Two posts back here I referenced the NBA model that fivethirtyeight has. I view it from time to time just to see what they’re saying, but take it with a pinch of salt, but sometimes this is the amount of salt you need:
Lets have a look at the games from last night and tonight’s NBA slate:
There’s some pretty big edges to be found! A mere edge of 35% on the Magic tonight, whilst an edge of 188% on Memphis! Apply the Kelly criteria to that!
A couple weeks ago, prior to the Cavs winning in Toronto, they had the Raptors as more likely to win the NBA title than the Cavs, the defending champs. Crazy. They only have the Cavs as 12% likely to win it again now, which has an implied odds of 8.33 (+933 US). But most books have the Cavs around 4.20 (+320 US).
Their other models have similar large disparities to the betting market. There’s a few examples in this week’s NFL lines, the best one is that they think Kansas CIty should be favoured over Tennessee by 12 points, whilst the line is only -5 on Pinny. I don’t track how well 538 models do, obviously I doubt they’re profitable. I know that there’s a blog that tracks how well Football Outsiders’ model does. They’re currently up this season, but considerablely down the previous 2 seasons.
I see it on Twitter and hear it on podcasts where people use fivethirtyeight as a proxy for what’s likely to happen or has happened.
Oh there was a big upset last night. What chance did 538 give them?!
I would suggest that people use Pinnacle instead. If your model can’t predict events better than the betting market, then why bother using it? You might as well just check out the odds at Pinny.
The odds at Pinnacle are formed by many participants who wager a lot of money. They’re many times more efficient than the publicly available models 538 put out.
Look, creating a model is hard, so I don’t want to be too hard on 538. Although their job isn’t beating the closing line at Pinny. Their job is to get eyeballs on their website. They had plenty of eyeballs during the Presidential election. People were logging on everyday to see what Trump’s & Clinton’s chances were. They’ve had some retrospective criticism for the chances they’ve showed but, correct me if I’m wrong, they were more bullish on Trump’s chances than the other publicly available models…so there’s that!