how do you know when a mechanical system is actually broken vs just in drawdown

[dreamchaser]

running a mechanical system and im in the worst drawdown ive seen with it. the agonising question every systematic trader faces: is the edge genuinely gone, or is this a normal drawdown i need to sit through? turn it off too early and i miss the recovery, leave it on too long and i bleed a dead system. how do you actually make this call with discipline rather than fear?

[carterw]

you decide it in advance, not in the drawdown. before going live you should know your systems historical maximum drawdown and losing-streak length from backtesting. if the current drawdown is still within what the backtest showed as normal, you sit through it, full stop. you only consider it broken when it exceeds the worst the system ever produced historically, or when the market structure it relies on has demonstrably changed. deciding in the pain is fear, deciding from the pre-set threshold is discipline.

[jeremyy]

pre-defining the kill threshold from backtested worst-case is the only way to do this without emotion. if your max historical drawdown was, say, x, then a drawdown approaching but not exceeding x is expected and you hold. exceeding it meaningfully is your signal to pause and investigate. the number has to be set when youre calm, because in the drawdown every instinct screams to either panic-quit or stubbornly hope, and both are wrong without the pre-set line.

[cloudyvision]

the backtested max drawdown is also the number that gets exceeded the moment you go live, because live always finds a worse stretch than the curve-fit history showed. trusting the backtest drawdown as your safety line is trusting the exact thing that overfit you into the system in the first place.

[bluedreams]

the troll lands a real one, in-sample max drawdown is usually optimistic and live often exceeds it, which is why a robust process pads it. i use roughly one and a half times the backtested max drawdown as the real kill line, on the assumption live will be worse than history. and i separate the two questions explicitly: has the drawdown exceeded my padded threshold, and separately, has the market mechanism the system exploits structurally changed. either one trips the review. the padding addresses exactly the overfit-optimism the troll described.

[mellowday]

how do you tell if the market mechanism changed versus just bad luck, that part sounds subjective?

[william_h]

its the hardest part and not fully objective, but there are concrete tells. if your system exploits, say, a volatility breakout pattern, check whether that pattern is still occurring and simply not paying, versus the pattern no longer forming at all because volatility regime changed. setups occurring but failing leans bad luck or normal drawdown. setups no longer occurring, or the conditions that created them structurally gone, leans real mechanism change. look at whether the opportunity still exists, not just whether its currently profitable.

[tylerg]

the setups-still-occurring check is the most practical objective handle on an inherently fuzzy question. combined with the padded drawdown threshold, you get a two-part rule that resists both panic and denial: hold while drawdown is within the padded historical range and the setups still appear, review seriously when either breaks. its not perfect but its a disciplined framework rather than a gut feeling in the middle of pain.

[Nicholas]

pre-set a padded kill line, check if setups still occur. decide calm, not in the pain.

[samuel.j]

this is exactly the spiral i was in, panic-wanting to kill the system mid-drawdown realising my current drawdown is actually still inside the backtested range and the setups are still firing. holding, per the framework. thank you, i was about to abandon a system on pure fear.