Roulette Mathematics · Applied · about 8 min read

Kelly Criterion and Roulette

The Kelly criterion tells a bettor with an edge how much to stake. For standard bets on an unbiased roulette wheel, its answer is zero.

Short answer

Kelly stake = edge / odds. With a negative edge the formula returns a negative fraction, which means do not bet. The criterion helps explain the difference between sizing a bet with a known advantage and setting a spending budget for ordinary roulette.

Kelly fraction
f* = (bp − q) / b
for red
(1 × 18/37 − 19/37) / 1 = −1/37
interpretation
bet nothing
for a 1/30 pocket
≈ +0.6% of bankroll

The criterion

John Kelly (1956) asked how a gambler with a known edge should size bets to maximize long-run growth of the bankrollmoneyBankrollThe amount set aside for play. Bankroll management concerns stakes and limits within that amount.Read full definition →. The answer, for a bet paying b to 1 with win probabilitymathProbabilityA measure of how likely an event is, from zero to one. Percentages express the same scale from 0% to 100%.Read full definition → p and loss probability q = 1 − p, is to stake the fraction f* = (bp − q) / b of the bankroll. Positive f* means bet that fraction; zero or negative means do not bet. Kelly worked at Bell Labs, and his paper, “A new interpretation of information rate”, set the problem in Claude Shannon’s theory of information: a gambler who gets tips over a noisy line, at fair odds, can grow the bankroll at best at the rate the line carries information. This is a growth objective under a probability model, not a promise to finish ahead after a particular number of bets. A spending limit, finite deadline or payoutmoneyPayoutThe amount paid on a winning wager. Check whether the quoted figure is net profit or total return including the stake.Read full definition → cap changes the decision problem.

Illustration of a bettor receiving information by telephone and staking part of a bankroll; not a historical photograph.

Kelly betting in roulette

For redbetRedAn even-money wager on the 18 red numbers. French: rouge.Read full definition →: b = 1, p = 18/37, q = 19/37; f* = (18/37 − 19/37)/1 = −1/37. For a straight upbetStraight upA wager on one number, ordinarily paying a profit of 35 to 1.Read full definition →: b = 35, p = 1/37, q = 36/37; f* = (35/37 − 36/37)/35 = −1/1295. Every standard bet gives a negative Kelly fraction, because every bet has a negative edge, and the growth-optimal stake is zero. Kelly betting on roulette is a formal way of saying what the house-edge page says: no positive stake maximizes expected logarithmic growth when the bet has negative expectation. A particular session can still end in profit. La partageruleLa partageA rule returning half of a qualifying even-money stake when zero lands. On a fair single-zero wheel it gives an edge of 1/74, about 1.35%.Read full definition → needs a separate three-outcome calculation: a win, a full loss or a half-loss on zerowheelZeroThe 0 outcome, conventionally shown in green. Ordinary color, parity, high/low, dozen and column bets do not include it.Read full definition →. Its expected log growth also falls as the stake increases from zero.

betbpf* = (bp − q)/bmeaning
red, European118/37−0.027do not bet
red, la partagethree outcomes18/37 win; 1/37 half-lossbinary formula does not apply0% growth-optimal stake
straight up351/37−0.00077do not bet
straight up on a 1/30 pocket (biased)351/30+0.0057bet 0.57% of bankroll
straight up on a 1/25 pocket351/25+0.0126bet 1.26% of bankroll

A hypothetical positive edge

Suppose a pocketequipmentPocketA compartment on the rotor in which the ball can settle. The number of pockets depends on the wheel.Read full definition → really has a stable, known probability of 1 in 30 and pays 35 to 1. The Kelly fraction is 0.57%: about $57 from a $10,000 bankroll. Under independent spinsmathIndependent spinsSpins for which learning an earlier outcome does not change the probabilities assigned to a later one.Read full definition → and exact fractional staking, the geometric growth factor over 1,000 spins is about 1.7. Expected bankroll grows faster because rare large wins pull up the average. These are model calculations, not evidence that any named historical player used Kelly or achieved this result.

Kelly stake as % of bankroll on a straight up, by true pocket rate%
1/37 (fair)
0% · do not bet
1/35
0.08%
1/33
0.26%
1/30
0.57%
1/25
1.26%
1/20
2.29%
%
01.152.29%
Even a large bias calls for small stakes: the payout is long and variance is high.

