Movatterモバイル変換

• new functions coefB(), coefA(), coefPI() returning the A, B and PIcoefs for VECMs
• new function VECM.sim()
• new function TVECM.boot()
• new method getTh.lstar(), getTh.default()
• new method vcov.setar()
• new method coef.setar(), coef.lstar()
• new method logLik.cajo.test()
• extend argument in lineVar/VECM: allow lag=0
• TVECM.sim() extend for r>1, new argument seed
• VECM(): extend ML estimation of restricted model
• lstar(): change coefficient naming, consistency with setar()
• minor: robustify OLS computations using lm.fit() in SETAR(), inTVAR()

Version 0.9-33:Matthieu Stigler (28/3/2014)

• minor: correct use of partial naming in all.equal(), which resultedin package being taken out of CRAN
• bug correction: correct VARrep in presence of external regressors.Reported by Paul Schreck
• bug correction: correct resVar
• minor: removed various calls of :::
• minor: change package call, most notably remove the enhanceforecast, as forecast already in Imports

Version 0.9-32:Matthieu Stigler (13/7/2013)

• change some tests to avoid difference in outputs depending onplatform/architecture
• remove the data(x) used at bottom of many scripts

Version 0.9-31:Matthieu Stigler (12/7/2013)

• adopt half roxygen documentation
• minor correction in rank.test: add print.summary()
• minor correction in predict_rolling for forecast objects.
• changes in tests/lstar and /compareVecm to lower printing digits, toreduce changes in architecture/platform

Version 0.9-3: MatthieuStigler (5/5/2013)

• change: package loading: put most packages from depends toimport
• change: TVECM.SeoTest, check with try() if inversion ok, otherwiseuse return NA
• change: VAR, check with try() if inversion ok, otherwise useginv()
• vignette: change GIRF section
• New argument: label in some toLatex functions
• New argument: exogen in VECM, VAR

Version 0.9-2: MatthieuStigler (20/12/2012)

• GUI: remove due to problems in assign(yy, globalEnv) not alowedanymore.
• New function: accuracy_stat: give forecast accuracy either comparingtwo datasets, or giving a pred_roll object
• Extend function: predict_rolling: allows to give rolling predictionsfor many models nlar, VAR/VECM, forecast

Version 0.9-12:Matthieu Stigler (11/12/2012)

• Correct bug: TVAR_LRtest now handles extern ThVar
• Correct bug: TVAR had bug in naming arguments
• Minor: many very minor changes, check class in predict, add messagefor ADF in VARrep, exclude lag <1 in lineVar

Version 0.9-1: MatthieuStigler (05/11/2012)

• Extend method: vcov.lstar
• Extend method: confint.lstar
• Extend method: getTh.lstar
• Extend method: regime.lstar
• New argument: thInt as control for starting values of lstar()
• New argument: trace in selectLSTAR and selectNNET
• New argument: strategy for selectLSTAR: allows recovering previousoptimal values for starting values in all LSTAR searches.
• Minor: changed procedure in lstar(), recover only parameters withoptimHessian, instead of runnign second optim round.
• Correct bug: corrected bug in exporting fake ca.jo object, reportedby Uwe Ligges.

Version 0.9-0: MatthieuStigler (26/10/2012)

• New function: rank.select() and derived lags.select(): compute therank and lags according to AC/BIC criteria.
• New function: rank.test() for Johansen LR rank test, using gammaapproximated p-values.
• New function: fevd() and irf() from package vars are nowavailable.
• New function: predict() now implemented for linear multivariatemodels.
• New function: predict_rolling(): allow rolling forecasts, usingsub-samples.
• New function: VARrep(): VAR representation of a VECM, or of a VAR indifferences.
• New function: VECM.sim(): wrapper of TVECM.sim()
• New function: VAR.sim(): wrapper of renamed TVAR.gen()
• New function: VAR.boot(): wrapper of renamed TVAR.gen()
• New argument: fitted() can be retunred either as in the model (sosometimes in diff) or on the original series with arg “level”.
• New argument: VAR.sim() can now fix seed when generating/resampinginnovations, with arg “seed”.
  
