Introduction

Abstract

This study examines multiple aspects of the way in which the development of computer software, specifically econometric software, has affected and, in the future, might affect the practice of econometrics. Its publication is preceded by over 50 years of software development and use, during which time economists have paid little attention to the possibility that the particular computational tools they employ might have any discernable impact on their research, and which therefore might imply the lack of perceived need for such a study. Except for the speed of calculation effect of the electronic computer, which has long been recognized, software as an affective econometric tool is a novel idea. Indeed, historically, econometricians have usually interpreted the “tools of econometrics” to be its conceptual methods, often considered somewhat abstractly. For example, in 1966, under this rubric and when considering the econometric testing of an hypothetical statement about the empirical world, Jacob Marschak (1966), following Harold Hotelling, distinguished between economic theory-based maintained, or “prior” propositions, as assumptions or “specifications,” in contrast to those properties to be immediately tested against observation. He optimistically characterized these specifications as possibly derived from prior observation, perhaps as a result of sequential testing, although he spoke of models and “structures” in a fashion that to modern ears might seem somewhat anachronistic. The idea of testing being a truth discovery process is implicit in his argument, perhaps stemming from a common acceptance then of at least quasi-axiomatic foundations for economic theory. Yet he also recognized, in a way that is still up-to-date (p. ix), the difficulty the economist has in assigning “future validity to the patterns of the past. For policy change may consist in changing the very mechanism by which the environment influences economic variables,” requiring that the economist “must therefore peek in the interior of the notorious “black box” that operated in the past and describe policy changes as specific changes of that interior.” This recognition of a policy inspired requirement to represent economic phenomena in a manner so as to permit the economist to make out-of-sample predictions using structural knowledge is historically significant, for the distinction that Marschak made between the “black box” reduced form and its corresponding, possibly changeable structural representation involves of course both the identification problem and the well-known Cowles Commission methodology (Marschak, 1953). Directly and indirectly, a consequence of this distinction was the adoption of a conceptual approach, beginning in the 1940s (Haavelmo, 1944; Hood & Koopmans, 1953; Koopmans, 1950; Morgan, 1990; Qin, 1993), that then gave rise to a substantial econometrics literature.