Session 22

Back to Schedule

Title of session: Quality of flash estimates

Chair: Jean -Pierre Poncelet

Room: S4A Mariacki

Time: 17:00 - 18:30

Date: 27 June

Session 22 - papers & presentations

Presenting AuthorAbstract
Cristina Lopez-Vilaplana
Title: <<< Flash estimates of income distribution indicators for the European Union: results 2016, methodology and quality assessment >>>
Indicators on poverty and income inequality represent an essential tool to monitor progress towards the Europe 2020 poverty and social exclusion target and to prepare the European Semester. They are based on EU statistics on income and living conditions (EU-SILC) and are available for all countries around 18 months after the reference period. In order to better monitor the effectiveness of social policies at EU level, it is important to have more timely indicators. A new approach was therefore proposed, which consists in the development of flash estimates (FE). FE have currently a release date approximately one year earlier than the actual data. The main methodology used is based on microsimulation techniques further enhanced to take into account the evolution in employment, population structure and indexation factors. Developing flash estimates on poverty and income inequalities in the ESS involves that their methods, sources and output adhere to a common quality framework. This includes: 1) quality checks concerning input coherence and intermediate steps 2) the historical performance of the model is defined as the ability to predict accurately the past changes as captured by EU-SILC and 3) the plausibility of the estimated changes, mainly by linking these to evolutions in observed indicators (e.g. employment trends, total household income in national accounts, national data) and disentangling the impact of simulated policies via EUROMOD. EUROSTAT has published for the first time FE 2016 as experimental statistics. While there are still limitations and we cannot expect the estimates to capture perfectly EU-SILC changes, the FE can provide useful information about the direction and magnitude of the change. The FE 2016 include several indicators, including the at-risk-of-poverty and interquartile share ratio. Deciles that measure changes at different points of the distribution seem to be important complements as early warnings for yearly changes.
Véronique Deneuville
Title: <<< ESS-cooperation on Employment Flash Estimates >>>
Following the successful introduction of EU and euro area GDP flash estimates after t+30 days in April 2016, Eurostat and national statistical institutes (NSIs) are cooperating on a new project: the estimation of EU and euro area employment flash estimates in national accounts, which could be published at t+45 or t+30 days. A task force has been set up in 2017 to work on test estimates and quality criteria to decide on the feasibility of such estimates. The task force allows NSIs to share their experience on the use of available source data as well as estimation techniques. These will be documented in a manual to provide methodological guidance to NSIs and information to users. While the estimation of European aggregates is obtained as a weighted average of countries' growth rates, econometric techniques will be explored to complement for missing country data and/or to address a systematic bias in the estimates reflecting countries' revisions. Following the first test estimates, country coverage and results look promising for a t+45 flash estimate. Estimates at t+30 will be explored, but are more challenging, since countries generally need to use modelling techniques to estimate missing data for the last month. The decision on the publication of EU and euro area employment flash estimates will be taken based on live estimates and back estimates and on quality criteria combining minimum country coverage and reliability of the estimate. While it is too early to draw conclusions at this stage, Eurostat is optimistic that the key macroeconomic indicators can be complemented by employment flash estimates. In any case, the work of the Task Force is an appreciated opportunity to share experience between NSIs and a good example of cooperation within the ESS to improve the availability of high quality statistics for users.
Kevin Schmitt
Title: <<< Flash estimate of the poverty rate using microsimulation : estimations based on French data >>>
Every year in September N+2, INSEE publishes the poverty rate and the main indicators of inequalities in standard of living for year N. This delay is unsatisfactory for meeting the social requirements of users of these indicators. Of the 21 months between the end of the year under consideration and the publication of the poverty rate, about three-quarters of this time is taken up collecting tax and social data, and about one quarter with statistically matching Labour Force Survey (LFS) data, from which the Tax and Social Incomes Survey (ERFS) is produced. Nowcasting consists of producing an earlier indicator of the poverty rate for the target year N (in autumn N+1) based on the ERFS N-1. The method to be used here is microsimulation, which creates individuals’ standard of living by imputing benefits and contributions on scales, and thus it is possible to take account of any legal changes made to these measures. The exercise is based on the INES model, which simulates the majority of French social security and tax legislation, based on any year of the ERFS. To implement nowcasting, one important step is ageing population by uprating incomes (using surveys about wages, aggregated tax data, inflation...) and calibration weighting (using margins from LFS and census). Reverse ageing is also used so that evaluations for year N and N-1 (and thus annual evolutions) are only based on the ERFS N-1 (that is minimising the sample bias). In this paper, we present the methodology and assess the quality of the early indicators thus produced. Indeed, we compare the results that would have been produced by microsimulation with those that were in fact disseminated from the ERFS. When applied to the target years 2010 to 2015, this method producedflash estimatesimilar to the actual figures published the following year.
Henri Luomaranta
Title: <<< Nowcasting Finnish Real Economic Activity: a Large Dimensional Approach >>>
Despite the evergrowing data availability, statistical institutes publish economic indicators with considerable lag and the initial estimates are revised considerably over time. In Finland, the first estimate of GDP provided by Statistics Finland is released 45 days after the end of the reference quarter (flash estimate), while the first "appropriate" version is released 60 days after the end of the quarter. We develop a nowcasting framework in order to provide faster estimates of the monthly real economic activity indicator, the Trend Indicator of Output (TIO), and of quarterly GDP. In particular we rely on firm-level turnovers, which are available shortly after the end of the reference month, to form our set predictors. Given the large dimension of our dataset, we rely on a set of statistical models and machine learning methodologies which are able to handle high-dimensional information set to compute the nowcasts. The results of our pseudo-real-time analysis indicate that it is possible to provide substantially faster estimates of the TIO and GDP without increasing the revision error. Finally, we examine the nowcasting accuracy obtained by relying on traffic data extracted from the Finnish Transport Agency website, and find that using machine learning methods in combination with this big-data source provides competitive predictions of real economic activity indicators.
Črt Grahonja
Title: <<< Use of alternative data sources as flash estimates of economic indicators – abstract >>>
At the Statistical Office of the Republic of Slovenia we calculate the statistics of gross domestic product (GDP) every quarter of a year. 60 days after the reference period is unfortunately the quickest we can publish such statistics as the timeliness of GDP data is limited by the survey evaluations of some of the components that make up GDP. However the use of flash (rapid) estimates could fasten this process. On the basis of investigation of various big data sources we had the idea to use the data we acquired from traffic sensors and use them as primary and secondary regressors in a linear regression model for nowcasting GDP 45 days after the reference period. Nowcasting is a method of calculating estimates on the basis of unknown present or near-future values with the use of a known correlator. In the following article we describe our work and the process of nowcasting indicators from the point of data acquisition to the end results on GDP and also on a known GDP correlator, the Industry Turnover Index. We also touch on what could be extended in the future like component estimation, model accuracy improvement and data processing improvements. We wish to show how useful such data can be and what was needed to be done, before this data could actually be used. Different types of knowledge were used while composing the selected process for economic indicator estimation. These include new skills in the fields of information technology and methodology, and knowledges in the respective subject matters.

Back to Schedule

Font Resize