David F. Hendry 57that embed the available economic theory as a special case, consistent with our
knowledge of the institutional framework, historical record, and the data proper-
ties. Once a congruent encompassing general model is established, an automatic
model selection approach based on general-to-simple principles could help bring
objectivity and credibility to empirical econometric modeling.
Economic observations are far from perfect, being subject to revision, and even
to conceptual changes, with important variables unobserved, and available prox-
ies of unknown quality. Theory constructs (such as “consumption,” or “user cost
of capital”) and their measured counterparts (consumers’ expenditure or after-tax
real interest rates adjusted for depreciation) can differ markedly, especially after
aggregation. Thus, a “pure” data-based approach can lack substance.
Economics has delivered a range of invaluable insights into individual deci-
sion taking, market functioning, and system-wide economies, with a vast body
of theory, which has made rapid technical and intellectual progress – and will con-
tinue to do so. Applied econometrics cannot be conducted without an economic
theoretic framework to guide its endeavors and and help interpret its findings.
Nevertheless, since economic theory is not complete, correct, and immutable, and
never will be, one also cannot justify an insistence on deriving empirical models
from theory alone. That paradigm encourages covert data mining, so the credibility
of existing evidence is unclear.
Data “mining” does not have pernicious properties when using a structured
approach, using appropriate significance levels that decline with both the number
of candidate variables and sample size: at 1% significance, one irrelevant variable
in 100 will be significant by chance, at the cost of raising the selectiont-ratio from
around±2.0 to±2.7. Parsimony is not a justification for arbitrarily excluding many
potentially relevant contenders, not even when doing so to avoid more initial vari-
ables than observations. While it is essential that the final model is much smaller
than the sample size, that does not preclude starting general and making the max-
imum use of our best available theory and econometrics to guide our empirical
endeavors and then interpret their outcomes. Thus, Frisch (1933) remains our best
advice: “mutual penetration’, which entails using economic analysis to guide an
applied study, but letting the empirical evidence play a real role.
Acknowledgments
Financial support from the ESRC under Research Grant RES-062-23-0061 is gratefully
acknowledged. I am indebted to Gunnar Baardsen, Julia Campos, Jennifer Castle, Guillaume
Chevillon, Jurgen Doornik, Neil Ericsson, Katarina Juselius, Søren Johansen, Bobby Mariano,
Jaime Marquez, Terry Mills, Mary Morgan, Bent Nielsen, Ragnar Nymoen, Kerry Patterson,
Duo Qin, James Reade, Aris Spanos and Pravin Trivedi for helpful comments.
Note
- Atkinson (2008) notes Robbins’ apparent dismissal of Richard Stone (1951) as “not
economics.”