What Is the Kelly Criterion?
The Kelly Criterion — also known as the Kelly formula, Kelly strategy, or Kelly bet — is a mathematical formula that calculates the optimal fraction of your capital to allocate to a bet or investment in order to maximize the long-term growth rate of your bankroll. It was developed in 1956 by John L. Kelly Jr., a scientist at Bell Labs, originally to solve a problem in information theory. Within years of its publication, gamblers and investors recognized its power as a position sizing tool, and it has been at the core of professional money management ever since.
The Kelly Criterion answers a deceptively simple question: given that you have an edge in a repeating game, what fraction of your capital should you bet on each round to make your bankroll grow as fast as possible over time?
The answer Kelly proved is exact: bet the fraction that maximizes the expected logarithm of wealth after each round. This is called the Kelly fraction or f* (f-star). Bet more than f* and your growth rate actually decreases — even though your individual bets are larger. Bet less and you grow slower than optimal but more smoothly. Bet exactly f* and no other strategy grows your bankroll faster over the long run.
Among the investors known to have used Kelly-based position sizing are Edward Thorp (who applied it to blackjack and then the stock market), Warren Buffett, Charlie Munger, Bill Gross, and the late Renaissance Technologies founder Jim Simons. It is considered the gold standard of position sizing theory.
The Kelly Formula — How It Works
The Kelly Criterion formula for a binary outcome (win or lose) is:
f* = (b × p − q) / b
Where:
f* = optimal fraction of bankroll to bet (0 to 1)
b = win/loss ratio (how much you WIN per unit risked)
p = probability of winning (0 to 1)
q = probability of losing = 1 − p
Simplified form:
f* = p − q/b
The formula has an elegant interpretation: f* equals your expected gain minus your expected loss, normalized by the win/loss ratio. When your edge is zero (b×p = q), Kelly says bet nothing. When your edge is positive, Kelly gives you the exact optimal fraction.
Breaking down each component
- b (Win/Loss Ratio) — The ratio of what you win to what you lose per unit bet. In stock trading, this is your risk-reward ratio: if your target is $150 above entry and your stop is $50 below entry, b = 150/50 = 3.0. In sports betting, if you bet $100 and win $200 net, b = 2.0.
- p (Win Probability) — Your estimated probability of a winning outcome, expressed as a decimal. A 55% win rate = p = 0.55. This is the hardest input to estimate accurately in real markets.
- q (Loss Probability) — Simply 1 − p. No separate estimation required.
- f* (Kelly Fraction) — The output: the optimal fraction of your total bankroll to risk on this trade. If f* = 0.25, risk 25% of your capital. If f* is negative, the trade has negative expected value — don't take it.
Worked example
- Win Probability: 55% → p = 0.55, q = 0.45
- Win/Loss Ratio: 2.0× → if you risk $1 and win, you gain $2
- Bankroll: $50,000
f* = (2.0 × 0.55 − 0.45) / 2.0 = (1.10 − 0.45) / 2.0 = 0.65 / 2.0 = 0.325 = 32.5%
Full Kelly says: risk $16,250 of your $50,000 on this trade.
Half Kelly (recommended in practice): risk $8,125.
Edge = b×p − q = 2×0.55 − 0.45 = +0.65 per $1 risked
Break-even win probability = 1/(1+b) = 1/3 = 33.33% — your 55% easily clears this.
The growth rate formula
At the Kelly fraction, the expected log-growth rate per round is:
G(f) = p × ln(1 + b×f) + q × ln(1 − f)
G is maximized when f = f* (the Kelly fraction)
At f = 2×f*, G = 0 (same as not betting at all)
At f > 2×f*, G < 0 (you're losing money long-term despite positive edge)
This is the mathematical proof that over-betting is as bad as under-betting — and betting more than twice the Kelly fraction actually destroys wealth over time. The growth rate curve is the visualization shown in the chart on our Kelly Calculator.
Purpose and Core Objective
The Kelly Criterion has one specific purpose: maximize the long-term compounded growth rate of your capital given a repeated series of bets or trades with known edge.
