4
$\begingroup$

I want to find the moment generating function (mfg) and mean deviation of this distribution:

$$f(x,\epsilon,k,\theta) = k\theta^{(1+1/k+\epsilon/k)}x^{(k+\epsilon)}\exp{(-\theta x^k )}/(\Gamma(1+(1+\epsilon))/k)$$

where $\epsilon, k, \theta$ are the three parameters of this distribution. $x$ ranges from $0$ to infinity.

$\endgroup$
  • 1
    $\begingroup$ Hi zahida and welcome to the site. I changed your formula to proper LaTeX code. Please check carefully if they are still correct. $\endgroup$ – COOLSerdash Jun 21 '14 at 17:24
  • 2
    $\begingroup$ This is not a well-defined distribution until you specify the range of $x$. $\endgroup$ – whuber Jun 21 '14 at 17:31
  • 1
    $\begingroup$ And specify any constraints on the parameters/ $\endgroup$ – wolfies Jun 21 '14 at 17:33
  • 3
    $\begingroup$ If we assume the range is $0\le x \lt \infty$ and reasonable values for the parameters, then when $k\gt 0$ is rational the mgf can be expressed as a finite algebraic combination of generalized hypergeometric functions. $\endgroup$ – whuber Jun 21 '14 at 17:37
  • 1
    $\begingroup$ It looks a bit like a Stacy distribution. $\endgroup$ – wolfies Jun 21 '14 at 17:38
2
$\begingroup$

This is a Generalized Gamma distribution: up to a scale factor, $x^k$ has a Gamma distribution.

E. W. Stacy studied the Generalized Gamma distribution in a 1962 paper available on Project Euclid. His parameterization is more natural for statistical applications and leads to simpler formulas. The density function is determined by the density for a unit scale factor $a=1$ by

$$f(x;1,d,p)\mathrm{d}x =\frac{1}{C(p,d)} x^d \exp(-x^p) \frac{\mathrm{d}x}{x}$$

for $0 \lt x \lt \infty$ (see Stacy equation (1)). The constant of proportionality is readily found via the substitution $x^p=u,$ yielding

$$C(p,d) = \int_0^\infty x^d \exp(-x^p)\frac{\mathrm{d}x}{x} = \frac{\Gamma(d/p)}{p}.$$

To find the $r^\text{th}$ moment, integrate $x^r$ against $f\mathrm{d}x.$ No additional calculations are needed, because the preceding formula already gives the result upon replacing $d$ by $r+d$:

$$\mu_r(p,d) = \frac{1}{C(p,d)} \int_0^\infty x^{r+d} \exp(-x^p) \frac{\mathrm{d}x}{x} = \frac{C(p, r+d)}{C(p, d)} = \frac{\Gamma((r+d)/p)}{\Gamma(d/p)}.$$

By definition, these give the terms in the expansion of the moment generating function (mgf),

$$M_{p,d}(t) = \sum_{r=0}^\infty \frac{\mu_r(p,d)}{r!} t^r = \frac{1}{\Gamma(d/p)} \sum_{r=0}^\infty \frac{\Gamma((r+d)/p)}{r!} t^r.$$

(See Stacy equation (3).) As usual, the scale factor $a \gt 0$ is accommodated by replacing $t$ by $a t$ in the mgf.

The parameterization given in the question identifies $k$ with $p,$ $1+k+\epsilon$ with $d$, and $1/\theta^{1/k}$ with $a.$

I presume the "mean deviation" refers to some function of these moments, but since I'm unsure which one, I will leave the algebra to the reader.

$\endgroup$

Your Answer

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

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