GMM Estimation
Posted: Mon Apr 26, 2010 4:49 am
hi,
I have Panel data of about 130 countries and would like to estimate the growth rate of yearly GDP as a function of institutions and other macroeconomic variables (e.g. inflation). For this purpose, I wanted to employ the GMM_Diff of AB (91) in order to account for endogeneity problems.
As far as I understand, in this case differences instead of levels are used as explanatory variables and the lags of the variables are employed as instruments. The wizard for example uses AB(91) company data. As a dependant variable, I used d(log(gdp))in order to explain the growth rate. In step 2, I put the explanatory variables in levels (e.g. institutions or inflation). In step 5, I chose the No.Transformation because I want to treat all variables as being endogenous.
Each time I run a regression using all variables, the results are inferior. If I only use part of the variables, the level of significance improves but I always get a negative autocorrelation of 2nd order or more. Can you tell me what I'm doing wrong in order to improve my results?
Should I write log(gdp) instead of d(log(gdp))?
Also, using 5 year instead of yearly averages does not solve my problem. Neither does using only observations in beginning or the end of each period.
Thank you for your help!
I have Panel data of about 130 countries and would like to estimate the growth rate of yearly GDP as a function of institutions and other macroeconomic variables (e.g. inflation). For this purpose, I wanted to employ the GMM_Diff of AB (91) in order to account for endogeneity problems.
As far as I understand, in this case differences instead of levels are used as explanatory variables and the lags of the variables are employed as instruments. The wizard for example uses AB(91) company data. As a dependant variable, I used d(log(gdp))in order to explain the growth rate. In step 2, I put the explanatory variables in levels (e.g. institutions or inflation). In step 5, I chose the No.Transformation because I want to treat all variables as being endogenous.
Each time I run a regression using all variables, the results are inferior. If I only use part of the variables, the level of significance improves but I always get a negative autocorrelation of 2nd order or more. Can you tell me what I'm doing wrong in order to improve my results?
Should I write log(gdp) instead of d(log(gdp))?
Also, using 5 year instead of yearly averages does not solve my problem. Neither does using only observations in beginning or the end of each period.
Thank you for your help!