The Probability Of Finding The Data Generation Process After A t2-Testing Sequential Procedure
Julia Campos · RePEc: Research Papers in Economics · 2004
Recently developed algorithms for model selection proceed by ordering the variables by their absolute t−values. So much testing involved in choosing the Þnal model, and the generality of the initial model have been questioned in the literature. This paper shows that increasing model generality and repeated t2−testing do not reduce the chances of Þnding the DGP. However, if somewhere along the procedure an irrelevant variable is incorrectly included then all following variables in the sequence are retained with probability one. JEL classiÞcation: C12, C22, C32, C51, C52.