Every trading course teaches you to build a system. Almost none teach you to kill one — which is the decision that decides whether a career survives a decade rather than a year.
Two failure modes, opposite and equally expensive:
Killing too early. Abandoning a working system during a normal losing run. This is the more common one, and it is why traders say "nothing works."
Killing too late. Continuing a system whose edge is genuinely gone, month after month, because stopping means admitting it.
The difference between them is not judgement. It is a decision rule written before either situation arrives.
Three explanations, in the order you must test them
When results deteriorate, there are exactly three possibilities. Confusing them is the expensive part, and reaching for the flattering one first is the reliable error.
1. Variance. Results are inside what your sample predicts. Change nothing.
2. Execution drift. You stopped following the rules. Check the adherence field.
3. Regime change. The market genuinely changed in a way that removed the edge.
Test in that order. Regime change is the most comfortable explanation — it blames the market — so it is the one people reach for first and the one that must be checked last.
Testing for variance
Take your historical standard deviation and ask: is the current drawdown inside the range your own Monte Carlo produced?
If your 95th-percentile shuffled drawdown was 19.8% and you are down 14%, you are inside your own forecast. Uncomfortable, and not evidence of anything.
Only when current performance falls *outside* what your own sample can produce does the question become interesting.
Testing for execution drift
Split the last 50 trades by the adherence field.
Finding
Diagnosis
Adherence 95%+, results poor
Not drift. Continue testing
Adherence fell from 90% to 60%
This is your answer. The system is untested, not broken
Almost every "the system stopped working" case that reaches this test stops here. The system did not change; the trader did, usually gradually and usually without noticing.
Testing for regime change
This requires a measurable change, not a feeling. Compare the last 60 trading days to the prior 250:
Measure
Regime signal
ATR vs its long average
Above 1.5× or below 0.7×
Trend persistence (% of days closing in the direction of the 20-day)
Shifted by more than 15 points
Average daily range
Halved or doubled
Correlation of your instrument to its usual driver
Broken down
If nothing measurable has changed, it is not a regime change — regardless of how different the market feels.
And note: a genuine regime change does not necessarily kill an edge. A trend-following system in a newly range-bound market has a *suspended* edge, not a dead one. The correct response may be to stop trading it and wait, not to discard it.
The decommission rule
Written before you need it, because the moment you need it is the moment you are least able to write it:
I will retire this system if, after 100 further trades taken with 90%+ adherence, expectancy is negative and the 95% confidence interval excludes my original measured edge.
Read what that demands:
100 further trades — not a bad month
90%+ adherence — you actually ran the system
Negative expectancy — not merely disappointing
Statistically distinguishable from the original edge
If all four hold, the edge is gone and continuing is not perseverance. It is a sunk-cost decision.
Suspending versus retiring
Not every deterioration requires a permanent decision.
Suspend when the regime measurably changed and your system depends on the missing condition. Stop trading it, keep the rules, keep watching. A trend system in a range is suspended, not dead.
Retire when the four conditions above are met, or when the structural reason the edge existed has gone — a spread that widened permanently, a session that moved, a market that closed.
Suspension is reversible and much less costly than a decision made under stress.
Worked example
A trader's system: 340 trades, +0.29R, 95% CI +0.12 to +0.46R. It has worked for fourteen months.
The last 70 trades produce −0.08R. They are down 11%.
Test 1 — variance. Monte Carlo on the 340-trade sample gives a 95th-percentile drawdown of 16%. 11% is inside it. No conclusion available.
Test 2 — adherence. Last 70 trades: 92% followed the plan. Not drift.
Test 3 — regime. ATR at 1.08× its long average. Trend persistence 54%, versus 58% historically. Daily range normal. Nothing measurable changed.
Verdict: variance. Continue at unchanged size.
The next 90 trades produce +0.33R. The 500-trade expectancy settles at +0.27R — statistically unchanged from before the drawdown.
Now the counter-case. A different trader, same tests, but test 3 shows ATR at 0.58× its long average and trend persistence down from 61% to 39%. Their system is a breakout system.
A measurable regime change, and it removed exactly the condition their edge required.
They suspend rather than retire. Rules unchanged, no trades taken, ATR monitored weekly. Four months later volatility returns, they resume at half size, and the edge returns with it.
The trader who retired that system permanently would have thrown away a working edge because the market was quiet for a season.
What to take away
Killing too early and killing too late are both fatal. A written rule prevents both.
Test variance, then adherence, then regime — never the other order.
Regime change requires a measurable change, not a feeling.
Write the decommission rule before you need it: 100 trades, 90% adherence, negative expectancy, statistically distinguishable.
Suspend before you retire. A suspended edge is usually just waiting for its conditions.
