I do get serial correlation and cross-sectional dependence when I run the model using EVIEWS 8. I would like to know if there is a way to overcome this. I appreciate your comments on this.
Hie Nosheen. To deal with serial autocorrelation, hetroskedasticity and cross sectional dependence in panel data go for the Feasible Generalised Least Squares (FGLS) and the Panel Corrected Standard Error (PCSE). The former works well ifT>N, while the latter is feasible when N>T. In STATA use the xtgls syntax for FGLS and xtpcse for PCSE.
In your information, I see that you have to put the lagged dependent var into model as independent var to solve the serial correlation. In the case of panel data, if we do this action, the panel data is called as dynamic panel data => the problem of endogeneity emerges so that i suggest you in using 2SLS or 3SLS (in the case of no heteroskedasticity), or GMM (in the case of heteroskedasticity) or LSDV.
In addition, I suggest you using the Stata to run the panel data model
For dealing with serial correlation in panel data model, the most straighforward tool is to cluster the standard errors at the unit level. This is readily available in most of the statistical softwares (e.g., Stata). It is a conservative strategy, as your errors would be robust to all sort of serial correlation. Another possibility, in line with the suggestion of a previous reader, is to try to model explicitly the serial correlation process.
Regarding cross-sectional dependence, I would first wonder myself if it is an issue and where it comes from. If for example, it could be related to groups (e.g., members of families, doctors in hospitals, firms in districts, students at schools, etc.), it can be easily fixed. In principle, to deal with possible intra-cluster correlation you would need to cluster at the unit-year level (school-year, hospital-year, etc.). If one suspects on the existence of serial correlation in a variable at the -let's say- school level, you would need to cluster at that level. That would take account of both issues. You will be also addressing heteroskedasticity. The alternative here is to model the process through a spatial econometric model.
If students can change their school, you can use a two-level clustering (available in stata with the command xtivreg2).
In sum, if you have units and a variable that are likely to be correlated across some units embedded into a larger one and you suspect also on serial correlation, the state-of-art solution is to cluster the errors at the level of the large unit. This is a conservative procedure and is considered appropiate as long as you have a large enough number of clusters (e.g., schools).
All the best and I hope it helps.
PS: Please, rate my answer if you are satisfied with the response.
For cross sectional dependence use spatial approach or factor structural approach. For details go through the paper "Cross-sectional Dependence in Panel Data Analysis".
For serial correlation in panel data refer to
Testing for serial correlation in linear panel-data models
Hie Nosheen. To deal with serial autocorrelation, hetroskedasticity and cross sectional dependence in panel data go for the Feasible Generalised Least Squares (FGLS) and the Panel Corrected Standard Error (PCSE). The former works well ifT>N, while the latter is feasible when N>T. In STATA use the xtgls syntax for FGLS and xtpcse for PCSE.
I have tried your advise but I would like to know how we can re-test for cross sectional dependence and serial correlation. As xtcsd pesaran and xtserial require to use xtreg. I need to report the results in my paper.
Additionally, is it possible to get data after correcting for cross sectional dependence and serial correlation as I amusing the data in different software so want to correct these problems before transferring the data.
Sarah SA you can run xtgls and use Wooldridge Test (Serial correlation), Breusch-Pagan LM Test (Cross-sectional Independence), and Modified Wald Test ( Groupwise Heteroskedasticity).
Cross-section dependence is the problem, but the question that how to solve this problem?? Well, this is not such problem that could be solved; somewhat almost every cross-section data have cross-section dependence. The existence of cross-sectional dependence presents misleading results when we use the models that are inconsistent to deal with the such issues.