Estimate Revisions
Question 17.1. It is reasonable to expect that revisions in analysts' earnings forecasts can be used to attain above-benchmark returns.
b. Fiction.
The preceding chapters have shown (I hope unambiguously) that today's stock prices reflect an amalgamation of uncertain expectations—including uncertain estimates of future earnings. As more and more time passes, more and more of the uncertainty about coming earnings is resolved. For some companies, earnings expectations reach a level above the initial expectation; for other companies earnings expectations move to a lower level. As time advances, earnings expectations reflect ever more complete information. Investment returns follow the resolution of this uncertainty in either direction.
One way to think about what I have presented thus far is that I have taken two snapshots. The first snapshot shows what we know today. The second snapshot moves one year ahead and looks at how earnings and stock prices have changed. What happens to stock prices as time unfolds through the year is the fertile ground I now explore.
Earlier I illustrated my discovery and naming of an extremely important phenomenon—the concurrent earnings-change/return-change effect. Over the course of each of the past 25 years the companies with the worst earnings changes concurrently had the worst investment returns; the companies with the best earnings changes concurrently had the best investment returns.
The existence of this phenomenon shows (1) that the year-ahead earnings expectations that are embedded in today's stock prices are wrong,1 and (2) that over the course of the next 12 months the portfolios of the stocks with the best unfolding actual earnings changes will concurrently provide their owners with the best investment returns. Thus, the correct answer to Question 17.1 is "a"—fact. It is reasonable to study revisions in analysts' earnings estimates for clues to understanding this unfolding earnings-change/return-change process.
Recognizing the interest in studying analysts' earnings expectations, the Institutional Brokers Estimate Service (I/B/E/S)—then part of the brokerage firm Lynch, Jones & Ryan—in 1971 began collecting and selling the earnings forecasts of institutional equity analysts. In 1981 Zacks Investment Research began providing another source of analysts' earnings forecasts.
Over the intervening years researchers have tested virtually every permutation and combination of the usefulness of earnings forecasts, unfolding estimate revisions, differences between the prognostications of "star" versus "also-ran" analysts, and measures of "diffusion" between the number of estimates raised and lowered as well as techniques to assign more weight to more recent estimate revisions.
In the old days, I/B/E/S and Zacks Investment Research released their revised data on certain preannounced release dates. This was an era when messengers on motorcycles, motors racing, wearing leather jackets and with white silk scarves blowing in the wind, eagerly awaited the release of the latest tapes. Once passed to the messengers, the tapes were scurried into the night to be analyzed before the market opened the following morning.
Today motorcycle messengers are gone, replaced long ago by banks of multicolored monitors that stream up-to-the-second data to voracious analysts. The clues to the usefulness of these data are found among a long list of research studies.
Question 17.2. The so-called "cockroach theory"2 holds that once an analyst raises or lowers his or her earnings forecast more such revisions are likely to follow. (As, so the theory goes, "Once you see one cockroach more will follow.")
This theory is largely:
b. Fiction.
An early study by Eugene Hawkins, Stanley Chamberlin, and Wayne Daniel,3 using a database that contained earnings estimates for over 2,400 stocks made by more than 70 brokerage firms for each of the 24 quarters from March 1975 through December 1980, found that month-to-month percent changes in consensus estimates could be used to predict changes in stock prices.
In 1984 Edwin Elton, Martin Gruber, and Mustafa Gultekin4 found that analysts had a tendency to overestimate earnings growth for companies they believed would do well and underestimate earnings growth for companies they believed would do poorly. Later studies by Edwin Elton, Martin Gruber, and Mustafa Gultekin;5 and D. van Dijk,6 using larger samples of analysts' expectational data, showed that revisions of earnings forecasts could be used to predict future returns.
In 1985 Robert D. Arnott, chief executive officer at First Quadrant, reported, "Although the market is relatively efficient in discounting current consensus, the startling fact is that it seems to reflect essentially none of the information in recent shifts in consensus!"7
Patricia O'Brien has found that analysts' most recent forecasts are more accurate. Thus, when aggregating forecasts, eliminating the most out-of-date forecasts improves the accuracy of the overall forecast.8
In 1991 Dan Givoly and Josef Lakonishok9 found that news of revisions is absorbed slowly, giving rise to the possibility that investors who act upon this type of publicly available information can earn above-benchmark returns. Also in 1991 Scott Stickel,10 after studying 173,620 revisions for 1,465 companies by 1,869 analysts from 83 brokerage firms, reported that "prices continue to drift in the direction of the revision for about six months after the revision."
In 1998 I participated in an internal study of the estimate-revision effect using 20 years of quarterly data on the 500 largest stocks. The results of this study, summarized in Table 17.1, showed that the estimate-revision effect was very strong from 1977 through 1981. The 120 stocks in the upward-revision portfolio gained 6.4 percent; the 120 stocks in the downward-revision portfolio fell 5.8 percent. The effect decreased somewhat but remained strong from 1982 through 1991. During the five-year period between 1992 and 1996 the effect vanished.
