Questions / Quant research, Research decisions
Resample the dependence structure
Daily residuals show material serial dependence. Which bootstrap is more appropriate to investigate than independently resampling individual days?
Answer it in the app It's free. You get hints, a second try if you miss, and a follow-up question.
Show the answer and worked solution
Answer: C. Resample contiguous time blocks, examining the block-length assumption.
- Serial dependence links nearby observations, so individual-day iid resampling changes the data-generating structure.
- Resampling contiguous blocks can preserve dependence within each sampled segment.
- Block size and other time-series assumptions require sensitivity checks; blocks are not an automatic guarantee of valid inference.
Interview shortcut. Serial dependence: resample blocks, not days.
Takeaway. Choose resampling units that retain the dependence relevant to the statistic.
Why the other choices fail
- Resample each residual independently within its calendar month. Month strata preserve some seasonality but independently resampling days within them still discards local serial dependence.
- Resample individual days after randomly permuting their temporal order. Random permutation destroys the serial structure before resampling. It does not preserve the dependence needed for inference.
- Resample individual days and widen the interval to compensate. Weekly blocks are a block bootstrap; resampling days ignores dependence.
In the app this has a follow-up where one thing about the setup changes. Try the follow-up.