Changed maintainer e-mail to [email protected].
Expanded the description and included the doi of the relevant papers.
All the real kinds in the Fortran sources have been changed in C_DOUBLE from the ISO_C_BINDING module, for portability.
Now TARMAGARCH.test uses the package rugarch to estimate in one step the parameters of the ARMA-GARCH null model. Checks for stationarity and invertibility have been added. Also, the routine has been improved to include the ARCH test when garch.ord = 0.
TARMA.test, has been improved in many places and the heteroskedastic robust version of the routines have been added and are now returned as output, together with their corresponding parameters. Checks for non-invertibility of the MA part are now consistent with those of TARMAGARCH.test.
All the routines for TAR and TARMA testing are now more consistent, especially in the computation of the threshold range.
In TARMA.fit fixed a bug in the computation of the Hessian matrix D2eps.R that was only triggered when the dimensions of tma1.lags and tma2.lagswere different. Also, only half of the Hessian matrix was populated.
Extensive improvement of TARMA.fit, see below.
The robust method in TARMA.fit has been reworked as to include Student's T innovations for the robust rho function.
The new switch innov allows to select the innovation density. Allowed options are norm for Normal and student for Student's T. The robust weights are also returned.
The HAC standard errors in TARMA.fit are now computed in a unified modular way. A small bug is also fixed (wrong scaling).
A new trimmed LS estimator for TARMA.fit is implemented (experimental).
The switch estimate.thd in TARMA.fit and TARMA.fit2 has been removed. Now if threshold is not NULL the
threshold is estimated over the threshold range.
the Iterative Re-weighted Least Squares algorithm has been improved and made more robust.
Moreover, now irls.control can be used to control the algorithm.
The documentation has been improved.
Fixed a bug in the creation of the residuals: the first k zeros were not removed (line 859)
Now print.TARMA prints the type of innovation density and its estimated parameters. Also, the output is more informative on the kind of estimation method selected.
Removed a leftover string in the doc of TARMAur.test.B.
Updated and improved the references.
Cosmetic improvement of print.TARMAtest.
print.TARMARemoved a leftover call to the Fortran intrinsic RANDOM_NUMBER function.
Fixed the docs to cope with the roxygen2 issue https://github.com/r-lib/roxygen2/issues/1491
Now the residuals in TARMA.fit inherit the time attributes of the original series.
Fixed print.TARMAtest when fixed is used to fix parameter values in TARMA.fit2.
In TARMAGARCH.test: set include.mean=TRUE in the ARMA fit and removed centering.
Minor fixes to the documentation of TARMAGARCH.test.
Updated the e-mail address of Greta Goracci in the docs.
Updated the Statistica Sinica 2023 reference.
Fixed missing dependency in MAKEVARS: the Fortran library TARMAurtest depends on the Fortran module TARMA_MOD
Fixed broken references in the README.md file. Now README.Rmd is used to generate it.
Fixed missing PACKAGE in several .Fortran calls.
First CRAN submission.
Fixed urls and DESCRIPTION file to avoid notes from CRAN.
Added value field to plot.tsfit, predict.TARMA, print.TARMA, print.TARMAtest.
Improved the examples of predict.TARMA.
Removed \dontrun or replaced it with \donttest.
The new functions TARMAur.test and TARMAur.B.test implement the asymptotic and bootstrap supLM unit root test for an integrated MA against a stationary TARMA process. The print method has been updated to handle them and uses tabulated critical values made available through the dataset supLMQur.
Added the Journal of Econometrics reference for TAR.test and TAR.test.B.
The new function TAR.test implements the robust supLM test for TAR nonlinearity.
The dataset ACValues contains the tabulated critical values of Andrews(2003), Econometrica.
TAR.test, TAR.test.B, TARMA.test, TARMAGARCH.test have been modified to use the new class TARMAtest. The print method for TARMAtest is implemented and extracts and prints the critical values from the dataset ACValues.
TARMA.test.B has been replaced by the new function TAR.test.B. Added the bibtex item for the relevant paper.
The Fortran module TARMA_MOD has been updated. Also, the ISO_C_BINDING facilities for floating point types are used.
Removed the switch shift from TARMA.test and TARMAGARCH.test.
TARMAGARCH.test now always uses method for arima fitting.
Minor cosmetic polishing of the documentation of TARMA.test and TARMAGARCH.test.
