| That econometric methodology remains in dispute partly reflects the lack of clear evidence on alternative approaches. This paper reconsiders econometric model selection from a computer-automation perspective, focusing on general-to-specific reduction approaches, as embodied in the program PcGets (general-to-specific). Starting from a general linear, dynamic statistical model, which captures the essential data characteristics, standard testing procedures are applied to eliminate statistically-insignificant variables, using diagnostic tests to check the validity of the reductions, ensuring a congruent final model. As the joint issue of variable selection and diagnostic testing eludes most attempts at theoretical analysis, a simulation-based analysis of modelling strategies is presented. The results of the Monte Carlo experiments cohere with the established theory: PcGets recovers the DGP specification with remarkable accuracy. Empirical size and power of PcGets are close to what one would expect if the DGP were known. |