Staking plans are rules for how much to bet each time. The four main kinds are level stakes, a percentage of your current bank, staking to win a set amount, and the Kelly criterion. Run on the same 100 hypothetical bets from a $1,000 bank, with Kelly staked in full and at half, they finished between +$43 and +$139 on a record close to its expected winners.
None of them changes the price you take or the chance of winning, so none can make bets with negative expected value (EV) profitable on average. What a plan decides is how big the swings are, how fast losses shrink the bank, and whether you can keep to it.
Four staking plans and the rule behind each
With an illustrative $1,000 bank, a $20 base stake and the three prices used in the comparison below:
| Plan | The rule | Stake at $1.85 | At $3.25 | At $8.00 |
|---|---|---|---|---|
| Level stakes | The same amount on every bet | $20 | $20 | $20 |
| Percentage of the bank | A fixed share of the current bank | 2% ($20 at the start) | 2% | 2% |
| Staking to win a set amount | Stake = target profit / (price - 1) | $23.53 | $8.89 | $2.86 |
| Full Kelly | Stake = bank x edge / (price - 1) | 3.53% | 2.22% | 0.86% |
| Half Kelly | Half the Kelly share | 1.76% | 1.11% | 0.43% |
The Kelly rows use edges of 3% at $1.85, 5% at $3.25 and 6% at $8.00, so the full Kelly share at $1.85 is 0.03 / 0.85 = 3.53%.
Level stakes
Level stakes keep records simple: profit / turnover is your yield, and results counted in betting units compare across bank sizes. The stake does not follow the bank, so after a long fall each bet is a bigger share of what is left.
Percentage of the bank
Also called proportional staking, this re-sizes every stake from the current bank. Stakes shrink as the bank falls, so a run of losses cannot empty it outright, and they grow as it rises. The cost is compounding: a 2% loss followed by a 2% gain leaves 0.98 x 1.02 = 0.9996 of the bank, a little below where it started.
Staking to win a set amount
Each stake is sized so that a winner makes the same profit: $20 to win takes $23.53 at $1.85 but $2.86 at $8.00. That puts the most money on the shortest prices, which steadies results when prices are mixed. It gets heavy at very short prices: to win $20 at $1.10 takes a $200 stake.
Sizing stakes to your edge
Kelly sizes each stake from your edge and the price, so a bigger edge or a shorter price means a bigger share. With the same edge on every bet it amounts to staking to win a set share of the current bank, because a Kelly stake x (price - 1) = bank x edge. Every edge is an estimate, so staking a fraction of Kelly is common. The Kelly criterion works through the formula and fractional Kelly the fractions.
The same 100 bets under each plan
Example: 100 hypothetical bets from a $1,000 bank, with the edges set by assumption. The true chance at each price is (1 + edge) / price.
| Price | Bets | Edge | True chance | Expected winners | Winners in the record |
|---|---|---|---|---|---|
| $1.85 | 40 | 3% | 55.7% | 22.3 | 22 |
| $3.25 | 40 | 5% | 32.3% | 12.9 | 13 |
| $8.00 | 20 | 6% | 13.25% | 2.65 | 3 |
Each plan's result on that record, and on an unlucky and a lucky version of the same 100 bets:
| Plan | Close to expected (22, 13 and 3 winners) | Unlucky (19, 10 and 1) | Lucky (25, 16 and 5) |
|---|---|---|---|
| Level stakes, $20 | +$139 | -$487 | +$765 |
| 2% of the current bank | +$87 | -$407 | +$994 |
| Staking to win $20 | +$48 | -$215 | +$311 |
| Full Kelly | +$61 | -$384 | +$829 |
| Half Kelly | +$43 | -$207 | +$371 |
The working for the first column:
- Level stakes: 22 x $17 - 18 x $20 + 13 x $45 - 27 x $20 + 3 x $140 - 17 x $20 = $14 + $45 + $80 = $139.
- Staking to win $20: 22 x $20 - 18 x ($20 / 0.85) + 13 x $20 - 27 x ($20 / 2.25) + 3 x $20 - 17 x ($20 / 7) = $16.47 + $20.00 + $11.43 = $47.90.
- 2% of the bank: each winner adds 2% x (price - 1) and each of the 62 losers takes 2%, so the bank ends at $1,000 x 1.017 to the power 22 x 1.045 to the power 13 x 1.14 to the power 3 x 0.98 to the power 62 = $1,087.19.
- Full Kelly: each winner adds exactly its edge to the bank and each loser takes the Kelly share, so the bank ends at $1,000 x 1.03 to the power 22 x 1.05 to the power 13 x 1.06 to the power 3 x (1 - 0.03 / 0.85) to the power 18 x (1 - 0.05 / 2.25) to the power 27 x (1 - 0.06 / 7) to the power 17 = $1,061.29.
Three things stand out. Level stakes made the most on the middle record, and $80 of its $139 came from the 20 bets at $8.00: 3 x $140 - 17 x $20 = $80. Each $8.00 winner is worth $140 to level stakes and $20 to the to-win plan, so with two of them instead of three, level stakes would have finished at -$21.
