Correlation Matrix with pooled data ...

[TheRocK]

Hi again,

as mentioned in other topics, I have a workfile consisting of 8 pools, each including 285 cross-sections and 67 time periods. I want eViews 6 to give me the correlation coefficient between the pools. However, then I mark every group and ask eviews to compute the correlations, its starting to calculate the correlations between every single cross section and any single variable.

How can i tell eviews to give me the correlation between the POOLs, e.g. market_cap_? & leverage_?, where the ? represents the cross section identifyer?

All the best and thanks

Kai

[EViews Glenn]

Perhaps it's just me, but I'm not quite certain exactly what your setup is.

From my reading, you have 285 cross-sections and 67 time periods. That defines your pool structure.

The pool is just a collection of cross-section identifiers. Presumably, for a pool (say POOL01), you'll have 285 cross-sections in the pool. For each variable, you'll have 285 individual series, which in pool language, we denote MARKET_CAP_? and LEVERAGE_?.

Now here is where I'm not quite sure what's going on. I don't quite know what you mean by "correlation coefficient between the pools", nor what you mean when you say you have "8 pools". Can you be a bit more specific about what correlations you wish to compute, and what assumptions you are making about variability across cross-sections?

[TheRocK]

Hi Glenn,

thanks for your answer. Actually I already found the explanation. I had my variables unstacked; now when they are stacked I have no problem in calculating the correlations.

Nevertheless, here is another question ;-) Since OLS regressions are quite sensitive to outliners/extreme values, I want to "winsorize" my variables.

However although I read and know how to do it in cross-section settings, how can you do this in a panel setting? Do you take all the cross sections at one distinct point in time, form your confidence interval and then winsorize? OR do you take a distinct cross section and all of its time periods, form your confidence interval over the time periods and windsorize then ?

Thanks

Kai