Games

Week 12, 2024

Last week recap

About

nfelo is a prediction model built on top of 538's Elo framework that uses unique dynamics about the NFL to improve prediction accuracy. It is one of the (if not the!) most accurate public models available on the internet.

The tables below show model performance for 538, nfelo, and the Vegas line across a variety of measures. "Straight up" describes a model's ability to pick winners, "RMSE" describes a model's ability to predict the final margin of victory, and "ATS" describes a model's ability to beat the spread.

Straight up and RMSE take all games into consideration while ATS only considers games rated break-even or better. Break-even for nfelo is defined by expected value, while 538 is defined by games with a delta between projected line and market line of over 1.5.

All Time (Since 2009)

538nfeloVegas (Open)Vegas (Close)

Straight Up %

% of winners picked correctly

63.0%

64.2%

62.7%

64.7%

RMSE

Error in predicting final margin (lower is better)

13.39

13.17

13.30

13.19

ATS (Open)

Record against the opening line on break-even or better plays

52.7%

57.3%

ATS (Close)

Record against the closing line on break-even or better plays

51.5%

54.4%

By Season

538nfeloVegas (Open)Vegas (Close)

Straight Up %

% of winners picked correctly

61.1%

68.1%

67.9%

69.7%

RMSE

Error in predicting final margin (lower is better)

13.27

12.97

13.21

12.86

ATS (Open)

Record against the opening line on break-even or better plays

51.5%

48.9%

ATS (Close)

Record against the closing line on break-even or better plays

45.1%

47.8%

Cumulative Charts

Brier score relative to the closing Vegas line. Brier Score is a modified version of RMSE that measures how accurately the model projects the final margin. Higher is better.

nfelonfelo (unregressed)538538 (old model)Vegas (Open)Vegas (Close)Season2,0122,0142,0162,0182,0202,0222,024-4,000-3,000-2,000-1,000