Issue: Vol 42 No 2 (2008)

One possibility to construct heavy tail distributions is to directly manipulate a
standard Gaussian random variable by means of transformations which satisfy
certain conditions. This approach dates back to Tukey (1960) who introduces the
popular -transformation. Alternatively, the -transformation of MacGillivray
& Cannon (1997) or the -transformation of Fischer & Klein (2004) may be used.
Recently, Klein & Fischer (2006) proposed a very general power kurtosis transformation
which includes the above-mentioned transformations as special cases.
Unfortunately, their transformation requires an innite number of unknown parameters
to be estimated. In contrast, we introduce a very simple method to
construct exible kurtosis transformations. In particular, manageable “superstructures”
are suggested in order to statistically discriminate between -, – and
-distributions (associated to -, – and -transformations).

Fulltext