Kelly Criterion vs Fixed-Fractional Sizing: Which Wins

The Kelly Criterion, developed by John Kelly at Bell Labs in 1956, tells you the mathematically optimal fraction of capital to risk on any bet with a known edge. In backtested equity strategies with Sharpe ratios above 1.5, full Kelly sizing can produce terminal wealth multiples that dwarf fixed-fractional approaches — but drawdowns routinely exceed 70% along the way. That single data point explains why most professional traders never use it in its pure form.

The stakes here are not academic. Position sizing is the variable that determines whether a profitable edge survives real-world execution or destroys an account before the edge has time to pay out. Getting this decision wrong — overcommitting to Kelly’s volatility or undercommitting through overcautious fixed fractions — directly governs your compounded return over any meaningful time horizon.

This page delivers a direct, honest comparison: how each method is calculated, where each breaks down, which trader profiles benefit from each, and a ready-to-use prompt to interrogate your own system’s parameters against both frameworks.

How the Kelly Criterion Actually Works

The Kelly formula is f* = (bp - q) / b, where b is the net odds received, p is the probability of winning, and q is the probability of losing (1 - p). For a trading strategy with a 55% win rate and a 1:1 reward-to-risk ratio, Kelly prescribes betting 10% of capital per trade. That fraction scales aggressively upward as your edge or reward-to-risk ratio improves — a 60% win rate at 2:1 reward-to-risk produces a Kelly fraction exceeding 40%.

The mathematical guarantee underlying Kelly is that no other strategy produces higher long-run geometric growth given accurate inputs. That caveat — accurate inputs — is where the framework collapses in practice. Win rates and payoff ratios estimated from historical data carry estimation error. When the true edge is lower than estimated, full Kelly bets are effectively overbets, and overbetting a Kelly system leads to ruin faster than underbetting it.

Most professional applications use Half-Kelly or Quarter-Kelly — dividing the output by two or four — to hedge against input uncertainty. This sacrifice roughly halves the expected growth rate but cuts peak drawdown from catastrophic to tolerable.

• Full Kelly maximizes geometric growth but produces drawdowns above 50% routinely
• Half-Kelly cuts growth rate ~25% while reducing drawdown by ~50%
• Kelly requires accurate win rate and payoff ratio estimates — small errors compound destructively
• Kelly fraction changes after every trade as capital fluctuates, demanding continuous recalculation
• Kelly is theoretically optimal only over infinite time horizons; finite-period performance varies widely

How Fixed-Fractional Sizing Works

Fixed-fractional sizing is the simpler of the two frameworks. You risk a constant percentage of current equity on every trade — commonly 1% or 2%. A $100,000 account risking 1% per trade bets $1,000 on each setup, regardless of the strategy’s current estimated edge. As equity grows, position sizes scale up proportionally; as equity declines, they scale down, providing automatic drawdown protection.

The method’s strength is its robustness to estimation error. Because the fraction is fixed by discretion rather than derived from statistical inputs, it doesn’t blow up when your win rate estimate is wrong by five percentage points. The tradeoff is suboptimal growth: a fixed-fractional trader with a genuine 2:1 edge and 60% win rate leaves a significant multiple of terminal wealth on the table compared to a Kelly trader with perfect input data.

Fixed-fractional sizing is the dominant approach at retail and institutional desks precisely because it survives model error. The 1-2% rule that pervades trading literature isn’t mathematically derived — it’s an empirical compromise between growth aspiration and ruin avoidance.

• Risk per trade is set by discretion, not statistical derivation — survives input error
• Position size adjusts automatically with equity changes, compressing drawdowns
• No ongoing recalculation of win rates or payoff ratios required
• Systematically underperforms Kelly when edge estimates are accurate
• Flat fraction ignores edge variation across setups — a 60% win rate trade gets the same size as a 51% win rate trade

Where Kelly Breaks Down in Live Trading

Kelly’s theoretical optimality rests on three assumptions that live markets violate constantly: that bet outcomes are independent, that the distribution of payoffs is known, and that the bettor has unlimited time. Equity trades are not independent — regime shifts, correlation spikes, and sector rotations create clustered losses that Kelly’s formula treats as uncorrelated draws from a fixed distribution.

Estimation error compounds destructively. A strategy that appears to have a 58% win rate over 200 backtest trades may have a true win rate anywhere from 51% to 65% at the 95% confidence level. At the lower bound, Kelly prescribes a fraction that is mathematically an overbet — a position size that, applied consistently, moves the expected geometric growth rate below zero. The trader is not just leaving money on the table; they are actively destroying capital in expectation.