This sounds similar to simply "maximizing expected value" — but it is fundamentally different. Maximizing expected value at each round would tell you to bet your entire bankroll every time you have a positive edge. That strategy leads to ruin with near-certainty over enough rounds, because one loss wipes you out. Kelly optimizes for the geometric mean of outcomes (the long-run compound growth rate), not the arithmetic mean (expected value of a single bet).
| Strategy | Optimizes For | Long-Term Result | Ruin Risk |
|---|---|---|---|
| Bet everything | Single-round EV | Near-certain ruin | Extremely high |
| Fixed dollar amount | Simplicity | Suboptimal growth | Low |
| Fixed % of bankroll (non-Kelly) | Drawdown control | Suboptimal growth | Low |
| Kelly Criterion (f*) | Long-run growth rate | Maximum possible growth | Theoretically zero |
| Half Kelly | Growth + drawdown balance | ~75% of Kelly growth | Very low |
The Kelly Criterion also has a remarkable theoretical property: given infinite rounds, a Kelly bettor will have more wealth than any other bettor with probability 1. No other strategy has this guarantee. This is why it is considered the theoretical optimum of position sizing.
Applications: Stocks, Betting, and Beyond
The Kelly Criterion was originally developed in the context of information theory but has found applications across any field where capital is repeatedly risked against uncertain outcomes with known or estimable probabilities.
Stock and equity trading
For individual stock trades, Kelly can be applied when the trader has a defined entry point, profit target, and stop-loss level. The win/loss ratio b is the risk-reward ratio (target distance ÷ stop distance), and p is the trader's estimated probability of hitting the target before the stop. Kelly then gives the fraction of the account to risk on that trade — where "risk" means the amount you would lose if the stop is hit.
- Entry: $100 | Target: $115 | Stop: $95
- Win amount per share: $15 | Loss amount per share: $5 → b = 3.0
- Estimated win probability: 50% → p = 0.50, q = 0.50
f* = (3×0.50 − 0.50) / 3 = 1.0/3 = 33.33%
On a $50,000 account: risk $16,667 (but risk here = stop loss amount, not full position size)
Options and asymmetric payoffs
Kelly adapts naturally to options because the win/loss ratio can be very asymmetric — a long option can gain 300% while the maximum loss is 100%. With b = 3.0 and a 40% estimated probability of the option expiring in the money: f* = (3×0.40 − 0.60)/3 = 0.60/3 = 20%.
Sports betting and gambling
The Kelly Criterion was popularized in gambling contexts first. Given decimal odds O (e.g., 2.50 means you win $1.50 on a $1 bet), b = O − 1. If you estimate the true probability of the outcome at 55%: f* = (1.50×0.55 − 0.45)/1.50 = 0.375 = 37.5%. Bet 37.5% of your bankroll. Professional sports bettors almost universally use Kelly or a fraction of it.
Venture capital and portfolio management
For concentrated portfolios with high-conviction positions, Kelly provides a framework for how large any single position should be relative to the portfolio. Managers like Buffett and Munger have referenced Kelly thinking when describing their willingness to hold very large positions in companies where they have high confidence in their edge estimate.
Cryptocurrency trading
Crypto traders use Kelly for sizing positions on individual trades, particularly in momentum strategies where win probability and risk-reward ratios can be estimated from backtests. Given the high volatility of crypto assets, most practitioners use Quarter Kelly or lower to manage drawdown.
Key Characteristics of the Kelly Criterion
1. Maximizes geometric growth rate — not expected value
Kelly optimizes the expected value of the logarithm of wealth — the geometric mean return — not the arithmetic expected value. This makes it inherently about long-term compounding. A strategy that maximizes expected value at each step (always bet 100% when edge exists) will almost surely go bankrupt over many rounds. Kelly avoids this by keeping position sizes conservative enough that losses can always be recovered.
