In one of Fisher's classical paper [1] I stumbled over the following:
If the frequency with which the variate $x$ falls into the range $dx$, be given by $$df = \frac{1}{\pi}\frac{dx}{1+(x-m)^2}$$ where $m$ is the unknown parameter representing the centre of the symmetrical frequency curve of $x$, then it is not difficult to show that the arithmetic mean of any number of independent values of $x$, will be distributed in exactly the same distribution as a single value of $x$.
As Fisher called this 'not difficult', I would like to prove the statement ... but I fail :-).
I assume $\frac{df}{dx}$ represents the pdf $g(\cdot)$ of a random variable. If I put Fisher's statement in other words, it says: if there are two random variables $X$ and $Y$, both with pdf $g(\cdot)$, their mean has also $g(\cdot)$ as its pdf. So I guess that the convolution of $g(\cdot)$ with itself and divided by two - which represents the pdf of the mean - should be equal to $g(\cdot)$. However, when I try to calculate this convolution, I arrive at complex integrals which are not at all 'not difficult'.
Does anyone know how to actually prove this in a simple way?
[1] Fisher, R. A. (1925, July). Theory of statistical estimation. In Mathematical Proceedings of the Cambridge Philosophical Society (Vol. 22, No. 05, pp. 700-725). Cambridge University Press.