Tools and Practices for Business Cycle Analysis in EU by xfz11675


									               Tools and Practices for
            Business Cycle Analysis in EU
            BUSY project FP5-IST-12654
        Gabriele FIORENTINI, Christophe PLANAS and Gilles TEYSSIERE
                          Institute for Systems, Informatics and Safety
                              Joint Research Centre of EC, TP361
                                          Via E.Fermi, 1
                                     I-21020 Ispra (VA) Italy

         Abstract: BUSY is a shared-cost action funded within the Information
         Society line of 5th Framework Program. The final objective is to implement
         a software tool able to guide official statisticians through the main steps of a
         standard business cycle analysis in an organized and informative way, so as
         to help in improving the knowledge of cycles in EU. Emphasis is put on
         statistical properties, on methods reliability, and on how suited is every
         method to the case of European economies and to the data typically
         available in Europe. The investigations balance theoretical and practical
         considerations, fundamental analysis and empirical experience. The project-
         term is 31 December 02.

         Keywords: Business cycle analysis, NBER, dynamic factor models, turning
         points analysis

1.        Introduction
BUSY is a shared-cost action funded within the Information Society line of 5th
Framework Program. The final objective is to implement a software tool able to guide
official statisticians through the main steps of a standard business cycle analysis in an
organized and informative way, so as to help in improving the knowledge of cycles in
EU. Emphasis is put on statistical properties, on methods reliability, and on how suited
is every method to the case of European economies and to the data typically available
in NSI’s. The investigations balance theoretical and practical considerations,
fundamental analysis and empirical experience. The project is led by the Joint Research
Centre of the European Commission, in partnership with CISI (Milan), INE (Madrid),
INSEE-CREST (Paris), ISTAT (Roma), GREQAM (Marseilles), and it is scheduled to
last three years starting in January 2000. Besides the kick-off meeting, the consortium
already met twice, in Paris in June 00 and in Roma in December 00. A third consortium
meeting will be held in Madrid in May 01. BUSY benefits from the comments and

                      Tools and Practices for Business Cycle Analysis in EU
                                 BUSY project FP5-IST-12654

suggestions of the external advisors Filippo Altissimo (Bank of Italy, Rome), Marco
Lippi (University La Sapienza, Rome) and Bjorn Fischer (European Central Bank,
Frankfurt). Their unquestionable technical knowledge and their empirical experience
are very much helpful to BUSY and the consortium gratefully acknowledges their
commitment. The project is regularly reviewed by Eurostat and by the chef supervisor,
G.Keogh (Central Statistical Office, Dublin). The consortium expresses its gratitude to
G.Keogh for his very constructive comments and suggestions. Special thanks are also
due to Gianluca Caporello for the outstanding quality of his commitment in software

Much has been done in this first half of the project. The first four work-packages, WP1-
Dating by INE, WP2-Composite Indexes ISTAT, WP3-Early prediction and forecasting
by JRC, WP4-Practices by INSEE-CNRS-DR01 are now completed. On that basis, it
has been possible to draw the general profile of the BUSY software, bearing in mind
that the overall objective is to give access to a large-scale business cycle analysis.
BUSY consortium made the choice to balance traditional tools, i.e. the NBER-
approach, with modern techniques related to dynamic factor models; these will be the
two main pillars of BUSY. The first type of approach involves descriptive statistics and
tools like for example cross-correlations, cross-spectra, phase analysis, Bry and
Boschan turning point analysis (see Bry and Boschan, 1971). Second, thanks to recent
developments in the generalised dynamic factor model by Forni, Hallin, Lippi, Reichlin
(1999,2000) and by Stock and Watson (1999), it is now possible to analyse common
cycles in hundreds or thousands of series without being faced with problems related to
the number of model parameters. For example, Stock and Watson (1999) considered
215 macroeconomic series in their analysis. Forni, Hallin, Lippi, Reichlin (1999, 2000)
also showed how information about leading and lagging behaviour of economic series
result from the dynamic principal component analysis. This is by far the most
promising approach available in modern economic and statistical literature.
Furthermore there are reasons to argue that this approach is particularly suited to the
EURO problem, due to the short length of the available series.

