Dummy variables in panel data and serial correlation
Posted: Fri Nov 25, 2016 3:40 am
Hi,
I'm using panel data with a sample of 49 countries and a sample period between 2001 and 2006. I use listed domestic firms/population as dependent variable and dummy variables for the civil law legal origin and whether the one share one vote principle is present in a country, I also use gdp growth, log gnp, log gnp per capita, the rule of law index and also the anti director rights index as independent variables. The latter variable does not change over time and has the same value for every year between 2001-2006 for a given country.
When I run my regression it has serial correlation (Durbin-Watson Stat: 0,1). When I try to remove the serial correlation by using the first difference of all variables, I get the near singular matrix error. I wonder if there are other methods to remove the serial correlation or if the serial correlation has something to do with my data and the manner that I put my variables in the excel workfile. For example did I put the dummy variables in the excel file in the right way or do I need to create the dummies in Eviews? The same for the non time-varying antidirector rights variable. Furthermore, is it correct to take the first difference of my two dummy variables and non time-varying variable in a similar manner as the other variables d(civil_law) d(one_share_one_vote) d(antidirector_rights) or is there an other way? I've attachted my workfile to this message.
Kind regards,
Alexander
I'm using panel data with a sample of 49 countries and a sample period between 2001 and 2006. I use listed domestic firms/population as dependent variable and dummy variables for the civil law legal origin and whether the one share one vote principle is present in a country, I also use gdp growth, log gnp, log gnp per capita, the rule of law index and also the anti director rights index as independent variables. The latter variable does not change over time and has the same value for every year between 2001-2006 for a given country.
When I run my regression it has serial correlation (Durbin-Watson Stat: 0,1). When I try to remove the serial correlation by using the first difference of all variables, I get the near singular matrix error. I wonder if there are other methods to remove the serial correlation or if the serial correlation has something to do with my data and the manner that I put my variables in the excel workfile. For example did I put the dummy variables in the excel file in the right way or do I need to create the dummies in Eviews? The same for the non time-varying antidirector rights variable. Furthermore, is it correct to take the first difference of my two dummy variables and non time-varying variable in a similar manner as the other variables d(civil_law) d(one_share_one_vote) d(antidirector_rights) or is there an other way? I've attachted my workfile to this message.
Kind regards,
Alexander