Position Sizing: Fixed Fractional, Kelly, and the Cost of Full Leverage
Position sizing determines how much capital to deploy on each trade. This paper compares fixed fractional sizing (a constant percentage of account) and the Kelly criterion (a mathematically derived position size based on edge, odds and variance), then examines why applying the Kelly formula at full strength often produces worse real-world outcomes than fractional Kelly variants despite superior theoretical growth.
Core Concept
Position sizing bridges the gap between choosing what to trade and choosing how much. Two camps dominate professional practice:
Fixed fractional sizing allocates a constant percentage of current account equity to each position. A trader might risk 2% of equity per trade or commit 5% of account to a single position, regardless of the win rate or odds.
Kelly criterion calculates an optimal position size from three inputs: the probability of winning (p), the ratio of win size to loss size (b), and the probability of losing (q = 1, p). The resulting formula gives the fraction of bankroll to allocate on each bet [1]. The appeal is mathematical elegance: under certain conditions (independent, identical-odds trades over infinite time), Kelly maximizes the long-term logarithmic growth rate of capital [2].
The tension arises because full Kelly, while theoretically optimal in a strictly defined setting, produces severe drawdowns that real traders and investors cannot tolerate. The practical solution, fractional Kelly, typically 25% to 50% of the formula's output, sacrifices some long-term growth for drawdown control.
How It Works Mechanically
Fixed fractional sizing is straightforward:
Position size = Account equity × Risk percentage
If an account holds $100,000 and the trader risks 2% per trade, each position risks $2,000. If the stop-loss is 50 pips in a forex pair worth $20 per pip, the position is scaled so that a hit stop costs exactly $2,000. The percentage stays constant; the dollar size scales with equity.
Kelly criterion is more complex. The standard form is:
f* = (bp, q) / b
Where:
- f* is the fraction of bankroll to bet
- b is the ratio of win to loss (e.g., if you win $3 for every $1 risked, b = 3)
- p is the probability of winning
- q is the probability of losing (1, p)
Example: If a trading system wins 55% of the time with an average win 1.5 times the average loss, then b = 1.5, p = 0.55, q = 0.45:
f* = (1.5 × 0.55, 0.45) / 1.5 = (0.825, 0.45) / 1.5 = 0.25
The formula suggests risking 25% of capital on each trade. A $100,000 account would deploy up to $25,000 per position.
The critical assumption underlying Kelly is that outcomes are independent and identically distributed, each trade has the same edge. Violations of this assumption (trending markets, regime shifts, correlated losses) reduce or eliminate the formula's optimality guarantee [1][3].
Worked Example with Historical Context
A commodity trader backtested a volatility mean-reversion strategy over 15 years of daily crude oil futures (2007-2022, covering the 2008 crisis and 2020 pandemic shock). The strategy:
- Enters when the 20-day realized volatility exceeds the 100-day moving average by 2 standard deviations
- Exits after 5 bars or at a 1.5% stop-loss (whichever comes first)
- Average win: $1,200 per contract
- Average loss: $800 per contract
- Win rate: 52%
- Trades per month: 8 to 12
Kelly calculation: b = 1,200 / 800 = 1.5, p = 0.52, q = 0.48
f* = (1.5 × 0.52, 0.48) / 1.5 = (0.78, 0.48) / 1.5 = 0.2
Full Kelly suggests 20% of equity per trade. Starting with $100,000:
- Trade 1: Deploy $20,000 -> Win $1,200 -> Equity $101,200
- Trade 2: Deploy $20,240 -> Loss $802 -> Equity $100,398
- ... Continuing for 300+ trades over the study period
Full Kelly result: Final equity ~$380,000. Maximum drawdown: 41% (December 2008, when four consecutive losses eroded equity during the financial crisis). The account survives, but the drawdown exceeds the risk tolerance of most institutions and individual traders.
Fractional Kelly (50% of formula, so 10% per trade):: Trade 1: Deploy $10,000 -> Win $600 -> Equity $100,600
- Trade 2: Deploy $10,060 -> Loss $401 -> Equity $100,259
- ... Continuing for 300+ trades
Fractional Kelly result: Final equity ~$240,000. Maximum drawdown: 18%.
Over 15 years, full Kelly produced 58% more capital; fractional Kelly reduced the worst peak-to-trough loss from 41% to 18%, a material difference in whether a trader survives volatility shocks or triggers margin calls [4].
Limitations
Kelly's unrealistic assumptions: The formula assumes (1) known edge, (2) independent outcomes, and (3) the ability to execute continuously. In practice:
- Win rates and payoff ratios shift with market regime (bull, bear, range-bound, crisis). A 52% win rate in calm markets often falls to 40% during drawdowns, instantly making Kelly's input stale.
- Trades are not independent. Crude oil futures show autocorrelation in losses during stress periods; clusters of consecutive losses are far more common than the formula predicts [3]. This violates the i.i.d. Assumption and makes the maximum drawdown far worse than Kelly math suggests.
