Panel data, 70 countries and multiple observations per year

[sam hermans]

Hi all,

Currently I am investigating cross border mergers and acquisitions and the effect of culture on this. And I determined all buy-and-hold returns (BHARs) and cumulative abnormal returns. Now I want to do regression with these BHARs as a dependent variable and culture as an independent variable and I want to take care of fixed country and fixed year effects. But I have multiple observations of for instance the United States in one year. So I tried to do a simple panel regression, with as panel ID the countries (for which I created a variable which gives a numerical value to the countries, and as time variable I want to use the years. But than I get an error if I want to take care of the fixed effects, specifically stata mentions the following: repeated time values within panel r(451).
Which is a logical error, but I do not now how to resolve this, without losing data.

Hopefully someone can help me out there. Thanks a lot in advance.

[Andrew Musau]

and as time variable I want to use the years

When you have annual data, a variable reflects some aggregation (e.g., sum or average) for months, weeks or days in the year. For example, annual GDP for country i in year t by definition represents the sum of the value of final goods and services for the months and weeks that make up the year. So if you want to use annual data, you must accept such a reduction to one observation per year. Any objections would arise if that observation is not representative of that country in that year. However, there is no requirement that your time variable is
a year. It could be quarters, months or days.

Currently I am investigating cross border mergers and acquisitions and the effect of culture on this.

Does culture vary much across time? You may have a problem with the fixed effects setup.

[sam hermans]

Dear Andrew,

Thank you for your reply. yes for me there is a requirement that the time variable is a year. And no culture does not vary much across time. So that is not the problem. The problem is that I have several observations for one country within one year... Hopefully my problem is a bit more clear, I hope you can help me out here.

[Andrew Musau]

Any specific advice would require knowledge of your sample and the underlying literature. Your description above can only elicit general comments like #2. My advice is that you talk to your advisor (if you are a student) or a senior colleague otherwise. You can also look at similar published papers and look at the common datasets used and what procedures and assumptions are employed. A number of empirical publications these days have a requirement that authors deposit their data and codes where it is accessible to others.

And no culture does not vary much across time. So that is not the problem

'A requirement of fixed effects is that there is temporal variation in your data. It may not be a problem now but it will be once you get to the estimation stage.

[Henry Arauz]

Hi Sam, I am currently going through a strikingly similar issue. I am researching the impact of corruption and country governance on cross-border M&A performance. I also have multiple observations for each year, which creates error 451 when running a fixed effects regression. Did you manage to get around this issue? - if so, would you mind sharing it with me? I will highly appreciate your help. Thanks in advance.