iGaming Journalist & Crypto Casino Analyst
Every sports bettor who has read about the Kelly criterion has seen the formula, but far fewer have seen what it actually feels like to follow it. This reference presents original Kelly criterion simulation data for sports bettors: 480,000 simulated 1,000-bet seasons at standard -110 odds, showing the median outcome, the spread of results and, most importantly, how deep the drawdowns run under full, half, quarter and double Kelly staking. If you are deciding how much to stake per bet, the drawdown tables below are the numbers that matter.
Key Findings: What Kelly Staking Does to a Betting Bankroll
For a bettor who truly wins 55% at -110, full Kelly stakes 5.5% per bet and suffers a 50%+ drawdown in 98.7% of simulated 1,000-bet seasons. Half Kelly cuts that to 40.2% while still producing a median 2.8x bankroll. Quarter Kelly drops it to 1.4%. Staking double Kelly turns a profitable bettor into a coin flip: only 49% finish above their starting bankroll.
Methodology: How the Simulations Were Run
We simulated a bettor placing 1,000 independent bets at American odds of -110 (decimal 1.909, net payout of $0.909 per $1 risked). For each true win probability from the break-even 52.38% up to 60%, and for each of four staking plans, we ran 20,000 independent 1,000-bet seasons using a seeded NumPy random generator (seed 20261102), for 6 win rates × 4 plans × 20,000 seasons = 480,000 seasons in total. Each bet risks a fixed fraction of the current bankroll (so stakes shrink after losses and grow after wins).
The Kelly fraction for a bet paying b per unit risked with win probability p is f* = (b·p − (1 − p)) / b. At -110, b = 100/110 = 0.9091. Fractional plans multiply f* by 0.25, 0.5 or 2. Because the break-even win rate has zero edge (f* = 0), we instead used flat fractions of 0.25%, 0.5%, 1% and 2% of bankroll for that case. "Drawdown" is measured peak-to-trough on the running bankroll at any point in the 1,000 bets. We assume the bettor's win probability is known and constant, which is the most favorable possible assumption; real bettors estimate their edge with error, which makes the real-world results worse than those shown.
The model ignores account limits, line movement, and the fact that real bets are not all at -110. It also assumes perfectly independent bets. These simplifications keep the numbers reproducible and isolate the pure effect of the staking rule.
The Kelly Formula at -110: Stake Sizes by Win Rate
The Kelly criterion originates in John L. Kelly Jr.'s 1956 paper "A New Interpretation of Information Rate" in the Bell System Technical Journal, and was popularized for gamblers and investors by Edward Thorp. It maximizes the long-run growth rate of a bankroll. At -110 odds, the stake it recommends is strikingly small for the edges bettors can realistically hold.
| True win probability (at -110) | Edge (ROI per $1 risked) | Full Kelly stake | Half Kelly | Quarter Kelly |
|---|---|---|---|---|
| 54.0% | 3.1% | 3.40% | 1.70% | 0.85% |
| 55.0% | 5.0% | 5.50% | 2.75% | 1.38% |
| 56.0% | 6.9% | 7.60% | 3.80% | 1.90% |
| 58.0% | 10.7% | 11.80% | 5.90% | 2.95% |
| 60.0% | 14.5% | 16.00% | 8.00% | 4.00% |
A 54% bettor risks just 3.4% per bet
Beating -110 odds by two percentage points (54% vs the 52.38% break-even) is an excellent long-run result for most bettors, and full Kelly suggests staking 3.4% of bankroll. The 5% staking that many casual bettors call "a big unit" is already above full Kelly for a 54% bettor. Understanding how betting odds work is the first step toward judging what edge you can honestly claim.
Why the numbers are so sensitive to edge estimates
Because the Kelly stake is proportional to edge, overestimating your win rate by two points at the 54% level more than doubles your stake. That is why professional bettors almost universally use a fraction of Kelly, a point we quantify next.
Drawdown Data: How Deep Do Kelly Bettors Fall?
The headline result of this study is that Kelly staking produces very large drawdowns even for bettors with a real edge. The table gives, for each scenario, the median maximum drawdown across 20,000 seasons, and the share of seasons containing a drawdown of at least 30%, 50% and 75%.
