← Back to all sparks
S

stochvol

INFRA · APIS
Velocity0.0

Efficient Bayesian Inference for Stochastic Volatility (SV) Models

A Bayesian volatility sampler in its maintenance decade, paying for its own speed

bayesian-inferencestochastic-volatilitymcmcrcppmaintenancecran-compliance
Current state
stochvol runs MCMC for stochastic volatility models, with a C++ sampler underneath an R interface. Five years of releases in this window contain no new models: the work is compiler and dependency compatibility, CRAN check notes, and a steady trickle of corrections to the sampler itself. Its methodological milestone, the Journal of Statistical Software paper, is recorded in a 2021 tag.
Where it's heading
This is what a finished computational package looks like. The formula interface arrived at 3.1.0 and nothing has been added since; what changes is the ground underneath — RcppArmadillo major versions, UBSan checks, error-handling conventions moving from Rf_error to Rcpp::stop for correct memory management. The recurring pattern worth watching is that several releases fix real errors in the sampler's proposal distributions, found by users and by CRAN's own instrumented checks rather than by the maintainer.
Prediction
Nothing in these notes suggests new methodology. Expect the next release when RcppArmadillo or a CRAN check flavour forces one, and treat any bug report against the samplers as the more consequential event.

Recent moves

  1. 5mo ago

    Sampler crash with constant parameters fixed, plus RcppArmadillo 15

    One tag carrying three versions of notes. The substance is 3.2.7's fix for a sampler crash when every parameter was held constant, surfaced by CRAN's UBSan check; the rest is RcppArmadillo 15 compatibility and swapping Rf_error for Rcpp::stop so error paths free memory properly.

    View source ↗
  2. 1y ago

    Two CRAN check notes cleared

    A missing import and an invalid URL, contributed by an automated agent making its first commit to the repository. No behaviour change.

    View source ↗
  3. 1y ago

    Proposal variance corrected in the centered parameterisation

    A user-reported error in the fast sampler: under the centered parameterisation the proposal distribution for phi and mu did not have the variance the documentation claimed. The kind of defect that degrades sampling efficiency without ever announcing itself.

    View source ↗
  4. 2y ago

    CRAN stochvol 3.2.4

    The release note is a bare compare link. Nothing about this version is recoverable from the feed.

    View source ↗
  5. 2y ago

    Rolling-window indexing and inverse gamma prior validation fixed

    Two user-reported bugs: svsample_roll() indexed expanding windows incorrectly, and validation of the inverted gamma prior on sigma squared was wrong. Both affect results rather than ergonomics.

    View source ↗
  6. 3y ago

    Core C++ sampler routines exported for reuse

    Cholesky tridiagonal solves, forward and backward algorithms, inverse transform sampling and the mixture indicator CDF are exported, which lets other packages build on stochvol's internals instead of reimplementing them.

    View source ↗