The Kelly Criterion, developed by John L. Kelly Jr. at Bell Labs in 1956, answers a specific question: given a bet with known probability of winning and known payoff ratio, what fraction of your bankroll should you bet to maximize long-term wealth growth? Applied to trading, it translates to: given a strategy with known win rate and average win/loss ratio, what fraction of your trading capital should you allocate per trade? The answer from Kelly: exactly the fraction that produces the maximum geometric growth rate without over-betting. Under-betting Kelly leaves returns on the table. Over-betting Kelly produces lower geometric returns than Kelly (due to excessive drawdowns reducing the compounding base) and eventually leads to ruin. For automated crypto trading, Kelly provides a principled framework for position sizing that goes beyond arbitrary percentage rules. This guide covers the Kelly formula, full vs fractional Kelly for crypto's volatility, and how to apply Kelly sizing to DennTech strategies.
Related guides: Position Sizing, Expectancy, Win Rate vs Profit Factor.
Kelly Criterion Formula
Kelly Fraction (f*) = W - ((1 - W) / R) Where: W = Win Rate (probability of a winning trade) R = Win/Loss Ratio (average win ÷ average loss) f* = Fraction of capital to bet per trade Example: Win Rate = 55% (W = 0.55) Average Win = $120, Average Loss = $80 R = $120 / $80 = 1.5 Kelly = 0.55 - ((1 - 0.55) / 1.5) Kelly = 0.55 - (0.45 / 1.5) Kelly = 0.55 - 0.30 Kelly = 0.25 → Bet 25% of capital per trade (Full Kelly) Half Kelly = 12.5% Quarter Kelly = 6.25% Negative Kelly (don't bet): If Win Rate = 40%, R = 1.0: Kelly = 0.40 - (0.60 / 1.0) = 0.40 - 0.60 = -0.20 Negative Kelly means this strategy has negative edge — do not trade
Full Kelly vs Fractional Kelly for Crypto
| Kelly Fraction Used | Growth Rate | Max Drawdown Risk | Recommendation |
|---|---|---|---|
| Full Kelly (100%) | Maximum theoretical | Very high — drawdowns can exceed 50% | Not recommended for crypto |
| Half Kelly (50%) | ~75% of max growth | Moderate — manageable for aggressive traders | Experienced traders only |
| Quarter Kelly (25%) | ~50% of max growth | Low-moderate — suitable for most traders | Recommended default |
| Tenth Kelly (10%) | ~25% of max growth | Low — very conservative | Capital preservation priority |
Why full Kelly is not recommended for crypto: Kelly assumes perfectly known win rate and R — but in crypto trading, both are estimated from backtests with uncertainty. When the true Kelly fraction is smaller than estimated (due to backtest overfitting), full Kelly over-bets and causes excessive drawdowns. Fractional Kelly (25–50%) provides a safety margin against estimation error while capturing significant growth potential.
Kelly with ATR-Based Stop Losses
When using ATR stops in DennTech, the loss per trade (L) is defined by the ATR stop distance: L = ATR × multiplier as a percentage of entry price. This makes Kelly calculation straightforward: calculate R = average take-profit / ATR stop loss ratio from your backtest, use your backtest win rate, and compute Kelly fraction. This integration of Kelly with ATR-based stops produces position sizes that are both mathematically optimal for growth and volatility-adjusted. See our ATR Stops guide and position sizing guide.
Frequently Asked Questions
- How accurate does my win rate estimate need to be for Kelly Criterion to be reliable in practice?
- This is the central practical limitation of Kelly for trading: Kelly requires accurate estimates of win rate and win/loss ratio, but backtested statistics are estimates with uncertainty. A 10-percentage-point error in estimated win rate can cause Kelly to recommend significantly different (and potentially dangerous) position sizes. For example, if your strategy's true win rate is 50% but your backtest estimates 60%, Kelly's recommendation for 60% win rate with R=1.5 is 0.60 - (0.40/1.5) = 33%; for the true 50% win rate, Kelly is 0.50 - (0.50/1.5) = 17% — nearly double the optimal size. Using quarter Kelly (25% of Kelly fraction) automatically corrects for this estimation uncertainty by providing a buffer. The more backtest data you have (500+ trades rather than 50), the more reliable the win rate estimate and the more confidently you can approach full Kelly. With limited backtest data, use 10th Kelly as a conservative starting point. See our backtesting guide. Compare editions at the pricing page.
- What are the practical limits on position size that Kelly might recommend for a strong crypto strategy?
- A high-win-rate crypto strategy with favorable R can produce surprisingly large Kelly fractions. Example: win rate 65%, R = 2.0 → Kelly = 0.65 - (0.35/2.0) = 0.65 - 0.175 = 47.5% of capital per trade. Full Kelly at 47.5% means nearly half your capital on a single trade — extremely risky in crypto's volatile environment. Quarter Kelly would give 11.9% per trade — much more manageable. In practice, consider capping Kelly-derived position sizes at a maximum regardless of the formula output: no more than 10–20% of total capital in any single trade. Even if Kelly mathematically recommends 40%, the fat-tailed nature of crypto returns (sudden exchanges hacks, flash crashes, adverse manipulation) makes that level of concentration inappropriate. DennTech's position sizing calculator applies Kelly with a built-in maximum cap for safety. Explore the live demo. Start at the pricing page.
- Should I recalculate Kelly sizing as my strategy's win rate and R change over time?
- Yes — Kelly sizing should be treated as a living calculation that updates as your strategy accumulates more live trading data. The win rate and R from a backtest are estimates; live trading produces the true parameters. As you accumulate live trades (ideally 100+ to have statistical significance), recalculate your Kelly fraction using live trade data rather than backtest estimates. If live win rate is lower than backtest (common due to backtest overfitting), reduce your Kelly fraction accordingly. If live performance matches backtest, you can potentially increase from quarter to half Kelly. Update Kelly calculation quarterly as part of your strategy review process. DennTech's monthly review framework includes win rate and R tracking — this data feeds directly into Kelly recalculation. See our monthly review guide. Start at the pricing page.
Position sizing: Kelly Criterion (this guide), Position Sizing, Expectancy. All at the strategies page.