Another week gone by, and another week I get to try to explain the piss poor performance of our college football picks. We don’t have the 3 seasons of documented profitable performance in college that we do with our NFL plays. This is our first season, and this is not the way we hoped to start. However, we have a model that we are confident in, a model that has done well in out-of-sample backtesting, a model that’s been on the right side of the line moves, a model I still put my own money behind.
One of the trends I’ve noticed is that we do seem to like a lot of big underdogs. The way the model is created, we have ratings for each team in a few different statistics (play efficiency, rush efficiency, play success, scoring efficiency). It’s explained a little more in my “Introduction to Massey-Peabody College Football Ratings” post. There are a couple of reasons to believe our predictions may not be as accurate for high point spread games:
- They did not do as well in backtesting. However, big underdogs as a whole profiled very poorly in that period, and as someone who believes that the market will tend to correct obvious inefficiencies, I’d expect big underdogs as a group to cover 50% of the time.
- Our system does not consider pace of play. With the exception of scoring efficiency, our other stats are taken at the play level. This means that we’re assuming a team that is Y% better than its opponent on a play level will be X points better, without considering whether that team is running 65 or 90 plays per game. It stands to reason that we’d expect a team Y% better per play than their opponent to win by more the more plays there are in the game. (Sidenote: The market may not fully incorporate this either, as large favorites with very high totals have consistently profiled as strong bets)
- Our system uses “situational weights”–every play is not created equal. The fourth quarter of the Louisville/FIU game last week (for those who weren’t watching, and that means probably everyone, Louisville was up 60-something to 0) is not weighted as heavily as the first quarter, when the game was still relatively close. Why do we do this? Because how a team plays when leading/trailing by 50 points is not as indicative of how they’d play in a more “normal” situation. Here’s the issue though: one team may pull it’s starters when they’re up 30 in the third quarter; another may wait till they’re up 50 with 3 minutes left in the game. One team may continue running their normal offense with the 2nd-stringers to give them game reps, while another team may be more sportsmanlike and run the ball up the middle every play to run out the clock. The point is that performance in low-leverage situations is more predictive for some teams than others, something our model does not consider. Teams have different motivations in these situations. Louisville has an incentive to run up the score, since they play such a soft schedule that they likely won’t get national championship consideration even if they go undefeated. Alabama has no such incentive.
This does not mean we expect our plays in large point-spread games to be unprofitable. In fact, we empirically correct for any biases based on the magnitude of the point spread. However, we are going to (a) use a higher threshold for picks with big point spreads, and (b) look at the pace of play in these games, and see if it could be a reason why we’re showing value.
There’s a good chance we don’t get back to 50% on the season, just due to the hole we’ve dug ourselves. But my expectation going forward is that these will be profitable. We only pick games that we are willing to bet ourselves and we’ve still been predominantly on the right side of the line moves. As is our credo, we will continue to look value process over outcome.
Here are this week’s plays. As usual, injuries are not considered. Plays are based on the Massey-Peabody ratings, and lines are widely available as of 1:30pm EDT on Wednesday.
Big Plays (5-7 YTD)
- Pittsburgh -6 vs. Virginia
- Oregon State -11 vs. Colorado
Other Plays (6-17 YTD)
- Texas State +12 vs. Wyoming
- Central Michigan +23.5 at North Carolina State
- Florida State -21.5 at Boston College
- Notre Dame +3.5 vs. Oklahoma
- Stanford -9.5 at Washington State
MP Leans (2-2 YTD)
- USC +5.5 at Arizona State (some +6s still available)
- Arizona +10 at Washington
- Vanderbilt -19.5 vs. UAB