Euro-indicators Working Group

Transcription

1 Euro-indicators Working Group Luxembourg, 9 th & 10 th June 2011

2 Item 9.3 of the Agenda Towards an early warning system for the Euro area By Gian Luigi Mazzi Doc 308/11

3 Introduction Clear picture of economic movements - timely available statistics and statistical indicators Relevant improvements of PEEIs since timeliness - coverage Timeliness gap between Euro Area and USA PEEIs improvements: - increasing timeliness without lost of accuracy - long time series - high frequency data - extracting signals by means of specific indicators

4 Limits of PEEIs further improvements of timeliness are required without a significant decrease of accuracy (more flash estimates and nowcasting); the time coverage of series is not always appropriate for economic modelling and business cycle analysis an effective monitoring of the economic situation requires at least a monthly follow up so that relevant indicators have to be available at least at monthly frequency; (construction of new monthly indicators); compiling composite indicators to extract signals and to fill the specific gaps in data availability of official statistics (cyclical estimates, turning points dating and detection, coincident and leading indicators). Ongoing studies aim to produce a more significant monthly indicator for business cycle purposes likely based on GDP

5 OUTLINE Back-calculation Real time database Increasing data timeliness: nowcasting techniques Increasing data timeliness: coincident indicators EuroMIND: euro area Monthly indicator of the economic activity Euro area turning point detection Growth cycle estimates

6 Back-calculation scope The exercise does not imply a micro-level back-calculation Mainly targeted to main aggregates with a limited breakdown. Its aim is not to change the past pattern of the series but to keep it unchanged (turning points, cyclical shape) Preserve the historical characteristics of the series We don t want to "rewrite" the history This exercise - homogenisation of existing, partially overlapping segments of time series eliminating breaks and inconsistencies. This is in line with users' expectations involved in BC analysis and econometric modelling

7 Back-calculation scope Simple and robust method based on linear regression models Hypothesis -> overlap between the target series and a proxy A similar approach is also adopted for annual national accounts. Overlapping sub-periods -> all the information at national level Temporal and sectoral, linear and non linear aggregation constraints are taken into account SA data -> consistency with the approach currently used in Eurostat

8 Experimental back-calculated series Indicator name Breakdown Starting date Industrial Production NACE divisions and MIGs 1970 Producers Prices NACE divisions and MIGs 1970 Turnover index NACE divisions and MIGs 1974 Retail Trade Food, Non food 1970 Unemployment Male, Female, under and over Employees, self-employees NACE Employment A Building permits Total 1970 Nights spent in Hotels Resident/Not resident 1990 Wages and salaries NACE A National Accounts main aggregates NACE A6 1970

9 Back-calculated monthly EA12 IPI comparison with main related indicators 125 Back calculated monthly index Growth rates in % of annualized indexes 10 0 EA12 Germany old France old Italy old Growth rates in % of annualized indexes EA12 EA11 old

10 The PEEIs real time database Daily snapshots from 16th November 2000 onward No delay in indicators updating All revisions are captured Presently vintages from 16th November level indicators (extended version of Shorties) covering 8 main categories: Balance of payments, Business and consumer surveys, Consumer prices, External trade, Industry, commerce and services, Labour market, Monetary and Financial Indicators and National Accounts.

11 Data and metadata requirements for building a real-time database Country / region description; Variable description (and its evolution over time); Variable measures; Identification of vintages; Length of vintages; Length of time series (within vintages); Ongoing updating; Data access

12 Usefulness of real Time Database Large availability of daily snapshots over the time allows Eurostat to reconstruct a real-time database covering a quite long time period which is particularly useful to: study the revisions behaviour of macro-economic statistics, perform simulation exercises concerning new estimates and/or indicators, monitor the evolution of the database quality across time, perform real-time simulations of all nowcasts and construction of new indicators

13 Increasing data timeliness Speeding up data production process - advanced survey techniques - simplification of questionnaires Now-casting by means of statistical and econometric techniques - incomplete information set EU sampling techniques - not significant national samples Construction of coincident indicators - anticipating latest economic trends

14 Use of statistical and econometric techniques Using forecasting techniques to estimates the recent past and the present Not all forecasting models are allowed in official statistics Key principles of the construction of Flash Estimates: - use of all partial information whenever available - use of soft data only combined with a minimum of hard data - use of related indicators only in case of unavailability of relevant partial information - avoiding economic hypotheses in the model specifications - discarding univariate model specifications - selecting simple methodologically sound models

