This is literally impossible.

Unskilled (naive 50:50) is 75%

If you take sports (most frequent event)

60% is brier of 0.16 so 84%

65.35% means predicting half the variance which is near the theoretical limit (since sports are somewhat random with the true Vegas odds usually being between 10 and 90%)

so that’s only half the way from 75% so 87.5%

90% on this scale means questions being asked on Kalshi are 90/10 questions which is quite unlikely

@ChinmayTheMathGuy the concept of the benchmark is cool but the results aren’t that clear / useful

Brier should be -300% to +100%

so 90% means explaining 60% of the variance.

Is this saying every model loses money since they’re all less than 100%

@ChinmayTheMathGuy also

the most risk averse version (gamma = 1)

uses logarithmic utility which is full Kelly which is still too risky for most, fractional Kelly is much better since the bets give some credence to the market.

gamma=2 roughly corresponds to half kelly

@kiudee I think you mean 1 - brier score, which is what they report. a 10% or less brier score (less is better) would mean a 90% or more score on the Prophet Arena

that's minor and presumably nobody would be confused but probably worth clarifying the description

To confirm, that's what I meant. Thanks for the keen eye.