1
$\begingroup$

There are two (price) distributions of the same class, but they differ in parameter values. One distribution has a smaller upper bound and a greater lower bound, so intuitively we know it has a smaller dispersion. Unfortunately, it is very difficult to calculate the variances. Is there another formal way to establish the claim (that one distribution has a smaller dispersion)?

$\endgroup$
3
  • $\begingroup$ How do you measure dispersion? $\endgroup$ Jul 13 at 21:25
  • $\begingroup$ Why can't you calculate variance? $\endgroup$
    – Tortar
    Jul 13 at 21:30
  • $\begingroup$ @Michael Greinecker It is a good question. If I am unable to calculate variance (because it is a complicated distribution function), what is a good alternative measure of "dispersion"? I don't necessarily need to use the word "dispersion". I used it only to convey the idea. $\endgroup$
    – Adam
    Jul 13 at 21:47
2
$\begingroup$

The following might help, although whether it's simpler than calculating the variances will depend on the particular functions. Suppose the two distributions are of random variables $x_1$ and $x_2$. First find the respective means $\mu_1$ and $\mu_2$. Then replace $x_1$ by $y_1=x_1-\mu_1$ and $x_2$ by $y_2=x_2-\mu_2$, with the effect of shifting the distributions so that they both have mean zero while preserving their respective dispersions. Let the cumulative probability distributions after this shifting be:

$$F_{Y_1}(y_1)=P(Y_1\leq y_1)$$ $$F_{Y_2}(y_2)=P(Y_2\leq y_2)$$

Comparing these distributions (the whole distributions, not just the upper and lower bounds), it may (or may not) be found that:

$$\forall y_1,y_2 <0, F_{Y_1}(y_1) = F_{Y_2}(y_2) \implies y_1>y_2$$

and

$$\forall y_1,y_2 >0, F_{Y_1}(y_1) = F_{Y_2}(y_2) \implies y_1<y_2$$

Those conditions, if satisfied, would show that distribution 1 is less dispersed than distribution 2.

$\endgroup$
2
  • $\begingroup$ Thanks for the suggestion. Unfortunately for me, even the mean does not have a closed-form solution. I should have also made it clearer, these are two (price) distributions of the same class that differ in only 1 parameter value. $\endgroup$
    – Adam
    Jul 14 at 17:32
  • $\begingroup$ Can you collect several sample points? It seems unreasonable that a price distribution is lacking a mean and variance $\endgroup$ Jul 15 at 13:50

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.