The Kelly Criterion Explained
The Kelly Criterion is the mathematically optimal formula for sizing bets when you have a genuine edge. Developed by physicist John L. Kelly Jr. in 1956, it tells you exactly what fraction of your bankroll to stake on any bet to maximise long-run growth — without risking ruin. This guide explains the formula, how to apply it in practice, why most bettors use fractional Kelly, and what happens when your probability estimates are wrong.
What Is the Kelly Criterion?
The Kelly Criterion tells you what percentage of your bankroll to bet on a given outcome to maximise the long-term growth rate of your bankroll. Bet more than Kelly says and you grow slower and risk ruin. Bet less than Kelly says and you leave growth on the table. The Kelly stake is the mathematically optimal point.
John L. Kelly Jr. published his formula in 1956 while working at Bell Labs — originally applied to signal transmission problems. Its application to gambling and investing was immediately recognised, and it has been used by professional gamblers and fund managers ever since. Ed Thorp — the mathematician who beat blackjack and later ran one of the most successful hedge funds in history — was among its earliest and most prominent proponents.
The Kelly Criterion solves a specific problem: given a bet with known odds and an estimated probability of winning that is higher than the implied probability in the odds, how much should you stake? Bet too little and you grow slowly. Bet too much and variance can wipe you out even with a genuine edge. Kelly finds the optimal middle ground.
The Kelly Criterion is a staking system for bettors who have identified a positive expected value opportunity — where their estimated probability of winning is higher than the bookmaker's implied probability. If you do not have a genuine edge, Kelly cannot create one — it only optimises how you exploit an edge that already exists. Applied to negative expected value bets, Kelly will tell you to bet zero (or close to it).
The Kelly Formula
The Kelly formula calculates the optimal fraction of your bankroll to stake as a single expression combining your edge and the odds on offer.
f* = (bp − q) ÷ b
Where:
f* = the fraction of bankroll to bet
b = the net decimal odds (decimal odds − 1)
p = your estimated probability of winning
q = your estimated probability of losing (1 − p)
A useful alternative expression using decimal odds directly:
f* = (p × d − 1) ÷ (d − 1)
Where:
f* = fraction of bankroll to bet
p = your estimated probability of winning
d = decimal odds offered by the bookmaker
If f* is negative, the bet has negative expected value — do not bet.
The result — f* — is expressed as a decimal fraction of your bankroll. If f* = 0.05, Kelly says to bet 5% of your total bankroll on this bet. If f* = 0.15, bet 15%. If f* is negative, the bet has negative expected value and should not be placed at all.
Worked Examples
Example 1 — A Clear Value Bet
You estimate a team has a 55% chance of winning a match. The bookmaker offers odds of 2.10 (implied probability 47.6%). You have a genuine edge.
p = 0.55 (your estimated win probability)
d = 2.10 (decimal odds)
b = d − 1 = 1.10 (net odds)
f* = (p × d − 1) ÷ (d − 1)
f* = (0.55 × 2.10 − 1) ÷ (2.10 − 1)
f* = (1.155 − 1) ÷ 1.10
f* = 0.155 ÷ 1.10
f* = 0.141 (14.1%)
Bankroll: £1,000
Kelly stake: £1,000 × 0.141 = £141
Example 2 — A Smaller Edge
You estimate a horse has a 30% chance of winning. The bookmaker offers 4.00 (implied probability 25%). Smaller edge, longer odds.
p = 0.30 | d = 4.00 | b = 3.00
f* = (0.30 × 4.00 − 1) ÷ (4.00 − 1)
f* = (1.20 − 1) ÷ 3.00
f* = 0.20 ÷ 3.00
f* = 0.067 (6.7%)
Bankroll: £1,000
Kelly stake: £1,000 × 0.067 = £67
Example 3 — Negative Expected Value
You estimate a team has a 40% chance of winning. The bookmaker offers 2.10 (implied probability 47.6%). The odds are wrong — you are the underdog here.
f* = (0.40 × 2.10 − 1) ÷ (2.10 − 1)
f* = (0.84 − 1) ÷ 1.10
f* = −0.16 ÷ 1.10
f* = −0.145
Why Kelly Maximises Growth
The Kelly Criterion is not arbitrary — it is derived mathematically as the staking fraction that maximises the expected logarithm of wealth over repeated bets. This is what makes it the theoretically optimal long-run growth strategy.
