9
$\begingroup$

I know first $N$ moments of some distribution. I also know that my distribution is continuous, unimodal and well shaped (it looks like gamma-distribution). Is it possible to:

  1. Using some algorithm, generate samples from this distribution, which in limit conditions will have exactly the same moments?

  2. Solve this problem analytically?

I understand that until I have an infinite number of moments, this question cannot have a unique solution. I would be happy to have any.

Due to the comments clarification: I don't need to restore original distribution. I need ANY with a given moments.

$\endgroup$
12
  • $\begingroup$ How do you define "well shaped" ? $\endgroup$
    – Tim
    Apr 19 '18 at 7:31
  • $\begingroup$ @Tim It looks like gamma-distribution. I've edited the question accordingly. $\endgroup$
    – zlon
    Apr 19 '18 at 7:43
  • 1
    $\begingroup$ You cannot generate from a distribution for which you only know moments. There even exist cases where the whole sequence of moments is not sufficient to specify the distribution uniquely. $\endgroup$
    – Xi'an
    Apr 19 '18 at 9:20
  • 1
    $\begingroup$ I don't need the unique distribution. I need ANY with given moments. $\endgroup$
    – zlon
    Apr 19 '18 at 9:21
  • 2
    $\begingroup$ If any solution suffices, use your data. $\endgroup$
    – Nick Cox
    Apr 19 '18 at 11:35
5
$\begingroup$

We really need that you give some more information as asked for in comments.

There is a monograph https://projecteuclid.org/euclid.lnms/1249305333 dedicted to your question.

Here: http://fks.sk/~juro/docs/paper_spie_2.pdf is another paper.

Some related posts on sister sites:

https://math.stackexchange.com/questions/386025/finding-a-probability-distribution-given-the-moment-generating-function

https://mathoverflow.net/questions/3525/when-are-probability-distributions-completely-determined-by-their-moments

Another paper is http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.106.6130 Its author lists some possible approaches, like maximum entropy methods (Jaynes 1994), a method of obtaining upper and lower bounds on the cumulative distribution function (cdf) using the first $n$ moments (https://www.semanticscholar.org/paper/A-moments-based-distribution-bounding-method-R%C3%A1cz-Tari/cd28087b8ead5c4d5c4eebc2b91e2a4b8caef3f3), but then diced to assume a unimodal distribution and fit to a flexible distribution family, like Pearson family, Johnson family or Generalized Tukey Lambda family. Finally she implements a solution based on fitting first four moments to the Generalized Lambda family.

$\endgroup$

This site is temporarily in read only mode and not accepting new answers.

Not the answer you're looking for? Browse other questions tagged .