| True win rate | Staking plan | Stake per bet | Median max drawdown | Drawdown of 30%+ | Drawdown of 50%+ | Drawdown of 75%+ |
|---|---|---|---|---|---|---|
| 54.0% | Quarter Kelly | 0.85% | 19% | 10.6% | 0.1% | 0.0% |
| 54.0% | Half Kelly | 1.70% | 36% | 73.1% | 14.6% | 0.2% |
| 54.0% | Full Kelly | 3.40% | 62% | 99.9% | 81.2% | 19.7% |
| 54.0% | Double Kelly (2x) | 6.80% | 90% | 100.0% | 100.0% | 89.7% |
| 55.0% | Quarter Kelly | 1.38% | 26% | 32.1% | 1.4% | 0.0% |
| 55.0% | Half Kelly | 2.75% | 47% | 96.6% | 40.2% | 2.5% |
| 55.0% | Full Kelly | 5.50% | 76% | 100.0% | 98.7% | 52.8% |
| 55.0% | Double Kelly (2x) | 11.00% | 98% | 100.0% | 100.0% | 99.7% |
| 56.0% | Quarter Kelly | 1.90% | 31% | 54.6% | 4.1% | 0.0% |
| 56.0% | Half Kelly | 3.80% | 54% | 99.8% | 64.6% | 6.3% |
| 56.0% | Full Kelly | 7.60% | 84% | 100.0% | 100.0% | 77.6% |
| 56.0% | Double Kelly (2x) | 15.20% | 99% | 100.0% | 100.0% | 100.0% |
| 58.0% | Quarter Kelly | 2.95% | 37% | 84.3% | 10.9% | 0.1% |
| 58.0% | Half Kelly | 5.90% | 63% | 100.0% | 90.9% | 16.5% |
| 58.0% | Full Kelly | 11.80% | 91% | 100.0% | 100.0% | 97.2% |
| 58.0% | Double Kelly (2x) | 23.60% | 100% | 100.0% | 100.0% | 100.0% |
| 60.0% | Quarter Kelly | 4.00% | 41% | 95.5% | 19.5% | 0.1% |
| 60.0% | Half Kelly | 8.00% | 69% | 100.0% | 98.5% | 27.9% |
| 60.0% | Full Kelly | 16.00% | 94% | 100.0% | 100.0% | 99.8% |
| 60.0% | Double Kelly (2x) | 32.00% | 100% | 100.0% | 100.0% | 100.0% |
Full Kelly is a rollercoaster
A 56% bettor staking full Kelly (7.6% per bet) sees a median maximum drawdown of 84% and a 50%+ drawdown in 100.0% of seasons. The growth is real, with a median 13.9x bankroll after 1,000 bets, but few humans can tolerate watching 84% of their bankroll evaporate on the way. Sportsbook account limits and the psychological toll make full Kelly impractical in reality.
Half and quarter Kelly buy enormous protection
Moving from full to half Kelly roughly halves the median drawdown in every scenario (for a 55% bettor, 76% down to 47%), while giving up much less growth than most expect: at 55%, median bankroll is 2.8x at half Kelly vs 4.0x at full. Quarter Kelly goes further, cutting the chance of a 50% drawdown to 1.4% at 55%, at the cost of a lower 1.8x median.
Final Bankroll Distributions After 1,000 Bets
Drawdown is only half the story. The next table shows the 10th percentile, median and 90th percentile of the bankroll multiple after 1,000 bets, plus the probability of finishing below the starting bankroll.
| True win rate | Staking plan | 10th percentile | Median | 90th percentile | Chance of finishing below start |
|---|---|---|---|---|---|
| 54.0% | Quarter Kelly | 0.91x | 1.26x | 1.74x | 18.1% |
| 54.0% | Half Kelly | 0.77x | 1.48x | 2.84x | 21.6% |
| 54.0% | Full Kelly | 0.46x | 1.69x | 6.21x | 29.7% |
| 54.0% | Double Kelly (2x) | 0.07x | 1.00x | 13.52x | 51.5% |
| 55.0% | Quarter Kelly | 1.08x | 1.83x | 3.17x | 7.3% |
| 55.0% | Half Kelly | 0.98x | 2.81x | 8.04x | 10.3% |
| 55.0% | Full Kelly | 0.48x | 3.97x | 32.63x | 19.9% |
| 55.0% | Double Kelly (2x) | 0.01x | 0.98x | 67.85x | 51.0% |
| 56.0% | Quarter Kelly | 1.53x | 3.16x | 6.53x | 2.2% |
| 56.0% | Half Kelly | 1.68x | 7.19x | 33.10x | 4.0% |
| 56.0% | Full Kelly | 0.75x | 13.93x | 257x | 12.1% |
| 56.0% | Double Kelly (2x) | 0.00x | 0.94x | 338x | 51.3% |
| 58.0% | Quarter Kelly | 5.18x | 16.01x | 49.47x | 0.1% |
| 58.0% | Half Kelly | 12.20x | 117x | 1,123x | 0.3% |
| 58.0% | Full Kelly | 6.15x | 581x | 54,979x | 4.1% |
| 58.0% | Double Kelly (2x) | 0.00x | 0.72x | 7,684x | 51.0% |
| 60.0% | Quarter Kelly | 35.56x | 164x | 760x | 0.0% |
| 60.0% | Half Kelly | 298x | 6,438x | 138,917x | 0.0% |
| 60.0% | Full Kelly | 254x | 125,438x | 62,001,455x | 0.8% |
| 60.0% | Double Kelly (2x) | 0.00x | 0.35x | 128,439x | 53.6% |
Overbetting destroys value even with an edge
Double Kelly is the single most important result in the data. At every edge level, staking twice the Kelly fraction gives a median final bankroll at or below the starting bankroll (for example 0.98x at 55%), and about half of seasons finish below the start. The theory predicts this: growth rate is symmetric around the Kelly stake and hits zero at twice Kelly. Every fixed fraction beyond twice Kelly has negative expected log growth, regardless of how good the bettor is.
