A system stops working for one of three reasons: the regime changed and will change back, the edge has genuinely decayed because too many people found it, or it never existed and the backtest was fitted. These look identical from inside a drawdown.
The three causes
| Cause | What it looks like | Correct response |
|---|---|---|
| Regime change | Losses concentrated in one market condition; the setup still triggers normally | Reduce size, wait — the system is intact |
| Edge decay | A slow, steady deterioration across all conditions over years | Retire it; the inefficiency has been arbitraged away |
| Never existed | Live results diverge from the backtest immediately, from the first month | Retire it and re-examine how it was tested |
Normal drawdowns are longer than expected
Before deciding a system is broken, check what its own history says a bad run looks like. Most people quit inside the range their backtest already contained.
Note how much steeper recovery becomes past about 30%. This is why kill criteria should sit well before that point.
Kill criteria, written in advance
The only defence against deciding during a drawdown is deciding before one. Write the conditions that would retire the system while you are calm, and treat them as binding.
- 1A drawdown limit beyond historical worst
For example 1.5× the worst drawdown in testing. Reaching it means live behaviour is outside anything the system has demonstrated.
- 2A time limit
Flat for longer than the longest flat period in the backtest, by a clear margin.
- 3A behavioural condition
Losing in conditions the system should have handled well. A trend system losing in a strong trend is far more damning than losing in chop.
- 4A sample-size floor
No retirement decision before a stated minimum number of trades. Thirty trades is noise, whatever they did.
When edges genuinely decay
Some edges do disappear permanently, and it is worth knowing which kind you are trading. The more mechanical and easily discovered an edge is, the shorter its life tends to be.
- Simple, published, easily coded rules
- Pure arbitrage or structural quirks
- Anything a broker can automate for everyone
- Edges dependent on a specific market rule
- Edges rooted in human behaviour under stress
- Requiring discomfort — holding through drawdowns
- Requiring patience most participants lack
- Capacity-constrained, so large money cannot crowd in
Your trend system is down 16% with a worst historical drawdown of 24%, and the losses are all from choppy months while ADX has been below 18. What should you do?
Aapka system 16% neeche hai — par backtest mein bhi kabhi 24% gira tha, aur abhi ke saare nuksaan sirf side-ways mahino mein hue hain. Yeh system ka marna nahi, mausam ka badalna hai. Nuksaan ek hi tarah ke market mein ho toh regime; har tarah ke market mein dheere-dheere ho toh sach mein edge khatam.
- Regime change, edge decay and a fitted backtest look identical from inside a drawdown.
- Clustered losses point to regime; diffuse gradual decline points to genuine decay.
- Check your own backtest's worst drawdown and longest flat period before concluding anything.
- Write kill criteria in advance and treat them as binding.
- Editing a system mid-drawdown leaves you with neither the old one nor a tested new one.
Mark it done to track your progress through the curriculum.
Common questions
Short, direct answers to what people ask about this topic.
- edge decay meaning in trading
- Edge decay is a trading edge permanently weakening because enough participants have found and traded the same inefficiency. It shows up as a slow, diffuse deterioration across every market condition over years, and — the part that identifies it — performance does not recover when the favourable regime returns. Simple, published, easily coded rules decay fastest; edges rooted in how people behave under stress have lasted far longer.
- the conditions written in advance that would cause a trading system to be retired are called
- Kill criteria. Written while you are calm and treated as binding, they typically specify a drawdown limit beyond the worst seen in testing, a flat period clearly longer than the longest the backtest contained, a behavioural condition such as losing in exactly the conditions the system should handle well, and a minimum number of trades before any retirement decision is allowed. Their entire purpose is to stop the decision being taken from inside a drawdown.
- how do i know if my trading system is broken or just in a drawdown
- Check whether the losses cluster. Regime problems concentrate in one market condition — a trend system bleeding in chop while its setups still trigger normally — and lift when that condition passes, whereas genuine decay is diffuse, gradual and visible in every condition. Then compare your current state against your own backtest’s worst drawdown and longest flat period, because most people abandon a system while still comfortably inside a stretch their own testing already produced.
- how many trades before you can judge a trading strategy
- More than most people use — thirty trades tells you almost nothing whatever they did, because a system with a 42% win rate can throw an eleven-trade losing streak in the ordinary course of things. There is no single magic number: the lower the win rate and the more variable the individual outcomes, the larger the sample needed before results separate edge from noise. That is why a sample-size floor belongs in the kill criteria alongside the drawdown limit.
- is it ok to tweak a trading system during a drawdown
- Doing so leaves you with neither the old system nor a tested new one. Adding a filter, shifting a parameter or skipping the signals that look obviously bad changes what is being judged halfway through the experiment, so you can no longer tell which version the results belong to. The clean options are to run it exactly as written or to stop it — any change belongs in a separate, properly tested version with its own record.