Week 4, 2023 NCAAF College Football Betting details:
- Last week was pretty devasting, nothing seemed to be working.
- I worked on the model this past week to make sure that backtesting results were set to their peak performance.
- It’s rough going right now, but the more data accumulated each week the more solid ground for the model as it’s set to achieve maximum profit potential based on game data for the season.
Season Results:
*Thru Wk3 – 9/15/23
- ML/Spreads: 40-47-2 | -17.50 units | -12.2% ROI
- Round Robin Parlays: 80-377 | -5.48 units | -28.0% ROI
- Total profit: -22.90 units | -14.0% ROI
Unit sizes:
- Week 4 the unit sizes are set to 1 units a bet.
- 77.0 games of churn
- The lower level of units is due to two weeks of losses in a row and the bankroll now deep in the red.
- It’s time to regroup and focus on Model adjustments and only can I raise the units again after making some solid gains back with a proven model.
Round Robin Parlays:
I’m doing 10% of the card in RRs. It’s split into two groups for Saturday, the afternoon (noon-5:30 games) and Evening games (6PM+)
The Model:
I expanded upon the adjustment factors from last week. Reviewing the results to the setup last week I identified a flaw in that it was adjusting each team in the overall Power Rankings a percentage amount based on game performance. This is sort of what I want done, but it really limited the movement as team’s that blow out opponents by 50 pts were getting about 25% adjustments up or down. This felt wrong as team’s that are deemed bottom barrel beating top competion wouldn’t be justly compensated where they would likely need a 400% move up the rankings.
This week I set the adjustment for each game performance to look at the final spread and assigned each point a percentage impact on what the true probability of each team should be. Then I worked backwards to figure out what the power ranking point totals should’ve been for each team that week based on the final result of that game.
- Note: I’m calculating what I deem the adjustment should be based on past performance and then only future games for the following week use that adjusted point total for the team’s power ranking.
Example:
- Last week (wk3) Vanderbilt had 84.8 pts in the power ranking and UNLV had 120.2.
- I’m currently giving Home teams a 30% boost to their Power Rankings. This seems large but when looking at how it’s applied it makes sense as low teams in the rankings get a very low boost and top teams with huge home field crowds based on their winning ways become even more formidable against similar competition at Home.
- Adjusted UNLV had 156.2
- Vandy 84.8 out of 241 total game points in power rankings between the two teams represented 35.2% of the points and that is what I use as Vandy’s WP% in Wk3.
- The final score ended up being Vandy 37- UNLV 40. A 3 point advantage for UNLV.
- Based on a 3 point difference and assuming each point difference is a 1% difference in WP%, Vandy should’ve had a 48.5% WP% based on the game results.
- 48.5% of the game’s power ranking points were 116.9 which will take over as their new Power Ranking rating in Wk4, an improvement based on their more competitive game here where they entered Wk3 with 84.8 points.
- UNLV underperformed this game and they captured the remaining share of the 241 game power ranking points, with 124.1.
- But remember they were the home team and had their points boosted 30%, so I have to deboost them to get their neutral power ranking back for Wk4. Adjusted down by dividing by 1.3 their Wk4 Power Ranking Points are equal to 95.5.
As you can see in the example over the season each team will play each team and help paint a story of who the team is based on how perform against certain competion. I’ll probably need to taper the changes as the season progresses as some games bad luck will happen and not be truly representative of who a team is, but for now I feel like with the low sample size of weeks to review this is the best I have for the moment.
Lastly I’m quite pleased with this setup because after backtesting some constants and cutoffs in the algorithim I’ve got the model settled at a point where it would’ve gained 21.4 units over 112 bets. Also each week shows a profit with the recent week’s having the most profit. All good trends I want to see in the backtested data.
Betting Cards:
_____________________________________________________________
Card for the Week:
Wk4 Card:
I’m going to try to keep things simple here going forward. Instead of a daily card I’m just going to stack all bets here for the week. Just know that I’m often packaging daily groups or afternoon/evening slates on Saturdays into RR packages worth 10% of the main bets here.
You’ll see that Friday’s bets are listed at top here, greyed out as completed for the week.
Round Robin Units Profit: X
___________________________________
___
