The Discussion Papers series presents results from ongoing research projects and other research and analysis by SSB staff, intended for international journals or books. The views and conclusions in this document are those of the author(s). 

While existing panel estimators address the simultaneity problem, the state-ofthe- art LIML estimator remains inconsistent and inefficient, and lacks a valid framework for inference that accounts for parameter constraints. We develop a constrained GMM (C-GMM) estimator that is consistent and derive a unified expression for its asymptotic variance that is valid also at the boundary of the parameter space. Furthermore, we apply the size-adjusted conditional likelihood ratio test statistic proposed by Ketz (2018) to conduct boundary-robust inference and demonstrate improved coverage of confidence intervals at or near the boundary compared with inference methods based on the normal distribution. A Monte Carlo study confirms the consistency of the C-GMM estimator and shows that it substantially reduces bias and root mean squared error relative to the LIML estimator. The C-GMM estimator, combined with boundary-robust inference, maintains high coverage of confidence intervals across a wide range of sample sizes and parameter values.