Does Fibonacci staking work?

The sequence came first, the staking plan much later. Leonardo of Pisa, the medieval mathematician better known as Fibonacci, set it out in his Liber Abaci in 1202, in a puzzle about breeding rabbits: each number is the sum of the two before it, so it runs 1, 1, 2, 3, 5, 8, 13, 21 and on forever. The sequence turned out to be one of the loveliest objects in mathematics. It hides in sunflower heads and snail shells, and the ratio between its neighbouring terms settles towards the golden ratio, the number artists and architects have chased for centuries.

Gamblers, being gamblers, saw a staking plan in it. The betting version works like the Martingale's more reasonable sibling: stake one unit, and after every losing bet step one place further along the sequence, so the stakes climb 1, 1, 2, 3, 5, 8. After a win, go back to the start. The climb is gentler than doubling, the recovery arithmetic still works on paper, and the whole thing carries that faint glow of natural mathematics, as if the same pattern that arranges a pine cone might arrange your betting bank.

Pascal brought it to the Lab with some confidence, and here is the honest twist that makes this the most interesting page on the board: when we replayed the plan over 27,355 real GB races, the headline number came back positive. And that positive number is precisely the mirage this page exists to take apart, because the profit sits in a handful of bets so large that no real bankroll could ever have placed them. The sequence that grows rabbits, it turns out, grows stakes exactly as fast.

The Fibonacci plan sells itself on one comparison: it is not the Martingale. Doubling after every loss is obviously reckless, everyone has heard how that ends, and the sequence looks like the grown-up alternative. Each step up multiplies the stake by roughly 1.6 rather than 2, so after five straight losses a Martingale wants 32 units where Fibonacci asks for a modest 8. Written out side by side, one column looks like a panic and the other looks like a plan.

Then there is the rhythm of it in practice, and this is the part that genuinely does feel good. The plan here backs the favourite, and the favourite obliges often, winning 34.8% of races in our record. So most sequences are short. You lose one, lose another, the favourite comes in, you reset to a single unit and bank a small recovery. Weeks can pass like that: little runs, little resets, the cumulative line ticking gently upward. The plan is doing exactly what its brochure promised, and every reset reads as proof that the mathematics is working.

And the mathematics itself is a quiet part of the sales pitch. A staking plan named after a casino feels like a hustle; one named after the pattern in a sunflower feels like wisdom. Punters who would never touch a doubling system will happily walk a sequence that Leonardo of Pisa wrote down eight centuries ago, because it carries a borrowed authority that plain arithmetic never gets.

The appeal, in short, is that Fibonacci softens everything: softer climbs, frequent resets, a respectable pedigree. What it does not soften, as the next section shows, is the one thing that decides the outcome. The climb is slower, but it is still exponential, and an exponential climb with a losing run underneath it always reaches the same place. It just arrives politely.

The catch has two layers, and the first is the losing run. Favourites win about a third of races, which means they lose about two thirds, and over 27,355 real races the bad stretches are not a theoretical risk, they are on the record. The worst of them ran 19 races without a single favourite winning. Walk the sequence through that run and the stakes go 1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144, 233, 377, 610, 987, 1,597, 2,584, 4,181, and then the plan asks for 6,765 units on one race. At a £10 base unit, that is a single bet of £67,650, placed by someone whose original idea of a bet was a tenner, with nineteen straight losers immediately behind it. No bookmaker would take it, and no sane person would place it. This is the same cliff the Martingale walks off; Fibonacci simply takes a few more steps to reach the edge.

The second layer is subtler, and it is what makes the headline figure a mirage rather than a lie. The +5.49% is turnover-weighted: total profit divided by total money staked. In a progression, the late-sequence bets are hundreds or thousands of times larger than the base stake, so those few monster bets dominate the average. The "profit" is not a gentle skim off 27,355 races. It is the outcome of a handful of enormous recovery bets that happened, in this particular replay of history, to land. The tens of thousands of ordinary one-unit and two-unit bets underneath are just the favourite's usual slow leak. The plan's result belongs almost entirely to bets that only a bankroll with no upper limit could have placed, and a bankroll with no upper limit is the one thing no punter has.

