An integrated system for euro area and member states turning points detection By: Gian Luigi Mazzi, Monica Billio and Laurent Ferrara Brussels, 11 March 2015 Eurostat – Unit C1: National accounts methodology. Sector accounts. Financial indicators.
Outline • Introduction • Choice of the reference cycle • Methodological aspects • Constructing composite indicators for turning points detection • Empirical results • Outcome summary: comments • Comments on latest results of coincident indicators • Conclusion and future activities
Introduction (I) • Accurate monitoring cyclical fluctuations is a key component for short term macroeconomic analysis and forecasts • Cyclical information is a crucial requirement for policy makers, macroeconomic analysts, forecasters and analysts • Timely and reliable turning point detection is the core of modern business cycle analysis • Since turning points are quite rare events and typically appearing as discontinuities or breaks in the series, they are not easily estimated • Non-linear modelling as the most promising approach for turning points • Binary regressions • Univariate and multivariate non-linear time-series models
Introduction (II) • In-depth theoretical and empirical investigation of alternative non-linear modelling approaches • Focusing on non-linear time-series models • Higher flexibility in comparison with other non-linear techniques • Markov-switching models considered the most reliable tool for turning points detection • Outcome of an empirical comparison, especially with SETAR models (Billio, Ferrara, Guegan, Mazzi, (2014)) • First attempt to compile euro area turning point indicators started in 2006 • Ferrara, Mazzi (2008)
Choice of the reference cycle (I) • Classical Business cycle (Burns and Mitchell definition) • Very relevant for detecting recessions • Not very informative during usually quite long expansion phases • Growth cycle (Output gap) • Very relevant to understand the position with respect to the potential output • More informative also during the expansion phases of business cycle • Anticipating business cycle peaks • Unable to detect the start and the end of recessions
Choice of the reference cycle (II) • Growth rate cycle (Acceleration cycle ) • Highest number of fluctuations • High degree of volatility • Anticipating growth cycle peaks and business cycle troughs • Jointly monitoring several reference cycles (Anas, Ferrara 2004) • Growth cycle and Business cycle (ABCD sequence) • Also including Acceleration cycle ( α AB β CD sequence) • Approach retained by Eurostat
Methodological aspects (I) • Multivariate Markov-Switching models to jointly estimate a pair of probabilistic coincident indicators of the classical business cycle and growth cycle • Euro Area as a whole (direct indicator) • Largest Member Countries • Unfeasible jointly modelling also of acceleration cycle for mathematical reasons • Each pair satisfies, by construction, the ABCD approach • Comparison of a huge number of alternative coincident indicators obtained by combining a variety of • model specifications • number of regimes • endogenous variables • rules to associate regimes and economic cycles
Methodological aspects (II) • Models only based on empirical results without any economic hypothesis or restriction • The accuracy of each pair of coincident indicators in locating economic fluctuations measured with respect to a benchmark • namely the historical turning points dating chronology developed by Eurostat • For the direct EA indicators and for each Member Country, a pair of coincident indicators was identified • one expressing the probabilities of a recession of the classical cycle • one expressing the probabilities of a slowdown of the growth cycle • Choice of each pair of indicators based on the Brier's score (QPS) and the Concordance index
Methodological aspects (III) • In addition, a pair of coincident indicators for the euro area’s classical and growth cycles obtained as a weighted average of Member States ones was developed • The considered Member Countries account for almost 90% of the Euro area GDP • The indirect coincident indicators can shed light on the economic fluctuations across the euro area • They can be the basis for computing in real-time the degree of diffusion and synchronisation of a given economic cyclical phase among member states
