Forecasting GDP over the Business Cycle in a Multi‐Frequency and Data‐Rich Environment
| Author | Marie Bessec,Othman Bouabdallah |
| DOI | http://doi.org/10.1111/obes.12069 |
| Date | 01 June 2015 |
| Published date | 01 June 2015 |
360
©2014 The Department of Economics, University of Oxford and JohnWiley & Sons Ltd.
OXFORD BULLETIN OF ECONOMICSAND STATISTICS, 0305–9049
doi: 10.1111/obes.12069
Forecasting GDP over the Business Cycle in a
Multi-Frequency and Data-Rich Environment*
Marie Bessec† and Othman Bouabdallah‡
†LEDa-CGEMP Universit´e Paris Dauphine, Place du mar´echal de Lattre de Tassigny
75016, Paris France (e-mail: marie.bessec@dauphine.fr)
‡European Central Bank, Kaiserstrasse 29, 60311, Frankfurt am Main Germany
(e-mail: Othman.Bouabdallah@ecb.europa.eu)
Abstract
This paper merges two specifications recently developed in the forecasting literature: the
MS-MIDAS model (Gu´erin and Marcellino, 2013) and the factor-MIDAS model (Mar-
cellino and Schumacher, 2010). The MS-factor MIDAS model that we introduce incorpo-
rates the information provided by a large data set consisting of mixed frequency variables
and captures regime-switching behaviours. Monte Carlo simulations show that this speci-
fication tracks the dynamics of the process and predicts the regime switches successfully,
both in-sample and out-of-sample. We apply this model to US data from 1959 to 2010
and properly detect recessions by exploiting the link between GDP growth and higher
frequency financial variables.
I. Introduction
The recent financial crisis has heightened practitioners’ interest in models that differentiate
GDP dynamics over the course of the business cycle, as that first initiated by Hamilton
(1989). In order to forecast GDP dynamics, macroeconomists are able to mobilize a very
large set of indicators, as suggested by Stock and Watson (2005). In this context, extract-
ing common factors reflecting the co-movements of these indicators has proved to be a
convenient way of summarizing the information contained in such large data sets. These
indicators are very often available at higher frequencies than the targeted variable (GDP).
This aggregation issue is quite successfully dealt with by MIxed DAta Sampling (MIDAS)
models introduced by Ghysels, Santa-Clara and Valkanov (2004) and Ghysels, Sinko and
Valkanov (2007). This paper is at the crossroad of these three strands of the literature.
*This workwas initiated when the two authors were in Banque de France.We wouldlike to thank Anindya Banerjee,
Gorkem Celik, Laurent Ferrara, Christian Francq,Ana Beatrix Galv ˜ao, Sophie Haincourt, Sheheryar Malik and two
anonymous referees for their comments and suggestions, as well as Michel Juillard for his help with computational
issues. The paper also benefited from discussions with the participants of the European Meeting of the Econometric
Society in Gothenburg, the CFE conference in Oviedo, the EACBNConference on Disagg regatingthe business cycle
in Luxembourg and the SNDE conference in Istanbul. Finally, we would like to thank Marie-Pierre Houri´e-Felske
for excellent research assistance. The views expressed in this paper are those of the authors and do not necessarily
reflect those of the European Central Bank.
JEL Classification numbers: C22, E32, E37.
Forecasting GDP with a MS-FaMIDAS model 361
MIDAS models are regressions that involvevariables sampled at different frequencies.
In this framework,a low frequency variable can be explained byhigher frequency indicators
without any time aggregation procedure.A distributed lagged function can be used to obtain
a parsimonious specification of the relationship between the dependent variable and the
higher frequency variables. While MIDAS models were first applied to financial data,1
they have also become a popular tool for forecasting macroeconomic variables such as
GDP growth. Forecasters use specifications that link the GDP variable to a handful of
monthly leading indicators or they rely on combinations of MIDAS models to deal with
the potentially large number of indicators.2See Andreou, Ghysels and Kourtellos (2011)
for a survey of this literature.
Two recent extensions are particularly designed to forecast macroeconomic variables:
factor MIDAS (FaMIDAS) models by Marcellino and Schumacher (2010) and Markov-
Switching MIDAS models by Gu´erin and Marcellino (2013). In addition to involving
mixed frequency data, the first class of models allows the use of information provided
by a large data set and can handle unbalanced samples that practitioners usually have to
work with because of different publication lags. The second class of models incorporates
regime changes in the parameters of the relationship between the low and high frequency
variables. Moreover, it gives qualitative information about the state of the economy. This
provides a useful tool for business cycle analysis.
In this paper, we introduce the MS-Factor MIDAS model (MS-FaMIDAS), which cap-
tures both co-movements and regime shifts in the dynamics of the variables and which
can be implemented with mixed frequency data. We consider the dynamic factor model of
Giannone, Reichlin and Small (2008), estimated with the two-stepmethod of Doz, Giannone
and Reichlin(2011). This approach can deal with the unbalanced data availability at the
end of the sample caused by uneven publication lags. It should be noted that we allow a
variation in the coefficients of the equation of the dependent variable (the coefficients of
the GDP equation in our application), but not on the factor dynamics.
The MS-FaMIDAS model helps improving the short-run analysis of business cycle
fluctuations. It provides both quantitative information (GDP growth rate) and qualitative
information (state of the economy). The MIDASspecification allows incorporating within-
quarter information to update directly the GDP forecast and the probability of recession
several times during that very quarter. Extracting factor from a larger set of indicators
provides less noisy indicators and might improve the inference about the business cycle.
Moreover, this approach overcomes the problem of missing data, as it can be implemented
even when some observations are missing at the end of the sample because of publication
lags. Finally,it can also deal with the ir regular pattern of missing observations (the so-called
ragged edge problem) caused by the different time releases of the indicators.
An obvious limitation of the model is its omission of potential switches in the factor
representation. It could be relevantto incor porate regime switchesin the factor dynamics, as
1See Ghysels, Santa-Clara andValkanov (2005, 2006) and Ghysels et al. (2007) for applications to equity returns,
Clements, Galv˜ao and Kim (2008) to exchange rates.
2For instance, see Clements and Galv˜ao (2008, 2009), Bai, Ghysels and Wright (2013), Armesto et al. (2009),
Armesto, Engemann and Owyang (2010), Andreou, Ghysels and Kourtellos (2013) for US GDP, Kuzin,Marcellino
and Schumacher (2011) for euro area GDP, Foroni, Marcellino and Schumacher (2013) for both US and euro area
GDP.
©2014 The Department of Economics, University of Oxford and JohnWiley & Sons Ltd
To continue reading
Request your trial