Strategy decay: when a scalping edge fades

[Fairman]

Strategy decay: when a scalping edge fades

Edges in short-term trading do not last forever. Market structure changes, other participants adopt similar approaches, brokers adjust execution, and volatility regimes shift. A system that worked for a year may slowly stop working.

Decay is hard to spot because results are noisy. A bad month could be variance, or it could be the start of a decline. Distinguishing them requires monitoring.

Set up control charts. Track rolling expectancy over, say, 200-trade windows, and compare it to the historical distribution. If it falls below a defined threshold, such as two standard errors under the long-term mean, flag it.

Also monitor components: win rate, average win, average loss, slippage, and spread. A change in one points to the cause. Rising slippage suggests execution problems. A falling win rate suggests a change in market behavior.

Have a plan for decay: reduce size, retest, and consider whether adaptation or retirement is appropriate.

How is a decaying edge distinguished from a normal bad patch?

[Amanhasnoname]

The thing that helped me most with decay was keeping a record of conditions along with results. When my rolling expectancy fell for about three weeks, I first assumed the edge was fading, but my notes showed that the average daily range on the pair had fallen by about a third because of a quiet summer period. The strategy was designed for movement, and with less of it the same rules naturally earned less.

So now I tag every week with the realised volatility, and I compare the current expectancy to weeks with similar volatility, not to the all-time average. When it's weak and volatility is normal, I take it seriously. When it's weak and the market is dead, I wait.

For the plan you mention, my first step is only to halve the size, not to stop. That keeps me trading the system, so I collect more data, and it removes the pressure of deciding whether the edge has gone. If another two hundred trades confirm the weakness at normal volatility, then I retest and decide about retiring it.