Last year I went 21-9 against Vegas [1]. My win delta was 5.9, compared to Vegas' 6.3 [2]. Despite my best two bets on the board missing, I’d call it a fairly median result. As I mentioned last year, I have a great deal of respect for Vegas where there is high volume allowed and a short turnover, and not a lot of respect where there is low volume and your money is tied up all season. I suspect I’ll be able to win about ⅔ of these bets going forward, but the juice and the length of the bet settlement will keep me from overspending my winnings on hookers and blow.

At first glance I expected Vegas to be a bit sharper this year, but I’m about .43 wins off per team, where last year I was only .40 off per team. This might be attributable to my unwillingness to disagree with the market *too* much last season. This year, I erred on the side of accuracy rather than forcing myself to add a few wins to the tire fire that is the Detroit Pistons.

One of the first things I had some trouble with is trying to model teams at the extremes. I would imagine there’s a difficult to quantify rubber band effect happening, just as there is during regular season games. The front office might (or might not!) have the luxury of waiting for regression, but even if a losing streak is caused by shooting luck the players and coaches are considering every possible adjustment. NBA teams play fairly hard, but some play a little harder. The resulting sense of urgency should compress every team a bit closer to the middle class.

Something that interested me this year is the team wide effects that I wish I had a better handle on. Two of them that stand out are from teams that I’m higher than market, the Raptors and the Nets.

For the Raptors, I’m worried about their spacing. To quantify my worry, I made a very simple spacing score for all teams coming into this season. To get this spacing score, I used last year’s stats and combined volume and accuracy from 3 pro rated for the minutes I'm projecting [3]. This is *extremely* crude and doesn’t include incoming rookies, but the scores do a good job of quantifying about how much spacing you’d expect teams to have.

MIL 3.46

BOS 3.42

GSW 3.14

DAL 3.08

IND 2.96

The top 5 teams include spacing big men and high volume off the dribble shot creation. This method easily passes the smell test for the best spacing teams in the league.

Now peep the bottom 5 teams, paying special attention to the outliers at the bottom.

DET 2.27

LAL 2.25

BKN 2.21

NOP 1.98

TOR 1.85

The bottom 5 teams include some suspect roster building and a reliance on this year's draft to contribute to spacing, rarely a good bet.

Though the Nets don't look very good on this list, I’m actually more worried about their lack of creation. To quantify my worry, I just looked at last year’s USG pro rated for the minutes I’m projecting [4]. The result is irrelevant on the high end (additions like Jrue Holiday and Damian Lillard show the Celtics and Bucks up top), but on the low end it's notable how much Brooklyn sticks out:

BKN 18.66

MEM 19.25

MIN 19.28

CHI 19.47

GSW 19.56

I'll hand wave away every other score except how big of an outlier the Nets are. They’ll be relying on a massive increase in workload from Mikal Bridges, which some people may consider him more capable of than I do, and a strong season from Spencer Dinwiddie, who has torn his ACL twice and was medically red flagged in the draft. Ben Simmons also plays for the team.

I underscore these problems with the Raptors and Nets primarily to expose the difficulty in projecting teams as a product of individual players. It's more useful to see them as an organism that needs a certain amount of creation, spacing, rebounding, and defense [5].

In that vein, the Timberwolves are another conundrum that comes to mind. It was clear last year that as a frontcourt Towns and Gobert wouldn’t be greater than the sum of their parts, but it’s hard to quantify overlapping roles. I’ll have them as a little worse than my projection last season, but not quite as low as the market has them this year.

And that leads me to my final point. As much as I’d love to rely solely on machines [6], I think the NBA is a sphere where there’s a bit too much of the human element at work to not think I can improve the model with some common sense adjustments [7]. I made notes last year where I made the biggest manual adjustments, and they were largely big upgrades on machine alone. Wisdom of the crowds has been proven to be very wise, and there are some instances where I try take into account public sentiment on a player before that player truly arrives and it shows up in advanced stats.

The best publicly available win projections are back:

[1] Though it should be noted that a half dozen of those came within 2 wins, I was even on the close ones as well.

[2] This is how many wins I was off on average from the actual result, obv.

[3] 3PA/100 * 3% * MIN / team minutes. It should be noted that I’m projecting about 85-90% of team minutes, where the remainder go to replacement level players. Or in James Wiseman’s case, below replacement level.

[4] USG * MIN / team minutes

[5] You could break this down much further into things like at rim finishing, at rim defense, off the dribble shooting, etc etc. Dereliction in any area can leave you vulnerable to predictability, an NBA no-no.

[6] Unlike most forms of poker, which are solved af.

[7] The base of this model is EPM, which passes the smell test as the best available one number metric. I've added an aging curve, taken into account other publicly available advanced stats, and added some common sense on players who have had injuries derail recent seasons.

[8] These projections assume that James Harden plays basketball for the Philadelphia 76ers. That, uh, seems unlikely.