Staking Plans Explained — Flat, Percentage and Kelly, and Which Survives a Bad Estimate

18+ only. Gambling can be addictive. Please play responsibly.

Deciding *what* to bet and deciding *how much* are two separate problems, and the second one is where more bankrolls are destroyed than the first. A bettor with a genuine edge and a reckless staking plan goes broke. A bettor with no edge and a perfect staking plan also goes broke, just more slowly.

This page is only about the second problem. If you have not set the bankroll itself yet — the money, the limits, the separation from everything else — start with bankroll management, because everything below assumes a fixed pot you have already decided you can lose.

One sentence before the maths, and it governs every word that follows: no staking plan creates an edge. Staking decides how fast money moves. Whether it moves toward you or away from you is decided entirely by whether your prices are better than the true probabilities, and by default they are not.

The three plans

Level stakes — the same amount every time

Every bet is the same fixed sum. £10 a bet, every bet, regardless of how confident you feel or what the odds are.

What it gets right. It is impossible to get wrong, it makes your results readable, and it removes the single most destructive habit in betting: raising the stake because this one feels certain. It is also the only plan under which your betting record actually measures your judgement, because every selection carries identical weight.

What it costs. It ignores information. A bet you rate a small edge on and a bet you rate a large edge on get the same money, which is not optimal in any formal sense.

Who it is for. Anyone who has not yet proved, from a written record over a meaningful number of bets, that their probability estimates beat the closing line. That is nearly everyone, nearly always. Level stakes is not the beginner option — it is the correct option until you have evidence to move off it.

Percentage of bankroll — the same *proportion* every time

Every bet is a fixed percentage of the current bankroll. At 2% of a £1,000 pot you stake £20; if the pot falls to £800 you stake £16; if it grows to £1,200 you stake £24.

What it gets right. It cannot mathematically ruin you. Losing bets shrink the stake automatically, so the pot approaches zero asymptotically rather than hitting it. It also compounds upward without any decision on your part.

What it costs. Recovery is slow by construction — after a 50% drawdown you are staking half as much, so you need considerably more than a 50% gain to get back. And it needs an honest rule about *when* you recalculate. Recalculating after every bet is fine; recalculating only after wins is a percentage plan in name and an escalating plan in practice.

Who it is for. Bettors with a record long enough to believe in a small edge, who want compounding without the volatility of the next option. Common sizing is 1–3%.

Kelly — stake in proportion to your edge

The formal answer, and the most misused idea in betting.

The mathematics comes from J. L. Kelly Jr., "A New Interpretation of Information Rate", published in the *Bell System Technical Journal* in July 1956 (volume 35, pages 917–926; the original is scanned at archive.org). Kelly was working at Bell Labs on information theory, not on gambling — the betting framing was an illustration of a channel-capacity result. His paper shows that a gambler using knowledge of the outcomes, at fair odds, can make capital grow exponentially, and that the maximum exponential rate of growth of the gambler's capital equals the rate of transmission of information over the channel — a result he then generalises to arbitrary odds.

For a single binary bet the formula reduces to:

f = (bp − q) / b

where f is the fraction of bankroll to stake, b is the profit per unit staked (decimal odds minus 1), p is your probability the bet wins, and q is 1 − p.

Worked example. Odds of 3.00, so b = 2.00. You rate the outcome at 40%, so p = 0.40 and q = 0.60. Then f = (2.00 × 0.40 − 0.60) / 2.00 = (0.80 − 0.60) / 2.00 = 0.10. Kelly says stake 10% of your bankroll.

On a £1,000 bankroll that is a £100 bet on a selection you personally rate at less than a coin flip. That number is correct, and it is also the reason full Kelly is the wrong answer for almost everybody.

Why full Kelly is the wrong answer

Kelly's formula maximises the long-run growth rate of your capital given that p is correct. Everything rests on that condition, and in sports betting p is not a known quantity. It is your opinion.

The failure is asymmetric, which is the part that catches people. If you overestimate your edge, Kelly does not stake slightly too much — it stakes dramatically too much, because f scales directly with the error. Rate a 33% shot at 40% and you have not made a 7-point mistake in a probability; you have turned a bet with no edge into a 10%-of-bankroll position. Do that across a season and the drawdowns are severe enough that most people abandon the method mid-slump, which is the worst possible moment to change plans.

There is also a structural reason to expect your p to be too high. Karl Whelan and Tadgh Hegarty's study of bookmaker margins (2023) finds that bookmakers set higher profit margins on lower-probability outcomes, so the average loss rate across bets actually placed is worse than the headline overround suggests. The longer the price, the more hidden margin you are paying — and long prices are exactly where a Kelly calculation produces the largest stakes.

The standard response is fractional Kelly: compute f, then stake a fixed fraction of it. Half-Kelly turns the 10% above into 5%; quarter-Kelly into 2.5%. This deliberately gives up some theoretical growth rate in exchange for a large reduction in variance and, more importantly, a large reduction in the damage done by an overestimated p. Quarter-Kelly is a common working choice, and it lands close to the 1–3% a percentage plan would have given you anyway — which tells you something about why the simple plan is not as naive as it looks.

Comparing them honestly

The row that decides it for most readers is the fourth one. Under level stakes, your profit and loss is a measurement of your judgement. Under any variable plan it is a measurement of your judgement *multiplied by* your sizing decisions, and separating the two afterwards is genuinely hard. If you do not yet know whether you can pick winners, a plan that obscures the answer is the wrong plan regardless of its mathematics.

The rules that matter more than the plan

Whichever you pick, these do more work than the choice between them:

• Fix the unit before the bet, never during. The stake is decided by the plan, not by how the selection feels. A stake raised because of confidence is not a staking plan.
• Never chase. Increasing stakes to recover losses is the Martingale family, and it converts a series of small losses into one catastrophic one. Table limits and account limits exist and the sequence hits them.
• Recalculate on a schedule, not on a mood. If you are on a percentage plan, decide in advance whether the bankroll is recomputed per bet, per week or per month, and stick to it.
• Log the closing price, not just the result. Whether you beat the closing line is the only early evidence that your p is worth anything at all — long before profit or loss says anything reliable. That is the argument in closing line value, the only scoreboard.
• Withdraw wins out of the bankroll periodically. Money that stays in the account is money still exposed to the edge, and the edge is against you.

Where to go next

Staking sits downstream of price. If the price you took has no value in it, no plan on this page helps — how to find value bets and expected value in betting, explained cover the upstream half, and our step-by-step value calculation tutorial walks a single bet through the arithmetic. The bankroll itself — the pot every percentage above is a percentage *of* — is set up in bankroll management. And why the favourite wins and you still lose is the plainest statement of what all of this is fighting against.

The honest summary: betting carries a negative expected return by design, and the margin is built into every price before you choose a stake size. A staking plan is damage control and variance management. It is not a source of profit, and anyone selling one as though it were is selling you something else.

18+ only. Gambling can be addictive. Please play responsibly. Nothing here is a prediction and no market is an income source. If gambling is causing you problems, contact BeGambleAware.org, or see our responsible gambling resources.