Bankroll maths: streaks, staking and the risk of ruin
A bankroll is the money set aside for betting. The maths of how it rises and falls is unforgiving, and it explains why staking plans cannot rescue a negative expected value.
Losing streaks are longer than they feel
Most people underestimate how long a run of losses can get by pure chance. The table below gives the exact probability of at least one losing run of a given length, for bets that each have a 50% chance to lose. The numbers come from a step-by-step calculation over every bet, not from a simulation.
| Losing run | Chance within 100 bets | Chance within 500 bets |
|---|---|---|
| 4 or more | 97.3% | > 99.9% |
| 5 or more | 81.0% | > 99.9% |
| 6 or more | 54.6% | 98.3% |
| 7 or more | 31.8% | 86.4% |
| 8 or more | 17.0% | 62.5% |
| 10 or more | 4.4% | 21.5% |
So in 100 coin-flip bets, a run of 6 or more losses in a row happens 54.6% of the time. Nothing has gone wrong when that happens. It is simply what independent 50/50 events look like, and a bankroll has to survive it. The gambler's fallacy guide looks at why streaks feel meaningful when they are not.
Staking as a share of the bankroll
A common way to describe stakes is as a percentage of the bankroll, or in fixed "units". The smaller the stake relative to the bankroll, the more losing bets it takes to empty it. With flat stakes of 2% of the starting bankroll, the bankroll is gone once losses exceed wins by 50 bets. With flat stakes of 10% it takes a gap of only 10.
Stake size changes the speed of the swings, not the direction. If each bet has a negative expected value, the expected loss per unit staked stays the same however you size the bets. Smaller stakes only make the decline slower and smoother.
The Kelly formula
The Kelly criterion is a formula from information theory. It gives the share of the bankroll that maximises the long-run growth rate, if you know the true probability and the bet has a positive edge. With decimal odds d, b = d − 1 and win probability p:
Worked examples
- p = 0.55 at 2.00 (b = 1): (0.55 − 0.45) ÷ 1 = 10.0% of the bankroll.
- p = 0.55 at −110 (decimal 1.9091, b = 0.9091): 5.5%.
- p = 0.50 at −110: the formula gives -5.0%. A negative answer means the bet has negative EV and the correct stake is zero.
Two warnings come with Kelly. First, it assumes p is known. If your estimate is too optimistic, Kelly tells you to stake too much, and over-betting is worse than under-betting. That is why people who use it often bet a fixed fraction, such as half of the Kelly amount. Second, even exact Kelly staking produces deep drawdowns along the way. It is a statement about the long run, not a safety device.
The risk of ruin
A classical problem in probability, known as gambler's ruin, asks: if you keep making the same bet until you either reach a target or lose everything, how often do you reach the target? For a bet that wins with probability p and pays even money, starting with i units and stopping at N units:
Worked example: trying to turn 50 units into 100
Bet on red on a single-zero wheel: p = 18/37, so q/p = 19/18. Starting with 50 units, staking 1 unit at a time and stopping at 100:
P(reach 100) = 6.3%.
If the game were fair (p = 0.5) the answer would be 50%. Staking all 50 units on one bet instead gives 18/37 = 48.6%.
This is a famous result. When every bet has a negative edge, many small bets give the edge many chances to work, so the long slow route to a target is far less likely to succeed than the short one. The point is not that large bets are a good idea. It is that the more you stake in total, the more of the margin you pay. The long-run guide explains why.