2026 CFB Preseason Model Intro
Now that summer is in full swing, it’s time to start getting ready for the upcoming college football season. I want to share my plans for how I’ll be betting this season, as well as the changes I’ve made to my model this year.
2025 Retrospective
2025 was a fantastic year for my model. I bet 412 units on the year and won 103 units, for a 25% return. My best bet of the year was Duke to win the ACC (where I won 40 units, although I could’ve won an additional 60 if I didn’t hedge my exposure before the ACC title game). However, even outside the Duke bet, I had a very strong year.
One big change I made in 2025 was to focus more on futures. I had seen in 2023 and 2024 that I had more edge betting futures than I did individual games, so I changed my bet split from about 25/75 futures/single games to 75/25. I’m quite happy with this decision, as I once again had a much better return (96 units on 326 units bet) on futures, than I did on single games (7 units on 86 units bet).
A major tailwind for my model has been the increased liquidity in CFB futures markets. As recently as 2022, it was hard to find reliable odds on teams to make the college football playoff. Now, those markets are quite liquid, and often poorly priced. Other futures markets now exist that previously weren’t listed- stuff like odds to win 10+ games, odds to win divisions, odds to make conference championship games, etc. The weirder and more esoteric the bet, the more likely I am to have an edge, so I’ve been very happy to see this growth in futures markets.
I plan on sticking with my futures-heavy strategy for 2026, perhaps tilting my bet split even further towards say, 80/20 futures as opposed to 75/25.
Tailwinds and Headwinds for 2026
There are some exciting changes to the CFB betting landscape that I hope to be able to take advantage of. The most obvious one is the growth of prediction markets. Last year, there was not any reason for me to bet on Kalshi- for individual games, the prices were the same as the listed sportsbooks, and the liquidity for futures sucked. Even deep into the season, the spreads for “odds to make the playoff” were often a dime wide. However, in the six months since the CFB season ended, Kalshi volumes have more than doubled. I’m hopeful that means better liquidity for CFB bets, and potentially more things to bet on than the listed sportsbooks.
I’m also encouraged by what I’m seeing on DraftKings and FanDuel right now. DraftKings has listed alternative win totals for every FBS team, which is something I’ve never seen before. This has me hopeful that they’ll expand their listings throughout the season. As always, the more bets they offer, the happier I am.
However, not all is rosy for me this offseason. The major headwind I’m facing is that the market has caught up to me a bit and I am already seeing my edges decay. I’ve always felt volatility is underpriced in CFB- in a sport where many players are unknown and half the roster turns over a year, no one (myself included) can be too confident in the preseason about how good teams will be. I’ve made a lot of money over the years betting on longshots in the preseason or early season (Indiana to win the Big Ten, Tulane to make the playoff etc.), because many longshots have odds that are too cheap. Unfortunately, looking at preseason odds this year, the bookmakers have clearly caught on. Teams that would’ve been +1000 to make the playoff last year in the preseason are now more like +700. The good news is that I think volatility in the sport is higher than it ever has been (and I have some evidence to prove it- stay tuned for more details). However, there are going to be fewer obvious longshots to bet on in the preseason this year than I’m used to.
Model Changes for 2026
I’ve made a few changes to my model this year that are worth mentioning. None of them are revolutionary, but they do make sense when you consider the direction the sport is heading.
One thing to establish before I get into model talk- my model, like many other CFB models, rates teams in terms of points relative to the average FBS team. E.g. a team with a rating of +14 would be expected to beat the average FBS team by 14 points. Historically, a rating of +14 means a team is about the 20th best team in the country. A handy rule of thumb is that the average P4 team has a rating of about +7, while the average G6 team has a rating of about -7. That is, the average P4 team is expected to beat the average G6 team by about two touchdowns.
Historically, my preseason model has worked as follows:
Start with last year’s final ratings for each team. This is by far the biggest factor- it’s hard to be multiple touchdowns better or worse than you were last year.
Move each team towards 0. This happens by roughly a factor of ⅔, although it depends a bit. By this, I mean that a team that had a rating of +21 last year will have a rating of about +14 this year.
Bump teams up or down based on how much returning production they have. I used to calculate returning production myself, but I found that my estimates ended up being quite close to Bill Connelly's estimates that he uses for SP+. It was quite time consuming to do this process myself every year, and Bill publishes his data for anyone to use (thanks, Bill!), so I now just use his numbers. It’s worth noting that Bill’s method of calculating returning production is maybe a bit surprising to most CFB fans- offensive line continuity is worth a lot more than you might expect. (I found the same result myself when I used to do this exercise every year). Note that transfers are accounted for in Bill’s returning production data.
Bump teams up or down based on how good their high school recruiting has been. I use a weighted average of the last few years of recruiting data.
The changes I’ve made to the model this year are:
Reduced the reliance on previous year’s preseason ratings. It’s still the biggest input into the model by far, but it’s smaller than it used to be. This makes some intuitive sense to me- as teams take more transfers than they had before, rosters are turning over more year to year. There’s thus less correlation between a team’s performance in one year and its performance in the next year.
Increased the reliance on high school recruiting. This surprised me, as intuitively you’d think that high school recruiting is the least important it’s ever been in the sport. My theory is that “high school recruiting score” is essentially serving as a proxy for “how much money do you have” in my model. I can’t directly measure how much money each team is spending on their roster (because no one knows the real answer), but their recruiting ranking is a decent proxy for it.
Discretionarily bumped some teams up or down for notable QB situations. The returning production formula measures returning QB production by looking at how many passing yards a team returns. However, I think there are some significant convexity effects here that are not being accounted for. To put it in layman’s terms, the model says that returning a 4000 yard passer is 33% better than returning a 3000 yard passer because he passed for 33% more yards. I think that returning a very good QB (or losing a very good QB) is more valuable than that. I thus have bumped up some teams (Cal, Oklahoma State etc.) who have much better QBs than you’d expect for their roster, and bumped down some teams (Duke, North Texas etc.) whose performance last year was largely attributable to a superstar QB who has since left.