R code for relevant functional mean changes

This repository contains the R programs for the paper Relevant Mean Changes in Functional Time Series under Weak Directional Moments by Zhuo Lin, Jiajing Sun, Wolfgang Karl Härdle, and Meiting Zhu.

The code implements adjusted-range and quadratic self-normalization and a same-block scalar Wald procedure. It covers the unknown-break simulations, nonuniform variance profiles, changing persistence, directional and orthogonal heavy tails, exact finite-grid Gaussian references, variance diagnostics, and the electricity-demand application. The empirical program downloads the public National Energy System Operator data and verifies the fixed file checksums before analysis.

Requirements

Use R 4.2 or later and install:

install.packages(c('digest','matrixStats','qrng','spacefillr','ggplot2','patchwork'))

Run

Run commands from the repository root.

Rscript run_simulations.R
Rscript run_empirical.R
Rscript make_figures.R
Rscript validate.R

run_simulations.R reproduces the primary Monte Carlo study with 2,000 data samples per configuration and is computationally intensive. run_empirical.R downloads the public source files into data/raw/ and writes calculated results to results/. make_figures.R writes the main simulation and empirical plots to figures/. Generated data, results, figures, and validation output are intentionally excluded from version control.

run_all.R executes all four steps in order.

Code map

• R/core.R: functional statistics, split estimation, block profiles, Gaussian references, and bootstrap kernels.
• R/simulation_helpers.R: data-generating processes and common simulation utilities.
• R/primary_simulations.R: unknown-break, observed-loading, and ordinary-mean experiments.
• R/size_adjusted_power.R: independent size adjustment and paired uncertainty intervals.
• R/variance_diagnostics.R: block-variance bias diagnostics.
• R/data_preparation.R: public data download, checksum verification, and curve construction.
• R/empirical_analysis.R: level/shape decomposition and calendar sensitivity analysis.
• R/figures.R: color-blind-friendly figures with line-type distinctions.
• R/validation.R: numerical checks for segment rounding, reference construction, degeneracy handling, and scale behavior.
• R/limiting_power.R: Brownian limiting-power calculations.

The data source is the NESO Historic Demand Data portal.