6

Here's something I've wondered about for a while, but haven't been able to discover the correct terminology. Say you have a relatively complicated density function that you suspect might have a close approximation as a sum of (properly weighted) simpler density functions. Have such things been studied? I'm particularly interested in reading about any applications.

Here's one example I've found:

Expansion of probability density functions as a sum of gamma densities with applications in risk theory

Christopher D. Long
  • 370
  • 2
  • 13
  • You could also look into the [saddlepoint approximation](https://stats.stackexchange.com/questions/191492/how-does-saddlepoint-approximation-work/191781#191781). – kjetil b halvorsen Sep 27 '19 at 17:26

3 Answers3

4

You can also use Edgeworth series, if your random variable has a finite mean and variance, which expands the CDF of your random variable in terms of the Gaussian CDF. At first glance it's not quite as tidy conceptually as using a mixture model, but the derivation is quite pretty and it gives you a closed form with very fast decay in the tail terms.

Rich
  • 4,336
  • 1
  • 20
  • 20
4

Histogram density estimator is estimating the density with a sum of piecewise functions (density of a uniform).

KDE is using a sum of smooth function (gaussian is an example) (as long as they are positive they can be transformed into a density by normalization)

The use of "mixture" in statistic is about convex combination of densities.

robin girard
  • 6,335
  • 6
  • 46
  • 60
3

You can do this with mixture modeling. There are a number of R packages on CRAN for doing this. Search for "mixture" at http://cran.r-project.org/web/packages/

Rob Hyndman
  • 51,928
  • 23
  • 126
  • 178