Hey ngawang-
Attached are two papers that should help you out. The first is on R-filters, a generalization of the HP filter. While HP uses only 2nd backward differences, (thus implying the series is I(2)), r-filters are a more general case for any (generally higher) order backward difference filters. The higher order R-filters have a much sharper frequency cutoff, but with the penalty that more future values are needed beyond the end of the series. There is, however, a tradeoff there which one can exploit.
The second paper is about TCS filters; essentially, the authors investigate the assumptions of the stochastic model underlying the HP filter. They present some ways to deal with endpoint problems, etc...
These articles are both innovative and well-written, hope they help.
-Gavner
Attached are two papers that should help you out. The first is on R-filters, a generalization of the HP filter. While HP uses only 2nd backward differences, (thus implying the series is I(2)), r-filters are a more general case for any (generally higher) order backward difference filters. The higher order R-filters have a much sharper frequency cutoff, but with the penalty that more future values are needed beyond the end of the series. There is, however, a tradeoff there which one can exploit.
The second paper is about TCS filters; essentially, the authors investigate the assumptions of the stochastic model underlying the HP filter. They present some ways to deal with endpoint problems, etc...
These articles are both innovative and well-written, hope they help.
-Gavner
Attached File(s)
r-filters HP Generalization.pdf
589 KB
|
1,220 downloads
TCS filter.pdf
1.4 MB
|
1,304 downloads
1