Estimating and Predicting the General Random Effects Model

ABSTRACTThis paper extends the ‘remarkable property’ of Breusch (Journal of Econometrics 1987; <b>36</b>: 383–389) and Baltagi and Li (Journal of Econometrics 1992; <b>53</b>: 45–51) to the three‐way random components framework. Indeed, like its one‐way and two‐way counterparts, the three‐way random effects model maximum likelihood estimation can be obtained as an iterated generalized least squares procedure through an appropriate algorithm of monotonic sequences of some variance components ratios, θ_i (i = 2, 3, 4). More specifically, a search over θ_iwhile iterating on the regression coefficients estimates β and the other θ_jwill guard against the possibility of multiple local maxima of the likelihood function. In addition, the derivations of related prediction functions are obtained based on complete as well as incomplete panels. Finally, an application to international trade issues modeling is presented. Copyright © 2014 John Wiley & Sons, Ltd.