2. The Kelly fraction scales naturally with edge
When your edge is small (e.g., win probability just slightly above break-even), Kelly prescribes a small bet. When your edge is large, Kelly prescribes a larger bet — but always with a built-in ceiling. The fraction f* cannot exceed 1 for positive-edge bets with a win/loss ratio greater than 1, and in practice it is usually well below 50%. This self-calibration is one of Kelly's most elegant properties.
| Win Prob | Win/Loss Ratio | Full Kelly | Half Kelly | Edge |
|---|---|---|---|---|
| 35% | 2.0× | 2.50% | 1.25% | +0.050 |
| 45% | 2.0× | 17.50% | 8.75% | +0.350 |
| 55% | 2.0× | 32.50% | 16.25% | +0.650 |
| 60% | 3.0× | 46.67% | 23.33% | +1.200 |
| 30% | 2.0× | −5.00% | — | −0.100 |
3. Over-betting is always worse than under-betting
The growth rate curve G(f) is concave — it rises from 0 to a peak at f*, then falls back to 0 at 2×f*, and becomes negative for f > 2×f*. This means that betting twice the Kelly amount gives the exact same long-run growth as not betting at all — despite having a positive edge. And betting more than 2× Kelly destroys capital long-term. The implication: when in doubt, bet less.
4. Kelly is path-independent in theory
Because Kelly betting is a fixed fraction of current wealth, the final result is independent of the order of wins and losses. A sequence of W-L-W-L-W produces the same bankroll as L-W-L-W-W when using Kelly sizing, because you always bet a fixed fraction of whatever you currently have. This is not true for fixed-dollar sizing strategies.
5. Half Kelly delivers ~75% of maximum growth at much lower risk
One of the most important practical results in Kelly theory is the efficiency of fractional Kelly strategies. Half Kelly (betting f*/2) captures approximately 75% of the maximum growth rate while dramatically reducing drawdown and variance. Quarter Kelly delivers ~44% of maximum growth. This efficiency curve means the cost of conservatism is smaller than most investors assume.
Limitations — Why Kelly Is Harder to Apply Than It Looks
The Kelly Criterion is theoretically perfect. In practice, it is extremely challenging to apply correctly — and the consequences of misapplying it can be severe. Understanding these limitations is essential before using Kelly sizing with real money.
1. Estimating win probability (p) is very difficult
The Kelly formula is highly sensitive to the win probability p. A small error in p produces a large error in f*. In a casino, p can be calculated exactly from known odds. In stock markets, p is unknown — it must be estimated from backtests, analysis, historical patterns, or subjective judgment. If you estimate p = 0.55 but your actual edge is 0.50 (no edge), Kelly will tell you to bet 32.5% of your bankroll on a trade that has zero expected value.
This is called edge overestimation — the most dangerous error in Kelly betting. Studies of professional traders consistently show that self-assessed win probabilities are overconfident by 5–10 percentage points. This alone is why most practitioners use Half Kelly or less: it provides a substantial buffer against overconfidence.
2. The win/loss ratio (b) is also difficult to estimate
In stock trading, the win/loss ratio is not fixed. Your actual exit may be different from your planned target or stop — slippage, gap downs, early exits, and holding winning trades longer all change the effective b. A defined risk-reward ratio at entry (e.g., 2:1) does not guarantee an actual 2:1 ratio at exit. The Kelly fraction calculated from estimated b may not match the true optimal fraction.
3. Kelly assumes bets are independent — markets are not
The Kelly formula in its basic form assumes each bet is independent of previous bets. In financial markets, trades are correlated — particularly during market stress, when many positions move in the same direction simultaneously. A portfolio of 5 positions sized at 20% Kelly each assumes they are independent; if they are all correlated long equity positions, the effective bet is much larger than 5 independent 20% bets.
4. Full Kelly drawdowns are psychologically intolerable
Even with a correctly estimated edge, Full Kelly produces very large drawdowns. The well-known Kelly result is that the probability of ever halving your bankroll at Full Kelly is approximately 50% — even on a winning strategy. Drawdowns of 70–80% from peak are not unusual for Full Kelly strategies. Most investors cannot tolerate this psychologically and exit at the worst possible moment, converting a mathematically sound strategy into a practical failure.
| Limitation | Severity | Practical Solution |
|---|---|---|
| Overestimating win probability p | Very High | Use Half Kelly or less as default |
| Inaccurate win/loss ratio b | High | Use historical actual exits, not planned exits |
| Correlated positions | High | Portfolio Kelly with correlation adjustments |
| Full Kelly drawdown | Very High | Half Kelly or Quarter Kelly |
| Short time horizon | Medium | Kelly is a long-run formula; accept variance |
| Non-stationary probabilities | High | Re-estimate inputs regularly; use conservative fractions |
5. Kelly is a long-run formula — results vary wildly in the short run
Kelly maximizes the growth rate asymptotically — over an infinite number of rounds. Over 50 or 100 trades, the actual result of a Kelly strategy can deviate dramatically from the expected median path. In any given 100-trade sample, a Half Kelly strategy might outperform Full Kelly simply due to variance. The mathematical optimality of Kelly is a statement about very long sequences — not short-term performance.
