Research Master Pre-Defense
Stationary or non-stationary? An investigation on the initial conditions for panel maximum likelihood estimation.
-
Speaker(s)Sander Tromp , Sander Tromp
-
LocationTinbergen
Amsterdam
-
Date and time
July 07, 2025
10:00 - 11:30
This thesis investigates some theoretical properties of the First-Difference Maximum Likelihood (FDML) estimator. The properties are derived by casting the estimator into the alternative Transformed Maximum Likelihood (TML) framework. Subsequently, the impact is investigated when one deviates from the covariance-stationary initial condition. Under this deviation, the expected score of the autoregressive parameter of interest will be biased. The analytical form of the bias is derived for both the time-series homoskedastic and heteroskedastic cases. Moreover, one can derive the limiting distribution of the estimator. The results imply that a deviation of the covariance-stationary initial condition impacts the possibility of inference on the estimated parameters. The analytical results are verified using a Monte Carlo study.