Course 29: Position Sizing & the Kelly Criterion

Advanced Track • Estimated reading time: 28 minutes

Of all the variables a trader controls, position sizing is simultaneously the most impactful and the most neglected. Entry timing, indicator selection, and strategy design receive exhaustive attention; the question of how much capital to deploy on any given trade is frequently answered with a round number or a gut feeling. This is a critical error. A mathematically superior sizing methodology can take a mediocre strategy and produce excellent long-term results; conversely, reckless sizing can take a profitable strategy and drive it to ruin through a perfectly normal drawdown sequence. The Kelly Criterion — and its practical derivative the half-Kelly — provides the rigorous, probabilistic foundation for optimal sizing that professional traders and institutional risk desks rely upon. This course builds that foundation from first principles.

The Mathematics of Ruin: Why Sizing Dominates Strategy

Before introducing the Kelly formula, consider a demonstration that makes the primacy of sizing concrete. A strategy with a 55% win rate and a 1:1 reward-to-risk ratio — a genuine, meaningful edge — produces vastly different long-run outcomes depending solely on sizing. Sized at 25% of account per trade, the probability of experiencing an 80% drawdown before tripling the account exceeds 90% due to natural variance. The same strategy sized at 2% of account carries a near-zero probability of an 80% drawdown over a 1,000-trade sample. The strategy is identical. The outcomes are categorically different.

The theoretical basis for this observation was formalised by John L. Kelly Jr. at Bell Labs in 1956 in his landmark paper "A New Interpretation of Information Rate." Kelly's insight, drawing on Claude Shannon's information theory, was that expected geometric (compound) growth rate — not expected arithmetic return — is the correct objective function for a trader managing a compounding account. Arithmetic return is the simple average of individual trade returns; geometric return is the actual compound growth rate, which is permanently impaired by large losses. A 50% drawdown requires a 100% subsequent gain to recover to the prior high. A 25% drawdown requires only a 33% gain. Position sizing that constrains drawdown magnitude directly accelerates the pace of compounding recovery — and that acceleration, compounded over hundreds of trades, dwarfs any improvement achievable through better entry timing alone.

Use the free risk management calculator to anchor these concepts to your own account size before proceeding. The foundational rules covered in Risk Management 101 apply throughout this advanced material.

The Kelly Criterion: Derivation and Formula

Kelly's formula addresses a specific question: given a sequence of independent bets with a known edge, what fraction of capital maximises the long-run compound growth rate? His answer is the celebrated Kelly fraction:

Applying the formula: win rate p = 0.55, reward-to-risk b = 1.5, loss probability q = 0.45. The Kelly fraction = (1.5 × 0.55 − 0.45) ÷ 1.5 = 0.375 ÷ 1.5 = 25% of capital per trade. This is the mathematically optimal fraction for maximising the long-run geometric growth rate given those exact inputs.

However, a critical limitation is the formula's sensitivity to estimation error. Your "true" win rate and reward-to-risk ratio are unknowns estimated from a finite sample of historical trades. If you overestimate either p or b — entirely plausible given the sample sizes available to retail traders — the formula returns an oversized fraction that operates in the overbetting regime, where geometric growth actually decreases. This asymmetry argues decisively for the half-Kelly as the professional operating standard. Use the free position size calculator to apply Kelly-derived fractions to your account without manual calculation errors.

Full Kelly vs Half-Kelly vs Fixed Fractional

Three sizing methodologies dominate professional practice, each occupying a different position on the growth-versus-safety spectrum.

Full Kelly maximises the median geometric growth rate but produces enormous outcome volatility. Even with genuine edge, drawdowns of 30 to 50% are common under full Kelly because the formula assumes infinite trials and zero estimation error — conditions that do not exist in finite trading careers. Most professional risk managers consider full Kelly psychologically and practically unacceptable.

Half-Kelly — deploying 50% of the Kelly-optimal fraction — produces approximately 75% of the geometric growth rate of full Kelly while reducing drawdown volatility by more than 50%. This asymmetry is remarkable: give up 25% of theoretical growth to halve your drawdown exposure. Most proprietary trading firms and systematic funds running individual strategy sleeves operate Kelly fractions between 0.25 and 0.5.

