A profitable signal with bad position sizing still loses money. A mediocre signal with disciplined sizing can survive long enough to compound. Sizing isn't a detail bolted onto a strategy — it's half the strategy.
The core tradeoff
Bet too small on a genuine edge and you leave most of the return on the table. Bet too large and a normal losing streak — which will happen, even with a real edge — wipes out capital before the edge has room to play out. Position sizing is the answer to one specific question: given a real edge, how much of your capital should a single bet risk?
What the Kelly criterion actually says
The Kelly criterion computes the position size that maximizes long-run compound growth, given a win probability and a payoff ratio. The formula itself (edge divided by odds) matters less than the intuition: bet size should scale with how confident you actually are in the edge, and shrink toward zero as that edge shrinks toward zero. A coin-flip edge deserves a tiny bet. A strongly confirmed edge deserves a larger one — but never all of it.
Why full Kelly is a trap in practice
Full Kelly sizing is famous for being right on average and brutal in the short run — it produces enormous, realistic drawdowns even when every input is exactly correct, because it assumes your edge estimate is perfect. It never is. Real-world edge estimates are noisy, and a full-Kelly bet on a slightly-overestimated edge oversizes the position in exactly the wrong direction. That's why almost nobody who understands the math trades full Kelly.
Fractional Kelly: the practical version
Most systematic traders who use Kelly-style sizing run a fraction of it — commonly a quarter to a half. Half-Kelly captures roughly three-quarters of full Kelly's long-run growth rate while cutting the worst-case drawdown dramatically. That asymmetry — give up a little expected growth, remove a lot of tail risk — is the entire argument for sizing below the theoretical optimum.
What this means for a real strategy
- Size scales with edge, not conviction alone. "I have a good feeling about this one" isn't an edge estimate; a backtested win rate and payoff ratio is.
- Uncertainty in the edge estimate should shrink the bet. A strategy validated on 40 trades deserves smaller sizing than one validated on 4,000, even with an identical headline win rate.
- Correlated positions aren't independent bets. Sizing five correlated crypto longs as if they were five independent edges understates the real portfolio risk.
- Re-check the inputs periodically. An edge estimate computed once and never revisited is a stale input driving a live decision.
Sizing discipline is the unglamorous half of every strategy Atradian backtests — the Strategy Lab shows position-sizing assumptions alongside win rate and drawdown for exactly this reason: a strategy's return figure is meaningless without knowing how much capital was actually put at risk to get it.