Quote:
I would simulate it using a BJ simulator which I have
My BJ simulator (CVSIM) will give me the the correct percentages for unit bets won/lost/, doubles won/lost, and splits won/lost. We know how many blackjacks we win. But unfortunately, the simulator doesn't actually simulate risk of ruin; it just uses the same formula that I described above based on the EV and standard deviation (I checked it) which may not be accurate for our case. So I would have to write my own sim in R and input the results from CVSIM. This shouldn't be so bad since I wouldn't be simulating the entire game of blackjack, just a game with the correct payouts and losses.
This is actually an important problem for which many people would be interested in some tabulated results. It would also be interesting to see at what point the risk of ruin formula breaks down and by how much. The risk of ruin formula will provide a good sanity check on the simulation.
The risk of ruin formula (which may not turn out to be accurate) works as follows. We actually want to compute the risk that the house loses a 1 unit bankroll before it wins a B unit bankroll, where B is 1,2,...12. Let ror be the risk that it loses a 1 unit bankroll if it played forever (not stopping after it wins B), and r be the risk that it loses 1 before winning B. Then we have
ror = r + (1 - r)*ror
(B+1)
That is, the house's risk ror of losing 1 unit playing forever is the risk r that it loses 1 unit before winning B, plus the probability it wins B units before losing 1 unit (1-r) times the risk of ruin after that point which is the risk of losing a B+1 unit bankroll ror
B+1. Solving for r gives
r = (ror - ror
B+1) / (1 - ror
B+1)
ror is computed from the standard risk of ruin formula
ror = exp(-2*EV*1/SD
2)
Where the 1 is the house's bankroll. For example, for a house edge of 1.2%, we have EV = 0.012. We'll take SD = 1.16 which is about right for a 6 deck game. This gives
ror = 0.982322209
Now if you have a 12 unit bankroll, use the equation for r with B = 12. This gives
r =~ 91.5%
So the player would have a 91.5% chance of winning 1 unit before losing 12 units according to this formula. That agrees with the number CVSIM spits out. We shall see how this compares to the simulation results, but I would still need the details.
Last edited by BruceZ; 01-05-2012 at 12:50 PM.