Kelly Criterion Betting: Optimal Bankroll Management Explained
Learn how the Kelly Criterion formula calculates your exact bet size to maximise long-run growth while protecting your bankroll from ruin.
Category: Guides · By Growl Games Editorial Team · Fri Aug 21 2026 · Updated Fri Aug 21 2026
Table of Contents
- What Is the Kelly Criterion?
- Breaking Down the Kelly Formula
- Step-by-Step Example Walkthrough
- Fractional Kelly: The Safer Alternative
- Kelly vs. Flat Staking vs. Proportional Betting
- Limitations and When Not to Use Kelly
- Kelly Criterion Do's and Don'ts
- Why Growl Games for Serious Bettors
- Frequently Asked Questions
- Sources & Further Reading
Most bettors either bet too much and blow their bankroll in a bad run, or bet too little and leave long-run growth on the table. The Kelly Criterion is the mathematical formula that solves this problem — it tells you the precise fraction of your bankroll to stake on any given bet, given your edge and the odds on offer. Originally derived by physicist John L. Kelly Jr. at Bell Labs in 1956, it has since become the gold standard of bankroll management for professional bettors, poker players, and even institutional investors.
This guide explains exactly how Kelly Criterion betting works, how to calculate it, when to use a fractional version for safety, and where the formula breaks down. If you manage a sports betting or casino bankroll and you want a principled, evidence-based staking system — not a gut-feel heuristic — this is the framework to understand.
What Is the Kelly Criterion?
The Kelly Criterion is a bet-sizing formula designed to maximise the long-run geometric growth rate of a bankroll. It is not a system for guaranteeing wins — no formula can do that. What it does is answer a specific, important question: given that I have a genuine edge on this bet, how much of my current bankroll should I risk?
Bet too small and your growth is slower than it could be. Bet too large and you risk ruin — even with a positive edge, oversizing stakes leads to devastating drawdowns that compound against you. Kelly sits at the mathematical optimum between these two failure modes.
The criterion was originally published in Kelly's 1956 Bell System Technical Journal paper, framed as an information theory problem about telegraph noise. Its application to gambling and investing came later, popularised by Edward Thorp (the blackjack card counter turned hedge fund manager) and later by quantitative traders at firms like Renaissance Technologies.
The core insight: bet a fraction of your bankroll proportional to your edge, and that fraction — no more, no less — maximises long-run wealth accumulation while making ruin theoretically impossible (assuming you never bet more than Kelly prescribes).
Breaking Down the Kelly Formula
The standard Kelly formula for sports betting is:
Where:
- f* — the fraction of your bankroll to stake
- b — the decimal odds minus 1 (i.e., the net profit per unit staked at those odds)
- p — your estimated probability that the bet wins
- q — your estimated probability that the bet loses (q = 1 − p)
An equivalent and often easier form to use with decimal odds is:
The formula has three components worth understanding separately:
Your Edge (the Numerator)
The numerator (bp − q) is your expected value per unit staked. If this is negative — meaning you have no edge — Kelly returns a negative number, which means "don't bet." This is crucial: Kelly only works when you genuinely have an edge over the market. In a standard sportsbook with a margin (overround) of 4–8%, you need your true probability estimate to beat the implied probability embedded in the odds before Kelly recommends any stake at all.
Your Odds (the Denominator)
The denominator b normalises the stake by the size of the potential win. Higher odds = larger potential return per unit = smaller Kelly fraction for the same edge, because you don't need to risk as much to generate the same expected growth.
Why This Fraction Specifically?
Kelly proved mathematically that this exact fraction maximises the expected logarithm of wealth — which is equivalent to maximising long-run compounding. Bet more than Kelly and your long-run growth actually decreases despite your edge. Bet double Kelly and, in the long run, you are expected to lose money even if every individual bet has positive expected value.
Step-by-Step Example Walkthrough
Scenario: You are betting on a football match. The bookmaker offers odds of 2.20 (decimal) on Team A to win. After your own analysis — form, injuries, head-to-head, market movement — you estimate Team A's true probability of winning at 52%. Your current bankroll is £1,000.
Step 1 — Identify your variables:
b = 2.20 − 1 = 1.20 (net profit per £1 staked)
p = 0.52
q = 1 − 0.52 = 0.48
Step 2 — Apply the formula:
f* = (bp − q) / b
f* = (1.20 × 0.52 − 0.48) / 1.20
f* = (0.624 − 0.48) / 1.20
f* = 0.144 / 1.20
f* = 0.12, or 12% of your bankroll
Step 3 — Convert to a stake:
12% of £1,000 = £120 stake
Step 4 — Check the implied edge:
The bookmaker's implied probability at 2.20 is 1/2.20 = 45.5%. Your estimate is 52%. That 6.5-percentage-point edge is genuine — Kelly confirms it's worth betting, and tells you exactly how much.
Step 5 — After the bet resolves:
If Team A wins, your bankroll becomes £1,000 + (£120 × 1.20) = £1,144. Your next Kelly stake is recalculated on £1,144. This is the compounding engine — the bankroll grows geometrically when you're right, and the loss on a defeat is bounded to 12%.
