1 Introduction
The results in this paper are a contribution to knowledge in two areas. First
of all we present new methods for tting the generalized lambda distribution
to data. Secondly, we show the utility of the generalized lambda distribution
in the area of nance where up until now, very little investigation of it has
been undertaken.
The four-parameter generalized lambda distribution (GLD) family is
known for its high
exibility, producing distributions with a huge range
of dierent shapes. The lambda distribution originated with Tukey [1962],
and was generalized by Filliben [1969, 1975], Joiner and Rosenblatt [1971],
and Ramberg and Schmeiser [1972, 1974].
The properties of the GLD were studied in detail by Ramberg et al.
[1979]. The monograph of Karian and Dudewicz [2000] summarizes the status
of research until 2000. Since then several papers have been published
which present and discuss dierent parameter estimation approaches to tting
empirical data to this distribution. Parameter estimation of the GLD
is notoriously dicult through the rapid change of the distributional shapes
by varying the parameters in the dierent regions of the two shape parameters.
In particular, the support of the distribution can change with the
value of the parameters from being the whole real line to an interval which
is innite it only one direction. It is our view that tting should be carried
out only in regions where such dramatic changes do not take place. Our
experience is that otherwise tting routines typically become stuck in local,
not necessarily global optima.
Little attention appears to have been paid to the GLD family in the
area of nance, with the exception of Corrado [2001] and Tarsitano [2004].
The methods we develop in this paper allow the GLD to be easily tted to
empirical data such as nancial returns. Calculations of important nancial
risk measures such as the value at risk, expected shortfall and tail indices
are shown to be very simple for the GLD. We are also able to realistically
simulate data arising from returns from equities. We show that using the
GLD we can simulate data which closely resembles returns of the equities
comprising the NASDAQ-100.
2 The Generalized Lambda Distribution
Ramberg and Schmeiser [1974] introduced the four parameter GLD dened