Fitting Power-Tail Data with Pareto–Pickands Distribution
The Pareto–Pickands distribution, also known as generalized Pareto distribution (GPD), was introduced to study extreme order statistics and is often used to describe events exceeding some high threshold. The Pareto–Pickands distribution finds applications in insurance, hydrology and climatology.
The Pareto–Pickands distribution can also be used to model exceedance for power‐tail distributions, many of which exist in the Wolfram Language, like CauchyDistribution, DagumDistribution, LogGammaDistribution, ParetoDistribution, StudentTDistribution and more.
The higher moments of this distribution do not exist.
The histogram shows that the sample has a long tail.
Truncate data by keeping large values from the tail.
Fit the tail data to a Pareto–Pickands distribution.
Check goodness of fit with a quantile plot.
Goodness-of-fit test.
An interesting factoid: standard Pareto–Pickands distributions are stochastically ordered; i.e. for 
, cumulative distribution functions are also ordered 
for all 
:
