Page 2 of 2

Re: Why Conviction-Based Sizing Is Dangerous

Posted: Fri Oct 23, 2026 10:00 pm
by PTScalper
The Math in Practice

The math seamlessly handles any strategy profile, including high-frequency scalping setups with inverted risk-to-reward ratios. For example, if you risk 3 units to make 1 ($b = 0.33$), and your price action setups have an 85% win rate ($p = 0.85$, $q = 0.15$), the formula dictates an optimal risk fraction of roughly 40% of your available capital ($0.85 - \frac{0.15}{0.33}$).

When allocating a specific pool of risk capital for a dedicated 2- to 6-week trading series, Kelly defines your absolute mathematical ceiling. Risking anything more than the Kelly fraction ($f^*$) doesn't just increase volatility—it mathematically guarantees a lower long-term growth rate and accelerates the path to ruin.

Re: Why Conviction-Based Sizing Is Dangerous

Posted: Fri Oct 23, 2026 10:00 pm
by PTScalper
The Danger of "Full Kelly"

In a casino, variables like $p$ and $b$ are fixed constants. In spot forex, they are dynamic estimates. If market conditions shift or you hit a string of bad execution, a "Full Kelly" sizing model will quickly trigger devastating drawdowns.Because of this estimation risk, quantitative traders almost universally operate on Fractional Kelly (typically "Half-Kelly" or even a quarter). Cutting the calculated Kelly fraction in half reduces your drawdown severity exponentially while still capturing about 75% of the optimal compounding curve, acting as a massive mathematical buffer against a changing market regime.

Re: Why Conviction-Based Sizing Is Dangerous

Posted: Thu Nov 05, 2026 10:03 am
by Fairman
Conviction really does feel like information when you're in it, and that's the trap. The trades I felt surest about were usually the ones where everything lined up on the chart, which is also what everyone else was looking at. My best "I can't lose" setups ended up as a few of my worst stops.

What changed my sizing was looking at my own numbers. Out of about 300 trades, I'd marked 60 as high conviction. Their average came out at +0.12R, against +0.10R for the rest. Basically no difference. But I'd been risking 2 percent on those and 1 percent on the others, so on a 10,000 dollar account a single bad cluster of three high conviction losers cost 600 dollars instead of 300.

Now the size comes from a checklist with fixed answers, and the stop distance sets the lots. If volatility is high I go smaller, not bigger. It's dull, and I don't miss the adrenaline.

The part I'd add to the points about confirmation bias is that conviction tends to rise right after a winning streak, which is exactly when your sample says the next trade is no better than any other. Have you ever tested your own high-rated trades? And did it surprise you?