Skip to contents

Simulate synthetic series with Wavelet Autoregressive Modeling (WARM)

Source: R/wavelet_warm.R
simulate_warm.Rd

Generates n_sim synthetic realizations of length n.

Two operating modes are supported.

Component mode (n >= bypass_n) - Each wavelet component is centered. - A stationary non seasonal ARMA(p,q) model is fit once per component using a grid search over p in 0..max_p and q in 0..max_q with AIC selection. - Simulations are generated from the selected ARMA model using a fixed simulator (no S3 simulate dispatch). - Component means are re added and components are summed. - If fitting fails for a component, a block bootstrap fallback is used for that component only.

Bypass mode (n < bypass_n) - Component simulation is skipped. - A stationary non seasonal ARMA(p,q) model is fit to the original series and simulated directly. - If fitting fails, a block bootstrap fallback is used.

Usage

simulate_warm(
  components = NULL,
  n = NULL,
  n_sim = 1000,
  seed = NULL,
  series_obs = NULL,
  bypass_n = 15L,
  match_variance = TRUE,
  var_tol = 0.1,
  check_diagnostics = FALSE,
  verbose = TRUE
)

Arguments

components

Matrix, data.frame, or list of numeric vectors. Wavelet components. In component mode, each component must have length n. In bypass mode, components are only used to reconstruct the original series if series_obs is not provided.

n

Integer. Length of each simulated series.

n_sim

Integer. Number of realizations to generate. Default is 1000.

seed

Optional integer. Base RNG seed for reproducibility.

series_obs

Optional numeric vector. Observed original series. Provide this in bypass mode to avoid reconstructing the series from components.

bypass_n

Integer. Threshold for bypass mode. Default is 25.

match_variance

Logical. If TRUE, rescale simulated outputs to match observed standard deviation when relative mismatch exceeds var_tol.

var_tol

Numeric in [0, 1]. Relative standard deviation tolerance.

check_diagnostics

Logical. If TRUE, runs a Ljung Box test on ARMA residuals and warns when residual autocorrelation remains.

verbose

Logical. If TRUE, prints informational messages.

Value

Numeric matrix with dimension n by n_sim. Each column is a simulated realization.

Details

ARMA configuration - Non seasonal: seasonal terms are not considered. - Stationary: fits enforce stationarity constraints via parameter transforms. - No drift and no mean: include.mean is FALSE. The code centers series before fitting and re adds the observed mean after simulation.

Fallback logic - Fallback is applied per component in component mode. Components that fit successfully use ARMA simulation; failed components use block bootstrap. - In bypass mode, fallback is applied to the whole series.