The two approaches complement themselves in a natural way. Also, the choice of
implementing the generalised dynamic factor model will make the final product
particularly attractive, as it is true that these recently developed techniques are not of
easy access to economists. Experiences conducted at Bank of Italy (see Altissimo,
2000) proved that dynamic principal component analysis is very powerful for
multivariate studies and deliver relevant information. As the partner INE, the Spanish
national statistical institute, also used of factor models, we can thus be confident that
the project will not turn into an academic exercise.

The software profile that been proposed by the JRC on 30-09-00 is currently under
development. The targeted dates are 30-09-01 for Alpha version and 31-12-01 for Beta

There is an urgent need to characterise the European Business Cycle in the EMU phase.
If the importance of the matter will help in broadening the dissemination of the project
achievements, a strong dissemination strategy is nevertheless required. Two important
pieces are first the maintenance of the cooperation established between the BUSY

                   Gabriele Fiorentini, Christophe Planas and Gilles Teyssiere

consortium, the JRC and DG ESTAT (especially the unit in charge of short-term
analysis) and second the successful set up of a Business Cycle interest group that
includes 45 experts from all over Europe.

Of course much remains to do. During the second project-half software programming,
debugging and validating will require many efforts and concentration. Also important
will be the preparative work for the organisation of a major conference on the
convergence of the EU economies in 2002, including the Business Cycle interest group

2.        The project at mid-term
Mid-term objectives

The objectives in the first year of the project were:
· to design the overall structure and procedures implemented in the BUSY software,
· to start the software development,
· to develop the dissemination and use plan (compulsory), and finally
· to set up an interest group

 Four preliminary investigations were necessary for reaching the first two objectives:
· to review and evaluate techniques for dating turning points (INE)
· to review and evaluate techniques for composing turning points (ISTAT)
· to review and evaluate techniques for prediction and early prediction of turning
    points (JRC)
· to review practices in official statistics (INSEE-CNRS),

 As originally programmed, all four tasks have been completed and the output has been
used to design an overall software profile. The software development has now started,
and an Alpha version is foreseen in September 01.

2.1 Reviewing and evaluating statistical methods for dating, by INE (see Abad et
    al., 2000)

Fluctuations, turning points and cyclical classification are three key concepts for the
short-term economic analyst. The fluctuations or business cycles are the raw material of
the short-term economic analyst, so that he devotes a large part of his time to its
estimation, assessment and prediction. Turning points are particularly relevant events in
the usual development of cycles, so that their presence marks phase changes in the
economic trend that must be examined carefully and analysed in their implications.
Furthermore, the cyclical classification is a task that substantially extends the
knowledge of the analyst about the economic system, so that he can explain with more
rigour the behaviour of the variables and prepare more accurate predictions.

The structure of the study is as follows. Two methods to identify empirically the
turning points of a time-series are analysed, the <F> procedure developed at INE and
the Bry and Boschan (1971) routine. Both share the use of linear filters and the coding

                      Tools and Practices for Business Cycle Analysis in EU
                                 BUSY project FP5-IST-12654

by programs of the decision rules that the cycle analysts have been using to detect
turning points. Then, two model-based procedures are reviewed, namely TAR models
and Markov switching models. These models, which allow for an express definition of
the concept of turning point, conceive the cyclical event as an element inseparable from
the mechanism of propagation of the shocks.

The study then focuses on bivariate cyclical classification. One of the most common
uses of cyclical chronologies consists of the detection of dynamic relations between
indicators, so that some can be used as leading indices allowing for the early
identification of phase changes in the business cycle. This dynamic identification task
can be also carried out by the analysis of cross correlation functions and spectral
coherence, which are also shown. Finally, a multivariate method of cyclical
classification based on a dynamic factor model is discussed. This model allows for
identifying common evolution patterns with subsequent classification.

The study concludes assessing the different methods and providing a structured guide
of analysis enabling the common application of these techniques in the analysis of the
cycle and the development of useful computer tools for this analysis. Complementary
results and algorithms about detrending and the estimation of the cyclical component
has been made available by the Institute of Systems, Informatics and Safety, Joint
Research Centre of European Commission. In particular the appendix reports C
algorithms for Baxter and King filtering and Hodrick-Prescott detrending procedures
(see Baxter and King, 1997; Hodrick and Prescott, 1998). Interested readers can also
consult Gómez (1999), Kaiser and Maravall (2000), amongst others.

