At some point, my bowl projections needed to be revisited this off season. After all, of what use is my putting them out there if no one goes back to see how accurate they were? I wouldn't expect you, dear reader, to waste time doing that, so here's a look from me.
I knew they were reasonably accurate in just determining winners, but for the first time, I had them go against the point spread and over/under as well. Straight up, they went 23-12 in the 35 bowl games for a respectable .657 winning percentage. Against the spread they weren't as good, going 19-14 for a .576 winning percentage (two bowls matched their spreads exactly). That's not shabby for against the spread though. For the over/under, they were a useless 11-19 for 36.7 percent (four games projected even with their over/under, so I didn't count them).
I broke my projections down by the kinds of participants, and as it turns out, they were quite good against the spread in games involving teams from non-automatic qualifying conferences:
|Kind of Bowl||Straight Up||vs. Spread||O/U|
|Overall||23-12 (.657)||19-14 (.576)||11-19 (36.7%)|
|AQ vs. AQ||12-7 (.632)||7-10 (.411)||4-14 (22.2%)|
|AQ vs. non-AQ||7-0 (1.000)||5-2 (.714)||4-3 (57.1%)|
|non-AQ vs. non-AQ||4-5 (.444)||7-2 (.778)||2-3 (40.0%)|
That really surprised me, especially with the AQ vs. non-AQ row. I have always believed that my method would work worst for those and better for AQ-only and non-AQ-only games. So much for that.
When it came to predicting individual teams' scores, they were a bit hit-and-miss. I projected at least one participant's score within a touchdown in 24 bowls (68.6%), though I only got 29 total teams' scores within seven points (41.4%). The score projections also got 14 bowls' margin of victory within a touchdown (40.0%).
How do these results compare with some others?
One of the gold standards of advanced college football metrics is the FEI ratings from Brian Fremeau. His bowl projections ended up fairly close to mine: 23-12 straight up (.657), 18-12 against the spread (.600), and 14-19 on the over/under (.424). His projections did better against the spread and for the over/under than mine did, though they didn't win by a mile. He also had 24 bowls with at least one team's score within a touchdown (68.6%) and 29 total team scores within a touchdown (41.4%), though he got two more margins within a touchdown at 16 (45.7%).
Of course, no mathematical model will get every bowl pick correct. There's no way to project, say, Nebraska's epic flop in a Holiday Bowl it didn't want to be in with a formula. How about someone making up picks without a mathematical model?
For that, I'm picking on ESPN's Pat Forde. His bowl picks went 19-16 straight up (.543), 19-13 against the spread (.594), and 17-17 on the over/under (50.0%). He was worse than us math guys straight up, but against the spread he was right there, and he was better on the over/under. He got only 20 bowls with at least one participant's score within a touchdown (57.1%), and he got only 27 total team scores within a touchdown (38.6%). He did get 19 games' margin of victory within one score (54.3%), the best of anyone.
So I guess the conclusion is that my bowl projections stacked up favorably when compared against a real football stats expert like Fremeau and when compared against someone who makes his living following the sport like Forde. I already have some ideas on how to make my projections better for next year, so keep an eye out next December for more score projections.