yes unit root test is to be performed.

Of course, Yes. You need perform stationarity test

Yes, you need to perform the stationarity test or the unit root test. Your underlying assumption that '  earnings revert back to normal' will also be valid only if the data series is stationary.

A stationary process is a process whose statistical properties do not change with change in time at which the series is observed. Consequently, statistical parameters like mean, variance, etc. stay steady over time and do not follow any trends.

The test of stationarity is related to the test of Unit Root. A stationary time series is said to be integrated of order zero i.e. I(0). If the time series needs to be differenced once to make it stationary, it is an I(1) series and is said to have a unit root. So, testing for stationarity is equivalent to testing for unit roots in the time series. A Unit Root Test tests whether a unit root is present in an autoregressive statistical model of the time series. There are a number of tests to ascertain the stationarity of the datasets or to test for unit roots. The most common unit root test is the Dickey-Fuller Test and the most commonly used stationarity test is the KPSS test. Hope this clarifies your doubt.

Dear,

If it a "Well designed" panel data set with sufficient number (?) of firms along with modest time range in years (?) in your case, then a stationary test is not mandatory given that your sample firms data are well distributed across the time parameter viz., for each nth year there is no bias of clustering the sample data around it and in another kth year there is little presence of the data. Therefore, in essence you need to see whether the data set represent homogeneous distribution or homogeneous across time frame. A well designed panel data is always a normally distributed sample so long the population data remains normally distributed. That is, the nature of the parent distribution is assumed here.

Now, say you are considering a longitudinal analysis, then time plays the major role not the firms as such, then you need to qualify all the criteria for time series analysis.

Now, come to your research quarry, to explore whether earnings (say EPS or DPS) revert back to normal. Hope, "normal" is refers to the average or a normal range of values. If a dynamic model of capital structure is incorporated into your decision choices, then payment of dividend becomes dynamic in nature (and please do not misguide that such dynamism means non-stationary) and it must be revert backing at positive NPV projects and forward moving at negative or Zero or indecisive NPV projects. Hence, if your idea is to explore the dynamism, then try to fit a dynamic partial model .

Best wishes.

Thank you all for the feedback. I am using eviews so the panel unit root tests are in built.

Thanks