My question may sound more econometrical than about economics at the begining, but it is the contrary.
I am working with a panel data of about 150 countries and it is cointegrated. My variables are GDP, capital accumulation and another variable which nexus with GDP is expected to be country specific. Therefore, I am using estimators which average country specific coefficients (I mean, not pooled estimators). When I observe the panel specific coefficients for capital accumulation, I got some coefficients which were significant and negative and others which surpass 1.
In theory, the coefficient for capital accumulation is expected to be between 0 and 1, but I have seen some published articles in prestigious journals where the coefficient for capital was a little larger than 1. On the other hand, I have never seen an article with a negative and significant coefficient for capital.
I have read that when you obtain outliers or no sense coefficients in some panels, it is good idea to remove those cases (those countries in my case). I am certain about deleting the countries which regression yielded negative and significant coefficient for capital. However, I am not sure about which cut off I should use for the coefficients which are larger than 1. I am considering to delete the countries where I obtained a significant coefficient larger than 1.1. Is this a reasonable procedure? Or should I remover the nations where the coefficient is just larger than 1? Or should consider another cut off? Or should I keep the panel as it is (not removing any country)?
Very much thank you in advance,