Question of the Day - 18 December 2023

[Editor's Note: The answer to this somewhat complicated question is supplied by Logan Fields, author of our book 20/20 Sports Betting -- Think Like a Pro. Nice to hear from Logan again; it's been awhile. His QoDs are always interesting and instructive.] 

In other words, is it possible for a prediction model to be so good that it consistently beats the “market,” where one could profit betting into a line whenever a discrepancy exists? Short answer: not likely.

Say the model of your choice gives the San Francisco 49ers a 60% probability of beating the Baltimore Ravens. This would mean the money line for the Ravens should be +150, which represents a 40% chance of the Ravens winning. Betting markets have settled on San Francisco as a -175 favorite and the Ravens a +155 underdog. If you have blind faith in your model, this would mean the Ravens are a good play at +155, since your model suggests they should be only +150.

Since betting markets reflect the collective knowledge of all bettors, including other models, you should ask yourself, what is my model picking up on that the rest of the free world seems to be missing?

In this scenario, you'd be better off putting your faith in the market, which is more likely to be correct over a particular model. The worse your model is, the more plays it will find, as it will be more prone to drift away from the wisdom of the crowds. If your model is finding 8 or 9 plays on an NFL Sunday, good luck with that!     

So if it's not possible for any one analytical model to consistently beat the market, should they be ignored? Not necessarily.

You raise a good point when stating that you wouldn’t necessarily bet into all discrepancies, but only when a certain gap exists. So rather than look at all games where a model might be finding value, perhaps looking closer at games where the discrepancy is the greatest might be the way to go -- the model’s “greatest hits,” so to speak.

Once one of these games is identified, look at the various intangibles and determine if something might be happening in this game that the market may not be factoring in. For example, the weather. Official game-time weather conditions have been classified as “snow” or “light snow” only 19 times since 2014. This is a tiny sample size and is it possible that your model be picking up on things the market is missing? Perhaps. Ask two professional bettors whether snow helps or hinders scoring and you might get two totally different answers. With this kind of uncertainty, it’s not farfetched to think a model might be better than the market in certain situations. (Fun Fact: In those 19 snow games, 10 went over the total, while 9 went under. So, the jury's still out).    

Or look at the handful of games played in London each year. Once again, a very small sample size. Might there be a difference in team performance for those teams arriving on Friday of game week versus those teams that can and do arrive in London a week early?

My conclusion: No models will consistently beat the market week in and week out. Ignore the minor discrepancies, but pay attention when the differences are greater. In certain game situations, a sophisticated model might outperform the market when there's significant uncertainty due to various intangibles that possibly come into play.