Skip to main content

Position Sizing for Systems

Updated Jun 24, 2026 at 8:22pm

Research Draft High 1,257 words

Position sizing for systems is the bet-sizing layer of a mechanical strategy: given an edge whose statistical properties you have measured (win rate, payoff, return distribution, volatility), it answers how much capital to commit to each signal. It is conceptually separate from the entry/exit logic — the same edge can be turned into a fortune or a wipeout purely by how it is sized. The core tension is that the bet size which maximizes long-run growth is dangerously close to the bet size which maximizes the chance of ruin: growth-optimal sizing produces brutal drawdowns, so practitioners deliberately bet less than the math says is optimal. A second tension is that all sizing formulas assume you know your edge's parameters, but in live trading those parameters are estimated with error and drift over time.

How it's calculated / formed

Several sizing models dominate systematic practice; they sit on a spectrum from fixed to growth-optimal.

Percent-risk (fixed-fractional). Van Tharp's most-cited model: risk a fixed fraction f of current equity per trade, where risk is defined by the distance to the stop. Position size = (equity × f) / (entry − stop) per share. Tharp catalogs four families — units per fixed amount, equal-units, percent-risk, and percent-volatility (Trade Your Way to Financial Freedom). Because f applies to current equity, position size shrinks automatically in drawdowns and grows in winning streaks — the defining property of fixed-fractional sizing.

Percent-volatility / ATR-based. Size so that a fixed fraction of equity is risked per unit of instrument volatility, typically using Average True Range. The Turtle system sized in "Units," each Unit set so that 1 ATR of adverse move equalled ~1% of account equity. Position size is inversely proportional to ATR, normalizing risk across instruments and across calm-versus-volatile regimes.

Volatility targeting. A portfolio-level cousin: scale gross exposure so realized portfolio volatility tracks a target (e.g., 10% annualized). When measured vol rises, leverage is cut; when it falls, leverage is added. Standard in managed-futures/CTA and risk-parity construction, where weights are set inversely proportional to each asset's volatility so each contributes equal risk (equal-risk-contribution).

Kelly criterion. Derived by John L. Kelly Jr. (Bell Labs, 1956), Kelly gives the fraction that maximizes the expected geometric growth rate of capital. For a binary bet, f = (bp − q)/b, where p is win probability, q = 1−p, and b is the win/loss payoff ratio. For continuous return streams the analogue is roughly f ≈ mean/variance.

Optimal f. Ralph Vince's generalization of Kelly to real trade streams with varying outcome sizes: the fixed fraction that maximizes geometric growth over an actual sequence of trade results, anchored to the largest historical loss. Closely related to Kelly but computed empirically from the return series rather than from a clean win/loss model.

How it's used in practice

In disciplined practice, position sizing is governed by a per-trade risk budget rather than a desired profit. The widely repeated convention is to risk 1–2% of equity per trade (frequently attributed to Tharp-style risk control and to O'Neil/CAN SLIM-era discipline), though the number is a heuristic, not a derived optimum. The workflow: define risk-per-trade as the distance from entry to a predefined stop, then back out share/contract count so that distance equals the dollar risk budget. This makes risk constant across trades regardless of price or volatility — a high-priced, low-volatility name and a cheap, jumpy one carry the same dollar risk.

For growth-optimal methods the universal practice is fractional Kelly — betting one-half, one-third, or one-quarter of the computed Kelly fraction. The reason is mathematical: half-Kelly retains roughly three-quarters of the maximum growth rate while cutting the variance (volatility) of the bet roughly in half — the trade-off documented by MacLean, Ziemba & Blazenko (Management Science, 1992) and echoed by Thorp. (Drawdowns are also greatly reduced, but the cleanly-quantified result is on growth and variance, not drawdown.) Most quantitative funds operate in the quarter-to-half-Kelly band, trading away some growth for survivability and robustness to parameter error.

At the portfolio level, single-trade sizing is overlaid with caps: total open risk (sum of per-trade risks), sector/correlation limits, and a portfolio volatility target. This is where sizing connects to portfolio-level risk: correlated positions sized independently can stack into one large hidden bet.

Adoption, debate & evidence

Volatility-scaling/targeting has the strongest empirical backing. Academic work (e.g., Moreira & Muir, Volatility-Managed Portfolios, Journal of Finance 2017; and managed-futures research from AQR and others) finds that scaling exposure inversely to volatility historically improved Sharpe ratios and reduced tail risk in many equity and futures series — partly because volatility is more persistent and predictable than returns. This is a genuine, replicated effect, though its magnitude is debated and not universal across assets.

Kelly/optimal-f are mathematically correct but practically hazardous and widely misapplied. Full Kelly's theoretical maximum drawdown can approach 100%; simulations routinely show 50–80% peak-to-trough declines along the path (QuantPedia, multiple Kelly tutorials). QuantPedia's framing — treat optimal f as a boundary not a target, and that "reality is usually worse than the backtest" — captures the consensus: overbetting past optimal f reduces growth while exploding risk, and since the edge is estimated, you are likely already overbetting. Bootstrap-resampled returns typically suggest roughly half the naive optimal fraction.

The honest base rate: there is no published evidence that any single sizing formula manufactures an edge. Tharp's stronger claim — that position sizing matters more than entries, and that "even a mediocre strategy can be profitable with proper sizing" — is influential but rests on his simulation game and trader anecdotes, not peer-reviewed study; it should be read as a discipline philosophy, not a proven theorem. What is well-supported is the converse: bad sizing destroys good edges.

Strengths & limitations

Strengths: sizing rules convert an abstract edge into survivable, repeatable execution; fixed-fractional and volatility-based methods are robust, simple, and self-correcting in drawdowns; volatility targeting has measured Sharpe/tail benefits.

Limitations and the #1 misuse: using full Kelly or raw optimal f on backtested parameters. Estimation error means the live edge is almost always smaller and noisier than the backtest, so the "optimal" fraction is an overbet that courts ruin. Other failure modes: ignoring correlation (independently sized positions becoming one macro bet), assuming returns are stationary when the edge is decaying, and stops that get gapped through so realized risk exceeds the budget. All formulas are only as good as the parameter estimates fed into them.

Sources