2.2 Reviewing and evaluating statistical methods for composing indexes, by
    ISTAT. (see Polidoro, 2000 and Polidoro and Bacchini 2000)

The aim of this study is the description and analysis of the different statistical methods
for building composite indexes. The research activity has gone in three directions:
Review of the NBER methodology. This part includes a brief history of the NBER and
the illustration of the fundamental characteristic of the NBER approach. It is also
presented an application of model based approach by Stock and Watson (1990) to the
Italian economy. This part also includes an the extension of the Stock and Watson
model to the Markov switching case (see Otranto 2000)
Dynamic factor model review. This part includes a review of the traditional factor
model with the first application to the time series domain. A detailed overview of the
recent contribution by Stock and Watson (1999) and Forni, Hallin, Lippi and Reichlin
(FHLR) (1999) is developed.
Description of the tools for evaluation and comparison of different methods. The
differences in the statistical methodology mainly concern the common cycle estimation
strategy, and can be related to different information set used in the estimation.
Moreover, it stresses the characteristics of “unified approach for the selection of the
coincident and the leading variables” instead of the traditional NBER empirical

                    Gabriele Fiorentini, Christophe Planas and Gilles Teyssiere

2.3 Early detection and forecasting of turning points, by JRC (see Fiorentini et al.,

The aim of this study is to review the methods available for the early detection and
prediction of turning points in the business cycle and, considering the specificity of
European statistics, to propose an approach that can be made available to economists
and statisticians through the software BUSY.

The JRC reviewed first the statistical techniques for early detection and prediction of
business cycle turning points that can be found in the international literature. Both
univariate and multivariate approaches are described. The univariate tools considered
are namely Bayesian techniques with the Wecker’s approach (see Wecker 1979) and its
Bayesian extensions (see Zellner et al., 1991), and switching regimes models with the
sequential probability model and Markov process. Multivariate techniques were also
illustrated through an application to the French economy. Some specificity of the
European statistics is also discussed; in particular, special consideration is paid to the
fact that while statistics are reported for a mass of economic variables, the amount of
past data available is typically short. Thanks to progresses in information technologies,
large databases about official statistics are now easily accessible. Meanwhile, it is still a
difficult task to consider a large number of variables in multivariate analysis with
standard statistical models. The statistical treatment of large-scale information requires
thus some information reduction-scheme. This is essentially the purpose of dynamic
factor models as studied by Stock and Watson (1999) and Forni et al. (1999, 2000).
Special attention is hence paid to these methods.

Given that the aim of BUSY is to make available methods for business cycle analysis in
Europe, given the context of European Statistics and given the state of the statistical
knowledge, this study supports in first place the implementation of information
reduction scheme of the dynamic factor model type. These techniques are recent and
offer statistical answers to problems like leads and lags relationships with respect to a
targeted variable.

2.4 Methods for official statistics – the practitioner side by INSEE (see Cudeville
    and Gregoir, 2000)

After a detailed presentation of the traditional NBER approach in compiling coincident
and leading indicators, the authors introduce more recent developments in these fields.
They also describe the current practice of the Italian, Spanish and French institutions.
The three approaches differ by the kind of data they use and the statistical treatment
they apply (or plan to apply) to these data, but basically they implicitly focus on the
same dynamic one-factor model. A detailed presentation of the qualitative Hidden
Markov Models used in the compilation of short-term indicator and their
implementation in France is then developed. The model has been estimated on balances
of opinions from French business surveys and allowed French statisticians to compute
on a regular basis a coincident indicator. Gregoir et al. discuss the validation procedure
that has been used by INSEE before releasing such a short-term statistic. It is
emphasized that short-term indicators are fragile statistical objects, whose quality must
be assessed in ``real time'' compiling conditions before any decision for a regular

                      Tools and Practices for Business Cycle Analysis in EU
                                 BUSY project FP5-IST-12654

release can be taken. Furthermore, its quality must be continuously monitored, where
quality here is to be understood in terms of accuracy of the released signal and its
timeliness. In this spirit, the study concludes with a list of recommendations for the
BUSY that put a strong emphasis on validation techniques.

