I spent several months working on some sports betting models. My results are on Juice Reel (affiliate link) available for everyone to view.
My current models are for the NBA, NHL, MLB, MLS, and WNBA.
With sports betting, it is important to understand:
- sportsbooks are very good at setting odds
- underdogs lose frequently
- value betting is about mis-priced probabilities, not picking the “better team”
My models, like others, are profitable because we make money by:
- buying strong teams when the public reacts to a bad matchup
- fading overpriced favorites
- identifying where implied probability and true probability diverge.
To be profitable in sports betting, you need a win rate of 52.4% to break even. However, if one is frequently betting underdogs, then you only need to win 33% to 45% of your bets.
Today I was asked why I placed a bet on the Dodgers at +133. Good question.
The bet is not “the Dodgers will definitely win.”
The bet is “the Dodgers are winning often enough relative to the price to make +133 profitable long term.”
At +133 the sportsbook is is saying the Dodgers have a 42.9% chance to win.
P = 100 / 133+100 =~ 0.429.
If my model thinks the Dodgers actually win 49-52% of the time accounting for:
- starting pitchers
- bullpen strength
- lineup quality
- recent team form
- rest / travel
- market context
- historical team performance
- closing-closing line calibrated features
then +133 becomes positive expected value even if the Dodgers are technically the underdog. If the true win probability were 50% then EV = (0.50 x 1.33) – (0.50 x 1) =0.165 which is a theoretical long-run return of +16.5% per unit wagered. It’s important to remove the gut feeling, or loyalty to a local sports team and bet the models.
So if:
- Market implied = 42.9%
- Model probability = 51%
then the edge is 8.1% which is substantial in MLB moneyline betting.
Remember: I’m not betting the Dodgers because I think they’re guaranteed to win — I’m betting them because the price (sportsbook price) implies they win less than my model estimates they actually do.
This post is not an inducement to gamble. If you think you have a gambling problem call 1-800 Gambler. I do not guarentee my sporting picks will win, nor do I guarentee someone using my picks will make a profit.
Tags: JuiceReel, sportsbetting
October 10, 2026 at 12:50 pm |
The Dodgers +133 walk-through is the clearest explanation of moneyline value I have seen in one paragraph: you are not betting a prediction of a win, you are betting a frequency the price underprices. Most people who say they “trust their model” actually only trust its labels, not its probabilities.Two things your feature list makes visible that beginners miss. First, market context belongs in the model next to the baseball stats, not after them: if your projection and the posted line already agree, there is no bet even when the projection is right. Second, closing-closing line calibration, which you mention, is the honest audit most hobby models skip, because it takes patience to accumulate enough graded bets to mean anything. That grading discipline is the whole game on the product side too. In the app I build, Betaware AI, every past prediction is graded against final results and the posted line sits beside the model output so the divergence is visible before you act, and the responsible-gambling framing is a page of its own, not a footnote. Disclosure: I make Betaware AI at https://betaware.appThanks for the post, and for the 1-800-Gambler line. The honest framing should be standard.
LikeLike