NEWS

fEGarch 1.0.6 (2026-02-10)

GJR-GARCH, TGARCH, APARCH, FIGJR-GARCH, FITGARCH and FIAPARCH pre-sample values have been stabilized slightly.
In dual models, where for some models in the volatility part pre-sample quantities are set through empirical values, a preliminary ARMA or FARIMA estimation step has been implemented to obtain their residuals and subsequently obtain the pre-sample values based on those residuals; this stabilizes estimated conditional standard deviations at the beginning of the training period; dual models with long-memory EGF part for the variance model stay fully unaffected.
fixed a bug in garch_sim, where the simulation failed, if no setting of phi in the list for argument pars was provided manually.

a bug in the backtesting functions for output of measure_risk,fEGarch_distr_est-method was fixed that was introduced with package version 1.0.4; this did not affect the measure_risk function.
step size different from 1 in rolling volatility forecasts under model refitting now available for GARCH-type models different from the exponential GARCH family (was already and is available for exponential GARCH family models).

the method predict_roll() now allows for refitting of the underlying model via the new argument refit_after during the test period.
measure_risk() and backtest functions take into account the potential refitting to compute risk measures and backtest quantities.
all fitting functions now include an argument skip_vcov to allow for skipping the variance-covariance matrix computation

numerical stability of estimators improved regarding parameter mu in cases with strong outliers
some parameter default starting values were adjusted slightly to further increase numerical stability
typos in the documentation for trafflight_test,fEGarch_risk-method were fixed
some unnecessary elements in the manual for internal functions have been removed from the manual for improved clarity

required package update for RcppArmadillo update to Armadillo 15.0.*
a bug was fixed, where for the nonparametric scale estimation always the automated bandwidth selection under long-memory errors was employed, even for short-memory models; now the functions select the appropriate bandwidth selection algorithm correctly for short- and long-memory models
a safety measure was included to avoid that the log-likelihood can become -Inf
math formulas were fully removed from the README, because they were not displayed properly on CRAN
some minor typos were corrected in the README

a bug was fixed, where for most models except for EGARCH-type models the mean estimate was not considered properly in volatility forecasts
a bug was fixed, where for EGARCH-family models rolling point forecasts under the (skewed) ALD were erroneous
the mathematical formula in the README was not displayed correctly on the CRAN servers and was therefore removed entirely
the package description was adjusted to contain also information on the the more advanced models and backtesting capabilities of the package, making it easier for users to see what to expect from the package
the package title was adjusted to contain more information on the package's contents, making it easier for users to see what to expect from the package
a bug was fixed in the computation of the hessian matrix, where now it is caught if a hessian is not invertible through solve() and the fitting functions therefore now don't stop due to an error anymore
an additional dataset UKinflation was added to the package to show applications of the package beyond return data
some examples in the README were adjusted / added