r.v. with the variance the half of the mean I am doing a random experiment. For each value of the variable "secu" I have a lot of points, so I record the mean and the variance.

The white dots are the mean and the black dots are the variance for the variable secu.
Well, it looks like the variance is a fraction of the mean, around the half. 
The experiment is discrete, so I am having discrete outcomes for each secu.
I am asking for a discrete r.v. X that satisfy something like:
E(X(secu))=secu/5
V(X(secu))=E(X)/2=secu/10
Any idea? it cant be binomial, negative binomial nor poisson... ty...
 A: You didn't answer to requests for (link to) data or a context---what does your observed data represent in reality? so we are reduced to guessing. But you do say data are discrete.  Are they half-integers? I ask because if $X$ is a poisson random variable (with equal mean and variance) then $\frac12 X$ do have variance equal to half the mean!  But of course other solutions are possible. Negative binomial cannot have variance less than mean, so is ruled out. A binomial distribution with $p=\frac12$ wil also do. 
Just check the wikipedia pages for discrete distributions and look at the mean/variance formulas, then you can find other possibilities. 
A: To complement kjetil’s answer, are your data derived from a combination of other random variables (even if you don’t model it explicitly)?  I ask because (for example) the sum of two anti-correlated poisson random variables will have variance less than the mean.  More generically, processes that derive from sums/products of anti-correlated variables often have low coefficients of variation.
A: 
I am asking for a discrete r.v. X such that its variance is half its mean

The simplest possible solution is a Bernoulli random variable with $p = \frac12$ (i.e. a fair coin), for which the mean = $\frac12$ and the variance = $\frac14$.
More generally, the same holds true for any $\text{Binomial}(n,\frac12)$ random variable, which has mean $\frac{n}{2}$ and variance $\frac{n}{4}$.
It also holds for:

*

*$\text{Geometric}(p)$ random variable defined on {1, 2, 3, ...} when $p = \frac23$, in which case the mean is $\frac32$ and the variance $\frac34$.

