Betting analytics · Risk

Bankroll management: flat staking vs the Kelly criterion

More winning systems are destroyed by stake size than by bad selections. The picks decide the direction; the staking decides whether you are still there when it arrives.

Published 3 September 2026 · ~8 min read · 18+ · Educational content, not financial advice

Two bettors receive identical selections with an identical genuine edge. One stakes 2% of bankroll on each; the other stakes 10%. After a long enough sample, the first is comfortably ahead and the second is broke. Nothing about the quality of the picks differed. Only the stake did.

This is the least glamorous part of betting and the part that decides outcomes most reliably.

Why the maths is asymmetric

Losses and gains are not symmetrical in percentage terms, and the gap widens fast:

DrawdownGain required to recover
10%11.1%
25%33.3%
50%100%
75%300%
90%900%

A bankroll down 50% needs to double just to return to level. This is why avoiding deep drawdowns matters more than capturing every last unit of growth — and why doubling your stake size does far more than double your risk.

The same picks, three stake sizes

Identical results, different stake sizes
Illustrative simulation — not SixAlgo results
200 100 0 Bankroll 0 250 bets 500
2% flat 5% flat 12% flat
All three curves come from the same sequence of wins and losses at the same positive edge. The 12% line does not fail because the picks were worse — it fails because compounding losses at that size makes recovery arithmetically impossible after a bad run. Schematic illustration of the compounding effect, not measured data.

The 12% curve is the shape almost every blown bankroll follows: strong early, a violent swing, and then a long flat crawl where each subsequent win is too small in absolute terms to repair the damage.

Flat staking

Stake the same percentage of your current bankroll on every selection, regardless of how confident you feel. One to two percent is the standard range.

The auditability point matters more than it sounds. Variable stakes are the easiest way to flatter a record retroactively — log winners at five units and losers at one, and a break-even set of picks produces a glowing ROI. We cover that trick in how to spot a fake tipster.

The Kelly criterion

Kelly calculates the stake that maximises long-run bankroll growth given your edge and the odds:

f = (bp − q) ÷ b

where b is decimal odds minus 1, p is your estimated true probability, and q is 1 − p.

At odds of 2.50 with an estimated 46% probability:

f = (1.50 × 0.46 − 0.54) ÷ 1.50 = 0.10

Kelly says stake 10% of bankroll. Which brings us to the problem.

Kelly assumes your probability estimate is exactly right. In sports betting it never is. And the formula is brutally sensitive to that error: overestimate your edge by a modest amount and Kelly overstakes by a large one. This is why full Kelly is almost never used in practice by people who bet for a living.

Fractional Kelly: the practical compromise

Stake a fixed fraction of the Kelly recommendation — commonly a quarter or a half. In the example above, quarter Kelly stakes 2.5% instead of 10%.

The trade is favourable. Halving the Kelly stake gives up a portion of theoretical growth but reduces volatility disproportionately, because variance scales with the square of stake size while expected growth scales roughly linearly. You keep most of the upside and shed most of the danger.

ApproachStake on the exampleBest suited to
Flat 2%2.0%Everyone, especially anyone following a service rather than modelling themselves
Quarter Kelly2.5%Bettors with their own model and a reasonable sense of estimate error
Half Kelly5.0%Well-calibrated bettors with a long verified sample
Full Kelly10.0%Essentially theoretical. Requires exact probabilities, which do not exist here

Rules that survive contact with reality

  1. Your bankroll is money you can lose entirely without consequence. If losing it changes your living situation, the stake is wrong regardless of the percentage.
  2. Recalculate from current bankroll, not starting bankroll. This is what makes losing runs self-limiting.
  3. Never chase. Increasing stakes after losses converts a survivable drawdown into a terminal one.
  4. Set the maximum in advance. Cap any single bet at 5% no matter what any formula says.
  5. Decide your stop point before you need it. Knowing you will pause at a 30% drawdown is worth more than any staking formula, because it is a decision made calmly rather than mid-panic.

Staking plans control the speed and volatility of the outcome, never its direction. No stake size turns negative expected value positive — it only changes how fast you find out.

Every signal ships with a stake unit

SixAlgo publishes a recommended stake unit alongside each signal, and our published results use flat staking at 2% of bankroll — the same plan on every selection, so the record stays auditable.

Join the free channel About SixAlgo

Frequently asked questions

How big should my starting bankroll be?

Large enough that a 1–2% stake is an amount you are comfortable losing, and small enough that losing all of it changes nothing material in your life. The absolute figure matters less than that condition.

Should I increase stakes on higher-confidence signals?

Only if you have a verified, well-calibrated sense of your own edge across a large sample. Most bettors overestimate confidence, and variable staking amplifies that error in both directions.

What drawdown should I expect?

Deeper than feels reasonable. At medium odds with a small edge, drawdowns of 20–30% occur in normal operation. Knowing the number in advance is what keeps you from quitting at the bottom.

Related reading