The Kelly Criterion is one of the most mathematically interesting ideas in trading — a formula that claims to calculate the precisely optimal fraction of your bankroll to risk on each trade to maximize long-run wealth growth. Developed by John L. Kelly Jr. at Bell Labs in 1956 as a solution to information theory problems, it was subsequently applied to gambling and investing by Ed Thorp (the MIT blackjack genius turned hedge fund manager) and has since become a standard reference in professional trading circles.
For crypto bot traders, the Kelly Criterion provides a rigorous mathematical framework for thinking about position sizing that goes beyond simple rules-of-thumb. However, it also has critical limitations that make direct application dangerous — and understanding those limitations is as important as knowing the formula. This guide is a deep dive covering the math, the practical limitations, how to use Kelly as a framework rather than a hard rule, and how it integrates with DennTech's position sizing module.
For the foundational position sizing concepts including the simpler fixed-percentage method that most traders should use first, see our position sizing guide.
The Kelly Formula
The basic Kelly formula for a simple win/lose bet:
f* = (b × p - q) / b
Where:
- f* = the optimal fraction of bankroll to bet
- b = the net odds received on the bet (profit per unit risked) = Reward/Risk ratio
- p = probability of winning = Win rate
- q = probability of losing = 1 - p
Simplified for trading (with Reward-to-Risk ratio = R):
Kelly % = W - (1 - W) / R
Where W = win rate and R = average winner / average loser.
Example Calculation
Strategy metrics from 3 months of paper trading: 56% win rate, 1.6:1 reward-to-risk ratio (average winner $160, average loser $100).
Kelly % = 0.56 - (1 - 0.56) / 1.6 Kelly % = 0.56 - 0.44 / 1.6 Kelly % = 0.56 - 0.275 Kelly % = 0.285 = 28.5%
Kelly says risk 28.5% of your account on each trade. In theory, this maximizes long-run wealth growth. In practice, this would blow up your account during a normal losing streak.
Why Full Kelly Is Dangerous for Crypto Trading
The Kelly formula's results assume:
- Your edge estimates are exactly correct. If your true win rate is 52% (not 56%) and your true R is 1.4 (not 1.6) — very plausible estimation errors — the formula gives you a very different result. Using Kelly based on imprecisely estimated inputs is a recipe for overbetting.
- There is no path dependency in practice, Kelly produces extreme bankroll volatility. Even with correct edge estimates, a losing streak will reduce your account dramatically before the long-run edge compounds. Most human traders (and automated bots with drawdown circuit breakers) would stop trading during this drawdown — preventing recovery.
- Your edge does not change over time. Markets are non-stationary — your strategy's edge changes with market conditions. A strategy with 56% win rate in a ranging market may have 44% win rate in a trending market. Kelly calculated on historical data does not account for regime changes.
Crypto's high volatility amplifies all these risks. The combination of imprecise edge estimates, regime changes, and the emotional/circuit-breaker pressure of volatile drawdowns makes full Kelly extremely dangerous in practice.
The Practical Kelly Variants
Half Kelly
Risk f*/2 per trade. In our example: 28.5% × 0.5 = 14.25%. Still high for crypto, but provides significantly reduced volatility compared to full Kelly while retaining approximately 75% of the long-run growth rate. Professional traders who use Kelly-based sizing generally cap at Half Kelly.
Quarter Kelly
Risk f*/4 per trade. In our example: 28.5% × 0.25 = 7.1%. More moderate, suitable for strategies where edge estimates are uncertain. Retains approximately 56% of the theoretical optimal growth rate while dramatically reducing drawdown volatility.
Capped Kelly
Calculate Kelly but cap the result at a maximum percentage regardless of what Kelly recommends. For crypto bots, a typical cap is 3–5% maximum per trade:
Bet size = min(Kelly %, Maximum Cap)
This is the most practical implementation for crypto bot sizing — use Kelly as a floor check ("is my current sizing at least remotely justified by my edge?") while a hard cap prevents dangerously large bets even when Kelly suggests them.
Using Kelly to Evaluate Your Current Sizing
Rather than using Kelly to set your position size, use it to evaluate whether your current sizing makes mathematical sense:
- Calculate your strategy's win rate and R ratio from at least 50 paper trades
- Apply the Kelly formula to get f*
- If your current risk per trade is below f*/4 (Quarter Kelly): Your sizing is very conservative relative to your estimated edge. You might be leaving returns on the table.
- If your current risk per trade is above f*/2 (Half Kelly): You are overbetting relative to your estimated edge. Consider reducing position size.
- If your current risk is between Quarter and Half Kelly: This is the reasonable zone for crypto trading given estimation uncertainty.
Kelly Applied to a Portfolio of Strategies
In DennTech Elite, where you can run multiple strategies simultaneously, Kelly can help allocate capital between them. The multi-asset Kelly formula allocates proportionally to each strategy's edge, giving more capital to higher-edge strategies and less to lower-edge ones. This is more sophisticated than equal allocation but requires reliable edge estimates for each strategy independently. See the position sizing guide for the simpler baseline approach before attempting multi-strategy Kelly allocation.
Frequently Asked Questions
- Should I use Kelly Criterion for my first bot deployment?
- No. For your first live bot, use the simple fixed-percentage approach (1% risk per trade) described in our position sizing guide. You do not have enough live trade data to estimate your edge reliably enough to apply Kelly responsibly. After 3–6 months of live trading with 100+ completed trades, revisit Kelly to assess whether your sizing is appropriately matched to your demonstrated edge.
- How do I handle a period where my win rate drops — does Kelly tell me to stop trading?
- If your win rate drops below the Kelly breakeven point (where the formula returns zero or a negative value), Kelly is mathematically telling you to not bet because you have no positive edge. In practice, this means pausing the strategy for review. Check whether market regime has changed, whether the strategy parameters need adjustment, and whether the recent run is within expected variance for your historical win rate. See the MDD guide for circuit-breaker logic that handles exactly this scenario.
- Is Kelly available as a position sizing mode in DennTech?
- DennTech's position sizing module includes Fixed Percentage, Fixed Dollar, and ATR-based modes. You can manually calculate a Kelly-derived percentage and enter it as your fixed risk percentage. The DennTech documentation covers all position sizing configuration options. See Elite pricing for multi-strategy portfolio management features.
Build your complete risk management framework by combining this Kelly guide with our position sizing guide, stop-loss guide, and MDD guide. Visit the pricing page to explore DennTech editions.