Infinite Variance Probability Distribution

219 Views Asked by At

I have data $\vec{X}=(x_{1}, ..., x_{n})$ with size $N$, sampled from a physical phenomenon. The data takes discrete values in the range $[-N,N]$. As $N$ is increased, the sample variance and fourth order moments explode. I obtained a theoretical result about the problem that implies as $N\rightarrow +\infty$, then $Variance(\vec{X})\rightarrow \infty$. The mean remains finite. If I am not mistaken, this means that as the distribution converges to be continuous, then the variance tends to $+\infty$. I know that a discrete distribution cannot have infinite variance, of course.

The following figure (1) shows one example sample obtained with $N=100$. The y-axis shows the relative frequencies and the x-axis the value of $\vec{X}$. This figure shows another sample with a kernel estimate 2.

I am looking for a distribution that could model this type of data, whether in its continuous limit or discrete. Distributions with infinite variance I am aware of are continuous and exhibit some type of power law, which my estimated samples PMF do not appear show. My question is: Do you know any distributions that could model this type of data? Please tell me if I am missing any important information. Any reference or hint would be extremely helpful. Thank you!

1

There are 1 best solutions below

0
On

Given the information you have provided it is hard to tell which limiting distribution best describes your data. However, e.g. the student's t-distribution with $1<v\leq 2$ degrees of freedom has a finite mean yet inifite variance. Its support is also the whole real line.