Staking Plans That Survive a Five- and Six-Figure Bankroll on UK Horse Racing

The £180,000 that taught me what staking plans actually do

A decade ago I sat down with a friend who had been told, by someone who should have known better, that his system was profitable enough to bet 4% of his bankroll on every selection. He had a bankroll of £180,000 from a property sale, a strike rate of 22% on his bets over a thoughtful 200-bet sample, and average odds of around 9/2. He believed — because he had been told — that this combination was an edge. We worked out the maths on the back of a beer mat in a pub in Camberwell and discovered, depressingly, that his system had close to zero expected value once Betfair commission was applied. The strike rate looked great. The prices he was getting on those strikes did not.

What that meant for his 4% staking plan was not that it was the wrong size. It was that the size didn’t matter, because the system underneath it had no edge to compound. The staking plan, in other words, was being asked to do something a staking plan cannot do. Six months later, after I had persuaded him to drop back to flat staking at 0.5% of bankroll while we rebuilt the underlying selection process, he had lost £14,000 — but he had paper-tested the difference between flat staking and 4% percentage staking on the same selections and discovered the percentage plan would have lost him £43,000 on the same bets.

That story sits at the front of this piece because the single most important truth about staking plans is the one I want you to hear before anything else. The plan does not generate the edge. The selections do. The plan shapes the variance of an edge that already exists or, in its absence, the variance of the bleeding. Everything below is built on that foundation.

What I’ll cover. Why staking plans cannot rescue a negative-EV system. Flat staking and where it earns its keep at scale. Percentage staking and where it doesn’t. The mechanics of full Kelly and why the textbook formula is rarely the right tool in practice. Fractional Kelly, drawdown, and the risk profile that actually survives a year. Confidence staking and the trap it sets. And stop-loss rules, monthly review processes, and the question of when to scale a bankroll up or hold it flat.

Why staking plans fail when there is no edge

I’m going to be uncharitable to a whole genre of betting content right at the top of this section. There is a category of writing on staking plans that promises mathematical certainty — usually built around a clever progression of stake sizes after losses or wins — and that obscures, often deliberately, the question of what underlying edge the plan is supposed to compound.

Independent stake-plan analysis is fairly settled on this point. The Betting Kingdom editorial summary of the staking-plan literature put it cleanly: several independent staking plan studies point toward a consistent pattern that no staking plan turns a negative-expectation system into a profitable one, and that what staking plans do is modify the risk profile of a system that already has value, with flat staking and percentage staking surviving long-term testing because they protect the bank. That formulation is worth committing to memory. The plan modifies the risk profile. It does not modify the expectation.

The maths underneath this is unforgiving and worth working through. Imagine a system with zero edge — a fair coin flip on each bet. No staking plan, applied to that system, has positive long-run expectation. Martingale (doubling after losses) gives huge short-run win rates and catastrophic eventual ruin. Anti-Martingale (doubling after wins) gives small short-run wins and steady long-run bleeding. Fibonacci progressions, Labouchere, D’Alembert — all of them simply rearrange the order in which the zero expectation arrives. None of them changes the fact that, in expectation, you finish where you started minus the commission.

So step one in any conversation about staking plans is to verify, with discipline and honesty, that your underlying system has positive expectation after all costs. That means tracking closing line value over a meaningful sample (200 bets minimum, ideally 500). That means measuring the gap between the price you struck and the closing price on the same selection, with positive average CLV being the cleanest single proxy for edge. If your average CLV is flat or negative across a 200-bet sample on UK horse racing, you do not have an edge, and no staking plan — no matter how cleverly calibrated — will save the bankroll. With YouGov reporting that only 8% of UK horse racing punters stake more than £100 a month, the population from which winning systems can be drawn is small enough that genuine edge is rare and worth verifying carefully.

Flat staking at scale

Flat staking — placing the same monetary stake on every bet, regardless of price or confidence — is the staking plan most professional punters with verified edges actually use. It is also the staking plan least talked about in the betting media, because it doesn’t generate clickable headlines.

What flat staking does, on a five- or six-figure bank, is provide the cleanest possible mapping between the underlying edge of the selections and the financial outcome of betting them. Each bet contributes equally to the eventual P&L. The variance of the system is exactly the variance of the underlying edge, multiplied by the stake. The drawdown depth is predictable from the strike rate and the typical odds. The monthly review process is simple: P&L divided by stake size gives you points won, which is directly comparable across periods.

The catch with flat staking on a large bank is the question of what the “flat” number should be. Too small, and you are dramatically under-utilising the bank. Too large, and a normal losing run can wipe out an unacceptable share of capital. The conventional wisdom — 1% to 2% of bankroll per bet — was developed for the era of smaller bankrolls and tighter operator limits. On a £100,000 bankroll, 1% is £1,000 a bet, which sounds right; on a £500,000 bankroll, 1% is £5,000 a bet, which is well above the effective limits at most UK operators for a non-Premier Fixture, meaning the plan starts colliding with stake-acceptance reality before the maths goes wrong.

