I run Augmented Dickey Fuller test on a time sereis variable to test its stationarity by using Eviews. the results disclose that the variable is not stationary.
As we know Eviews produces ADF test output in three different parts.
1. first part - ADF test resutls
2. second part - regressin
3. third part - model diagnosis.
my doubt is as follows:
t- statistic report in firt part(ie., ADF part) and t-statistic for one period lagged value of the variable(reported in second part i.e, regression part) are the same.
but the 'p' value are different. what is the reason for such difference?
i am attaching the output also.
thanks in advance.
In ADF test the null hypothesis is usually the statement of the existence of a unit root of the time series which amounts to a claim of nonstationarity. If the p-value is less than the specified alpha, then this null hypothesis is rejected and the series is adjudged as stationary. Otherwise the series is adjudged as non stationary. There can be no confusion.
Dear Srikanth,
You are trying to compare p-values for different tests, which is not possible. The one shown in the regression corresponds to a standard significance student test (t-test). The test above the regression results corresponds precisely to a Dickey Fuller test, which takes into account the fact that under the null (i.e. the series is non-stationnary), the statiscal properties of the t-statistic are quite different compared to a traditional student test. DF have shown what these differences are and how to account for them in their very influential JASA paper: https://doi.org/10.1080/01621459.1979.10482531
This is also why this test carries their names. In a nutshell, you should always trust the p-value above rather than the one that you obtain in the regression. If the series is non-stationary (i.e. the null cannot be rejected), the p-value in the regression part is simply wrong.
Best regards
thank you prof Ahmadi Murjani
your answer is very specific and clear
t- test is not too good for complex data, F-test is better in such a scenario you are to consider more robust test like chi-square test and compare notes for authentication.
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series