Dynamic Position Sizing: Fractional Kelly and Volatility Targeting
Dynamic position sizing covers fractional Kelly, volatility targeting, and concrete formulas for scaling positions as edge and volatility shift over time.
Les outils interactifs peuvent ne pas fonctionner dans la vue traduite.
Dynamic Position Sizing: Fractional Kelly and Volatility Targeting
Fixed fractional sizing ("risk 1% per trade") is a starting point, not a finished rule. It ignores that edge and volatility move. Dynamic sizing scales position size to current conditions — larger when edge is high and vol is low, smaller when the reverse holds.
The Kelly formula
Full Kelly maximizes long-run growth:
f* = (p * b - q) / b
where p is win probability, q = 1 - p, and b is the win/loss ratio. For a strategy with 55% win rate and 1.5 payoff ratio: f* = (0.55 × 1.5 − 0.45) / 1.5 = 0.25 — 25% of capital per bet.
Full Kelly is mathematically optimal for growth and psychologically untradeable: it produces drawdowns of 50%+ with regularity. A 50% drawdown requires a 100% gain to recover; most traders quit before that.
Fractional Kelly
Run a fraction of full Kelly — typically ¼ to ½. The math: half-Kelly captures 75% of the growth rate at 50% of the volatility. Quarter-Kelly captures about 44% of growth at 25% of volatility. The growth sacrifice is small; the drawdown reduction is large.
For the example above, half-Kelly is 12.5% per trade, quarter-Kelly is 6.25%. Most professional strategies run 0.25–0.5 Kelly.
Volatility targeting
Combine Kelly with a volatility target. Scale each position so its contribution to portfolio volatility is constant:
size = target_vol_per_position / asset_vol
If target risk per position is 1% daily and an asset's 20-day realized vol is 2% daily, size it to 50% of the notional you would use at 1% vol. This prevents the high-vol position from dominating the portfolio.
Use EWMA or GARCH vol estimates; a 20-day exponential weighted average tracks regime shifts faster than a simple 60-day window.
Scaling by edge confidence
Edge is not constant. Scale size by a confidence multiplier:
- Backtested, walk-forward validated edge: 1.0× base size
- Live, but edge confirmed over 100+ trades: 1.0×
- New strategy, under 30 live trades: 0.25×
- Strategy in drawdown beyond historical norm: 0.5× until recovered
The anti-martingale rule
Increase size after wins (within Kelly bounds), decrease after losses. This is the opposite of the gambler's instinct to "win it back." Concretely: after a 10% equity drawdown, cut all position sizes by 30% until equity recovers to within 5% of peak. After a 20% drawdown, cut to 50% indefinitely.
Practical implementation
Daily, compute: each strategy's recent vol, its Kelly fraction given recent win rate and payoff, and the volatility-targeted size. Cap the combined portfolio risk at 6% daily. Reallocate weekly, not per trade — per-trade resizing over-trades and increases costs.
Dynamic sizing is the difference between a strategy that compounds and the same strategy that grinds sideways. The edge is in the sizing as much as in the entry.
Live Chart
Open full chart →Related market data, powered by TradingView.