The Kelly criterion (f* = (bp - q) / b) calculates the optimal fraction of capital to risk per trade to maximize geometric growth over time, where b is the average win/loss ratio, p is the win rate, and q = 1 - p. Derived by J.L. Kelly Jr. in 1956, it is one of the mathematical pillars of systematic money management. Most retail traders either risk too much (threatening account survival) or too little (leaving compounding gains on the table). This guide walks through the formula step by step, its real-world limitations, and how to apply it directly from your backtest results.
What Is the Kelly Criterion?
Mathematical Origin
The Kelly criterion was published in 1956 by J.L. Kelly Jr. of Bell Labs in the Bell System Technical Journal, originally designed to optimize information transmission rates over telephone lines. It was quickly adopted by professional gamblers and quantitative traders, most notably by Ed Thorp, who applied it first to blackjack and then to financial markets in the 1960s.
The core insight: there is a single fraction of your capital that maximizes the logarithmic growth rate over a large number of trades. Bet less and you underperform; bet more and variance increases until ruin becomes inevitable on a long enough timeline.
From Theory to Trading Practice
In trading, the three variables map directly to your backtest results:
- b: average win / average loss (your R:R ratio)
- p: win rate (percentage of winning trades)
- q: 1 - p (percentage of losing trades)
Kelly and prop firm accounts
According to the ESMA, between 74% and 89% of retail CFD accounts lose money. Poor capital allocation is a primary contributor. The Kelly criterion provides a rigorous mathematical framework for sizing positions to avoid overrisking your account on any single trade.
How to Calculate Position Size Using Kelly
The Exact Formula: f* = (bp - q) / b
The full formula:
f = (bp - q) / b*
- f*: fraction of capital to allocate to this trade
- b: win/loss ratio (e.g., 1.5 if you win $150 for every $100 risked)
- p: win rate (e.g., 0.55 for 55%)
- q: 1 - p (loss rate, e.g., 0.45)
With a 1:1 R:R (b = 1), you need a win rate above 50% for f* to be positive. With a 2:1 R:R (b = 2), a win rate of 34% is sufficient. A negative or zero result means the strategy has no edge and should not be traded.
Worked Example with Win Rate and R:R
Take a backtested SMC strategy with the following results over 200 trades:
- Win rate: 55% (p = 0.55, q = 0.45)
- Average win/loss ratio: 1.5 (b = 1.5)
Calculation:
- f* = (1.5 x 0.55 - 0.45) / 1.5
- f* = (0.825 - 0.45) / 1.5
- f* = 0.375 / 1.5
- f = 0.25, or 25% of capital per trade*
This may seem high. It is precisely why virtually all practitioners use a fraction of the full Kelly value rather than the raw calculated number.
| Win Rate | R:R | Kelly f* | Half-Kelly | Quarter-Kelly |
|---|---|---|---|---|
| 45% | 2:1 | 2.5% | 1.25% | 0.6% |
| 50% | 1.5:1 | 8.3% | 4.2% | 2.1% |
| 55% | 1.5:1 | 25% | 12.5% | 6.25% |
| 60% | 1:1 | 20% | 10% | 5% |
| 65% | 1:1 | 30% | 15% | 7.5% |
Kelly Fractions: Half-Kelly and Quarter-Kelly
In practice, systematic traders typically use 25% to 50% of the calculated Kelly value:
- *Half-Kelly (f/2)**: preserves roughly three-quarters of full Kelly's geometric growth while significantly reducing drawdowns. The most widely used fraction among professional systematic traders.
- *Quarter-Kelly (f/4)**: recommended when your backtest sample is under 150 trades, where estimation uncertainty is still too high to justify a larger fraction.
- Fixed 1-2% risk: the practical standard for beginners and prop firm challengers operating under strict drawdown rules.
Full Kelly warning
Full Kelly can produce drawdowns of 40% to 60% of capital during losing streaks, even for strategies that are profitable long-term. According to PropJournal, only 8 to 10% of traders pass FTMO Phase 1. Oversized position fractions are among the most common reasons for failure in prop firm challenges.
Limitations of the Kelly Criterion in Trading
Sensitivity to Win Rate Estimates
The Kelly criterion is highly sensitive to its inputs. A 5% error in win rate or R:R estimation dramatically changes the recommended fraction. Backtest statistics are estimates over a finite sample, not absolute truths.
With 100 trades and an observed win rate of 55%, the statistical confidence interval is wide. The true win rate might be anywhere between 45% and 65%, producing very different Kelly fractions. The smaller your sample, the more conservative your chosen fraction should be.
The solution: wait until you have at least 150 to 200 backtested trades before applying Kelly, and always start with half-Kelly or quarter-Kelly. See the expectancy and profit factor guide for a rigorous approach to evaluating the statistical reliability of your backtest results.
Significant Drawdowns During Losing Streaks
Full Kelly maximizes growth at infinity but allows substantial drawdowns on short horizons. Running a thorough multi-timeframe backtest across multiple years gives you a realistic picture of the worst consecutive losing streaks your strategy has historically produced. This information is essential for choosing a Kelly fraction compatible with your risk tolerance.
When Not to Use Kelly
Avoid using Kelly in these scenarios:
- Fewer than 100 backtested trades, where win rate estimates are too imprecise
- Strategies with correlated trades, such as simultaneous open positions on correlated currency pairs
- Prop firm accounts with strict drawdown rules: prop firm risk management requirements may conflict with the calculated Kelly fraction
- Strategies with fat-tailed return distributions, where rare extreme losses invalidate the formula's underlying assumptions
Alternatives and Complementary Approaches
Fixed Percent Risk per Trade (1-2% of Capital)
The simplest and most robust method: risk a fixed percentage of capital on every trade. This approach requires no win rate estimates, guarantees strong protection against ruin, and is compatible with all prop firm constraints.
