Changelog

Source: NEWS.md

bsvarPost 1.1.0

• Support bsvarSIGNs 3.0 log-weight native helpers while retaining compatibility with 2.0. Narrow upstream namespace imports to avoid forecast-class name collisions with bsvars 4.0.

Correct posterior aggregation and inference

• Grouped historical decompositions and event distributions now aggregate within each posterior draw before computing medians and credible intervals. Combining multiple shocks, including an Other category, requires draw-level input. Ungrouped summary-table plots remain supported.
• The baseline uses all shock contributions before display filtering. Totals are computed from draw-level sums; marginal component medians need not add up to the median total.
• Event and cumulative plots use the same event-based contributor ranking. Array event summaries select and aggregate the window before constructing a long table.
• Model comparisons reject duplicate explicit names and generate collision-free missing names. Tuple grouping keeps dotted and delimiter-containing labels distinct.
• Forecast horizons are numeric, so periods beyond nine plot in temporal order. Nonnumeric original labels are retained as horizon_label.
• Pairwise draw-level hypotheses now align RHS labels with each draw. Structural restrictions accept model variable labels, and one-period narrative diagnostics request a valid upstream computation horizon.
• Hypothesis results render correctly through automatic plotting, report bundles and publication helpers. Unsupported draw-level hypothesis plots fail immediately; report bundles retain their table with a NULL plot.
• Simultaneous bands use an empirical order statistic covering at least the requested fraction of stored draws. achieved_coverage reports empirical coverage, which can exceed the requested probability when draws tie. This does not guarantee exact coverage of the underlying posterior distribution. Floating-point boundaries are adjusted outward when necessary to retain the promised empirical coverage.

Provenance, validation and compatibility

• Posterior tables expose analysis type, interval probability and method, normalisation, known units, current upstream software versions and draw count as ordinary columns. Unknown values are typed NA. See help("posterior_provenance") for column definitions.
• draw_id preserves original posterior labels alongside positional draw. Representative summaries add selected_draw_id alongside draw_index.
• Default reports and CSV exports include provenance columns. Existing estimate columns, function names and object-type attributes are retained. Code relying on an exact column list should select columns explicitly or use compact reports. APRScenario adapters retain their external column schema.
• Restriction constructors and native-bridge operations validate scalar parameters, effective-sample bounds and posterior dimensions. Native calls verify registered argument counts and report actionable compatibility errors.
• R 4.1 remains supported. Minimum dependencies are bsvars 3.2, bsvarSIGNs 2.0, ggplot2 3.4.0 and vctrs 0.5.0.
• CI exercises minimum R, absent/present optional integrations and runtime coverage. Static release QA no longer treats function-name mentions in tests as proof of coverage.

bsvarPost 1.0.0

Initial CRAN release.

Breaking changes and deprecations

• The variable and shock (singular) arguments are deprecated in favour of variables and shocks (plural) throughout the package. The singular forms still work and produce a warning; they will be removed in a future version.

Core posterior extraction

• tidy_irf() extracts impulse responses from PosteriorBSVAR, PosteriorBSVARSV, PosteriorBSVARMIX, PosteriorBSVARMSH, PosteriorBSVART, and PosteriorBSVARSIGN objects into tidy tibbles with credible-interval columns.
• tidy_cdm() extracts cumulative dynamic multipliers for all supported posterior types.
• tidy_fevd() extracts forecast error variance decompositions (reported on the 0–100 percentage scale used by ‘bsvars’).
• tidy_shocks() extracts structural shocks.
• tidy_hd() extracts historical decompositions.
• tidy_forecast() extracts out-of-sample forecasts.
• cdm() estimates cumulative dynamic multipliers directly from any supported posterior object, bridging ‘bsvars’ and ‘bsvarSIGNs’ in a single call.

Hypothesis testing and inference

• hypothesis_irf() and hypothesis_cdm() compute posterior probabilities for user-specified inequality constraints on impulse responses and cumulative dynamic multipliers.
• joint_hypothesis_irf() and joint_hypothesis_cdm() extend hypothesis testing to joint constraints across multiple variable–shock–horizon combinations.
• simultaneous_irf() and simultaneous_cdm() compute simultaneous credible bands over selected response paths.

Representative model selection

• representative_irf() and representative_cdm() identify the posterior draw closest to the pointwise median across all impulse-response or CDM paths.
• most_likely_admissible_irf() and most_likely_admissible_cdm() select the draw with the highest kernel score among sign-restriction-admissible draws.

Response summaries

• peak_response() locates the horizon of maximum absolute impulse response and its posterior distribution.
• duration_response() summarises the number of horizons for which a response remains above a threshold.
• half_life_response() computes the posterior distribution of the half-life of a response decay.
• time_to_threshold() computes the posterior distribution of the first horizon at which a response crosses a user-specified threshold.

Model comparison

• compare_irf(), compare_cdm(), compare_fevd() compare tidy extractions across two models side-by-side.
• compare_peak_response(), compare_duration_response(), compare_half_life_response(), compare_time_to_threshold() compare response-shape summaries across models.
• compare_restrictions() compares restriction-audit results across models.
• compare_hd_event() compares event-window historical decomposition summaries across models.
• compare_acceptance_diagnostics() compares sign-restriction acceptance diagnostics across models.

Restriction and acceptance auditing

• restriction_audit() evaluates whether posterior draws satisfy a set of structural sign or narrative restrictions.
• magnitude_audit() checks whether impulse responses satisfy magnitude-based restrictions with explicit variable and shock arguments.
• acceptance_diagnostics() reports draw-level admissibility statistics for bsvarSIGNs posteriors, with summary() and print() methods.

Event-study helpers

• tidy_hd_event() summarises historical decomposition contributions within a user-specified event window.
• shock_ranking() ranks shocks by their contribution to a target variable within an event window.

Plotting and reporting

• autoplot() support for all tidy extractors (PosteriorIR, PosteriorCDM, PosteriorFEVD, PosteriorSHOCKS, PosteriorHD, PosteriorFORECAST), all comparison outputs, simultaneous bands, and joint hypothesis objects.
• theme_bsvarpost() provides a minimal publication-ready ggplot2 theme.
• style_bsvar_plot() applies consistent axis, legend, and colour styling to any bsvarPost ggplot2 output.
• publish_bsvar_plot() applies family-aware publication templates across comparison, representative, diagnostics, event-study, and joint-inference outputs.
• as_gt(), as_flextable(), and as_kable() convert tidy bsvarPost tables to formatted gt, flextable, and knitr::kable outputs for publication documents.
• write_bsvar_csv() writes any tidy bsvarPost table to CSV.
• report_bundle() produces a paired plot-and-table bundle suitable for publication workflows, with support for representative-response, acceptance-diagnostics, simultaneous-band, joint-hypothesis, and event-study objects.
• report_table() generates compact or default publication-oriented table layouts with family-specific labels and subtitles.