A History of Econometric Metaphysics and their Paradigms
In the 1950s and 1960s, the dominant approach in econometrics was the CC approach (Hildreth 1986; Epstein 1987; Christ 1952, 1994). The Cowles Commission for Research in Economics (CC) was set up in 1932, being funded by Alfred Cowles (1891-1984) specifically to undertake econometric research.
In 1939, it moved to the University of Chicago, and in 1955 it moved to Yale University where it was renamed as the Cowles Foundation for Economic Research. The Econometric Society’s journal, Econometrica, was published by the Commission. The CC’s econometric approach, developed in the 1940s and 1950s, became the standard approach found in most econometric textbooks; this approach is also known as the simultaneous equation modelling (SEM) approach.One of the most popular of these textbooks, Jack Johnston’s Econometric Methods, identifies the basic task of econometrics as being “to put empirical flesh and blood on theoretical structures” (Johnston 1984: 5, original emphasis). For Johnston, this involves three distinct steps:
1. The model must be specified in explicit functional - often linear - form.
2. Decide on the appropriate data definitions, and assemble the relevant data series for those variables included in the model.
3. Form a bridge between theory and data through the use of statistical methods. The bridge consists of various sets of statistics, which help to determine the validity of the theoretical model.
The most important set of statistics consists of numerical estimates of the model’s parameters. Further statistics enable an assessment to be made of the reliability or precision with which these parameters have been estimated. Further statistics and diagnostic tests will help to assess the performance of the model.
Trygve Haavelmo’s 1944 paper, “The probability approach in econometrics”, laid the groundwork upon which the CC developed a more advanced methodology for macroeconometric modelling than Tinbergen’s (1936, 1939).
Haavelmo (1911-1999) had been a student of Frisch and later his research assistant. The Haavelmo paper moved the emphasis of econometrics from the measurement of parameters (for example, demand elasticities) and the problem of the quality of statistical data, to the testing of theories. It introduced the probabilistic approach into econometrics by showing how to use the Neyman and Pearson statistical theory of hypothesis testing. It became the “blueprint” of an approach which is now identified with the CC approach, and which emerged via two monographs (Koopmans 1950; Hood and Koopmans 1953) published by the Commission.From the mid-1970s onwards, there was a growing scepticism towards the value of the more theoretical CC approach for applied econometric research. There was a gap between what theoretical econometricians were developing and what applied econometricians actually needed. These discussions lead to Hendry’s (1980) “Alchemy or science?” paper and Leamer’s (1983) “Let’s take the con out of econometrics” paper, but also to Christopher Sims’s (1980) “Macroeconomics and reality”.
A result of these publications, for the first time, a special invited symposium on econometric methodology was held in 1985 at the Fifth World Congress of the Econometric Society. The invited speakers were Hendry, Leamer and Sims. Adrian R. Pagan (1987) was invited as discussant and published an “appraisal” of this debate (see also Hendry et al. 1990). To prevent econometrics from becoming statistical alchemy, Hendry, Leamer and Sims developed their own methodologies: the general-to-specific approach (Hendry and Richard 1982), the Bayesian approach (Leamer 1978), and the vector autoregression (VAR) approach (Sims 1980), respectively.
The VAR approach arose as an answer to the growing criticism of the CC structural approach with its assumptions of zero restrictions to achieve identification of a model and the division of the variables into endogenous and exogenous ones.
The questioning of the role of zero restrictions had already started in the 1960s, due to the work of Ta-Chung Liu (1914-1975) (Chao and Huang 2011). Liu (1960) had noticed that the number of restrictions needed to identify large-scale macro-econometric models far exceeded the number that economic theory could be confidently relied upon to provide. The danger is that models may be formulated and estimated with some variables added to equations without much economic justification and other variables deleted (that is, zero restrictions imposed), in order to achieve identification. The a priori restrictions imposed on structural simultaneous equation models were viewed by Sims as “incredible”.Sims’s proposal was to remove the pretence of applying theoretical structure to the data and, instead, to use unrestricted systems of reduced form equations or vector autoregressions to model the responses of variables to shocks. This does not mean however that the VAR approach should be considered as “atheoretical macroeconometrics”. Stimulated by the rational expectations movement in macroeconomics, the VAR approach offers a systematic procedure to tackle the issue of model choice. It shifted the focus from measurement of given theories to identification of data-coherent theories (Qin 2011).
The model of James E.H. Davidson, Hendry, Frank Srba and Stephen Yeo (1978) of UK aggregate consumption expenditure, published in 1978 was a major influence on the way econometricians now use time series data to model economic relationships. One of its main methodological innovations was the general-to-specific approach to deal with time series data. General-to-specific modelling is the formulation of a fairly unrestricted dynamic model which is subsequently tested, transformed and reduced in size by performing a number of tests for restrictions (Hendry 1995). This approach also came to be known as the LSE approach and is in the spirit of the Box-Jenkins approach.
The approach proposed by Box and Jenkins (1970) introduced a number of concepts that were either missing or neglected in the CC approach: the principle of parsimony in model formation, diagnostic checks on model adequacy, and a data-led identification procedure.
Leamer observed a wide gap between the formal textbook approach and its practised variant, which Leamer called “cookbook econometrics” (Leamer 1978):
As it happens, the econometric modeling was done in the basement of the building and the econometric theory courses were taught on the top floor (the third). I was perplexed by the fact that the same language was used in both places. Even more amazing was the transmogrification of particular individuals who wantonly sinned in the basement and metamorphosed into the highest of high priests as they ascended to the third floor. (Leamer 1978: vi)
He therefore suggested the making of more “honest” modellers out of cookbook econometricians by formalizing their ad hoc procedures using informal Bayesian procedures such as extreme bounds analysis. This should expose the possible fragility of estimated relationships by testing their robustness to changes in prior information (Leamer and Leonard 1983).