Halved-Kelly, explained without the textbook
Why we cap stakes at 5 units per pick, why "full Kelly" is too aggressive for almost everyone, and what the halved version is actually optimising for.
Kelly's formula gives you the bet size that maximises the long-run growth rate of a bankroll, given your true edge and the price you're offered. It is mathematically optimal under one hard assumption: that your edge estimate is exactly right.
Outside the textbook, your edge estimate is never exactly right. That mismatch matters more than people expect.
Why full Kelly is too much
Full Kelly sizes assume the model is perfectly calibrated. If your model's edge estimate is even a little optimistic, full Kelly stakes overshoot and the variance becomes brutal. Drawdowns of 50% or more are normal in the upper half of a Kelly distribution. Most users cannot tolerate that emotionally, and many bankrolls cannot survive it operationally.
What halved-Kelly buys you
- Roughly 75% of the long-run growth rate of full Kelly.
- Roughly 25% of the drawdown depth.
- A useful margin of safety against the model overstating its own edge.
Cutting Kelly in half is not a half-measure. It's a trade — sacrificing a small slice of theoretical growth in exchange for a much steadier ride, while staying robust to the gap between modeled and real edge.
Why we cap at 5 units
Even halved-Kelly will occasionally recommend stake sizes that feel too large for a single game's noise. The 5-unit per-pick cap and 25-unit per-day cap are guard rails — not because the math says they're necessary, but because the math assumes you'll wake up tomorrow and place the next bet calmly. The caps make that more likely to be true.