A Preliminary Discussion of Heavy- and Light-Tailed Distributions
User Rating: / 1
PoorBest 
Written by theoretic   
The Danish fire insurance data and the US industrial fire data presented in Figures 3.2.5 and 3.2.15, respectively, can be modeled by a very heavy-tailed distribution. Such claim size distributions typically occur in a reinsurance portfolio, where the largest claims are insured. In this context, the question arises:
What determines a heavy-tailed/light-tailed claim size distribution?
There is no clear-cut answer to this question. One common way to characterize the heaviness of the tails is by means of the exponential distribution as a benchmark. For example, if
F(x)
limsup г— < oo for some A > 0,
— ЛХ
x—>oo e
where
F(x) = 1 — F(x), x > 0,
denotes the right tail of the distribution function F, we could call F light-tailed, and if
F(x)
liminf г— > 0 for all A > 0,
x^cx, e ~Xx
we could call F heavy-tailed.
Example 3.2.4 (Some well-known heavy- and light-tailed claim size distributions)
From the above definitions, the exponential Exp(A) distribution is light-tailed for every A > 0.