Stability of Extracted Cycles : A Case of Adjustment Factors in Quarterly Data

P(論文)This paper studies stability of cyclical components extracted from economic time series. We examine the cyclical components extracted with the Butterworth filter in Gomez (2001) and Pollock (2000), the Hammingwindowed filter in Iacobucci and Noullez (2005), and the ChristianoFitzgerald (CF) filter in Christiano and Fitzgerald (2003). We use two types of stability diagnostics discussed in Findley et al. (1998): sliding spans and revision histories. The main findings are as follows. First, the tangentbased Butterworth filtering produces more stable adjustment components than other three filtering methods. Second, the estimates are very stable in the middle range of the series. Finally, as a rule of thumb, we should use series longer than 120 sample points for filtering in practice so that data revision would least affect subsequent analyses.departmental bulletin pape