On the correspondence between data revision and trend-cycle decomposition