Forecasting Euro‐Area Macroeconomic Variables Using a Factor Model Approach for Backdating
| Date | 01 February 2015 |
| DOI | http://doi.org/10.1111/obes.12053 |
| Author | Ralf Brüggemann,Jing Zeng |
| Published date | 01 February 2015 |
22
©2013 The Department of Economics, University of Oxford and JohnWiley & Sons Ltd.
OXFORD BULLETIN OF ECONOMICSAND STATISTICS, 77, 1 (2015) 0305–9049
doi: 10.1111/obes.12053
Forecasting Euro-Area Macroeconomic Variables
Using a Factor ModelApproach for Backdating*
Ralf Br ¨uggemann† and Jing Zeng†
†University of Konstanz, Department of Economics, Box 129, 78457 Konstanz, Germany
(email: ralf.brueggemann@uni-konstanz.de) (email: jing.zeng@uni-konstanz.de)
Abstract
We suggest to use a factor model based backdating procedure to construct historical Euro-
area macroeconomic time series data for the pre-Euro period. We argue that this is a useful
alternative to standard contemporaneous aggregation methods. The article investigates for
a number of Euro-area variables whether forecasts based on the factor-backdated data are
more precise than those obtained with standard area-wide data. A recursive pseudo-out-
of-sample forecasting experiment using quarterly data is conducted. Our results suggest
that some key variables (e.g. real GDP, inflation and long-term interest rate) can indeed be
forecasted more precisely with the factor-backdated data.
I. Introduction
With the creation of European Monetary Union (EMU), the focus of macroeconomic anal-
ysis has shifted towards the analysis of the Euro area as a whole. Econometric models for
area-wide variables have been used for forecasting and structural analysis.As actual EMU
time series data are only available from 1999 onwards, synthetic time series data for the
pre-EMU period are in use. Often the construction of historical (pre-EMU) Euro area data
is based on contemporaneous aggregation of time series from the EMU member coun-
tries. Different aggregation methods have been suggested in the literature and Marcellino
(2004) points out a number of drawbacks inherent in these methods. The choice of a par-
ticular aggregation method is a very important practical issue that impacts any following
econometric analysis. For instance, Bosker (2006) illustrates that estimated cointegration
parameters change substantially with the choice of the aggregation method. Given these
drawbacks of standard methods, it is worth to consider the merits of alternatives to aggre-
gation. In this article, we therefore consider a factor model based alternative to the standard
method of contemporaneous aggregation and analyze the usefulness of this approach in
forecasting Euro area aggregates.
*Wethank par ticipants in the DIW Berlin MacroeconometricWorkshop 2009, the research seminar of the Center
of Quantitative Methods and Survey Research, Konstanz, the EEA 2010 Glasgow, the CFE 2010 London, and the
ESEM 2011 Oslo as well as two anonymousreferees and the editor for helpful comments and suggestions. Financial
support by the Deutsche Forschungsgemeinschaft, project number BR 2941/1-1, is gratefully acknowledged.
JEL Classification numbers: C22, C53, C43, C82.
Forecasting using a factor model for backdating 23
One of the standard aggregation methods suggested in the literature has been discussed
by Fagan,Henry and Mestre (2001, 2005). Their approach has been used to create a database
of historical Euro-area time series data for estimating theArea Wide Model (AWM)in use at
the European Central Bank (ECB).1The AWM data is based on cross-country aggregation
of log-level variables with fixed weights (referred to as FHM weights). The FHM weights
are obtained as shares of GDP at constant 1995 prices. Anderson et al. (2011) point out
that the use of fixed weights will tend to undervalue the importance of the countries, which
hold a leading role in the European markets and suggest extending the FHM weights with
a sliding factor which measures the relative distance from economic integration to EMU.
Using fixed weights mayalso be problematic because it does not take changes in exchange
rates between member countries into consideration.Therefore, Beyer, Doornik and Hendry
(2001) suggest to aggregate growth rates of the variables with time-varying weights based
on previous period’s real GDP share (henceforth BDH weights) and find that in their method
the aggregates of the individual deflators correspond to the deflator of the aggregate. Beyer
and Juselius (2009) show that results based on BDH weights are sensitive to the choice
of the base year and therefore suggest to use weights based on previous period’s nominal
GDP. None of the proposed methods seems optimal in all respects.
Alternatives to standard aggregation have also been considered in the literature. For
instance, Br¨uggemann and L¨utkepohl (2006) and Br¨uggemann, L¨utkepohl and Marcellino
(2008) argue that the use of synthetically constructed, aggregated data is inappropriate
especially in the presence of structural changes induced by adjustment processes required
in some countries prior to EMU in order to satisfy the Maastricht criteria. They suggest
a representative country approach which combines German data until 1998 with actual
Euro-area data after 1999. They find that at least for some variables like interest rates and
prices using German data rather than aggregated EMU data for the pre-EMU period is
preferable when forecasts of EMU aggregates are of interest.
This article proposes to use another alternative method for constructing historical Euro-
area data. We use the idea put forward inAngelini et al. (2006) and Angelini and Marcellino
(2011), where a factor based approach is used to construct time series of macroeconomic
variables for unified Germany prior to 1991. In the factor model approach, a small number
of factors are extracted from a large set of time series from individual EMU member
countries using the Stock and Watson (2002a) principal component based estimators.The
estimated relation between the factor time series and the actual Euro-area time series
of interest is used to construct time series data for the pre-EMU period. This method
is referred to as factor-backdating. Advantages of this method include its ability to use
more time series information than standard aggregation methods and its ability to handle
situations with missing time series data in some of the cross-sectional units (countries).
Against the background of future EMU enlargement and the doubtful quality of historical
data in some of the future member countries, the factor-backdating procedure may be an
attractive and useful alternative to standard aggregation methods.
We analyze the usefulness of this approach in forecasting a number of macroeconomic
Euro-area variables by conducting a forecast comparison. We compare the accuracy of
1Updates of this database are available from the Euro Area Business Cycle Network (EABCN) at
http://www.eabcn.org/.
©2013 The Department of Economics, University of Oxford and JohnWiley & Sons Ltd
To continue reading
Request your trial