🐻⬇️🏀

2025-2026 NBA Model Performance Analysis

Scope
All Games Ranked Games

All scored games in the selected league and season. AP Poll is excluded here.

Season
2025-2026

Comparing prediction accuracy across 1257 games using multiple rating models.

Model Catalog

7-day holdout coverage: 16/17 models .

Rolling Holdout Curves

Each point is a strict weekly holdout: train on all games before that week, test on that week. This first version uses a 21-day warmup, then 7-day holdouts stepped forward weekly.

Log Loss Brier AUC Accuracy

Weekly strict holdout log loss. Lower is better. Showing 16 models across 26 windows. Click legend items to hide/show series.

Recent Window Winners

Holdout Best Log Loss Runner-up Models
Apr 15 - Apr 21 Home Team Baseline 0.646 Bradley-Terry Recency (0.709) 16
Apr 8 - Apr 14 Elo 0.522 Dynamic Bradley-Terry (0.529) 16
Apr 1 - Apr 7 Elo 0.479 Adjusted Context Blend (0.493) 16
Mar 25 - Mar 31 Dynamic Bradley-Terry 0.510 Adjusted Context Blend (0.510) 16
Mar 18 - Mar 24 Elo 0.475 Recency Ensemble (0.502) 16
Mar 11 - Mar 17 Adjusted Context Blend 0.560 Dynamic Bradley-Terry (0.562) 16
Mar 4 - Mar 10 Adjusted Efficiency 0.590 Log Adjusted (0.590) 16
Feb 25 - Mar 3 Log Adjusted 0.495 Adjusted Efficiency (0.495) 16

Model Performance Leaderboard

Models ranked by strict holdout AUC when available (fallback: full-season AUC). Hover over column headers for explanations.

# Model 7d Split AUC Acc Brier LogLoss n AUC 7d Acc 7d Brier 7d n 7d
1 Efficiency Efficiency Tempo-adjusted efficiency version of Pythagorean ratings. More → FULL
no 7d
 0.752 68.1% 0.203 0.591 1106 - - - 0
2 Home Team Baseline Home Team Baseline Always favor the home team with a fixed prior. More → STRICT
18g
 0.550 55.0% 0.250 0.693 1106 0.662 66.7% 0.227 18
3 Elo Elo Streaming paired-comparison rating with recency baked into sequential updates. More → STRICT
18g
 0.743 68.5% 0.206 0.600 1106 0.588 55.6% 0.273 18
4 Adjusted Context Blend Adjusted Context Blend Experimental context-heavy win model blending strong team components with rest and venue context. More → STRICT
18g
 - - - - 0 0.562 50.0% 0.263 18
5 Dynamic Bradley-Terry Dynamic Bradley-Terry Time-evolving paired-comparison model with latent team strength drift. More → STRICT
18g
 - - - - 0 0.537 61.1% 0.257 18
6 Bradley-Terry Recency Bradley-Terry Recency Static Bradley-Terry with exponential recency weights on newer games. More → STRICT
18g
 - - - - 0 0.500 61.1% 0.258 18
7 Core Ensemble Core Ensemble Equal-logit blend of Elo, recency BT, recency margin, log-adjusted pyth, and points off/def. More → STRICT
18g
 - - - - 0 0.500 44.4% 0.273 18
8 Recency Ensemble Recency Ensemble Equal-logit blend of Elo, recency BT, recency margin, log-adjusted pyth, and recency points off/def. More → STRICT
18g
 - - - - 0 0.500 50.0% 0.275 18
9 Margin Margin Linear team-strength model fit on point differential instead of binary wins. More → STRICT
18g
 0.743 67.7% 0.211 0.611 1106 0.463 44.4% 0.268 18
10 Margin Recency Margin Recency Margin regression with exponential recency weights on newer games. More → STRICT
18g
 - - - - 0 0.463 38.9% 0.285 18
11 Points Off/Def Points Off/Def Raw points regression with separate offensive and defensive team parameters. More → STRICT
18g
 0.751 67.5% 0.208 0.605 1106 0.463 44.4% 0.268 18
12 Points Off/Def Recency Points Off/Def Recency Off/def points regression with exponential recency weights. More → STRICT
18g
 - - - - 0 0.463 38.9% 0.283 18
13 Adjusted Efficiency Adjusted Efficiency Opponent-adjusted efficiency model with separate offensive and defensive components. More → STRICT
18g
 0.749 67.9% 0.204 0.594 1106 0.438 44.4% 0.295 18
14 Log Adjusted Log Adjusted Log-scale adjusted efficiency model that downweights blowout leverage. More → STRICT
18g
 0.749 67.9% 0.204 0.594 1106 0.438 44.4% 0.295 18
15 Pythagorean Pythagorean Pythagorean win expectation from raw points scored and allowed. More → STRICT
18g
 0.751 68.4% 0.214 0.620 1106 0.425 44.4% 0.264 18
16 Avg Margin Baseline Avg Margin Baseline Predict from simple average scoring margin in the training window. More → STRICT
18g
 0.763 69.6% 0.201 0.589 1106 0.425 44.4% 0.280 18
17 Bradley-Terry Bradley-Terry Static logistic paired-comparison model with one team strength parameter. More → STRICT
18g
 0.748 68.2% 0.205 0.598 1106 0.412 38.9% 0.262 18

Methodology

ELO / Bradley-Terry

  • ELO: Iterative updates, K=64, HCA=100
  • BT: Static logistic regression on all games
  • Both model win probability, not margin
  • ELO updates after each game; BT fits once

Pythagorean Models

  • Raw: Classic points scored/allowed formula
  • Efficiency: Pace-adjusted (pts per possession)
  • Adjusted: Opponent-adjusted efficiency
  • Log: Log-linear multiplicative scale

Margin Regression

  • Team-level ridge regression on point margin
  • Linear Bradley-Terry (margin target)
  • Alpha=0.05 (CV-tuned)
  • Learns home advantage from data (~6 pts)

Baselines

  • Home Team: Always predict home wins (60%)
  • Avg Margin: Higher average margin wins
  • Models should beat these to add value