Sizing leverage in volatility units, not in raw multiples.

[01]The volatility-adjusted leverage framework

Position risk depends on two factors: how leveraged the position is (the multiple of equity you are exposed to) and how volatile the underlying instrument is (the typical magnitude of price moves). The product of these factors is what matters; either factor alone tells you nothing useful.

Operationally, define:

• L = effective leverage = position value / equity
• σ_annual = annualised volatility of the underlying instrument
• σ_daily = daily volatility = σ_annual / √252
• R_position = effective position-level annualised volatility = L · σ_annual

R_position is the volatility of your equity, not the volatility of the underlying. Fund-management practitioners typically target portfolio-level R values of 10–15 % annualised; aggressive hedge funds run 20–30 % annualised; long-only retail accounts implicitly run whatever R the underlying produces (since L = 1).

[02]Reading historical volatility

Annualised volatility for liquid US equities can be estimated from the daily log returns over a trailing period (90 days for short-term reading; 252 days for the standard 1-year reading). Public references typically quote the trailing 30-day annualised volatility for “current” readings. Approximate annualised volatility for major asset classes as of early 2026:

• Large-cap defensive equity (utilities, consumer staples): 12–18 %
• Broad large-cap equity (S&P 500 index): 14–20 %
• Mid-cap equity: 18–26 %
• Small-cap equity: 22–32 %
• Tech-sector concentrated: 25–40 %
• Biotech / single-name speculative: 40–80 %
• Bitcoin and major crypto: 50–90 %
• Leveraged ETFs (3× products): roughly 3× the underlying index volatility

Use the trailing 60–90 day reading for sizing decisions; the 30-day reading is too noisy and can be substantially elevated or depressed by recent specific events.

[03]Worked example: 3× long the S&P 500

Position: 3× leverage on the S&P 500 (achieved either via 3× leveraged ETF, futures, or a margin position). S&P 500 annualised volatility: ~16 %. R_position = 48 % annualised.

Translation: the standard deviation of one-year return on this position is approximately 48 %. A two-sigma adverse year produces approximately a 96 % loss — account effectively zero. A one-sigma adverse year produces approximately 48 % loss — survivable but devastating.

Daily volatility: 48 % / √252 = 3.0 %. A 3× long S&P position has daily standard deviation of 3 %; a 2-sigma adverse day produces 6 % loss to equity; a 3-sigma adverse day produces 9 % loss. With 25 % maintenance margin, the position survives 33 % adverse cumulative move — a few standard-deviation days clustered together.

The 2008 financial crisis and the 2020 COVID crash both produced 5+ sigma daily moves in concentrated bursts. 3× long S&P positions held through either event were typically liquidated.

[04]The Kelly-criterion bound on leverage

The Kelly criterion (J.L. Kelly, 1956) sets the optimal fraction of bankroll to wager on a positive-expected-value bet. Adapted for continuous-time investment with normally-distributed returns, the Kelly-optimal leverage L* is:

L* = (μ − r_f) / σ²

Where μ is the asset’s expected return, r_f is the risk-free rate, and σ is the asset’s volatility (all annualised). For the S&P 500 with historical μ ≈ 10 %, r_f ≈ 4 %, σ ≈ 16 %, the Kelly-optimal leverage is approximately 2.3×.

The Kelly-optimal leverage maximises long-run compound growth. It does not minimise drawdown; the drawdown profile of a Kelly-leveraged portfolio is brutal. Practitioners typically size at “half-Kelly” (in this case ~1.15×) to retain most of the long-run growth benefit at substantially reduced drawdown. Full Kelly is for traders who can tolerate 50 % drawdowns without changing behaviour; half-Kelly is more humanly sustainable.

The framework changes when expected return is unknown or contested. For most retail traders, μ on a specific position is a guess, not a measurement. Kelly-style sizing assumes you know μ, which most traders don’t. The pragmatic application: use Kelly to set an upper bound on plausible leverage, accept that your operational leverage should be well below that bound to allow for parameter uncertainty, and treat any leverage above 2–3× on broad indices as aggressive even before instrument-specific analysis.

[05]The fat-tail correction

The volatility framework above assumes normally-distributed returns. Real return distributions, especially for equity, are leptokurtotic: extreme moves occur much more frequently than the normal distribution predicts. The 1987 crash was a 22-sigma event under normal-distribution assumptions, but it happened. The COVID crash of March 2020 produced multiple 5+ sigma days clustered together, again a vanishingly improbable sequence under normality but observed reality.

The implication for leverage sizing: the 2-sigma and 3-sigma calculations above understate the true risk by a substantial factor. A position designed to survive a 3-sigma adverse move under normal-distribution assumptions may not survive the empirically-observed worst-case scenarios. The correction: size leverage as if 5-sigma events occur once every few years (which empirically they do) rather than once in tens of thousands of years (which the normal distribution claims).

Practical translation: cut your “sigma-survival” target by roughly half from what the normal distribution suggests. If normality says a 3-sigma move occurs once in 730 days, treat it operationally as if it occurs once every 100–200 days. Your position should survive that. The Taleb argument on Black Swans is essentially this correction at greater length.

[06]The correlation-amplification trap

For traders running multiple leveraged positions concurrently, the position-level volatility framework above understates portfolio-level risk if the positions are correlated. Two 2×-leveraged S&P 500 long positions held simultaneously do not have the volatility of one position; they have the volatility of two positions whose returns are perfectly correlated, which is the same as a single 4×-leveraged position.

The implication: total portfolio leverage matters, not individual position leverage. A “diversified” portfolio of 5 leveraged positions, all in the same sector, all correlated above 0.85, is a single 5×-equivalent position dressed up to look like diversification. The portfolio-margin framework discussed on the portfolio-margin page formalises this insight at the broker level; retail margin accounts apply position-level margining without correlation offsets, which means the broker is effectively forcing the trader to over-collateralise correlated positions.