Skip to contents

Changelog

Source: NEWS.md

exuber (development version)

New methodologies from the docs/enhancements/ research programme, each independently validated against a published number (formula-exact check, table lookup, or a direct Monte Carlo reproduction of the source paper’s own theorem) — see docs/enhancements/README.md for the full record of what was checked and how.

Volatility-robust tests

  • radf_sbz()/radf_sbz_cv()/radf_sbz_union() — Herwartz & Siedenburg’s WLS/kernel-volatility SBZ test, split 2026-08-22 into a statistic (radf_sbz()), its bootstrap critical values (radf_sbz_cv(), full datestamp()/autoplot() support), and the union-of-rejections test against classic supDF (radf_sbz_union(), renamed from the original bundled radf_sbz_cv() — see vignette("volatility-robust-radf").
  • radf_kp() — kernel-purge heteroskedasticity test.
  • radf_wb_cv(..., dist_skew = TRUE) — Hafner (2020) skewness-corrected wild bootstrap.
  • radf_sign()/radf_sign_cv() — Harvey, Leybourne & Zu (2020) sign-based sGSADF, invariant to volatility with no bootstrap needed.
  • ssu_test() — Kurozumi & Nishi (2025) stochastic explosive-coefficient test (minimum-viable subset).
  • datestamp(..., option = "svadf") — Sarkar & Wells (2026) SV-ADF asymmetric-threshold dating, folded into datestamp() rather than shipped as a separate radf_svadf() entry point. Caveat: the source is a non-peer-reviewed preprint, flagged at call time and in ?datestamp, Caveats section.

Dating and root inference

  • dating_pdc() — PDC/KS sequential sample-splitting dating, plus type = "wls" for Kurozumi & Skrobotov (2023)’s time-varying-volatility correction.
  • radf_recovery()/radf_recovery_cv() — Phillips & Shi (2014) reverse-regression crisis-origination/recovery dating. Caveat: f_c and the overall false-detection rate are exploratory pending further validation, flagged at call time and in ?radf_recovery.
  • dating_hls() — Harvey, Leybourne & Sollis (2017) SSR/BIC single-bubble dating.
  • dating_hlw() — Harvey, Leybourne & Whitehouse (2020) SSR/BIC multi-bubble two-step wrapper.
  • dating_knp() — Kejriwal, Nguyen & Perron (2025) bias-corrected dating.

Real-time monitoring

  • monitor() — Phillips & Shi (2020) training/monitoring orchestration (Family A), plus Kurozumi (2020) closed-form SADF/ GSADF_s0 boundaries and Homm & Breitung (2012)’s FLUC boundary.
  • monitor_cusum() — Homm & Breitung (2012) CUSUM monitoring, plus Astill et al. (2023)’s volatility-robust CUSUMV kernel variant and HB’s finite-sample boundary.
  • lbi_test()/monitor_lbi() — Breitung & Diegel (2025) static LBI test and its sequential mCUSUM/wCUSUM extension.

