i am currently doing a research to study the granger cause and effect among female labor force participation rate (FLFPR), fertility rate (FR), infant mortality rate (IMR) and female tertiary education (TER). i have very little knowledge about Granger Causality test but based on my reading, if i understood correctly, i have to run a unit root test and cointegration test prior to the granger test. correct me if im wrong. for your information, im using Eviews9 software to perform this test.
as for the unit root test, i performed Augmented Dickey Fuller (ADF) test on each of my variable, and found that all of my variables are I(1).
1st question:
to proceed with the cointegration test, i am not sure if i should run the test using the raw (level) data, or the first-differenced (re-calculated from raw data) data.
2nd question:
my data consists of 36 observations, so i wonder if i should use Johansen cointegration test, or ARDL Bound test?
3rd question:
say, if im using the Johansen cointegration test, on the test specification part, which of the five options should i click? (intercept, trend, etc)
i think i have more questions to ask, these are all i could think of right now. i might ask more questions later. i hope someone can help me and answer my questions. thank you in advance.
Another option for the cointegration test is the engle-Granger method, that is when you had all of your variables with integration order I(1) the residuals of your equation in level (or with raw data) must be I(0), one order of integration below the variables. This test is very easy to implement in eviews. So far i think i covered the first two questions, as for the third if you use Johansen (with raw data) you must graph your variables to check if the have a visible trend, the constant just ensures that your variable doesn't take value 0, I reccomend always use at least constant. Also remember when you check unit roots in fisrt differences that the constant you use in raw data must be suppressed, as the derivative of a constant is 0.
1. Yes. Use the level (raw data) as an input for Johansen cointegration test. But test to ensure that your data are non stationary at level and stationary at first difference.
•
2. Your observations are too few, because cointegration being is a long run concept, require a reasonable sample size, so as to capture the long run dynamics. So increase your observations to (like 72 sample size or more) reasonable size.
•
3. Test for unit roots, and then decide what to choose in the option. If your data is trend stationary and unit roots, choose the 4th option (i.e.Linear trend, intercept and trend), or the 3rd option (i.e.Linear trend, intercept and no trend) If your data is not trend stationary and unit roots.
•
Good luck
- Badges
- Science topic
- Similar topics
- Cognitive Science
- Thinking