This is a trivial known problem but for every one that is not trolling (the old classic martingale) and is for real, there will be a first time of course this came as a thought.
How much money do you have to use for this? If you have 2^n-1 available money and you start with 1 unit of bet and then go to 2, 4 etc until you win, then what happens if you hit n times the opposite of what you are betting for?
You would have used after n bets 1+2+4+..2^(n-1)=2^n-1 ie all your money. This is why i chose the total to be that way so that the last bet would still be enough to recover everything lost up to that point+1 unit of profit.
So all it takes to run into trouble is to use all your money and then nobody will help you to bring the next bet to deliver a possible victory. What happens then? You have lost 2^n-1, that is what happens.
So if you have a chance to win that is p per bet then with a chance (1-p)^n you lose all your money in that trial meeting a deadly n-streak and with chance 1-(1-p)^n you will always walk away with +1.
As a result your expectation for this system is
(1-(1-p)^n)*1-(2^n-1)*(1-p)^n
For this to be larger than 0 ie the strategy to make sense as a plus EV effort (it may still be a bad way to play though even then as you will eventually learn/recognize)
you need to have (1-(1-p)^n)>(1-p)^n*(2^n-1)
First of all realize that if p<1/2 no amount of betting (in a game that you win 1 with probability p and lose 1 with probability 1-p) will ever make playing that game positive in return and no sum of any such bets however structured will lead to an average result that is larger than 0. This is because the average of the sum (of any sequence of bets) is the sum of the averages and the average each time for any bet is negative so the average of the sum which is the sum of the individual averages will be negative too no matter how you construct the sum.
You see whatever x you bet the expectation will be E=x*p-(1-p)*x=x*(2*p-1) and since p<1/2 => 2p-1<0 so E<0. An elaborate sequence of bets wont change the average result.
What if p=1/2
In this case
(1-(1-p)^n)*1-(2^n-1)*(1-p)^n becomes 1-1/2^n-1/2^n*(2^n-1)=1-1/2^n-1+1/2^n=0
This is expected of course because any bet you make in that game has expectation 0 since its 50-50.
Now things can get better if p>1/2 but you do not have this here.
Even then this is not the best way to use your money when p>1/2 as you will tend to be bankrupt typically even then unless p or n or both are very large.
What would p need to be to have a nonzero probability to never lose? This may be true because although there is always a chance to have a streak that kills everything, your position also grows every time you win so you will eventually get big enough that the chance to deplete everything is much smaller than what it was when you started.
For example if p>1/2 one can claim that
To double your bankroll from K=2^n-1 to say 2*(2^n-1)=2^(n+1)-2 you need to survive 2^n-1 attempts of this strategy (
https://en.wikipedia.org/wiki/Martin...betting_system) )
The chance to survive them is P(n)=(1-(1-p)^n)^(2^n-1)
As a result the chance to survive an infinite sequence of doublings (ie never die) when you start with 2^n-1 bankroll is
S= survival probability
S=P(n)*P(n+1)*P(n+2)....
Log S= Sum[(2^k-1)*log(1-(1-p)^k), {k,n, Infinity}]
Given that (1-p)^k is typically very small<<1 if we pick 2^n-1 the starting bankroll big enough
We can say log(1-(1-p)^k)~-(1-p)^k
So LogS~Sum[(2^k-1)*(-(1-p)^k), {k,n, Infinity}]=
Sum[-(2*(1-p))^k+(1-p)^k,{k,n,Infinity}] =
= -(((1 - p)^n (1 - 2 p + 2^n p))/(p (-1 + 2 p))) (When p>1/2)
So survival probability S of infinite doublings is
S~e^(-(((1 - p)^n (1 - 2 p + 2^n p))/(p (-1 + 2 p)))
So eg for p=0.55 and n=10
S~e^(-3.486)~0.03 or 3%
But if p=0.55 and n=20
S~e^(-1.21577)~29.6%
If p=0.55 and n=40 (trillionaire)
S~e^(-0.1478)~86%
See even then with such edge its kind of risky and much better ways exist to play in those cases that not only grow you faster but can take the ruin probability to much smaller levels even with real life min bets restrictions in place. As always see this
https://en.wikipedia.org/wiki/Kelly_criterion
PS: Hopefully i didnt make any stupid math error in these last parts. Will revisit if so.
Last edited by masque de Z; 06-16-2016 at 12:41 AM.