The Kelly Criterion for Horse Racing: Optimal Bet Sizing in Practice

Kelly Tells You Exactly How Much to Bet — If You Trust Your Edge

I discovered the Kelly criterion during a losing run in 2018. My selections were profitable over the season, but my staking was erratic — big on hunches, small on strong opinions, completely backwards. A friend who traded financial markets handed me a photocopied page from a probability textbook and said “this will fix your sizing problem.” He was right, but it took me two years of misapplying it before the fix actually worked.

The Kelly criterion is a mathematical formula that determines the optimal fraction of your bankroll to stake on a bet, given the odds available and your estimated edge. It was developed by John L. Kelly Jr. at Bell Labs in 1956 for an entirely different purpose — optimising signal transmission — but its application to gambling and investment has been studied and debated ever since. The appeal is seductive: a single formula that tells you exactly how much to risk, maximising long-term growth without ever going broke. The catch is that it demands one thing most punters cannot provide with confidence — an accurate estimate of the true probability of winning.

The Kelly Formula Applied to Horse Racing Odds

The formula itself is simple enough to fit on a napkin. In its most common form for fixed-odds betting:

f = (bp – q) / b

Where f is the fraction of your bankroll to stake, b is the decimal odds minus 1 (so for 5/1, b = 5), p is your estimated probability of winning, and q is the probability of losing (q = 1 – p).

Take a concrete example. You are looking at a horse priced at 4/1 (decimal 5.0, so b = 5). After studying the form, you believe the horse has a 25% chance of winning. That makes p = 0.25 and q = 0.75.

f = (5 x 0.25 – 0.75) / 5 = (1.25 – 0.75) / 5 = 0.50 / 5 = 0.10

Kelly says stake 10% of your bankroll. On a 1,000-pound bank, that is 100 pounds on a single 4/1 shot. If that feels aggressive, you are not alone — and your instinct is pointing at the formula’s biggest practical limitation.

Now consider a tighter edge. Same horse at 4/1, but you estimate a 22% win probability instead of 25%.

f = (5 x 0.22 – 0.78) / 5 = (1.10 – 0.78) / 5 = 0.32 / 5 = 0.064

The stake drops to 6.4% of your bank. A three-percentage-point shift in your probability estimate nearly halved the recommended stake. That sensitivity is the formula’s defining characteristic: small errors in estimating p produce large changes in f. And in horse racing, where nobody knows the true probability of any outcome, p is always an estimate, never a measurement.

Full Kelly vs Fractional Kelly: Why Caution Pays

Full Kelly staking — betting exactly the fraction the formula prescribes — maximises the long-term growth rate of your bankroll in theory. In practice, it produces volatility that most human beings cannot stomach. I ran full Kelly for six months in 2019 and watched my bank swing from 2,400 pounds to 900 to 3,100 to 1,200 within a single National Hunt season. The end result was a profit, but the journey was genuinely unpleasant, and I made two emotion-driven errors during drawdowns that cost me more than the formula saved.

The standard solution is fractional Kelly — staking a fixed fraction of what full Kelly recommends. Half Kelly (f/2) is the most common variant. It produces roughly 75% of the long-term growth rate at substantially lower variance. Quarter Kelly (f/4) is even more conservative, and some professional bettors go as low as one-eighth Kelly, treating the formula as a ranking tool rather than a precise prescription.

Betting turnover per race across UK racing fell roughly 8% year-on-year against 2023/24 figures, and 19% against 2021/22 — a sustained contraction that reflects tighter markets and increased regulatory pressure. In that environment, protecting your bank from unnecessary drawdowns is more important than maximising theoretical growth. Half Kelly gives you the benefits of proportional staking — betting more when your edge is larger, less when it is smaller — without the white-knuckle swings that full Kelly inflicts.

My own practice is to calculate full Kelly, then apply a half-Kelly stake, then round down to the nearest sensible figure. If half Kelly says 47 pounds, I stake 45. The rounding costs almost nothing in expected value but keeps my records clean and my emotional state steady. The goal is not mathematical perfection; it is a staking method that I will actually follow consistently, even during losing runs. For a broader comparison of staking approaches, the betting strategy guide covers level stakes and percentage staking alongside Kelly.

Overestimating Edge: The Biggest Kelly Pitfall

Here is the part that nobody tells you when they first recommend Kelly: the formula only works if your estimate of p is accurate. And in horse racing, it almost certainly is not — at least not with the precision Kelly demands.

If you estimate a horse has a 30% chance of winning when it actually has a 22% chance, full Kelly will instruct you to massively overstake. Repeated overstaking on inflated edge estimates does not just slow your bankroll growth — it can destroy it. The Kelly formula is ruthlessly honest: feed it garbage inputs and it produces garbage outputs, but it does so with the same confidence it would show with perfect data. The formula does not know you are wrong.

Problem gambling among horse racing bettors sits at around 2.8% — below the industry average — but over-staking driven by overconfidence is a risk factor that affects disciplined punters too, not just compulsive ones. The Kelly criterion, misapplied, can accelerate losses precisely because it gives a mathematical veneer to what is ultimately a subjective judgment.

The safest approach is to assume your probability estimates are wrong and build that assumption into your method. Half Kelly already provides a buffer. Some punters go further, applying a “confidence discount” to their edge estimate before running the formula: if they think the horse has a 30% chance, they enter 25% as a conservative adjustment, then apply half Kelly to that figure. The result is a stake that is survivable even if the estimate is significantly off.

Another safeguard is minimum sample size. I do not use Kelly staking for a bet type or race category unless I have at least 200 tracked bets in that category showing a positive ROI. Below 200 bets, variance can masquerade as edge. A 15% ROI over 50 bets might be skill or might be luck, and Kelly cannot tell the difference. A 6% ROI over 500 bets is much more likely to reflect genuine ability, and Kelly can be applied to it with greater confidence.

Kelly as a Framework, Not a Religion

The punters I know who use Kelly successfully treat it as a framework for proportional staking, not as a set of commandments. They use the formula to differentiate between a strong-edge bet that deserves a larger stake and a marginal-edge bet that deserves a smaller one. They do not agonise over whether the fraction should be 6.2% or 6.8% because they know the input probability is a rough estimate, not a precise measurement.

What Kelly teaches, more than any specific stake size, is a principle: bet more when your edge is bigger, and bet less when it is smaller. That principle holds true regardless of whether you run the formula precisely or apply it loosely. A punter who stakes 3% of their bank on strong-opinion bets and 1% on weaker opinions is following the spirit of Kelly even if they have never seen the formula. The formula just formalises what good bankroll instincts already suggest.

If you want to start using Kelly, begin with a simple version. Track your bets for a season. Calculate your average estimated probability versus actual win rate by price range. Use those historical figures as your p values rather than estimating from scratch for each bet. Apply half Kelly or quarter Kelly. Review after 200 bets. Adjust. The process is iterative, not one-and-done, and the punters who stick with it tend to converge on a staking rhythm that feels natural and produces steady, undramatic growth — which is exactly what long-term profitable betting looks like.