The routines tarma.fit and tarma.fit2 have been renamed TARMA.fit and TARMA.fit2, for consistency.
In tarma2.fit a warning is issued if the regressors and the series have different lengths.
Fixed a bug in tarma2.fit. The residuals lost the tsp attributes of the original series.
Fixed a bug in tarma.fit. Now the robust standard errors routine for the robust method works also for TMA order greater than one. The internal routine D2eps.R has been fixed.
The documentation of tarma.fit has been improved. Most notably, the examples section has been expanded and is now run.
Fixed the examples of plot.tsfit and predict.TARMA: they had not been updated after the change to tarma.fit in version 0.1-4.
print.TARMA now reports also the value of alpha and qu if the estimation method is robust.
plot.tsfit gains more flexibility by allowing control over: line type with ltype, line width with lwdt.
Also, the default line colours and widths have been changed. The examples are now run.
Minor improvements to the docs: the defaults of predict.TARMA have been changed. Added a reference for the robustness paper.
Now tarma.fit2 allows to specify different subsets of lags for the two regimes: tar1.lags and tar2.lags. tarma.fit has been changed accordingly and now passes the exact subsets to tarma.fit2 for initial estimation. The documentation of tarma.fit2 has been updated.
print.TARMA has been improved and the methods for the TARMA class have been moved to a separate file.
plot.tsfit now uses inherits to avoid the note from checking.
Minor corrections and improvements to the documentation of tarma.fit.
Minor improvements of Description and Readme.
Fixed a bug in predict.TARMA: the number of discarded observations did not take into account ar.lags.
TARMA.test and TARMA.test.B now always use method for arima fitting.
TARMA.test and TARMA.test.B now also work with zoo objects.
The robust version of tarma.fit has been polished and is no more experimental.
Fixed a bug in the derivation of robust standard errors.
Fixed a in bug in tarma.fit for which the trimmed estimation that was meant to be used only for the initial estimation was used also in the subsequent steps. Fixed alpha in line 407.
tarma.fit has been completely rewritten and uses an experimental additional iteratively reweighted least squares step to estimate the robust weights.The routines ARMAvsTARMA.test and ARMAvsTARMA.B.test have been renamed TARMA.test and TARMA.B.test , respectively.
Added the routine TARMAGARCH.test .
The method print.TARMA has been rewritten.
The docs have been rewritten to use the package mathjaxr.
Fixed a bug in line 259 of tarma.fit2: replaced > with !=.
The output slot method is added to tarma.fit2 for consistency with tarma.fit. It is set
to MLE and has no effect other than allowing to differentiate between the two routines.
The output slots arlag and include.int are added to tarma.fit2.
tarma.fit has been rewritten. Now it can fit a full subset TARMA(p1,p2,q1,q2) model.
Some ergodicity bounds for the optimization routines have been also added.
The objective functions have been rewritten in a recursive fashion and the code now is
several times faster than before. predict.TARMA has been modified to cope with the new routine.
Added a warning upon failed convergence in tarma.fit.
Added the S3 method residuals.TARMA.
tarma.fit now outputs the values of the AIC and BIC computed
according to Chan & Cryer (2008) and Chen & Tsay (2019).
The output of tarma.fit2 has been modified to match that of the new version of tarma.fit.
ARMAvsTARMA.test now includes the full test for ARMA versus TARMA when ma.fixed=FALSE.
The output of the test now includes the information on the parameters tested. The documentation
has been amended accordingly.
Fixed a bug in predict.TARMA: now it handles correctly time series objects
with frequency different from 1.
Now tarma.sim simulates from a full TARMA(p1,p2,q1,q2) model where q1 and q2 can be different.
tarma.fit2 now gains the possibility to add a seasonal MA part. The output has been amended accordingly.
Fixed a bug in tarma.sim when the (T)MA order was greater than the (T)AR order.
Now the output of tarma.fit2 has no NAs and can be used in tarma.sim.
Changed the option fitted in plot.fit in plot.tsfit.
Fixed a bug in tarma.sim: now it handles correctly time series objects with frequency different from 1.
ARMAvsTARMA.test now includes the test for AR versus TAR when ma.ord=0.
Fixed a bug in the output of ARMAvsTARMA.test.B: Tb was outside the output list.
tarma.fit2 now allows the inclusion of covariates in the model.