Version 0.9-47: Matthieu Stigler (15/5/2018)
 -fix small issue in lstar code.

Version 0.9-46: Matthieu Stigler (21/1/2018)
 -new: predict.TVAR
 -new: rewrite TVAR.gen, cleaner TVAR.boot and TVAR.sim
 -fix bug: nameB in TVAR had issue for some cases.
 -many minor updates in doc

Version 0.9-45: Matthieu Stigler (29/8/2017)  never published on CRAN
 -minor: change registration of c code
 -minor: add dataset m.unrate, previously in (now archived) FinTs
 -minor: small documentation updates, use canonical urls, etc

Version 0.9-44: Matthieu Stigler (22/5/2016)
 -new: vcov.TVAR
 -minor fix: TVAR_LRtest now uses robust QR instead of manual solve()...
 -fix bug: vcov.VAR corrected, was wrong
 -minor fixes:  VAR.sim did not return all args, internal insertCol did not handle well just one row, typos in man file VECM
 -minor new: summary.VAR now outputs component coefMat, more consistent output
 -minor new: TVAR outputs df.residual
 -vcov: improve vcov for VAR/VECM, using faster chol2inv saving previous qr
 -minor fixes in NAMESPACE, urls citations, etc 

Version 0.9-43: Matthieu Stigler (23/4/2015)
 -minor fixes: add checking in selectLSTAR in case of convergence issue, rename lstar.Rout.save as created issues on other platforms

Version 0.9-42: Matthieu Stigler (15/4/2015) never published on CRAN
 -full rewrite of VAR.sim, VAR.gen, now cleaner, allows for exogeneous predictions, etc.
 -remove predict.VECM
 -fix small issue in DESC file (pkg description in title case, only one maintainer).

Version 0.9-41: Matthieu Stigler (26/8/2014)
 -change description for new url to github wiki, as well as new field mailinglist
 -fix suggests issues (sm::foo()) newly detected by R CMD check devel

Version 0.9-40: Matthieu Stigler (12/8/2014), never published on cran
- new functions coefB(), coefA(), coefPI() returning the A, B and PI coefs 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(), in TVAR()


Version 0.9-33: Matthieu Stigler (28/3/2014)
-minor: correct use of partial naming in all.equal(), which resulted in 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 enhance forecast, as forecast already in Imports

Version 0.9-32: Matthieu Stigler (13/7/2013)
-change some tests to avoid difference in outputs depending on platform/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, to reduce changes in architecture/platform

Version 0.9-3: Matthieu Stigler (5/5/2013)
-change: package loading: put most packages from depends to import
-change: TVECM.SeoTest, check with try() if inversion ok, otherwise use return NA
-change: VAR, check with try() if inversion ok, otherwise use ginv()
-vignette: change GIRF section
-New argument: label in some toLatex functions
-New argument: exogen in VECM, VAR

Version 0.9-2: Matthieu Stigler (20/12/2012)
-GUI: remove due to problems in assign(yy, globalEnv) not alowed anymore.
-New function: accuracy_stat: give forecast accuracy either comparing two datasets, or giving a pred_roll object
-Extend function: predict_rolling: allows to give rolling predictions for 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 message for ADF in VARrep, exclude lag <1 in lineVar

Version 0.9-1: Matthieu Stigler (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 previous optimal values for starting values in all LSTAR searches. 
-Minor: changed procedure in lstar(), recover only parameters with optimHessian, instead of runnign second optim round. 
-Correct bug: corrected bug in exporting fake ca.jo object, reported by Uwe Ligges. 


