2014-08-13 Alexander Ristig
* The quasi Maximum Likelihood estimation procedure (method = 1) is modified, such that the pseudo variables are computed from the diagonal of Archimedean copulas. As a result, the estimates at higher hierachical levels are consistent and (asymptotically) unbiased.
* The function nacopula2hac() was added. It transforms 'nacopula' objects into 'hac' objects.
2014-06-09 Alexander Ristig
* The updated version of the HAC package is referred to the JSS manuscript:
Ostap Okhrin, Alexander Ristig (2014). Hierarchical Archimedean Copulae: The HAC Package.
Journal of Statistical Software, 58(4), 1-20. URL http://www.jstatsoft.org/v58/i04/.
* The assignments ML = 1, FML = 2, RML = 3, HAC_GUMBEL = 1, AC_GUMBEL = 2, HAC_CLAYTON = 3, AC_CLAYTON = 4, HAC_FRANK = 5, AC_FRANK = 6, HAC_JOE = 7, AC_JOE = 8, HAC_AMH = 9 and AC_AMH = 10 are removed, as they might cause error in the underlying functions. To avoid this kind of errors, please use the integer values themselves instead of the objects.
* to.logLik has an additional argument, called sum.log = TRUE (by default). If sum.log = FALSE, the log of the log-likelihood contributions are returned.
* A bug in the estimate.copula() function has been fixed, which might slightly improve the estimation results for large dimensional HAC based on complicated structures.
2014-02-13 Alexander Ristig
* estimate.copula() was debugged. Even though the argument `epsilon > 0« in estimate.copula(), the structure was not aggregated anymore. After the last update, `epsilon = 0« was set by default for `type = ML«.
2014-01-12 Alexander Ristig
* type = GAUSS is removed.
* The estimation method (method = TAU) is removed, as the asymptotic theory for Kendall's $\tau$ is usually restricted to the two-dimensional case and cannot be transferred to a higher-dimensional framework as necessary for the considered purpose.
* The package also covers the following generator families: Frank, Joe, Ali-Mikhail-Haq.
* From now, the simulation methods of all copula families are based on those of the copula-package.
2013-05-03 Alexander Ristig
* A vignette and the dataset finData are added.
2012-12-22 Alexander Ristig
* rHAC returns non-zero values for 'large' copula parameters, say up to theta = 30.
2012-10-11 Alexander Ristig
* to.logLik is debugged for the argument eval = TRUE. This causes changes in the value of the log likelihood, if the values of the vector returned by get.params() are not ascendingly ordered.
2012-08-06 Alexander Ristig
* Updates regarding changes of the copula package.
2012-06-19 Alexander Ristig
* The Recursive ML (RML) procedure is added. It provides almost precise estimates as the FML procedure, but builds the structure simultaneously.
* rAC is removed. Use rHAC to sample from Archimedean copulas.
2012-05-16 Alexander Ristig
* Warnings are added to dHAC and pHAC, as an useless error message was shown, when the colnames of X and the variables' names of hac were not identical.
* type = HAC_ROTATED_GUMBEL is removed.
* Restrictions of full ML estimation are optimized.
2012-05-04 Ostap Okhrin and Alexander Ristig
* dHAC and pHAC work for estimated ACs.
* to.logLik provides the log-Likelihood function for arbitrary HAC.
* Based on to.logLik, full Maximum-Likelihood estimation `FML' is implemented.
* par.pairs returns the pairwise parameter of a hac object.
2012-04-20 Ostap Okhrin and Alexander Ristig
* The package is generally upgraded and allows for general structures. The density is available for arbitrary structures as well. Note, the construction of `hac´ objects is modified.
2011-12-24 Ostap Okhrin and Alexander Ristig
* rHAC with type = HAC_CLAYTON works for small copula parameters.
2011-12-05 Ostap Okhrin and Alexander Ristig
* Due to an update of the copula-package, the dependencies of HAC were adjusted.
2011-10-18 Ostap Okhrin and Alexander Ristig
* estimate.copula is debugged. Estimation of a 2-dim HAC was not possible.
* estimate.copula is upgraded. The argument `margins´ specifies how the margins are computed. They can be determined nonparametrically denoted by "edf" or in parametric way , e.g. "norm". Following the latter approach the parameters of the selected distribution are estimated by ML. Basing on these estimates the values of the univariate margins are computed. If the argument is defined as scalar, all margins are computed according to this specification. Otherwise, one can define a different margin for each column of \code{X}, e.g. \code{c("norm", "t", "edf")} for a 3 dimensional sample. Almost all continuous functions of \code{\link{Distributions}} are available. Wrong usage of this argument might result in misspecified margins, e.g. application of \code{"exp"} though the sample contains negative values. If the sample contains the margins yet, the default value is to use.
* pHAC, dHAC, emp.copula and emp.copula.self are upgraded with the argument `margins´.
* pHAC and dHAC are upgraded. The arguments na.rm and max.min are introduced. So far the arguments are used for the functions estimate.copula, emp.copula and emp.copula.self
2011-10-10 Ostap Okhrin and Alexander Ristig
* as.hac is introduced as further function to create hac objects.
* aggregate.hac and estimate.copula are upgraded. The variables "method" and "agg.method" are introduced. It allows to determine the new dependency parameter to be the mean, min or max of the two previous, if epsilon > 0.
2011-09-18 Ostap Okhrin and Alexander Ristig
* The manual is revised.
2011-08-02 Ostap Okhrin and Alexander Ristig
* emp.copula is upgraded. If na.rm = TRUE, values < 0 and > 1 are not removed anymore. However, the variable max.min is introduced. If max.min = TRUE, values > 1 are set to 1 and values < 0 are set to 0.
* estimate.copula is upgraded. Two further variables are introduced. na.rm and max.min enables the user to remove NA values from the data and to set values >= 1 to 1-10^-6 as well as values <= 0 to 10^-6.
* estimate.copula is upgraded. Argument type of tau2theta was not specified in two cases.
2011-07-28 Ostap Okhrin and Alexander Ristig
* The description is cut.