I just want to reiterate that turning a profit via gambling is not the goal of this project. I am estimating what the expectations are if every organization followed the cyclical nature of the league. I am not predicting what will happen on the field. Through the first 3 weeks of the 2024 season, underdogs of 5.5 or more points were 10-4, including 5-0 in Week 3 alone. In that same time period, NFL betting public (51%+ of the bets) was 17-30-1 against the spread (betting using point spread), which is the second worst performance through that time span in the past 20 years. If one were to follow those bets, to use $10 as a unit, and to bet 1 unit on each bet, he or she would be down $145.30 (-30%). While the league had certainly undergone a period of instability from 2021 to 2023, I would be very suspicious of anybody who is confident about what will happen in any individual NFL game regardless.

“If this is based on data, why doesn’t this win 100% of its bets?”

“All models are wrong, but some are useful.” Article 1 / 2

Deviation from expectation can actually be more useful

Example: firing Head Coaches when they underperform

Other potential internal reasons for inaccuracy

Misreading prior data, leading to an incorrect prediction

Could simply be that model could be adjusted/improved

Other potential external reasons for inaccuracy

Weather, refs, random bounces, injuries (especially QBs)

Order of opponents, schedule timing, placement of byes