The practical consequence: Kelly is defensible for traders with extremely large sample sizes, stable statistical properties across market regimes, and disciplined use of fractional Kelly. It is not defensible for discretionary traders, traders in low-liquidity markets, or anyone operating on fewer than 500 historical trade observations.

You are a quantitative risk advisor.
My trading strategy has the following characteristics:
- Historical win rate: [X]%
- Average win / average loss ratio: [Y]
- Number of historical trades: [N]
- Market: [equity / futures / forex / crypto]
Calculate the full Kelly fraction, Half-Kelly fraction, and the 95% confidence interval around my edge estimate.
Then tell me whether fixed-fractional at 1% or 2% produces better risk-adjusted outcomes given my sample size uncertainty.
Flag any conditions under which Kelly would actively destroy my geometric returns.

Where Fixed-Fractional Falls Short

Fixed-fractional sizing’s core limitation is that it treats all setups as equivalent. A trader running a multi-strategy system that generates both high-conviction setups with 65% win rates and marginal setups with 52% win rates assigns identical position sizes to both. The result is systematic under-betting on high-edge opportunities and over-betting relative to Kelly on marginal ones — a portfolio construction error that erodes compounded returns over time.

The second limitation is that the fraction itself is arbitrary. The 2% rule has no mathematical derivation; it was popularized through trading literature as a heuristic that produced survivable drawdowns across a wide range of strategy types. Traders with high win rates and fat tails in their payoff distributions could defensibly risk 4-5% per trade without approaching ruin probability thresholds that Kelly would flag as dangerous.

In practice, sophisticated fixed-fractional systems partially address this by tiering position sizes — risking 0.5% on marginal setups, 1.5% on core setups, 2.5% on high-conviction setups. This hybrid approach captures some of Kelly’s edge-sensitivity while retaining the robustness that makes fixed-fractional operationally viable.

Which Method Fits Which Trader

Kelly Criterion, in fractional form, is best suited to systematic traders with large, stable trade histories, clearly quantifiable edges, and the psychological capacity to tolerate the higher variance that even Half-Kelly produces. Algorithmic traders running mean-reversion strategies in liquid futures markets with 1,000+ trade samples are the core use case. The math works when the inputs are trustworthy.

Fixed-fractional is the correct default for discretionary traders, traders in early-stage strategy development, and anyone whose strategy parameters shift meaningfully across market regimes. It is also the correct choice when the cost of a large drawdown exceeds the cost of suboptimal growth — which describes most retail traders and many prop desks operating under drawdown-based risk limits.

A practical hybrid for traders sitting between these profiles: use Kelly to compute the theoretically optimal fraction, then apply a 25-50% scaling factor as an uncertainty discount, and cap the result at your maximum tolerable single-trade loss. This preserves Kelly’s edge-sensitivity while hard-coding a floor against estimation error.

• Use Kelly: systematic strategies, 500+ trade samples, stable win rates, high Sharpe
• Use Fixed-Fractional: discretionary strategies, new systems, drawdown-constrained accounts
• Use Hybrid: size by Kelly output scaled to 25-50%, capped at max tolerable loss
• Never use Full Kelly on live capital without a statistically validated edge across multiple market regimes

Running the Numbers on Your Own Strategy

The comparison between these frameworks becomes concrete only when applied to your specific parameters. A strategy with a 55% win rate and 1.5:1 reward-to-risk produces a Kelly fraction of 18.3%. Half-Kelly puts you at 9.15%. A fixed-fractional trader at 2% is operating at roughly one-fifth of even the conservative Kelly prescription — leaving substantial compounded growth on the table if the edge estimate is accurate.

Conversely, if that same strategy’s true win rate is 51% due to overfitting, Full Kelly at the estimated fraction is an overbet by a factor of three. The 2% fixed-fractional trader survives this miscalibration comfortably. The Kelly trader faces accelerating drawdown precisely when confidence in the system should be lowest.

Screening your strategy’s statistical properties — win rate confidence intervals, payoff distribution skew, regime stability — before choosing a sizing framework is not optional. It is the analysis that determines whether Kelly is an accelerant or a grenade.

You are a position sizing analyst.
Given my strategy metrics below, model the terminal wealth and maximum drawdown outcomes for:
1. Full Kelly sizing
2. Half-Kelly sizing
3. Fixed-fractional at 1%, 2%, and 3%
Assume 250 trades per year over a 3-year horizon, starting capital $100,000.
Strategy metrics: win rate [X]%, average win [Y]x average loss, max consecutive losses observed [Z].
Present results as a comparison table. Highlight which method produces the best risk-adjusted outcome and explain why.