HTDV 0.2.0

Validation release. The framework now ships with the executable evidence of its calibration claims.

New: validation evidence shipped with the package

• htdv_sim_summary: aggregated output of the pre-registered factorial Monte Carlo study (1024 cells × 500 replications × 3 inferential layers, 31 hours wall-clock on 16 cores, sign-corrected coverage).
• htdv_empirical_benchmarks: three-dataset external validation against Stock-Watson (2007), Campbell-Shiller (1998), and the iid Welch baseline on US-Canada 10y yields.
• New vignette HTDV-validation consolidates both validations with the per-layer reading and the persistence-ladder cross-cut.

New: simulation infrastructure

• htdv_simstudy(): end-to-end factorial Monte Carlo runner with cell-level parallelism via parallel::mclapply and atomic per-cell caching that survives worker failures and supports resume.
• htdv_simstudy_summary(): aggregate per-cell results.
• htdv_simstudy_warnings(): flag cells where the Bayesian diagnostic-pass rate falls below a user-set threshold (typically high-persistence × small-sample corners where the AR(1) likelihood approaches its limit of identification).
• New Stan model simstudy_two_sample.stan for two-sample AR(1) inference with shared autocorrelation.

Bug fixes

• Sign convention in the simulation’s HAR-Wald and stationary-bootstrap layers aligned with the Stan parameterization (delta = E[X_2] - E[X_1]). Earlier output had spuriously low HAR and bootstrap coverage for delta > 0 cells; the recover_coverage.R helper of the companion paper repository regenerates the corrected aggregates from the raw RDS.
• Documentation references for Betancourt (2016) corrected to use <doi:10.48550/arXiv.1604.00695> (CRAN-mandated DOI form).
• Package metadata: GitHub URL points to IsadoreNabi/HTDV; ORCID, email, and BugReports URL aligned in DESCRIPTION and CITATION.

HTDV 0.1.0

Initial public release.

• Hierarchical Bayesian fit via HMC (htdv_fit) with TAC, WSC, MPC, Whittle, and Composite likelihood backends.
• Berger-robust posterior envelope (htdv_envelope).
• Long-run variance with Andrews bandwidth (htdv_lrv).
• Fixed-b HAR Wald test (htdv_fixedb).
• Block bootstrap with Patton-Politis-White automatic block length (htdv_boot).
• Adaptive conformal inference for dependent data (htdv_conformal).
• ROPE decision, bridge-sampling Bayes factor, WAIC, leave-future-out CV, and predictive stacking (htdv_rope, htdv_bf, htdv_waic_lfo, htdv_stack).
• MCMC diagnostics and posterior-predictive checks on dependence statistics (htdv_diagnostics, htdv_ppc).
• Explicit numerical constants for TAC/WSC/MPC metric equivalence (htdv_equivalence_constants).