Updated Oct 1, 2022 after a comprehensive study on historical results with data back to 2010.

For a while, I was not comfortable suggesting any single spread or O/U to bet on. The randomness is simply too large!

However, I do suggest an approach based on portfolio optimization: i.e., maximizing returns while minimizing risk.

• I recommend betting on multiple game lines each week (a combination of spreads and O/U bets). The number of bets has normally been around 15 per week, however fluctuating between 8 and 25, depending on the specific week and the season. (Current 2022 season has fewer bets— maybe only 10).

• I give the relative amounts to bet on each given line.

  • The numbers assume you have a weekly pot of approximately "10,000 money-units".

  • Added to this “10k” or whatever you choose, the suggestion is that you keep that same amount in reserve— as safety for future weeks.

  • The total suggested amount to bet will change every week, depending on whether there are good bets or not. Sometimes there can be a slump of only 8-10 bets in a given week, and the total amount to bet will also be lower. You should still consider the “10k money units” relative to the amount you target on average = amount kept in reserve on average.

The recommended betting proportions are based on a simple formula I derived, which approximates an optimal "investment portfolio". It is not the same as the Kelly Criterion since it is adapted for multiple simultaneous bets.

If my betting lines table does not show a certain game line, it is because that bet is not worth betting on.

Diversification

I suggest ca. 15 wagers each week. My models usually do better on Totals than Spreads, so a higher fraction of bets will be totals. The idea is to reduce risk while increasing gains, as a statistical consequence of placing a wider range of bets. Making just few "high likelihood" bets can backfire too easily. So, in contrast to focused betting, spreading out wagers sets a floor against %losses.

What I like about doing this:

1. Reducing risk means you'll most likely not lose your pot at once, so you can have enough for betting next week. And you can feel a little safer throwing a somewhat larger pot.

2. The goal is to work on gains over weeks, instead of in one go. If you can return 8% each week (+/- 40%) with this lower risk, then 16 weeks in a row would give a 160% return on the season (simple betting... 350% compounded— though that is not recommended unless you otherwise feel like expanding your overall pot).

3. And it can be fun to be involved in more wagers.

Accuracy measurement

This approach directly suggests a good way of measuring accuracy.  Unlike my usual trusted approach (correlation coefficients), reporting ROI is more suggestive of usefulness.  And unlike measuring "%hits", it more appropriately weights bets with higher likelihood. 

What goes into the model?

The approach is the result of a bunch of different models working together.  There is of course the Spread model and the O/U model.  There are also models that compare current Vegas lines to my expectations, which required analyzing their past relationship.  Most importantly, there are updated Probability (likelihood) models for Spread and O/U.  Finally, the probability models are processed and converted to a diversification scheme.