• New argument: predict() now can predict using bootstrap or MC, witharg “type”.
• New argument: lineVar() can compute also model with type=“ADF” as inunivariate case.
• New argument: AIC, BIC are corrected and have new argumentfitMeasure.
• New help page: document lokLik()
• New test page: nlar-methods.
• New test page: VAR.
• Correct bug: Correct bug in resVar()
• minor: TVECM.sim(): print more information on input coefficientmatrix when demanded.
• minor: plot.setar(): change lty
• minor: adapt citation to new person() function.
• minor: adapt documentation pages for cross-links to other packagesusing
• minor: corrections in VECM() for restricted constant and interceptsin MLE.
• minor: corrections in VECM() for restricted constant and interceptsin OLS.
• minor: getTh() now alos works for linear multivariate models.

Version 0.8-1: MatthieuStigler (13/02/2012)

• minor: avoid partial matching in call to legend() inTVECM.HStest
• minor: re-run tests with newest R 2.14 in english
• minor: move vignettes from /inst/doc to new folder vignettes/

Version 0.8: MatthieuStigler (12/02/2012)

• New feature: Add argument include in lstar: possibility to estimatelstar with constant, none, trend or both.
• New feature: Add argument starting.control in lstar: possibility tospecify grid search parameters
• New feature: Add argument hpc in selectSETAR to run on parallelcores.
• Minor: add explicit warning in lstar when grid search selectbound.
• Minor: add explicit warning in lstar when optim code is 1,referencing argument maxit.
• Minor: Cleaning of help files, correcting many typos.

Version 0.7-62:Matthieu Stigler (19/12/2011)

• Add standard errors for lstar()
• Minor: change sd(matrix) to apply(matrix,2,sd) (chage in R2.14)
• Minor: change partial argument matching in a few calls
• Minor: correct vignette ThCointegration

Version 0.7-61:Matthieu Stigler (02/08/2011)

• TVECM: allow for K>2 variables when the cointegrating vector ispre-specified
• TVECM summary: change output, adding newlines after coint vector andpercentage of observations

Version 0.7-6: MatthieuStigler (14/05/2011)

• correct bug in star (bugreport by Petri H) by Christoph B
• change return value of star() in case of no star regime detected:returns linear object instead of series itself

Version 0.7-52:Matthieu Stigler (08/05/2011)

• add tests for lstar
• correct lstar so that optim function re-evaluate each time linearparameters
• remove VECM.Rout as useless

Version 0.7-51: MatthieuStigler

• update NAMESPACE as previous (0.7-50) did not work
• add VECM.Rout
• add econometrica .bst and amsplain.bst

Version 0.7-50:Matthieu Stigler (30/03/2011)

• remove own BIC function and use that of pkg nlme, as recommend byProf Ripley
• fix bug in Rd files, and description (missing suggest packages)
• fix bug in ‘autopairs’ and ‘autotriples’: bogus ‘showvalue’ optionwas passed to the ‘scale’ widget (bugreport by Budi Hari Priyanto)

Version 0.7-40:Matthieu Stigler (11/08/2010)

• Add paralell option for TVAR.LRtest(), setarTest(),TVECM.Seotest()
• regime() works also for nlVar, also add arguments initVal andtimeAttr
• RENAME function TVECM.HSTest in TVECM.HStest
• correct bug in TVECM.HSTest

Version 0.7-30:Matthieu Stigler (30/06/2010)

• Add function TVECM.HStest (before was HanSeo_TVECM). Has newargument hpc for parallel computing.
• Add function VECM(), simple wrapper for lineVar(…,model=“VECM”)
• Add Johansen estimator for VECM()
• Add methods AIC, BIC, logLik for class VECM
• Correct numerous bugs

Version 0.7-23:Matthieu Stigler (01/04/2010)