2.5 General software profile (see Planas et al., 2000)

 On the basis of the recommendations in previous four studies, a general design for the
software BUSY has been developed. Two main methodologies for business cycle
analysis will be made available: an empirical approach of the NBER-type and dynamic
factor models. Accessory modules will include data set manager, data pre-filtering,
forecasting models, validating tools based on real-time analysis, and a special module
for business survey data. The profile described is general and it cannot be excluded that
complementary features will be added in the future. The ALPHA version is planned for
end September 00.

3.        Dissemination activities
BUSY's dissemination is based on four main lines of action.

1) The first dissemination action is the set up of an interest group. Since the first
months of the project, a community made up of 45 potential users is been constituted,
on the basis of a precise identification of institutions and relevant departments.

Two-way communication within this large community of statisticians will be animated
and fostered with the support of:
    · A Web-space, located within the site "Time Series Analysis for Official
         Statisticians" located at http//
    · Regular meetings: the consortium will organise the first meeting of that group
         in 2002. Advertisements will be made in the statistical press, including the
         journal Research in Official Statistics.

2) The second action is a direct support to the dissemination of the software BUSY. It
will be based on two action-types: a Web-supported Help-Desk within the site "Time
Series Analysis for Official Statisticians", and implementation in the premises of a list
of test-users. These test-users will be mobilised as soon as BUSY will have been found
satisfactory by the NSI's which compose the consortium.

3) The third line of action concerns support in post-project phase. The following
operations are foreseen: maintenance, animation of interest group, training, and

4) The fourth line of dissemination is a standard one for projects financed under FP5. It
includes presentation to conferences and special events, such as NTTS-ETK 2001,
articles and reports in the specialised literature, and use of web technologies. The
BUSY project has been advertised to delegates of European National Statistical
Institutes and Central Banks at the occasion of the Informal Meeting of the Seasonal
Adjustment Working Group hold in Luxembourg, Eurostat, the 7th of February. It has

                   Gabriele Fiorentini, Christophe Planas and Gilles Teyssiere

also been advertised to officers from the European Bank of Investment, European
Central Bank, EC-DG ECFIN, EC-DG REGIO, EC-DG EUROSTAT, National Bank
of Belgium and the National Institute for Economic and Social Research during a
meeting organised by DG-CCR in Brussels, on 13 June 00.

4.        Conclusion and next steps
The main objective of next second half of the project is to make available a Beta
version for 30-06-02 and to actively support its dissemination in the next semester. The
main intermediate objectives are:
     · Alpha version for 31-06-01
     · Test-reports on BUSY-Alpha by every NSI-partner
     · BUSY interest group meeting during 02

Still in support of the product dissemination, a business cycle course maybe set up.
Some targeted studies will also need to be produced within the year in order to build
more applications and extension of the dynamic factor model. Following a suggestion
of G.Keogh, empirical comparisons between NBER-methods and the use of dynamic
factor models will be developed.

The contribution of the project to business cycle analysis in practice is unquestionable:
such analysis requires to perform a lot of operations on large dimension multivariate
data set, and as far as we know no statistical software giving an easy access to standard
and to best practices do exist. Furthermore, some tools typically involved can be
complex enough. Our final aim is to provide economists with a simple to use product
for understanding either with simple tools or with some modern statistical techniques
the general path of a national economy or of a set of economies.

BUSY will help economists to assess the state of EU national economies and of the
Euro-area. Based on a better understanding of the behaviour of the economies, policy-
makers will be in a better position to make informed choices for short-term macro-
economic policy making. For that reason, access to business cycle analysis tools is
most expected by important actors on the EU scene like for example the European
Commission, with its general directorates EUROSTAT and ECFIN, the EU central
bank ECB. National actors like NSI, central banks of the ESCB. Member States
government economic services, and other institutes like the OECD, as well as public
and private economic research institutes, are also potential customers of BUSY.

