A look at statistical win projections for the 2019-20 NBA season could be helpful for bettors looking for opportunities on over/unders.
With just two weeks until the NBA’s opening night, the boundless optimism of summer has shifted towards cold rationality and more honest assessments of team quality. Nowhere is that tonal shift more obvious than in the annual pastime of picking team over/unders. Whether you’re actually betting money or just using these as a line in the sand to measure how far your personal assessment is from the more conventional wisdom, digging deep on the over/under lines Vegas sportsbooks set for each team gives us plenty to unpack.
Another useful measuring stick you’ll see proliferating this time of year is statistical win projections. There are a variety of systems publicly available, relying on different metrics and methodologies to project win totals for each team, and comparing these projections to the Vegas lines can highlight some places where objectivity seems to differ from the conventional wisdom.
The graph below looks at win projections for all 30 teams, from a variety of statistical systems and individuals — 538’s Elo, 538’s CARMELO, ESPN’s BPI, ESPN’s Kevin Pelton, Jacob Goldstein, Kevin Ferrigan. (h/t to Nicolas Canova for collating these projections). You’ll also see a mark for the average of the statistical models and the Vegas over/under win total, per the Westgate SuperBook. The graph lets you see teams for whom the model average diverges from the Vegas assessment, as well as the variation for how the different models see each team.
If you were analytically inclined, had to money burn and were looking to place some over/under wagers without doing your own research or creating your own analytic projection models, simply looking for betting opportunities where the Vegas line and the model average differ the most would be an interesting place to start.
However, you can see from the visual above that the model average doesn’t necessarily represent a broad consensus. For example, the model average estimates both the Miami Heat and Portland Trail Blazers to finish roughly 2.5 wins below their Vegas line. However, the spread of projections on the Trail Blazers is much broader, ranging from 538’s pure Elo projection of 50 wins to a projection of just 40 wins with their CARMELO system. The spread of projections on the Heat goes from 43 with 538’s CARMELO system to 38.9 from Jacob Goldstein and his Player-Impact Plus-Minus. In this case, if someone was inclined to bet the under, it might make more sense to go with the Heat since there was a greater consensus among the models, all of which had them finishing under their Vegas line.
This next graph marks all 30 teams by the difference between the Vegas line and the model average for their projected win totals and the variance in projections among the seven models.
Here we can see the difference in implied risk between taking the under on the Trail Blazers and Heat, as well as several other projections that jump out. The Pacers (under), Bucks (under), Jazz (under), Wizards (over) and Timberwolves (over) are all teams for whom the projection models tend to agree and have an average projection significantly above or below their Vegas line.
Of the 30 NBA teams, there were only six for which all seven models landed on the same side of the Vegas line — Clippers (under), Heat (under), Bucks (under), Timberwolves (over), Spurs (under), Jazz (under) and Wizards (over) — which is another interesting batch of teams for potential bettors to consider.
Reasonable people can disagree about the accuracy of the models, and most of the 18 win projection models in last season’s APBRmetrics Forum win projection contest did finish with average absolute errors between six and seven wins. However, 10 of those 18 models had an average error smaller than that of the Vegas line.
It’s also important to remember that minute allocations can explain a healthy percentage of the difference between a win projection and a team’s actual win total — that is to say unexpected injuries or trades could have a dramatic impact on how close these projections finish to the actual mark and so the perceived roster stability of each team is an important factor to consider.
As always, the conventional wisdom will be way off on a handful of teams this season. The statistical projections will be way off on a few as well. In a few weeks, we’ll start to see which is which.