Overbetting Kelly — Growth Slows Then Reverses
Bet more than the Kelly fraction and your expected growth rate decreases. At 2× Kelly you have the same expected growth as betting nothing at all. At more than 2× Kelly your expected bankroll decreases — you are on a path to ruin even with a genuine edge.
Underbetting Kelly — Safe But Slower
Bet less than the Kelly fraction and your bankroll grows slower than it could — but you are never at risk of ruin. This is why most practitioners use fractional Kelly: you sacrifice some growth rate in exchange for dramatically reduced variance.
Kelly Is the Sweet Spot
The Kelly stake maximises the geometric growth rate of the bankroll. Over a large number of bets, no fixed fractional staking strategy produces higher expected bankroll growth than full Kelly — given the same edge and the same bets.
Kelly and Variance
Even with a genuine edge and perfect Kelly sizing, losing runs are inevitable. Kelly does not eliminate variance — it simply optimises the growth-to-risk ratio. Drawdowns of 50% or more are possible even following Kelly perfectly on a sequence of high-edge bets.
Fractional Kelly — The Practical Approach
Full Kelly is theoretically optimal — but in practice, virtually every professional bettor uses a fraction of Kelly. The most common approach is half Kelly (betting 50% of the full Kelly stake) or quarter Kelly (25%). Here's why.
The Kelly formula assumes your probability estimate is perfectly accurate. In reality, your estimated probability is always uncertain — and if you overestimate your edge, full Kelly will cause you to overbet. A 10% overestimate of your edge at full Kelly produces the same long-run result as overbetting. Fractional Kelly provides a safety margin against probability estimation errors.
75% of Full Kelly Growth
Half Kelly produces approximately 75% of the long-run growth rate of full Kelly — but with dramatically reduced variance. The maximum drawdown is roughly halved. For most practical bettors, half Kelly is the optimal balance between growth and risk of ruin. It is the most widely recommended fractional Kelly size.
Very Conservative — High Safety
Quarter Kelly produces around 56% of full Kelly growth but is extremely conservative. Suitable for bettors who have low confidence in their probability estimates or who are just beginning to apply Kelly to their betting. Almost eliminates ruin risk from estimation errors at the cost of slower bankroll growth.
Fractional Kelly — Worked Example
Full Kelly stake calculated at 14.1% of a £1,000 bankroll = £141.
Full Kelly
Stake: £141 (14.1% of bankroll)
Max growth rate.
High variance.
Sensitive to estimation errors.
Half Kelly
Stake: £70.50 (7.05% of bankroll)
~75% of max growth.
Significantly lower variance.
Recommended for most bettors.
Quarter Kelly
Stake: £35.25 (3.5% of bankroll)
~56% of max growth.
Very low variance.
Suitable for uncertain edge estimates.
Risks and Limitations
Kelly Requires Accurate Probability Estimates
The formula is only as good as your probability input. If you consistently overestimate your edge — a common problem — you will systematically overbet and your bankroll will grow more slowly or not at all. The discipline required to assign accurate probabilities is at least as important as the staking formula itself.
Full Kelly Produces Large Swings
Full Kelly bets can be uncomfortably large — 10%, 15% or even 20%+ of bankroll on a single bet. The psychological pressure of watching a 15% bankroll stake lose is significant. Many bettors who intellectually accept Kelly find they cannot emotionally execute it at full size. Fractional Kelly solves this.
Kelly Is Designed for Single Independent Bets
The formula in its basic form is derived for sequential, independent bets — not simultaneous bets. When placing multiple bets at the same time (e.g. across a Saturday's matches), the stakes interact. Simultaneous Kelly requires portfolio-level adjustment that is complex to implement in practice.
Bookmaker Limits Can Make Kelly Irrelevant
If Kelly says bet £200 on a selection but the bookmaker's maximum stake is £20, the theoretical Kelly stake is academic. In practice, bettors with genuine edges are frequently restricted by bookmakers before Kelly-sized stakes become viable — making exchange betting and exchange-focused strategies more practical for serious value bettors.
Kelly vs Flat Staking
Flat staking — betting the same fixed amount on every bet — is the most common approach among recreational bettors. How does it compare to Kelly over the long run?
Simple — But Suboptimal
Fixed stake per bet regardless of edge or odds. Easy to implement, easy to track. Does not account for bet quality — a bet with a 20% edge gets the same stake as one with a 2% edge. Over the long run, flat staking grows a bankroll with an edge but at a slower rate than Kelly. The bigger the variation in edge between bets, the more flat staking underperforms Kelly.