15 Nowcasting strategies for PPI and GDP regression based methods to produce nowcasts national and Euro indicator variables in estimated models BIC criterion used recursively to select the best single model real time simulation of the flash estimates for the GDP against the Eurostat first and final estimates for the euro area preferred model uses two months of within quarter IP data and the latest quarter's value of the Economic Sentiment Indicator real time simulation of the flash estimates for the Producer Price Index against the Eurostat first and final estimates for the euro area. The selected model is based on German industrial output price and energy prices data according to the BIC selection criterion often the flash are quite close to Eurostat first estimates

16 Euro Area Nowcast of GDP Growth Flash Eurostat Eurostat Error Error t+15 days First estimate Final estimate First estimate Final estimate 2008q q q q q q q q

17 Euro Area Nowcast of Producer Price Index Flash t+16 days Eurostat Eurostat Error Error First Final First estimate Final estimate estimate estimate 2008m m m m m m m m m m m m m m m m m m m m m m m m m m m m m m m m

18 Density nowcasts to provide a complete description of the uncertainty associated with the point nowcast implies advantages when communicating with the public: - analys themselves expect the point nowcast to be wrong - users assess the balance of risks associated with the nowcasts Key principles to construct density nowcasts: statistical models seek to explain and then nowcast GPD growth by exploiting information on indicator variables distinguish between quantitative ( hard ) and qualitative ( soft ) indicator variables focus on simple components models estimate a linear regression of quarterly GDP growth on a single indicator variable combin the component density nowcasts using the linear pool approach estimate three component models for each indicator

19 Recursive Weights on the soft indicators, hard indicators and lagged values of GDP growth

20 Results more hard data available, higher weight in the combined density, during the recession, weights of soft data increases at t-30, t-15 and t+0, close to unity during the depth of the recession, AR hard to beat when point nowcasting There is always an issue about how to choose the length of the training period to calibrate the weights The shorter the length of the training period the more quickly the combined density can adjust to changes over time in the performance of the different models the longer the length of the training period the better the combination weights are estimated

21 Probability of recession

22 Probability of recession

23 Results RW combined densities pick up earlier the recession than hard IP data, on average over the evaluation period, nowcast density produced by survey data alone is not well calibrated, Waiting for second month s IP data, RW combined densities anticipate both the beginning and the end of recession earlier than EW combined density

24 Coincident indicators Forecasting the target variables during the reference period or right after its end - similar philosophy of Leading Indicators Fewer constraints in model specifications than Flash estimates - still avoiding economic hypotheses in model specifications Ongoing Eurostat projects on euro area coincident indicators - GPD, Employment, IPI Unsatisfactory results for IPI - high degree of volatility Alternative specifications of euro area GDP coincident indicators - three different Bridge equation models - a dynamic factor model

25 Coincident indicators - different models LARS algorithm - do not eliminate series from the data set ; series are ranked by decreasing predictive power according to the selection criterion of the algorithm. BHS - bridge model containing hard and soft data (BHS): Industrial Production Index, Construction Output Index, Consumer opinion over next 12 months, Employment expectations in construction, Construction confidence indicator and the euro/dollar real exchange rate FHS - first factor model, with factors built with soft and hard data FS - second factor model, with factor constructed exclusively with soft data. Factor models include: survey data: Industry, Consumers, Construction, Retail Trade hard data: Industrial Production excluding construction, Construction production, exports, retail sales and unemployment rate

26 Coincident indicators of the GDP Growth End of Quarter T 2008Q1 2008Q2 2008Q3 2008Q4 2009Q1 2009Q2 2009Q3 2009Q4 2010Q1 2010Q2 2010Q3 Models FS BHS FHS FS BHS FHS FS BHS FHS FS BHS FHS FS BHS FHS FS BHS FHS FS BHS FHS FS BHS FHS FS BHS FHS FS BHS FHS FS BHS FHS Estimates Estimates T 30 T Estimates T Eurostat Flash T

27 Monthly indicator of the economic activity Declining role of the IPI as reference variable for business cycle analysis service activities also characterised by cyclical movements industrial fluctuations not necessarily determine cycles for the whole economy GDP: ideal reference variable for business cycle analysis only available at quarterly basis construction of monthly GDP based on EA principles still problematic Ongoing Eurostat projects on the construction of euro area monthly indicator of the economic activity

