The information Laundromat is beating the best minds of Wall Street -- but nobody really knows how.

Published October 26, 1999 4:00PM (EDT)

Imagine getting a hot stock tip in a conversation with a friend. The friend tells you that the tip comes from a number of "inside sources." At least he thinks so. But he's not really sure because he himself gets his tips in the mail, in unmarked envelopes. He doesn't know who his sources are, where their information comes from or why they're distributing the hot tips. Many of his informants don't even agree with each other. But he's used these tips before, and often they've worked. He's not really sure why.

Do you trust the tip?

I found out about from a story on the financial news site CBS Marketwatch. Last week, IBM reported its earnings, and they came in a little above what analysts had forecast, 93 cents a share instead of 90. Marketwatch duly reported that IBM had beaten the analysts' projections, but more intriguing was the report that it had also beaten the "whisper number" of 91 cents a share -- a number that Marketwatch attributed to

"Whisper numbers" have for some time been a subject of debate on Wall Street. Routinely, company profits exceed the "official" expectations of investment bank analysts. They almost have to: Companies that merely meet expectations are pummeled by the marketplace. The "whisper number" is a mysterious beast, a number leaked by insiders and traded furtively by analysts -- the "real" number that reflects what the big players on the Street really think a company's earnings will be for the quarter.

The details of the earnings forecasts don't matter that much -- in this case, after all, analysts' guesses and the "whisper number" weren't really that far apart (though, for IBM a difference of a penny a share in earnings does translate into millions of dollars). The really interesting thing was that a site called "" sounds a lot like "" or "" -- just a little too good to be believed. If the information was that good, where did it come from and why would anybody put it online? And is it really that good?

It turns out that the answers are a lot more complicated than one might expect. It turns out that the data is both surprisingly useful and troublingly mysterious. is a stunning case study in the genealogy of information. You can think of it as the ultimate information Laundromat: Information from users and the rest of the Web gets thrown into a machine, washed clean of its origins and air-dried into a single consensus "whisper number." That number then gets picked up by a major news source -- in this case, CBS Marketwatch -- whereupon it turns into still one more piece of data that number-starved investors grab at to get a little edge in their trading., the site, is barely 5 months old, founded in June by Paul Hauck and John Scherr, two former Dunn & Bradstreet marketing directors. It's currently owned by the Internet Financial Network, itself partially owned by the huge financial services provider CitiGroup. But, the phenomenon, is something much more than just another Web site aimed at capturing the eyeballs of day traders and other information-hungry investors.'s murky, untraceable numbers may actually be more accurate than the "official" numbers -- a disturbing fact that throws into question our whole understanding of how to value "information."

Here's how it works: begins by using a search engine to comb Internet message boards, published articles and chat rooms for whisper numbers. In addition, visitors to the site can add to the mix, putting in their own whisper numbers for the companies whose performance they follow closely. then posts the resulting numbers. In some cases, the site draws on a great many sources -- in the case of America Online, 254. In others, the whisper number can be drawn from a small handful of mentions.

The reason all this matters is that immediately after a company reports its earnings, the stock price tends to move. Missing projections by just pennies a share can mean the stock price plummets. Beating them by a lot can mean the stock goes up. A difference of even a few pennies per share matters, because the earnings reflect how close the company is to meeting its goals and whether it justifies the investment pros' expectations for the future.

The strange thing about the whisper numbers that float on the Web is that, in many cases, they are eerily accurate -- according to a paper published by three business professors in the Journal of Accounting & Economics that surveyed whisper numbers posted on Internet discussion boards between January, 1995 and May, 1997. The professors found that the Net rumors were actually more effective at predicting company earnings than analyst forecasts.

One explanation is that analyst estimates tend to be too low -- CEOs are worried about missing their numbers and downplay expectations -- and the whisper numbers are usually higher. (Susan Watts, one of the authors of the study, said that FirstCall, the information service that distributes analyst estimates, is already looking into adjusting the numbers for this downward bias). But even when the whisper numbers are lower than analysts' estimates, the study shows that they are still more accurate.

The unexpected accuracy of the whisper numbers is strange indeed -- a lot like random surveys of political discussion sites regularly predicting elections better than network news polls.

Especially interesting to investors are stocks that beat analysts' official expectations but fall short of the whisper number. Often these stocks go down when their earnings are announced even though the company meets the official projections. The accuracy of using the "whisper number" to predict this kind of move can vary wildly, but overall, according to's own studies, they fall almost 70 percent of the time.

Something definitely appears to be going on, but it's hard to put your finger on what. Paul Hauck, one of the founders of, says that some of the site's heaviest usage comes from computers at investment banks. It's also likely that many Web posters are repeating information they read somewhere else. And many of the whisper numbers found on the Net are likely to come from research and guesswork generated by individual investors.

"If self-directed investors want to enter whisper numbers, they've probably done their homework," says Hauck. "They are providing good financial opinions. They have the right to give opinions, because of their role in the financial markets."

This is an odd way to look at information. After all, whisper numbers are supposedly the "insider" numbers passed around Wall Street. The right to give an opinion doesn't seem to correspond with the accuracy of an earnings estimate.

But, on the other hand, the proof is ultimately in the results. If the whisper numbers collected from the Web work -- that is, if they are more accurate than analysts' estimates and can be used to predict how a company's stock will perform, does it really matter where they come from?

Clearly, the thought that information might be bubbling up from a mysterious well, a collection of Internet sources impossible to trace, leaves nobody feeling fully comfortable. Information coming from a few known sources is much less scary than a soup whose composition no one quite understands.

Even Thom Calandra, the editor-in-chief of Marketwatch, which references's figures, isn't entirely comfortable with the data. "Whisper numbers," says Callandra, "are nebulous in the first place. Since they're based on unofficial communications [from analysts], I'd generally feel more comfortable if a reporter got them from speaking to some analysts than from a Web site."

It is likely that a lot of readers will share Calandra's unease. It is easy, for instance, to envision a situation in which's results could be manipulated. While often has many sources -- "whispers" -- for technology company earnings, big industrial companies are discussed on the Web much less frequently;'s whisper number for Ford, for instance, was compiled from just five whispers. Just a handful of people can change the projections for a multi-billion dollar corporation.

That said, the fact that's numbers might be susceptible to manipulation, or even the fact that nobody seems to know exactly where the information that goes into them originates, does not invalidate them. Actually, there is more hard evidence for their usefulness than for that of a great deal of hocus-pocus stock analysis.

"Investors are starting to see the information they get from analysts as biased and skewed," argues Hauck, explaining why users go to his site. "They're giving you what the company wants you to have."

He has a point. Investors are increasingly unwilling to take at face value information that comes from conventional sources and the usual suspects, and the numbers back their suspicions. And if Hauck's numbers work, who is to say that investors shouldn't use them?

Still, it's an unusual circumstance: We are used to trusting information whose provenance we know, and discarding information whose history we don't know. Yet here, investors have one source of information they do not trust -- the investment analysts -- and another that seems to work, but that no one quite understands. Consider it a metaphor for the expanding information economy. Like a sophisticated machine, as the information available to us gets richer and more complex, it paradoxically becomes both easier to deploy and harder to dissect.

By Mark Gimein

Mark Gimein is a staff writer for Salon Technology.

MORE FROM Mark Gimein

Related Topics ------------------------------------------

Stock Market