Resampling individual trades ignores clustering

Question

An analyst has 10{,}000 individual trades spread over 200 days (about 50 trades per day) and wants a confidence interval for the average PnL per trade. He bootstraps by resampling individual trades with replacement and takes the percentile interval. It is very tight. A colleague warns that trades placed on the same day are highly correlated (they share the day's market moves and the same signal state). Explain why resampling individual trades understates the uncertainty and what resampling scheme is appropriate.

Accepted Answer

The core mistake: ignoring within-cluster correlation. The trades are not 10{,}000 independent observations; they cluster by day, and trades on the same day move together. The effective sample size is far closer to the number of independent days (200) than to the number of trades (10{,}000). Resampling individual trades i.i.d. treats every trade as independent information, so it acts as if there are 10{,}000 independent data points. That massively overstates the information content and produces a standard error (and interval) that is far too small, exactly the deceptively tight interval he se…

Resampling individual trades ignores clustering

Your answer

Discussion

Learn the concepts

Related questions