Panel data:independent variables are collinear with the panel variable

[Sandra Rhouma]

Dear readers,

I have a problem with estimating a static panel model. I have collected data on GFCF and 11 other variables from 1960 to 2013 on the World Bank website and data on federal reserve rate (fedrate) treasuries and MBS held on the Federal Reserve website, which I incorporated manually on excel. So I have 59 countries and 3186 observations, country is my panel variable and time my time variable . When trying to estimate the following model using the options for fixed effect or random effect (so to determine afterwards wich one to use):

xtreg gfcf fedrate, fe

I have the following message: " independent variables are collinear with the panel variable country"

However, when estimating the following:

xtreg gfcf gdp, fe

I don't have any problems.

I think the problem comes from the variables collected from the Fed and added manually to the dataset (as it occurs only when these are used as independent variables). Those variables are only available for the US, so there are dots for all other countries. Please find attached my dataset.

I tried to find a solution on the forum and else where, but didn't find anything that can help me fix the problem. Any help/advice would be really appreciated.

Thank you in advance for your answer.

[Clyde Schechter]

There is no solution to this problem. As fedrate is only available for one country, there is no magical way to use it as an independent variable in a multi-country model: there is no information about it for any other country than the US. Perhaps you want to run some separate regressions (not panel) for just the US that include fedrate as a predictor. But there is simply no way to incorporate it into an analysis using other countries.

[Sandra Rhouma]

Dear Clyde,

Thank you for your answer, I was suspecting this problem, now I am sure.

My purpose is to estimate the impact of the Fed policies on the GFCF of other countries. Would you recommend running a simple OLS on a selected number of countries/group of countries, keeping only the time series dimension of the data?

Thank you again.

[Maarten Buis]

If that is your aim, then substantively the fedrate is no longer a characteristic of the USA, but a characteristic of a point in time. In that case, you can and should use all countries: The fedrate of 1990 should in that case be attached to all countries in 1990.

[Sandra Rhouma]

Dear Maarten,

Thank you for your answer. I see what you mean. I just need to have the fedrate data in a separate excel file, which I will join by date to the panel data set that includes all other 12 variables (GFCF and so on).

Thank you both for your help