• correct bug TVAR.sim
• export function VAR.sim, simpler wrapper for TVAR.sim

Version 0.7-22:Matthieu Stigler (22/02/2010)

• correct bug in HanSeo_TVECM, which can now be used (but still notexported)

Version 0.7-2: MatthieuStigler (20/02/2010)

• Correct degrees of freedom for VAR covariance matrix
• Standardize output of lineVar with lm
• update slightly vignette
• add specific vignette on threshold cointegration

Version 0.7-1: MatthieuStigler (04/09/2009)

• Change args of TVECM from nn to ngridBeta, from ngridG to ngridTh,model to common
• Add functions regime to extract variabl indicating state regime
• Add function getTh to extract threshold values
• Adapt toLatex.setar to handle MTAR models. Still needsimprovement
• Fix bugs in TVECM, in HansSeo_TVECM and in setar

Version 0.7-0: MatthieuStigler (02/07/2009)

• added possibilty to have two thresolds and hence three regimes insetar and selectSETAR (arg nthresh)
• created new functions for unit roots tests: KapShinTest() andBBCTest()
• created new functions for estimating VAR and VECM: lineVar
• created new function for estimating TVECM: function TVECM()
• created new function for estimating TVAR: function TVAR()
• created new function to test for setar: function setarTest()
• created new function to test for TVAR: function TVAR.LRtest()
• created new function to test for TVECM: functions TVECM.SeoTest()and HanSeo_TVECM()
• created new function to simulate/bootstrap a TVAR: functionTVAR.sim()
• created new function to simulate/bootstrap a TVECM: functionTVECM.sim()
• created new function to simulate/bootstrap a setar: functionsetar.sim()
• created new function to estimate regime-specific variance in setar:function resVar()
• created new function to extend a bootstrap replication in setarTest:function extendBoot()
• added ‘zeroyld’ dataset
• added ‘unemp’ dataset
• added ‘IIPUs’ dataset
• added in selectSETAR() and setar() following args: include, common,model, trim, MM, ML, MH, model, restriction
• added MakeThSpec() function for specification of the threshold inthe search
• added in selectSETAR(): criterion “SSR” (sum of squares residual)and argument max.iter
• extended arg th in selectSETAR to search inside an intervall oraround a point or on the whole grid
• added k parameter to the AIC method and fixed a bug
• standardized license naming in description
• fixed typo in print.star

Version 0.6-1: AntonioFabio di Narzo

• fixed CITATION file to the new R format
• fixed bug in nnetTs: control args were not passed to ‘nnet’ fitterreported by prof. Joachim Zietz
• fixed bug in llar.predict (spotted by Markus Gross)

Version 0.6-0: AntonioFabio di Narzo

• fixed Matrix dependency in DESCRIPTION
• added minimal docs for newly created internal objects
• added STAR model option to the tcltk GUI
• added a new STAR-fitting function with automatic selection of thenumber of regimes
• fixed bugs in lstar
• fixed bugs in some internal functions
• added google codebase project URL to DESCRIPTION

Version 0.5-6: AntonioFabio di Narzo

• fixed bug in setar model when mL and/or mH=0

Version 0.5-5 released

• fixed bug in lstar model out-of-sample forecasting

Version 0.5-4 released

• fixed some warnings issued in R-2.4.0 check script

Version 0.5-3 released

• fixed some warnings issued in R-devel check script
• added URL field to DESCRIPTION

Version 0.5-2 released

• fixed gui design error in autopairs and autotriples

Version 0.5-1 released

• added draft of internal design vignette
• minor fixes in vignette, added AAR model fitting example
• fixed bug in setar and lstar (introduced in version 0.5)

Version 0.5 released

• added minor fixes for passing checks under R-2.3.0
• added tsDyn package overview man page, with getting startedindications
• upgraded vignette to the new models usage
• fixed bug in plot.llar
• changed llar interface to nlar models schema
• eliminated nlar function: now each model has its ownfunction. Using more explicitely classes inheritance and polymorphism.Adding more nlar models now should be much simpler.

