Cross‐data‐vintage Encompassing*
| Date | 01 December 2008 |
| Published date | 01 December 2008 |
| Author | Steve Cook |
| DOI | http://doi.org/10.1111/j.1468-0084.2008.00533.x |
849
©Blackwell Publishing Ltd and the Department of Economics, University of Oxford, 2008. Published by Blackwell Publishing Ltd,
9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.
OXFORD BULLETIN OF ECONOMICS AND STATISTICS, 70, SUPPLEMENT (2008) 0305-9049
doi: 10.1111/j.1468-0084.2008.00533.x
Cross-data-vintage EncompassingÅ
Steve Cook
School of Business and Economics, Swansea University, Singleton Park, Swansea SA2 8PP,
UK (e-mail: s.cook@swan.ac.uk)
Abstract
The issues of model comparison and evaluation in the presence of data revision are
examined. The initial analysis revisits the model validity and data accuracy tests of
Hendry [Oxford Review of Economic Policy (1994) Vol. 10, pp. 86–106] to develop
the concept of model and vintage coalescence (MVC). A taxonomy of MVC tests is
proposed, with the finite-sample properties of the tests examined via Monte Carlo
simulation. The analysis proceeds to extend the encompassing principle [see Mizon
and Richard, Econometrica (1986), Vol. 54, pp. 657–678; Contributions to Operations
Research and Econometrics: The XXth Anniversary of CORE, Hendry and Richard
(1989) MIT Press, Cambridge, MA] to permit the comparison of econometric models
developed upon alternative vintages of data. The resulting cross-data-vintage encom-
passing tests are discussed, with their finite-sample properties compared with those of
conventional encompassing tests. The paper concludes by considering implications
of the present analysis for econometric modelling.
I. Introduction
Economic data are regularly revised, resulting in the frequent appearance of different
vintages of economic variables over a particular time period. The practical impor-
tance of data revision to the applied researcher is reflected in the growing literature
ÅI am extremely grateful to David Hendry for many helpful comments and suggestions which have improved
both the content and presentation of this paper. I am indebted to Grayham Mizon and two anonymous referees
for additional comments. However, the usual disclaimer applies. It is my privilege to contribute to a volume
in honour of Grayham Mizon. Like very many others my understanding of econometrics has been greatly
enhanced by Grayham’s research. In addition, I am also in the fortunate position of having benefited from
Grayham’s inspirational teaching during my time as a postgraduate student and his continual support and
insights while working as his research assistant. Being able to draw upon the above has helped me enormously
and for that I am extremely grateful.
JEL Classification numbers: C12, C15, C52.
850 Bulletin
examining the existence and properties of alternative vintages of economic data and
the related issue of real-time data (see, inter alia, Mankiw, Rankle and Shapiro,
1984; Mankiw and Shapiro, 1986; Patterson and Heravi, 1991; Egginton, Pick and
Vahey,2002; Patterson, 2002; Croushore and Stark, 2003; Garratt and Vahey, 2006).1
In a more specific analysis of data revision, Hendry (1994) considers the role such
revisions may have played in the observed breakdown of models of UK aggregate
consumers’ expenditure. This is achieved by analysing the reproducibility of the
Hendry and von Ungern-Sternberg (1980) (HUS) model of consumers’ expenditure
on a more recent vintage of the sample data upon which it was developed. The analysis
conducted on the HUS model of consumption leads Hendry to conclude that a differ-
ent model would have been selected on the newer vintage. In the process of examining
the impact of data revision upon aggregate consumption functions, Hendry proposes
a number of tests to investigate which data vintage is ‘accurate’ and which model
is ‘valid’ when examining differing specifications derived on alternative vintages of
data.2The fact that data revision can have such an impact on econometric modelling
and the idea that the accuracy of alternative vintages and the validity of models can be
tested provide the initial motivation for the present study. This relationship between
model validity and data accuracy is discussed in the following section, where the issue
is developed to arrive at the concept of ‘model and vintage coalescence’ (MVC), a
term which emphasizes the two-way nature of the relationship between the data and
the model, rather than the individual claims of validity and accuracy. Section II also
provides a taxonomy of MVC tests and revisits the empirical application presented in
Hendry (1994). In section III a Monte Carlo analysis of the MVC tests is undertaken.
This simulation analysis helps explain an anomaly arising from consideration of a
previously unreported test.
In section IV the concept of MVC is extended to arrive at the focus of this pres-
ent study, the notion of cross-data-vintage encompassing, where the encompassing
principle (see Mizon, 1984; Mizon and Richard, 1986; Hendry and Richard, 1989)
is generalized to permit the comparison of models developed on alternative data
vintages while allowing for the influence of these differing data series. It has been
noted previously in the literature that by allowing the relative strengths and weak-
nesses of rival models to be assessed, the encompassing principle can lead to the
development of a sequence of models, with each successive model representing an
advance on its predecessor. However, for real progress to be made in econometric
modelling it is vital that the role of the data in the performance of models, and hence
in any model comparison exercise, is taken into consideration. As noted by Hendry
(1994), data revision can lead to alternative models being chosen, and so the apparent
encompassing by one model of its predecessor may be due to the latter failing to cope
1Similarly, the importance of data revision is evident from the increasing number of real-time data sets
provided by, inter alia, the OECD, Federal Reserve and Bank of England.
2Hendry (1994) applies these tests to the Davidson et al. (1978) model of consumers’ expenditure, rather
than the HUS model, over original and revised data sets because of the unavailability of the original HUS data
on liquid assets.
©Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2008
To continue reading
Request your trial