I am interested in the Bounded Pareto Distribution but I can't find any reference that supplies me the moment generating function. This does however seem like something that is known.

More general : Is there some (online/book/...) reference where I can find a list of all (or at least many) distributions and their most important properties (like moments, PDF, CDF,...) I have been mainly using wikipedia but it has now let me down.

EDIT:

From the provided comment, I have found that a general formula for the moment generating formula for the MGF might not be avaible. I do however still wonder how to obtain it numerically. When I use the moments (for which there does exist a general formula) and the taylor series expansion to obtain the MGF, I find that (when the upper bound is somewhat large) the moments quickly explode and (at least numerically) the Taylor series diverges.

  • One canonical reference for probability distributions is Johnson-Kotz-Balakrishnan's discrete/continuous (both multivariate and univariate) books. – StubbornAtom Apr 9 at 9:39
  • Don't think the mgf exists; it doesn't exist in the usual sense even for the standard pareto distribution. – StubbornAtom Apr 9 at 10:43
  • 1
    For a bounded Pareto it must exists since if $L \leq X \leq H$ then we have for any $t: \mathbb{E}[e^{tX}] \leq \max\{e^{-tL}, e^{tH}\}$ – Darkwizie Apr 9 at 10:45
  • You can see it form a incomplete gamma integral when the argument of the mgf is negative. – BGM Apr 9 at 11:54
  • @Darkwizie - yes, and since $0 \le L$ that simplifies to $\mathbb{E}[e^{tX}] \leq e^{tH}$ – Henry Apr 9 at 16:59
up vote 1 down vote accepted

I'm assuming that if you had a general formula for the moments that you wouldn't need the MGF.

Using Mathematica the general formula for the $k$-th moment is found in the following manner:

f[x_] := \[Alpha] L^\[Alpha] x^(-\[Alpha] - 1) /(1 - (L/H)^\[Alpha])
muk = Integrate[x^k f[x], {x, L, H}, Assumptions -> {\[Alpha] > 0, L > 0, H > L, k >= 0}]

$$\mu_k=\frac{\alpha \left(H^{\alpha } L^k-H^k L^{\alpha }\right)}{(\alpha -k) \left(H^{\alpha }-L^{\alpha }\right)}$$

This formula clearly has a problem when $k=\alpha$. So when $k=\alpha$ we can solve for that moment directly:

mualpha = Integrate[x^\[Alpha] f[x], {x, L, H}, Assumptions -> {\[Alpha] > 0, L > 0, H > L}]

$$\frac{\alpha L^{\alpha } \log \left(\frac{H}{L}\right)}{1-\left(\frac{L}{H}\right)^{\alpha }}$$

This is also what one obtains when taking the limit of the general formula as $k\rightarrow\alpha$.

* 2nd Update*

There is a relatively closed-form solution for the MGF but one needs to be a bit more careful when extracting the moments. First, the closed-form solution:

$$\text{mgf}=\sum _{k=0}^{\infty } \frac{t^k \left(\alpha \left(H^{\alpha } L^k-H^k L^{\alpha }\right)\right)}{k! \left((\alpha -k) \left(H^{\alpha }-L^{\alpha }\right)\right)}$$

$$=-\frac{\alpha \left(L^{\alpha } (-H t)^{\alpha } \Gamma (-\alpha ,0,-H t)-H^{\alpha } (-L t)^{\alpha } \Gamma (-\alpha ,0,-L t)\right)}{L^{\alpha }-H^{\alpha }}$$

where $\Gamma(a,z_0,z_1)$ is the generalized incomplete gamma function (just as suggested in a comment by @BGM).

mgf = Sum[(((H^\[Alpha] L^k - H^k L^\[Alpha]) \[Alpha])/((H^\[Alpha] -
        L^\[Alpha]) (-k + \[Alpha]))) t^k/k!, {k, 0, \[Infinity]}]

To obtain the $k$-th moment one would usually differentiate $k$ times and then set $t=0$ but that won't work with this MGF. If $\alpha$ is not a positive integer, then you'll need to take the $k$-th derivative followed by taking the limit of that derivative as $t\rightarrow 0$. Using Mathematica commands that translates to the following when $k=3$:

Limit[D[mgf, {t, 3}], t -> 0]

$$\frac{\alpha \left(L^3 H^{\alpha }-H^3 L^{\alpha }\right)}{(\alpha -3) \left(H^{\alpha }-L^{\alpha }\right)}$$

If $\alpha$ is a positive integer (say $\alpha=5$, then two limits need to be taken:

$$\lim_{\alpha \to 5} \, \lim_{t\to 0} \, \frac{\partial ^3\text{mgf}}{\partial t^3}=\frac{5 \left(H^5 L^3-H^3 L^5\right)}{2 \left(H^5-L^5\right)}$$

with the associated Mathematica commands:

Limit[Limit[D[mgf, {t, 3}], t -> 0], \[Alpha] -> 5]
  • I really like/appreciate your answer (therefore I have upvoted it) but I do in fact take interest in having a way to compute the MGF – Darkwizie Apr 9 at 19:22
  • However, I accept your answer :)! – Darkwizie Apr 9 at 19:22
  • @Darkwizie. General form of the MGF added. – JimB Apr 10 at 4:15
  • Wonderful, thank you! – Darkwizie Apr 10 at 21:05

Your Answer

 

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Not the answer you're looking for? Browse other questions tagged or ask your own question.