Trends and cycles in real-time

Abstract

This paper compares the results of applying several detrending methods to the Chilean monthly economic activity index (IMACEC) using real-time data sets. We show that data revisions are extremely important and that they can lead to systematically inconsistent estimates of the trend component. Furthermore, most of the filters commonly used to detrend time series in practice, are highly unstable and unreliable for end-of-sample estimation.