Trading System Decay Detection and Iteration

Edges erode. Markets adapt, participants copy, and the structure that made a system profitable shifts. Decay is inevitable; the question is whether you detect it before a full drawdown forces the issue.

Early Decay Signals

Define decay triggers in advance, before deployment, so emotion does not govern them.

• Rolling 50-trade Sharpe drops below 50% of the backtest Sharpe, the earliest quantitative signal, firing before drawdown peaks.
• Win rate falls more than 1 standard deviation below backtest win rate over 50 trades.
• Average R declines steadily across three consecutive 20-trade buckets.
• Max drawdown exceeds the backtest 95th-percentile Monte Carlo drawdown.

Any single trigger is a warning; two firing together is action: cut size by 50% and investigate.

Distinguishing Decay from Variance

Not every drawdown is decay. Systems have normal variance. Before declaring decay:

• Compare the live drawdown to the Monte Carlo 95th-percentile drawdown. Within it, the system is behaving as expected.
• Check whether the regime matches the system's design. A trend system losing in a choppy range is out of regime, not decaying.
• Confirm trade execution matches backtest assumptions. Slippage or missed fills can mimic decay.

Decay is present only when performance deteriorates beyond what variance and regime explain.

The Iteration Process

When decay is confirmed, follow a fixed process, do not improvise.

1. Reduce size first. Cutting risk to 25-50% of full size stops the bleeding while you investigate.
2. Re-audit the edge. Has the underlying market structure changed? Did a competitor copy the signal? Is the data still clean?
3. Re-test on recent data. Re-run the backtest on the most recent 2-3 years. If the edge is gone even in-sample, the system is dead, not decaying.
4. Adjust one variable at a time. Change a filter, a parameter, or an exit rule singly, and re-run walk-forward. Multiple simultaneous changes obscure what worked.
5. Forward-test the iteration. Do not return to full size until 50 forward trades confirm the fix.

When to Kill vs Iterate

Iterate when:

• The edge survives in recent in-sample testing but parameters shifted.
• The market regime is transitional and likely to revert.

Kill when recent in-sample testing shows no edge, when the structural reason for the edge is gone (regulatory or microstructure shift), or when three iterations have failed. Killing a dead system is cheaper than nursing it; capital and attention are finite.

Portfolio-Level View

Track decay at the portfolio level, not only per system. A single system decaying is manageable; multiple decaying simultaneously signals a regime shift. Then reduce total portfolio risk and revisit the strategy mix rather than patching systems one by one.

The Discipline

Decay detection only works with a live journal recording every trade, trigger check, and decision. Without the log, you rationalize losses as variance until the account forces the truth. Define triggers, log the data, act on them, and decay becomes a managed cost rather than a surprise.