My own approach with a six-figure bank, settled over many years of testing, is flat staking at 0.4% to 0.6% of bankroll per bet on Core Fixtures and 0.8% to 1.2% on Premier Fixtures. The asymmetry reflects the BHA’s own Q3 2025 data: average turnover per race at Premier Fixtures was up 2.7% year on year while Core Fixtures dropped 8.6%, which tells me that recreational money — and therefore stake-acceptance headroom — is concentrating at the flagship meetings. Sizing accordingly lets the plan extract more from the meetings where it can extract more, without trying to force a stake the trading desk will not accept.

Percentage staking at scale

Percentage staking — placing a fixed percentage of current bankroll on each bet — is, in theory, an elegant solution to the question of how to compound a bank through good runs and protect it through bad ones. The bet size grows as the bank grows; it shrinks as the bank shrinks. In a positive-expectation system, the geometric growth rate of the bank under percentage staking can substantially exceed flat staking over a long period.

In theory.

In practice, on a large bank with the operator limit picture I’ve described, percentage staking runs into three structural problems that the textbook treatments don’t address.

Problem one: the percentage you actually want to stake is much smaller than the percentage the textbooks recommend. The usual range cited is 1% to 5% of bankroll. On a system with realistic UK racing edge — say 4-5% net of commission and slippage — 5% of bankroll per bet produces astronomical drawdown depths in normal sample variance. The optimal percentage, calculated from a properly estimated edge, is typically below 1%, and frequently below 0.5%. The headline 1-5% range is, in most cases, dramatic over-staking.

Problem two: the percentage compounds operator limits. As the bank grows, the percentage stake grows. As the percentage stake grows, it more rapidly collides with the effective stake limit the operator will accept. So percentage staking on a winning system tends to produce a path where the early bets are sized at the calculated percentage, then bets start getting refused or stake-factored, then the percentage you actually stake diverges from the percentage you wanted to stake, and the elegant geometric growth becomes a messy hybrid of percentage staking on the way up and flat-capped staking from the point where operators start saying no. The plan breaks.

Problem three: the psychology of percentage staking on drawdown is genuinely harder than the psychology of flat staking. When the bank drops 20%, the percentage stake drops 20% — which feels, in the moment of a losing run, like surrender. Punters routinely abandon percentage staking exactly at the point in the drawdown when it is doing its job, switching back to flat (and therefore over-stating relative to the diminished bank) and accelerating the loss.

My take, after years of running both: percentage staking is the right plan for a clean-room implementation on a verified edge, executed with discipline, on a bank size that doesn’t collide with operator limits. For most six-figure UK racing punters, the operator-limit collision happens early enough that flat staking — perhaps with a periodic upward revision of the flat number when the bank has grown materially — is the cleaner practical implementation.

Full Kelly: what it actually says

The Kelly criterion is the most famous of the staking formulas and the most misunderstood. The formula itself is straightforward: optimal stake as a fraction of bankroll equals (edge × odds minus 1) divided by (odds minus 1), where edge is your estimated probability of the bet winning and odds are the decimal odds offered. On a bet at 4.0 (3/1) where you believe the true probability of winning is 30% (so the fair price is 3.33), the Kelly fraction is (0.30 × 4 − 1) / (4 − 1) = 0.20 / 3 = 6.67% of bankroll.

That is what the formula says. What it means is more nuanced.

What Kelly is actually solving for is the maximisation of the long-run geometric growth rate of the bankroll. It does this by trading off variance against expected value: betting too little leaves growth on the table; betting too much produces drawdowns deep enough to destroy the geometric average. The Kelly fraction is, in expectation, the stake size that produces the highest compounded growth over an infinite sequence of bets.

Two things follow from that, and both of them matter. First, “infinite sequence” matters. Kelly is asymptotically optimal — meaning it dominates other staking plans as the number of bets gets very large — but on any realistic finite sample (say 1,000 bets in a year), the variance of outcomes under full Kelly is enormous. A 30% drawdown is normal. A 50% drawdown is not unusual. Most punters cannot psychologically tolerate the drawdowns that full Kelly produces, which is why they abandon the plan partway through and end up worse off than if they had used a less aggressive scheme. Second, Kelly is extremely sensitive to the accuracy of the edge estimate. If you believe your edge is 5% but it is actually 2%, full Kelly over-stakes by a factor of more than 2 — which translates into materially deeper drawdowns and a long-run growth rate substantially below what Kelly would have delivered with the correct estimate.

So the headline answer on full Kelly is that the maths is correct and the implementation is rarely advisable. The variance is too large for human bankroll management, and the edge estimate is rarely accurate enough to trust the formula at its full setting. For a much more detailed step-by-step on the formula applied to a real race, the Kelly criterion worked example on a UK handicapwalks through the arithmetic on a concrete bet.

Fractional Kelly and drawdown

Fractional Kelly — staking a fraction of the calculated Kelly amount, typically half or a quarter — is the practical response to full Kelly’s variance problem. The mathematics is straightforward. If full Kelly produces an expected long-run growth rate of G with standard deviation S, then half-Kelly produces an expected long-run growth rate of 0.75G with standard deviation 0.5S. You give up a quarter of the expected growth to halve the variance. On any realistic sample, the trade is overwhelmingly favourable.

