Picking horses is the half of betting people like talking about. Staking is the half that decides whether you are still around in six months. It has no stories in it. Nobody stands at the rail telling you about the time they staked 1.5 per cent of bank instead of 2. But almost every punter who has ever blown up did it through staking, not selection.
Two facts sit underneath everything in this article. First, the bookmaker's margin — the overround — means that most punters lose over time. That is not cynicism; it is arithmetic built into the prices. Second, even a genuinely good judge hits losing runs that are far longer than intuition suggests, and a staking approach that cannot survive those runs will fail no matter how well its owner reads a race.
So the honest job of a staking plan is not to make you money. No staking plan can do that. Its job is to control what a bad stretch costs you, so that a normal run of losers stays an annoyance rather than a disaster.
A bank is money you have decided you can afford to lose
Before stakes, the bank. A betting bank is a fixed sum, set aside in advance, that you have decided you can genuinely afford to lose — all of it — without it touching rent, bills, savings or anyone else's life. If losing the whole sum would matter beyond disappointment, the sum is too big. There is no clever workaround for this. The overround guarantees that the typical outcome of betting is a slow loss, so the only sensible way to fund it is as entertainment money, priced like any other hobby.
Keep the bank separate from everyday money — a separate account or wallet works — and never top it up mid-losing-run. If it empties, that is the season over, and the records (more on those later) will tell you why. Deciding the number is the single most important staking decision you will make, and it happens before the first bet.
Level stakes: the honest baseline
The simplest plan is level stakes: divide the bank into points and put one point on everything. A £500 bank at 100 points is £5 a bet; an each-way bet is two points, not one (see each-way value for why that second stake needs to justify itself).
Level staking has two virtues. It is impossible to fool yourself with — profit or loss at level stakes is the plainest measure of whether your opinions beat the prices — and it never asks you to bet more because things are going badly. Tipping results are quoted at level stakes for exactly this reason. If your selections lose money at level stakes over a few hundred bets, no staking arrangement will rescue them; anything a cleverer plan appears to add is rearrangement, not improvement.
How many points? A common range is 100 to 200 points for someone backing at typical win odds. The reason it needs to be that many is the next section.
Losing runs are longer than you think
Suppose you back horses around 3/1 (4.00) and you are decent at it, winning 25 per cent of the time. How likely is a run of ten straight losers?
The working is simple. Each bet loses with probability 0.75, so any particular stretch of ten bets is all losers with probability 0.75¹⁰ ≈ 5.6 per cent. Small. But over 200 bets — one or two a day through a season — there are 191 places such a run could start. The stretches overlap, so you cannot just multiply, but an exact streak calculation (stepping through the 200 bets one at a time and tracking the current run) gives the true figures. They are startling.
Probability of at least one losing run of the given length, somewhere in 200 bets:
| Run of at least… | 20% strike rate (≈4/1 backer) | 25% strike rate (≈3/1 backer) | 33% strike rate (≈2/1 backer) |
|---|---|---|---|
| 8 losers | >99.9% | 99.8% | 94.9% |
| 10 losers | 99.6% | 96.1% | 71.1% |
| 12 losers | 96.2% | 81.6% | 41.4% |
| 15 losers | 77.8% | 48.6% | 14.3% |
| 20 losers | 36.1% | 13.8% | 2.0% |
Read that middle column again. A punter with a perfectly respectable 25 per cent strike rate is more likely than not to hit twelve consecutive losers at some point in a 200-bet season, and has roughly a coin-flip's chance of hitting fifteen. Over a shorter 100-bet stretch the numbers soften but not by much: the same punter still faces about a 79 per cent chance of a ten-loser run. And these are the runs of an unlucky good judge — nothing has gone wrong with the selections at all.
This is why stakes must be small relative to the bank. At £10 level stakes from a £500 bank, a twelve-loser run costs £120 — about a quarter of the bank, unpleasant but survivable. At £50 stakes the same entirely ordinary run takes £600, and the bank is gone before the variance has finished being variance.
Percentage-of-bank staking
A refinement is to stake a fixed percentage of the current bank — say 2 per cent — recalculated as the bank moves. Stakes shrink automatically in a bad run and grow slowly in a good one, and in theory the bank can never quite reach zero.
The drawdown arithmetic is gentler than level stakes. Twelve straight losers at 2 per cent of a rolling bank multiplies it by 0.98¹² ≈ 0.785 — a fall of about 21.5 per cent, against 24 per cent at the equivalent level stake. The difference grows as runs lengthen, because each loss is a slice of a smaller number.
The honest cost is that percentage staking flatters your records slightly (wins after a losing run are struck at reduced stakes) and, taken to extremes, leaves you betting pennies to recover pounds. It is a perfectly sound approach at modest percentages — 1 to 2 per cent is common — but it is a comfort mechanism, not an edge. Nothing about it changes whether the bets themselves are any good.
