Two bettors receive identical picks with a genuine 3% edge. After a year, one has grown their bankroll steadily and the other has gone broke. The difference is not selection. It is stake sizing, and it is the part of betting that gets the least attention relative to how much it determines the outcome.
Flat staking means risking the same amount on every bet — typically 1% to 2% of bankroll, expressed as one unit. It is the approach almost every bettor should use, for reasons that have little to do with mathematical optimality.
It is immune to overconfidence. It cannot be quietly abandoned in the middle of a losing run. It produces a clean record, because every result is weighted equally and your return on turnover is directly interpretable. And it does not require you to estimate your own edge, which is fortunate, because most bettors cannot.
The cost is that it leaves some growth on the table. A 5% edge and a 0.5% edge get the same stake, which is not optimal. For most people, that inefficiency is a price worth paying.
Kelly answers a precise question: what fraction of bankroll maximises the long-run growth rate, given a known edge and known odds? The formula for a simple win-or-lose bet is:
f = (b × p − q) / b
where f is the fraction of bankroll to stake, b is the decimal odds minus one (your profit per unit staked), p is your estimated probability of winning, and q is 1 − p.
Worked example. You bet at -110, so b = 0.909. You believe your true win probability is 55%, so p = 0.55 and q = 0.45.
f = (0.909 × 0.55 − 0.45) / 0.909 = (0.500 − 0.45) / 0.909 = 0.05 / 0.909 = 5.5% of bankroll
On a $10,000 bankroll, that is a $550 bet. Compare that to a 1% flat stake of $100. Kelly is recommending five and a half times more.
The formula is correct. The problem is the input.
Kelly assumes you know p. You do not. You have an estimate, and the previous article in this series explains why that estimate cannot be verified without on the order of a thousand bets. If your true edge is smaller than you believe — the overwhelmingly common case — Kelly’s output is not slightly too large. It is dangerously too large, because the formula is highly sensitive to overestimated probability.
Consider what happens to that $550 stake during a normal eight-loss run, which a 55% model should expect roughly once per 500 bets. Eight consecutive losses at 5.5% of a declining bankroll costs around 37% of the total. That is survivable but brutal. Now suppose your true edge was 52% rather than 55% — Kelly was sizing you for an edge you never had, and the drawdown compounds against you indefinitely.
Full Kelly also produces stake volatility that most people cannot tolerate psychologically. Betting $550 today and $80 tomorrow, then $600 the day after, is a recipe for improvised deviation — and an improvised staking plan is no staking plan.
The standard practical compromise is to stake a fraction of the Kelly recommendation — commonly a quarter or a half.
| Approach | Stake on the example above | Character |
|---|---|---|
| Full Kelly | $550 (5.5%) | Maximum theoretical growth, severe drawdowns, unforgiving of estimation error |
| Half Kelly | $275 (2.75%) | About three-quarters of the growth rate with substantially reduced volatility |
| Quarter Kelly | $137 (1.37%) | Conservative, tolerant of overestimated edge, close to a disciplined flat stake |
| 1% flat | $100 | Simplest, requires no probability estimate at all |
Quarter Kelly is a reasonable starting point for anyone who wants edge-proportional sizing without betting their bankroll on the accuracy of their own probability estimates. It is notable that quarter Kelly on a strong edge lands close to a standard flat stake — which is a decent indication that conventional flat staking is not far from sensible.
Kelly is mathematically elegant and practically hazardous, because its accuracy depends entirely on an input you cannot measure. Flat staking is mathematically suboptimal and practically robust. For the large majority of bettors, 1% flat with an occasional 2% on genuinely stronger positions will outperform an ambitious Kelly implementation — not because the maths is wrong, but because the discipline holds.
And it is worth stating plainly: no staking plan converts a negative expectation into a positive one. Sizing determines how a real edge compounds and how a losing run is survived. It cannot manufacture an edge that is not there.
One to two percent is the conventional range. Lower is appropriate if your edge is uncertain, if you bet frequently, or if a drawdown would affect your finances outside betting.
Confidence is not the same as verified accuracy, and full Kelly punishes overestimation severely. Quarter or half Kelly captures most of the growth with far less exposure to being wrong about your own probabilities.
No. Progressive systems such as Martingale rearrange when losses occur without changing expectation, and they increase the risk of catastrophic loss. No sizing method creates positive expectation.
Weekly or monthly is sufficient and considerably more stable than recalculating after every settled bet.
Flat staking works best with a steady stream of selections rather than an unpredictable flood. Members get one considered pick a day — and a clear pass when the price is not there.
18+ · Please gamble responsibly.
69 Advisory provides informational sports analysis only. Nothing above is a guarantee of results and past performance does not indicate future outcomes. Only stake what you can afford to lose.
Free confidential support: National Gambling Helpline (UK) 0808 8020 133 · National Problem Gambling Helpline (US) 1-800-GAMBLER · begambleaware.org
Cardinals @ Phillies: How We Read the F5 Under (5.0)
MLB · Match Analysis How we read it: Cardinals @ Phillies F5 under 5.0 St. Louis Cardinals @ Philadelphia Phillies · Dobbins vs Luzardo Analysed by OPTIMUS II The read MLB · First 5 innings First 5 Innings – Under 5.0 Runs @ 1.649 1.00Unit 1.649Odds 57.8%Win prob. +10.7%Model EV Note: this analysis is published […]
Read articleSample Size and Variance: How Many Bets Before You Know a Model Works
A month of results tells you almost nothing. Here is the actual arithmetic on how long it takes to distinguish skill from luck — and why the answer is uncomfortable.
Read articleRun Line Betting Explained: When +1.5 Is Value and When It Is a Trap
The baseball spread is fixed at 1.5 runs, which makes it a very different instrument from a football or basketball line. Here is how to price it properly.
Read article
Back to blog