DATA MINING RECONSIDERED: ENCOMPASSING AND THE GENERAL-TO-SPECIFIC APPROACH TO SPECIFICATION SEARCH

The effectiveness of one aspect of the London School of Economics (LSE) approach to econometrics is assessed in a simulation study. The paper uses a data set and nine models analogous to those in Lovell''s (1983) study of data mining. A simplified general-to-specific algorithm is tested in a simulation framework. While the study documents some of the pitfalls of the general-to-specific approach, it is, on the whole, supportive of the effectiveness of the LSE methodology as applied to stationary data with relatively simple dynamics. The general-to-specific methodology clearly dominates the alternative search methodologies investigated by Lovell.