NEWS

lgarch 0.7 (2025-03-25)

class checking in rss() is now done with is()
minor changes to DESCRIPTION file
minor changes to virtually all the .Rd files (i.e. the help-files)
fix: - mlgarch(): bug-fix (thanks to Rik Wienke!) that affected the dimension of the 'xreg' argument

lgarch(): - formula for the variance-covariance matrix of the ARMA-representation corrected when method = "ls" - improved column-names handling of X-regressors - the dates/index of regressor(s), i.e. the xreg argument, is automatically matched with dates/index of the regressand

lgarchSim: c.code argument added with default TRUE (i.e. compiled C-code is used for the recursion; this speeds up simulations considerably)
S3 method summary() added for objects of class lgarch and mlgarch
lgarch() and mlgarch() functions: backcast.values argument removed
lgarchRecursion1(): for improved numerical stability, the backcast values of ln(y^2) was changed to the empirical mean. Also, a bug that occurred whenever c.code = FALSE was corrected

lgarch(): - mean-correction as estimation-option added - a third estimation method was added: QML via the centred Chi-squared distribution as instrumental density - fitted.lgarch(): bug fix (the bug induced incorrect fitted values at zero-locations) - stylistic changes to the documentation

functions and S3 methods for the simulation and estimation of the multivariate CCC-log-GARCH(1,1) model were added
gdiff() function added
rss.lgarch() function changed name to rss
zoo-related bug corrected in glag()
glag() function: improved further, and the pad argument in the glag function acquired a new default (TRUE)
minor improvements throughout, and several stylistic changes made to the documentation

lgarch(): - a couple of bugs corrected in the parameter-indexing, which ocurred whenever the garch order argument was set to 0 - argument logl.penalty changed name to objective.penalty - the argument method=c("ml","ls") was added. If method="ml", then estimation is with Gaussian QML via the ARMA representation. If method="ls", then estimation is with least squares via the ARMA representation. Although asymptotically equivalent in most respects, the latter is slightly faster since one parameter less is estimated
glag(): Completely rewritten. Now it can also lag matrices, and it gives a "special treatment" to zoo-objects (the indexing is retained)
lgarchLogl() function changed name to lgarchObjective()
new functions: - rss.lgarch(), extract the Residual Sum of Squares of the ARMA representation an object of class lgarch - mlgarchSim(), simulate from a multivariate log-GARCH(1,1) - rmnorm(), simulate from multivariate normal distribution