They seem to be computing the correlation of price levels (which are completely meaningless) rather than the correlation of price difference or returns. Any tw

They seem to be computing the correlation of price levels (which are completely meaningless) rather than the correlation of price difference or returns. Any two time series which broadly moved up over the same time period will look correlated if you compute the correlation of levels. But they could in fact be completely uncorrelated (and therefore have zero predictive power) when you compute the correlation on returns, which is what you should be doing.