Movatterモバイル変換

tsgarch 1.0.3

• Added the log-likelihood vector to the returned fitted and filteredobject as this will be needed in the calculation of the standard errorsin the upcoming multivariate GARCH package.
• Added an extra option to the estimation method which adds the TMBobject to the returned estimation object. This can then be used todirectly vary the parameters and quickly extract information from thefiltration. This could be also achieved by the tsfilter method with aspecification object but is much slower. Use case is for themultivariate GARCH partitioned hessian calculation.
• Added plus overload method to combine together GARCH specificationsto generate a multi-specification object which can then be estimated inparallel. This is required for 2-stage multivariate GARCH models. Aseparate to_multi_estimate function is also added to instead convert alist of estimated objects to a validated multi_estimate class.Extractors include fitted, residuals and sigma.
• Removed RcppArmadillo dependency and converted code to RcppEigensince it is already in use by TMB.
• Switched to using simulate for the parametric simulation for thepredict method. The bootstrap still remains the most valid approach asthe out of sample distribution is best approximated by the bootstrappedresiduals rather than the imposed parametric distribution with estimatedparameters.
• For the bootstrap simulated prediction, the re-sampled standardizedinnovations are now scaled to avoid bias.
• Fix to h=1 and nsim. Previously when h=1, nsim was set to zero.

tsgarch 1.0.2

• Moved a unit tests back to original folder and added a tolerance tothe expectation per CRAN maintainers directions.

tsgarch 1.0.1

• Moved a couple of unit tests to other folder to avoid checking onCRAN since the M1 mac had different rounding errors than otherarchitectures for simulation tests.

tsgarch 1.0.0

• Initial CRAN submission.
• Changes to initialization of recursion in the ARCH equation to bemore consistent with the literature. This leads to a more than doublingin the accuracy against the FCP benchmark.
• Correction to EGARCH forecast to account for log bias.
• Added a couple more data series for benchmarking.
• Added a pdf vignette.
• Added demo html vignettes.
• Added extensive unit tests.