the difference between the actual EPS and the forecasted EPS. The latter is measured by the consensus EPS. The result is a measure of the earnings surprise. The problem with using this measure of earnings surprise is that it does not take into account the severity of the error based on the level of EPS. For example,

a $0.02 difference between the actual and consensus EPS is more signif- icant for a company with actual EPS of $0.20 than it is for a company with actual EPS of $10. Thus, the dollar difference in the error must be deflated or standardized by the level of EPS. Two measures have been used by researchers. Earnings surprise can be standardized by dividing by either actual EPS or consensus EPS. In fact, because researchers want to know the bias of the forecast error (overestimate or underestimate), the practice is to divide by the absolute value of the actual EPS or con- sensus EPS. That is, the forecast error can be measured in either of the following ways:

Actual EPS Consensus EPS –

Forecast error = -------------------------------------------------------------------------------------

Absolute value of the Actual EPS or,

Actual EPS Consensus EPS –

Forecast error = -----------------------------------------------------------------------------------------------

Absolute value of the Consensus EPS The forecast errors as measured above are also referred to as a measure

of “standardized earnings surprise.” Using the Qualcomm example, the two measures of forecast error are:

Actual EPS Consensus EPS –

Forecast error = -------------------------------------------------------------------------------------

Absolute value of the Actual EPS $0.33 $0.26 –

= ------------------------------------ = 21.21%

and

Earnings Analysis

Actual EPS Consensus EPS –

Forecast error = -----------------------------------------------------------------------------------------------

Absolute value of the Consensus EPS $0.33 $0.26 –

= ------------------------------------ = 26.92%

To assess the forecasting ability of analysts, researchers then analyze these forecasting errors by looking at the mean absolute forecasting error and the proportion of the sample of forecasts outside of practical error bands (e.g., the percentage of forecasts that fall outside a plus or minus a 10% interval around the actual earnings).

Recent studies by David Dreman and Michael Berry 13 and by Lawrence Brown 14 have examined the ability of analysts to forecast quarterly EPS and whether or not there is a bias in analyst forecasts. Dreman and Berry found that the average forecast errors are too high— more than 20% when not standardized and double that amount when standardized. They also found that when a 10% forecast band is used, more than half of the forecasts were outside the band. Dreman and Berry used other bands but the 10% figure is what they state is “a level that many Wall Street professionals consider minimally acceptable.” 15 Moreover, they find that forecasts overestimate actual earnings. That is, analysts tend to be optimistic about a firm’s future earnings. In conclud- ing their study, they write:

The observed frequency, size, and increasing trend of all of the error metrics for quarterly estimates bring into ques- tion many important methods of stock valuation, which rely on precise earnings estimates sometimes years into the future. The growth, earnings momentum, discounted cash flow, and earnings yield techniques, for example, require fine-tuned estimates often a decade or more into the future. Thus, a significant portion of current security anal- ysis requires a precision in earnings forecasts that is increasingly difficult for analysts to meet. 16

13 David N. Dreman and Michael A. Berry, “Forecasting Errors and Their Implica- tions for Security Analysis,” Financial Analysts Journal (May/June 1996), pp. 30–41.

14 Lawrence D. Brown, “Analyst Forecasting Errors: Additional Evidence,” Financial Analysts Journal (November/December 1997), pp. 81–88.

15 Dreman and Berry, “Forecasting Errors and Their Implications for Security Anal- ysis,” p. 39

16 Dreman and Berry, “Forecasting Errors and Their Implications for Security Anal- ysis,” p. 39.

FINANCIAL STATEMENT ANALYSIS

The database used in the Dreman-Berry study was the Abel-Noser database. This database uses information from Value Line, I/B/E/S, Zacks Investment Research, and First Call. The potential problem with such a database is that providers define actual earnings and forecasted earnings differently and as a result this could make the forecast errors

larger than they actually are. 17 In a study published a year after the Dreman-Berry study, Brown reexamined the ability of analysts to fore- cast earnings relying only on the I/B/E/S database. He also reported results using the Abel-Noser database used in the Dreman-Berry study. Brown found that for the two databases, the results supported the posi- tion that analyst forecasting errors are large. In addition, he finds that there is an optimistic bias in the forecasts.

Brown extended his investigation to determine whether the types of firms that analysts follow have an effect on their forecasting ability. Brown examined this question by looking at analyst forecasts based on the following firm-specific factors: whether a firm is included in the S&P 500, market capitalization, the absolute value of earnings forecast, and analyst following. He found that for firms in the S&P 500, the forecast- ing errors are smaller compared to firms not in the S&P 500. For firms with comparatively large capitalization, absolute value of earnings fore- cast, and analyst following, the forecasting error was relatively small. He continued to observe an optimistic bias. When Brown investigated analyst forecast errors for 14 industries, he found that the forecasting errors for some are substantially greater than for others.

There are other findings reported in the Brown study that warrant noting because they shed some light on other questions we raised earlier in this chapter regarding earnings management. Brown, as well as Dreman and Berry, found that the median and modial value of earnings surprise was zero, suggesting that analysts forecasts tended to be on tar- get (although the average forecast was too large relative to actual earn- ings). Brown found that the number of small positive errors was greater than the number of small negative errors. Based on this finding, Brown suggested that corporate managers may manage earnings so as to not fall below the consensus estimate. Moreover, Brown also found that the number of large negative errors was greater than the number of large positive errors. This finding sheds some light on the “big bath” observa- tion that we discussed earlier, whereby managers create large negative earnings surprises relative to the number of large positive earnings sur- prises.

17 D.R. Philbrick and W.E. Ricks, “Using Value Line and I/B/E/S Analysts Forecasts in Accounting Research,” Journal of Accounting Research (Autumn 1991), pp. 397–

Earnings Analysis