4
$\begingroup$

Given the probability density function $f(x) = \frac{1}{x^2}$ for $x > 1$. The expected value of this function is: $$E[X] = \int_1^\infty \frac{1}{x} dx = \infty$$ Can someone please explain this? Is $E[X]$ only finite under certain condition? Is $\infty$ really the expected value for this probability density function?

$\endgroup$
  • 3
    $\begingroup$ Everything is fine. The expectation is infinite, or, others might say, does not exist. For a more important example, please see the Cauchy distribution (Wikipedia). $\endgroup$ – André Nicolas Nov 16 '15 at 7:49
  • $\begingroup$ Thank you. I never encountered this before. $\endgroup$ – Paichu Nov 16 '15 at 7:51
  • 2
    $\begingroup$ You are welcome. For a fun example, please see the St. Petersburg Paradox. $\endgroup$ – André Nicolas Nov 16 '15 at 7:53
2
$\begingroup$

This is an example of a Pareto distribution which typically has a density function of the form $$f(x) = \dfrac{\alpha x_{m}^{\alpha}}{x^{\alpha+1}} \text{ for }x \gt x_m $$ and so a cumulative distribution function of $$F(x) = 1-\left(\dfrac{x_{m}}{x}\right)^{\alpha} \text{ for }x \gt x_m $$ where $x_m \gt 0$ is lowest value of the support (a location parameter, $x_m=1$ in your question) and $\alpha \gt 0$ is a shape parameter ($\alpha =1$ in your question).

Trying to derive the mean would give $\dfrac{\alpha}{\alpha-1} x_m$, but this is only meaningful when $\alpha \gt 1$. For $0 \lt \alpha \le 1$ the mean is infinite, as you found with your example.

Something similar happens with the variance: you can derive $\dfrac{\alpha^2}{(\alpha-1)^2(\alpha-2)}x_m^2$ but this is only valid for $\alpha \gt 2$. For $1 \lt \alpha \le 2$ the variance is infinite, while for $0 \lt \alpha \le 1$ the variance is undefined (since there is no finite mean) and the second moment is infinite.

The Pareto distribution is a fat-tailed distribution. The smaller $\alpha$ is, the fatter the tail becomes. With $\alpha = 1$ the tail is sufficiently fat to prevent the distribution having a finite mean.

It is still possible to descibe a centre of the distribution: for example in general the median of a Pareto distribution is $2^{1/\alpha} x_m$ and in your particular example the median is $2$.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Thank you, that was really helpful. $\endgroup$ – Paichu Nov 16 '15 at 15:42

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.