Most backers who have read anything about staking eventually run into the same uncomfortable idea: that betting a flat percentage of your bankroll quietly puts your bigger stakes on losers and your smaller stakes on winners. It sounds wrong the first time you hear it. Then somebody walks through an example and it starts to feel right. Then the term “reverse Kelly” gets attached to it and the conversation drifts into staking systems that promise to flip the curve in your favour.
The concept comes from the American bettor David Schwarz, and the original Race Advisor article on it from 2011 ran into a copyright wall before the maths could be shown in full. Fifteen years on, with Race Advisor sitting on more than 1.6 million UK racing runners, the more useful question is no longer “what did Schwarz mean” but “does the supposed bias actually show up in real bet sequences, and what would fixing it cost you?”
Short answer: not in the way the legend says.
Last updated: 18 May 2026.
What is reverse Kelly betting?
Reverse Kelly betting is a staking idea (popularised by US bettor David Schwarz) which argues that flat-percentage-of-bankroll staking quietly biases your stakes towards losers and away from winners, because your bank size at the moment of each bet rarely correlates with whether that specific bet wins. The “reverse” approach pushes back against that by adjusting stake sizing in the opposite direction to a normal percentage plan. Race Advisor’s 13 years of UK racing data suggest the bias is far smaller than the story implies, and that the cure is usually worse than the disease.
Why Flat-Percentage Staking Can Quietly Under-Stake Your Winners
The intuition is easy to follow. You bet 1% of your bank on every selection. You win the first one, your bank grows, so your next stake is bigger. You lose the next one, the bank shrinks, so the stake after that is smaller. Wins compound your stake size; losses shrink it. The story goes that, because losing runs tend to cluster, you spend a lot of bets staking down just at the moment you would have wanted to be staking up.
It is a tidy story. The problem is that it assumes a pattern: losing streaks systematically followed by winning streaks at predictable points. In horse racing, where each result is largely independent, that pattern doesn’t really exist. What you actually get are runs that look like they should mean something, scattered through a much longer flat distribution of independent outcomes.
Race Advisor put the claim through a 13-year stress test on UK racing using PR Rank 1 as the signal, the top horse by Race Advisor’s proprietary PR rating in each race. Across 154,124 UK runners between 1 July 2013 and 31 March 2026, PR Rank 1 won 33.45% of the time at an A/E of 1.022 against Betfair SP. The longest losing streak in that 13-year sequence was 26 consecutive losses, with 59 individual bets falling inside losing runs of 20 bets or more. That is the variance reality any staking plan has to absorb.
Run a flat 1% staking plan on that sequence from a £10,000 starting bank, and the stake-weighted win rate comes out at 33.520%. The raw, unweighted win rate is 33.452%. The “bias against winners” arrives at roughly seven hundredths of a percentage point, and it points the wrong way. At a flat 0.5% the gap is the same size and the same direction. The supposed structural disadvantage of percentage staking is statistically invisible when the bets are sequenced as they actually occurred.
What “Reverse Kelly” Actually Says, In Plain English
The Kelly Criterion is a formula. It takes your estimated win probability and the price on offer and produces an optimal stake as a fraction of bank for an edge-positive bet. Fractional Kelly (typically a quarter or a half of the full Kelly fraction) is what most disciplined bettors actually use, because full Kelly is brutally volatile.
“Reverse Kelly”, as Schwarz used the term, isn’t a literal inversion of the Kelly formula. It is a staking philosophy that says: if standard percentage-of-bank staking shrinks your stake when you most want it large, then deliberately do the opposite. Increase stake size after losses, hold or reduce after wins. The variant tested in Race Advisor’s simulation took a 1% base, lifted it by 25% for each consecutive loss up to a four-loss cap, and trimmed it by 25% for each consecutive win up to a two-win cap.
That is closer to a fractional Martingale than to anything Kelly would recognise. Which is the first warning sign.
Strict Martingale doubles after every loss until a single win clears the sequence. Schwarz’s version steps up more politely, in 25% increments with a cap. The mechanism is the same in both cases: more money on bets that follow losses. So is the failure mode. If the underlying signal does not have edge, the drawdowns compound and the bank is gone before the assumed bounce-back arrives.