Multivariate / panel tests

Alternative paradigms

Naming

  • 12 of the functions above (cobubble_test, contagion_reg, monitor_cusum, dating_hls, dating_hlw, dating_knp, lbi_test, monitor_lbi, dating_pdc, monitor_quantile, quantile_test, ssu_test) were named radf_* in earlier development snapshots of this unreleased version; renamed before release since none of them are actually recursive-ADF-based tests. No deprecated aliases were kept, as the old names never shipped in a CRAN release.
  • radf_monitor() renamed to monitor() (via a brief intermediate monitor_radf(), never released), the flagship of the real-time monitoring family alongside monitor_cusum()/monitor_lbi()/ monitor_quantile() — deliberately carrying no radf/sadf token so it can’t be mistaken for a radf_*() variant, ADF-family internals notwithstanding — see vignette("naming-and-analysis").
  • exuber_functions() added: a queryable registry of every exported function’s family (adf, test, dating, monitor, root, regression), so “what monitoring functions exist” is an actual function call, not a naming convention to memorize.
  • radf_wb_cv2()/radf_wb_distr2() renamed to radf_wb_ps_cv()/ radf_wb_ps_distr() – the 2 suffix named nothing (just “the second wild bootstrap added”); _ps identifies it as Phillips & Shi (2020)’s wild bootstrap (fits a null AR model, resamples residuals, supports a tb training-window boundary), as opposed to radf_wb_cv()’s Harvey et al. (2016) non-parametric multiplier bootstrap, matching this file’s own internal naming (radf_wb_dgp_ps/radf_wb_ps vs. radf_wb_dgp_hlst/radf_wb_hlst) and the package-wide radf_<method>_<qualifier>_cv pattern (radf_sign_dm_cv()). Unlike the renames above, radf_wb_cv2() shipped in the 1.0.0 release (the JSS paper), so this one keeps radf_wb_cv2()/radf_wb_distr2() as deprecated aliases (.Deprecated(), warn-and-forward, see ?exuber-deprecated) rather than a clean break.
  • radf_tt_cv(), radf_sign_cv(), and radf_sign_dm_cv() now all compute badf_cv/bsadf_cv (a time-varying boundary), not just the three scalar critical values — radf_tt()/radf_sign()/ radf_sign_dm() results now work with the full summary()/ datestamp()/tidy()/autoplot() pipeline, not just summary()/ tidy(). Found first in radf_tt_cv() (a user-reported datestamp() crash on radf_sign() prompted checking all three GLS-demeaned-family functions), then confirmed the identical fix applies to the other two. Validated per function: badf_cv‘s last row is bit-identical to adf_cv (a hard identity), empirical false-alarm rate at or below nominal (radf_tt 3.3%, radf_sign 5.5%, radf_sign_dm 3.5%, vs. 5% nominal), and detection power on an identical synthetic bubble in the same range as the established radf()/radf_mc_cv() baseline (16%): radf_tt 18%, radf_sign 20%, radf_sign_dm 8% (lower power a known, expected property of the sign-based tests’ heteroskedasticity invariance, not a validation concern) — see vignette("naming-and-analysis").

Other

  • rootstamp() — confidence interval and doubling time on the explosive root, via S3 dispatch: the default method fits a single sub-sample, the radf_obj method runs every datestamp() episode at once (previously three separate functions – explosive_root(), root_ci(), root_ci_datestamp() – consolidated before release).

exuber 1.1.0

CRAN release: 2025-08-31

  • Fixed targets not in the package itself nor in the base packages to use package anchors, i.e., use

exuber 1.0.1

CRAN release: 2023-02-12

Maintenance release to accommodate breaking changes in dplyr 1.1.0.

exuber 1.0.0

CRAN release: 2022-08-19

This first major release accompanies the publication of an article in the Journal of Statistical Software:

Vasilopoulos, K., Pavlidis, E., & Martínez-García, E. (2022). exuber: Recursive Right-Tailed Unit Root Testing with R. Journal of Statistical Software, 103(1), 1–26. https://doi.org/10.18637/jss.v103.i10

augment method for radf_obj and radf_cv

  • New arg trunc

  • Fixed inconsistencies among functions.

  • Now radf stores the data that are later can be accessed with mat+

  • Advanced features on datestamping: New columns that indicate:

    • Signal
    • Peak
    • Ongoing
    • Nonrejected

  • New datestamping procedure rev_radf etc.

  • New bootstrap procedure radf_wb_cv2 and radf_wb_distr2

  • New coloring convention for plotting ds and obj classes

Bug Fixes

  • Now autoplot can include periods that have an ongoing bubble

exuber 0.4.2

CRAN release: 2020-12-18

  • Include printing methods for radf_obj and radf_cv.
  • Removed unused class definitions.
  • Using progress package for progress_bar.

exuber 0.4.1

CRAN release: 2020-05-12

Maintenance release for compatibility with dplyr v1.0.0.

exuber 0.4.0

CRAN release: 2020-05-04

Design

We have the following design in mind for future scalability. If you want make inference about radf models, then the estimation can be achieved with radf() function and return an object of class radf_obj, and the critical values can be achieved with radf_*_cv() and return an object of class radf_cv.