Optimal — But Requires Estimates
Stake proportional to edge — bigger bets on bigger edges, smaller bets on smaller edges. Maximises long-run growth rate. Requires accurate probability estimates for every bet. Stakes vary in size, which complicates record keeping. Stakes shrink after losses (protecting the bankroll) and grow after wins — which is the correct mathematical behaviour but can feel counterintuitive.
Many professional bettors use a tiered flat staking approach as a practical approximation of Kelly — betting one unit on small edges, two units on moderate edges and three units on high-confidence large edges. This captures most of the benefit of differential staking without requiring a precise probability estimate for every bet. It's not mathematically optimal but it's a significant improvement over pure flat staking.
How to Apply Kelly in Practice
This is the most important discipline in applying Kelly correctly. Form your probability estimate independently — then compare to the bookmaker's implied probability. If you look at the odds first, you are likely to anchor your estimate to them and undermine the entire process. The estimate comes first; the odds comparison comes second.
✅ Estimate first — odds secondIf the formula returns a negative number, Kelly is explicitly telling you not to bet. This is one of the most valuable features of the system — it provides a clear, unemotional signal for when to pass on a bet entirely. Discipline in not betting when f* ≤ 0 is as important as sizing correctly when f* is positive.
📊 Negative f* = no bet, no exceptionsUnless you have exceptional confidence in your probability estimates — built from a large, well-recorded track record — default to half Kelly as your stake size. The growth rate reduction is modest; the protection against estimation error and variance is significant. Most professionals operate at between 25% and 50% of full Kelly.
Kelly stakes are a percentage of your current bankroll — not your starting bankroll. After a winning run, your bankroll grows and Kelly stakes increase proportionally. After a losing run, Kelly stakes shrink — protecting what remains. This dynamic adjustment is what makes Kelly superior to flat staking over the long run. Recalculate your bankroll regularly — at minimum after each significant winning or losing period.
✅ Kelly stake = % of current bankroll, not starting bankrollThe only way to validate your probability estimates over time is to record them and compare to actual outcomes. If you estimate 60% probability across 100 bets and win 55% of them, your estimates are systematically overconfident — and your Kelly stakes have been too large. Calibration tracking is the feedback loop that makes Kelly-based betting genuinely powerful over time.
Common Questions
If you can't form a genuine probability estimate independent of the bookmaker's odds, Kelly cannot be properly applied. In this case, flat staking or a conservative percentage staking plan is more appropriate. Kelly is a tool for bettors who have developed a systematic approach to estimating probabilities — through statistical models, deep domain knowledge or a combination of both. Without a genuine independent estimate, the formula produces an arbitrary number that has no meaningful interpretation.
In theory yes — you would apply the same formula using the combined odds of the accumulator and the combined probability (your estimated probability for each leg multiplied together). In practice, accumulator bets compound both the odds and the variance significantly, and Kelly stakes on accumulators are typically very small — even on high-edge legs. Most professional bettors who use Kelly focus on single bets rather than accumulators, as accumulators reduce the edge per pound staked due to the compounding bookmaker margin.
Yes — Kelly or fractional Kelly is used by a significant portion of professional sports bettors and quantitative traders. Ed Thorp applied it famously in blackjack and later in financial markets. Many professional betting syndicates use Kelly as the basis for their staking models — though almost always in fractional form (half or quarter Kelly) with additional constraints for simultaneous bets and maximum exposure per event. It is also widely used in algorithmic trading as the basis for portfolio sizing.
There's no hard rule — but most experienced practitioners cap individual bets at 5–10% of bankroll regardless of what the full Kelly formula suggests, particularly for sports betting where probability estimates are inherently uncertain. A full Kelly stake of 25%+ of bankroll suggests either an exceptionally large edge (which should be scrutinised carefully) or an overconfident probability estimate. When in doubt, apply a percentage cap as a safety measure alongside fractional Kelly.
Simple percentage staking — betting a fixed percentage of your bankroll on every bet (e.g. always 2%) — is similar to Kelly in that it scales with bankroll size, but it bets the same percentage regardless of the size of the edge or the odds. Kelly varies the stake based on the specific edge and odds for each individual bet — betting more when the edge is large and less when it's small. This dynamic adjustment is what makes Kelly mathematically superior to fixed percentage staking for bettors with varying-size edges across their bets.
Kelly staking starts with finding genuine value — odds that are higher than the true probability warrants. Our live odds comparison shows prices across all major bookmakers simultaneously, making it easier to spot where the market is mispriced.
View Live Odds Comparison →