BillC
11-19-2003, 04:28 PM
You should have about
100/k big bets
where k is your Kelly fraction in blackjack (perhaps as implied by your acceptable risk of ruin). This is equivalent to a utility function u(x)=-x^(1-1/k) for k>1 and log(x) if k=1. The utility specifies risk tolernace and can be expressed in terms of risk of ruin and other drawdowns. So if you play .5 Kelly, k=.5 (and you will eventually be reduced to half your original BR about half the time). Since most teams use around .3 Kelly. That puts the required bank at around 300 big bets, pretty close to the standard Malmuthian 300 BB, arrived at totally differently in GTOT. Virtually all bj teams have k between .25 and .35.
I am assuming one big bet per hour and a variance of 100 big bets per hour. The 100 can be replaced by var./EV in general.
The 100/k expression means that a full Kelly bettor (i.e. one with a log utility function) should be playing a game where the big bet is exactly 1% of the BR. This then is optimal Kelly betting for poker.
the value of k is a question of risk tolerance, which in turn a question of utility function. A convenient way to get a utility function for poker is to use "implied utility" from BJ. There is also a complicated estimator for the value of k based on historical betting and returns (I don't know if it is known in the lit.)
I jsut did this on the back of an envelope, so it might be wrong. Details to follow if there is any interest.
100/k big bets
where k is your Kelly fraction in blackjack (perhaps as implied by your acceptable risk of ruin). This is equivalent to a utility function u(x)=-x^(1-1/k) for k>1 and log(x) if k=1. The utility specifies risk tolernace and can be expressed in terms of risk of ruin and other drawdowns. So if you play .5 Kelly, k=.5 (and you will eventually be reduced to half your original BR about half the time). Since most teams use around .3 Kelly. That puts the required bank at around 300 big bets, pretty close to the standard Malmuthian 300 BB, arrived at totally differently in GTOT. Virtually all bj teams have k between .25 and .35.
I am assuming one big bet per hour and a variance of 100 big bets per hour. The 100 can be replaced by var./EV in general.
The 100/k expression means that a full Kelly bettor (i.e. one with a log utility function) should be playing a game where the big bet is exactly 1% of the BR. This then is optimal Kelly betting for poker.
the value of k is a question of risk tolerance, which in turn a question of utility function. A convenient way to get a utility function for poker is to use "implied utility" from BJ. There is also a complicated estimator for the value of k based on historical betting and returns (I don't know if it is known in the lit.)
I jsut did this on the back of an envelope, so it might be wrong. Details to follow if there is any interest.