Hansen's (1995) Covariate-Augmented Dickey-Fuller (CADF) test. The only required argument is y, the Tx1 time series to be tested. If no stationary covariate X is passed to the procedure, then an ordinary ADF test is performed. The p-values of the test are computed using the procedure illustrated in Lupi (2009).
Changes in Version 0.3-3
Some compilation warnings corrected. Vignettes have been updated.
Changes in Version 0.3-2
Vignettes migrated to new directory "vignettes".
Changes in Version 0.3-1
Corrections to methods: - A minor bug has been corrected in the plot() method. Documentation: - Some typos corrected and other minor fixes in the .Rd files.
Changes in Version 0.3-0
A new improved version of the "CADFtest" package (version 0.3-0) is available. The new version includes some minor adjustments to the code and adds a number of new features. New methods added in Version 0.3-0: - residuals() retrieves the model's residuals; - plot() plots the model's residuals (standardized residuals, residuals density and Jarque-Bera test for normality, ACF of residuals, partial ACF of residuals). - an update-method has been added to update model specification, lags, options. - the summary-method now produces an object of class "CADFtestsummary". A new print-method has been added for this class. This allows the user to control the usual options (digits, signif.stars, etc.). The p-value of the lagged dependent is now consistent with the CADF distribution. The F-statistic refer to the joint significance of the stationary regressors in the auxiliary test regression. New options and arguments: - in the new version, the model can be specified also as "y ~ 1". This means that no stationary covariates are used, and an ordinary ADF model will be carried out. However, ``1'' does not necessarily mean that a onstant is used in the model: as usual, the type of model (without constant, with constant, with constant and linear trend) is chosen according to the argument "type". - instead of specifying "Auto=TRUE" for automatic model selection (as in the previous version of the package), the criterion argument now takes the form criterion=c("none", "BIC", "AIC", "HQC"). The default is "none" that corresponds to the "Auto=FALSE" argument of the previous version of the package. The Hannan-Quinn criterion ("HQC") has been added for completeness to the already present Akaike ("AIC") and Schwarz ("BIC") criteria; - the user can control all the argument passed to "KernHAC", instead of just "kernel" and "prewhite". However, the default has been changed to quadratic spectral kernel with VAR(1) prewhitening. If one wants to use Hansen's (1995) setting kernel="Parzen" and prewhite=FALSE have to be specified; - the argument `trend' has changed to type = c("trend", "drift", "none") to be consistent with package "urca". New data: - The subdirectory ./data contains the tables of parameters estimated in Costantini et al. (2007) in binary (.rda) format. The tables can be accessed using the usual procedures. Dependencies: - The package now depends on "urca" rather than on "fUnitRoots", given that the latter will become deprecated and the relevant functions have been already ported to "urca". Dependency on "tseries" has been added to allow for the new features of version 0.3-0 (the Jarque-Bera test is computed using package "tseries"). Documentation: - Some typos corrected and other minor fixes in the .Rd files; - documentation on the new features has been added; - an URL tag has been added to ./inst/CITATION.