I saw Peep mention to you a simple formula that can be used to figure out your chnaces of going broke. Do you know anything about that?
I assume you're referring to this statement by Peep:
Originally Posted by
Peep
If we knew that we could win say 30% of our bets at average odds of 9/2, [the formula] let us know what our chance of going broke was if we bet ____ percent of our bankroll.
The simplicity of the formula to which he's referring, however, is rather muted by its general lack of utility.
The risk-of-ruin approximation (it's only an approximation because it ignores the higher-order moments of a betting distribution and then appeals to the Central Limit Theorem, which will only be true asymptotically) is given by:
R ≈ e-2μ(x) / σ2(x)
where R represents our risk-of-ruin, μ our expectation, and σ our standard deviation (the latter two both given in units of initial bankroll and as a function of bet size, x).
Note that even as an approximation this is only really applicable for constant μ and σ. While this assumption remains robust across multiple classes bets of varying size, it won't hold for a player sizing his bets as a percentage of bankroll. Furthermore this formulation refers to a player's probability of going broke before the End of Time, rather than of going broke before reaching a certain goal. As such, it's of no use to the recreational bettor (whose probability of eventually going broke approaches 100%). We see this by noting that when μ < 0, the exponent goes positive and R would be > 1, a logical impossibility.
So for a bet size of x (as a fraction of initial bankroll) we have:R ≈ e-2μ / (x * σ2)
Recall that with binary outcome wagers, if p represents the win probability, d the decimal payout odds, and x the bet size, then we have:
μ(x) = x * (p*d - 1)
σ2(x) = x2 * (d - μ/x - 1) * (1 + μ/x)
σ2(x) = x2 * d2 * (1-p) * p
So in Peep's example of a 9/2 dog winning at a 30% clip we'd have for a unit bet size:
μ = 11/2 * 30% - 1 = 65%
σ2 = (11/2)2 * 70% * 30% = 6.3525
e-2μ / σ2 = e-2 * 65% / 6.3525 ≈ 81.494%
So our risk-of-ruin approximation would be given by
R ≈ 81.494% 1/x
So for a player betting 5% of bankroll, risk-of-ruin would be 81.494%
1/5% ≈ 1.669%, while for a player betting 25% of bankroll it would be 81.494%
1/25% ≈ 44.106%.
Note that we can restate the above formula as follows:
R = e-2μ / (x * σ2)
ln(R) = -2μ / (x * σ2)
x = -2μ / ( ln(R) * σ2)
x = -2*(p*d - 1) / ( ln(R) * d2 * (1-p) * p)
Which tells us our maximum bet size if we're willing to accept a given risk-of-ruin.
So using the above example of a 9/2 dog winning at 30%, if we're willing to accept a risk-of-ruin no greater than R*:
x ≤ -20.464% / ln(R*)
So for R* = 5%, we'd have:
x ≤ -20.464% / ln(5%) = 6.831%
Implying we'd need to select a bet size of no more than 6.831% of initial bankroll in order to keep our risk-of-ruin at 5% or less.