Hybrid generalized empirical likelihood estimators: Instrument selection with adaptive lasso

In this paper, we use the adaptive lasso estimator to choose the relevant instruments and eliminate the irrelevant instruments. The limit theory of Zou (2006) is extended from univariate iid case to heteroskedastic and non Gaussian data. Then we use the selected instruments in generalized empirical likelihood estimators (GEL). In this sense, these are called hybrid GEL. It is also shown that the lasso estimators are not model selection consistent whereas the adaptive lasso can select the correct model with fixed number of instruments. In simulations we show that hybrid GEL estimators have smaller bias and mean squared error than the other estimators in certain cases.