Global identification of dynamic panel models with interactive effects

Abstract

We investigate the problem of global identification in dynamic panel models with interactive effects, under the large-N, fixed-T setting. While local identification, typically established via the Jacobian matrix, is well understood, global identification has remained a more elusive and challenging issue. It is commonly believed to be unachievable in this context. However, we demonstrate that the model is, in fact, globally identified. Our analysis also covers models with additive fixed effects, including cases with unit roots, where previous studies have reported non-identification based on certain estimators. We show that, even in these settings, global identification holds.