Every system has losing periods, and every system that has genuinely stopped working also has a losing period. From inside a drawdown the two look identical, which is what makes this the hardest judgement in systematic trading.
Getting it wrong is expensive in both directions. Abandoning a working method during a normal bad stretch is the most common way people discard something that would have paid. Continuing to fund a dead one is the most expensive.
Decide what abnormal looks like before you need to know
The single most useful thing you can do about strategy decay is done before deployment, while you are calm and have no money at risk.
Your backtest already contains the answer to the only question that matters: what does a bad stretch look like for this method? Pull it out explicitly rather than leaving it in a summary statistic.
The deepest drawdown in testing, and how long it lasted before recovery. Not the average. The worst one.
The longest run of losing trades and the longest period without a new equity high. Most people know their win rate and have never looked at the longest losing sequence it implies, which is usually longer than they would guess.
The distribution of monthly results, so you know how often a losing month happened historically. A method that lost money in four months out of twelve in testing will lose money in four months out of twelve live, and each one will feel like evidence of failure.
Write those numbers down and keep them somewhere you will find them later. They convert an unanswerable question — is this normal? — into a checkable one. If the current drawdown is inside what testing produced, you are experiencing the thing you signed up for. That is not proof the method still works, and it is proof that nothing unusual has happened yet.
The evidence that actually indicates decay
Losses alone are weak evidence, because losses are what a normal bad period is made of. Four things are stronger.
The drawdown exceeds anything in testing, on any measure. Deeper than the worst historical drawdown, or longer than the longest, or more consecutive losers than the method ever produced. One of these can happen by chance; testing never captures the true worst case. Several at once is a different signal.
The character of the losses has changed. This is the most informative and the least looked at. Compare the current trades to the historical ones on their own terms — average holding period, average size of win against average size of loss, how often the exit is a stop rather than a target. A method still producing its usual shape of trades and losing is having a bad run. A method whose trades no longer look like its own trades is meeting a different market.
The structural reason it worked has changed. Every method exploits something specific. If you cannot say what yours exploits, that is a separate and larger problem. If you can, ask whether it still exists: a rule change, an instrument's expiry cycle being altered, spreads narrowing, a participant type entering the market, a tick size or lot size revision. Decay usually has a cause, and the cause is often visible if you know what to look for.
It stopped working everywhere at once. If the method is applied to several instruments and all of them turned at the same time, that points to a market regime change rather than to one instrument's noise. Independent failures are noise; simultaneous failure is information.
The rules worth committing to in advance
The reason to write these before deployment is simple: during a drawdown, every rule you have not already written will be reasoned away. That is not a character flaw, it is what a drawdown does to judgement.
A reduction threshold. At a defined drawdown level, cut position size by a defined proportion. This is more useful than a single on-off decision because it keeps you in the method if it recovers while reducing the damage if it does not. It also means the account that survives is large enough to matter.
A stop threshold. At a worse level, stop and do not restart until a stated condition is met. Define the condition now, because “until I feel better about it” is not one.
A no-modification rule while live. Adjusting parameters during a drawdown is fitting the method to the recent past, which is the same error as overfitting a backtest, except done with real money and without a test set. Change nothing while running. Take the method offline, change it deliberately, test the change on data it has not seen, and redeploy as a new system with its own numbers.
A review date rather than a review trigger. Look at the method on a schedule — monthly, quarterly — rather than whenever it hurts. Reviewing after every painful week guarantees that every decision is taken at the worst possible moment.
When it really has stopped, and what to keep
Retiring a method feels like a failure and is usually a normal event. Approaches stop working. That is why the ones that still work are still paying.
What is worth keeping from a retired system is more than people realise.
The infrastructure. The data handling, the execution layer, the logging, the monitoring and alerting, the deployment. None of that was specific to the method, and rebuilding it is where most of the original effort went.
The record. A retired strategy with a complete history of what it did and when it stopped is a genuine reference point. The next method's drawdown can be compared against something you actually lived through rather than against a number in a spreadsheet.
The reason it worked, and the reason it stopped. Written down honestly, this is the most valuable output of the whole exercise. A method that stopped because an exchange changed a rule teaches something different from one that stopped because too many people found it.
The mistake to avoid at this point is the natural one: immediately deploying a replacement built from the same data, during the same period, by the same person, with the same assumptions. That is not a new system. It is the previous one with different parameters, and it will meet exactly the same market.
An account with a full trade history
Every fill and charge, downloadable, for the honest comparison. Free to open
This is educational material and not advice. Nothing here recommends any strategy or position. We run no tips group and no signal service, we manage nobody's money, and no return is promised. Automated trading carries a real risk of loss, and a tested method can still lose money.
Disclosure: the account-opening link on this page is under Atul Shrivastava's Zerodha Authorised Person registration (NSE AP Reg: AP2516003481; Zerodha Broking Ltd. SEBI Reg: INZ000031633) and earns a revenue share. TheFinBaba is not a SEBI-registered Investment Adviser — this content is educational, not investment advice.
Frequently Asked Questions
How do I know if my strategy has stopped working or is just in a drawdown?
Compare the present against what your own testing produced. If the drawdown is deeper or longer than anything in the test, if the trades no longer resemble the method's historical trades, if the structural reason it worked has changed, or if it failed across several instruments at once, that is evidence. Losses alone are not.
Should I change parameters while a strategy is running?
No. Adjusting during a drawdown fits the method to the recent past, which is overfitting done with real money and no test set. Take it offline, make the change deliberately, test it on data it has not seen, and treat the result as a new system with its own numbers.
What drawdown level should make me stop?
Whatever level you defined before deploying, based on the worst drawdown your testing produced. A reduction threshold followed by a stop threshold works better than a single decision, because it keeps you in the method if it recovers while limiting the damage if it does not.
Why do strategies stop working at all?
Usually because the specific thing they exploited changed. A regulatory change, an altered expiry cycle, narrower spreads, a revised lot size, or simply more participants doing the same thing. Decay generally has a cause, and it is often identifiable if you know what your method was exploiting.
What should I keep when I retire a strategy?
The infrastructure, which was never method-specific and is where most of the work went. The complete record of what it did. And an honest written account of why it worked and why it stopped, which is the most useful input into whatever you build next.
Related Reading
- Loss limits, taught for Delhi's traders
- Getting the worst-case numbers out of a backtest
- The errors that come before decay
- Sizing rules that survive a bad run
- Testing a replacement before funding it
- Building and deploying the system properly
Disclaimer: TheFinBaba provides educational content only - this is not investment advice. Trading involves risk of loss.