Version 0.9-0: Matthieu Stigler (26/10/2012)
-New function: rank.select() and derived lags.select(): compute the rank and lags according to AC/BIC criteria. 
-New function: rank.test() for Johansen LR rank test, using gamma approximated p-values. 
-New function: fevd() and irf() from package vars are now available. 
-New function: predict() now implemented for linear multivariate models. 
-New function: predict_rolling(): allow rolling forecasts, using sub-samples. 
-New function: VARrep(): VAR representation of a VECM, or of a VAR in differences. 
-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 (so sometimes in diff) or on the original series with arg "level". 
-New argument: VAR.sim() can now fix seed when generating/resamping innovations, with arg "seed".  
-New argument: predict() now can predict using bootstrap or MC, with arg "type". 
-New argument: lineVar() can compute also model with type="ADF" as in univariate case. 
-New argument: AIC, BIC are corrected and have new argument fitMeasure. 
-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 coefficient matrix when demanded. 
-minor: plot.setar(): change lty
-minor: adapt citation to new person() function. 
-minor: adapt documentation pages for cross-links to other packages using \pgk{}
-minor: corrections in VECM() for restricted constant and intercepts in MLE. 
-minor: corrections in VECM() for restricted constant and intercepts in OLS. 
-minor: getTh() now alos works for linear multivariate models. 

Version 0.8-1: Matthieu Stigler (13/02/2012)
-minor: avoid partial matching in call to legend() in TVECM.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: Matthieu Stigler (12/02/2012)
-New feature: Add argument include in lstar: possibility to estimate lstar with constant, none, trend or both.
-New feature: Add argument starting.control in lstar: possibility to specify grid search parameters
-New feature: Add argument hpc in selectSETAR to run on parallel cores. 
-Minor: add explicit warning in lstar when grid search select bound. 
-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 R 2.14)
-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 is pre-specified
-TVECM summary: change output, adding newlines after coint vector and percentage of observations

Version 0.7-6: Matthieu Stigler (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 linear parameters
-remove VECM.Rout as useless

Version 0.7-51: Matthieu Stigler 
-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 by Prof Ripley
-fix bug in Rd files, and description (missing suggest packages)
-fix bug in 'autopairs' and 'autotriples': bogus 'showvalue' option was 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 and timeAttr
-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 new argument 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 not exported)

-version 0.7-2: Matthieu Stigler (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: Matthieu Stigler (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 needs improvement
-Fix bugs in TVECM, in HansSeo_TVECM and in setar

-version 0.7-0: Matthieu Stigler (02/07/2009)
-added possibilty to have two thresolds and hence three regimes in setar and selectSETAR (arg nthresh)
-created new functions for unit roots tests: KapShinTest() and BBCTest()
-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: function TVAR.sim()
-created new function to simulate/bootstrap a TVECM: function TVECM.sim()
-created new function to simulate/bootstrap a setar: function setar.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 in the search
-added in selectSETAR(): criterion "SSR" (sum of squares residual) and argument max.iter 
-extended arg th in selectSETAR to search inside an intervall or around 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: Antonio Fabio di Narzo
-fixed CITATION file to the new R format
-fixed bug in nnetTs: control args were not passed to 'nnet' fitter
	reported by prof. Joachim Zietz
-fixed bug in llar.predict (spotted by Markus Gross)

-version 0.6-0: Antonio Fabio 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 the number of regimes
-fixed bugs in lstar
-fixed bugs in some internal functions
-added google codebase project URL to DESCRIPTION

-version 0.5-6: Antonio Fabio 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 started indications
-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 own function. 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, with pass-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' model fitting
-fixed latex apparence of some manual pages
-fixed notational errors in vignette: Z_t=Z_{t-d}, z_t changed in Z_t in 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 palette for 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 300 points are to be plotted
-fixed notation error in vignette (equation (4) on neural networks)
-deleted 0-lag in acf, pacf and ami summary plots for nlar.fit objects
-extended threshold variable printing informations for setar and lstar 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 relative documentation
	
-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 (Functional AutoRegressive) model
-added early, undocumented version of AAR (Additive AutoRegressive) model
-modified selectSETAR, selectLSTAR, selectNNET functions: now more coherent 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 output object

-version 0.2-4 released
-fixed vignette
-fixed external variable management in predict.setar and predict.lstar

-version 0.2-3 released
-added vignette
-improved SETAR and LSTAR models: now 'predict' interface is sensible to the type of threshold variable
-fixed and exported delta, delta.test, delta.lin and delta.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 one page per model
-added possibility of external threshold variables in setar models
-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, for automatic 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 nnet subclass
-added utility 'sum-of-squares' functions for implemented models
-added plot method to nlar.fit class

- version 0.1: initial release
