Model Performance · NBA

Is a 65% projection actually winning near 65%? Calibration is checked out-of-sample on prior games only.

Selections
4044
Record (W-L-P)
2280-1764-0
Win %
56.4%
ROI (simulated)
+9%
364.14 units
Avg model prob
63%
Avg implied prob
51.8%
Brier score
0.245
lower is better calibrated
Max drawdown
-37.6 u
longest loss streak 9

Simulated equity curve (units)

4122131415161728395108121134148162175188203218232247262277292307322337352367381400-8081624

Calibration by probability bucket

BucketNActualPredicted
50-55%99146.3%52.4%
55-60%85653.5%57.4%
60-65%67057.2%62.4%
65-70%57058.2%67.4%
70-75%41964.7%72.3%
75-100%53870.1%80%

By market

MarketNWinsWin %
assists36018551.4%
blocks967275%
pa36020356.4%
points36018651.7%
pr36019353.6%
pra36019353.6%
ra36020556.9%
rebounds36020556.9%
steals34825372.7%
stocks36019754.7%
threes36018751.9%
turnovers36020155.8%

Walk-forward: each prediction used only games played before the graded game. Results are simulated on DEMO DATA. Model v1.0.