Ideas on NBA Bets
As long as I’ve been interested in following sports, I’ve been very into the idea of sports betting. I’ll be the first to admit it’s not exactly a friendly market to bettors. It’s a zero-sum game where no money is actually being created, meaning every penny that betting sites earn has to be taken from someone else’s pockets. When you consider how huge companies like Draftkings and FanDuel have gotten, each generating hundreds of millions in annual revenue, that’s some compelling evidence that the betting sites are pretty darn good at what they do. It’s basically hopeless for a casual bettor to profitably “truly beat” any of these betting sites.
That said, the sportsbook and bettor have totally different constraints on how they make their predictions. The two key advantages for the bettor I think are:
Volume of predictions
You can go on to any of these sportsbooks and you’ll find hundreds of different lines you can bet on, updated live. As a bettor you’re under no obligation to generate hundreds of live predictions daily. You can be selective and only bet on lines that you find most favorable. My hope then is that in order to cast as wide a net as possible, the sportsbooks have sacrificed some level of rigor in their predictions, while still generating net profit on each line.
When setting lines, incentives for the sportsbook aren’t perfectly aligned with accuracy
The way for a sportsbook to make money is through the “juice” in the lines they set. On standard lines, the house has the equivalent of a 2.4% edge on every bet (it’s frequently even higher). So their optimal scenario is an even 50–50 split from bettors leaving them a guaranteed profit. To accommodate this, they’ll often shift lines away from their original predictions to entice bettors to even out the spread, even as they’re sacrificing accuracy in the process.
The short of this is, this opens some window of opportunity for you as the bettor to move your opponent from the sportsbook to other bettors. If you can tease out where these weaknesses exist, you’ll be able to bet profitably without ever needing to dream up some sort of prediction model that’s more accurate than the sportsbook’s.
I’m aiming to turn this into a series where I test out various hypotheses and see which could be winning. For now, I’m only looking at player prop bets since I’m hopeful there’s more room for (likely intentional) inaccuracies in their lines.
The average bettor is likely to overestimate the performance of stars. Accordingly, sportsbooks may intentionally overestimate scoring lines for star players.
To test this out, I scraped thirty games worth of player prop betting lines off of DraftKings, from August 25th — September 15th. That translates to 300 lines, averaging out to the equivalent of one betting line per starter (not actually the case game-by-game). The number of betting lines to test shrinks down to just 89 lines, once you get down to only considering all-star players. As it turns out, over those 89 lines the bet under on all-stars strategy works shockingly well.
By always betting the under, on all star players points lines, you win 56 of 89 bets. Assuming you bet $100 every game, you’d end up profiting $1657.52. Below I’ve also plotted the performance of this strategy over the 89-bet duration.
Testing against zero-info bettors
Asserting all-star lines are overestimates provides no additive information. Equivalently, betting under on every line should result in equal return to betting over/under randomly.
Ea = Total earnings when betting under on all-stars.
Eo = Total earnings when betting randomly
Null Hypothesis: Ea = Eo
Alternative Hypothesis: Ea > Eo
After testing 5000 simulations of the random bettor, I find that the all-star pessimist strategy performs better than 4951 (99%) of the random bettors. That even with all the extenuating factors, seems like compelling evidence that my theory has additive predictive value toward the betting lines.
I’ll be honest, I’m still skeptical that this would be a consistently winning strategy, and that’s mostly due to the small sample size of this test. Particularly because these games took place in unique circumstances, (i.e. being in the bubble, and then transitioning to playoffs) it’s entirely possible that even if this trend wasn’t just noise, that it’s not sustainably exploitable outside of this brief period.
