Moneyball 2.0

Baseball history is replete with free agent deals—A-Rod to the Yankees for $275 million, anyone?—that probably seemed like good ideas at the time.

The problem is that it’s pretty easy for Major League Baseball teams to target players they want in their rosters, but not so simple to know what they’re actually worth. Even the rise of baseball analytics in the “Moneyball” era hasn’t stopped teams from shoveling millions to players who underperform.

A new analytical tool created at the University of Maryland to find free agent bargains just might change that. In a forthcoming paper in Statistical Analysis and Data Mining: The ASA Data Science Journal, Sean Barnes and Margrét Bjarnadóttir, assistant professors in the Robert H. Smith School of Business, show how analyzing an unprecedented breadth of hitting, pitching and fielding statistics can identify the best and worst deals in free agents.

“We can predict future performance in baseball players, and use that to estimate how much value their performance would generate,” Barnes says. “Then when we compare that to their market value, we can actually identify players with excess value—those whose expected future performance value exceeds their market value.”

The use of cold, hard statistical analysis rather than more subjective methods like scouting took off when the method was employed by Oakland A’s manager Billy Beane, and chronicled in Michael Lewis’ 2003 bestseller, “Moneyball.”

But while Beane’s early analysis mainly targeted two measures—on-base percentage and slugging—as better indicators of player performance than traditional metrics like RBIs and batting average, the UMD researchers’ performance gobbles a lot more than that, from WPA (Wins Probability Added) to WAR (Wins Above Replacement). “We look at everything,” says Bjarnadóttir, a native of Iceland who began cheering for the Red Sox as a grad student in the Boston area. (Barnes is a dyed-in-the-wool Chicago Cubs fan.)

By retrospectively comparing player performance before and after signing free agent deals, Barnes and Bjarnadóttir found their models successfully identified players like Ichiro Suzuki who were on their way to becoming superstars, but still relative bargains. According to the model, Suzuki was worth $9.8 million more than the deal he signed with the Seattle Mariners in 2003. Pitcher Chris Sale, signed in 2013 by the Chicago White Sox, was the biggest steal of all, with $11.7 million in excess value.

On the other hand, A.J. Burnett, signed by the Pittsburgh Pirates in 2013, underperformed to the tune of $16 million, they report, making him the most overvalued free agent in the study.

The next step in their research, Barnes and Bjarnadóttir say, is developing models that help managers predict not only how well individual players are likely to perform, but also which players fit best on their roster.

Other endeavors besides sports could potentially be analyzed—HR departments could someday deploy the workaday equivalent of baseball analytics to decide on hires. But America’s pastime was a natural, the Smith School researchers say, because teams and fans alike obsessively track statistics, even as the kinds of stats being tabulated continue to multiply.

“There’s probably not a team that’s not using some kind of data-driven models,” Bjarnadóttir says. “If you’re not taking it into account, you’re a step behind the others.”