Fat tails in scalp P&L and what they do to risk estimates

[Amanhasnoname]

Standard risk calculations assume results follow a bell curve. Scalp results often do not. Most trades cluster near small gains and losses, but a few extreme outcomes are far larger than a normal distribution would predict.

These tails come from slippage during news, gaps, system failures, and rare events. A 5 sigma loss in a normal model should almost never happen. In practice, it shows up every few months.

Risk numbers based on the average and standard deviation understate the danger. Value-at-risk calculations calibrated to normal assumptions are too optimistic.

To account for fat tails, use empirical percentiles from the actual distribution, such as the 1st percentile of trade results. Stress test with historical worst cases and then add a margin.

For position sizing, base it on the worst plausible loss per trade, not the average loss. A system that normally loses 3 pips but occasionally loses 25 requires size set by the 25.

Examining the tail separately provides realistic planning figures.

What is the largest single trade loss relative to the average loss, and was it expected?