A large number of people have developed models for predicting the point spreads of college basketball games. For those that have made their picks publicly available, ThePredictionTracker does a great service by tracking the live performance of each model over the course of the season. But, it's difficult to do an apples-to-apples comparison using the the Tracker, since each model has predicted a different subset of games (mostly this is random/accidental, but some models start submitting picks later in the season). Here, I show a subset of models which have made picks since the beginning of the season, and I throw out games for which any of those models did not make a pick. In a (somewhat lazy) attempt to address misprinted lines (which have appeared occasionally), I filtered out any games for which the opening and closing Vegas lines differed by more than 5 points (it's very rare that this really happens).

Results are shown as of 2020-03-13 for a set of 2967 games. First, let's look at some statistical benchmarks. Below we show the mean-squared error (in predicting the margins of victory), the binary accuracy (in predicting the win/loss outcomes), and the average bias of each model:
Model Mean Squared Error (MSE)
Line 130.341
Erik Forseth 130.767
Opening Line 131.220
TeamRankings 134.255
Dokter Entropy 134.393
Sagarin Predictor 134.697
Sagarin Golden Mean 136.588
ESPN BPI 136.654
Sagarin Rating 136.780
DRatings.com 140.724
StatFox 141.467
Kenneth Massey 142.543
Sonny Moore 144.145
ComPughter Ratings 148.258
Sagarin Recent 149.613

Model Binary Straight Up (%)
Line 73.63
Opening Line 73.54
Erik Forseth 73.51
TeamRankings 73.10
ESPN BPI 73.00
Sagarin Predictor 73.00
Sagarin Golden Mean 72.85
Dokter Entropy 72.77
Sagarin Rating 72.72
StatFox 72.14
ComPughter Ratings 71.91
DRatings.com 71.87
Kenneth Massey 71.77
Sonny Moore 71.12
Sagarin Recent 70.63

Model Average Bias
Opening Line 0.007
Sagarin Golden Mean -0.039
Erik Forseth 0.060
Sagarin Predictor -0.135
Line -0.157
Sagarin Rating -0.175
ComPughter Ratings -0.191
Sagarin Recent -0.330
Kenneth Massey 0.349
StatFox 0.361
Sonny Moore -0.533
ESPN BPI 0.559
DRatings.com -0.596
TeamRankings 0.603
Dokter Entropy -0.934


We might also ask how each model would have done betting against the Vegas line, shown below:
Model Against the Spread (%)
ESPN BPI 51.52
DRatings.com 50.79
Erik Forseth 50.69
Dokter Entropy 50.56
StatFox 50.10
Opening Line 50.03
TeamRankings 50.00
Sagarin Golden Mean 50.00
ComPughter Ratings 50.00
Sagarin Rating 49.87
Sagarin Predictor 49.36
Sonny Moore 49.26
Sagarin Recent 49.09
Kenneth Massey 48.67


Let's regress the observed margins of victory onto the predictions of each model, while constraining the regression to have nonnegative coefficients. This gives the optimal (backward-looking and subject to nonnegativity constraints) mixture of predictors. We find:
Model Coefficient
Line 0.482
Erik Forseth 0.328
DRatings.com 0.098
ESPN BPI 0.092
Opening Line 0.000
Kenneth Massey 0.000
Sagarin Rating 0.000
Sonny Moore 0.000
TeamRankings 0.000
Dokter Entropy 0.000
StatFox 0.000
Sagarin Recent 0.000
Sagarin Predictor 0.000
Sagarin Golden Mean 0.000
ComPughter Ratings 0.000

The MSE of this hypothetical predictor would be 129.81. (This is overly optimistic for obvious reasons.)

Finally, we can ignore the other models and look only at games for which I made a pick this year, constituting a sample of 4112 games:
Model Mean Squared Error (MSE)
Line 123.489
Erik Forseth 123.931
Opening Line 124.659

Model Binary Straight Up (%)
Erik Forseth 72.67
Line 72.51
Opening Line 72.35

Model Average Bias
Erik Forseth 0.018
Opening Line -0.029
Line -0.190

Model Against the Spread (%)
Erik Forseth 50.63
Opening Line 49.40