Retrospective on ‘Econometric Modelling: The Consumption Function in Retrospect’, Scottish Journal of Political Economy, 30 (1983), 193–220'
| Date | 01 November 2013 |
| DOI | http://doi.org/10.1111/sjpe.12028 |
| Published date | 01 November 2013 |
RETROSPECTIVE ON ‘ECONOMETRIC
MODELLING: THE CONSUMPTION
FUNCTION IN RETROSPECT’,
SCOTTISH JOURNAL OF POLITICAL
ECONOMY,30 (1983), 193–220’
David F. Hendry
BACKGROUND
Hendry (1983) was written to explain and systematise the approach used by
Davidson, Hendry, Srba and Yeo (1978) for modelling real consumers’ expendi-
ture, c
t
, in the light of the various aspects of econometric theory that had been
developed between 1975 and 1982. The literature on modelling consumers’
expenditure had bifurcated into Euler-equation approaches (see Hall, 1978) and
so-called ‘reduced form’ models based on real personal disposable income, y
t
,
and although I regarded the former as rejected by Davidson and Hendry (1981),
it remained popular. An early discussion of the theory of reduction and its asso-
ciated information taxonomy providedthe framework. The papers on exogeneity
(see Engle, Hendry and Richard, 1983), modelling concepts (Hendry and Rich-
ard, 1982) and autoregressive conditional heteroskedasticity (ARCH: see Engle
1982) had barely been published, and those on encompassing were in process(see
Mizon and Richard, 1986), so there was much new materialto explain. The novel
empirical component was modelling annual United Kingdom data for the inter-
war period, 1920–1938, using an ‘error-correction’ mechanism (ECM), prior to
the start of the cointegration era (see Engle and Granger, 1987).
SUBSEQUENT DEVELOPMENTS
Modelling theory, in the form of cointegration analysis and model selection
methods, and computer software implementing those methods have advanced
almost out of recognition since 1983. It took but a few moments to model
that earlier data set using Autometrics within PcGive. The 1983 paper had an
extensive discussion of parameter constancy, but since then, impulse-indicator
saturation (IIS: see e.g., Johansen and Nielsen, 2009) greatly extends the num-
ber and types of outliers and location shifts that can be tackled jointly with
model selection over variables, lags and non-linear forms. Instead of inflation,
the model selected over T=1920–1936 using IIS and reported in equation (1),
included indicators for 1926 and 1932, both salient dates, with a smaller equa-
tion standard error of ^
r¼0:43%. Are c
t
and y
t
actually cointegrated over
1920–1938? Phew: despite the short sample, the PcGive test t
ur
=4.07**
Oxford University
Scottish Journal of Political Economy, Vol. 60, No. 5, November2013
©2013 Scottish Economic Society.
523
To continue reading
Request your trial