When customers order goods from online stores or bid for items in Internet auctions, they enter a bond of trust: They pay in advance with the assumption that they will receive the products advertised, and that those products will arrive in good condition and will remain so during their advertised lifetimes. Disappointed buyers can take legal action against sellers they suspect of cheating them; but court cases can involve high costs and provide uncertain results.

A more effective incentive for sellers to act honestly comes from the paper trail. The buyer and seller can leave public feedback about every online transaction—feedback on which future buyers can rely to determine sellers’ reliability. But even that record carries no guarantee of accuracy; a single negative report can needlessly damage a seller’s reputation.

Can the situation be improved? Mehmet Ekmekci, an assistant professor of managerial economics and decision sciences at the Kellogg School of Management, believes it can. He has devised a system for rating transactions that selectively forgets some of the information about sellers’ behavior. “My rating system will provide the feedback, but not make it stick forever,” he explains. “The further back in time, the more it forgets, and the more new information it includes. I’m always keeping a certain amount of data, but that amount never reaches full learning.”

Censoring Information
Ekmekci proposes “a new mechanism that determines the information each short run player observes about the past play of the game. In particular, a central authority that observes the full sequence of past signals censors the information that the short run players observe. We show that it is possible to censor the information in such a way that enables the long run player to build reputation at all times.” In other words, the system pools ratings in such a way that the most recent ones carry the most weight.

“The main message is that a lot of information is not necessarily good,” Ekmekci explains. “The idea is that a bad hit shouldn’t harm you forever, because you then have no incentive to improve. After all, if your credit score goes really bad, does it mean that you are doomed for the rest of your life? My model is suggesting a way in which markets might be replenishing effects by forgetting information. The market intentionally forgets some of the information—something that has merit. My research says that this could be the optimal thing that we might have expected to observe anyway. I’m saying: ‘Look guys. Getting rid of some information might have some merit as well.’ ”

Inspiration from a Native Land
The inspiration for the project came from a much larger arena than interactions between individual buyers and sellers. Ekmekci became intrigued by the credit score of Turkey, his native land. “The score moves up and down very slowly,” he recalls. “The country can’t keep itself at the best rating because it will be punished right away if it does anything bad.” That led him to consider why teachers and professors use grades rather than exact scores—a B rather than a score of 84 and an A rather than 86, for example. “My work doesn’t apply to student grades, but it got me thinking about how the market reveals information,” he says. “I realized that, in a dynamic setting, incentives can be mitigated more easily by having a forgetful rating.”

Trained as a game theorist, Ekmekci applied the methodology of repeated games to help him understand how forgetting specific incidents can improve means of rating sellers’ reputations. The key to his system is the limited freedom that the forgetfulness gives sellers to compensate for single—or even multiple—poor interactions with buyers in the past. “Leaving the buyers with some residue of uncertainty will have huge implications on the sellers’ incentives; they’ll always try to look good,” he asserts. “Even in the worst cases, there’s always a chance that the seller can change things. The amount of uncertainty can be made very small so that there’s not much of a loss, but gains can be high for society.”

The Value of Uncertainty
Ekmekci began his project with two goals: providing maximum benefit to both sellers and buyers, and understanding how to enhance the usefulness of information about interactions between them. “My contribution is that in the process of learning, if we leave a residual amount of uncertainty it will be even better for people,” he summarizes. “If you punish sellers for a long time, you’ll reduce any incentive to improve.”

Ekmekci’s system is strictly theoretical. However, the general field of rating reputations has already stimulated active experimentation. “There are growing amounts of work in the economics literature about rating systems,” he explains. “They don’t directly test my model, but research is definitely under way on how you can gain by having a high, rather than a low, reputation.”

His system has potential applications beyond Internet transactions. As Ekmekci sees it, it can be used in rating credit scores and other aspects of the finance industry, grading students, storing and prioritizing information in crime databases, and informational planning in economics.

And what does Ekmekci see as the bottom line of his project? “To sellers, never give up,” he says. “To buyers: Pay attention to ratings.”