Racing guide · 21 September 2026 · 5 min read
Staking plans, honestly compared
What a staking plan actually decides
A staking plan is just a rule for how much you put on each selection. It doesn't touch the selection process itself. Whether you're picking horses off ratings, form study, or a system that scores runners on speed and class, the staking plan sits entirely downstream of that decision and has no say in whether the horse can run a yard.
This matters because staking plans get sold, implicitly or otherwise, as a way to turn a shaky method into a solid one. They can't. Changing how much you stake on a set of selections changes the shape of your results, sometimes the swings get bigger, sometimes smaller, but it doesn't change whether those selections were sound in the first place. A method that loses money at level stakes will still lose money, just with different peaks and troughs, under almost any other staking approach.
This piece looks at the maths behind the main approaches, level stakes, percentage-of-bank, and the various progression systems built around chasing losses. None of it is advice on what to stake. It's an explanation of why some of these plans behave the way they do, and why one of them in particular carries a risk that isn't obvious until you've written the numbers down.
Level stakes: the honest benchmark
Level stakes means putting the same amount on every selection, regardless of odds, confidence, or how the last few results went. It's the plan used almost universally when judging a record, because it's the only one that doesn't let the staking distort the story. If a set of selections shows a level-stakes loss over a few hundred runners, no amount of clever staking on top changes the fact that the selections themselves lost.
The appeal of level stakes is that it isolates the selection method from everything else. You can see the win rate, the place rate, the average odds, and the running total, and know that what you're looking at reflects the picks, not a staking pattern layered over them. That's why any serious backtest, including the ones published against The Racing Bot's scores, uses level stakes as the baseline before anything else gets discussed.
It isn't the most exciting way to stake in practice, since it doesn't scale with a bank the way percentage staking does. But for comparing one method against another, or one period against another, it's the closest thing racing has to a fair test.
Percentage-of-bank staking
Percentage staking means betting a fixed slice of your current bank each time, say 2%, rather than a fixed sum. As the bank grows the stakes grow with it, and as the bank shrinks the stakes shrink too. This has a genuine mathematical effect: it's structurally impossible to lose your entire bank on a percentage plan, because you never stake all of what's left.
That protection comes at a cost. Percentage staking tends to smooth out both the peaks and the troughs of a level-stakes record, so a long losing run does less damage in cash terms, but a long winning run also builds less dramatically than it might feel like it should. It also makes a record harder to compare against anyone else's, because the actual amounts staked depend entirely on the starting bank and how it moved along the way.
None of that changes the underlying point. If the selections beneath a percentage plan don't hold up at level stakes, percentage staking will still produce a declining bank, just more gradually. It manages the rate of loss. It doesn't reverse it.
Progression systems: Martingale and its relatives
Martingale and its variants work on a simple idea, double the stake after every loss so that a single win recovers everything lost so far plus a small amount extra. On paper, provided you never run out of money and never hit a maximum bet limit, it looks unbeatable, because a win has to arrive eventually.
In practice both those conditions fail. A run of eight or nine losers, which happens more often than most people expect even with a genuinely well-supported horse at short odds, requires a stake that has grown many times over from where you started. Bookmakers impose maximum bet limits that will stop the progression dead well before that point, and even without a limit, a finite bank runs out before an infinite run of losers does. The system assumes both unlimited stakes and unlimited losing streaks aren't a problem. In racing, both are.
Other progression systems, like increasing stakes after a win rather than a loss, avoid the worst of that trap but introduce a different issue: they only compound gains if the underlying selections keep winning, which brings the conversation straight back to whether the method behind the selections is sound. Every version of a progression plan is really a bet on the shape of future results, and no staking rule can predict that shape in advance.
Why no plan fixes a losing method
Every staking plan, level, percentage, or progression, is a way of arranging bets around a set of selections that already exist. The maths behind each one is well understood and none of it can look at a horse and know something the selection method didn't already know. If the selection process is generating losers more often than winners can cover, staking arranges how the losses land, not whether they happen.
This is worth being blunt about because staking systems get marketed as if they're a shortcut round bad form study, and they aren't. The only way to see whether a method holds up is to test it honestly, at level stakes, over a large enough sample that a short run of luck in either direction doesn't distort the picture. A dozen selections tell you almost nothing. A few hundred, across different tracks and conditions, start to tell you something real.
That's the reason a level-stakes simulator is useful for anyone looking at a published record, including The Racing Bot's own. It lets you replay a set of scored selections at a fixed stake and see the running total exactly as it happened, race by race, rather than taking a headline figure on trust. What it won't do, and what no staking plan will ever do, is turn a weak set of selections into a strong one.