NEWS
BGVAR 2.7.0
- Fixed FEVD handling of explicitly supplied rotations and empty variable selections; historical decomposition now reports its identification scheme correctly.
- Corrected vcov() posterior quantiles by transforming covariance matrices for each posterior draw before summarizing. Fixed repeated logLik() calls and quantile-specific likelihood summaries; aligned likelihood and DIC trends with estimation.
- Fixed country-column matching in matrix inputs, unnamed data-list error messages, numeric Excel sheet exclusions, and conditional-forecast dimension validation.
- Fixed plotting with unequal lag orders, single forecast quantiles, and histories shorter than the plotting window.
- Fixed deprecated Minnesota shrinkage aliases and SV variance hyperparameter names (Bsigma_sv is accepted as an alias for Bsigma; explicitly supplied canonical names take precedence). Implemented documented all/half core presets with core-count validation.
- Fixed shock-information defaults and validation for signs, identification schemes, and row counts; added regression tests for these audit findings.
- Improved estimation performance using weighted cross-products, cached constant-variance cross-products, and precision Cholesky solves. Covariance summaries now use bounded buffers and exploit symmetry without changing median definitions; disabled Normal-Gamma diagnostics no longer inherit SSVS storage dimensions.
- Fixed a covariance-coefficient row assignment in the C++ sampler that failed for models with more than two equations; added direct Rcpp regression tests for three and four equations under all four priors.
- Corrected cumulative Normal-Gamma shrinkage updates in both samplers: full conditionals now include all affected lag blocks, exclude the sampled factor from the rate product, use actual coefficient counts, and start from positive factors.
- Corrected Minnesota shrinkage acceptance targets, foreign-shrinkage proposals, deterministic-row variances, and auxiliary AR scaling with higher or unequal lag orders.
- Corrected Horseshoe auxiliary rates, C++ Gamma shapes, and shrinkage around nonzero prior means. SSVS component probabilities now use stable log odds, including probability endpoints.
- Added conditional-distribution and R/C++ agreement regression tests for these sampler corrections, including constant-variance storage.
- Change of mailing address of maintainer due to change of affiliation: [email protected]
- Added testthat regression tests for bgvar() inputs and numerical boundaries; Minnesota, SSVS, Normal-Gamma, and Horseshoe priors in both the C++ and R samplers; impulse responses, FEVD, historical decomposition, prediction and forecast evaluation; higher lag orders and trends; and eigenvalue calculations and stability trimming.
- Fixed input validation for missing values in W and Ex: bgvar() now reports the intended validation messages instead of a generic missing-value error.
- fixed issue with Calls to Rf_error() in Rcpp
- Fixed reversed SSVS mixture weights for p_i and q_ij. These parameters now represent prior inclusion probabilities, as documented. Results using the default values of 0.5 are unaffected. To reproduce the previous prior specification with nondefault values, replace each probability by its complement (1-p_i and 1-q_ij).
BGVAR 2.6.0
- added new argument in function 'bgvar', within the 'expert' settings of 'OE.weights': OE.exo
- this new argument manipulates the 'getweights' function to allow other entities exogenous variables to influence other entities as well
- bug fix regarding estimation with truly exogenous variables
- adding of some checks regarding truly exogenous variables
BGVAR 2.5.9 (2025-09-22)
- updated documentation regarding 'prmean'
- fixed bug in .getweights
- fixed bug in irfs
- fixed bug regarding weakly exogenous variables
BGVAR 2.5.8 (2024-09-30)
- update mail adress of maintainer
- fix for hyperparameters of Minnesota prior
BGVAR 2.5.6 (2024-07-03)
- some minor bug fixes (numerical issues when estimating single-VARs)
BGVAR 2.5.5 (2023-12-13)
BGVAR 2.5.3 (2023-12-08)
- small change in estimation procedure: countries can now have asymmetric effects, i.e., they are important for other countries
- but do have own spillovers (no weakly exogenous variables for particular countries)
- add JSS paper to vignette
- update maintainer information (mail address)
- change 'itemize' to 'describe' in markdown help files.
- irf command returns plotted data in list format.
- fixed a bug in .getweights().