The 10th percentile is the number to read
A bettor needs to survive their bad luck. At 54%, full Kelly yields a 10th percentile of only 0.46x (a 54% loss after 1,000 bets despite a real edge), while quarter Kelly's 10th percentile is 0.91x. Half Kelly sits between at 0.78x. Always compare the bottom of the distribution before choosing a plan.
What Happens at Break-Even Win Rates
At a true win rate of 52.38%, the bettor has no edge before accounting for the sportsbook's vig. Kelly says to stake nothing, and the simulation shows why: any positive stake loses bankroll over time through variance drag.
| Staking plan (flat 0.25%/0.5%/1%/2% of bankroll) | Stake per bet | Median final bankroll | Chance of finishing below start | Median max drawdown |
|---|---|---|---|---|
| 0.25% flat stake | 0.25% | 1.00x | 51.7% | 8% |
| 0.50% flat stake | 0.50% | 0.99x | 53.9% | 16% |
| 1.00% flat stake | 1.00% | 0.96x | 57.1% | 30% |
| 2.00% flat stake | 2.00% | 0.84x | 61.4% | 53% |
Variance drag with zero edge
Even small flat stakes show a median slightly below 1.0x, and the chance of finishing below the start is above 50% in every case (51.7% at 0.25% staking up to 61.4% at 2%). This is the geometric-mean penalty: wins and losses of equal size leave you worse off than you began. It is one reason why sports betting at break-even is a losing proposition once you factor in the cost of the vig.
Practical Staking Guidelines From the Data
1. Use a fraction of Kelly, and make it small
Our simulations suggest that quarter to half Kelly captures most of the growth with drastically lower drawdowns. If you cannot say with confidence that your edge is more than a percentage point or two above break-even, quarter Kelly or less is the more defensible choice.
2. Define your drawdown tolerance before choosing a stake
If a 50% drawdown would make you quit betting, avoid any plan where the dd50 column is above a few percent. For a 55% bettor, that means quarter Kelly (about 1.4% per bet) or less, rather than half Kelly where a 50% drawdown happens 40% of the time.
3. Remember that your edge is an estimate
Backtests overstate edges, and closing-line value is a noisy proxy. Because overestimating edge pushes you toward the dangerous region above twice true Kelly, uncertainty is a strong argument for fractional staking. For U.S. market context, see the guide to legal sports betting in the United States.
Limitations
This is a stylized model: constant -110 odds, constant known win probability, independent bets, and no betting limits. In practice, correlated parlays, line shopping across books, soft-book limiting, and changing edges all affect outcomes. The results are illustrative of the staking mathematics, not a forecast of any bettor's results, and are not financial advice. Betting involves risk; if gambling is a problem for you or someone you know, call 1-800-GAMBLER.
Frequently Asked Questions
What is the Kelly criterion in sports betting?
It is a formula that gives the fraction of your bankroll to wager so as to maximize long-run growth. At -110, f* = (0.9091·p − (1 − p)) / 0.9091, where p is your true win probability.
How much should I bet at a 55% win rate?
Full Kelly is 5.5% of bankroll. Our simulation shows that stake produced a 50%+ drawdown in 98.7% of 1,000-bet seasons, so half Kelly (2.8%) or quarter Kelly (1.4%) is more practical.
Why is half Kelly so popular?
It gives up part of the growth rate but reduces drawdowns sharply. In our data, half Kelly at 55% cut the median maximum drawdown from 76% to 47%, and produced a median 2.8x bankroll vs 4.0x.
What happens if I bet more than Kelly?
Up to twice Kelly, growth is still positive but volatility explodes. At exactly twice Kelly, expected growth is zero. Beyond that, a bettor with a real edge is expected to lose money in the long run.
What win rate do I need to break even at -110?
52.38%, because you must risk $110 to win $100: 110 / 210 = 0.5238.
Sources and Further Reading
- Original Monte Carlo simulation (480,000 seasons of 1,000 bets, seed 20261102), DeucesCracked Editorial, November 2026.
- J. L. Kelly Jr., "A New Interpretation of Information Rate," Bell System Technical Journal, 1956. See the overview at Wikipedia: Kelly criterion.
- Fox Sports, What Is the Vig in Sports Betting?
- TheSpread, Vig and Juice Explained: Hold and Break-Even
- OpticOdds, Hold vs. EV
Cite This Article
If you use data from this article, please link back to https://www.deucescracked.com/blog/kelly-criterion-drawdown-sports-betting-simulation-data. Suggested citation: DeucesCracked Editorial (2026). Kelly Criterion Drawdown Data for Sports Bettors. DeucesCracked.com.
Join the Conversation
Be respectful. No spam. Strategy discussion welcome.