The Kelly criterion, done properly
In 1956 John L. Kelly Jr, a physicist at Bell Labs, published "A New Interpretation of Information Rate" in the Bell System Technical Journal. Dressed up as a problem about noisy communication channels, it answered a gambling question: if you genuinely know the true probabilities better than the odds imply, what fraction of your bank should you stake to maximise long-run growth? The answer, for a simple win bet, is:
f* = (bp − q) / b
where b is the fractional odds as a number (7/2 = 3.5), p is your estimated probability of winning, and q = 1 − p.
A worked example. A horse is 7/2 (4.50), and after your own work you make its true chance 25 per cent. Then:
- b = 3.5, p = 0.25, q = 0.75
- f* = (3.5 × 0.25 − 0.75) / 3.5 = (0.875 − 0.75) / 3.5 = 0.125 / 3.5 ≈ 0.036
Kelly says stake about 3.6 per cent of your bank — £18 from £500. Notice what the formula does when you have no edge: 7/2 implies a probability of 1 ÷ 4.5 ≈ 22.2 per cent, and if your estimate is 22.2 per cent or lower, f* is zero or negative and Kelly says do not bet at all. Kelly never asks you to bet without an advantage, which already makes it more honest than most staking talk.
Why full Kelly is too aggressive for real people
Here is the caveat, and it is not a footnote. The formula's output is only as good as p, and you do not know p. You estimated it. In the example above, staking 3.6 per cent rests on the claim that the market has this horse nearly three percentage points wrong — a strong claim about a market full of informed money. If the true chance is what the price says, your "3.6 per cent edge bet" is just a bet, made at four times a sensible stake.
The mathematics of overbetting is brutally asymmetric. Edward Thorp, who used Kelly in blackjack and later in fund management, showed in "The Kelly Criterion in Blackjack, Sports Betting, and the Stock Market" that betting twice the true Kelly fraction reduces long-run growth to zero, and beyond that growth turns negative — you can hold winning opinions and still grind a bank to nothing purely through stake size. Understaking merely costs some growth; overstaking can destroy the bank. And since punters systematically overestimate their edges, the practical error is almost always on the ruinous side.
This is why serious Kelly users use fractional Kelly: half, or more often a quarter, of the formula's output. Thorp himself favoured half Kelly, which keeps roughly three-quarters of the theoretical growth rate at a fraction of the volatility. MacLean, Thorp and Ziemba's 2011 collection The Kelly Capital Growth Investment Criterion makes the deeper point: Kelly stakes are acutely sensitive to errors in the estimated probabilities, and bettors whose estimates run optimistic are led into overbetting and possible ruin. Quarter Kelly on our example is about 0.9 per cent of bank — £4.50 from £500 — which looks timid until you set it against the losing-run table above. If your carefully estimated edges are partly imaginary, and some of them are, fractional Kelly is what stands between you and the drawdown that proves it.
Never increase stakes after a loss
Every generation reinvents the recovery plan: double after a loss (the martingale), or add a unit, or stake "to win back" the deficit. They all share one mechanic — stakes rise exactly when the bank is falling — and they all fail the same way.
The arithmetic: doubling from a £5 stake, the eleventh bet in a losing run is £5,120, with £5,115 already gone, all to chase a few pounds of profit. Now look back at the table. Runs of that length are not freak events; at ordinary strike rates they are close to certainties over a season. A recovery plan does not remove the losing run — it arranges for the run to arrive at your maximum possible stake. Chasing losses is also one of the clearest warning signs of gambling harm that support services such as GamCare describe. If a plan ever asks you to stake more because you lost, the plan is the problem.
No staking plan can turn losing bets into winning ones. The only thing staking controls is what a losing run costs you — and every backer, however good, gets losing runs.
Keep records, and keep them honestly
Discipline without records is a feeling, and feelings about our own betting are unreliable in one consistent direction. Keep a simple log: date, race, selection, the price taken and whether it was a board price, SP or BSP (the differences matter), stake, result, and the actual return — after any Rule 4 deductions — plus a running bank total. A spreadsheet is plenty.
Record every bet, especially the impulsive ones, and review monthly rather than daily; day-to-day results are nearly all noise. What the record gives you is the one thing memory will not: a level-stakes truth about whether your betting beats the prices over hundreds of bets. If it does not — and for most people, honestly kept records eventually say exactly that — the answer is to bet less or stop, not to find a cleverer staking plan. That is not a defeat. It is the record doing its job, and it is knowledge most punters never let themselves have.
A note on limits
The tools for keeping betting in its box are ordinary and free, and using them is no more dramatic than setting a budget for anything else. Every UK-licensed bookmaker offers deposit limits and time-outs in account settings; GAMSTOP self-excludes you from all British-licensed online operators in one go; and if betting ever stops feeling like entertainment, BeGambleAware.org and GamCare's National Gambling Helpline (0808 8020 133, free, 24 hours) are there for exactly that conversation. A fixed bank, small stakes, honest records, and limits set in advance — that is the whole of the unglamorous half. The rest of the Academy is about the interesting half, and it only works if this half is already in place.