Elo

Just in time for the last few live games I got a crude Elo predictor working. As I detailed earlier Elo makes sense to be used in the framework of a Monte Carlo harness. Here’s the diff where we slot Elo in as a new MC_Predictor subclass. Results:

$ python3 ./mcc_schedule.py  -v
San José State 7 at USC 30 on Sep 04, 2021
Stanford 42 at USC 28 on Sep 11, 2021
Fresno State 40 at UCLA 37 on Sep 18, 2021
UCLA 35 at Stanford 24 on Sep 25, 2021
San Diego State 19 at San José State 13 on Oct 15, 2021
Fresno State 30 at San Diego State 20 on Oct 30, 2021
UCLA 62 at USC 33 on Nov 20, 2021
California 41 at Stanford 11 on Nov 20, 2021
Fresno State at San José State on Nov 25, 2021
California at UCLA on Nov 27, 2021
USC at California on Dec 03, 2021

Full Enumeration Simulation:
UCLA 2 [25%]
California 3 [37%]
Fresno State 3 [37%]

Monte Carlo [Sampled Home Margin Predictor] Simulation:
Fresno State 3598 [35%]
California 3296 [32%]
UCLA 3105 [31%]

Monte Carlo [Elo Predictor] Simulation:
California 1666 [16%]
Fresno State 7651 [76%]
UCLA 682 [6%]

Fresno State            2-0
California              1-0
UCLA                    2-1
San Diego State         1-1
Stanford                1-2
USC                     1-2
San José State          0-2
2021, 11, ,

The new Elo Predictor block consistently gives Fresno State a 75% chance to win the whole thing. That huge delta to the sampled margin predictor is because the cfbd elo data estimates Fresno has a near 90% chance of beating San Jose State in San Jose. Of course this first pass doesn’t factor in home field advantage (but as we just discussed, that might be a good thing.) It also doesn’t do “real life” scoring margins yet.

Since the Fresno State / San Jose State game is next up, we have the advantage of getting Las Vegas’s perspective about what the true odds are. The SJ State money line is +240, which converts to an implied chance of 29% to win the game. We can see the weaknesses of our two predictors on a spectrum. CFBD Elo data applied naively with no home field advantage says 10%. Sampled margin that only uses home field advantage says 55%. The best market estimation is 29%. Oddly enough that’s just about between the two.

For Cal at UCLA on Saturday we can also get the Vegas prediction. Elo says UCLA has a 71% chance and Vegas says 72%.

There are a few easy tweaks we can make to the Elo predictor that we’ve already identified so let’s refine it before we worry too much.