There was a thread on Twitter recently that got me thinking about delving a little further into the topic of analytic models. The twitter user notes that not one model listed on The Prediction Tracker performs well over the long term against the spread.
It's kind of interesting that according to The Prediction Tracker, not one single model listed, BPI (@SabinAnalytics), @masseyratings, @DRatings, etc, performs well longterm in any sport against the spread.
To be fair, some do have edge in specific, granular situations.
— FiftyKelly (@fiftykellypicks) June 21, 2022
This brings up a few interesting questions and thoughts as to what value analytical models have in the realm of betting. What are the advantages and disadvantages of using modeling? How accurate of a model is it possible to create? What are the things that can not be accounted for in a model? And finally, is it possible to create an analytical model that can beat the closing line over the long term? Let’s dissect all of this right now!
Advantages and Disadvantages of Modeling
There are a variety of benefits and drawbacks to modeling outcomes in sports. These lists are infinite and constantly changing, but we try and tackle the big issues here…
Where Do Models Thrive?
While models have a few hurdles to overcome, there are some clear benefits to using models to gain a betting advantage.
Firstly, models have no bias, feelings or emotion. This is perhaps the most important advantage that a model can have. Whereas, a bettor can get bogged down in meaningless streaks and biases, a model can simply ignore all of the noise. Secondly, in games without huge markets, a good model (with excellent inputs) could be more knowledgeable than the crowd. This is because there are less sharp eyes and lower limits on these contests. Lastly, though it seems counter-intuitive, models can pick up trends that the eye might not be able to see. This can be a positive as a model may be able to pick up a trend before it is seen by the naked eye.
I Can’t Model This
There are also some drawbacks to modeling. Simply, models can not account for all the factors that affect probabilities to win and final scores.
Firstly, take injury data on it’s face. How is a model supposed to value an injury report that reads, “LeBron James questionable with lower body injury”? Obviously, this can mean a variety of different things. Did LeBron just tweak his pinky toe or is he playing on a torn MCL? A sharp bettor will have much more valuable information (or dare I say inside info) as to the true extent of the injury. LeBron healthy is worth up to eight points against the spread versus a LeBron that isn’t playing. It’s easy to see how detailed information that a model can’t account for is huge here.
Secondly, a model usually doesn’t know how the motivation factor plays into probabilities. I don’t mean this as it relates to playing for playoff spots versus being eliminated from the playoffs, because this can be accounted for. The tricky part is the day-to-day regular season stuff that goes on behind the scenes.
The Vegas Golden Knights played their first season in the NHL starting in the fall of 2017. Few thought they had much of a chance to make the NHL Playoffs at the beginning of the season. They ended up winning ten of their first eleven games and sixteen of their first nineteen at home. How did this happen? The answer is twofold. First, the Knights were better than expected. Second, teams who played Vegas had the added drag of having stayed in Vegas the night before. In essence, do you think that a 24 year-old hockey player is going to land in Las Vegas and sit in his room at Mandalay Bay the night before game day? Of course not! The hangover effect is real. Models can’t see it, but sharp bettors will find out if the team is at the strip club until 3am the night before a game.
Can a Model Beat the Closing Line Over Time?
This answer is a little more complicated than it appears. The evidence to-date shows that models have greatly struggled against the closing line over the long-term. However, this doesn’t mean that models haven’t had winning seasons or aren’t valuable. Further, I’d argue that some sports are much easier to model than others. The NHL and MLB, with their troves of data are a lot more predictable than the NBA, for example. From a bet type perspective, over/unders are easier to project than money lines and spreads. At the end of the day, the less eyes that are on a bet, the better the chances that a model can win in the long term.