Conversion between parametrizations of a skew-elliptical distribution
Convert direct parameters (DP) to centred parameters (CP) of a skew-elliptical distribution and vice versa.
1 2 3 4
a vector (in the univariate case) or a list (in the multivariate
case) as described in
a vector or a list, in agreement with
a vector or a list, in agreement with
a characther string with the family acronym,
as described in
optionally, an S4 object of class
character string, which has effect only if
numeric value (in
dp2cp, a matching vector (in the univariate case) or a list
(in the multivariate case) of
op2dp, a similar object of
provided the set of input parameters is in the admissible region.
dp2op, a similar set of
For a description of the DP
parameters, see Section ‘Details’ of
CP form of parameterization is cumulant-based. For a univariate
distribution, the CP components are the mean value (first cumulant),
the standard deviation (square root of the 2nd cumulant), the coefficient of
skewness (3rd standardized cumulant) and, for the ST,
the coefficient of excess kurtosis (4th standardized cumulant).
For a multivariate distribution, there exists an extension based on the
same logic; its components represent the
vector mean value, the variance matrix, the vector of marginal coefficients of
skewness and, only for the ST, the Mardia's coefficient of excess
kurtosis. The pseudo-CP variant provides an ‘approximate form’ of
CP when not all required cumulants exist; however, this parameter set
is not uniquely invertible to DP. The names of pseudo-CP
components printed in summary output are composed by adding a
after the usual component name; for example, the first one is denoted
Additional information is provided by Azzalini and Capitanio (2014). Specifically, their Section 3.1.4 presents CP in the univariate SN case, Section 4.3.4 CP for the ST case and the 'pseudo-CP' version. Section 5.2.3 presents the multivariate extension for the SN distribution, Section 6.2.5 for the multivariate ST case. For a more detailed discussion, see Arellano-Valle and Azzalini (2013).
The OP parameterization is very similar to DP, from which it differs only for the components which regulate dispersion (or scatter) and slant. Its relevance lies essentially in the multivariate case, where the components of the slant parameter can be interpreted component-wise and remain unaffected if marginalization with respect to some other components is performed. In the multivariate SN case, the components of OP, denoted ξ, Ψ, λ, are associated to the expression of the density function (5.30) of Azzalini \& Capitanio (2014); see pp.128–131 for more information. In the univariate case, the slant component of DP and the one of OP coincide, that is, α=λ, Parameter ξ and other parameters which may exist with other families remain the same of the DP set. The term OP stands for ‘original parameterization’ since this is, up to a negligible difference, the parameterization adopted by Azzalini & Dalla Valle (1996).
While any choice of the components of DP or OP is
admissible, this is not true for CP. An implication is that a
cp2dp may fail with an error message
for certain input values. The most extreme case is represented by the
SC family, for which CP never exists; hence it makes
to sense to call
It is possible to call the functions with
cp having more
components than those expected for a given family as described above and in
makeSECdistr. In the univariate case, this means that
cp can be vectors of longer length than indicated earlier. This
occurrence is interpreted in the sense that the additional components after
the first one are regarded as regression coefficients of a
and they are transferred unchanged to the matching components of the
transformed parameter set; the motivation is given in Section 3.1.4 of
Azzalini and Capitanio (2014). In the multivariate case,
cp[] can be matrices instead of vectors; the rows beyond the first
one are transferred unchanged to
Arellano-Valle, R. B. and Azzalini, A. (2013, available on-line 12 June 2011). The centred parameterization and related quantities of the skew-t distribution. J. Multiv. Analysis 113, 73-90.
Azzalini, A. with the collaboration of Capitanio, A. (2014). The Skew-Normal and Related Families. Cambridge University Press, IMS Monographs series.
Azzalini, A. and Dalla Valle, A. (1996). The multivariate skew-normal distribution. Biometrika 83, 715–726.
the ‘Note’ at
summary.selm for the reason why
CP is the default parameterization in that function and in
the ‘Examples’ at
rmsn for use of the CP
1 2 3 4 5 6 7 8 9 10 11 12
# univariate case cp <- dp2cp(c(1, 2222, 3333, 2, 3), "SN") dp <- cp2dp(cp, "SN") # notice that 2nd and 3rd component remain unchanged # # multivariate case dp3 <- list(xi=1:3, Omega=toeplitz(1/(1:3)), alpha=c(-3, 8, 5), nu=6) cp3 <- dp2cp(dp3, "ST") dp3.back <- cp2dp(cp3, "ST") # op3 <- dp2op(dp3, "ST") dp3back <- op2dp(op3,"ST")
Want to suggest features or report bugs for rdrr.io? Use the GitHub issue tracker.