Position Sizing with the Kelly Criterion

The Silent Killer: Variance

I have seen incredibly smart people build highly accurate predictive models, only to completely drain their accounts in three weeks. Why? Because the sharpest sports data analytics platform in the world is entirely useless if poor money management bankrupts you during an inevitable statistical downswing.

In sports modeling, realized outcomes vary around estimated probabilities, so losing runs remain possible even for a process that later performs well over a larger sample. The Kelly Criterion is one well-known mathematical framework for relating an assumed probability and price to a growth-oriented allocation; it is not a guarantee and can be dangerous when those inputs are wrong.

Decoding the Kelly Formula

The Kelly Criterion estimates a growth-oriented allocation from your assumed probability and the available price. Its result is only as reliable as those inputs and the assumptions behind the formula.

f = (bp - q) / b

• f = Fraction of the reference unit base to allocate
• b = Decimal odds - 1
• p = Probability of winning
• q = Probability of losing (1 - p)

If the Kelly formula returns 3%, that is the full-Kelly allocation under the supplied assumptions. It does not eliminate drawdowns or the risk of ruin, especially when the estimated probability is wrong.

The Danger of Full Kelly

The formula has strict assumptions. It depends on an accurate win probability and price, while sports models work with estimates that carry uncertainty.

If your model overestimates your edge, applying the "Full Kelly" percentage can be dangerously aggressive and expose you to devastating drawdowns. Because of this, professional syndicates almost exclusively use Fractional Kelly.

By utilizing a Quarter-Kelly (dividing the recommended size by 4) or Half-Kelly approach, you dramatically smooth out the volatility curve. Your compound growth is slightly slower, but you mitigate the risk of overestimating your edge. EdgeSlate provides integrated tools to help you identify model-vs-market opportunities while maintaining disciplined, mathematical growth of your reference unit base.