Automatic Variable Selection

Abstract

In practice, specification of empirical econometric models relies on search. Researchers use both general-to-specific (Gets) and specific-to-general searches. The chapter explains how Gets can be semiautomatized to aid economists in making decisions about model specification. We first assume that the aim of model specification is to conclude with an empirical model which is a close approximation to the local data generating process (LDGP). Gets represents a sensible strategy that hinges on the premise that the LDGP is contained within the general unrestricted model (GUM), which is the starting point of the specification search. Over the past decades, Gets has evolved to become a theoretically well-founded methodology, and recently has also been supported by computer algorithms that make automatized variable selection possible. In this chapter, we review the main concepts and theory needed to use Gets, manually, or with the aid of an computer algorithm such as Autometrics. In the last part of this chapter, we discuss the usefulness of automatic Gets when the LDGP is not necessarily contained within the GUM, and show that the method can deliver a meaningful approximation to the LDGP in this case also with the aid of extended methods such as Impulse Indicator Saturation (IIS).