Stat Pundit Rankings: MLB win over-unders
About two weeks ago, I used some familiar metrics to analyze how analytics-based websites performed as far as predicting MLB win totals. With the regular season now complete, winning bets have been cashed, and the official performance for each site is listed below:
O/U: The Hilton’s over/under for each team
BP: Baseball prospectus
TR: Team Rankings (caveat on the linked page: the site stresses their MLB predictions are a work in progress)
Zips: ZIPS projection system (espn.com)
PM: Prediction Machine
TB: Trading bases, an avid blogger and book-writer
Here are my metrics
MSE: Averaged squared error between the prediction and the win totals (lower is better)
MAE: Averaged absolute error between the prediction and the win totals (lower is better)
Corr: Correlation between the predicted and the win totals (higher is better)
Baseball Prospectus and Trading Bases appear to offer the only clear advantage over the Las Vegas line, at least among these predictions, as judged by a higher correlation and a lower MSE between observed and predicted values. On average, TB was the only prediction site to finish, on average, within seven wins of the actual results.
A savvy bettor would’ve finished 12-9 on bets where BP differed by the Las Vegas O/U by more than two wins, and 10-6 using the same cutoff for TB. Picks that BP and TB agreed (by more than 2 predicted wins) on finished 7-4
Here are the Vegas lines and each site’s picks. In some cases, the projected total wins might not add up to 82 per team, most likely due to rounding errors.