Full Kelly itself has a cousin of this problem. The formula is mathematically optimal when you know your edge, and brutally volatile when you do not. A miscalibrated edge estimate of a few percentage points is enough to flip Kelly from optimal to ruinous, which is why fractional Kelly exists. Any staking plan that scales aggressively with perceived edge is sensitive to errors in the edge estimate, and edge estimates are almost always less precise than the bettor believes.
Three Staking Strategies Tested On 13 Years Of UK Racing Data
Race Advisor’s analysis of those same 154,124 PR Rank 1 bets, all priced at Betfair SP with 2% commission on net winnings, sets out the choice cleanly. All three lines start from a £10,000 bank.
- Flat fixed £10 per bet. Survives the full 154,124 bets. Final bank £29,576. ROI on staked turnover +1.27%. Peak-to-trough drawdown 21.5%. Average stake on winners £10.00. Average stake on losers £10.00.
- Flat 1% of bank per bet. Also survives the full 154,124 bets, but only just. Final bank £2,160. ROI -0.03%. Peak-to-trough drawdown 99.94%. Average stake on winners £198.37. Average stake on losers £197.77.
- Reverse-increment plan (1% base, +25% per consecutive loss, -25% per consecutive win). Busted at bet 95,601. ROI on staked turnover -1.38%. Peak-to-trough drawdown 100%.
Two things are worth pausing on. First, the flat 1% plan’s average stake on winners and losers are within sixty pence of each other across over 150,000 sequenced bets; the famed “bigger stake on losers” effect simply isn’t there at any meaningful size. Second, the plan that deliberately staked up after losses lost everything. Not because the underlying selections got worse, but because the staking strategy concentrated capital into exactly the moments when variance was doing its worst damage.
The same exercise on single Betfair-SP favourites (160,512 UK races where one horse was unambiguously the shortest price, A/E 1.042 after commission) makes a different point. Flat £10 per bet from a £10,000 bank ends at £82,929, a +4.54% ROI with a maximum drawdown of just 10.5%. Flat 1% compounds upward to absurd numbers because the signal had genuine edge. Reverse-increment also compounded upward but with a 99.4% drawdown along the way. A signal with real edge can survive most staking choices. A signal without one cannot be rescued by any of them.
When Reverse Kelly Helps, And When It Wrecks You
There is a narrow theoretical case where increasing stake size after losses is mathematically defensible: when losses are genuinely independent, your edge is real, and your bankroll is large enough that the drawdown won’t break you. In practice, the second and third conditions are where everyone’s planning falls apart.
If your underlying selections have an A/E below 1.00 on Betfair SP after commission, increasing stake size during losing runs is a one-way ticket to ruin. The reverse-increment line on PR Rank 1 demonstrates this exactly: PR Rank 1’s A/E of 1.022 was thin enough that boosting stakes during the losing streaks of 20+ bets it had to absorb 59 times pulled the bank to zero before it could ever cycle back up.
If your underlying selections do have an edge, you don’t need reverse Kelly to capture it. Flat fixed stakes on Betfair-SP favourites delivered +4.54% ROI with a drawdown profile that any serious bettor could sit through. The signal was the story.
A More Honest Staking Question To Ask Yourself
The Race Advisor house position on staking has not changed. Edge first, price second, staking third. If the first two are absent, no staking plan can manufacture a winning bettor, and clever variants that promise to invert the variance curve are usually the most efficient way to bust a small bank fast. If the first two are present, the staking decision is mostly about how much volatility you can tolerate, not whether you can squeeze a hidden few basis points out of the sequencing of your bets.
The better question is not “should I reverse my percentage staking” but “do my selections currently sit above 1.05 A/E on Betfair SP after commission, and across how many bets?” Until that question has a confident answer, every staking conversation is downstream of a much more important one. A/E ratio is the framework that answers it. Bankroll management is the framework that lets you sit through the variance without breaking your discipline. The shortcut is not a clever stake plan; it is a clean edge with enough bank behind it to outlast the inevitable losing run.
If you want to see where your own analytical signal sits on the edge question, Race Advisor’s betting systems framework is the right starting point, and the daily PR-ranked racecards are where the day-to-day work happens. Get the selection right, and the staking question turns out to be much smaller than the staking books make it look.
