Resamples contiguous, disjoint blocks so that dependence WITHIN a
block survives the resampling.
Usage
morie_boot_nonoverlap_block(
x,
block_len = NULL,
stat = NULL,
B = 500,
seed = 0
)
Arguments
- x
series.
- block_len
block length; defaults to n^(1/3).
- stat
function of a numeric vector; defaults to the mean.
- B
replicates.
- seed
integer seed. R's generator is not numpy's, so replicate
values differ across languages; the estimate and the block geometry
do not.
Value
list with estimate, se, ci,
replicates, block_len, n_blocks.
Details
The iid bootstrap destroys serial dependence entirely and therefore
understates the standard error of anything computed on a dependent
series – often by a large factor. Block length is the whole design
choice: too short and the dependence is broken anyway, too long and
there are too few blocks to resample. Non-overlapping blocks give
only n / block_len resampling units; the moving-block variant
is more efficient but its blocks are correlated.
References
Carlstein, E. (1986). The use of subseries values for
estimating the variance of a general statistic from a stationary
sequence. Annals of Statistics, 14(3), 1171-1179.