Full Kelly vs Half Kelly — What Professionals Actually Use
The gap between the theoretical Kelly Criterion and how professionals apply it in practice comes down to one key adaptation: fractional Kelly. Rather than betting the full f*, most experienced practitioners bet a fraction of it — typically Half Kelly (f*/2).
Why Half Kelly is the industry standard
Half Kelly captures approximately 75% of Full Kelly's maximum growth rate while reducing variance by approximately 75%. This means:
- You give up only 25% of potential growth
- Your drawdowns are dramatically smaller
- Your bankroll is far more robust to edge overestimation
- You are much more likely to actually follow the strategy through losing periods
Even if your true Kelly fraction is 25% and you use Half Kelly (12.5%), your growth rate is still ~75% of optimal. But if your true Kelly is 20% and you bet Full Kelly at what you estimated to be 25% (a 5-percentage-point overestimate), you may be in the over-betting zone — potentially destroying wealth despite having a genuine edge. Half Kelly provides a crucial buffer against this error.
| Fraction | Kelly Used | Growth Rate Efficiency | Recommended For |
|---|---|---|---|
| Full Kelly | f* | 100% | Theoretically optimal only — high confidence required |
| ¾ Kelly | 0.75×f* | ~94% | Very high confidence in edge estimate |
| Half Kelly ⭐ | 0.5×f* | ~75% | Recommended default for most traders |
| ⅓ Kelly | 0.33×f* | ~57% | Conservative traders, uncertain edge |
| Quarter Kelly | 0.25×f* | ~45% | Very conservative, smooth equity curve |
| 10% Kelly | 0.1×f* | ~20% | Beginners, very uncertain edge |
The practical rule of thumb, widely used in professional trading: calculate Full Kelly, then use Half Kelly as your actual position size. If you are uncertain about your edge estimate, use Quarter Kelly. Never use more than Full Kelly.
How to Use Our Kelly Criterion Calculator Pro — Tab by Tab
Our Kelly Criterion Calculator Pro has five tabs, each designed for a specific use case — from finding your optimal position size in seconds, to comparing how different Kelly fractions compound over hundreds of trades.
Tab 1: Kelly % — Calculate your optimal position size
The core calculator. Enter your win probability, win/loss ratio, account size, and the Kelly fraction you want to apply (default: 50% = Half Kelly). Results update in real time as you type. You'll see:
- Full Kelly % — the mathematically optimal fraction (hero display)
- Half Kelly and Quarter Kelly percentages
- Position size in dollars at Full Kelly and at your applied fraction
- Edge — expected value per $1 risked
- Rounds to double your bankroll at Full Kelly
- Break-even win probability — minimum win rate needed for positive edge
- Edge strength rating and contextual alert (No Edge / Aggressive / Solid)
- Growth rate curve chart — shows G(f) vs fraction, with markers at your key fractions
- Win Probability: 55% | Win/Loss Ratio: 2.0×
- Account: $50,000 | Apply Fraction: 50%
→ Full Kelly: 32.50% | Half Kelly: 16.25% | Position Size (Applied): $8,125 | Rounds to Double: ~7 | Break-even: 33.33%
Tab 2: Scenarios — See how edge changes across win probabilities
Fix your win/loss ratio and define a win probability range (e.g., 30% to 75%). The calculator generates a full table showing Full Kelly %, Half Kelly, edge, growth rate per round, rounds to double, and color-coded status (red = no edge, amber = low edge, green = bet) for every probability step in your range. The break-even win probability is shown as the hero, and a dual-line chart plots Full Kelly vs Half Kelly vs win probability.
This tab is ideal for backtesting different strategies: if you run a strategy that wins 60% of the time with a 2:1 risk-reward ratio, the Scenarios tab instantly shows you the Kelly sizing for that setup and how much it changes if your win rate drops to 50%.