Percentage staking trailed level stakes on the middle record, but it lost less on the unlucky one and made more on the lucky one. Its stakes grow and shrink with the bank, so it bends with the run, and on a record that ends near its start, compounding costs a little.
Staking to win $20 and half Kelly swung least, from -$215 to +$311 and from -$207 to +$371. They put the least on the $8.00 bets, where one result moves the most money.
Each plan sets every stake from a fixed amount or a fixed share of the bank, so the order of the results changes the path, not these final figures.
Risk: Betting involves risk. These bets have an edge built in by assumption; a real edge is an estimate, and a plan that made money on this record can lose on real bets. See responsible gambling for limits and support.
Why staking cannot turn negative EV positive
Expected profit = the sum over all bets of stake x expected value per $1. A staking plan only chooses the stakes, so when every bet has negative expected value, every term in that sum is negative and so is the total, whatever the plan.
Here are the same 100 bets, staked the same ways, with the expected (average) result over every possible outcome. The second column assumes every price is 5% below fair, so each bet's expected value is -5%: true chances of 51.4% at $1.85, 29.2% at $3.25 and 11.9% at $8.00.
| Plan | Expected result with the edges assumed above | Expected result if every bet is -5% |
|---|---|---|
| Level stakes, $20 | +$88 | -$100 |
| 2% of the current bank | +$92 | -$95 |
| Staking to win $20 | +$49 | -$68 |
| Full Kelly on the assumed edges | +$102 | -$116 |
| Half Kelly on the assumed edges | +$50 | -$60 |
| Kelly sized on the true edges | +$102, the same as full Kelly | $0: it places no bets |
The working for level stakes: 100 x $20 x -0.05 = -$100. For 2% of the bank: $1,000 x (1 - 0.02 x 0.05) to the power 100 - $1,000 = -$95.21.
Every plan's expected result takes the sign of the bets' expected value: a plan only decides how much of it you take on. Full Kelly on edges that are not there loses the most, because it stakes the most. Kelly stakes nothing on a bet it knows has no edge, but it only knows what your estimate tells it.
Plans that raise stakes after losses, such as the martingale system, are chasing losses written as a rule, and betting systems that don't work tests the others. Price and selection are where a betting strategy finds its edge; staking decides how much of the swing you carry.
Risk: Betting involves risk. No staking plan changes the price or the chance of winning, and there is no guarantee of profit. See responsible gambling for limits and support.
Choosing a staking plan you will actually follow
The best staking plan is the one that suits your prices and your edge, and that you will keep to through a losing run. A plan dropped halfway through a bad stretch protects nothing.
| Your situation | A plan that fits | Why |
|---|---|---|
| No edge estimate for each bet, or a method still being tested | Small level stakes | Results read straight as a yield, and the record stays comparable |
| Prices mostly between $1.50 and $2.50 | Level stakes or a percentage of the bank | Near $2.00, level stakes and staking to win are almost the same size |
| Short prices mixed with long shots | Staking to win a set amount, or fractional Kelly | Both put less on long prices, where one result swings the bank most |
| An edge estimate for every bet, checked against the closing price | Fractional Kelly | Stakes follow the edge, cut back for estimate error |
| A losing run tempts you to raise stakes | A percentage of the bank, with a written maximum stake | Stakes fall with the bank, and the cap is set before the run |
Whatever you choose, write the staking strategy down before the first bet: the bank, the rule, the maximum stake, when you will review it and what ends it, such as the bank running out. Set the bank first, as bankroll management explains, and review after a set number of bets, such as 500, rather than after a bad week.
Edge cases a staking plan has to cover
| Case | How to handle it |
|---|---|
| A bookmaker accepts only part of your stake | Bet what is accepted and record that amount; a limited account is a fact to plan around, never something to route through another person's account |
| Each way bets | Count the whole outlay against the plan: $20 each way is $40, two bets of $20 |
| Bonus bets | Keep them outside the plan: a bonus bet that loses pays nothing, and the bonus bet converter values one by what it converts to |
| Several bets open at once | Percentage and Kelly stakes all come from the same bank, so five open bets at 2% each put 10% of it at risk together |
| Stakes that are not round numbers | Round down: 2% of $1,087.19 is $21.74, and rounding it up to $25 stakes 2.3% of the bank |
Where staking plans go wrong
| Mistake | What it costs |
|---|---|
| Staking to win at very short prices | To win $20 at $1.10 takes a $200 stake, so one loser wipes out ten winners |
| Switching to a bigger share after a fall | Moving from 2% to 5% after the bank drops from $1,000 to $800 lifts the next stake from $16 to $40, double the original $20 |
| Full Kelly on edges that are not there | On the 100 bets at -5%, it loses $116 on average against $60 at half Kelly |
| Doubling after losses | From $20, six losses in a row make the seventh stake $1,280, more than the whole $1,000 bank |
| Judging a plan on one record | On the middle record level stakes made $52 more than 2% of the bank; on the unlucky one they lost $80 more |