Specification Error in Macro-Econometric Models: The Influence of Policy Goals-Comment
Stephen Goldfeld's conclusions that analysis of my article is flawed by a statistical confusion and that the extra equation is a red herring pure and simple are based on a set of assumptions which are virtually opposite of ones I worked with, and depend on an excessively restrictive interpretation of what is implied by use of Theil certainty equivalence framework. Under assumptions stated in my article, my conclusions remain valid. It is Goldfeld's new assumptions that create errors he then discovers in my paper. The policy problem that I modelled was one in which state had relatively firm policy targets and adjusted its policy instruments in response to shifts in expected of uncontrolled variables. In order to emphasize potential problem, I assumed that X* and Y* were constant, so that X was chosen to minimize equation (1), quadratic loss function, as S, expected value of S, shifted from to period.' As I stated in article, clearly changes every period (p. 1027). But I used symbol S to denote policy maker's certainty-equivalent forecast of S, i.e., Goldfeld's S. My policy examples all contained same assumptions. Instrument movements would ... counterbalance other forces tending to move goal variables away from their desired values (p. 1029) represents a typical sample from my discussion. Following Theil, I assumed that policy maker behaved as if S, expected value of his subjective probability distribution (or forecast distribution) of S, was known with certainty, but not that forecast distribution or S never shifted or changed over time.2 Goldfeld reverses these assumptions to describe and analyze a fickle policy maker in a world of constant expected of uncontrolled variables, very opposite of my assumptions and a relatively uninteresting case at that. He argues that natural presumption is X* and Y* vary nonrandomlv over sample and, simplicit , that S is constant over sample. Only for case where S is constant is Goldfeld's conclusion in footnote 5 that constancy of X* and Y* implies constancy of policy instruments X correct. In footnote 4 Goldfeld suggests that assumption that S is constant is convenience only, but with respect to problem addressed in my article, this is not so. To see this, define Et=St-st.3 Here e is a random variable with an expected value of zero and constant variance, and true reduced form is y=rx+-+E. Under Gold-
- Export
- BibTeX
- Sources
- openalex