0
$\begingroup$

I have two distributions $A$ and $B$ that are i.i.d.

I want to create two distributions $A'$ and $B'$, that have the "same distribution" as $A$ and $B$ (meaning the same probability distribution function), but are correlated (pearson correlation) with correlation $\rho$.

My first approach was to set $A' = B' = (A+B)/2$, but this works only for perfect correlation (or $\rho = -1$). Furthermore the distribution of $A'$ is not neccessarily equal to the distribution of $A$.

If the general case is to complicated, the problem I work on actually only requires $A$ to be uniform and not any given distribution.

$\endgroup$
1
$\begingroup$

This trick works for any $\rho\in[0,1]$: Flip a biased coin whose heads probability is $\rho$. If heads comes up, let $(A',B') = (A,A)$; otherwise let $(A',B') = (A,B)$.

If you want a negative $\rho$, and if $A$ and $B$ are to be uniform on $[0,1]$, flip a coin with heads probability $|\rho|$ and let $(A',B') = (1-A,A)$ if heads comes up and $(A',B') = (A,B)$ otherwise.

$\endgroup$

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.