The hardest thing about running a strategy is that a bad month and a dead strategy look identical from the inside. Both hurt. Both make you want to change something. And the reason people get this wrong so consistently is that they try to make the call while they are losing money, which is exactly the moment your judgment is worth the least. I have shut down strategies that were fine and kept ones that were quietly rotting, and in both cases the mistake was the same. I was deciding in the middle of the pain instead of before it started.
So the whole game is moving the decision earlier. You write down, while you are calm and the equity curve is boring, the specific numbers that would tell you the edge is gone. Then when the drawdown comes, and it will, you are not asking yourself an impossible question under stress. You are just checking whether a line you drew months ago has been crossed.
A drawdown is normal until it beats your backtest
Every strategy has a worst historical stretch. If you backtested it honestly, you know roughly how deep the deepest drawdown went and how long the longest flat period lasted. That number is your first tripwire. A normal drawdown stays inside the envelope your backtest already showed you. The strategy has been this deep before and came back. Nothing new is happening.
The trouble starts when live results go somewhere the backtest never did. If your worst historical drawdown was, say, in the low twenties as a percentage, and you are live and pushing past thirty with no sign of a floor, that is not the same event. You are in territory the system was never shown to survive. It might recover. But you no longer have evidence that it will, and "it might recover" is how people ride a dead strategy all the way down.
One honest caveat. A single backtest gives you one path, and the real worst case is almost always deeper than the one you happened to observe. So do not treat your backtest max drawdown as a hard ceiling. Treat it as a soft one, and set your real tripwire a bit beyond it, sized to how much of that path you are willing to sit through. The point is that the number exists before you need it.
Watch the trend, not the last few trades
Drawdown tells you how much it hurts. It does not tell you why. For the why, I care more about whether the engine that produces the edge is still turning, and the cleanest read on that is rolling expectancy. Expectancy is just your average result per trade, win rate times average win minus loss rate times average loss. Compute it over a rolling window of your last N trades and watch which way it drifts.
A strategy in a normal drawdown still has positive expectancy over a reasonable window. It is losing because of variance and clustering, not because the average trade has gone bad. A decaying edge looks different. Rolling expectancy grinds down toward zero and stays there, or goes negative and does not bounce. When the equity curve and the expectancy trend disagree, the expectancy trend is the one telling the truth, because equity can be dragged down by one ugly cluster while the underlying average is fine. If the average itself is sinking, the machine is broken, not just unlucky.
The other tool I lean on is a win-rate control chart, borrowed straight from manufacturing quality control. You know your strategy's baseline win rate and roughly how much it wobbles from sample to sample. Draw control bands around that baseline, a couple of standard deviations wide. As long as your rolling win rate bounces around inside the bands, it is behaving. When it drops below the lower band and stays there across several independent windows, that is a signal the process shifted, not that you hit a rough patch. One window outside the band is noise. A run of them is a message.
Why edges die in the first place
It helps to know what you are actually watching for, because the cause tells you whether decay is likely to be permanent. A few of the usual ones:
- Crowding. Your edge was a discount other people were leaving on the table. Enough of them find it, the discount closes, and your fills get worse right as your signal gets more popular. This one rarely comes back.
- Regime change. The strategy was fitted, knowingly or not, to a particular volatility and trend environment. The environment turns and the assumptions underneath it stop holding. This one can come back if the regime cycles, which is exactly why you do not want to bury the system permanently on the first bad stretch.
- Venue mechanics. Fee schedules change, a venue tweaks its matching engine, liquidity migrates somewhere else, funding behaves differently. The signal is intact but the cost of acting on it quietly ate the margin. This one is worth checking first because it is often fixable without touching the logic at all.
Crowding and venue changes tend to be structural. Regime shifts are the ones that tempt you into holding forever on the theory that it will all come back. Sometimes it does. The tripwires are how you avoid betting the account on that hope.
Cut size before you cut the system
Here is the part people skip. Retiring a strategy is not a single yes-or-no switch, it is a staircase, and the response should scale with how many independent signals have fired. Reserve the full shutdown for the case where the evidence has genuinely stacked up.
- One tripwire trips (say, drawdown past backtest max, but expectancy still positive). Halve the size. You stay in the game and keep collecting data at lower risk. You have not concluded anything, you have just stopped paying full price to find out.
- A second, independent signal confirms (rolling expectancy now flat or negative, or win rate breaks the lower control band and stays there). Cut size again, hard, down to a token allocation. You are close to a verdict but you want a bit more evidence.
- The edge is structurally explained and gone (you can name the cause, crowding or a venue change, and the numbers agree). Now you retire it. Fully. No token position kept alive out of sentiment.
Staging this way kills the two extreme errors at once. Cutting at full size on the first bad week means you murder working strategies over noise. Holding full size until you are certain means you donate the whole drawdown to the market before you admit it. Sizing down in steps keeps you alive through the false alarms and caps the bill on the real ones.
None of this works if you invent the numbers mid-drawdown. Write them before you go live, ideally in the same document as the strategy itself: your drawdown tripwire, your expectancy window and threshold, your control-chart bands, and the size cut at each stage. On Blockcircle I lean on the backtester for the historical drawdown and expectancy figures and on the scorecards to watch the live versions drift, but the tool matters far less than having the lines drawn in advance. The whole point of a tripwire is that it makes the decision for you at the exact moment you would otherwise make it badly. A strategy that has quietly stopped working will keep taking your money for as long as you let it, and hope is not a stopping rule.