Fixed fractional: the sensible default
Risk a constant percentage of equity per trade, compute lots from the stop distance, and let compounding do the rest. It is unglamorous, it adapts automatically to account growth and decline, and it is nearly impossible to get catastrophically wrong. For the overwhelming majority of operators it is also the correct final answer.
Volatility parity
Instead of constant currency risk, target constant volatility contribution — smaller positions when the instrument is wild, larger when it is calm. On gold this smooths the equity curve noticeably, at the cost of underparticipating in the explosive moves that sometimes deliver the year's best trades. It is a genuine trade-off rather than an upgrade.
Kelly and why we do not use full Kelly
The Kelly criterion maximises long-run geometric growth given a known edge. The clause doing all the work is 'known'. Edge estimates from finite samples are noisy, and Kelly is extremely sensitive to overestimation — a 20% overstatement of edge produces sizing that can halve an account during a normal losing sequence.
Half-Kelly gives you roughly three quarters of the growth with a fraction of the variance. Full Kelly is a bet on the accuracy of your own estimate.
Equity-curve scaling
Some operators reduce size during drawdowns and increase it during expansions, effectively trading their own equity curve. It reduces depth of drawdown and slows recovery, because you are smallest exactly when the recovery begins. We consider it acceptable as a psychological aid and rarely optimal as a mathematical one — but a system you can actually hold through a drawdown beats an optimal one you abandon.
Why full Kelly is the wrong answer in practice
Kelly maximises the long-run growth rate given perfectly known edge and distribution, and neither is available to anyone trading a real market. Because your estimate of edge carries error, sizing at the theoretical optimum reliably overshoots, and the penalty for overshooting is severely asymmetric: the drawdowns compound faster than the growth.
Fractional Kelly as the practical compromise
- —A quarter to a half of the Kelly figure retains most of the growth with a fraction of the volatility.
- —Recompute edge estimates on out-of-sample data only.
- —Cap the result with an absolute ceiling that no estimate can breach.
- —When the estimate and the ceiling disagree, the ceiling wins.
Equity-curve scaling and the trap inside it
Reducing size after losses and increasing it after wins is intuitive and, on many return series, actively harmful. If outcomes are close to independent, the rule mostly guarantees you are small during the recovery that follows a drawdown and large going into the next one. It only helps if losing streaks genuinely predict further losses — which requires evidence, not intuition.
Test the assumption before encoding it
Our default is a fixed fractional model with a hard ceiling, because it is the version that stays correct when the market changes character and nobody is watching.
Everything discussed here is applied on a public, third-party verified account — updated continuously, losing weeks included.