September 3, 2023 NCAAF College Football Betting Cards deta
- Yesterday was the first big Saturday of College Football this season. The model had a rough start in Wk0 and Thursday, but it performed big on Friday and Saturday.
- The total results yesterday were +6.65 units on Spreads/ML bets and another +1.29 units from RRs
Season Tracking:
I’m using a starting bankroll of $1,000 to start the season here. I don’t plan on replacing this with extra funds during the season, so this is all we will be playing with.
1 unit bets equal to $10 as my baseline bet for each game This money equivalent will not change throughout the season on what my units are worth. But the Unit sizes will be adjusted accordingly based on bankroll performance and model confidence.
Season Results:
- ML/Spreads: 19-14-1 | +5.07 units | 14.9% ROI
- Round Robin Parlays: 20-39 | +1.4 units | 39.6% ROI
- Total profit: +6.47 units | 17.2% ROI
Unit sizes:
- The unit sizes were previously set to 1 unit and the rough equivalent of a churn of 100 games of bankroll.
- This is ultra conservative and not an efficient use of bankroll. In MLB I’ve currently got churn set to 30 games as a hard floor before lowering unit sizes.
- What is Churn? Churn represents how many games I can bet before exhausting my bankroll if I were to lose each one. It’s the inverse of the percentage bet and I think it provides a clearer picture to help visualize the impact of each bet.
- Due to college sports having more volume the bankroll is more exposed to game volume at one time and one bad turn of the model could hurt results. For that reason I will be setting my bankroll floor to 40 games before lowering unit sizes.
- The Bankroll is a tool and I want to try to use it as efficiently as possible while not being overexposed on risk. I had no idea how well my College Football model would be to start out, hence why I was conservative with 1 unit games to start, but due to the recent successes over the larger volume I feel more comfortable getting the unit size to where I’d like it.
- Starting today the unit sizes will increase to 2 units.
- Based on my current bankroll of $1064.6 this means I currently have 106 units and 2 units of bet sizes represents 53 games of churn,
- This unit size is a relatively safe amount, especially compared to my MLB sizes, yet allows me to be more aggressive with the wager total.
- The hard floor of 40 units means I will need to lose a net of 13 games and have my bankroll drop down to $800 or -20 units on the season before I need to safely reset it back to 1 unit in size.
- On the flip side, I will be able to scale up if the model continues to perform. I want a hard floor of 40 units before lowering, but I want a floor of 50 units on the next size up before raising it. In order to raise the bet sizes up to 3 units I will be looking for the bankroll to get up to 150 units before making this leap up.
Round Robin Parlays:
Now you’ve been told that Parlays are suckers bet, but that’s not necessarily true. Parlays are like options in the stock market. They are a way to provide leverage to your betting. It can lever up or lever down. Since most sportsbettors are making longterm -EV bets than the leverage is more often working against them. However if one is making +EV bets, proven out by postive results straight betting than the leverage will work in their favor.
This is why I’m including small unit amount each day on the card for RR Parlays. Round Robins are just a quick way to get a parlay going for every game on the card as the combinations can be quite large.
I’ve been breaking them into smaller packages when possible to allow variance to do it’s thing as its better to sweep 4 on a card versus haviing a weaker card with more losers. 1 or 0 wins results in the same thing of a losing parlay so it would be better to try to pair them together with smaller cards. I’m starting to think the math will work out the same over the long run and the variance of smaller pairings is an illusion as it should all even out in the long run as then winners aren’t always paired with other winners. So perhaps there will be bigger RR parlay combos in my future here.
Due to them showing a whopping 39% ROI to start the season I’m going to increase the wager amount on them going forward. Before I had it set to about 10% total of the card, I’m going to increase it up to 20%. I don’t have a set floor or anything, I’d like to keep it at 20% above 0 in profit, but I’ll likely give it the same floor as regular units where each unit size in increase the leverage will increase 10%.
The Model:
Now this is the driver of the success of the bets. Without a good model how does one infer whether a bet is good? The key to sportsbetting is determining +EV. In order to do that one needs to know the Win Probability in order to compare it to the odds being offered. It is impossible to truly know what the WP% is of a certain for each team, but we can backtest and test various WP% calculations and determine whether they are performing close to their suggested amounts.
My model has the WP% calculated based on my own power ranking of teams that I’m tracking and with a few adjustments for Home advantage and soon to be Strength of Schedule performance I’ll have what I consider a very solid baseline for developing WP%. To test it I just run every game I have data on and test whether the game would’ve received a bet based on the model parameters knowing what I knew about each team at that time. By continually backtesting and trying to game out the most units wons possible across every game it should tailor the model to a winning spot going forward. One can make any sample size work, but if the backtested model shows strong low volatitly results, meaning the trend is on a firm path upward, then one likely has a good mooel.
After yesterday’s games I backtested the model and made a few changes to my current algorithm. Before I was taking ML bets on teams with WP% at 45%+, backtesting shows I should set these to 55%+ and opt for Spreads instead if the calculated edge is above 10%. These changes are specific to my model only as its dependent on all my inputs which is going to be vastly different than anyone elses. So you will need to do your own tinkering to backtest your model appropriately.
Here is the current backtested model evaluating all 90 completed games in NCAAF so far this season. My model currently would’ve bet these 30 games for a result of 19 units won. That is an insane 63% ROI.
I don’t expect that to hold, but it gives me a great baseline to say, “Hey this is how games of similar profile preformed, so let’s make bets on future games with these profiles”
Here’s to hoping the model can continue to perform.
DAILY CARD:
- Today’s slate only has 3 games and there was only one qualified bet on the docket
Round Robin Units Profit: 0
- Total Day Profit = X
___________________________________
___
