Skip to content
❗️ This is a read-only mirror of the CRAN R package repository. MSBVAR — Markov-Switching, Bayesian, Vector Autoregression Models
C++ R Fortran C
Branch: master
Clone or download
Patrick Brandt and cran-robot version 0.9-3
Latest commit ed3f8da Nov 15, 2016
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
R version 0.9-3 Nov 15, 2016
data version 0.7-0 Apr 30, 2012
man version 0.9-2 Feb 4, 2015
src version 0.9-2 Feb 4, 2015
COPYING version 0.1 Sep 15, 2005
DESCRIPTION version 0.9-3 Nov 15, 2016
LICENSE version 0.9-3 Nov 15, 2016
MD5 version 0.9-3 Nov 15, 2016
NAMESPACE version 0.9-3 Nov 15, 2016
README version 0.9-3 Nov 15, 2016

README

Package: MSBVAR
Version: 0.9-3
Date: 2016-11-14
Title: Markov-Switching, Bayesian, Vector Autoregression Models
Author: Patrick T. Brandt <pbrandt@utdallas.edu>
Maintainer: Patrick T. Brandt <pbrandt@utdallas.edu>
Depends: R (>= 2.11.0), KernSmooth, xtable, coda, bit, mvtnorm
Description: Provides methods for estimating frequentist and
  Bayesian Vector Autoregression (VAR) models.  Functions for reduced
  form and structural VAR models are also available. Includes
  methods for the generating posterior inferences for VAR
  forecasts, impulse responses (using likelihood-based error bands),
  and forecast error decompositions.  Also includes utility functions
  for plotting forecasts and impulse responses, and generating draws
  from Wishart and singular multivariate normal densities.  Current
  version includes functionality to build models with
  Markov switching.
LazyLoad: yes
License: MIT
URL: http://www.utdallas.edu/~pbrandt/