Tab 3: Portfolio — Multi-position Kelly allocation
Enter up to 8 simultaneous positions — each with a name, win probability, and win/loss ratio. Set your account size and the fraction of Kelly to apply. The calculator shows per-position Kelly %, applied percentage, position size in dollars, and edge for each trade. The summary shows total Kelly allocation, remaining cash, positions with positive edge, and average edge per position. A grouped bar chart compares Full Kelly % vs Applied % for each asset.
- AAPL: Win Prob 60%, W/L 2.5 → Kelly 44.00%
- TSLA: Win Prob 45%, W/L 3.0 → Kelly 26.67%
- BTC: Win Prob 40%, W/L 4.0 → Kelly 25.00%
Account $100K | Apply 50%: Total Kelly 95.67% | Total Positions $47.8K | Cash $52.2K
Tab 4: Simulator — Watch your bankroll grow over N rounds
Enter your starting bankroll, win probability, win/loss ratio, and number of rounds (up to 500). Optionally add a custom Kelly fraction. The simulator plots four expected median growth trajectories simultaneously — Full Kelly, Half Kelly, Quarter Kelly, and your custom fraction — using the log-normal growth formula. Final bankroll values for all four fractions are shown in the result cards. This tab makes it viscerally clear why Full Kelly grows explosively in expectation but also why Half Kelly is often the more practical choice.
- Full Kelly (32.50%): $190.67M expected median
- Half Kelly (16.25%): $18.04M expected median
- Quarter Kelly (8.13%): $872K expected median
- Custom 75% (24.38%): $105.94M expected median
Note: These are theoretical median trajectories. Actual results vary significantly due to variance — Full Kelly has much higher variance than Half Kelly.
Tab 5: Fractions — Compare all Kelly fractions side by side
Enter win probability, win/loss ratio, and account size. The calculator automatically generates a comparison table for six standard fractions (Full, ¾, Half, ⅓, Quarter, 10%) plus an optional custom fraction — showing Kelly % used, position size in dollars, growth rate per round, efficiency vs Full Kelly (with a visual bar), and rounds to double. The hero shows the recommended fraction. The growth rate curve chart shows the full G(f) parabola with scatter-point markers at each fraction, making the over-betting penalty visually obvious.
Common Mistakes When Applying Kelly
Using Full Kelly with estimated (not backtested) probabilities
The single most dangerous Kelly mistake is using Full Kelly sizing based on intuitive estimates of win probability rather than rigorously backtested data. If you feel that your strategy wins 60% of the time but it actually wins 50%, Full Kelly will have you over-betting significantly — destroying wealth while appearing to have a sound strategy. Always backtest your win probability and win/loss ratio over at least 100 trades before applying any Kelly fraction larger than Quarter Kelly.
Treating the Kelly fraction as a position size — it is a risk fraction
In stock trading, Kelly gives you the fraction of your capital to risk — not the fraction to invest in the position. If your account is $50,000, Kelly says 20%, and your stop loss is 5% below entry, you should risk $10,000 — but your position size is $10,000 / 0.05 = $200,000. This is a highly leveraged position, not a $10,000 position. Confusing the risk fraction with the investment fraction leads to dramatic under-sizing.
Applying Kelly to a single trade without considering the rest of the portfolio
Kelly sizing for a single trade assumes all your capital is available for that one trade. If you have 5 simultaneous positions, sizing each at Full Kelly could result in total allocation far exceeding 100% of your capital. Use the Portfolio tab to see your aggregated allocation. As a rule: if the sum of all Kelly fractions across your portfolio exceeds 50%, consider reducing each position.
Changing strategy during a losing streak
Kelly strategies go through extended losing periods — sometimes dramatically so at Full Kelly sizing. The correct response is to continue applying Kelly sizing to each subsequent trade (which will naturally be smaller, since Kelly is a fraction of current bankroll). The catastrophic error is to increase bet size to "make back" losses — the opposite of Kelly — or to abandon the strategy entirely after a drawdown that is statistically expected within the strategy's variance.
