[Gretl-users] Re: Running a 2SLS model with an interaction term

Am 17.04.2023 um 11:05 schrieb elfsog(a)yahoo.gr: First of all, from a gretl-specific perspective, I'm not aware of any tool in gretl that focuses on this nonlinear-in-variables IV setup. However, on a general econometric level, this is still a model which is linear in the parameters. I'd expect the product X1*X2 to be endogenous (= correlated with the true disturbances) as well, although I believe that this is not always clear and in general I expect that it depends on the joint distribution. But if Z1 is known to be a valid instrument for X1, my pragmatic approach would be to simply compute Z1*X2 and use that as an additional instrument in a traditional linear IV setup. (I guess it would also be possible to try other nonlinear transformations of Z1, see the pretty large literature on non-linear instruments.) If non-normal distributions are involved, another quite recent approach might be one based on (nonparametric) control functions, see this one: https://arxiv.org/abs/2207.09246 (Breitung, Mayer & Wied). cheers sven