Learn
The Kelly Haircut: a fractional Kelly betting table for when your edge is guessed high
Fractional Kelly betting is the fix for the most common Kelly mistake, an edge guessed too high, and the Kelly Haircut table shows why. If you think a -110 bet wins 55 percent but it really wins 53, full Kelly shrinks your bankroll, half Kelly treads water, and quarter Kelly still grows. This page gives you the formula, the full table at 53 and 55 percent, and the steps to pick your own haircut.
This is the table from the video. It follows the same four steps, in the same order.
The four steps (all hypothetical):
- The guess. You think a -110 bet wins 55 percent, so full Kelly says stake 5.5 percent of your bankroll.
- The miss. The bet really wins 53 percent. At a 5.5 percent stake your bankroll shrinks over time, not just grows slower.
- Half Kelly, about 2.75 percent, lands right around zero growth.
- Quarter Kelly, about 1.4 percent, still grows.
The formula
The Kelly criterion picks the stake that maximizes the long-run growth of your bankroll. Edward Thorp describes it as the strategy that maximizes the expected value of the logarithm of wealth. For a single bet:
f = (b x p - q) / b
where f is the share of your bankroll to stake, b is the profit per 1 staked, p is your win chance and q = 1 - p.
At -110, b = 100 / 110 = 0.909 (the moneyline profit per 1 staked). With p = 0.55, f = (0.909 x 0.55 - 0.45) / 0.909 = 0.055, so 5.5 percent. On a hypothetical $10,000 bankroll that is $550 full Kelly, $275 half Kelly and $137.50 quarter Kelly.
The catch is in the inputs. The Kelly criterion requires accurate probability values, and overestimating the probability raises the risk of ruin. Betting more than the Kelly amount increases that risk too. Nobody knows their true win rate, so every real Kelly stake is built on a guess.
The Kelly Haircut table
Each row is a stake size. The columns show the expected growth per bet (the average change in the log of the bankroll) and what that growth rate compounds to over 1,000 bets, for a bettor whose true rate is 53 or 55 percent at -110. These are our calculations on a hypothetical bettor, not results, and the 1,000-bet figure is the typical (median) path, not a promise of any outcome.
| Stake | Share of bankroll | Growth per bet if true rate 53% | Typical change after 1,000 bets at 53% | Growth per bet if true rate 55% | Typical change after 1,000 bets at 55% |
|---|---|---|---|---|---|
| Double Kelly | 11% | -0.43% | about -99% | -0.002% | about -2% |
| Full Kelly (guess 55%) | 5.5% | -0.073% | about -52% | +0.138% | about +297% |
| Half Kelly | 2.75% | -0.002% | about -2% | +0.103% | about +181% |
| Quarter Kelly | 1.375% | +0.008% | about +8% | +0.060% | about +83% |
Read it in two directions.
Down a column: if your guess of 55 percent is right, full Kelly grows fastest, half Kelly keeps about three quarters of that growth rate and quarter Kelly about 44 percent of it. Betting double Kelly grows about as slowly as betting nothing.
Across a row: the same stake gives very different results depending on the truth. Full Kelly built on a 55 percent guess goes from the best choice to a shrinking bankroll if the truth is 53. Quarter Kelly stays positive in both columns. That is the haircut: you give up some growth when you're right to stay alive when you're wrong.
The Wikipedia summary of fractional Kelly says the same thing in general terms: betting a fraction reduces volatility and the chance of ruin, at the cost of a lower growth rate, and investing less than the Kelly amount is a way to allow for risk you can't measure.
Pick your haircut, step by step
- Write down your estimated win chance and the price for the bet.
- Compute full Kelly with the formula above. If it is zero or negative, there is no bet (at -110 that happens at or below about 52.4 percent).
- Ask how wrong your estimate could be. If you can't rule out being two points high, look at what full Kelly would do at the lower number. In the table, a two-point miss turns full Kelly negative.
- Cut the stake to a fraction that stays positive at your pessimistic estimate. For the hypothetical here, quarter Kelly does; half Kelly only breaks even.
- Cap any single bet at a fixed share of your bankroll as a second guardrail, whatever the formula says.
- Re-size as the bankroll changes. Kelly stakes are shares, so they shrink after losses and grow after wins.
How to estimate your true rate better
The table shows the stake depends on a number you can't see directly, so get the best estimate you can.
- Use a sharp market's fair price as your starting point. Researchers who used probabilities implied by public odds treated the closing market as a strong estimate of the true chance.
- Track closing line value. If your bets keep beating the close, your edge is probably real. Positive CLV means your price beat the closing price, measured after removing the margin.
- Shrink your estimate toward the market when you're unsure. A model that says 58 percent where the market says 50 is more often wrong than brilliant.
What to watch out for
- Simultaneous bets. The single-bet formula assumes one bet at a time. Several bets at once on the same bankroll need smaller stakes.
- Correlated bets. Two bets that tend to win or lose together act like one bigger bet.
- Limits. A stake the formula likes may be more than a book will take.
- Short runs. Even the right stake on a real edge has long losing stretches. The growth figures above are long-run averages.
How to tell it worked
Your sizing is working when a bad month dents your bankroll but doesn't change how you bet, and when your stakes are small enough that you'd still be fine if your edge turned out two points lower than you think. Run your own numbers in the Kelly criterion calculator, and use the expected value calculator to check the edge before you size it. The Quant+ scanner's stake drawer uses fractional Kelly and defaults to quarter.
Quant+ is analytics for informational purposes. 21+. Nothing is guaranteed. If you or someone you know has a gambling problem, call 1-800-GAMBLER.
Questions people ask
What is fractional Kelly betting?
Staking a fixed fraction of the Kelly amount, such as half or a quarter. It lowers the growth rate when your estimate is right but greatly reduces the damage when your estimate is too high.
Is half Kelly or quarter Kelly better?
It depends on how sure you are of your edge. In our hypothetical, half Kelly breaks even if your 55 percent guess is really 53, while quarter Kelly still grows, so the less sure you are, the smaller the fraction.
What happens if I bet more than full Kelly?
Growth falls and the risk of ruin rises. In the hypothetical, double Kelly grows about as slowly as not betting even when the 55 percent estimate is right.
Does Kelly work for parlays and props?
The formula works for any single bet with a known price and chance, but the chance is harder to estimate on props and parlays, which argues for an even smaller fraction.
Sources
- The Kelly Criterion in Blackjack, Sports Betting, and the Stock Market, Edward Thorp (Occidental College copy)
- Kelly criterion, Wikipedia
- Beating the bookies with their own numbers, arXiv (Kaunitz, Zhong, Kreiner)
- Moneyline odds, Wikipedia
- Closing Line Value (CLV), SharpAPI docs
