If I have a sequence of random variables $X_1, X_2, \ldots, X_n$ (possibly infinite) such that all pairwise cdf's are factorized:

$$F(X_i, X_j) = F_i(X_i) F_j(X_j)$$

for all pairs $(X_i, X_j)$, does it mean that the joint cdf is also factorized? That is:

$$F(X_1, \ldots, X_n) = \prod_{i=1}^{n} F_i(X_i)$$

In other words, if I prove that each pair in the sequence is statistically independent of each other, can longer sequences still be non-independent?

It seems to me that they can, but I can't come up with a counter example.


The simplest of the many standard counterexamples is when $(X_1,X_2,X_3)$ takes the values $(1,1,1)$, $(1,0,0)$, $(0,1,0)$ and $(0,0,1)$ all equiprobably.

  • $\begingroup$ hummmm! I see. Pairwise they are all independent, but knowing any two of them completely determines the third. Neat! Thanks. $\endgroup$
    – Rorsa
    Feb 8 '13 at 13:38
  • $\begingroup$ See also mathoverflow.net/questions/23478/… $\endgroup$
    – Terry Tao
    Feb 8 '13 at 16:58

Steven's example is indeed the simplest. See chapter 3 of this book for counterexamples to lots of similar possibilities.

  • $\begingroup$ The link is for: J. M. Stoyanov, Counterexamples in Probability. (With the copyright page ommitted.) I don't know whether this is an authorized on-ine version, or a pirated copy. $\endgroup$ Feb 8 '13 at 15:21
  • $\begingroup$ It's only the table of contents. $\endgroup$ Feb 9 '13 at 0:15

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.