Notice that Kelly is a living fraction, not a fixed stake. Every bet is recalculated on your current bankroll, so stakes grow with wins and shrink with losses — automatically scaling with your financial position. This is fundamentally different from fixed-stake betting or percentage-of-opening-bankroll systems.
Fractional Kelly: The Safer Alternative
Full Kelly is mathematically optimal only under perfect conditions — namely, that your probability estimates are exactly correct. In practice, your p is always an estimate, and overestimating your edge is far more common than underestimating it. Using full Kelly on an edge you've slightly misjudged leads to painful variance.
The standard professional solution is Half Kelly (staking 50% of the full Kelly fraction) or Quarter Kelly (25%). This sacrifices some theoretical growth rate in exchange for dramatically lower drawdowns.
Research by Wizard of Odds and betting mathematicians shows that Half Kelly reduces variance by approximately 75% while only reducing the long-run growth rate by around 25%. For most bettors, that is an excellent trade-off.
Common Fractional Kelly Variants
| Fraction | Stake (from our example) | Variance Reduction | Growth Rate vs. Full Kelly | Best For |
|---|---|---|---|---|
| Full Kelly (1×) | £120 | — | 100% | Mathematically inclined, high confidence in edge estimates |
| Half Kelly (0.5×) | £60 | ~75% | ~75% | Most serious recreational bettors |
| Quarter Kelly (0.25×) | £30 | ~94% | ~44% | High uncertainty, early-stage models, casino play |
| Tenth Kelly (0.1×) | £12 | ~99% | ~19% | Testing a new market, minimal confidence in edge |
The practical recommendation: unless you have a rigorously back-tested, quantitatively validated edge (and most bettors don't), start with Half Kelly. You will still outperform flat staking in the long run, but your bankroll will survive a losing run that would devastate a full-Kelly player.
Kelly vs. Flat Staking vs. Proportional Betting
Understanding how Kelly compares to common alternatives helps clarify when to use each approach.
| Staking Method | Stake Size | Adjusts to Bankroll? | Requires Edge Estimate? | Ruin Risk | Best Use Case |
|---|---|---|---|---|---|
| Flat Staking | Fixed £/$ amount | No | No | Low (if stakes are small) | Casual betting, no edge analysis |
| Level Stakes (%) | Fixed % of current bankroll | Yes | No | Very low (never reaches zero) | Simple, disciplined recreational play |
| Full Kelly | Kelly-optimal % of bankroll | Yes | Yes (critical) | Low if accurate; high if edge overestimated | Professional sports bettors with validated models |
| Half Kelly | 50% of Kelly fraction | Yes | Yes | Very low | Serious bettors wanting growth + safety |
| Martingale | Doubles after each loss | No (grows dangerously) | No | Extremely high | Not recommended for any bankroll management |
The Martingale comparison is instructive. It is the archetypal anti-Kelly: it ignores edge entirely, grows stakes into losing runs (the opposite of what Kelly recommends), and leads to ruin with near-certainty given a sufficiently long session. Kelly explicitly punishes overbetting — and Martingale is nothing but overbetting dressed up in a system.
Limitations and When Not to Use Kelly
Kelly is a powerful tool, but it rests on assumptions that frequently fail in practice. Knowing where it breaks down is as important as knowing the formula.
Garbage In, Garbage Out: The Probability Estimation Problem
Kelly's output is only as good as your probability estimate. If you think a team has a 55% chance of winning and the true probability is 48%, Kelly will tell you to bet a positive fraction — but you are actually on the wrong side of the market. Systematic overconfidence in probability estimates is the single biggest failure mode for Kelly users.
It Assumes Unlimited Divisibility
The formula assumes you can stake any fraction of your bankroll. In reality, sportsbooks have minimum bets, and table games have minimum chips. On very small bankrolls, the prescribed Kelly stake may be unworkable.
It Ignores Bet Correlation
Standard Kelly assumes each bet is independent. In practice, accumulator bets, parlays, and correlated markets (e.g., betting on both the match winner and total goals in the same game) require a multi-asset version of Kelly that is significantly more complex to calculate.
Casino Games: Kelly Has Limited Application
In most casino games — slots, roulette, standard baccarat — the house edge is fixed and negative for the player. The UK Gambling Commission is clear that house-edge games are negative expected value by design. Kelly applied to these games returns a negative number, meaning: don't bet. Where Kelly can apply in a casino context is in games with a genuine player edge — specifically, card counting in blackjack — but even then, the bet-spread constraints imposed by casinos complicate pure Kelly implementation.
Psychological Tolerance for Drawdowns
Full Kelly drawdowns can be severe. In a sequence of independent bets, even correct Kelly staking produces drawdowns of 30–40% of peak bankroll with uncomfortable regularity. Most bettors cannot maintain discipline through those swings. A staking system you can follow consistently is better than a theoretically optimal one you abandon under pressure.
