Wall Street Analysts

Excerpted from A Random Walk Down Wall Street, 7th Edition, by Burton G. Malkiel.  W. W. Norton & Company, 1999.  Pages 166-171 from Chapter 7 of the hardcover edition.

Are Security Analysts Fundamentally Clairvoyant?

Forecasting future earnings is the security analysts' raison d'être.  As a top Wall Street professional put it in his fraternity magazine, Institutional Investor:  “Expectations of future earnings is still the most important single factor affecting stock prices.”  As we have seen, growth (in earnings and therefore in the ability to pay dividends or to engage in stock buy-backs) is the key element needed to estimate a stock's firm foundation of value.  The analyst who can make accurate forecasts of the future will be richly rewarded.  “If he is wrong,” Institutional Investor puts it, “a stock can act precipitously, as has been demonstrated time and time again.  Earnings are the name of the game and always will be.”

To predict future directions, analysts generally start by looking at past wanderings.  “A proven score of past performance in earnings growth is,” one analyst told me, “a most reliable indicator of future earnings growth.”  If management is really skillful, there is no reason to think it will lose its Midas touch in the future.  If the same adroit management team remains at the helm, the course of future earnings growth should continue as it has in the past, or so the argument goes.

Such thinking flunks in the academic world.  Calculations of past earnings growth are no help in predicting future growth.  If you had known the growth rates of all companies during, say, the 1980-90 period, this would not have helped you at all in predicting what growth they would achieve in the 1990-2000 period.  And knowing the fast growers of the 1990s will not help analysts find the fast growers of the early twenty-first century.  This startling result was first reported by British researchers for companies in the United Kingdom in an article charmingly titled “Higgledy Piggledy Growth.”  Learned academicians at Princeton and Harvard applied the British study to U.S. companies – and, surprise, the same was true here!

“IBM,” the cry immediately went up, “remember IBM.”  I do remember IBM:  a steady high grower for decades.  For a while it was a glaring exception.  But after the mid-1980s, even the mighty IBM failed to continue its dependable growth pattern until growth resumed in the late 1980s under a new management.  I also remember Polaroid, Apple Computer, and dozens of other firms that chalked up consistent large growth rates until the roof fell in.  I hope you remember not the current exceptions, such as Microsoft, but rather the rule:  There is no reliable pattern that can be discerned from past records to aid the analyst in predicting future growth.

A good analyst will argue, however, that there's much more to predicting than just examining the past record.  Rather than measure every factor that goes into the actual forecasting process, John Cragg and I decided to concentrate on the end result:  the prediction itself.

Donning our cloak of academic detachment, we wrote to nineteen major Wall Street firms engaged in fundamental analysis.  The nineteen firms, which asked to remain anonymous, included some of the major brokerage firms, mutual-fund management companies, investment advisory firms, and banks engaged in trust management.  They are among the most respected names in the investment business.

We requested – and received – past earnings predictions on how these firms felt earnings for specific companies would behave over both a one-year and a five-year period.  These estimates, made at several different times, were then compared with actual results to see how well the analysts forecast short- run and long-run earnings changes.  The results were surprising.

Bluntly stated, the careful estimates of security analysts (based on industry studies, plant visits, etc.) do little better than those that would be obtained by simple extrapolation of past trends, which we have already seen are no help at all.  Indeed, when compared with actual earnings growth rates, the five-year estimates of security analysts were actually worse than the predictions from several naive forecasting models.

For example, one placebo with which the analysts' estimates were compared was the assumption that every company in the economy would enjoy a growth in earnings approximating the long-run rate of growth of the national income.  It often turned out that if you used this naive forecasting model, you would make smaller errors in forecasting long-run earnings growth than by using the professional forecasts of the analysts.

