NBA Model Accuracy

Out-of-sample backtests of the per-game projection engine — projections vs. what actually happened, compared to the Marcel baseline
Per-game projections
Beat the Marcel baseline by 4% on per-game accuracy · p10–p90 coverage 66% (2025-26 backtest, 2,148 preds)
Player Accuracy
per-game rate vs naive · 2,148 predsskill 4%p10–p90 cov 66%
StatNModel/gBase/gSkillCov
Scoring
Points3582.532.68+6%64%
3PM3580.350.36+5%66%
Playmaking
Assists3580.70.73+5%63%
Defense & Boards
Rebounds3580.860.89+4%68%
Steals3580.20.2-0%65%
Blocks3580.140.15+5%67%
Team Game Model vs Home-Court Baseline
backtest
SeasonGamesBrierBaseAccMargin MAETotal MAE
2025-261,1610.20460.247469.0%11.5516.2
2024-251,1620.21310.248066.3%11.3516.0
2023-241,1530.21330.248264.4%11.2115.6
2022-231,1570.23170.243962.5%10.3815.0
2021-221,1590.22720.248363.6%11.3017.9
2020-211,0340.23000.248961.2%11.0624.3
2019-208900.21510.247966.2%10.4415.7
Overall7,7160.21930.247564.7%11.0617.2
Opponent-adjusted point-differential ratings (SRS) → predicted margin, win probability, and total, scored walk-forward (ratings built only from games played before each matchup — no leakage). A lower Brier than the home-court baseline means real predictive skill. The model never sees the betting line.