How to Detect Roulette Wheel Bias

Typical and average

The gap between the typical bankroll and the average one is wide. Take a pocket that hits 1 time in 25, a hypothetical edge of 44% per unitmoneyUnitA base amount used to express stakes or results.Read full definition → staked. Kelly stakes 1.26% of the bankroll on it. After 1,000 spins the typical bankroll has grown about 11.5 times, while the average, pulled up by a few very lucky runs, has grown about 249 times. Betting more than Kelly raises the average and lowers the typical result: at twice the Kelly stake the typical bankroll only doubles, and by two and a half times the Kelly stake it shrinks, although every bet still carries a 44% edge. In this example, half-Kelly retains about 78% of the expected log growth. This fraction depends on the probabilities and payout.

stake, share of Kellystake, % of bankrolltypical bankroll after 1,000 spins
½0.63%× 6.7
11.26%× 11.5
1½1.89%× 7.4
22.51%× 2.2
2½3.14%× 0.38
33.77%× 0.04
One pocket hitting 1 in 25, paid 35 to 1. The displayed growth factor is exp(1,000 × expected log return). Here it also matches the bankroll after 40 wins and 960 losses. It is not a guaranteed return.

When the edge is a guess

Kelly uses the true probabilities, but a player usually has only estimates. Selecting the busiest pocket from a past record creates a selection problem: even a fair wheel has a busiest pocket. Its observed rate can overstate its future probability. Choosing the stake from that same record compounds the error. An independent follow-up sample helps test a proposed bias (see How to Detect Roulette Wheel Bias). Reducing the calculated fraction reduces exposure, but cannot turn a false edge into a real one.

People with an edge

Haghani and Dewey (2016) studied 61 participants given $25 and 30 minutes to bet on a simulated coin with a disclosed 60% chance of heads. The payout cap was $250; its value was revealed when a proposed winning bet would reach it. Their reported 21% success category included balances of at least $200, while the 28% “bust” category meant less than $2, not necessarily zero. Eighteen participants bet their entire balance on one flip. The uncapped Kelly fraction is 20%, but the cap and deadline change the optimization problem. The experiment illustrates how stake choices matter even with a known advantage; it does not establish one best strategy for every objective.

Illustration contrasting large and small coin-toss stakes; not a depiction of the 2016 experiment.

What to use instead

Without an edge, sizing is a budget question, not a growth question: stake = what you are willing to spend per spin, derived from the hourly budget and the table’s pace (bankroll management page). Kelly’s contribution to the ordinary player is the clarity of its answer: zero. Everything you stake above zero is the price of playing, chosen by you, not optimized by any formula.

A note on fractional Kelly and “money management”

Betting a fixed percentage of the current bankroll is proportional staking: the stake rises after wins and falls after losses. Flat bettingstrategyFlat bettingUsing the same stake on each bet or spin, as specified by the strategy.Read full definition → instead keeps the cash stake fixed. Neither creates an edge on ordinary roulette. With infinitely divisible stakes and a fraction below 100%, proportional staking never reaches exactly zero in finitely many losses, but the balance can become arbitrarily small. Real table minimums can prevent further play much sooner. The growth-optimal fraction remains zero.

Flat Betting Roulette Strategy

Formula

f* = (b·p − q) / b
red: f* = −1/37 · straight up: f* = −1/1295
biased pocket at rate r: f* = (35r − (1 − r)) / 35

Questions

Does Kelly ever say to bet on roulette?

Only when the actual probabilities, payouts and conditions give a credible positive advantage. Standard bets on an unbiased wheel have no such advantage.

Why half-Kelly?

Using half the calculated fraction reduces exposure to errors in the estimated edge. The familiar three-quarters-growth rule is an approximation for small advantages, not a universal identity. It does not protect against an edge that does not exist.

Is Kelly the same as bankroll management?

No. Kelly optimizes growth given an edge. Bankroll management controls spending given none.

Fun with Math

Try these yourself first. Each solution is folded; open it when you are ready.

  1. Suppose a hypothetical bet on one number has a known 44% edge under standard payoutsmoneyPayoutThe amount paid on a winning wager. Check whether the quoted figure is net profit or total return including the stake.Read full definition → and independent spinsmathIndependent spinsSpins for which learning an earlier outcome does not change the probabilities assigned to a later one.Read full definition →. Kelly gives a 1.26% bankrollmoneyBankrollThe amount set aside for play. Bankroll management concerns stakes and limits within that amount.Read full definition → fraction. What if the same 44% edge came on one bet spread over five neighboring numbers?

    Show the solution

    Answer The growth-optimal fraction in this model rises to 7.1%. Expected log growth per bet is about 5.75 times as high. Over 1,000 bets, the corresponding geometric growth factors are about 1.25 million and 11.5; neither is a guaranteed result.

