56 Methodology of Empirical Econometric Modeling
̂σ=0.10FAR1− 2 (2, 220)=2.59FARCH 1 (1, 220)=0.01χnd^2 ( 2 )=3.72FHet(17, 204)=1.21FRESET(1, 221)=3.72.Most of the effects found make economic sense in the context of the limited
information set used here as an illustration. In reverse order, real equity prices are
near a random walk, but respond positively to changes in output, and negatively to
changes in bankruptcies. In turn, bankruptcies fall with increased output or equity
prices, but rise with patent grants. Neither equation has many outliers, whereas the
patents equation does, especially in the eighteenth century. Patents fall initially as
output, equity prices, bankruptcies rise, but adjust back later. Finally, changes in
output respond positively to patents and changes in equity prices, but negatively
to bankruptcies.
A substantive exercise would involve additional variables like interest rates and
human and physical capital; would check whether bankruptcies and patents should
also be per capita; and investigate cointegration reductions. Are the long lags
‘spurious’? The general historical record suggests that major innovations are both
creative and destructive of output, the former by the enlargement of the production
frontier, and the latter through the negative impact on those already engaged in
the occupations concerned (spinners, weavers, etc., initially; clerks and secretaries
in more modern times), so a “generation” is required for the new state to dominate
- that motivated the original choice of 25 lags. Innovations take time to develop
and be adopted; and the seeds for bankruptcy are often sown well before the reap-
ing, even if the span is not quite “clogs to clogs in three generations.” Notably,
the equation for equity prices still has short lags despite the “opportunity” to find
other correlations.
1.9 Conclusion
Ever drifting down the stream –
Lingering in the golden gleam –
Life, what is it but a dream? (Lewis Carroll, 1899)“Applied Econometrics” has a vast range of empirical issues to investigate: the very
non-stationarity of economies keeps creating new topics for analysis. However,
so long as “Applied Econometrics” is just a calibration of extant economic the-
ory, it will never make much of an independent contribution: in that sense, one
must agree with Summers (1991) but for completely opposite reasons. Much of
the observed data variability in economics is due to features that are absent from
most economic theories, but which empirical models have to tackle.Ceteris paribus
conditions can sometimes be justified for theoretical reasoning, but do not provide
a viable basis for empirical modeling: only a “minor influence” theorem, which
must be established empirically, will suffice.
This implication is not a tract for mindless modeling of data in the absence of
economic analysis, but instead suggests formulating more general initial models