How I stopped believing pretty backtests 😅
I found a rule for a screener. From August to November: +16…+35% over 3 months, with a reasonable drawdown. I was already rubbing my hands 🙌
Then I downloaded the data for May–July. The months the rule “hadn’t seen.” I ran it.
−21%. Red July, red June.
Turns out the whole “discovery” hinged on one lucky September. A classic case of overfitting to history.
Since then, my rule is simple: if you find something, test it on data you haven’t touched. If it doesn’t hold up there, it’s not a strategy—it’s a coincidence.
Measure, don’t believe.
#трейдинг #бэктест #crypto
I found a rule for a screener. From August to November: +16…+35% over 3 months, with a reasonable drawdown. I was already rubbing my hands 🙌
Then I downloaded the data for May–July. The months the rule “hadn’t seen.” I ran it.
−21%. Red July, red June.
Turns out the whole “discovery” hinged on one lucky September. A classic case of overfitting to history.
Since then, my rule is simple: if you find something, test it on data you haven’t touched. If it doesn’t hold up there, it’s not a strategy—it’s a coincidence.
Measure, don’t believe.
#трейдинг #бэктест #crypto