Right Now | Soar Like Soros
Stars, Bulls, and Bears
Picture an investor trying to decide which mutual fund will make the most of her retirement savings. As she reads prospectuses, the fine print and pie charts begin to blur before her eyes and all funds start to sound the same. She might choose based on reputation or past performance although, as every prospectus warns, "past performance is no guarantee of future results." Furthermore, track records say little about start-up funds that could be winners. To buy one of these unproven funds is, in essence, gambling.
For evaluating funds that are relatively new, "the existing methods are basically useless," says assistant professor of business administration Randolph Cohen. That's important because, according to the financial data firm Lipper Inc., 55 percent of all diversified and sector equity funds in the United States are five years old or younger.
The standard methods for rating mutual funds by their performance records date from the 1960s, and have had only minor modifications since then. Now, Cohen and Joshua Coval, associate professor of business administration, along with Lubos Pástor of the University of Chicago, have developed an algorithm that compares the decisions a fund manager makes with decisions made by all other managers, including those with proven track records. Their formula shows that if a fund holds the same stocks as other funds with a history of doing well, it is in good company and is likely to perform well itself.
Because even new funds normally hold dozens of stocks, creating an abundance of data points, the researchers say their method allows a degree of statistical confidence four to eight times higher than a traditional results-based analysis, depending on how the traditional approach adjusts its performance measures for risk.
Coval compares the method to a game of blackjack. After just a few hands, it's impossible to tell which of two players is more skilled, because neither has won much money yet. But suppose there is a third player at the table who has been playing for a while and has amassed a huge stack of chips. We can compare the strategies of the first two players to the one used by the big winner. The more skilled player is probably the one whose strategy most resembles that of the man who has beaten the house so far.
To some degree, this all seems obvious. Why wouldn't any blackjack player try to copy a winner? Why wouldn't a fund manager buy the same stocks held by a fund that usually gets good returns? In fact, "This paper started out with my asking why no one had done this," Coval says. As with any mathematical discovery, adds Cohen, "If it doesn't eventually seem obvious, then it probably isn't quite right." What's less obvious is that, according to the researchers, investors who followed their system from April 1977 to December 2000 would have made as much as 1 percent per year more than they would have by consulting track records alone.
The three scholars present their findings in an as-yet-unpublished paper (available on-line at http://ssrn.com/abstract_id=353620). Meanwhile, they are working on making their system available to the public. Pástor, for example, has given a presentation to Morningstar Inc., which rates mutual funds on its website. Like the researchers, Morningstar judges funds against others with comparable investment profiles, but unlike them, Morningstar would not assign a higher rating to a fund that had been outperformed by others of its type, however closely that fund's current holdings matched those of the top-ranked fund.
The ranks of successful fund managers include both innovators and copycats. In some cases, it doesn't matter whether the manager leads or follows. For cases where an assessment of a manager's true skill matters, however, the researchers developed a formula based on trading transactions that rates managers on the timing of their purchases and sales as well as on the list of stocks they hold. Most mutual funds publicly disclose their transactions quarterly, so if a manager picked a good stock before it was disclosed that other high-performing managers bought it, he gets credit for the decision.
If this new method emphasizes current holdings and recent transactions, downplaying the fund's performance history, could a fortunate winning streak sway the ratings? The researchers say that the new method actually uses more data points than the traditional one. Suppose a fund has been operating for five years. With the traditional rating method, there are essentially five data points. "You'd need hundreds of years of history to get a very accurate evaluation," Cohen points out. But when you analyze every transaction completed in those five years, your measure is much more precise. "When you look at the decisions that are being made, rather than the returns that are earned," says Coval, "you filter out luck."
Randolph Cohen e-mail address: firstname.lastname@example.org
Joshua Coval e-mail address: email@example.com