Global identification of dynamic panel models with interactive effects
arXiv:2504.14354
The paper proves that dynamic panel models with interactive effects are globally identified for almost all factor configurations, even when additive fixed effects or unit‑root processes are present, by leveraging the level covariance structure.
Abstract
We investigate the problem of global identification in dynamic panel models with interactive effects, in 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 for almost all configurations of the factors. Our analysis also covers models with additive fixed effects, including unit-root cases in which previous studies have reported non-identification from differenced moments. We show that, even in these settings, the level covariance structure delivers global identification.