A year after Erik Lie, a finance professor at the University of Iowa, uncovered the widespread practice of options backdating, three academics have uncovered circumstantial evidence of what appears to be either: a) another example of Wall Street-types trying to fix results ex post facto; b) a significant mishandling of data by a company that sells data for a living; or c) a big misunderstanding.
Felicia Marston of the University of Virginia, Christopher Malloy of London Business School, and Alexander Ljungqvist of New York University last month published a paper that appears to call into question the integrity of I/B/E/S, a huge database of stock upgrades and downgrades compiled and published by Thomson Financial. For example, it would record that on March 17, 2000, Joe Analyst of Morgan Stanley raised IBM from "buy" to "sell." I/B/E/S, which is used by the Wall Street Journal to compile analyst rankings, serves the same function for Wall Street as Major League Baseball statistics do for rotisserie league players.
I/B/E/S is also widely used by academics. Back in 2002, in the wake of the Wall Street research scandals, Ljungqvist, Marston, and a third researcher, William Wilhelm, looked into the question of whether investment banks issued biased investment advice for the purpose of winning business with corporate clients. The paper, published in the Journal of Finance, was based on data from the I/B/E/S database between 1993 and 2002. In 2004, Ljungqvist, Marston, and Malloy returned to the same 1993-2002 records and "noticed there was a difference in the data," said Ljungqvist.
To be specific, they noted four types of changes in the database, which consisted of 280,463 analyst actions. Most significantly, analysts' names had apparently been removed from about 20,000 recommendations. When this was brought to Thomson's attention, a spokesperson told the Financial Times last November that, in fact, the names were in the database, but that "they were not visible on the files seen by the academics due to an incomplete data feed." According to Ljungqvist, I/B/E/S has since been corrected to undo these anonymizations.
But the professors have not received what they regard as satisfactory answers to questions they raised about the other three types of changes. For example, there were 10,698 alterations between 2002 and 2004—records where a recommendation might have been changed from a "buy" in 2002 to a "hold" in 2004—as well as 4,923 deletions and 19,204 additions.
Now, any database can wind up with errors that can be cleaned up through routine system maintenance. Think how many of your holiday cards are returned because of wrong addresses. In theory, such database errors should be random. But the professors found that the changes weren't random. The changes generally appeared to make analysts seem smarter—less bullish during the tail end of the boom, for example—than they were in the first iteration. Most of the additions were in the "hold" and "sell" categories. The authors note: "[I]n the case of one prominent brokerage firm, 91.5 percent of its 234 additions are sells, and these increase the number of sells the firm has on the 2002 tape by a factor of 20." Meanwhile, the deletions were "disproportionately strong buys." While the additions, alterations, and deletions affected only a minority of the data, the end result was to make the analysts' advice seem substantially better. (To me, the biggest mystery is the 20,000 additions.)
The professors take pains to note that they are not accusing anyone of malfeasance, and they don't hypothesize about how such large-scale changes could have been made. "I'm not alleging anything, it's just that is what is in the data," said Ljungqvist.