Common change point estimation in panel data from the least squares and maximum likelihood viewpoints

We establish the convergence rates and asymptotic distributions of the common break change-point estimators, obtained by least squares and maximum likelihood in panel data models and compare their asymptotic variances. Our model assumptions accommodate a variety of commonly encountered probability distributions and, in particular, models of particular interest in econometrics beyond the commonly analyzed Gaussian model, including the zero-inflated Poisson model for count data, and the probit and tobit models. We also provide novel results for time dependent data in the signal-plus-noise model, with emphasis on a wide array of noise processes, including Gaussian process, MA$(\infty)$ and $m$-dependent processes. The obtained results show that maximum likelihood estimation requires a stronger signal-to-noise model identifiability condition compared to its least squares counterpart. Finally, since there are three different asymptotic regimes that depend on the behavior of the norm difference of the model parameters before and after the change point, which cannot be realistically assumed to be known, we develop a novel data driven adaptive procedure that provides valid confidence intervals for the common break, without requiring a priori knowledge of the asymptotic regime the problem falls in.