Fixed fractional sizing — risking a predetermined constant percentage (commonly 1 to 2%) on every trade — is the most widely used method among retail traders and the approach recommended in Risk Management 101. It does not maximise geometric growth but provides predictable, psychologically manageable drawdown profiles and requires no edge estimation. It is the correct starting point for traders who have not yet accumulated a statistically validated track record large enough to reliably estimate Kelly inputs.

Position Sizing Across Win Rates and Reward Ratios

The Kelly formula produces dramatically different optimal fractions depending on strategy characteristics. A scalping strategy with 65% win rate and 0.8:1 reward-to-risk yields a very different Kelly fraction than a swing trade with 40% win rate and 3:1. Both can be profitable; both demand different sizing. The matrix below shows half-Kelly fractions across the practical range of strategy parameters.

Two cells in the matrix deserve special attention. A 50% win rate at exactly 1:1 reward-to-risk produces a Kelly fraction of zero — no edge exists and the formula correctly advises no trade. A 60% win rate at 4:1 reward-to-risk produces a half-Kelly of 37.5% — extraordinarily large, reflecting a genuine statistical advantage that few real strategies possess. If your calculated Kelly fraction exceeds 20%, verify your sample size is statistically credible (minimum 300 trades) before trusting the estimate.

Practical Implementation: Inputs, Recalibration, and Ruin Probability

Implementing Kelly sizing requires three inputs: your historical win rate, your average reward-to-risk ratio, and your current account equity. From these you derive the Kelly fraction, halve it, then express it as a dollar risk amount using the position size calculator. The output is the maximum dollar risk — not position size — to accept on any single trade. Convert to position size by dividing by your stop-loss distance in price terms.

Critically, Kelly inputs must be recalibrated regularly. Strategy performance degrades as market conditions shift. A trend-following strategy that produced 58% win rate in a trending market may produce 44% in a ranging one, driving the Kelly fraction close to zero and signalling that you should reduce size dramatically or stand aside. Running rolling 100-trade performance reviews and recalculating Kelly inputs quarterly is professional discipline, not optional refinement. The journaling and performance metrics required to do this are covered in Course 39: Trading Journals & Performance Metrics.

Common Sizing Mistakes and How to Avoid Them

Emotional scaling is the most common error: increasing position size after wins because confidence is high, or increasing after losses in an attempt to recover quickly. Both behaviours violate the statistical independence that Kelly analysis assumes and consistently move traders into the overbetting regime. The correct response to a losing streak is to reduce position size — which Kelly does automatically by scaling to current equity — and investigate whether strategy edge has degraded.

Correlation blindness is the second major error: sizing multiple simultaneous positions as if they are independent when they are correlated. Three altcoin positions during a broad crypto selloff do not represent three independent risks — they represent one correlated risk multiplied across three instruments. Accounting for cross-asset correlation is covered in Course 35: Correlation & Portfolio Risk.

Insufficient sample size is the third common error: applying Kelly to fewer than 100 trades. At 20 to 50 trades, estimation noise in your win rate and reward ratio is so large that the Kelly formula is essentially useless — any output should be disregarded. Track every trade with discipline from day one using a structured journal and the free crypto risk management tools on this site, building toward a sample large enough to trust. Read the blog for practical journaling templates and examples.

Key Takeaways

• Position sizing dominates all other trading variables in determining long-run outcomes — more so than entry timing or indicator choice.
• The Kelly Criterion calculates the exact fraction of capital that maximises geometric (compound) growth rate for a strategy with a known edge.
• Full Kelly is too aggressive in practice due to estimation error. Use half-Kelly as the professional operating standard.
• Before accumulating 300+ trades, use conservative fixed fractional sizing at 1 to 2% per trade to protect your account.
• Recalibrate Kelly inputs at least quarterly — win rates and reward ratios drift as market conditions change.
• Multiple correlated positions must be sized with portfolio-level risk in mind, not as independent bets.
• Use the free position size calculator to apply these frameworks with precision and protect your trading account from ruin.