I am trying to make my data stationary.
I tried ADF on EVIEWS and got that my data is already stationary at level. However when looking at graph and correlogram it was clearly not the case.
I did the frequency converter both linear last and quadratic average.
Unemployment rate + poverty gap rate were I(0) in both cases
GDP per capita rate was I(0) with linear last and I(1) with quadratic average
When the ADF yielded I(0), it did not make sense at all when looking at the graph + correlogram.
I do not know how to proceed.
I attached my data
The first thing to do is to make sure that you are testing and plotting the same data set. Plot the data and visually check for a trend and intercept. Based on your visual inspection, move to check for the presence a unit root using the ADF test. If results are still different and the problem persists check for structural change in the data set.
Perform the ADF test with (i) intercept (ii) trend and intercept. See if it is stationary at level. Also check for structural break.
I have tried the ADF test in eviews on a seasonal time series without trend and the impression was that of I(0). This impression of stationarity is wrong because a seasonal time series is not stationary. Therefore exclusive use of the ADF test is not advisable. The correlogram may also be used to corroborate the ADF test decision.
I think it depends on your data
sometimes we have to give up the unit root test to the PP. test (nonparametric correction of DF)
At the practical level, the best to use the Urall (free) function which allows with Eviews 10 to see all the results (With Constant, With Constant & Trend, Without Constant & Trend) at the level and at the first difference and decide.
unit root tests are not as systematic as it seems. You have to use the right test (with intercept, without intercept, with trend, without trend...). Dickey-Fuller test have to be carried on three types of medels, you have to begin with global one (model C: with intercept, trend and of course stochastic trend), then B and A.
Please see page 56 in this file for the FD strategy test:
https://www.univ-orleans.fr/deg/masters/ESA/CH/CoursSeriesTemp_Chap2.pdf
Besides, the DF test assumes white noise residuals, if not you have to use augmented DF and add the adequate lags.
If you are interested in understanding the unit root tests you have to read some econometric references. For the easy and rapid way, try to plot your differentiated data (first difference) and see if the curve is around 0, so you don't have to add an intercept nor a trend. if it has a mean, but does not a deceasing or increasing shape, so you have to use the FD test with simply an intercept.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Graphs