Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

The following is a classic example that pairwise independent does not necessarily imply mutually independent:

Let $X_1$ and $X_2$ be independent r.v.'s with distributions $$P(X_i=1)=P(X_i=-1)=\frac{1}{2}\quad\tag{*}$$ for $i=1,2.$ Let $Z=X_1X_2$. Then $X_1,X_2,Z$ are pairwise independent but they are not mutually independent.

In this case $X_1$ and $Z=X_1X_2$ are independent. Now let $(X_i)_{i=1}^{\infty}$ be a family of independent variables which satisfy $(*)$ and let $Z_i=X_1X_2\cdots X_i$. How can I show that $(Z_i)_{i=1}^{m}$ are independent for any finite $m\in{\Bbb N}$?


When $m=2$, it is done. If one needs induction, then a key step is to show that the $n-1$ dimensional random vector $(Z_1,Z_2,\dots,Z_{n-1})$ and $Z_n$ are independent. This where I have no idea how to go on after writing down the definition.

share|improve this question

2 Answers 2

up vote 2 down vote accepted

Let $\epsilon_i=-1$ or $1$. We have $$P\left(\bigcap_{j=1}^n\{Z_j=\varepsilon_j\}\right)=P\left(\bigcap_{j=1}^n\{Z_j=\varepsilon_j\}\mid \{X_n=-1\}\right)P(X_n=-1)+P\left(\bigcap_{j=1}^n\{Z_j=\varepsilon_j\}\mid \{X_n=1\}\right)P(X_n=1),,$$ hence $$P\left(\bigcap_{j=1}^n\{Z_j=\varepsilon_i\}\right)=\frac 12P\left(\bigcap_{j=1}^{n-1}\{Z_j=\varepsilon_i\}\cap\{Z_{n-1}=-\epsilon_n\}\right)+\frac 12P\left(\bigcap_{j=1}^{n-1}\{Z_j=\varepsilon_i\}\cap\{Z_{n-1}=\epsilon_n\}\right).$$ We just deal with the case $\epsilon_{n-1}=\epsilon_n$. Then $$P\left(\bigcap_{j=1}^n\{Z_j=\varepsilon_j\}\right)=\frac 12P\left(\bigcap_{j=1}^{n-1}\{Z_j=\varepsilon_j\}\right)=\frac 12\prod_{j=1}^{n-1}P\left(\{Z_j=\varepsilon_j\}\right),$$ which is what we want. The other case is similar.

share|improve this answer
    
But you've already have $P(X_n=-1)$, how did you come up with $\frac 12$? –  Jack Oct 17 '12 at 15:37
    
I don't understand: this probability is $1/2$. –  Davide Giraudo Oct 17 '12 at 15:41
    
What i'm saying is that $$ P\left(\bigcap_{j=1}^n\{Z_j=\varepsilon_j\}\mid X_n=-1\right)P(X_n=-1)=P\left(\bigcap_{j=1}^{n-1}\{Z_j=\varepsilon_i\}\cap\{Z_{n‌​-1}=-\varepsilon_n\}\right).$$ Why do you need $\frac 12$? –  Jack Oct 17 '12 at 15:45
    
Also, why is $\varepsilon_i=0,1$ instead of $\varepsilon_i=-1,1$? –  Jack Oct 17 '12 at 15:51
    
I go back to the definition of conditional probability, using independence of $X_n$ with respect to $(Y_1,\dots,Y_{n-1})$. –  Davide Giraudo Oct 17 '12 at 19:45

For example, $P(Z_4 | Z_3 Z_2 Z_1) = P(Z_4 | Z_3)$ (because $Z_4 = X_4 Z_3 $, with $X_4$ independent from $Z_3$; hence, if $Z_3$ is known, $Z_2$, $Z_1$ do not give more information; in other words, $Z_i$ is a first-order Markov sequence)

But, say $P(Z_4 = 1| Z_3=1)= P(X_4=1) =\frac{1}{2} $ and the same happens in all cases, hence $P(Z_4 | Z_3 Z_2 Z_1) = P(Z_4 | Z_3) = P(Z_4) = $ and $Z_4$ is independent of previous values; applying induction, you get the independence of the full set $(Z_i)_{i=1}^{m}$.

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.