Knowing When a Trading Strategy Has Stopped Working

Every strategy you trade long enough will eventually lose money for a while. That sentence is obvious in the abstract and uncomfortable in practice. The equity curve on a system that has been profitable for years turns down, stays down longer than feels comfortable, and somewhere in that stretch a quieter, more consequential question starts running in the background of everything else you do — one that has nothing to do with today’s number and everything to do with what happens next.

There is no way to ask the market directly whether the conditions that made your edge work are still in place, or whether they ended quietly some time before your first losing trade showed up. All you have to look at is the equity curve, and in the moment, an ordinary bad stretch and a genuinely dead strategy produce exactly the same picture: a line going down and to the right, with no label attached explaining which one you are looking at. The question every systematic trader eventually has to answer is this: has the edge actually gone, or is the strategy simply doing what it has always done? Answer it wrong in either direction and it costs you. Pull a strategy that was only ever in an ordinary drawdown, and you lock in the loss and then, almost certainly, miss the recovery that was already priced into its long-run statistics. Keep trading one that has genuinely stopped working, and there is no floor — markets move on, the edge is gone, and every trade from here on is just financing the decline with fresh capital, on the hope that whatever made it work will come back on its own.

The way out is not a better feeling about the drawdown — feelings do not distinguish between the two cases, which is exactly the problem, since a temporary dip and a permanent decline feel identical from inside a losing streak. What actually resolves the question is an objective, pre-committed rule: defined before you are in a drawdown, applied without renegotiation once you are in one. This is less an analytical exercise than a risk-management one. The hard part was never running the numbers — it is having a rule, and then honouring it precisely when honouring it hurts the most.

The approach I use for building that rule is what StrategyQuant X calls a WHAT-IF analysis, and it comes in two parts that between them answer the two questions that actually matter here: what does normal actually look like for this particular strategy, and at what point has the current stretch moved outside of normal and into something you should act on.

First, separate the two questions

It helps to keep these apart, because collapsing them into one vague feeling of unease is exactly what leads to bad decisions:

What does normal actually look like for this strategy? Not a feeling — a measured quantity. If the strategy was properly walk-forward tested, validated on data it never saw during optimisation, the out-of-sample segment already gave you a real, if limited, preview of how deep its drawdowns run, how long they last, and how choppy the recovery gets before it resumes making new equity highs. That segment is not a formality completed to satisfy a checklist before deployment. It is the baseline everything afterwards gets measured against.
At what point does the current stretch stop being explainable by that baseline? This is the harder question, because there is no natural line — a drawdown does not announce the moment it crosses from ordinary into terminal. Somebody has to draw that line. The only real choice is whether you draw it in advance, while calm, or discover you are drawing it under pressure, which in practice means you will not draw it at all — you will keep moving it.

Skip the first question and you have no way to tell an outlier from an ordinary bad month, so every drawdown looks equally alarming. Skip the second and you have a diagnosis with no decision attached to it — you can always tell yourself the current stretch is “probably still normal,” right up until the account has absorbed damage no backtest ever justified. A rule you can actually use needs an answer to both.

1. Compare live behaviour against the baseline — does this still look like the strategy you tested?

What it defends against: panic — treating an ordinary losing streak as proof of decay simply because it is happening to you, right now, with real money attached.

The first half of a WHAT-IF analysis puts that baseline to work directly. You line up the strategy’s recent or live performance against its established training and out-of-sample behaviour and ask a narrow, answerable question: is the current segment consistent with what validation already showed you, or does it look like something new? Not “is it profitable this month” — strategies are not profitable every month, and expecting otherwise is its own kind of naivety — but whether the depth and duration of the current stretch sit inside the range the strategy has already demonstrated it can produce.

Framed this way, a drawdown stops being a verdict and starts being a data point. If it falls comfortably inside the distribution validation already produced, you have not actually learned anything new — the strategy is doing what it told you it would do, on a schedule that was never going to be convenient. That is uncomfortable, but it is not evidence of anything. The mistake most traders make at this stage is treating discomfort as information. It is not. Only a genuine departure from the established baseline is.

