Regime change or random walk? Telling a real shift from ordinary noise

When a strategy that worked stops working, two very different things could be happening: the world changed (a regime shift), or nothing changed and you are watching ordinary randomness (a drawdown inside a still-valid process). Reacting to noise as if it were a regime change is how people abandon good strategies; ignoring a real regime change is how people keep trading dead ones.

Why it is genuinely hard to tell

Random series produce streaks that look meaningful. A run of losses feels like a broken strategy the same way a run of wins feels like a discovered edge, and in both cases the feeling is unreliable. The honest position is that a short bad run is usually just variance, and declaring a regime change requires more evidence than discomfort — the same scepticism the why backtests fail piece applies to good-looking results applies here to bad-looking ones.

What actually helps you decide

Define what "working" means in advance, with a threshold you set before the drawdown, so you are not negotiating with yourself mid-pain. Look at whether the mechanism still holds — has the reason the strategy worked changed, or only the recent outcome? And judge over enough samples that variance has room to even out; a regime call made on a handful of trades is almost always noise. The lessons from a hundred experiments are mostly lessons in not over-reading short runs.

The asymmetry to respect

False alarms and missed shifts are not equally costly. Switching strategies on every wobble guarantees you churn and never let an edge compound; never switching means a genuinely dead strategy bleeds slowly. Most people err toward the first. The discipline is to require real evidence — a mechanism story plus a sustained, pre-defined breach — before concluding the world moved.