Position Sizing: Fractional Kelly and Why Full Kelly Ruins People

What Kelly actually says

The Kelly criterion gives the fraction of capital that maximises expected logarithmic growth, given an edge and odds. In the simple even-money case it reduces to something like twice the win probability minus one. In continuous settings it is roughly expected excess return divided by variance. Applied repeatedly, it maximises the growth rate that matters for compounding.

The practical problem is that edge and variance are estimated from noisy, non-stationary data. Small errors in win probability or payoff produce large errors in the recommended fraction. Full Kelly is therefore rarely used by people who are still in the game after a bad run. Half-Kelly, quarter-Kelly, or volatility targeting (scaling size so each position contributes roughly constant risk) are the standard practical responses. They give up some theoretical growth in exchange for a much lower chance of a catastrophic drawdown and more robustness to being wrong about the inputs.

1–2 per cent is an example, not a commandment

Many active practitioners risk 1–2 per cent of equity per idea, measured to the invalidation. That keeps the recovery maths tractable: a 50 per cent loss on a position requires a 100 per cent gain to get back to even, but if the position was only a small slice of the portfolio the portfolio-level hole is manageable. The number is a starting discipline for people who have evidence of an edge. It is not a universal default for a passive core, and it is not safe if several positions are really the same bet in different wrappers.

Distinguish capital allocated from planned loss. A large position with a tight invalidation can risk the same dollars as a small position with a wide one. Gap risk means the planned loss is an estimate, not a ceiling.

The two sizing errors

Too large relative to true edge produces fat-tailed losses that then trigger the disposition effect the sizing was meant to prevent. Too small relative to a genuine edge produces disappointing compounding and, eventually, frustration that leads to style drift and a sudden increase in size at the wrong time. The feedback loop between size, outcome and subsequent behaviour is tight. Most people underestimate it.

What to do with this

For your next active idea, write the invalidation first, then the dollar loss if it is hit, then the position size that keeps that dollar loss inside your pre-set risk budget. If the resulting size feels “too small” for your conviction, the conviction is not a sizing input. The budget is.