Estimating and Predicting the General Random Effects Model

This paper extends the ‘remarkable property’ of Breusch (Journal of Econometrics 1987; 36: 383–389) and Baltagi and Li (Journal of Econometrics 1992; 53: 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, θi (i = 2, 3, 4). More specifically, a search over θiwhile iterating on the regression coefficients estimates β and the other θjwill 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.