TABLE 17.1 Relative Returns: Earnings Estimate Revisions (Quintiles)
5 years (1977 to 1981) 10 years (1982 to 1991) 5 years (1992 to 1996)
Note: Returns are annualized, using quarterly rebalancing. Source: Morgan Stanley Investment Management.
In 1994 Langdon Wheeler,11 president of Numeric Investors L.P., tested the usefulness of several estimate-revision measures to predict investment returns within a broad universe of stocks. He found that data on analysts' revisions of consensus earnings estimates were useful predictors of coming changes in stock prices in each of the years he studied. Further, he showed that a more accurate predictor of coming price changes could be created by assembling several measures—such as changes in the consensus estimate and the number of estimates raised or lowered in the last few months—into a combined score.
Finally, Wheeler tested the effectiveness of his score in predicting returns in different market environments and in different market sectors. He found that his score was most effective when used across all economic sectors and when applied to smaller-capitalization stocks.
As you might expect, many other researchers (and practitioners) have mined the historical estimate-revision data in hopes of finding more useful measures of the earnings forecasts that are embedded in the consensus aggregation.
Specifically, researchers have shown that more useful data can be extracted from the following: more recent forecasts (see Patricia O'Brien,12 Scott Stickel,13 and Lawrence Brown14); forecasts that deviate from the average forecasts (see Haim Mozes and Patricia Williams15); company preannouncements (see Sandip Bhagat16 and Leonard Soffer, Ramu Thiagaragan, and Beverly Walther17); forecasts by historically more accurate forecasters (see Parveen Shinka, Lawrence Brown, and Somnath Das18); and classifications of analysts as "leaders, followers, or rebels" (see Ronald Kahn and Andrew Rudd19).
Martin Herzberg, James Guo, and Lawrence Brown20 have found that a composite measure incorporating the newness of the forecast, deviation of recent forecasts from the consensus measure, and the better-than-average skill of particular forecasters provides more accurate forecasts.
Today aggregations of many of the aforementioned categories are available through StarMine. Collecting every estimate and recommendation compiled by Thomson Financial I/B/E/S and First Call, StarMine monitors the roughly 5,200 estimates that are published each day. In turn, this service identifies the persons whom it considers the top analysts, tracks estimate revisions by the top analysts, predicts earnings surprises, and ranks stocks by changes in analysts' sentiment.
Several of these researchers have found persuasive evidence that analysts nibble away at forecast revisions. One revision is, indeed, likely to be followed by another. Thus, the answer to Question 17.2 is "a"—fact.
Before turning from the subject of estimate revisions it is useful to contemplate the fate of a perfect forecaster in a world in which portfolio managers devour estimate-revision data. In Richard Bernstein's book Navigate the Noise: Investing in the New Age of Media and Hype21 he spins a wonderful story of a security analyst who is a great forecaster. She makes her full-year forecast on January 1. Over the course of the next 12 months her prognostications materialize exactly as she had forecasted.
Succumbing to the belief that each analyst's forecasts must be up-to-the-minute to be useful—with no interim dribble coming from the analyst of the "up a few cents" or "down a few cents" variety, this great forecaster might actually be dropped from some real-time databases because she has not updated her earnings estimate within a certain number of months.
Contrast Bernstein's great forecaster with one of her contemporaries down the hall who is continually revising his forecasts. His phone rings constantly with calls from investors who want to know the reasons for his most recent changes. When he calls the supposedly impossible to reach manager of a large mutual fund with a message like "Tell her I have revised my estimate on Dell," he always gets through.
Question 17.3. Imagine you are a Wall Street analyst who makes quarterly earnings estimates. You note that 12 other Wall Street analysts routinely make quarterly forecasts of ABC Widget's earnings.
Your firm meticulously tallies the number of telephone calls that you make to and receive from your firm's clients. Your "contact scorecard"—which is reviewed at compensation time—gives many more points to contacts that you make with the firm's largest clients—which, not incidentally, produce the largest percentage of the firm's commissions. (Such "contact scorecard" systems for security analysts are the norm with Wall Street firms.)
Which of the following statements are true?
a. Your earnings forecasts are less likely to be wrong if they "hide" near the 12-analyst consensus.
b. You know from experience that your telephone never rings (with clients' calls) when your estimate "hides" near the consensus.
c. Not yet enjoying "star" status on the Institutional Investor All-Star Analyst Team, you discover that an extremely high or low forecast entices your clients to call—and to accept your calls.
d. All of the statements are true.
The correct answer is "d"—all of the statements are true.
Question 17.4. If you are the analyst, what would you do?
a. Hide near the consensus for fear of being marred with a bad forecast.
b. Start your telephone ringing.
Many researchers assume the former—that Wall Street analysts operate in fear of making poor forecasts. But in my considerable experience working in Wall Street research departments, I believe many analysts "game" the system—placing their estimates where their telephones will start ringing.
Post a comment