Methodology · NBA
How the NBA model works
Playoff, play-in, conference, and championship odds for all 30 teams.
- Version
- v1 (Pythagorean net rating, regression 0.35; market win-total anchor + roster overlay)
- Updated
- Preseason (daily once the season tips off)
- Data source
- ESPN (records + point differential); market season win totals; documented roster overlay
- Calibration
- Preseason projection; game grading begins at tip-off
The question
Each team's odds to make the playoffs, avoid the play-in, reach the Finals, and win the title — plus a projected win total for the 82-game season.
The data
ESPN's free feed: last season's records and, crucially, point differential — the single best simple predictor of next-season strength.
How strength is measured
Talent comes from the Pythagorean expectation on points scored and allowed (exponent 13.91, the NBA-standard value), a net-rating proxy that's steadier than win-loss record.
Recency & regression
Because rosters turn over every summer and no games have been played, last season's rating is regressed 35% toward league average for the preseason projection. That pull shrinks once real games arrive.
Injuries & roster
Last season's point differential can't see the summer — trades, the draft, and free agency. Two mechanisms bring it in. Preferred: the posted market season win total, which already prices every roster move; when entered, the projection anchors hard to it (90% market) and supersedes everything else for that team (the same approach as our NFL model). Fallback, until a total is entered: a documented human overlay — a dated, reasoned win-total adjustment per team (a club that lands a star gets, say, a +12-win bump), applied on top of the regressed rating with a written reason on every entry.
Home advantage
A home bump is applied in the playoff series model (with the standard 2-2-1-1-1 home-court pattern for the higher seed).
The simulation
log5 turns two ratings into a game win probability. A 10,000-run Monte Carlo plays each team's 82-game season against a conference-weighted schedule, seeds both conferences, runs the play-in (7 vs 8, 9 vs 10, then the 8-seed game), and simulates four best-of-seven rounds to a champion.
Postseason format
Play-in tournament for seeds 7–10 in each conference, then a 16-team bracket of best-of-seven series through the conference finals and the NBA Finals.
Backtesting
This is a preseason projection, so there's nothing to grade yet; once the season tips off, game-level predictions will be locked and graded on the Ledger like MLB.
Known weaknesses
The 2026-27 schedule isn't published until August, so the season is simulated against a representative conference-weighted slate rather than the real one; there's no player-level modeling; and offseason roster moves rely on the human overlay above until we anchor to market win totals. The regressed rating is the floor, not the whole story.
Every published number is timestamped to the Ledger the moment it's made and graded against results — nothing here is backdated. Percentages are rounded for readability.