Constrained and Unconstrained Maximum Likelihood Estimation of a Variance Components Model of Cross- Sections Pooled Over Time

Abstract

The authors report on simulations on the quality of parameter estimates of regression coefficients with lagged variables. Results showed that the quality of estimates varied with the amount of serial error correlation and with the relative strength of effects of lagged variables. Estimates of the coefficient of an exogenous variable should be very similar by maximum likelihood estimates and modified generalized least squares. If they are not close to identical, an investigator should suspect misspecification of the model.