Enhancing MIP data coverage: back-calculation and estimates of missing values ESTP Course Luxembourg 9-11 December 2014 Ferdinando Biscosi, MIP TF
Outline Background Methodology Approval workflow An application: HPI Other applications
Policy makers need a complete picture of the economy Background Policy makers need a complete picture of the economy Back-calculation & Missing values estimation Revisions New production standards New classifications Missing values
When do we need to back-calculate? Changes in methodology Change in classification systems Introduction of new regulations Major revisions
Back-calculation (1) Based on transparent and robust estimation techniques Targeted horizon back in time Available primary data at national level, historical context
Back-calculation (2) Choice of the model Set of validation criteria Simple, parsimonious, limited need for subjective judgment Set of validation criteria Measurable criteria, in agreement with MSs
Back-calculation (3) First step: Planning Variable to be back-calculated and possible related variables to be used as proxy Optimal horizon
Back-calculation (4) Second step: Data availability Search for: alternative sources alternative frequencies (quarterly, etc.) proxies
Back-calculation (5) Third step: Model selection Two scenarios: Only the target series is available There are related series with greater coverage
When only target series is available… series transformation (inverse) ARIMA model selection test parameters significance back series calculation
Issues: When only target series is available… Only limited numbers of steps in the past Feasibility depends on length of target series
When additional data is available… data transformation test correlation between additional and target series selection of explanatory variable OLS regression on overlap period obtained parameters used to estimate the back series
Issues: When additional data is available… Additional data not correlated to target series Feasibility depends on length of common period
Model selection Second scenario: additional data
Approval workflow: internal, at Eurostat level MIP back-calculations technical note feedback 1st Responsible Unit in Eurostat
Approval workflow: external, at MS level MIP Task Force feedback Responsible Unit in Eurostat technical note feedback 2nd Country contact NSI/NCB
The technical note Introduction about the MIP exercise Note explaining the additional data used and the model selected Schedule by which data should be validated
An application: HPI (1) Background New legislation entered into force in 2013 not covering past periods Some series rather short, starting in 2009 Data available for some past periods from other sources (NCB, ECB, BIS, OECD) Aim: ESTAT (2010=100) series for the period 2000q1 onward Quarterly series used to derive annual ones
An application: HPI (2) Coverage: 10th October 2014
An application: HPI (3) Coverage: 10th October 2014 13 Member States had at least one missing value Corresponding to 17.5% of the total number of required values
An application: HPI (4) Current coverage na Enlarged
An application: HPI (5) Work done: CZ NSI has provided back data for the period 2000-2007 BG NSI has provided estimates for the period 2001-2008 ES, LV, LT* for the period 2000-2005 and CY for 2002- 2004: NSIs have approved Eurostat estimates 2000-2005 (published flagged as estimates) EL: NCB data used by Eurostat as estimates AT: NCB data used by Eurostat as estimates until NSI data become available * LT MIP headline indicator not available for years previous 2006 due to derogation on the deflator
HPI – metadata file see: http://epp. eurostat. ec. europa
Example: Latvian HPI (1) 2000Q1 2006Q1 2013Q4 overlapping series NSI series ECB series Back-calculated Available
Example: Latvian HPI (2) High correlation index on the common period Indexes log transformed and first order differentiated to achieve stationarity OLS regression ran on common period Obtained parameters used to back-calculate ESTAT quarterly series Annual figures derived by quarterly ones
Example: Latvian HPI (3)
Other applications (1) When do we make estimates? Why/what for? Missing annual figures of the previous year Why/what for? Used only for Commission internal purposes Values are not officially disseminated/neither communicated to the MS DG ECFIN uses these figures in their forecast models
Other applications (2) Only target series available Test for parameters Estimation of missing figures using ARIMA models
Other applications (3) Additional related series available Test correlation between additional and target series Proxy selection Estimation of missing figures using proxy
Other applications: issues As for the back-calculation: Length of the target series Correlation between proxy and target series Parameters not significant
Future applications Unit Labour Cost The auxiliary indicator expressed as percentage change over 10 years ESA 2010 entering into force in 2014 NACE Rev. 2 entering into force in 2008
Open discussion Methods used in NSIs Exchange of experiences