Constructing composite indicators for turning points detection (I) • Step 1: middle-size dataset mainly based on PEEIs and opinion surveys data • Performing most appropriate data transformation to highlight cyclical movements • Step 2: variable selection based on the ability of timely and precisely detecting turning points within a real-time simulation exercise against the non-parametric historical turning point dating
Constructing composite indicators for turning points detection (II) • Step 3: selected variables are used to identify and estimate a number of autoregressive Markov-Switching models (MS- VAR) MSIH (K) – VAR (L) Where H indicates the presence of heteroskedasticity, (K) is the number of regimes and (L) the number of lags of the autoregressive part • Step 3 remark : dealing simultaneously with growth cycle and business cycle implies a number of regimes not smaller than 4 , while the heteroskedastic part can or cannot be present depending on the degree of asymmetry of fluctuations
Constructing composite indicators for turning points detection (III) • Step 4: from step 3, N best fitting models are identified, each of them producing a pair of coincident indicators: MS-VAR GCCI (j) and MS-VAR BCCI (j) j=1 …n indicates the j th • Step 4 remark 1: each composite indicator is defined between 0 and 1, and can be viewed as a composite probability of being in a recessionary phase for the MS-VAR BCCI (j) and in a slowdown phase for the MS-VAR GCCI (j) • Recession/slowdown regions defined on the basis of a threshold, usually equal to 0,5 • Step 4 remark 2: MS-VAR BCCI (j) > 0.5 = recession MS-VAR GCCI (j) > 0.5 = slowdown By construction, MS-VAR BCCI (j) > 0.5 MS-VAR GCCI (j) > 0.5 • ABCD sequence always fulfilled
Constructing composite indicators for turning points detection (IV) • Step 5: within a real-time simulation exercise, the N pair of composite coincident indicators is compared with the non- parametric historical turning point dating
Constructing composite indicators for turning points detection (V) • Step 6: identification of the best performing pair of coincident indicators based on the outcome of step 5 using the following criteria: • Maximisation of the Concordance Index • Minimisation of the Brier's Score (QPS) • Minimisation of type-2 errors: detection of false cycles • Minimisation of type-1 errors: missing cycles • Step 6 remark: due to the trade-off between type-2 and type-1 errors, the simultaneous minimisation of both is unachievable • A conservative approach suggest to privilege the minimisation of type-2 errors: detection of false cycles
Empirical results Model Summary for the EA direct Variables (differentiation order) Country Model Recessions Slowdowns IPI UR NPCR INDEA7 4+ EA MSIH(4)-VAR(0) R1 R1+R2 6 1 1 MA(6)
Euro Area ACCI (direct) 1 Probability of Being in a Deceleration of the Acceleration Cycle 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Acceleration Cycle Reference Chronology Provisional Dating Chronology Ending Date of Provisional Chronology ACCI 0.5 Threshold
Euro Area MS-VAR GCCI & BCCI (direct)
Models Summary for the major MS Variables (differentiation order) Country Model Recessions Slowdowns IPI UR BUIL IND CONS RETA Belgium MSI(4)-VAR(0) R1 R1+R2 6 3 6 3 - 3 France MSIH(4)-VAR(0) R1 R1+R2 6 1 3 - 1 12 Germany MSIH(4)-VAR(0) R1 R1+R2 3 3 3 - 6 12 Italy MSIH(5)-VAR(0) R1 R1+R2 3 3 - 12 12 3 Netherlands MSIH(4)-VAR(0) R1 R1+R2 12 - 6 3 1 1 Portugal MSI(5)-VAR(0) R1+R2 R1+R2+R3 6 - 3 3 12 1 Spain MSIH(4)-VAR(0) R1 R1+R2 12 12 3 6 12 -
Growth Cycle Outcome Summary Average Accuracy in signalling slowdown delay in Slowdown False locating Country missed slowdown Slowdowns Brier’s Score (QPS) Concordance Index start (in months) Belgium 0 1 (2005) 0.7 0.16 0.83 France 1 (1998) 0 3.2 0.16 0.82 Germany 1 (1998) 0 2.5 0.20 0.79 Italy 0 0 4.3 0.22 0.77 1 Netherlands 0 2.0 0.18 0.80 (1995 – 1997) Portugal 0 3 0.6 0.18 0.80 1 Spain 0 4.3 0.27 0.72 (1997-1998) 1 EA direct 0 2.0 0.15 0.83 (2004-2005) 1 EA indirect 0 3.0 0.10 0.87 (1998)
Business Cycle Outcome Summary Average Accuracy in signalling recessions delay in Recessions False Country locating missed recessions peaks Brier’s Score (QPS) Concordance Index (in months) 2 Belgium 0 6 0.15 0.84 (1998 - 2000) France 1 (2012) 0 2.5 0.06 0.94 1 Germany 0 3.4 0.08 0.92 (2001 - 2002) 1 Italy 0 2.8 0.12 0.87 (2001) Netherlands 0 0 3.5 0.12 0.88 Portugal 0 0 4.3 0.17 0.82 Spain 0 0 1.3 0.05 0.94 EA direct 0 0 2.3 0.06 0.94 1 EA indirect 0 2.5 0.06 0.90 (2011-2013)
Recommend
More recommend