Breaking changes

exuberdata

  • We created a new package called exuberdata that accommodates critical values for up to 2000 observations. Critical values can be examined with exuberdata::radf_crit2. The package is created through drat R archive Template, and can be easily installed with install.packages('exuberdata', repos = 'https://kvasilopoulos.github.io/drat/', type = 'source') or through install_exuberdata wrapper function that is provided in exuber.

Improvements

  • The package zoo has been used as a dependency to import the method index(). We made the decision to remove zoo and create a new method index() internally.

exuber 0.3.0

CRAN release: 2019-07-15

Breaking changes

  • Changed opt_bsadf = conservative for the simulated critical values (crit), also reduced the size of the crit from 700 to 600 due to package size restrictions.
  • sim_dgp1() and sim_dgp2() have been renamed to sim_psy1() and sim_psy2() to better describe the origination of the dgp.
  • sim_dgp1() and sim_dgp2() have been soft-deprecated.
  • autoplot_radf() arranges automatically multiple graphs, to return to previous behavior we included the optional argument arrange which is set to TRUE by default.

Three new functions have been added to simulate empirical distributions for:

  • mc_dist(): Monte Carlo
  • wb_dist(): Wild Bootstrap
  • sb_dist(): Sieve Bootstrap

and a function that can calculate the p-values calc_pvalue() given the above distributions as argument.

Also methods tidy() and autoplot() have been added to turn the object into a tidy tibble and draw a particular plot with ggplot2, respectively.

New features

  • tidy() methods for objects of class radf, cv.
  • augment() methods for objects of class radf and cv.
  • augment_join() to combine object radf and cv into a single data.frame.
  • glance() method for objects of class radf.

Improvements

  • New printing output for the functions summary(), diagnostics() and datestamp().
  • New improved progressbar with more succinct printing for wb_cv()
  • seed argument to functions that are using rng. Also the option to declare a global seed for reproducibility with the option(exuber.global_seed = ###)

Bug Fixes

  • sb_cv() and wb_cv()now can parse data that contain a date-column. Similarly, to what radf() is doing.

exuber 0.2.1.9000

  • Website development

exuber 0.2.1

CRAN release: 2019-03-01

  • Changed DESCRIPTION to include sb_cv reference.
  • Renamed boolean to dummy from datestamp and diagnostics.
  • datestamp dummy is now an attribute.

exuber 0.2.0

CRAN release: 2019-02-04

Options

Some of the arguments in the functions were included as options, you can set the package options with e.g. options(exuber.show_progress = TRUE).

  • parallel option boolean, allows for parallel in critical values computation.
  • ncores option numeric, sets the number of cores, defaults to max - 1.
  • show_progress option boolean, allows you to disable the progress bar, defaults to TRUE.

New features

  • Panel estimation in radf()
  • Added sb_cv() function: Panel Sieve Bootstrapped critical values
  • Default critical values are supplied directly into summary(), diagnostics, datestamp() and autoplot(), without having to specify argument cv. The critical values have been simulated from mc_cv() function and stored as data. Custom critical values should be provided by the user with the option cv.
  • Added ggarrange() function, that can arrange a list of ggplot objects into a single grob.
  • Added fortify to arrange a data.frame from radf() function.

Improvements

  • Parallel and ncores arguments are now set as options.
  • Ability to remove progressbar from package options.
  • radf() can parse date from ts objects.
  • report() has been renamed into summary().
  • plot() has been renamed into autoplot().
  • plot() and report() are soft deprecated.

Bug Fixes

  • Progressbar appears in the beginning of the iteration
  • Plotting date now works without having to to include any additional plotting option