Our method of determining the efficacy of the security analyst's diagnoses of his companies is exactly the same as was used before in evaluating the technicians' medicine.  We compared the results obtained by following the experts with the results from some naive mechanism involving no expertise at all.  Sometimes these naive predictors work very well.  For example, if you want to forecast the weather tomorrow you will do a pretty good job by predicting that it will be exactly the same as today.  Although this system misses every one of the turning points in the weather, for most days it is quite reliable.  How many weather forecasters do you suppose do any better?

When confronted with the poor record of their five-year growth estimates, the security analysts honestly, if sheepishly, admitted that five years ahead is really too far in advance to make reliable projections.  They protested that although long-term projections are admittedly important, they really ought to be judged on their ability to project earnings changes one year ahead.

Believe it or not, it turned out that their one-year forecasts were even worse than their five-year projections.  It was actually harder for them to forecast one year ahead than to estimate long-run changes.

The analysts fought back gamely.  They complained that it was unfair to judge their performance on a wide cross section of industries, because earnings for electronics firms and various “cyclical” companies are notoriously hard to forecast.  “Try us on utilities,” one analyst confidently asserted.  So we tried it and they didn't like it.  Even the forecasts for the stable utilities were far off the mark.  Those the analysts confidently touted as high growers turned out to perform much the same as the utilities for which only low or moderate growth was predicted.  This led to the second major finding of our study:  Not one industry is easy to predict.

Moreover, no analysts proved consistently superior to the others.  Of course, in each year some analysts did much better than average, but no consistency in their pattern of performance was found.  Analysts who did better than average one year were no more likely than the others to make superior forecasts in the next year.

My findings with Cragg have been confirmed by several other researchers.  For example, Michael Sandretto of Harvard and Sudhir Milkrishnamurthi of M.I.T. completed a massive study of the one-year forecasts of the 1,000 most widely followed companies.  Estimates were available from five or six analysts for each company.  The staggering conclusion of the study was that the average annual error of the analysts was 31.3 percent over a five-year period.  The error rates each year were remarkably consistent – the lowest error rate was 27.6 percent, the highest 33.5 percent.  Financial forecasting appears to be a science that makes astrology look respectable.  Amidst all these accusations and counterassertions is a deadly serious message.  It is this:  Security analysts have enormous difficulty in performing their basic function of forecasting earnings prospects for the companies they follow.  Investors who put blind faith in such forecasts in making their investment selections are in for some rude disappointments.

Why the Crystal Ball Is Clouded

It is always somewhat disturbing to learn that a group of highly trained and well-paid professionals may not be terribly skillful at their calling.  Unfortunately, this is hardly unusual.  Similar types of findings could be made for most groups of professionals.  There is, for example, a classic example in medicine.  At a time when tonsillectomies were very fashionable, the American Child Health Association surveyed a group of 1,000 children, eleven years of age, from the public schools of New York City, and found that 611 of these had had their tonsils removed.  The remaining 389 were then examined by a group of physicians, who selected 174 of these for tonsillectomy and declared the rest had no tonsil problem.  The remaining 215 were reexamined by another group of doctors, who recommended 99 of these for tonsillectomy.  When the 116 “healthy” children were examined a third time, a similar percentage were told their tonsils had to be removed.  After three examinations, only 65 children remained who had not been recommended for tonsillectomy.  These remaining children were not examined further because the supply of examining physicians ran out.

Numerous studies have shown similar results.  Radiologists have failed to recognize the presence of lung disease in about 30 percent of the X-ray plates they read, despite the clear presence of the disease on the X-ray film.  Another experiment proved that professional staffs in psychiatric hospitals could not tell the sane from the insane.  The point is that we should not take for granted the reliability and accuracy of any judge, no matter how expert.  When one considers the low reliability of so many kinds of judgments, it does not seem too surprising that security analysts, with their particularly difficult forecasting job, should be no exception.

There are, I believe, four factors that help explain why security analysts have such difficulty in predicting the future.  These are (1) the influence of random events,  (2) the creation of dubious reported earnings through “creative” accounting procedures,  (3) the basic incompetence of many of the analysts themselves, and  (4) the loss of the best analysts to the sales desk or to portfolio management.

Back