tuning stops from MAE can quietly curve-fit you

[Shadow Trader]

The standard advice goes like this. Plot MAE for winning trades. If 90% of winners never moved more than 3 pips against entry, a stop at 6 pips looks generous and a 3.5 pip stop looks tight enough. Put the stop where winners stop needing room.

It sounds neat, and there is a trap. Contrarian view: MAE tuning on a small log mostly fits the past.

Reasons it goes wrong. First, the sample is small. The 90th percentile of 60 winners is set by about six trades, and one odd day moves it. Second, the data comes from the old stop. Trades that would have survived with a different stop were already closed as losers, so the winners list is shaped by the stops that existed. Third, a tighter stop changes the sequence of trades afterwards, which the log cannot show.

That is no reason to ignore MAE. It is a reason to treat it as a hint and to test on trades the tuning has not seen. Splitting a log in two, tuning on the first half, then checking on the second half, is a cheap sanity check.

Volatility matters as well. A stop that worked in a quiet London morning may be far too tight when the same pair is moving twice as fast on a data release, and MAE averaged over both mixes the two worlds together.

A stop that only works on the data it was built from is not a stop, it is a memory.

Do you set stops from the data, from market structure, or from volatility like ATR?