28 Euro-Mind methodological description Construction based on quarterly GR-DP, output and expediture side components Selection for each component of a set of quantitative and qualitative monthly relate indicators Modelling monthly indicators using a dynamic factor analysis casted in a state- space form Dealing with multi frequency data in a state-space form in order to convert temporal aggregation into a systematic sampling problem Achieving computational efficiency by converting a multivariate estimation problem into a univariate filtering and smoothing Dealing with chain-linking measures by means of a multistep procedure exploiting the additivity property of the Laspeyers volume on previous year basis Estimating Euro-Mind as a combination of output and expeiture estimates using weights reflecting the relative precision

29 Euro-MIND growth rate on previous month Jan-05 May-05 Sep-05 Jan-06 May-06 Sep-06 Jan-07 May-07 Sep-07 Jan-08 May-08 Sep-08 Jan-09 May-09 Sep-09 Jan-10 May-10 Sep-10 Monthly growth rates Upper 95% Lower 95%

30 Extension of Euro-Mind Current ongoing projects: generalization of the model with better forward looking properties - two factors in the construction of the composite indicators, where the second one contains business and consumer surveys data. - this version of Euro-Mind increases its nowcasting and forecasting abilities at one-two-three steps ahead to jointly estimate a monthly indicator of economic activity at euro area and member state level in order to asses the relevance of national information to increase the reliability of euro area estimates to reconstruct Euro-MIND back in time possibly to the eighties and even seventies to produce an accurate estimate from mindgap

31 A statistical framework for business cycle analysis Not all needed information explicitly available by a simple data inspection Extracting signals and producing estimates for a better understanding of cyclical movements Eurostat ongoing activities aiming to the construction of a statistical framework for business cycle analysis - construction of chronologies growth cycle, classical cycle - construction of coincident turning point indicators - estimates of growth cycle (i.e. output gap of GDP) univariate and multivariate methods

32 Euro area turning point coincident indicators simultaneous analysis of classical business cycle and growth cycle in the so called ABCD framework statistical dating of euro area turning points by means of a simple non parametric dating rule comparison of euro area and Member States dating to achieve a final statistical chronology ensuring the maximum degree of consistency between the two approaches preliminary investigation of alternative models for the construction of coincident turning point indicators for classical business cycle and growth cycle variables selection for the growth cycle coincident indicators - Employment expectation, Construction confidence indicator, Financial situation of the last 12 months, IPI, Imports of intermediate goods construction of the growth cycle coincident indicators (GCCI) as a weighted mean of the transition probability returned by the five univariate two regimes Markov Switching models - equal weighting scheme variables selection for the business cycle coincident indicators - IPI, Unemployment rate, New cars registration construction of the business cycle coincident indicators (BCCI) as a weighted mean of the transition probability returned by the three univariate three regimes Markov Switching models - weighting scheme: 0.34, 0.46, 0.20 respectively

33 GCCI Probability of Being in a Recession of the Growth Cycle GCCI Forecasts 0.5 Threshold

34 GCCI Probability of Being in a Recession of the Growth Cycle Growth Cycle Reference Chronology Provisional Dating Chronology Ending Date of Provisional Chronology GCCI 0.5 Threshold

35 BCCI Probability of Being in a Recession of the Classical Business Cycle BCCI Forecasts 0.5 Threshold

36 BCCI Probability of Being in a Recession of the Classical Business Cycle Business Cycle Reference Chronology Provisional Dating Chronology Ending Date of Provisional Chronology BCCI 0.5 Threshold

37 Extension models inclusion of the accelerating cycle following the schema: α AB ß CD Developing coincident indicator for the acceleration cycle Developing multivariate models for simultaneous detection of turning points - this approach, among others, has the advantage of explicitly imposing the constraints derived by the ABCD approach

38 Growth cycle estimates Crucial importance of growth cycle estimates for policy makers and analysts - monitoring the inflationary pressures - designing a monetary policy oriented to inflation control Trend cycle estimated by means of filtering techniques - data revisions - end points estimates Ongoing Eurostat activities for the estimation of growth cycle - regular production of univariate growth cycle estimates Hodrick-Prescott, Christiano-Fitzgerald, Unobserved Components models Studying alternative multivariate growth cycle estimates - structural VARs, Unobserved components

39 EA GDP trend-cycle decomposition using HP filter (Cycle) (GDP, Trend) q q q q q q q q q q q q q q q q01 GDP Trend Cycle