The NBA board opens: why a 65% model still loses at −200
The NBA hub opens today, two weeks before the regular season starts on 2026-10-21. Game lines fill in as the schedule comes into range. Before reading any of them, it helps to know how little a good prediction is worth at the wrong price.
Accuracy is not profit
We ran a walk-forward rating model across more than 1,200 regular-season games from a full NBA season. It called about 65% of winners. Bet at closing prices, it lost about a third of a simulated bankroll. The model wasn't bad at picking winners; it picked favorites, and favorites are priced to match.
Where 65% stops paying
Hold the win rate at 65% and change only the price, and the expected result per $100 staked runs like this:
- −110: +$24.09
- −150: +$8.33
- −170: +$3.24
- −186: −$0.05, roughly break-even
- −200: −$2.50
- −250: −$9.00
- −300: −$13.33
A 65% hit rate makes money at −170 or better and loses at anything shorter. In our study, NBA moneyline favorites routinely sat between −200 and −400 — which is how a model can be right two times in three and still lose.
What the hub shows instead of a pick
- Each game's consensus price with the margin removed, beside the best price actually available, so the cost of the margin is visible game by game.
- The consensus spread and total as the line most books quote — never a quarter point.
- The move since the line opened. Movement is market information, not a tip.
- An AI read of each matchup. Nothing on the hub is a pick.
What we haven't measured
We haven't measured the NBA's in-season margin, so we won't estimate it. Our margin study covered six two-way markets over roughly six weeks and found consensus margins clustered between 4.54% and 4.78%. The NBA wasn't in it. That figure gets published once a season of in-season prices exists, whatever it says.
Bankroll Guardian is a bet-tracking and analytics tool — not a sportsbook, and none of this is betting advice. Betting carries risk; please bet responsibly.