EViews.com

Hi all,

My apologises for the question, though am struggling with data pooling in eviews and estimation respectivelly.
My data consists of 24 countries, period between 1995 2007, 3 indexes and in every country I have 2-10 companies, which is in Excel file and looks like this:
Country--Year--Company ID--Index1--Index2--Index3
AU--1995--Company A--
AU--1996--Company A--
...
AU--2007--Company B--
AU--1995--Company B--
AU--1996--Comapny B--
...
AU--2007--Company B--
UK--1995--Company C--
UK--1996--Company C--
...
UK--2007--Comapny C--
UK--1995--Company D--
UK--1996--Company D--
...
UK--2007--Company D--

The problem is, as long as I add companies into the countries, the Eviews does not want to pool the data - i.e. if I had only observations between 1995 2007 for every country it is easy to pool data. But, when I expand every country having different number of companies in every country I start to struggle with data pooling...

Does anyone know how to do this?

Thank you very much!

If I understand your description of your data...

The issue is that EViews only handles pooled/panel data where there are no "repeats" within a cross-section unit. In your context, we need the "year" indicator to be unique within a cross-section. When you add a "company", you are creating data with multiple observations for a given year within a country.

From EViews point of view, what you really have is a pool where "company" is the cross-section unit and year is the "time" unit. In your case, it so happens that there is a higher level category of country within which you can group companies, but from the point of view of pool structure that doesn't matter. If you wish to use this for analysis, you'll have to create a set of dummy variables for each company.

What you probably can do is to use year and company as your indicator, then create the dummies for country which will allow you to do analysis at that level of aggregation if you wish...

Although I tried that with other software, but might work with EViews. In what follows I assume that companies are unique across countries and company B in one country is not the same B company in other countries.
I create a numerical code for the country and multiply it by 1000000
I create a numerical code for the company (assuming that there is less than 100 of them), then multiply it by 10000
I use the usual code for the year

then I add the three codes together by normal addition, the resulting number is
ccxxyyyy
cc for country,
xx for company, and
yyyy for year.

I use the new variable as time series indicator (I'm not sure if EViews will accept it, but some other software do)

Then, in regression, I use the @expand function to make indicators for the variable I want to test (company for example).

If your companies are really unique across countries, and assuming you have a unique numeric or alphanumeric indicator for each company, don't you only need to use company and year as your identifiers?

The assumption of uniqueness of company is not necessary, even invalid (as Glenn pointed out), however, the exposition is still valid.
The use of ccxxyyyy is for time consistency if dynamics (lags) are used.
Importing the data into EViews and using time series structure and the Integer Date format should do the work; that allows for testing all or some of the Country, Company, Year effects by the @expand command and dropping one category (using the @drop… option in @expand) to deal with dummy variable trap.

However,

What you probably can do is to use year and company as your indicator, then create the dummies for country which will allow you to do analysis at that level of aggregation if you wish...

doesn't fix the problem because year will be repeated within company.