Tuesday, December 2, 2014

T-Rank versus Vegas 12.2

Since T-Rank predicts the score of every D1-D1 college basketball game, you might be wondering whether you should take T-Rank to Vegas and try to beat the house.

I frankly don't have the time to figure this out -- managing T-Rank eats up all my hobby time, and I am not a gambler. But I did look at the results over last weekend. There were 100 non-pushed games where T-Rank differed from the consensus Vegas line by at least a half point, and T-Rank's suggested pick covered in 52 of them. So that's 52% over a small sample size, and it tells us basically nothing.

I will tell you that my intuition is that T-Rank is useless for gambling, because the people who set the lines probably have something similar to but better than T-Rank, and they can adjust it to account for news like injuries, momentum, etc.

Of course when T-Rank and Vegas disagree by a point or so, I'd expect the results to be essentially random (particularly because T-Rank predicts scores in whole numbers).

But it might be fun to track the biggest outliers, where T-Rank disagrees with Vegas by a large margin. In those scenarios either the conventional wisdom is missing something or T-Rank is missing something. Usually it's going to be T-Rank, but it's also possible that T-Rank could be on to something for a given team. That's the only way I could see T-Rank being helpful to a gambler: if a certain team is consistently under- or over-performing, and T-Rank predicts it, that might be a sign that T-Rank sees something eludes human eyes. Unlikely, but possible. (And you can bet that the window of any such advantage would close very fast.)

So here are the top five games with the biggest "delta" between T-Rank and the Vegas line. Conveniently, three of them are B1G-ACC challenge games - and in each of those, T-Rank is pulling for the ACC. #bias

Matchup T-Rank Prediction Vegas Line T-Rank Suggests Delta
269 UT Arlington @ 7 Texas Texas, 90-58 (99%) Texas -23.5 Texas -23.5 8.5
125 Central Michigan @ 234 Bradley Central Michigan, 66-63 (64%)  Bradley -3.5 CMU + 3.5 6.5
14 Syracuse @ 29 Michigan Syracuse, 61-60 (51%)  Michigan -4.5 Syracuse + 4.5 5.5
47 Pittsburgh @ 80 Indiana Pittsburgh, 72-71 (51%)  Indiana -2.5 Pitt + 2.5 3.5
52 Minnesota @ 99 Wake Forest Minnesota, 65-64 (54%)  Minnesota -4.5 Wake + 4.5 3.5

Subjectively, I do not like Syracuse over Michigan (though a close game is by no means out of the question), and Wake has been so horrible this year that I'd hesitate to bet on them even at home without more points. Texas has an injured player, which might be effecting that line, and the top of T-Rank has frankly gotten a bit out of whack (next year I will probably adjust the system so that preseason ratings maintain more influence longer into the season to avoid this). The reversal of the CMU - Bradley line is intriguing, though.

Anyhow, I'll be following these games to see whether T-Rank looks like an idiot or a genius.


  1. Rough night for T-Rank, as only the Syracuse game hit, and none of the others were particularly close. Wake Forest was just as awful as I thought they were. Pitt was really terrible last night, and it looks like a lost seasons for them. Texas struggled mightily against UT-Arlington - the game was never in doubt, but it was about 9-10 spread throughout the second half.

    Maybe we could make money looking or T-Rank outliers and betting against them?

  2. Today's T-Rank Picks:

    San Fran +11 over Colorado. T-Rank says Colo by 6.
    Fresno St +11 over UC Irvine. T-Rank says UC Irvine by 6.
    Cal -13.5 over Montana. T-Rank says Cal by 19.
    Butler -7.5 over Indiana St. T-Rank says Butler by 12.
    Idaho +8 over Wash. St. T-Rank says Wazzou by 4.

  3. 3-2 yesterday, so back within shouting distance of respectability. No predictions tonight, as I spent my time fiddling with the algorithm, and there weren't many games.

  4. Picks for today:

    Butler -24.5 (T-rank: -29)
    Providence -15.5 (T-rank: -19)
    Purdue -14 (t-Rank: -18)