I'm looking for a univariate probability distribution defined for $x \in (-\infty, \infty)$ with the following properties:

  • The PDF is symmetric around the origin ($p(x)=p(-x)$).
  • The derivative of the PDF at $x=0$ (with the limit taken from above) can be specified to lie in some interval $-\infty \dots 0$ (not necessarily inclusive). Note that this implies that the derivative of the PDF may be discontinuous at that point.
  • The distribution should interpolate between a heavy tailed and a normal distribution. The normal distribution must be a special case.
  • The CDF and inverse CDF can be efficiently evaluated (i.e., using well-known functions and without integrating numerically in a typical computer algebra system).
  • The PDF should be as smooth as possible (everywhere except at the origin).

The generalized normal distribution (with $p(x)\sim\exp(-|x|^\beta)$) comes close to this, but $\beta$ specifies the tail behavior as well as its “peakiness” at $x=0$. I would like to be able to specify these two things as independently as possible.

One idea would be to convolve the PDF of a generalized normal with a normal PDF of a specified scale (i.e., adding a normal random variable to the generalized normal), thereby “smoothing” it at the origin. However, I am stuck with computing the resulting distribution in closed form.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.