This article identifies caveats of using the correlation coefficient between two fractionally integrated time series as a measure of association between them, and compares various procedures for removing the long-range dependence component in the output series of the G7.