2
$\begingroup$

Let $X_{(1)}\leq X_{(2)}\leq \cdots X_{(n)}$ be the order statics for a random sample from a continuous distribution with c.d.f. $F(x)$ and density $f(x)$. Define $U_{i}$, $i=1,2\ldots,n,$ by $$U_{i}=\frac{F(X_{(i)})}{F(X_{(i+1)})}, \quad i=1,\ldots,n-1, \: \mbox{and }\: U_{n}=F(X_{(n)}).$$

Find the joint distribution of $U_{1},\ldots,U_{n}$.

Remark: I need a suggestion or someone tell me in which book I can find this exercise or related theory that allows me to solve it.

$\endgroup$
1
  • $\begingroup$ The problem is equivalent to finding the joint distribution of $(\frac{Y_{(1)}}{Y_{(2)}},\cdots,\frac{Y_{(n-1)}}{Y_{(n)}}, Y_{(n)})$ where $Y_1,\cdots,Y_n$ is a random sample from $U(0,1)$ population. As one knows the joint distribution of the $Y_{(i)}$'s, a simple change of variables $Z_i=\frac{Y_{(i)}}{Y_{(i+1)}}$ for $i=1,2,\cdots,n-1$ and $Z_n=Y_{(n)}$ does the job. Indeed, $Z_1,\cdots,Z_n$ are independent. $\endgroup$ Commented Apr 13, 2018 at 18:35

2 Answers 2

1
$\begingroup$

As noted in previous answer, $V_{(i)}=F(X_{(i)})$ are order statistics for independent Uniform samples $V_1,\ldots,V_n$, where $V_i=F(X_i)$. Joint density function for $(V_{(1)},\ldots, V_{(n)})$ is (see here) $$ f_{V_{(1)},\ldots, V_{(n)}}(v_1,\ldots,v_n)=\begin{cases}n!, & 0\leq v_1\leq\ldots\leq v_n\leq 1,\cr 0 & \text{otherwise}.\end{cases} $$ Let us find c.d.f. for $U_{1},\ldots,U_{n}$. For all $t_1,\ldots,t_n\in(0,\,1)$ $$ P(U_{1}<t_1,\ldots,U_{n}<t_n) = P\left(\frac{V_{(1)}}{V_{(2)}}<t_1,\ \ldots, \ \frac{V_{(n-1)}}{V_{(n)}}<t_{n-1},\ V_{(n)}<t_n\right) = $$ $$=P\left(V_{(1)} < t_1V_{(2)},\ \ldots, \ V_{(n-1)} < t_{n-1}V_{(n)},\ V_{(n)}<t_n\right) = $$ $$ =\int\limits_0^{t_n}dv_n\int\limits_0^{t_{n-1}v_n}dv_{n-1}\cdot\dots\cdot\int\limits_0^{t_3v_4}dv_3\int\limits_0^{t_2v_3}dv_2\int\limits_0^{t_1v_2}n!\,dv_1 = t_1t_2^2t_3^3\cdot\dots\cdot t_{n-1}^{n-1}t_n^n. $$

$\endgroup$
2
  • $\begingroup$ I think the condition $0\leq v_{1}\leq \cdots \leq v_{n}$ over $f_{V_{(1)},\ldots,V_{(n)}}(v_{1},\ldots,v_{n})$ has influence in the process of integration, this implies that the distribution function must wrote by cases. $\endgroup$ Commented Feb 23, 2017 at 3:41
  • $\begingroup$ @DiegoFonseca, this inequalities are consequences of inequalities $0\leq v_1$, $v_1\leq t_1v_2$ and so on. Say, $v_1\leq t_1v_2$ implies $v_1\leq v_2$ since $t_1\leq 1$. $\endgroup$
    – NCh
    Commented Feb 23, 2017 at 3:50
0
$\begingroup$

First, for any random variable $X$ with cdf $F$ which is invertible away from 0,1 (so density positive everywhere on the domain would do it), $F(X)$ is $\text{Uniform}(0,1)$. So let $V_i=F(X_i)$ and notice that $V_{(i)}=F(X_{(i)})$ as well. So you're looking at the ratio of ordered uniform random variables $V_{(i)}$. This reminds me of similar questions on Poisson-Dirichlet Distributions, but they're not ordering Uniform random variables and they have an infinite number of random variables. An indepth introduction into this can be found here and a simpler explanation of the distribution here, Section 3.

The best hope on finding more information like this is to maybe look at just the Dirichlet distribution, see if anyone has worked on the ratio of the ordering of its coordinates, or see if you can work through the Poisson-Dirichlet material and come up with your own proofs.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .