This issue has received little attention in the literature in the past to the surprise of many. In fact, most studies assume that the initial value is either zero or bounded in analyzing the role of initialization when testing for a unit root in panel data. In response to this I prefer to consider a model in which the initialization is in the past, which is shown to have several distinctive features that makes it attractive, even in comparison to the common time series practice of making the initial value a draw from its unconditional distribution under the stationary alternative. The results have implications not only for theory, but also for applied work. In particular, and in contrast to the time series case, in panels the effect of the initialization need not be negative but can actually lead to improved test performance.

Panel data refers to collection of data from the sample units at different time periods. As time element is included in panel data, testing stationarity is required. If, unit root is present in the panel data, the regression results will be spurious. Because, presence of unit root makes the mean value of the series less representative.

Hi Ashu,

Panel data has cross-section as well as time-series elements. Since it has time-series element, Unit root test is required for testing stationarity in panel data as results will be spurious if data doesn't satisfy the stationarity assumption implicit in most tests. One can use Levin-Lin-Chu test for the purpose which is considered to be the ADF equivalent for panel data. However, if the panel data has very small T, i.e. data for a small number of time periods, stationarity testing is not an essential pre-requisite.

Hope it helps clarify.

Hemlata

Yes, theoretically it is true but some studies with sufficiently longer T, do not report it. That is where I was wondering. Is there something special about dynamic panel regression and unit root test?

Some authors like Maddala have criticized the panel unit root tests of not being similar to time series unit root tests & of low power. Some others opine that unit root test is not necessary if you do a panel using GMM. The stationarity of variables also tend to be ignored in Arellano-bond & Arellano-Bovver Dynamic Panel approach although it is not too much of a problem as these models are generally restricted to short time series.

Econometrics is an evolving discipline & there are not rigid rules laid down in many cases. Although there are issues in some cases in use of unit root tests in panels like different unit root tests giving different results, I would still recommend testing for stationarity of panel data. The test used may be LLC, IPS or a Fischer PP test as one may think appropriate.

If T is greater than N, we generally go for dynamic panel data analysis (non-stationary panel data analysis) and test for the existence of unit root. However, if N is greater than T, it does not require stationarity testing.

Additionally, I also agree to the detailed explanations of Miss Hemlata Chelawat.

Best wishes,

Muhammad Atif Nawaz . "However, if N is greater than T, it does not require stationarity testing" Reference?

@Sumya Abdi

If T is sufficiently large (generally 30 or above), we go for testing the presence of unit root even if N is greater than T. This is a rule of thumb that if N > T, Wo go for traditional panel data analysis. On the other hand if T > N, we go for non-stationary panel data analysis.

However, I will send you reference for the above statement too.

Best wishes,

Atif

Muhammad Atif Nawaz can you please send me the reference regarding your statement? This problem has been buzzing me for a while. Thank you!

Muhammad Atif Nawaz I would be really grateful if you could send me that reference about necessity of unit root test in small T data! Thank you.

Please @ Muhammad Atif Nawaz

I will really appreciate the said reference, I have a time dimension of just 5 Years. I am Using System GMM

Going for unit root when t is small may be misleading in the sense that no long run relationship is expected to exist.

Muhammad Atif Nawaz Please I need the reference for the statement you made above on when to run stationarity test based on the size of T and N

Muhammad Atif Nawaz Please send the reference for the statement

Please check Econometric Analysis of Panel Data by Badi H. Baltagi for reference. Sandeep Singh Chika Agnes Ifejirika Seb Ako Xinyu Zhang Khanh Hoang

Regards

Thanks @ Muhammad Atif Nawaz

Dynamic panel estimation with system gmm(n:50 x:15) do not require panel unit root test?

Muhammad Atif Nawaz I have a similar doubt regarding use of unit root test in panel data. Could you send me the reference for your statement.

Please check Econometric Analysis of Panel Data by Badi H. Baltagi for reference. Anuradha Saikia

Thank you Sir Muhammad Atif Nawaz

Thank you for sharing your tips and the reference book. I have a look at the 2008 version of the book (4e). I see a similar statement about doing unit root tests for small T panels can be misleading, but i cannot find the rule of thumb that the min T=30 for doing unit root. Could you suggest which chapter, please? Thank you so much. Muhammad Atif Nawaz

I have a similar doubt regarding use of unit root test in panel data. Could you send me the reference for your statement

Muhammad Atif Nawaz Can you share the exact edition and page number of the statement regarding rule of thumb to apply panel unit root test.

This topic is covered in chapter 12 of 'Econometric Analysis of Panel Data' for Baltagi (2005).

Many well-established resources on econometric analysis for panel data do not pay much attention to the stationarity issues since that we traditionally used to deal with micro panels (i.e., large N and small T). However, with the rise of macro panels in economic applications, this issue became increasingly relevant. Macro panels usually refer to large T and large N.

There seem no definitive answer as to how many time periods are needed to start worrying about nonstationary in panel data. This field is still largely underdeveloped compared to its counterpart in time series analysis, in fact, many applications of nonstationary tests in panel data context are merely adaptation from time series applications. Moreover, the length of time periods is not the only factor to consider in deciding whether you have to examine panels stationarity.

I highly recommend chapter 12 in Baltagi (2005) for more on this issue where you might find what is relevant to your specific case.