Version 0.4-1 released

• updated all package dialogs to the new GUI system
• written a whole (basic) object oriented GUI system, withpass-by-reference semantics
• added input checking to first dialog of nlarDialog()

Version 0.4 released

• added hidden ‘series’ name argument to ‘nlar’
• added early version of nlarDialog, a GUI for ‘nlar’ modelfitting
• fixed latex apparence of some manual pages
• fixed notational errors in vignette: Z_t=Z_{t-d}, z_t changed in Z_tin LSTAR model formulation
• added nlar.fit class documentation
• changed text entry with slide bar in autopairs and autotriples
  
• added availableModels function, upgraded docs
  
• added and documented aar model
  
• added and documented tarch model
• added and documented arch model

Version 0.3-5 released

• largely improved apparence of regime-switching plot, changed palettefor lstar diagnostic plots
• added latex printing of fitted setar models

Version 0.3-4 released

• changed lag-plots for setar and lstar models when more than 300points are to be plotted
• fixed notation error in vignette (equation (4) on neuralnetworks)
• deleted 0-lag in acf, pacf and ami summary plots for nlar.fitobjects
• extended threshold variable printing informations for setar andlstar models
• added alternative selection criteria for selectSETAR
• added regime proportions info in print.setar and summary.setar
• added regime switching plot to plot.setar and plot.lstar
• fixed graphical pars. resetting in plot.nlar.fit
• added convenience wrappers ‘setar’ and ‘lstar’, improved relativedocumentation

Version 0.3-3 released

• fixed minor bug in selectSETAR output presentation

Version 0.3-2 released

• changed general AIC formula
• added dummy (non-working), undocumented version of FAR (FunctionalAutoRegressive) model
• added early, undocumented version of AAR (Additive AutoRegressive)model
• modified selectSETAR, selectLSTAR, selectNNET functions: now morecoherent results

Version 0.3-1 released

• added autotriples.rgl (now also the rgl package is suggested)
• added the possibility to stop the llar procedure
• fixed y-scale in llar diagnostic plot
• improved autotriples, upgraded vignette

Version 0.3 released

• added llar.predict and llar.fitted functions, improved llar outputobject

Version 0.2-4 released

• fixed vignette
• fixed external variable management in predict.setar andpredict.lstar

Version 0.2-3 released

• added vignette
• improved SETAR and LSTAR models: now ‘predict’ interface is sensibleto the type of threshold variable
• fixed and exported delta, delta.test, delta.lin anddelta.lin.test

Version 0.2-2 released

• added optional legend to plot.setar and plot.lstar

Version 0.2-1 released

• added autopairs and autotriples functions, with a minimal (optional)TclTk GUI

Version 0.2 released

• changed builtin nonlinear models documentation structure: now onepage per model
• added possibility of external threshold variables in setarmodels
• added unexported delta tests of indipendence and linearity
• added hidden tests in nlar-methods and llar man pages examples
• improved summary method for linear and setar models
• added llar routine, i.e., Casdagli test of nonlinearity

Version 0.1-2 released

• removed unuseful sum-of-squares functions
• added selectSETAR, selectLSTAR and selectNNET functions, forautomatic selection of model hyper-parameters
• minor bug fixes
• added ‘predict’ method to nlar.fit objects, subclasses linear,setar, lstar and nnet
• fixed nlar.fit objects time series storage redundancy

Version 0.1-1 released

• added neural network model documentation
• added dummy predict method to nlar.fit
• minor bug fix in lstar model
• added lstar and setar plot method
• fixed lstar model documentation
• bug fix in summary.lstar when embedding dimension is 1
• bug fix in nlar and plot.setar when embedding dimension is 1
• added dummy ‘summary’, ‘print’ and ‘plot’ function to nnetsubclass
• added utility ‘sum-of-squares’ functions for implemented models
• added plot method to nlar.fit class

Version 0.1: initial release