deseats (version 1.1.0)

bwidth_confint: Bootstrapping Confidence Intervals for Locally Weighted Regression Bandwidths

Description

A stationary block bootstrap is applied to resample from a time series that was decomposed into a trend, a seasonal component and a remainder by means of data-driven local polynomial regression with automatically selected bandwidth. Bandwidth re-estimation from each bootstrapped sample results in confidence bounds for the bandwidth.

Usage

bwidth_confint(
  nonpar_model,
  blocklen = NULL,
  npaths = 1000,
  parallel = TRUE,
  num_cores = future::availableCores() - 1,
  ...
)

Value

A list with the following elements is returned.

conf

A vector with named elements that gives the original bandwidth estimate as well as the bootstrapped bounds of the 95 and 99 percent confidence intervals of the bandwidth.

bwidth_estimates

a vector with all the obtained bandwidths for the bootstrapped series.

se_bwidth

the sample standard deviation of bwidth_estimates.

Arguments

nonpar_model

the object with the nonparametric trend and seasonality estimation results returned by for example the function deseats.

blocklen

a numerical vector of length one that indicates the average block length to be drawn from the detrended series; the default is NULL, which means 8 for quarterly and 24 for monthly data; selecting a suitable expected blocklength and checking the sensitivity of the blocklength are left for the user.

npaths

a numeric vector of length one that indicates the number of bootstrap paths; the default is npaths = 1000.

parallel

a logical vector of length one that indicates whether or not to employ parallel programming for the resampling and the subsequently data-driven bandwidth estimations from the bootstrapped samples; the default is patrallel = TRUE.

num_cores

a numeric vector of length one that indicates the number of CPU cores to use for parallel programming, if parallel = TRUE; the default is num_cores = future::availableCores() - 1.

...

further arguments to pass to deseats.

Author

• Dominik Schulz (Research Assistant) (Department of Economics, Paderborn University),
  Author and Package Creator

Details

Confidence bounds for the bandwidth in local polynomial regression for identifying the trend in a trend-stationary short-memory time series are obtained via a block bootstrap, which ensures that no specific model assumptions are required for the detrended series.

This function makes use of the future parallel programming framework to ensure exactly the same results regardless of whether sequential or parallel programming, and then also regardless of the number of workers, is employed.

Examples

# \donttest{
xt <- log(EXPENDITURES)
est <- deseats(xt, set_options(order_poly = 3))
conf <- bwidth_confint(est, npaths = 200, num_cores = 2)
conf
# }