Dear all,

we need to estimate a system of three linear equations with an endogenous variable on panel data and we want to allow for individual random effects. The independent variables are the same in all equations and we have no lagged dependent variables among regressors.

At the moment we are treating it as a multilevel model (level 1: time-period observations; level 2: individuals) and using the command gsem. We specify 4 equations (3 equations for our system and 1 equation for the endogenous variable), and add a level 2 latent variable for individual random intercepts in each equation, i.e. (y1 <- x1 x2 L1[ind]); (y2 <- x1 x2 L2[ind]); (y3 <- x1 x2 L3[ind]); (x1 <- z1 z2 x2 L4[ind]) where z1 and z2 are two instrumental variables.

Do you think that this way of considering panel data estimation is correct?
Do you have any suggestions for alternative panel data estimators that would allow us to estimate this model?

Thanks in advance for your attention

Gabriella