Apart from what any reference book says (or you 've heard it says), you always have to make sure your variables are stationary (or cointegrated) before you analyze your findings. In other words, you have to do unit root tests, anyway.

As for the 15 years rule. first time I 've heard it. I really doubt it.

unit-root tests are a must in panel data. However, with only 7 periods, I doubt the reliability of unit-root tests which are based on the interdependence of observations through the time. These tests are based on the assumption that N and T tend to infinity. they are implemented on time series in the last step have low power in small sample sizes T. For this reason, Bai and Ng propose the use of a mean statistic à la Im, Pesaran and Shin (2003), or a statistic à la Maddala et Wu (1999). Still, I'm not confident that 7 periods in enough.

Here an interesting website of Prof Hurlin where you can run unit root tests

http://www.execandshare.org/CompanionSite/site.do?siteId=109

Prof Hurlin has a distinguished paper on unit-root tests for panel data, you can send him an email (he is very kind, he was my professor!).

Here the link of his paper:

Article Second Generation Panel Unit Root Tests

Panel unit root test needs to be done.

Unit root seems to be a prerequisite for panel data analysis most especially when dealing with time series panel data .. I mean when T>N. More so, u might not be needing it when dealing with discrete or ordinal depenedent variable analysis..

Since panel data is a mix of cross-section and time series, the problem of none stationarity could be resolved. Testing the data for stationarity would be recommended to be sure that your analysis of the data is validated

Hello, good day. In Iranian universities, I have heard that there is no need to perform a single root test for data under 15 years old based on Mr. Baltaji and data under 12 years old according to Mr. Andres. But to better understand this concept, it is better to use a local teacher related to your work. But it is important that the variables are consistent in each test. You can use the Jarkiu-Bra or Kolmogorov-Smirinov test to detect panel panel data. good luck.

Agree with Chiraz Karamti

Dear Sajjad,

Unit root test is to test Fr the stationary of your variables before analyzing. It's important to be on the safe side to avoid spurious regression when dealing with time series panel data.

Regards

Dear Sajjad,

Indeed, Baltagi (2005, 3rd edn) does not say explicitly T

Please be careful. Say yet Mohammed advices to use Jacque-Bera test which is a normality test not a unit-root test. Normality and stationarity are different concepts.

There is no need for unit root test for your variables because you are dealing with panel data. Instead, do panel unit root test. This is appropriate for panel data.

Testing for panel unit roots / stationarity

Panel unit root test (PURT) emerged from time series unit root testing. The major difference to time series testing of unit roots is that asymptotic behaviour of the time-series dimension T and the cross-sectional dimension N need to be considered. The way in which N and T converge to infinity is critical if one wants to determine the asymptotic behaviour of estimators and tests used for non-stationary panels.

Levin-Lin-Chu (2002) test was used to test the stationary for the balanced panel data. LLC tests work with the restriction that all panels share a common autoregressive parameter. The LLC test assumes that error term is independently distributed across panels and follows a stationary invertible autoregressive moving-average process for each panel, i.e., this has the null hypothesis that all the panels contain a unit root, if the significant result gives the panels are stationary. LLC recommend using their test with panels of “moderate” size, which they describe as having between 10 and 250 panels and 25 to 250 observations per panel. Hence LLC panel unit root test is used for the common balanced panels.

Reference: R. Gangai Selvi and K. Mani, 2015, Land Use Dynamics in Tamil Nadu - A Spatial Econometric Analysis, PhD Thesis submitted at Tamil Nadu Agricultural University, Coimbatore, Tamil Nadu, India

If t=30 years and N = 15 it would be better to use unit root test in panel data . In the case of panel data if values are found non- stationary , it is necessary to use unit root test for the expectation of good results.

If N

Yadav Sharma Gaudel Perhaps you mean T

A random walk with drift and trend as well as grafical analysi s in panel data is the initial process for checking whether the data are stationary or nonstationary.The question arises why some economic and financial time series data are stationay and some are nonstationary ?

In practice, most economic and financial time series data are found nonstationary. Unit root test is a conditional statistical test in econometric analysis, however, the question of not necessary may be unconditional and the unbounded solution needs further investigation.

Prof. Wooldridge told me, he dicussed this issue in his 2010 mit press. when N>T, I don't need to do unit root test. He suggested to use fixed effects, put in time effects and cluster for serial correlation.

I tend to side with Prof. Woolddridge. Unit root test is not usually conducted in Panel data.It is done in time series analysis. However, the LLC and several other tests are usually done to see how balanced your data is. Professor Wooldridge is an authority in Econometrics, and has a book on it.You can check for more insights !!

Sajjad Sargolzaei any theoretical reference why unit root test are not required for short panels?

Anuradha Saikia "wooldridge econometric analysis of cross section and panel data mit press"

Chapter7, Note on Time Series Persistence

Thank you Sajjad Sargolzaei

Honestly, am as crossed as you are. Unfortunately, had submitted my work without testing for unit root in my Panel study. Am heavily relying on basic literature arguments that says when your T is

Can somebody share some papers/literature that state that stationarity is not to be tested when T is less than a certain amount (say 30 or 15, as being discussed above)? It would be extremely helpful for my thesis.

A reference for the minimum values for T is to be found in a table of critical values for Dickey Fuller Tests for testing F ratio, AR model , AR model with constant, AR model with constant and time trend on p.1063 of " Dickey. D and W. Fuller: Likelihood Ratio Tests for autoregressive time series with a unit root " Econometrica, 49, 1981, pp1057-1072 ) which gives the critical values of sample sizes to test for unit root in AR models.

You will also find a good discussion quoting the table from Dickey Fuller in

"Econometric Analysis" 3rd edition (International edition, 1997 Prentice-Hall International Inc. ) of William H Greene on p 849.

Sample sizes at which tests were reported in the Econometrica paper are 25, 30, 100 and unrestricted. William H Greene also refers to Fuller W " Introduction to Statistical Time Series" , New York John Wiley & sons 1976, p373.

No one seem to answer this question in it's simple term. Am on a crossroad right now & no one seems to give accurate and acceptable answer.

Do you need a Unit root test ( either Panel Unit root or otherwise) in estimating a panel data? Y/N. If Y, why & if no Y?

2. I want to find answer to this panel problem:

I have 5 hyp to test. Each five is grouped in:

H1 - containing 17 countries

H2- containing 13 countries

H3- containing -15 countries

H4 - containing 10 countries

H5 - 6 countries

All same variables.

What is the advice of how to go about it.

Explanation of the answers given above: if T is