/* Revisions and History */
2005-03-17 Initial version
2005-07-21 Revised to new site and location.  Revised forecasting
	   functions to match the Political Analysis paper by Brandt
           and Freeman.
2005-09-15 Revised documentation and code to pass checks for generics.
	   Included Israeli-Palestinian dataset for examples.
2005-10-18 Incorporates preliminary functions for Bayesian SVAR models.
2006-01-19 Updated calls to Matrix package so they do not default to
           using sparse matrices in MSBVAR functions.
2006-01-29 Revised impulse response plotting functions to plot shocks
           / impulses in the columns and responses in the rows. The
	   code was documented to work this way, but delivered an
           incorrect plot.  Also changed the irf.var() function so
	   that it used an UPPER triangular A0 matrix rather than a
           lower one (the former is consistent with the return from
           chol()).
2006-02-01 Updated plotting functions for IRFs to use correct ylims
           for response plots.
2006-02-02 Removed creation of "L" in the LU decomposition used to
           backsolve structural VAR systems in "drawA0".  The U's from
	   the lu(t(A0)) are valid -- sometimes the Matrix library
           cannot generate a well behaved L, but it is not necessary
	   for the Gibbs sampler.
2006-02-10 Changed starting value computation for structural.szbvar()
	   from a set of N(0,25) to N(0,1).
	   Added a tolerance of 1e-12 to QR decompositions in drawing
           A(0) for B-SVAR models.
2006-02-18 Version 0.1.6
	   Replaced LU decompositions with QR
           decompositions in the computation of the orthonormal basis
           of the B-SVAR model posterior sampling.
	   Removed dependency on matrix package by replacing LU
           decompositions with QR decompositions
2006-02-24 Corrected limits for plots in plot.irf.var().  Previous
           versions could truncate the plot axes incorrectly.
2006-03-02 Corrected an error in the computation of the marginal data
           density for B-SVAR models.  Optimized code for computing
           posterior fit measures for
	   posterior.fit.structural.szbvar()
2006-03-07 Corrected overflow error in some computations of the log
	   determinants in posterior.fit.structural.szbvar() by
	   changing the computation of the log determinant for a
	   diagonal matrix to the simpler sum of the diagonal
           elements or using the determinant(,logarithm=T) function.
2006-03-15 Version 0.1.9
	   1) Added option to change normalization rules for A(0) in
           sampling the structural parameters of B-SVAR models.
           Previous versions had hard coded the normalization.  New
	   version supports all of the normalization rules in
	   normalize.svar().  Documentation for this change has been
           updated as well.
2006-03-15 Version 0.1.10
	   1) Corrected matrix inversion of A(0) in
	   mc.irf.structural.szbvar() and mcmc.structural.szbvar().
	   Previous version were possibly zeroing out A(0) terms
	   incorrectly when computing IRFs.  This did NOT affect the
	   sampling of IRFs using gibbs.A0.structural.szbvar().
	   2) Corrected indexing of AR coefficients and intercepts for
           the output object in reduced.form.var().  Previous version
	   selected the wrong coefficients for the intercept and AR(1)
	   terms.
2006-03-17 Version 0.1.11
	   Changed order of arguments in gibbs.A0.structural.szbvar()
	   so that the normalization options do not break older
	   code.
2006-03-26 Added forced matrix via as.matrix for testing Ui object
	   dimensions in b2a function in hidden.R
2006-04-01 Version 0.1.12
	   Now requires "coda" package.
	   Flattened the A0 and W objects that are output by gibbs.A0
	   function to conserve memory.  Lowers the memory overhead
           for B-SVAR models by nearly 70% allowing for longer runs of
	   the Gibbs sampler.
	   Added function "A02mcmc" to convert A0 posterior draws to
           coda mcmc objects.
2006-04-06 Changed selection criteria in var.lag.specification to
           include the Hannan-Quinn measure and to minimize fit
	   criteria.
2006-04-07 Added a check on the rank of the exogenous variables to the
	   structural.szbvar() function.
2006-06-18 Implemented classes for VAR, BVAR, and B-SVAR models.
	   Class based dispatching is now available for most
	   post-estimation tasks such as impulse responses, Monte
	   Carlo sampling, plotting, etc.   Some function names have
	   changed from previous (post 0.1.1) versions.  Please be
	   aware of this.
2006-07-15 Revised documentation to acccount for new class-based
	   dispatching.
2006-09-19 Corrected incorrect handling of intercepts in
           reduced.form.var() function.  Previous version thought the
	   intercept was the first, when it should be the last
	   coefficients.
2007-06-18 C++ implementations added
2007-10-23 Bug fixes: corrected summary() methods for VAR, BVAR and
           BSVAR classes and fixed bugs in gibbs.A0(), the Gibbs
	   sampler for B-SVAR models.  Thanks to Tatevik Sekhposyan
	   for helping find this bug.
2007-10-29 Updated uc.forecast() to allow for forecasts with exogenous
	   regressors.  The previous version would crash if there were
	   exogenous regressors to be included in the forecast
	   horizon.
2007-12-06 Fixed classes / defaults for plot.mc.irf() related
	   functions.  Previous versions had hard coded values rather
	   than allowing the user to control the arguments.  Thanks
	   to Mike Colaresi for finding this bug.
2008-04-22 Updated defaults to plot.mc.irf to properly pass user
	   arguments (which may have been arbitrarily ignored in the
	   past).  Also updated the sanity.check.prior() functions
	   that are internal to the B-(S)VAR estimators.  Some of
	   these recommendations were non-sensical or required
	   exogenous variables (a very VAR taboo).
2008-10-10 Upgraded version to 0.3.2, requires R 2.7.0 or higher,
	   updated UTD address info, and copyright dates.
	   Cleaned up empty documentation sections.
2008-10-13 Implemented storing of state matrices for MS functions with
	   the "bits" package.
2008-10-19 Fixed a but in the szbvar() code for the prior degrees of
	   freedom computation.
2008-11-20 Fixed scaling of prior for the intercept (lambda4) in the
 	   szbsvar() function.  Previously is was entering the prior
	   covariance as lambda4 rather than lambda4^2
2008-12-11 Revised mc.irf.BSVAR() to use means rather than medians for
	   central tendency for IRFs.
2009-06-12 Updated documentation to function under R 2.9 and the new
           Rd parsing.  Included MSBVAR estimation (initial cut) using
	   msbvar() and related functions.  This will handle the 2
	   state MS case with random permutation sampling.  Added
	   documentation for the new functions as well.
2010-02-24 Added classing mechanism to the forecast() function.  It
	   can now handle VAR, BVAR and BSVAR models.  Related summary()
	   and plot() methods have been added for forecast().  Also
	   updated the msbvar() and gibbs.msbvar() functions to handle
	   h>2 states.  New C backend for the state-space sampling has
	   been implemented and is much faster and more robust (i.e.,
	   it will not segfault like the old one).
2010-03-29 Implemented a set of posterior fit measures (llf, logMDD,
	   etc.) for the MSBVAR models.
2010-06-10 Changed prior variance computation of szbvar() and
           szbsvar() to used OLS and estimate an intercept.  This
	   better cover the case of non-stationary data or data that
	   are not demeaned, since we are not forcing the prior
	   intercepts to zero when computing the variance of the
	   prior. Thanks to Gareth Thomas for comments on this.
2010-06-23 Added full documentation for the MSBVAR model gibbs
           sampler.  Example included for gibbs.msbvar().
2010-11-28 Updated NAMESPACE and DESCRIPTION to comply with R 2.12.0.
	   Update BCF citations in documentation.
	   Update BCFdata files to fit documentation and published
	   papers.
	   Updated version to 0.5.1.
2011-01-13 Added attribute objects to VAR, BVAR, MSBVAR models so we
           have sensible variable names to use in the plots and
	   analyses.
2011-01-27 Added plotregimeid() function to plot post-posterior
	   sampling clustered draws from the Gibbs random permutation
	   sampler. This can be used for graphic identification of the
	   Markov-switching regimes based on intercepts and variances
	   of the VAR equations.  Since there are h! orderings of the
	   regimes (and identical posterior modes) such a function
	   helps identify a single posterior mode for refined
	   sampling.
2011-01-31 Updated package to version 0.6-0 to reflect milestones in
	   MSBVAR development.  Added initial documentation for
	   plotregimeid().
2011-02-11 Began works on forecasting MSBVAR models.  This is part of
           a wholesale rewrite of how the forecasts are computed,
	   presented and summarized in the package.
2011-03-01 Corrected sample indexing for the msbvar() function.  It
	   was carrying along one extra observation that should have
	   been use for computing lags rather than part of the data.
	   Changes down-stream functions like gibbs.msbvar(), etc. for
	   compliance.
2011-03-04 Degrees of freedom corrections in the prior and the
	   posterior computations for the msbvar() models.
2011-12-07 Adjusted the underflow threshold for regime probabilities
	   from 10^{-30} to 10^{-100} to improve the accuracy of
	   regime classifications in the C++ code.
2012-01-20 Updated MSBVAR model functions to use revised compiled
	   Fortran code that is more numerically accurate.  New code
	   follows Fruhwirth-Schnatter's (2006) recommendations for MS
	   models.  Includes new code for MSBVAR models with all
	   switching parameters, posterior summaries, etc.  See code
	   and package details.
2012-04-24 Revised computations of Sigma x (X'X)^{-1} computations in
	   msbvar() to be more robust to near singularities.
2012-07-21 Version 0.7-0 ready for release, including new code for
           msbvar() modeling
2012-07-25 Version 0.7-1, corrects a bug in Fortran allocations.
2012-07-26 Updated to correct Fortran allocation error.  Change
           initialization of msvar() function.  Corrects how the state
	   space filter compiles across archs.
2014-05-12 Corrected "extern C" errors in C++ code.  Version 0.9-0
2014-06-09 *Bug fixes*