The test is a full chronological replay, because a staking plan only means anything in the order history actually happened. Progressions live and die by the sequence of results, so unlike the flat-stake systems on the board, this one cannot be measured as a simple average over races. We lined up every qualifying GB race in our record, 27,355 of them, sorted strictly by date and then by off time, and walked the plan through them one bet at a time, exactly as a punter following it would have.

The rules were the plan's own. Back the favourite in every race. Stake in units along the Fibonacci sequence, starting at 1: after any race the favourite loses, step one place further along; after any race a favourite wins, reset to the start. Joint favourites split the stake between them, exactly as the board's other favourite systems handle them.

Settlement follows the Lab's standard honest conventions, the same ones behind the 24 systems on the leaderboard as of 25 June 2026: returns to industry Starting Price, non-runners voided and refunded, fallers and pulled-up horses counted as the losing bets they are, and every race required to have a recorded winner. Nothing about the settlement is softened for the staking plan; the only thing that changes from the flat-stake favourite system is how much goes on each bet. See how we settle every bet for the full conventions.

Two measurement notes, both of which matter for reading the result. The ROI is turnover-weighted: total net profit divided by total money staked, which is the only honest way to score a plan whose stakes vary enormously. And there is deliberately no confidence interval, because a single ordered staking path through history has no exchangeable resamples to bootstrap; you cannot shuffle a story and keep it the same story. What the path gives us instead is its shape, and the shape is the finding.

The numbers below are pulled live from the study file, so they update whenever the record is recomputed.

Read the grid left to right and it tells the whole story in miniature. The turnover-weighted return is +5.49%, and on any other page of this Lab a plus sign would be the headline of the year. The favourite won 34.8% of the 27,355 races, which is why the plan spent most of its life resetting comfortably after short sequences. And then the last two tiles quietly explain where the plus sign came from: a longest losing run of 19 races, and a peak demanded stake of 6,765 units.

Hold those two figures together. A punter staking £10 units needed to put down £67,650 on a single race, having just watched nineteen favourites in a row get beaten, to stay on the path that produces the +5.49%. Miss that bet, or any of the monster bets on the other deep runs, and the paper profit is gone, because turnover-weighting means those few bets are where nearly all the money, and therefore nearly all the result, actually sits. The plan's fate was decided in a handful of moments of maximum absurdity, and history happened, this time, to be kind in those moments.

Strip the progression away and the underlying bet is one we have already measured. Backing every favourite flat to SP over this record returns -8.6%, a leak of −8.6p from every pound staked, and that page shows the loss is structural, built out of the bookmaker's margin. Fibonacci staking does not repair that leak; it wallpapers over it with escalating stakes until, in this replay, a few colossal winning bets happened to land on top. A different shuffle of the same results, one where a deep run arrives when the sequence is already stretched, ends in ruin instead. The favourite's margin is certain; the rescue is luck.

So does Fibonacci staking work? No, and this page earns its place on the board precisely because the raw number says otherwise. The +5.49% is real arithmetic over real races, and it is the most interesting near-miss we have tested, a genuine curiosity worth understanding rather than a claim worth acting on. It exists only for an imaginary punter with a bottomless bankroll, a bookmaker willing to take a 6,765-unit bet from a losing account, and the iron nerve to place it after nineteen straight beaten favourites. Remove any one of those three fictions and the plan collapses into what it always was underneath: the favourite's steady leak of −8.6p per pound, dressed in a more elegant sequence than the Martingale wears.

The general lesson is the one every staking-plan page in this Lab keeps arriving at, from a different direction each time. Staking plans change the shape of your results, not their direction. A progression takes the favourite's small, certain disadvantage and repackages it: long comfortable stretches of small recoveries, paid for by rare moments where everything depends on a bet you cannot afford. Sometimes the repackaging even produces a positive paper number, as it did here, and the number is honest about the history and silent about the ruin required to collect it.

If Fibonacci tempted you because it felt more measured than doubling, the instinct behind that feeling is sound; the sequence is gentler. It is just not gentle enough, because nothing exponential is. The honest tools remain the dull ones: a flat stake you can afford, or a fractional Kelly if you genuinely hold a measured edge. And whether anything we test holds such an edge is a question we publish either way, losses included, in the track record, with the Martingale's fate sitting one page away as this plan's blunter twin.