2. Set a drawdown margin — how far past the backtest is too far?

What it defends against: the opposite failure — no predefined point at which a losing streak stops being tolerable, so every new low gets rationalised as “still within reason” until the account has absorbed far more than any backtest ever justified.

The second half is more mechanical, by design: a WHAT-IF snippet that tracks the strategy’s live drawdown as a percentage and compares it against the maximum drawdown the strategy actually produced in backtest. Before going live, you decide how far beyond that historical maximum you are willing to let the strategy run before you treat the result as broken rather than unlucky. Not at the historical maximum — a margin above it.

The margin matters more than it might look like it should. Set the threshold exactly at the backtested worst case and you will get stopped out by the first ordinary bad patch, because live trading extends the sample, and an extended sample will eventually produce a new worst case even on a strategy that is perfectly healthy — that is simply what happens when you keep drawing from the same distribution for longer. Set no threshold at all and you have no rule, only a hope that you will recognise decay when you see it, which is the same hope that already failed you in the opening paragraphs of this piece. The margin is what separates the two: it turns “the drawdown is now in territory the backtest never saw” from a source of panic into a specific, pre-agreed fact. Breach it, and the decision has already been made. You are not deciding whether to stop in the middle of a losing streak — you decided, long before now, exactly what would make you stop, and what you are watching is simply the rule executing.

The rule I use

Running either half of a WHAT-IF analysis is the easy part. Actually honouring what it tells you is the real difficulty, and it is where my practical rules live:

Set the margin before you go live, not during a drawdown. Decide it when you are calm and looking at a distribution of outcomes, not when you are deep in red numbers and looking for permission to hold on.
Anchor the margin to the strategy’s own history, not to a number that simply feels safe. A strategy with a naturally choppy backtest needs more room than one with a naturally smooth equity curve, or you will end up killing it on a predictable schedule instead of for cause.
Treat a breach as a stop-trading signal, not a redesign brief. The rule’s job is to tell you when to stop, not what to build next. The moment you catch yourself thinking “let me just tweak a parameter and keep it running,” you have already broken the rule — you are negotiating with it instead of honouring it.
Keep watching inside the margin, too. A drawdown threshold is a hard stop, not the only signal worth tracking. Ongoing equity-curve monitoring, and a general sense of whether live performance is quietly degrading relative to the out-of-sample baseline, can flag a problem before the hard stop is ever reached — which matters, because by the time the hard stop fires, you have already absorbed the full drawdown it permits.
Do not move the goalposts mid-drawdown. Revisit the threshold between deployments, on strategies you are actively redesigning — never on the strategy that is currently losing, while it is losing. Adjusting the rule under the exact pressure it exists to resist defeats the entire point of having one.

None of this is complicated. All of it is hard to do the first time you watch a real threshold approach on a strategy you like.

The point

No version of this rule is perfect, and it is worth being honest about that rather than selling a certainty nobody actually has. A margin-based drawdown threshold will, sometimes, stop you out of a strategy that would have recovered — that is the cost of having a rule at all, paid in the false positives it inevitably produces. Trade without one, and you avoid that particular cost, but you take on the other one instead: no defined point at which you stop financing a decline that was never coming back. You are not choosing between risk and no risk. You are choosing which risk you would rather carry, and how much of it. I would rather carry a known amount of the first risk than an unbounded amount of the second, which is the entire reason I set the margin before I go anywhere near a live account — but that is my trade-off, not a law of markets, and yours might reasonably sit somewhere else.

That is the actual decision, and it only needs to be made once, in advance, while you can still think clearly about it: how much of a beating are you willing to let a strategy take before you stop trusting it, weighed against how much of a recovery you are willing to risk missing in exchange for the discipline to walk away from the ones that do not come back. Decide that on a calm afternoon with the numbers in front of you, and a losing streak becomes something you already have a plan for. Decide it in the middle of the losing streak, and you are no longer managing risk — you are guessing, with money attached.

If you are staring at a live drawdown right now and want an outside read on whether it is normal or terminal, that is exactly the kind of work I do.