	   Corrected summary.BSVAR to include "p" (long-standing bug,
	   	   output now works).
	   New startup message code.
	   Added some gc() calls for memory cleanup after some
	   	 intensive functions.
	   Added par() reset to plotregimeid() so subsequent plot
	   	 calls have the correct layouts and expected margins.
		 Corrected startup of plotregimeid so it now produces
		 a plot rather an a user prompt for the first one.

	   *Features*

	   Added correct time series axis to plot.SS

	   Separated code to compute IRF modes and error bands from
	   the plotting.  This makes the package easier to maintain
	   and extend for other methods and plotting.

	   Added impulse response functions for the MSBVAR models via
	   plot.mc.irf().

	   Added plot.ms.irf() functions as an experiment in
	   combining IRFs across regimes in one plot.

2014-06-20 *Features*
	   regimeSummary() command added to provide summary statistics
	   for estimated MSBVAR regime probability densities and
	   durations.

	   Added forecast() example for MSBVAR models

	   *Bug fixes*

	   Added length check on sign.list in mc.irf for BSVAR models.
	   Added dimension check on prior transition matrix to MSBVAR
	   models,
2015-02-04 Updated DESCRIPTION files.
	   Added wrapper.c so that RNG calls point to unif_rand for
	   use in the FFBS.f Fortran code.  This way the RNG is all
	   seeded by R.
2016-11-14 Updates to correct use of return in hidden.R

MSBVAR Package
Copyright (C) 2005-2016 Patrick T. Brandt
Support provided by the U.S. National Science Foundation
(Grants SES-0351179, SES-0351205, SES-0540816, and SES-0921051)

Patrick T. Brandt
School of Economic, Political and Policy Sciences
University of Texas, Dallas
800 W. Campbell Rd, Green 31
Richardson, Texas 75080
http://www.utdallas.edu/~pbrandt
pbrandt@utdallas.edu
You can’t perform that action at this time.