A fixed 1% risk per trade is the industry standard for FTMO and MFF challengers starting out. At 2%, you can sustain 50 consecutive losing trades before wiping an account, which is statistically improbable for any viable trading strategy.
Optimal F by Ralph Vince
Ralph Vince developed Optimal F as an empirical alternative to the Kelly formula. Rather than using the theoretical formula, Optimal F tests different fractions directly on historical trade data to find the one that maximizes geometric capital growth.
Advantage: it captures the real distribution of returns, including extreme outliers. Drawback: it is susceptible to overfitting the historical data. Always validate on out-of-sample data before applying Optimal F to a live account.
Volatility-Based Position Sizing (ATR)
A complementary approach: adjust position size based on market volatility using the ATR (Average True Range). When volatility is higher, reduce your position size to maintain constant monetary risk per trade. When volatility drops, scale up proportionally.
This method works especially well for equity and futures strategies where volatility varies significantly across market regimes. It can be combined with Kelly for more sophisticated risk management.
| Method | Complexity | Robustness | Best For |
|---|---|---|---|
| Fixed 1% risk | Very low | Very high | Beginners, prop firms |
| Half-Kelly | Medium | High | 100+ backtested trades |
| Full Kelly | Medium | Low | 300+ trades, advanced traders |
| Optimal F (Vince) | High | Medium | Out-of-sample validated strategies |
| ATR-based sizing | High | High | Equities, futures, variable volatility |
Integrating Position Sizing Into Backtesting
Impact of Sizing on Metrics (Sharpe Ratio, Max Drawdown)
Your choice of position sizing method directly affects every performance metric in your backtest. With fixed 1% risk, max drawdown is predictable and proportional to the number of consecutive losing trades. With full Kelly, metrics behave non-linearly and drawdown peaks can be severe.
Analyzing the impact of sizing on your Sharpe ratio is particularly important: increasing your fraction may boost absolute returns while degrading the risk-adjusted ratio if drawdowns grow disproportionately faster than gains.
Backtesting Different Kelly Fractions
The recommended workflow for integrating Kelly into your backtesting:
Extract base metrics
Calculate f* and its fractions
Simulate capital curves
Validate on out-of-sample data
Implement and monitor
Backtrex lets you backtest strategies across years of historical data and extract the exact metrics needed for Kelly calculations, all without writing a single line of code. The anti-repainting simulation engine ensures that win rates and R:R ratios from backtests reflect real-world performance as closely as possible.
Important Risk Warning
Conclusion
The Kelly criterion is a powerful tool for calibrating trade size from backtest results. It provides a mathematically optimal fraction for long-term geometric growth, but it is highly sensitive to win rate and R:R estimates. In practice, half-Kelly combined with a rigorous backtest on at least 150 trades is the ideal starting point for systematic traders.
For a deeper dive into evaluating your strategy's statistical quality, see the backtest metrics guide and the complete guide to avoiding backtesting mistakes. Start backtesting Kelly configurations on Backtrex without any code.
FAQ
The Kelly criterion is a mathematical formula (f* = (bp - q) / b) that calculates the optimal fraction of capital to risk per trade to maximize long-term geometric growth. It uses win rate (p) and win/loss ratio (b) as inputs. A negative or zero result means the strategy has no mathematical edge and no position should be taken.
The full Kelly criterion is mathematically optimal but rarely used in practice because it requires precise win rate estimates and can generate 40-60% drawdowns during losing streaks. Half-Kelly (50% of the calculated value) is the practical standard: it preserves most of the growth benefit while significantly reducing variance. For prop firm accounts, a fixed 1-2% risk per trade is often the safer choice.
Extract win rate and average R:R from your backtest results, then apply f* = (bp - q) / b. Multiply f* by your total capital to get the dollar amount to risk per trade. On Forex, divide this by the monetary value of your stop loss in pips to get your lot size. Backtrex provides win rate and R:R directly from your backtest results.
Most systematic traders use between 25% and 50% of the calculated Kelly value. Half-Kelly is the standard for strategies with 150-300 backtested trades. Below 100 trades, quarter-Kelly or fixed 1% risk are preferable because the estimation uncertainty is too large to justify a higher fraction.
The Kelly criterion uses a theoretical formula based on p and b. Ralph Vince's Optimal F tests different fractions directly on historical trade data to optimize geometric growth empirically. Optimal F better captures the real return distribution including extreme outliers, but carries a significant risk of overfitting the historical data. Always validate Optimal F on out-of-sample data before going live.
With a 1:1 R:R (b = 1), you need a win rate strictly above 50% for f* to be positive. With a 2:1 R:R (b = 2), a 34% win rate is sufficient. The minimum win rate formula is p_min = 1 / (1 + b). Any strategy below this threshold has no mathematical edge according to Kelly and should not be traded.
You can use Kelly as a reference calculation, but most prop firm challengers use more conservative fractions (1-2% fixed risk) due to strict trailing drawdown rules. FTMO's 10% maximum drawdown and 5% daily loss limits constrain position sizing independently of what Kelly recommends. Always check your prop firm's specific rules against your calculated Kelly fraction before applying it to a funded account.