Panel unit root, testing for trend significativity

[lauren29]

Hello,
I am using Eviews 8. I am trying to test my panel data for stationnarity. However, I m having trouble to apply the same approach as for time series.
I explain : usually, for a time series Y, I choose to use a given test (let say PP) . I first test the specification with trend + intercept, if the T-stat associated to the trend is above the corresponding Dickey Fuller (1981) value (approximately 2.78) then I finish the stationanrity test here and conclude. If the trend is not significant, I re-apply the test on the specification with only the intercept. If the T-stat associated to the interecept is above the Dickey Fuller value (approximately 2.52) I conclude, otherwise I perform the last specification of the unit root test : with no contant and no trend.
When I tried to apply the same method to Panel Data I could not find information about the significativty of the trend nor the constant. The problem is that depending on the specification I run (trend + interecept, only intercept) the test conclusions are contradictory.
Basically, I want to obtain the following for a panel data workfile ( N=3139, T=14) in order to get the T-stat:

illustration.jpg (120.54 KiB) Viewed 7877 times

I thought I could just run OLS such as

Code: Select all

LS D(DAX) C DAX(-1) @TREND("1998m01")

However since my panel data is unbalanced I m not sure the trend specification would be adequate. What do you thnik?

Thanks in advance

[adnma]

Hi!

I have a question relating to panel unit roots:

Do I have to do panel unit root tests for my data????


My work deals with executive remuneration data with N>T (or) Is it that panel unit root tests are only applicable to macro economic variables such as GDP, Imports, Exports etc.,???


Very very thks in advance...

[Max]

Hello,
I am using Eviews 8. I am trying to test my panel data for stationnarity. However, I m having trouble to apply the same approach as for time series.
I explain : usually, for a time series Y, I choose to use a given test (let say PP) . I first test the specification with trend + intercept, if the T-stat associated to the trend is above the corresponding Dickey Fuller (1981) value (approximately 2.78) then I finish the stationanrity test here and conclude. If the trend is not significant, I re-apply the test on the specification with only the intercept. If the T-stat associated to the interecept is above the Dickey Fuller value (approximately 2.52) I conclude, otherwise I perform the last specification of the unit root test : with no contant and no trend.
When I tried to apply the same method to Panel Data I could not find information about the significativty of the trend nor the constant. The problem is that depending on the specification I run (trend + interecept, only intercept) the test conclusions are contradictory.
Basically, I want to obtain the following for a panel data workfile ( N=3139, T=14) in order to get the T-stat:

illustration.jpg

I thought I could just run OLS such as

Code: Select all

LS D(DAX) C DAX(-1) @TREND("1998m01")

However since my panel data is unbalanced I m not sure the trend specification would be adequate. What do you thnik?

Thanks in advance

Hi, I have the same question. Does anyone have any idea?

[popcorn]

Okay... I will try and respond to all the questions above and (hopefully it will help)

I think that the main thing is to ask (do you have to really identify the t-stat or computed tau in your workings)?

My advice is ... even if you are using unbalanced panel... to do a group panel test. Instead of selecting individual test, choose the summary and then look at intercept etc. If the variable is stationary, then the probability will be less than say 0.005. This means that it is stationary at 1% This is much simpler than looking up in separate tables and you get to use all the tests.

In terms of unit root tests, yes, sometimes they do give conflicting results. This may due to 0 values being included, the nature of the unbalanced panel or other reasons.

In terms of a panel where N>T, you do unit root tests ONLY if the variables are time variant. A variable such as female =1 would be time invariant.

Hope that this clarifies things