Squawka Signal is our in-house NBA model. It gives every team a probability — to make the playoffs, win its conference and win the Championship — and a projected final record. We then set those numbers against the live prediction market, so you can see where our model and the money disagree. Here is how it works, in plain terms.
Team power ratings
At the core is a single strength rating for every team, built with the Elo system — the same method chess uses and FiveThirtyEight popularized for the NBA. Win and you take rating points off the team you beat; win by more, or beat a stronger side, and you take more. The ratings run on every game of last season and this one, and update after every result. Home advantage is built in — we calibrate it to the modern league, where home teams win a little under 55% of the time, rather than the bigger edge of past eras. At the start of each season every team is regressed a quarter of the way back toward the league average, because last year’s form only tells you so much about this year’s roster.
Simulating the season
From those ratings we play out the rest of the season ten thousand times. Each simulation runs every remaining game, then the play-in tournament, then the full best-of-seven playoff bracket round by round, and finally the Finals, with home-court going to the better regular-season record. Count how often each team reaches each stage across all ten thousand runs and you have its playoff, conference and Championship probabilities, plus a projected win total. It all refreshes as results land, so the numbers move with every night’s games.
Why the favorite isn’t 50%
A team rating is a best estimate, not a certainty, so we don’t treat it as exact. Every simulation nudges each team’s strength up or down by a random amount before the games are played, which spreads the outcomes realistically. It is why even a clear favorite sits around one in five rather than running away with it: winning four best-of-seven series in a row is genuinely hard, and the model reflects that. Tighten that uncertainty and the favorites look stronger than the market believes; our setting is tuned so the top of the board lines up with what Kalshi and Polymarket are actually paying.
What we publish
For the season the model gives each team’s playoff, conference and Championship odds and a projected record. We set the title odds against the live prediction-market price — a de-vigged, volume-weighted blend of Kalshi and Polymarket — and flag the gap: where our probability sits above the market we call it value, below it the market is warmer than us. It is a read, not a tip.
What it does not do
A word on what the model is not. It rates teams on strength and results and simulates the whole season; it does not try to price who is injured or resting on a given night, and in the NBA that is the single biggest thing a pure ratings model can miss — one star out can swing a title race. It is not a tip service, and on plenty of teams it will simply agree with the market. Treat it as one informed input, not a guarantee.