Could one assume the existence of a Fractional Integral component when Dickey–Fuller or Phillips–Perron reject H0 ( p < 0.05 - i.e. The absence of a Unit Root) but also the KPSS Trend Stationarity test fails on H0 ( p < 0.05 - i.e : The series is non stationary ) ?
I would not assume fractional integration analyzed time-series based on such outcomes of the tests. The may result from number of reasons like for example inappropriate specification of the test equation (dynamic structure, lack of trend etc.). Instead I would recommend simple fractional integration tests like Whittle estimator or Geweke- Porter-Hudak test - the tests are implemented in most popular econometric packages.
I would not assume fractional integration analyzed time-series based on such outcomes of the tests. The may result from number of reasons like for example inappropriate specification of the test equation (dynamic structure, lack of trend etc.). Instead I would recommend simple fractional integration tests like Whittle estimator or Geweke- Porter-Hudak test - the tests are implemented in most popular econometric packages.
Mr. Cizkowcz
Firstly, Thank you very much for the advice.
I'm now refferring myself to the R fracDiff package (https://cran.r-project.org/web/packages/fracdiff/fracdiff.pdf) which implements both GPH and Whittle estimators and also to fArma package (https://cran.r-project.org/web/packages/fArma/fArma.pdf) which also implements a couple more estimators.
Best Regards
Hi Jorcy,
Following your KPSS test results, the nonstationary of the series is maybe due to the presence of structural breaks in his deterministic trend. See for more information as example the contributions of Bai-Perron(1998, 2003), Lee-Strazicich(2003, 2004).
Best regards.
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series