Not adjusting Kelly inputs as market conditions change
Win probability and win/loss ratios are not constant. A momentum strategy that wins 60% of the time in trending markets may win only 45% in choppy markets. Applying the same Kelly fraction in both environments means over-betting during unfavorable conditions. Re-estimate your inputs periodically — at minimum quarterly — and adjust your Kelly fraction accordingly.
Ignoring the difference between expected and median outcomes
The Simulator tab shows "expected median trajectory" — the 50th percentile outcome. At Full Kelly, the expected (arithmetic mean) bankroll is much higher than the median, because the distribution of outcomes is highly right-skewed. This means the "average" outcome looks fantastic, but the typical outcome is much more modest. Half Kelly reduces this skew, making the typical outcome much closer to the expected outcome.
Frequently Asked Questions
What is the Kelly Criterion in simple terms?
The Kelly Criterion is a formula that tells you exactly what fraction of your money to bet on a trade or wager to make your bankroll grow as fast as possible over time. It takes two inputs — your win probability and your win/loss ratio — and outputs the optimal position size. Bet more than Kelly recommends and your long-run growth rate actually decreases. Bet less and you grow slower but more smoothly.
What is the Kelly formula?
The Kelly formula is f* = (b × p − q) / b, where f* is the optimal fraction of your bankroll to bet, b is the win/loss ratio (how much you win per unit risked), p is the probability of winning, and q = 1 − p is the probability of losing. A simpler form is f* = p − q/b. If f* is negative, the bet has negative expected value — don't take it.
How do I find the win probability and win/loss ratio for my trades?
For backtested strategies, win probability = (number of profitable trades) / (total trades). Win/loss ratio = average profit on winning trades / average loss on losing trades. For forward-looking trades, win probability must be estimated based on your analysis — and should be treated conservatively. For defined risk-reward setups (fixed target and stop), win/loss ratio = target distance / stop distance.
Why should I use Half Kelly instead of Full Kelly?
Half Kelly gives approximately 75% of Full Kelly's growth rate while reducing variance and drawdown by approximately 75%. In practice, win probabilities are almost always estimated with some error, and even small overestimates cause significant over-betting at Full Kelly. Half Kelly provides a safety margin against estimation error while still capturing the vast majority of Kelly's growth advantage. Most professional traders use Half Kelly or less.
What does a negative Kelly fraction mean?
A negative Kelly fraction means the trade has a negative expected value — your estimated edge is negative. Kelly says the optimal bet is $0: do not take this trade. This happens when your win probability is below the break-even threshold 1/(1+b). For example, with a 2:1 win/loss ratio, you need a win probability above 33.33% for the trade to have positive expected value.
What is the break-even win probability?
The break-even win probability is the minimum win rate needed for a trade to have a positive expected value — the point where Kelly fraction = 0. It is calculated as p_breakeven = 1 / (1 + b), where b is the win/loss ratio. For a 2:1 risk-reward ratio: 1/(1+2) = 33.33%. You need to win more than 1 in 3 trades to have a positive edge at 2:1 payoff. The Scenarios tab shows the break-even probability for any win/loss ratio you enter.
Can I use the Kelly Criterion for a portfolio of multiple positions?
Yes, for independent positions. Calculate Kelly % for each position separately, then add them up. If the total exceeds 100%, you would theoretically need leverage — which is a warning sign to reduce position sizes. In practice, positions are often correlated, which means the true aggregate risk is higher than the sum of individual Kelly fractions suggests. The Portfolio tab calculates per-position Kelly and shows the total allocation at a glance.
Does the Kelly Criterion guarantee profits?
No. Kelly guarantees that no other long-term strategy grows faster, given correctly estimated inputs. It does not guarantee profits over any finite number of trades, and it does not protect against estimation errors. In the short run, Kelly strategies go through significant drawdowns. The guarantee is strictly asymptotic — over a very large number of trades with correct inputs, Kelly outperforms every other strategy with probability approaching 1.
Is the Kelly Criterion Calculator free to use?
Yes. The Kelly Criterion Calculator Pro on StockToolHub is completely free with no registration, account, or subscription required. All five tabs — Kelly %, Scenarios, Portfolio, Simulator, and Fractions — are fully accessible with no limitations and no sign-up required.
Calculate your optimal position size now
Free, instant, no sign-up — five Kelly calculators in one professional tool.
Open Kelly Criterion Calculator →