At the first time, I processed my data which consisted of three variable using Augmented Dickey–Fuller (ADF) test and each variable stasionary at different level. Then I tried to use Phillips-Perron (PP) test, then the result inidicates that all variable stationary at level or I(0).
Why is this happen, and which result or test should I use for my research (ADF or PP)?
These unit root tests have low statistical power in small samples. That might be a possible reason if we assume there are no extreme observations and/or structural breaks in the data series. If the sample size is big enough, these tests results typically do not contradict each other. The PP test is similar to the ADF test but allow for the residuals to be serially correlated. However, in the ADF equation, we need to make sure the residuals are not serially correlated even though EViews suggests the optimal model by augmenting the combinations of lagged dependent variable.
When dealing with time-series analysis, it is good to plot the data and visually inspect it. The ADF test could falsely report I~(0) or I~(1) due to a structural break! Again visual inspection could reveal such a problem.
absolutely agree with Manhal and Akram. Graphics permit detection of structural breaks (SB) and so, DF is no more reliable. There are unit-root tests in STATA that allow for SB.
I would add four more points to those already set out
https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=&cad=rja&uact=8&ved=2ahUKEwjZqoCKzLHtAhVc5eAKHUxBBtUQFjAKegQIAxAC&url=http%3A%2F%2Fwww.eviews.com%2Fhelp%2Fcontent%2Fadvtimeser-Unit_Root_Tests_with_a_Breakpoint.html&usg=AOvVaw25OE0t1rhPYANNtzgfDDYc
Dear Firsty,
I feel you are using a small sample. Why not try another unit root test and compare the result with the other two.
Or carry out a visual test like CUSUM or CUSUMsq to check for structural break.
Regards
- Badges
- Science topic
- Similar topics
- Correction