Skip to contents

Changelog

Source: NEWS.md

tsorder 0.1.0

  • New estimators: jirak_order() (Jirak’s refitted q-hat(5), also as order_select(method = "jirak_refit")) and the experimental "acf_robust" statistic (Ma-Genton Q_n-ratio ACF with a Monte Carlo-calibrated Bartlett plug-in studentisation; 25% breakdown against additive outliers, MA-targeted).

  • fable/tidyverts integration: TSORDER() is a fabletools model definition fitting through fit_arima(), so tsorder works inside model(), forecast(), accuracy() and friends.

  • as_tibble() on order_select() results returns the tidy decision path.

  • New vignette comparing tsorder with stats::ar() and forecast::auto.arima().

  • fit_arima() stores the series on the returned fit ($x, $series), so forecast::forecast() recovers the data instead of re-evaluating the internal call (which returned NAs or errored depending on the caller’s variable names).

  • First release: information criteria (AIC, AICc, BIC, HQ, FPE, CAT, MDL over Yule-Walker, Burg or conditional-OLS fits), Fine-Hwang threshold estimators with studentised statistics (PACF, Bartlett ACF, inverse PACF, Jirak band, robust PACF), McMillan-degree estimators (Bauer’s SVC, Hankel singular-value thresholding, sequential rank test), ARMA identification (EACF, Hannan-Rissanen with the 1983 correction pass), integration order by sequential unit-root testing, and the Q_n scale estimator - all in a dependency-free Rust core, with the high-level order_select() interface.

  • Defaults chosen by the reproducible benchmark in the repository (poly-log memory; SVC window d = 1).

  • Test suite cross-validates against stats::ar.yw (1e-10), robustbase::Qn (1e-12) and TSA::eacf (digit for digit), and replays a golden fixture shared with the Python binding.