mhpfilter 1.0.0

Changes visible to users

• Every function keeps its name, its arguments and its results, so existing scripts run unchanged; the results of every example of version 0.1.0 are reproduced to 1e-8. New arguments are appended with defaults that give the old behaviour (log_scale = FALSE, criterion = "approximate", fmhp = FALSE), and the frequency "monthly" (lambda = 129600) is added to the comparisons.
• Every function that takes a series or a result is an S3 generic with a .default method. A ts keeps its time and frequency in the results.
• With as_dt = FALSE the filters return lists of class c("mhp_<filter>", "mhp", "mhp_result", "list") with print(), summary(), plot(), ggplot2::autoplot() and as.data.table() methods; the printed result ends with the sources of the method.
• No startup message.

Speed

• Every filter is one pentadiagonal solve, O(T), instead of a dense one, O(T^3): mhp_filter() searches 100000 values of lambda for 200 observations in 1.3 s (6.4 s for 0.1.0), and 3000 values for 500 observations 20 times faster than 0.1.0 (vignette “Validation and speed”).

Corrections

• plot_batch() passed linewidth twice (a warning) and ignored highlight in the faceted plot.
• The help page of mhp_filter() printed the factor of the criterion as 1 + 2/(T lambda); the code used, and still uses, 1 + 2T/lambda (eq. 6 of Choudhary et al. 2014).
• mhp_batch() and batch_compare() accept series of different lengths (leading and trailing missing values).

New

• fmhp_filter() and fmhp_batch(): the fully modified HP filter of Hanif, Iqbal and Choudhary (2017), with endogenous lambda and endogenous linear or quadratic end-point weights; it reproduces the published MATLAB code (form = "rows", the default) and offers the penalised least-squares form (form = "differences") and a fixed lambda (lambda =).
• fmhp_compare(): HP, modified HP and fully modified HP filters compared by the moments of the cycles and the end-point loss.
• hp_loss(), hp_gain(), hp_weights(), hp_schedule(): the end-point loss of Bloechl (2014), the gain functions, the filter weights and the end-point schedules of any HP-type filter. The end-point losses of the HP filter in Tables 3a and 3b of Hanif et al. (2017) are reproduced, and so are the annual losses of the paper’s own scheme (quadratic weights, k = 20, alpha = 10) and the weights of its Figure 1a; the quarterly losses of its scheme are not, because the paper does not report the weights behind them.
• mhp_filter(criterion = "exact"): exact generalised cross-validation (eq. 5 of Choudhary et al. 2014) beside the published approximation; on the US series it is smallest at lambda = 1, which the help page explains.
• mhp_designs(), mhp_simulate(), fmhp_simulation(): the simulation designs of the 2014 and 2017 papers, simulated trend-plus-cycle series and the power comparison of the filters. The results of the comparison, for every design, are shipped and discussed in a vignette; they do not confirm that the fully modified filter wins in every replicate.
• mhp_interpret(), mhp_cite(), mhp_app(): a plain-language interpretation of every result, the references of every method and a built-in shiny application; mhp_theme(), mhp_save() (with WebP), mhp_notation(), mhp_reference().
• Data us_macro: quarterly US real GDP, consumption and investment, 1959 to 2009.
• Quarto vignettes in HTML and PDF; complete help pages with the formulas; tests against both MATLAB codes of the authors and against the published values.

Internal

• collapse is no longer imported; the help pages, README and NEWS are generated from roxygen and Quarto.

mhpfilter 0.1.0

Initial CRAN Release (2026-02-06)

This is the first official release of mhpfilter to CRAN.

Overview

mhpfilter implements the Modified Hodrick-Prescott (HP) Filter for decomposing macroeconomic time series into trend and cyclical components. The package automatically selects the optimal smoothing parameter λ using generalized cross-validation (GCV), based on the methodology of Choudhary, Hanif & Iqbal (2014).

Core Features

Filtering Functions

• mhp_filter() - Modified HP filter with automatic λ selection via GCV
• hp_filter() - Standard HP filter with fixed λ parameter
• mhp_compare() - Compare HP vs Modified HP for single series
• mhp_batch() - Batch processing for multiple time series
• batch_compare() - Batch comparison of HP vs Modified HP

Visualization Functions

• autoplot.mhp() - ggplot2-based visualization for mhp objects
• plot_comparison() - Visual comparison of HP vs Modified HP results
• plot_batch() - Visualization of batch processing results

Utility Functions

• get_lambda() - Extract optimal λ from results
• get_gcv() - Extract GCV value from results
• print.mhp() - Print method for mhp class objects

Technical Implementation

High Performance

• Fast C++ implementation using RcppArmadillo for matrix operations
• Optimized grid search algorithm for λ selection
• Efficient memory management for large time series
• Linear scaling for batch processing of multiple series

Modern R Integration

• Full compatibility with data.table (≥ 1.14.0)
• Integration with collapse (≥ 2.0.0) for fast statistics
• Support for kit (≥ 0.0.11) utilities
• Works seamlessly with tidyverse (≥ 2.0.0) and fastverse (≥ 0.3.0)
• ggplot2-based professional visualizations

Documentation

Comprehensive Vignettes

• Introduction - Quick start guide with practical examples
  • Basic usage and workflow
  • Comparison with standard HP filter
  • Real-world applications
• Methodology - Mathematical theory and derivations
  • Standard HP filter formulation
  • Generalized cross-validation criterion
  • Optimal λ selection procedure
  • Performance characteristics

Function Documentation

• Complete roxygen2 documentation for all functions
• Mathematical formulas properly rendered with LaTeX
• Extensive examples with commentary
• Author affiliations and ORCID identifiers

Package Website

• Professional pkgdown site with full documentation
• Download statistics and badges
• Interactive examples and tutorials
• API reference with search functionality

Methodology Reference

This implementation is based on:

Choudhary, M.A., Hanif, M.N., & Iqbal, J. (2014). “On smoothing macroeconomic time series using the modified HP filter.” Applied Economics, 46(19), 2205-2214. https://doi.org/10.1080/00036846.2014.896982

Authors and Contributors

Package Author and Maintainer

Muhammad Yaseen
Postdoctoral Research Fellow
School of Mathematical and Statistical Sciences
Clemson University

Original Methodology Authors

Javed Iqbal - State Bank of Pakistan
M. Nadim Hanif - State Bank of Pakistan

Installation

# From CRAN (recommended)
install.packages("mhpfilter")

# From GitHub (development version)
devtools::install_github("myaseen208/mhpfilter")

System Requirements

• R ≥ 4.0.0
• C++ compiler for building from source
• Operating systems: Windows, macOS, Linux

Testing and Quality Assurance

• Comprehensive test suite with testthat (≥ 3.0.0)
• Continuous integration via GitHub Actions
• R CMD check passes on all platforms
• CRAN policy compliance verified
• No NOTEs, WARNINGs, or ERRORs

License

MIT License

Bug Reports and Issues

Please report bugs at: https://github.com/myaseen208/mhpfilter/issues

Acknowledgments

• Original methodology: Choudhary, Hanif & Iqbal (2014)
• RcppArmadillo team for excellent C++ integration
• R Core Team for the R statistical computing environment
• CRAN team for review and hosting

For more information, visit: https://github.com/myaseen208/mhpfilter