4
$\begingroup$

Let's say we have a known normal distribution $N(\mu,\sigma^2)$.

I now draw 2 points $p1$ and $p2$ randomly from this Gaussian distribution for every observation, and repeat this process large number of times.

  1. What will the distribution of $\frac{p1}{p2}$ look like? Will it be normal? Can we say something about it's mean and standard deviation?

  2. What will the distribution of $\operatorname{max} (\frac{p1}{p2},\frac{p2}{p1})$ look like? Will it be normal? Can we say something about it's mean and standard deviation?

  3. What will the distribution of $\frac{e^p_1}{ e^p_2}$ and the distribution of $\operatorname{max} (\frac{e^p_1}{ e^p_2}, \frac{e^p_2}{e^p_1})$ look like? Will it be normal? Can we say something about it's mean and standard deviation?

$\endgroup$
  • 3
    $\begingroup$ Just about all conceivable variations of (1) are addressed in other threads: search our site. Those answers will also reply to question (2). Question (3), presumably about $\exp(p_1)/\exp(p_2)$ and its reciprocal, is about the lognormal distribution with parameters $\mu-\mu=0$ and $\sigma^2+\sigma^2=2\sigma^2$, which you can find answered in many places. The second part of (3) may be new (albeit straightforward). I would therefore encourage respondents to focus on that. $\endgroup$ – whuber Oct 22 '15 at 5:33
  • 3
    $\begingroup$ Gaussian Ratio distribution $\endgroup$ – Glen_b -Reinstate Monica Oct 22 '15 at 7:04
0
$\begingroup$

For Q 1 and Q 2: This is a gaussian ratio distribution, see https://en.wikipedia.org/wiki/Ratio_distribution#Gaussian_ratio_distribution and https://www.amazon.com/Probability-Distributions-Involving-Gaussian-Variables/dp/0387346570 and search this site. Si I will concentrate an Q3. First part is straightforward, as noted in whuber's coooent, it is lognormal with parameters $\mu=0, 2\sigma^2$. The second part is the a maximum of two related lognormals: $$ \max\left( e^{p_1-p_2}, e^{p_2-p_1} \right) $$ Note that this maximum is given by $e^{|p_1-p_2|}$ so we could call it a log-halfnormal distribution, in this case a half-normal (or absolute value of normal with zero-mean) based on normal with parameters $\mu=0, 2\sigma^2$ with density function $$ f(x) =\frac{2}{\sqrt{2\pi}\sqrt{2\sigma^2}}e^{-\frac12 \frac{x^2}{2\sigma^2}} = \frac1{\sqrt{\pi}\sigma} e^{-\frac14(\frac{x}{\sigma})^2}, \quad x\ge 0 $$ The density of such a half-normal variate exponentiated can be found from first principles. The result becomes $$ f_Y(y) = \frac1{\sqrt{\pi}\sigma} e^{-\frac14 (\frac{\log y}{\sigma})^2}, \quad y> 1 $$ expectation and variance can be found to be $$ 2 e^{\sigma^2} \Phi(\sqrt{2}\sigma), \\ 2e^{4\sigma^2} \Phi(2\sqrt{2}\sigma) - 4e^{2\sigma^2} \Phi(\sqrt{2}\sigma)^2 $$ (calculated with help of maple).

$\endgroup$

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.