Re: [Gretl-users] GMM Instruments Lagged Independent Variables

On Fri, 15 Apr 2011, Dempsey, Keely wrote: Yes, but you have to decide how you want to calculate R-squared, for a model not estimated via least squares. One approach that people use is to compute the square of the correlation between actual and predicted y (dependent variable). But GMM is general enough that gretl doesn't necessarily know what the "dependent variable" is; you'll have to tell it. Here's a simple example script: open data4-1 # run OLS and record R-squared ols price const sqft R2_ols = $rsq # set-up for GMM matrix X = { const, sqft} matrix y = { price } matrix B = { 50, 0.1 }' matrix I2 = I(2) matrix e = y - X*B # emulate OLS via GMM gmm e = y - X*B orthog e ; X params B weights I2 end gmm # calculate fitted values from GMM series yhat = price - e R2_gmm = corr(price, yhat)^2 printf "R2_ols = %.4f, R2_gmm = %.4f\n", R2_ols, R2_gmm Allin Cottrell