More specifically BUSY merges advanced statistical techniques with the practices in
economic, monetary and statistical institutions. It strengthens the link between
advanced academic research and the daily practice of statistics, improving the
accessibility of statistical tools.

                             Tools and Practices for Business Cycle Analysis in EU
                                        BUSY project FP5-IST-12654


[1]    Abad A.M., Cristóbal, A., Quilis E. (2000), “Economic fluctuations, turning points, and cyclical
       classification”, report for IST-12654 project BUSY, Instituto Nacional de Estadistica, Madrid.
[2]    Altissimo , F. (2000), `Coincident and leading indicators of the business cycle: a dynamic principal
       component approach”, Bank of Italy, mimeo.
[3]    Bacchini, F. and Polidoro F. (2000), “Business cycle in practice: the case of Italy”, report for IST-12654
       project BUSY, Instituto Nazionale di Statistica, Rome.
[4]    Baxter M. and R.G. King (1999) “Measuring Business Cycles: Approximate Band-Pass Filters for
       Economic Time Series”, Review of Economic and Statistics, 81, 4, 575-593, (also NBER WP 5022,
[5]    Bry G. and C. Boschan (1971), “Cyclical Analysis of Time Series: Selected Procedures and Computer
       Programs”, NBER Technical Paper 20.
[6]    Cudeville, E. and Gregoir S. (2000), “Economic composite indicators: methodology of compilation and
       validation”, report for IST-12654 project BUSY, Institut National de la Statistique Economqieu et des
       Entreprises, Paris.
[7]    Fiorentini, G. and Planas, C. (2000), “Early detection and forecasting of turning points”, report for IST-
       12654 project BUSY, Joint Research Centre of EC, Ispra.
[8]    Forni M., M. Hallin, Lippi M. and L. Reichlin (2000), “The Generalized Dynamic Factor Model:
       Identification and Estimation”, Review of Economics and Statistics, 82, 4, 540-554.
[9]    Forni M., M. Hallin, Lippi M. and L. Reichlin (1999), “Reference Cycles: the NBER Methodology
       Revisited”, ULB, Brussels, mimeo.
[10]   Gomez, V. (1999), “The use of Butterworth filter for trend and cycle estimation in economic time
       series”, mimeo.
[11]   Baxter M. and R.G. King (1999) “Measuring Business Cycles: Approximate Band-Pass Filters for
       Economic Time Series”, Review of Economic and Statistics, 81, 4, 575-593, (also NBER WP 5022,
[12]   Hodrick R. and Prescott E. (1997), “Post-war business cycles: an empirical investigation”, Journal of
       Money, Credit and Banking, 29, 1, 1-16.
[13]   Kaiser, R. and Maravall, A. (2000), “Long-term and short-term trends: the Hodrick-Prescott filter
       revisited”, mimeo, Bank of Spain.
[14]   Otranto, E. (2000) “The Stock and Watson model with Markov switching dynamics”, report for IST-
       12654 project BUSY, Instituto Nazionale di Statistica, Rome.
[15]   Planas, C., Fiorentini, G. and Teyssiere G. (2000), “General software design fore business cycle
       analysis”, report for IST-12654 project BUSY, Joint Research Centre of EC, Ispra.
[16]   Polidoro F. (2000), “Description and analysis of empirical and model-based procedures for building
       composite indexes”, report for IST-12654 project BUSY, Instituto Nazionale di Statistica, Rome.
[17]   Stock J.H. and M.W. Watson (1999), “Diffusion indexes”, NBER Working Papers 6702 (Revised).
[18]   Stock J. H. and Watson M.W. (1990), "Business cycle properties of selected U.S. economic time series,
       1959-1988", NBER Working paper 3376/A.
[19]   Wecker W.E. (1979) “Predicting the Turning Points of a Time Series”, Journal of Business 32, 1, 35-50.
[20]   Zellner A., Hong C. and Min C-K (1991), “Forecasting Turning Points in International Output Growth
       Rates Using Bayesian Exponentially Weighted Autoregression, Time-varying Parameter and Pooling
       Techniques”, Journal of Econometrics, 49, 275-304.


To top