Sizing Up the Analysts' Stock Picks and Earnings Forecasts

The math gets a bit tedious, but it's all for a good cause: finding the analysts who best serve investors when they recommend stocks or predict corporate earnings.

The Wall Street Journal's All-Star Analysts Survey ranks eligible securities analysts on the basis of their stock-picking and earnings-forecasting skills. For stock picking, an analyst's performance is measured by estimating the return an investor would have achieved following his or her recommendations. Earnings forecasting is evaluated separately by measuring how accurately each analyst predicted the earnings of the companies he or she followed.
In both cases, the survey's methodology is designed to be objective, accurate and fair. The calculations were performed by Zacks Investment Research Inc., Chicago.

Stock-Picking Skill
Each analyst in the 1999 survey was judged on a hypothetical portfolio that included every eligible stock within a given industry category designated as a "buy" during 1998. Although some firms have several gradations of buy recommendations -- using terms such as "attractive," "accumulate," "outperform" or "strong buy" -- the survey treated all buy recommendations equally.

The performance of the individual stocks in the portfolio was measured for only that part of 1998 when the analyst recommended them. Because a "hold" recommendation has long been considered by many people on Wall Street to be a euphemism for "sell," hold recommendations were treated as sells.

All stocks in an analyst's portfolio were given equal weight. That is, it was assumed that the portfolio started with an equal amount of money invested in each of the stocks an analyst recommended. The holdings were rebalanced -- with the portfolio's then-current value divided equally among all stocks recommended -- quarterly and whenever a stock was added or subtracted.

If an analyst was actively covering an industry but had no buy recommendations at a particular time, the hypothetical portfolio was invested in Treasury bills until the analyst did recommend a stock. This practice is designed to reward analysts who correctly warn investors away from a slumping sector.

Results are total returns, including capital gains (or losses) and dividends, which were assumed to be reinvested in the portfolio at large. Although whole numbers are shown in the accompanying tables, the rankings reflect results that were carried out to several decimal places; there were no ties.

For those who want to know still more, and have the patience to match their curiosity, here is a simple example:
Consider an analyst who started 1998 with a buy recommendation on Consolidated Widget. Its shares rose 4% through Feb. 15. On that day, the analyst downgraded Consolidated Widget and recommended zap-widget.com. Shares of zap-widget rose 25% between then and March 31 and another 50% through June 30. On June 30, the analyst downgraded zap-widget to a "hold" -- thereby sidestepping an 88% plunge in the fledgling company's share price -- and again recommended Consolidated Widget. Consolidated Widget rose 7% through Sept. 30 and another 4% through year end. The analyst was still recommending Consolidated Widget at the end of the year. He made no other recommendations, and these companies didn't pay dividends.

To figure the portfolio performance, start the year with $1,000 and multiply it by 1.04 to reflect Consolidated Widget's gain through Feb. 15. Then multiply the result ($1,040) by 1.25 to reflect zap-widget's gain through March 31. That result ($1,300) is then multiplied by 1.5 to reflect the stock's gain through June 30. The value of the portfolio at that time would be $1,950. That amount is then multiplied by 1.07 to reflect Consolidated Widget's rise through Sept. 30, and the result ($2,086.50) is multiplied by 1.04 to reflect the stock's gain through year end. The ending value of the portfolio would be $2,169.96 -- 117% more than at the beginning of 1998.

Got it? The important thing to remember is that the All-Star survey measures the aggregate gain of each analyst's hypothetical portfolio, not the average gain of the stocks or the sum of the individual gains.

Earnings-Estimate Accuracy
Analysts' skill in predicting corporate earnings was measured by comparing the accuracy of their earnings estimates with the accuracy of the estimates of other eligible analysts. (In the case of real-estate investment trusts, the relevant measure typically was "funds from operations." FFO -- which is calculated by taking net income and adding back depreciation, amortization expenses and nonrecurring and extraordinary items -- generally is considered the most useful performance measure for real-estate companies.) By using the relative error score, the survey doesn't penalize an analyst for following stocks with earnings that are hard to estimate.

To understand how this works, take the hypothetical analyst in the stock-picking example. If the analyst predicted Consolidated Widget would earn $2.15 a share in 1998 and actual earnings were $2.50 per share, the analyst's error was 35 cents a share. Whether that error is good or bad depends on how other analysts were doing.

Say this hypothetical analyst is one of four in the survey who covered the industry in 1998; the estimated earnings of two of the others would be $3 a share, while the fourth expected earnings of $3.15. Thus, the estimated errors of the four widget analysts would be 65 cents, 50 cents, 50 cents and 35 cents. (Under the survey's system, it doesn't matter whether an error is on the side of optimism or pessimism. Only the size of the error matters.) The average, or consensus, error would be 50 cents, and the hypothetical analyst's error would be 70% of the consensus error.

So in this example, the analyst would have an error-factor score of 0.70. An error factor of 1.00 is exactly as accurate as the consensus; an error factor of 0.50 indicates that an analyst's error was only half that of the consensus error. In reality, this multistep calculation was performed numerous times for every analyst in the survey. For each company, analysts were scored on the basis of 12 estimates of 1998 earnings -- those made at the end of December 1997, January 1998, and so on through November 1998.

The individual scores for each company were averaged together to form a composite score covering all the eligible companies on which the analyst prepared estimates.
--C. Frederic Wiegold

Return to Listing