serial correlation in panel data
Posted: Thu Dec 08, 2016 4:47 am
Hi,
I'm analysing panel data using 49 different countries and a time span between 2001-2006. I have three dependent variables: debt market capitalization/GNP equity market capitalization/GNP and number of listed firms per millions inhabitants. I also have various independent and control variables: civil law dummies, dummy for if the one share one vote principle is present in a given country, LogGNP, LogGNP per capita, Lag1GDP growth, rule of law and number of antidirector rights present in a country (variable that doesn't change over time). Hypothesis: civil law countries have higher: debt and equity market cap and more listed firms per million inhabitants. When I run my regressions my results are fine and quite significant, but it has a durbin watson value of around 0, indicating strong serial correlation. When I try to remove the serial correlation using first differences, my coefficients and results change a lot and all but one or two become very insignificant. Should I always remove serial correlation and are there perhaps other ways to do it that fit my data better? I've added one of my datasets to this message.
Kind regards,
Alexander
I'm analysing panel data using 49 different countries and a time span between 2001-2006. I have three dependent variables: debt market capitalization/GNP equity market capitalization/GNP and number of listed firms per millions inhabitants. I also have various independent and control variables: civil law dummies, dummy for if the one share one vote principle is present in a given country, LogGNP, LogGNP per capita, Lag1GDP growth, rule of law and number of antidirector rights present in a country (variable that doesn't change over time). Hypothesis: civil law countries have higher: debt and equity market cap and more listed firms per million inhabitants. When I run my regressions my results are fine and quite significant, but it has a durbin watson value of around 0, indicating strong serial correlation. When I try to remove the serial correlation using first differences, my coefficients and results change a lot and all but one or two become very insignificant. Should I always remove serial correlation and are there perhaps other ways to do it that fit my data better? I've added one of my datasets to this message.
Kind regards,
Alexander