Bootstrap Estimation for Block Maxima

Performs bootstrap resampling for various block maxima estimators (mean, variance, GEV parameters, quantile, return level) using either disjoint or sliding block methods.

Usage

maxbootr(
  xx,
  est,
  block_size,
  B = 1000,
  type = "sb",
  seed = 1,
  p = NULL,
  annuity = NULL
)

Arguments

xx

A numeric vector or array containing univariate samples. For multivariate cases, samples should be arranged in rows.

est

A string specifying the estimator to apply. Must be one of "mean", "var", "gev", "quantile", or "rl".

block_size

Integer. Size of each block used in the block maxima extraction.

B

Integer. Number of bootstrap replicates to generate.

type

Type of block bootstrapping: "db" for disjoint blocks or "sb" for sliding blocks (internally approximated via circular blocks).

seed

Integer. Seed for reproducibility.

p

Optional numeric value in (0,1). Required if est = "quantile".

annuity

Optional numeric value > 1. Required if est = "rl" for return level estimation.

Value

A numeric vector with B rows for scalar estimators. If est = "gev", a matrix with B rows is returned. Each row contains 3 estimated GEV parameters (location, scale, shape).

Examples

if (requireNamespace("maxbootR", quietly = TRUE)) {
  library(maxbootR)
  set.seed(123)
  x <- rnorm(100)

  # Bootstrap mean using sliding blocks
  boot_mean <- maxbootr(x, est = "mean", block_size = 10, B = 20, type = "sb")

  # Bootstrap variance using disjoint blocks
  boot_var <- maxbootr(x, est = "var", block_size = 10, B = 20, type = "db")

  # Bootstrap 95%-quantile of block maxima using sliding blocks
  boot_q <- maxbootr(x, est = "quantile", block_size = 10, B = 20, type = "db", p = 0.95)
}