Combination of 2 random variables that is perfectly correlated with another random variable

I'd appreciate some pointers for the following question: Given 3 random variables $X_i$, $i = 1, 2, 3$ such that none of them have a pairwise correlation of $+1$ or $−1$. Can we find a combination of $X_1$ and $X_2$ that has a correlation of $+1$ with $X_3$?

-
 This is going to depend on the nature of the underlying relationships amongst these variables. BTW, is this homework? If so, please edit to add the homework tag. – gung Oct 21 '12 at 21:23

Yes, it's possible, but not likely to occur in real life. An artificial example

x1 <- rnorm(100)
x2 <- rnorm(100)
x3 <- x1+x2
cor(x1, x2)
cor(x1, x3)
cor(x2, x3)


and, of course, since x3 = x1 + x2, the correlation between (x1 + x2) and x3 is 1.0.

If you had specified that the three variables are independent, I don't think it's possible, but I am not 100% sure.

-
If $X_1$ and $X_2$ are uncorrelated with $X_3$, the linear combination $aX_1+bX_2$ is uncorrelated with $X_3$ since covariance is bilinear. – Douglas Zare Oct 21 '12 at 11:02
The question didn't say $X_1$ and $X_2$ were uncorrelated with $X_3$, just that the correlations weren't 1.0 or -1.0. – Peter Flom Oct 21 '12 at 11:13
Yes, I was referring to your last line, which talks about what might happen if the variables are pairwise independent, hence uncorrelated. – Douglas Zare Oct 21 '12 at 11:26
Oh, sorry. Cool, thanks for that. That was my intuition, but it's nice to know I was right. :-). – Peter Flom Oct 21 '12 at 11:30