PRACTITIONERS' CORNER: Using PC‐GIVE in Econometrics Teaching
| Author | David F. Hendry |
| Date | 01 February 1986 |
| Published date | 01 February 1986 |
| DOI | http://doi.org/10.1111/j.1468-0084.1986.mp48001007.x |
OXFORD BULLETIN OF ECONOMICS AND STATISTICS. 48, 1(1986)
0305-9049 $3.00
PRACTITIONERS' CORNER
Using PC-GIVE in Econometrics Teaching
David F. Hendry
I. INTRODUCTION
PC-GIVE is an interactive, menu-driven econometrics program designed
for modelling time-series data in the light of economic theory when the
exact specification of the relationship of interest is not known for
certain a priori. Deriving from GIVE in the AUTOREG library (see
Hendry and Srba, 1980), PC-GIVE uses fast and tested FORTRAN sub-
routines for accurate numerical calculations, embedded in a user-
friendly and highly protected environment which avoids having to learn
any idiosyncratic or complicated command language. As such, it is easy
to use and is designed to cover both elementary classroom demonstra-
tions as well as research which requires powerful diagnostic testing
options.
This note briefly describes the structure and main functions of
PC-GIVE in Section II, and its usage for teaching in Section III. The
remainder of this section sketches the rationale for developing a specific
program for econometric modelling of economic time series.
The philosophy underlying PC-GIVE is that observed data are most
usefully viewed as arising from a data generation process (DGP) of
immense generality and complexity. The econometrician seeks to model
the salient features of the DGP in a simplified representation based on
the observables and related to prior economic (or other subject-related)
theory. Because many important data features are inevitably assumed
absent in any theory, researchers must discover model forms which (in
the senses to be explained below) adequately characterize the data
consistently with the theory (although perhaps only weakly so).
For example, while a theory-model might assume white-noise errors,
the lack of any precise mapping of decision periods to data observation
periods may mean that the estimated model manifests substantial
residual autocorrelation. Alternatively, 'parameters' may not be
constant over time. If modelling proceeds from simple initial specifica-
tions, it is essential that researchers be aware of such 'problems'.
Current modelling practice suggests that an important component of
any data characterization exercise is to estimate the most general model
which it is reasonable to entertain a priori. However, modelling based
87
To continue reading
Request your trial