Skip to main content

Feature

EA Drift Detection

Catch Expert Advisor degradation weeks before your equity curve shows it. CUSUM statistical monitoring on every live MT5 strategy.

Why drift kills profitable EAs

Every Expert Advisor eventually stops working. Market regime shifts, broker spreads widen, volatility compresses — the specific conditions that made your backtest profitable disappear. The problem is not that this happens. The problem is that you usually discover it after five figures of drawdown, when the equity curve finally bends down hard enough to ignore.

Most traders spot drift too late for two reasons. First, equity curves are noisy — a real drawdown looks identical to normal variance until it is obviously not. Second, human pattern-matching on P&L charts is terrible. You see what you want to see, and the EA keeps trading. Drift detection fixes both problems with statistics instead of intuition.

How CUSUM detects drift before the chart does

Algo Studio uses CUSUM — cumulative sum control charts — the same statistical technique that flags manufacturing defects in semiconductor fabs and catches quality regressions in large software systems. The method is simple: accumulate every deviation from your backtest baseline, and when the running sum crosses a calibrated threshold, flag the strategy.

Unlike a simple rolling-window P&L check, CUSUM accumulates directional evidence over time. A run of small losses that would individually look like noise adds up. By the time the sum crosses threshold, you have high confidence the shift is real — not a coincidence — and you can act before the raw equity curve makes it visually obvious.

Each live strategy gets its own CUSUM chart computed against its own uploaded backtest baseline. Thresholds are tuned per strategy based on backtest variance, not global defaults. This is why drift detection is a baseline-required feature: without a backtest, there is no statistical reference point.

What drift detection is not

Drift detection is not a stop-loss. It does not kick in on a single bad trade, a drawdown, or a single losing day. Those are normal. CUSUM specifically looks for persistent directional change in the distribution of outcomes — the kind of slow bleed that degrades a strategy over weeks.

It is also not a prediction of future losses. It is a current-state signal: your EA is now trading outside its historical distribution. What you do with that signal is up to you — pause the strategy, reduce size, investigate broker changes, or re-optimize. Algo Studio surfaces the alert; you decide.

From alert to action: auto-halt governance

If you don't want to be woken up by alerts, Algo Studio can halt a drifting EA automatically. When CUSUM crosses your configured threshold, the system issues a governance action to the monitor EA running inside MetaTrader. The EA stops opening new trades on the next tick. Existing positions are untouched — you decide whether to close them manually or let them run.

Auto-halt is opt-in per strategy and off by default. For most traders, a Telegram alert plus a 15-minute cooldown to investigate is the right workflow. But for prop firm accounts where a single runaway day can blow the account, auto-halt is the safety net that makes algorithmic trading survivable.

Frequently asked questions

How is drift detection different from a trailing drawdown alert?

A drawdown alert fires after losses. Drift detection fires when the pattern of outcomes shifts, regardless of current P&L. A strategy can be in drift while still up on the month — CUSUM catches the distribution change early, before losses accumulate.

Does drift detection work without a backtest baseline?

No. Drift is measured against your own backtest — without a reference distribution, there is nothing to deviate from. Live monitoring still works without a baseline, but drift detection and health scoring require uploading one.

How often does the CUSUM run?

After every closed trade. The algorithm is incremental — each new trade updates the running sum in O(1) time, so there is no performance cost to checking continuously rather than on a schedule.

Can I tune the sensitivity?

Yes. The default CUSUM parameters are calibrated to flag drift with ~95% confidence after a statistically meaningful sample. Aggressive users can tighten thresholds; defensive users can loosen them. Sensitivity is set per strategy.

Related features

Related reading

Try ea drift detection on your own EA

Free Baseline plan includes every feature. 1 monitored account, no credit card, no time limit. Upload your backtest and start monitoring in 5 minutes.