How to Detect Edge Decay From Your Trading Journal

A strategy doesn’t send a memo when it stops working. It just starts losing in a way that feels exactly like a normal rough patch — until the rough patch never ends. The hard part isn’t noticing the losses; it’s telling the difference between a strategy that’s dying and one that’s simply variance-ing through a bad stretch it will recover from.

Why a strategy stops working before you notice

Every edge lives in a market condition. Trend-following works when there’s trend; mean-reversion works when ranges hold; a news-fade works when liquidity behaves a certain way. When the underlying condition shifts, the edge degrades — but your rules don’t change, so you keep taking the same setups into a market that no longer rewards them.

The delay is the danger. Edge decay is gradual and losing streaks are loud, so you attribute the early decay to bad luck and keep pressing. By the time the losses are undeniable, you’ve given back weeks of progress. On a prop account with a hard drawdown, that lag can be the difference between a recalibration and a breach.

The only defense is measuring the edge continuously instead of feeling for it. Your journal already holds the evidence — the question is how you read it.

Tracking rolling expectancy instead of a single number

A lifetime expectancy figure is nearly useless for detecting decay, because a strong first hundred trades will mask a weakening recent hundred for a long time. The average dilutes the signal you actually care about.

Track rolling expectancy instead — expectancy over your last N trades, recomputed as each new trade posts:

A 50-trade rolling window smooths noise while still turning when the edge turns.
Compare the rolling figure against your calibration baseline — the expectancy the strategy showed when you validated it.
Watch the slope, not just the level. A rolling expectancy trending down over successive windows is more informative than any single reading.

If you want to see the mechanics of how the number is built from wins, losses, and average sizes, work an example through an expectancy calculator and read the expectancy primer. Expressing results in R-multiples rather than raw currency keeps the rolling series comparable even when your position size changes.

Confidence intervals: signal vs noise in a drawdown

Here’s the trap: a drawdown and edge decay produce the same recent numbers. A profitable strategy will, by chance, string together losing trades that look identical to a broken one. Reacting to every dip means abandoning good strategies at their worst moment; ignoring every dip means riding a dead one into a breach.

The way out is to stop reading expectancy as a single point and start reading it as a range. With a modest sample, your true edge could sit anywhere inside a band around your measured figure — and the width of that band is the whole story.

If the entire confidence interval stays above zero, a drawdown is very likely variance. The edge is intact; you’re just in the left tail of a positive distribution.
If the interval has slid to straddle zero, you can no longer say the edge is real from the data alone. That’s the moment to get cautious.
A narrowing band that’s drifting down is the clearest decay signal — more data, worse center.

This is exactly what a Wilson confidence interval is built for: it’s honest about small samples and doesn’t overclaim. Shibiki computes a live edge-health band per strategy this way, so instead of staring at a jittery point estimate you can see whether your recent slump is still inside the range a healthy edge would produce — or whether it’s fallen out of it.

Segmenting by regime to isolate what changed

Once the data says something changed, the next question is what. A blended number tells you the strategy is weaker; it doesn’t tell you why. Segment the log to find the leak.

Cut your trades by the dimensions most likely to hold a regime shift:

Volatility environment — high vs low; many edges are regime-specific.
Session or time of day — an edge can die in one session and survive in another.
Instrument — decay is often concentrated in a subset, not across the board.
Setup variant — if your strategy has flavors, one may have rotted while the others hold.

Frequently you’ll find the decay is localized: the edge is fine in the morning session and gone in the afternoon, or intact on one instrument and dead on another. That turns a scary “my strategy stopped working” into a surgical “stop trading this slice.” Segmenting is only possible if your journal tags these fields — another reason automatic capture of time, instrument, and conditions beats hand-entry that skips them under pressure.

Deciding to pause, recalibrate, or hold

The diagnosis maps to three actions:

Hold — the interval is still comfortably above zero. This is a drawdown, not decay. Keep trading your rules and let variance resolve.
Recalibrate — the edge survives in a subset. Tighten the strategy to the conditions still paying, and reset your baseline to the smaller, honest edge.
Pause — the interval has crossed zero and the slope is down across segments. Stop, protect the account, and rebuild the case from fresh data before committing size again.

The worst outcome is doing none of these because you couldn’t tell which situation you were in. Continuous edge-health measurement is what makes the call obvious instead of emotional — and on a funded account, making that call a week early is often what keeps the account alive.