Also, I don't buy the correlation evidence - if you stretch any time series long enough, everything is pretty much correlated to one another and falsely yields a high R-squared - But the ebb and flows in between aren't captured and it creates a conditional bias in the regression analysis - meaning you miss out on short-term trading opportunities...
That being said, correlation analysis is still being used by professional funds - I have colleagues that utilize them, but they are for diversified portfolio worth billions and these funds generally don't have the freedom to trade the short-term ebb-and-flows due to liquidity constraints. So they rely on a more systematic approach (regression analysis and multi-factor APT models) to determine their asset weightings over a longer time horizon.