I have independently generated two (pseudo-)random numbers in Excel.I have then applied these to the CDFs to get two random values (eg. Unit Sales) R = [614,037 66,735].These two random variables have an historical correlation of 0.68, ie. Choleky Decomp (C=LL') gives:L = [1.00 0.00 0.68 0.73]When I generate RL' = [695,436 48,913] and repeat this 1000 times, these resulting (correlated) Unit Sales vectors have a correlation of 0.35 and not 0.68.I must be missing something as I want to generate two (0.68) correlated random numbers. Can someone tell me where I've gone wrong, please?

r ur numbers distributed normally? not uniform, i suppose

QuoteOriginally posted by: ShagbagWhen I generate RL' = [695,436 48,913] and repeat this 1000 times, these resulting (correlated) Unit Sales vectors have a correlation of 0.35 and not 0.68.I must be missing something as I want to generate two (0.68) correlated random numbers. Can someone tell me where I've gone wrong, please?Assuming from your topic title that you intended to write 1.35 rather than 0.35, I notice this is just twice 0.68, rounded down (which might be correct if 0.68 is rounded up). This suggests you have an erroneous factor of 2 in your computation somewhere.

jawabean,My (psuedo-)random numbers are just Excel's RND() function which, I believe, is a uniform distribution.I then map them on to two separate (almost normal) CDFs of Unit Sales figures (one for Console Games, the other for Handheld Games). I believe these resulting Unit Sales figures are '(almost) normal random' figures (am I correct?).Also, are you saying that if I don't have normally generated random numbers then they won't correlate properly?Fermion,The topic title is wrong. I forgot to change it when I edited my original post.

QuoteOriginally posted by: Shagbagjawabean,My (psuedo-)random numbers are just Excel's RND() function which, I believe, is a uniform distribution.I then map them on to two separate (almost normal) CDFs of Unit Sales figures (one for Console Games, the other for Handheld Games). I believe these resulting Unit Sales figures are '(almost) normal random' figures (am I correct?).imortant thing here that chol decomposition will work if your inputs have variance =1 and mean =0, like std gaussian randoms. i dont know if your inputs r standard randoms.