Kelly Criterion Do's and Don'ts
✓ Do's
- Always recalculate your stake based on your current bankroll, not your starting bankroll
- Use Half Kelly (or less) until you have rigorous evidence your probability estimates are calibrated
- Track your estimated probabilities vs. actual outcomes over time to validate your edge
- Apply Kelly only to markets where you have a genuine analytical edge
- Treat the Kelly fraction as an upper bound, not a target you must hit exactly
- Set a hard stop-loss percentage (e.g., 20% of bankroll) and stop betting if you hit it — revisit your model
✗ Don'ts
- Don't apply Kelly to every bet indiscriminately — it only works where you have quantified edge
- Don't use full Kelly unless your edge estimates come from a back-tested, validated model
- Don't combine Kelly with parlays/accumulators without adjusting for correlation
- Don't mistake variance for model failure — a losing run doesn't automatically mean your edge estimate was wrong
- Don't bet the Kelly fraction and then manually override upward on "strong feelings"
- Don't use Kelly as a system for chasing losses in casino games with a fixed house edge
Why Growl Games for Serious Bettors
Applying the Kelly Criterion demands access to sharp, competitive odds — because even a small improvement in the odds on offer meaningfully improves your edge calculation. At Growl Games, the sportsbook is built for bettors who think in edges and fractions, with a broad market selection and fast withdrawals so your bankroll is liquid when you need to recalculate. New players can access the welcome bonus to build starting capital — just factor the wagering requirements into your edge calculation before deploying a Kelly stake against bonus funds.
Frequently Asked Questions
What is the Kelly Criterion in betting?
The Kelly Criterion is a mathematical formula that calculates the optimal fraction of your bankroll to stake on a bet, given your estimated probability of winning and the odds offered. Derived by John L. Kelly Jr. in 1956, it maximises the long-run geometric growth rate of a bankroll. The formula is: f* = (bp − q) / b, where b is the net decimal odds, p is your win probability, and q is your loss probability (1 − p). A positive result tells you how much to bet; a negative result means you have no edge and should not bet.
What is Half Kelly and why do most bettors use it?
Half Kelly means staking 50% of the fraction that the full Kelly formula prescribes. Most bettors — even professional ones — use it because the full Kelly formula assumes your probability estimates are perfectly accurate, which they rarely are. Overestimating your edge by even a few percentage points causes full Kelly to overstake dangerously. Half Kelly reduces variance by roughly 75% while only sacrificing about 25% of the maximum theoretical growth rate, making it a much more practical and psychologically manageable approach for real-world betting.
Can I use the Kelly Criterion in casino games?
In most casino games — slots, roulette, standard baccarat — the house edge is fixed and the player has no positive expected value, so Kelly returns a negative number (meaning: don't bet, or bet zero). Kelly cannot turn a negative-EV game into a profitable one. The one exception is games where a skilled player can gain a genuine edge — most notably blackjack card counting — where Kelly can be applied to the bet spread. But even then, casino countermeasures (shuffles, table limits) constrain pure Kelly implementation.
What happens if I bet more than the Kelly fraction?
Betting above the Kelly fraction — known as "overbetting" — is mathematically proven to reduce your long-run growth rate, even if every individual bet has positive expected value. At double the Kelly stake, expected long-run growth is actually zero. Beyond double Kelly, you are expected to lose money in the long run despite having an edge. The further above Kelly you go, the greater the drawdowns and the higher the risk of ruin. This counterintuitive result is one of Kelly's most important insights: more is not more.
How accurate do my probability estimates need to be for Kelly to work?
Kelly is highly sensitive to the accuracy of your probability estimates. A systematic bias of just 5 percentage points — believing a 45% event has a 50% chance — can cause full Kelly to recommend stakes that are significantly too large, leading to heavy losses over time. This is why Half Kelly or Quarter Kelly is recommended for bettors whose probability models haven't been extensively validated. If you track your estimated probabilities against actual outcomes over hundreds of bets and find your model is well-calibrated, you can move toward a larger Kelly fraction with more confidence.
Does the Kelly Criterion guarantee I won't lose my bankroll?
Theoretically, a bettor who always stakes exactly the Kelly fraction (or less) and never bets more than their current bankroll can never reach zero — because stakes always shrink proportionally with the bankroll. In practice, ruin risk comes from overestimating your edge (making Kelly recommend too large a stake), betting in correlated markets without adjusting the formula, or abandoning the system after a losing run and increasing stakes emotionally. Kelly provides a mathematically sound framework, but it doesn't remove the variance inherent in betting, and it doesn't work without a genuine, verifiable edge.
"The Kelly fraction is not just an optimal bet size — it is a measure of your conviction. If you find yourself wanting to override it upward, that's usually ego talking, not edge."
— Daniel ColeSources & Further Reading
princeton.edu/~wbialek/rome/refs/kelly_56.pdf
wizardofodds.com/gambling/kelly-criterion/
gamblingcommission.gov.uk
eecs.harvard.edu — Thorp Kelly Criterion 2007
sportsbettingdime.com/guides/strategy/kelly-criterion/
igamingbusiness.com