Quarter-Kelly drops the growth rate to roughly 0.4G but cuts the standard deviation to 0.25S. For most punters managing a bankroll they care about — meaning, in practice, almost every reader of this article — quarter-Kelly is the level at which the psychology of drawdown becomes tolerable enough to actually stay with the plan through a bad run. And staying with the plan, week after week, is what generates the compounded return. A theoretically optimal plan you abandon at the bottom of a drawdown is worse than a less optimal plan you stick to.

Drawdown specifically deserves more attention than it usually gets. The Q1 2025 data on the British racing market — turnover down 9%, average turnover per Core Fixture down 14.4% — is not just industry context; it is also a reminder that the operating environment for punters changes faster than the historical samples we calibrate plans against. A drawdown that would have been a normal 20% blip in 2022 can be the early stages of a structural problem with the underlying edge in 2026, because the market itself has reshaped. Fractional Kelly, by smoothing the variance, gives you more weeks to detect that distinction before the drawdown becomes existential.

The practical rule I use, on the bets where I am most confident in my edge estimate, is half-Kelly. On bets where I am less confident — typically novice hurdles, two-year-old races, or any market where my edge estimate is based on a smaller sample — I drop to quarter-Kelly or even eighth-Kelly. The variance reduction in those bets is worth the foregone growth, because the precision of my edge estimate doesn’t justify staking at the higher level.

Confidence staking and where it fails

Confidence staking — varying the stake based on how strongly you feel about each selection — is the staking plan most amateur punters adopt by default, often without realising they have adopted it. They put a bigger stake on their “best bet of the day” and a smaller one on their “speculative outsider”. This is intuitive and, in almost all cases, wrong.

Two things go wrong with confidence staking. The first is that “confidence” is, for most punters, a noisy signal that correlates poorly with actual edge. Punters feel most confident on the bets that fit their narrative most cleanly — the trainer they like, the jockey on a hot streak, the horse with the obvious story. Those bets are also the bets the market has priced most efficiently, because everyone else’s narrative aligns with the same factors. Genuine edge is far more often found in bets that don’t fit a clean narrative, and confidence is exactly the wrong heuristic for sizing those.

The second is the variance effect. Even if confidence does correlate with edge, the act of betting more on high-confidence selections and less on low-confidence selections increases the variance of the system as a whole, because more of the bank is concentrated on a smaller number of bets. Sharpe-ratio-style measures of risk-adjusted return typically fall when confidence staking is layered on top of a positive-EV system, not rise.

The exception, and there is one, is when confidence is operationalised through a quantitative edge estimate rather than a feeling. If you can convert “confidence” into a numerical probability of winning, you can plug it into Kelly and stake mathematically. That isn’t confidence staking; that’s just Kelly with variable edge inputs, which is the right way to do it. The version that fails is the version where confidence is a vibe.

Stop-loss rules and monthly review

Last piece. The stop-loss rule and the monthly review are the two pieces of operational discipline that separate a staking plan that survives a year from one that doesn’t, and they don’t get the attention they deserve in most treatments of this subject.

The stop-loss rule defines the drawdown depth at which you stop betting, review the system, and either restart or quit. It is not about protecting the remaining capital — flat or fractional Kelly staking already does that — but about creating a decision point at which you are forced to ask whether the system that generated the drawdown still has the edge you thought it had. My rule, on a six-figure bank running a fractional Kelly plan, is 25% drawdown from peak. At 25%, I stop. I don’t reduce stakes. I don’t switch systems. I stop, for at least a week, and I work through the data.

The Q1 2025 figure of -9% turnover I keep coming back to is important here too. When the market itself is contracting, the relationship between historical edge estimates and current edge estimates can break in ways that no in-sample testing would have predicted. A 25% drawdown in a contracting market is more likely to be a signal of structural change than the same drawdown in a stable market. The stop-loss forces you to make that diagnosis explicitly, rather than running the same plan into a 50% drawdown while telling yourself it’s just variance.

The monthly review is the lower-stakes version of the same discipline. Once a month, I sit down with my logged bets and look at three numbers: realised P&L, CLV across the month’s bets, and the strike rate against expected strike rate from the implied probabilities of the prices I struck. If the three move together in a way that’s consistent with the underlying edge thesis, I carry on. If P&L is down but CLV is positive, it’s variance and I carry on. If P&L is down and CLV is also down, the underlying edge is degrading and I need to investigate why before I bet another pound.

The thing about discipline is that it’s boring. The headline-grabbing parts of betting writing are the staking formulas and the Cheltenham anecdotes. The parts that actually compound a bankroll over a decade are the boring ones — the monthly review, the stop-loss, the willingness to size bets based on the operating reality rather than the textbook ideal. Get those right, and most of the rest of this article is implementation detail. Get them wrong, and no formula in the world will save you.

Published by the High-Stakes Horse Racing Betting team.