Sit and go and the Kelly criteria

[BobCarr]

I am trying to decide how the kelly criteria applies to choosing the size of the sit and go you play. In theory,if all players have equal likelyhood to win, you would be guaranteed a loss equal to the fee for the tourney. If I am better than the other players, I should get a win in the long run. Any ideas on how to handle this?

[BillC]

Figure your win rate r and variance v for large sample. Your bankroll should be v/r for full Kelly. So lets say you make 10$ per tourney with a standard deviation of 100$, your optimal bank will be 10^4/10=1000. There is a paper about this at bjmath.com.
Scale down for fractional Kelly.

[Doug Funnie II]

I tried to tackle this problem earlier, except for 18 tabling and increasing the stakes as my bankroll grows. The problem I ran into was that I couldn't accurately predict my winrate at higher stakes for which I don't have a meaningful sample yet. Also if you hit a rough patch, you cannot move down stakes enough to guarantee the success of using Kelly, and the jumps between levels makes a lot of the computations by Kelly meaningless. I figured it would probably just be better to trust STTF convention of move up at 30 buyins and down at 20.

[BillC]

But that 20-30 buy-in rule is based on the Kelly criterion.
Whether you like it or not, your choice of stakes is based on some estimate of your win rate or SD (unless you are overbankrolled or risk-loving). It is probably better to err on the conservative side and use maybe a 30-40 rule, since as you say, you don't really know your win rate, esp. when you move up.
I don't see why discrete jumps renders Kelly meaningless. In sit-n-gos, Kelly is maybe more relevant b/c of the relatively low SD.