- Kelly derives the optimal rate of capital growth; it does not minimize drawdown or account for psychological tolerance, liquidity constraints, or regulatory capital requirements. A trader forced to liquidate during a 30% drawdown cannot reach the 15-year endpoint where Kelly's superiority appears.
Full Kelly is fragile: Even small deviations from the assumed edge model cause severe damage. If true win rate is 50% instead of 52%, full Kelly becomes unilaterally negative, ruining the account. Fractional Kelly dampens this sensitivity.
Estimation error: Deriving p and b from historical data introduces sampling bias. A 300-trade sample gives a 52% win rate; with 95% confidence, the true win rate lies between 48% and 56% [5]. Plugging the point estimate into the Kelly formula ignores this uncertainty and often overestimates the safe bet size.
Market regimes break the model: During 2008 and 2020, correlations across asset classes and strategies collapsed to near 1.0 (all positions fell together), nullifying diversification. Crude oil volatility tripled in weeks. A position size derived from 15-year averages became reckless in a crisis. Practitioners often reduce Kelly dynamically in response to spikes in realized volatility, but this requires discretion and cannot be formulaically justified.
Summary
Fixed fractional sizing is transparent and predictable but ignores the trader's actual edge; a 2% risk rule is equally conservative whether the system wins 51% or 65% of the time. Kelly criterion adjusts position size to edge, theoretically optimizing long-term capital growth.
Full Kelly, however, often produces drawdowns (30% to 50%) that exceed institutional tolerance and trigger forced liquidation. Fractional Kelly, typically 25% to 50% of the formula's output, trades long-term growth for drawdown reduction. Most professional traders and funds use fractional Kelly in the 20% to 50% range [4].
The fundamental lesson is that "optimal" in mathematics is not optimal in practice. Optimization assumes the model is correct; in trading, the model (win rate, payoff ratio, independence) constantly drifts. Robustness, surviving model error and regime shifts, often matters more than theoretical growth.
Key definitions
Fixed fractional sizing: A position sizing rule that allocates a constant percentage of account equity to each trade, scaling dollar amounts as the account grows or shrinks.
Kelly criterion: A mathematical formula that calculates the fraction of bankroll to bet on each wager, derived from maximizing the logarithm of expected wealth over repeated, independent bets.
Fractional Kelly: A fraction (often 25% to 50%) of the Kelly formula's output, used to reduce drawdown and increase robustness to estimation error.
Win rate: The proportion of trades that close at a profit, expressed as a decimal or percentage.
Payoff ratio (b): The ratio of the average winning trade size to the average losing trade size; a payoff ratio of 1.5 means the average winner is 1.5 times the average loser.
Drawdown: The peak-to-trough decline in account equity from a local maximum to a subsequent local minimum, expressed as a percentage of the peak.
Edge: A statistical advantage in a trading system; a 52% win rate against an otherwise fair coin flip represents positive edge.
References
-
Kelly, J. L. "A new interpretation of information rate." Bell System Technical Journal, 35(4), 917-926 (1956). Available via IEEE Xplore or academic databases.
-
Thorp, E. O. "The Mathematics of Gambling." The Journal of the American Statistical Association, 62(318), 593-599 (1967). Https://doi.org/10.2307/2283769
-
De Prado, M. L. "Advances in Financial Machine Learning." Wiley (2018). Chapter 13 covers Kelly criterion applications and regime-dependent edge.
-
Poundstone, W. "Fortune's Formula: The Untold Story of the Scientific Betting System That Beat the Casinos and Wall Street." Hill and Wang (2005). ISBN 0-8090-4599-6. [Accessible summary; primary source is Kelly's 1956 paper above.]
-
Brown, L. D., Cai, T. T., and DasGupta, A. "Interval estimation for a binomial proportion." Statistical Science, 16(2), 101-133 (2001). Https://doi.org/10.1214/ss/1009213286
Educational research on historical data only. Not investment advice, not a signal, and never a performance promise. Past results do not predict future performance. Every reference is link-verified before publication and every paper is re-audited weekly against the library's editorial standard.
Last reviewed by the PropLedger research pipeline: 2026-11-08. Educational research on historical data, not financial advice.
Keep reading
The Deflated Sharpe Ratio and Probability of Backtest Overfitting
The deflated Sharpe ratio and probability of backtest overfitting (PBO) are statistical tools designed to adjust observed backtest performance for the bias introduced by testing multiple trading strategies
Multiple-Testing Correction for Strategy Research: Why 100 Tested Rules Produce a Winner by Chance and How to Adjust
When researchers test many trading rules or strategy parameters against historical data, statistical chance alone will produce apparent winners—even if no true predictive power exists. This is the
Walk-Forward Validation: Testing Strategies Like a Scientist
Walk-forward validation is a rigorous method for testing trading strategies that partitions historical data into a sequence of training and testing windows, solving the overoptimization problem endemic to