Changes in version 1.2.0                        

New features

  - Add S3 method plot.barma() to generate diagnostic plots for fitted
    barma objects. It displays a four-panel grid by default: observed
    vs. fitted values, residuals over time, residual ACF, and a residual
    histogram with a density overlay.
  - Add penalty argument to barma(), enabling ridge-penalized
    conditional maximum likelihood estimation as proposed by
    Cribari-Neto, Costa and Fonseca (2025, BJPS). When penalty = TRUE,
    the penalized log-likelihood is optimized and the penalized FIM is
    returned.
  - Add internal helper utils_internal.R with .barma_ridge_lambda() to
    centralize the ridge penalty scalar lambda_n = 1/(n-a)^0.9, shared
    across loglik_barma(), score_vector_barma(), and fim_barma().
  - Add utils_optimizer.R with internal helpers make_par_indices() and
    run_optimizer(), centralizing parameter indexing and optimization
    logic previously embedded in barma().

Improvements

  - Forward penalty argument through run_optimizer() to loglik_barma()
    and score_vector_barma(), ensuring the penalized likelihood is used
    consistently during optimization when requested.
  - Forward penalty argument to fim_barma() in barma(), so the returned
    Fisher Information Matrix reflects penalization when penalty = TRUE.

Documentation

  - Add two new vignettes: one detailing the numerical stability and
    ridge penalization approach (Cribari-Neto, Costa, and
    Fonseca, 2025), and another demonstrating applied
    hydro-environmental time series modeling (Costa, Cribari-Neto, and
    Scher, 2024).
  - Add Cribari-Neto, Costa and Fonseca (2025, BJPS,
    DOI: 10.1214/25-BJPS645) to @references in barma(), reflecting the
    new penalty argument.
  - Update @param optimization in barma() to use \describe instead of
    \itemize for correct rendering of named list fields in HTML help
    pages.

                 Changes in version 1.1.0 (2026-04-15)                  

Bug fixes

  - Fixed incorrect computation of ystar in score_vector_barma();
    previously the estimates were only correct for the logit link
    function.

New features

  - Export loglik_barma(), score_vector_barma(), and fim_barma() to
    allow advanced user access to model internals.

Breaking changes

  - Deprecate ar = NA and ma = NA for absent AR/MA components. Use ar =
    integer(0) and ma = integer(0) or omit the arguments. Backward
    compatibility is preserved in barma() with a deprecation warning.

Improvements

  - Simplify internal handling of absent AR/MA components using
    vectorized operations (drop(crossprod())) in loglik_barma(),
    score_vector_barma(), and fim_barma(), replacing conditional if/else
    branching.
  - Standardize parameter order across core functions to (alpha, varphi,
    theta, beta, phi).
  - Set default link to "logit" in make_link_structure() and all
    exported functions; improve error messages for unsupported links.
  - Rename internal helper ._get_phi_start to .get_phi_start for
    consistency with R conventions.
  - Improve documentation of precision parameter initialization in
    start_values().

                 Changes in version 1.0.1 (2026-03-29)                  

Documentation

  - Update README for CRAN release.
  - Add CRAN status badge.
  - Use install.packages() as primary installation method.
  - Update citation to version 1.0.1 with proper BibTeX key.
  - Exclude generated README.html from version control.

                        Changes in version 1.0.0                        

  - Initial CRAN submission.