The same signal can produce a safe strategy or a dangerous one
Position sizing converts a trading decision into portfolio exposure. A signal may say “long,” but sizing determines whether that means one share, one futures contract, 10% of capital, or enough leverage to threaten the account.
This makes sizing part of the strategy, not a presentation choice applied after the backtest. It changes profit and loss (P&L), drawdown, margin use, turnover, and sometimes the sequence of trades if limits bind.
Risk limits serve a different but related role. A sizing rule proposes exposure; limits constrain it. A volatility target might propose a large position during a calm period, while a gross-exposure cap, contract limit, or liquidity rule prevents the strategy from taking the full amount.
Four common sizing approaches
| Approach | Basic idea | Main strength | Main weakness |
|---|---|---|---|
| Fixed quantity | Hold the same units each trade | Transparent | Risk changes with price and volatility |
| Fixed capital fraction | Allocate a set percentage of equity | Scales with account value | Equal capital is not equal risk |
| Stop-based risk | Choose size so the planned stop implies a set loss | Connects size to trade structure | The stop may gap or fill worse |
| Volatility scaling | Reduce size when measured volatility rises | Makes ordinary risk more comparable | Volatility is backward-looking and can jump |
No method removes uncertainty. It decides which uncertainty the strategy is trying to stabilize.
A simple sizing example
Illustrative example. Assume an account has 100,000 in equity and a rule budgets 0.5% of equity to the distance between entry and a protective stop. The loss budget is therefore 500.
If entry is 100 and the stop is 98, the planned distance is 2 per unit. Ignoring fees, slippage, and contract multipliers for the moment, the proposed size is 250 units. If the stop distance is 5, the proposed size falls to 100 units.
That arithmetic does not cap realized loss at 500. Price can gap through the stop, liquidity can be unavailable, and slippage and fees can add cost. The calculation is a sizing convention based on a modeled exit distance, not an insurance contract.
Notional exposure, margin, and loss budget are different
The same illustrative position contains several numbers that answer different questions:
| Quantity | Value in the example | Meaning |
|---|---|---|
| Account equity | 100,000 | Capital base used by the sizing policy |
| Position notional | 25,000 | 250 units multiplied by the 100 entry price |
| Planned stop loss | 500 | 250 units multiplied by the 2-point stop distance |
| Cash or margin required | Instrument-dependent | Capital the broker or clearing arrangement requires to hold the position |
A 500 planned loss is not a 500 position, and a 25,000 notional position does not necessarily require 25,000 in cash. Futures multipliers, leverage, options exposure, currency conversion, and margin rules can make these quantities very different. Research should name the quantity being capped instead of using “risk,” “size,” and “exposure” interchangeably.
Volatility scaling changes exposure after the fact
A volatility-targeting rule often chooses exposure roughly in inverse proportion to a recent volatility estimate. If measured volatility doubles, target exposure is cut in half.
This can make different periods or strategies more comparable, but it creates important timing choices:
- Which return frequency and lookback estimate volatility?
- Is the estimate known before the rebalance?
- How quickly may exposure change?
- What happens when volatility is extremely low or missing?
- Are leverage and turnover capped?
A sudden volatility jump can arrive before the estimator reacts. Rapid deleveraging can then occur in the same stressed market for many participants, increasing costs and correlation.
Risk limits form a hierarchy
A robust sizing process typically applies several constraints:
- Instrument limit — maximum units, notional, or percentage in one market.
- Strategy limit — maximum gross, net, or volatility contribution.
- Portfolio limit — aggregate exposure, leverage, concentration, and shared-capital constraints.
- Liquidity limit — participation or capacity relative to tradable volume.
- Loss limit — daily, drawdown, or tail-risk boundary with a declared response.
The final position is the sizing proposal after all applicable limits. Research should record which limit bound the position; otherwise performance changes caused by constraints may be misattributed to the signal.
Drawdown limits need an action rule
“Stop at a 10% drawdown” is incomplete. Does the strategy close immediately, stop opening new positions, reduce risk gradually, or require a later review? Is the threshold based on live marked equity, realized P&L, or end-of-day equity? Can the portfolio recover while disabled?
Choosing the threshold after observing the historical maximum creates a backtest-shaped limit. Test a reasonable policy family and report the trade-off: lower thresholds may cap some losses but can also lock in ordinary drawdowns and miss recovery.
Kelly sizing is a boundary, not a default
The Kelly criterion maximizes expected long-run logarithmic wealth under known probabilities and payoffs. In trading, those inputs are estimated and unstable. Small errors in edge can produce large errors in proposed size, and full-Kelly paths can have severe drawdowns even when the model is correct.
For that reason, Kelly is best understood as a theoretical upper boundary and a demonstration that edge and size are inseparable. Conservative fractions, hard leverage limits, and stress tests are more realistic than treating an estimated Kelly fraction as permission to bet aggressively.
Evaluate the size rule separately from the signal
Compare at least:
- fixed capital or fixed-unit exposure as a transparent baseline;
- the proposed adaptive sizing rule;
- the same rule under an exposure cap;
- a delayed or smoothed response to changing volatility;
- higher costs during large rebalances.
Report returns with exposure, turnover, margin use, risk contribution, and drawdown. A higher Sharpe achieved by extreme low-volatility leverage may not be a better strategy.
Common sizing mistakes
- Reporting a signal result without the position rule that produced it.
- Treating equal capital as equal risk.
- Assuming a stop guarantees the budgeted loss.
- Letting an inverse-volatility formula explode when estimated volatility is near zero.
- Ignoring contract multipliers, currency conversion, or margin.
- Choosing a drawdown threshold from the same loss path it is meant to control.
- Applying Kelly sizing to a noisy in-sample edge estimate.
- Showing return without the exposure and leverage needed to obtain it.
Further reading
- Kelly, “A New Interpretation of Information Rate” (1956) — Derives the log-growth criterion under known probabilities and payoffs, making clear why sizing is only as trustworthy as those assumptions.
- Moreira and Muir, “Volatility-Managed Portfolios” (2017) — Supplies a concrete empirical example of reducing exposure when estimated volatility rises. It is evidence about one volatility-management construction, not a universal endorsement of volatility targeting.
- Rockafellar and Uryasev, “Optimization of Conditional Value-at-Risk” (2000) — Shows how expected losses beyond a selected tail threshold can enter allocation objectives and constraints.