    A straight upbetStraight upA wager on one number, ordinarily paying a profit of 35 to 1.Read full definition → with a 44% edge wins with chance 0.04, since 36 × 0.04 = 1.44. Kelly’s stake is the edge divided by the odds: 0.44/35 = 1.26%. The typical growth per bet is 0.04 × ln(1 + 35 × 0.0126) + 0.96 × ln(1 − 0.0126) ≈ 0.00244, so 1,000 bets multiply the bankroll by about e^2.44 ≈ 11.5.

    Spread one stake over five numbers and the bet wins with chance 0.20 and pays 36/5 for 1, that is 6.2 to 1: still a 44% edge, since 0.2 × 7.2 = 1.44. Kelly’s stake becomes 0.44/6.2 = 7.10%, and the growth per bet 0.01404, about 5.75 times as much. Over 1,000 bets that is e^14.04, about 1.25 million.

    The edge is the same; the swings are smaller. A bet that wins one time in five has a larger growth-optimal fraction in this model, although losses can still reduce the bankroll severely. This comparison assumes the same known edge for both bets; actual predictions need independent evidence.

  2. Assume two numbers are adjacent on the betting layoutequipmentLayoutThe marked betting surface, whether printed on a table or displayed on a screen.Read full definition → and each has a stable, known probabilitymathProbabilityA measure of how likely an event is, from zero to one. Percentages express the same scale from 0% to 100%.Read full definition → of 1 in 25 on independent spins. Compare a splitbetSplitA wager on two numbers sharing an allowed layout boundary, ordinarily paying 17 to 1.Read full definition → between them at 17 to 1 with equal straight-up stakes on the two numbers. How much does Kelly say, and how fast does the bankroll grow compared with one such pocketequipmentPocketA compartment on the rotor in which the ball can settle. The number of pockets depends on the wheel.Read full definition →?

    Show the solution

    Answer It makes no difference: 2.59% of the bankroll on the split, or 1.29% on each number. The typical bankroll grows about 155 times in 1,000 spins, against 11.5 times with one pocket.

    The split wins with probability 2/25 = 0.08 and pays 17 to 1, so it returns 18 × 0.08 = 1.44 per unitmoneyUnitA base amount used to express stakes or results.Read full definition → staked: an edge of 44%, the same as one biased pocket. Kelly’s stake is the edge divided by the odds: 0.44/17 = 2.59%.

    Two straight-up bets of 1.29% each give exactly the same results. When either number comes up, one chipmoneyChipA token used to place a wager, with a printed or assigned value. Also called a check in some casino terminology.Read full definition → wins 35 to 1 and the other is lost, a net of 34 × 1.29% = 44% of the bankroll, as the split pays; when neither does, both are lost.

    The growth per spin is 0.00505, against 0.00244 for one pocket, so over 1,000 spins the typical bankroll grows e^5.046 ≈ 155 times rather than e^2.44 ≈ 11.5 times. The geometric growth factor is larger because the combined bet wins more often at the same 44% edge. It is a model result, not a guaranteed bankroll.

Comic relief

For a standard straight-up bet on a fair single-zero wheel, the growth-optimal stake is zero.

For a standard straight-up bet on a fair single-zero wheel, the growth-optimal stake is zero.

Sources

  1. Kelly, J. L. “A new interpretation of information rate.” Bell System Technical Journal 35(4), 1956.
  2. Breiman, L. “Optimal gambling systems for favorable games.” Fourth Berkeley Symposium on Mathematical Statistics and Probability, vol. 1, 1961.
  3. Thorp, E. O. “Optimal gambling systems for favorable games.” Review of the International Statistical Institute 37(3), 1969.
  4. Thorp, E. O. The Mathematics of Gambling. Gambling Times / Lyle Stuart, 1984.
  5. Epstein, R. A. The Theory of Gambling and Statistical Logic, 2nd ed. Academic Press, 2013.
  6. Taylor, D. G. Games, Gambling, and Probability: An Introduction to Mathematics, 2nd ed. CRC Press, 2021.
  7. Ethier, S. N. The Doctrine of Chances: Probabilistic Aspects of Gambling. Springer, 2010. Chapter 13, roulette.
  8. Haghani, V. & Dewey, R. “Rational decision-making under uncertainty: observed betting patterns on a biased coin.” Working paper, October 2016; SSRN 2856963, arXiv:1701.01427.
Default model: European, single zero; exceptions are stated in the articlemethodology ↗sources ↗