Re: [Gretl-users] problem with system instrumental variables

Hi Sven Well, so for the system ! It work quite well and is a neat feature. As you may micro data analysis often done with system estimators Yep ! Well, up to now I was working on the AIDS demand system on microdata (I sent a few days ago a sample of my data and the accompanying script). In this context all was easy because in the standard formulation of AIDS the same variables appear in each equation. To sum up, because of theoritical constraints, the system covariance matrix is singular. Estimation is done by dropping one equation and homogeneity in prices is imposed costlessly by the use of price relatives of deleted good. My concern now is that I have a nasty problem with my data : some are censored with nil consumption. Thats means serious biais proportional to the degree of censoring. Unfortunately, one of my goods suffers from substantial censoring (>40 %). To overcome this I use first a probit model and compute the fitted Mills ratios [ f(x) /F(x) = dnorm(yhat) / cnorm(yhat) ]that are used as extra instruments in the AIDS system. Finally, my problem is that to do so, I have a system where each equation have all the same variables in common, except one => the Mills ratio. GRETL does not estimate the system when i specify using the tsls syntax which can in principle allow equation specific instruments lists. here is where i am. cheers & thanx Franck