Bankroll Management: The Survival Skill That Beats Picking Winners
Your picks decide how fast you win; your bankroll decides how long you survive
Start with the math of losing streaks
Even with a 55% long-run win rate (already elite), the chance of hitting at least one 6-game losing streak within 100 bets is still over 50%. Stake 20% of your bankroll per bet and a 6-game skid busts you outright — not because you lacked an edge, but because your position size never let you survive to see the mean revert.
Three mainstream approaches
| Method | How it works | Characteristics |
|---|---|---|
| Flat staking | A fixed amount per bet (e.g. 1-2% of your bankroll) | Simplest; best for beginners |
| Percentage staking | A fixed percentage of your current bankroll per bet | Scales up on wins, down on losses; never goes to zero but recovers slowly |
| Kelly criterion | Stake fraction = (win rate × odds − 1) ÷ (odds − 1) | Theoretically optimal growth, but extremely sensitive to errors in your win-rate estimate |
In practice, use "half Kelly" (divide by 2) or even quarter Kelly, because your estimated 55% might really be 52%.
Practical recommendations
① Stake 1-2% of your bankroll per bet, with total daily exposure capped at 5%; ② Set aside a separate, smaller entertainment budget for parlays (see the parlay guide); ③ During a losing streak, the single most important discipline is not chasing losses — chasing means sizing up exactly when your emotions are at their worst; ④ Log every bet (odds, your reasoning, the result), and after a season you'll know your true hit rate — which is the only reliable input the Kelly criterion can actually use.
Frequently Asked Questions
How much should I stake per bet?
The common recommendation is 1-2% of your total bankroll. This lets you absorb an extreme run of 20+ losses without damaging your capital.
Can I use the Kelly criterion straight up?
Full Kelly isn't recommended — it assumes your win-rate estimate is accurate. In practice, use half Kelly or quarter Kelly to buffer against estimation error.