BGVAR 2.5.2 (2022-10-26)
BGVAR 2.5.1 (2022-09-02)
BGVAR 2.5.0 (2022-05-02)
- bugfix in pesaranData (rowSums did not sum up to one)
- final version for JSS publication
BGVAR 2.4.6 (2022-04-12)
- improved memory usage and garbage collection in irf function
BGVAR 2.4.4 (2022-04-01)
- Implementation of the corrigendum of Carriero, Chan, Clark, and Marcellino (2021, Journal of Econometrics)
- added the horseshoe prior (Makalic and Schmidt, 2015, IEEE Signal Processing Letters)
- better memory usage
- new features:
- plag can now be differently specified for endogenous and weakly exogenous variables
- bugfix in Ex argument in 'bgvar'
- bugfix in summary command
- bugfix with Wex.restr
- bugfix in monthlyData
- bugfix in GFEVD
- bugfix in summary
BGVAR 2.4.2 (2021-11-04)
- bugfix in fevd; added additional check
BGVAR 2.4.1 (2021-11-04)
- bugfix in Rcpp code (occured when using SSVS prior and save.shrink.store=TRUE)
- bugfixes in vignette
- update of monthlyData dataset
BGVAR 2.4.0 (2021-10-06)
- bugfix in plotting fitted values
- per default no save of containers of shrinkage parameter - decreases significantly memory usage
- restructure of example data sets
- change print method of rmse / lps
BGVAR 2.3.1 (2021-09-10)
- small adaptions in help files.
- changed some functions to S3 method
- changed default options of bgvar()
- renaming of functions
- bugfix(es) with external variables
BGVAR 2.3.0 (2021-08-10)
- added reader function for Excel
- updated vignette with own section for reading data from excel
BGVAR 2.2.2
- small fix in BVAR_linear.cpp
BGVAR 2.2.1
- small update to clear errors in CRAN Checks
- added output of 'bgvar' function. Now directly shows the number of stable draws when using the function.
- added a missing reference in vignette
BGVAR 2.2.0 (2021-05-03)
- complete redesign of 'irf' function
- computation of irf now in C++
- makes use of parallel computing via RcppParallel
- additional argument with 'shockinfo'
- two new functions: get_shockinfo() and add_shockinfo() as helper function for designing shockinfo argument in irf computation
- introduction of 'expert' settings to reduce number of arguments in main functions (bgvar and irf)
- conditional predictions now possible within predict function
- synchronization of parameter names with accompanied paper
- minor bug fixes
BGVAR 2.1.4 (2020-11-14)
- updated backends of new 'stochvol' package
- added dataset 'eerDatasmall' to check functionality of package faster
- minor bug fixes
BGVAR 2.1.3 (2020-09-30)
BGVAR 2.1.2 (2020-09-14)
- adapted various roxygen articles
- summary output needs now less space
- citations are adapted with Working paper
BGVAR 2.1.1 (2020-09-07)
- changed argument names of 'bgvar' to be more similar to existing Bayesian packages
- 'saves' is now 'draws'
- 'burns' is now 'burnin'
- changed argument name of 'predict' to be more similar to existing Bayesian packages
- 'fhorz' is now 'n.ahead'
- added new class 'bgvar.summary' with print method
- loglikelihood is saved in class 'logLik'
- print methods have now no own helpfile (not necessary)
- deleted some printing to the console and use now consistently cat()
- rename of some functions
- changes related to multithreading
- not dependent on foreach and doparallal any more
- parLapply for Windows platforms
- mclapply for non-Windows platforms
- user can also specify own apply function
- adapted class 'logLik.bgvar' to class 'logLik'
- added 'print' method for classes 'bgvar.hd', 'bgvar.irf', 'bgvar.fevd'
- changed argument of 'nhor' to 'n.ahead' of function irf() and gfevd()
BGVAR 2.0.1 (2020-06-24)
- bug fixes in Rcpp code (wrong overloading of std::pow)
BGVAR 2.0
- added truly exogenous variables to the estimation function
- verbosity option now available for all functions
- user par settings are not affected by BGVAR any more
- First CRAN release version.