The Income, Consumption and Wealth project at Eurostat
PierreLamarche ()[1], SigitaGrundiza
Keywords:Wealth, saving rates, distributional National Accounts, statistical matching, survey integration.
1.Introduction
The new European Commission has stressed the need to bring social indicators on a par with macroeconomic indicators within the macroeconomic governance.Income, Consumption and Wealth (ICW) are three key dimensions that determine the economic well-being of people or material inequalities. ICW situation for the individual describes the level and realization of economic opportunities. The data availability of the joint the distributions of IWC and their dynamics are needed to fill the data gaps for policy making.
Meanwhile, the recommendations in [1] summarise the current orientations. They insist on the need for focusing on the household perspective, considering income and consumption jointly with wealth. They also underline the necessity of giving more prominence to distributions and better link macro-data with micro-data.
In addition, several rounds of consultations have been launched by Eurostat in order to better assess the users' needs and draw the state of play in terms of statistics on ICW in the European Union. As a result, one day of the last DGINS conference host by Statistics Austria was dedicated to the measurement of ICW by European National Statistical Institutes; the conclusions of the discussion recalled the importance of such statistics and agreed on the way forward for such a project in the European Statistical System.
2.Methods
Eurostat has identified the data gaps related to the distributional aspects of the household ICW:
- Currently joint distribution of micro data is not available. The three components traditionally are collected by three different household surveys: EU-SILC (Eurostat, run every year), HBS (Household Budget Survey – Eurostat, every 5- years) and HFCS (Household Finance and Consumption Survey, every 3-years, is coordinated by ECB and implemented in the euro area).
- There are notable conceptual and methodological differences between macro (National accounts) and micro statistics (household survey data) that do not allow making the distributions of aggregates directly;
2.1.ICW comprehensive distributions based on micro-data
The first possibility would be to improve the harmonisation of demographic variables in the different surveys, thereby increasing integration potentialities across surveys. The ultimate goal could be the design of one single European survey, containing a core module dedicated mainly to demographics, along with modules alternatively implemented on wealth, income and consumption.
The way to go as a "second-best" solution in the current setting consists of the implementation of statistical matching procedures in order to link the different surveys and have at one's disposal joint information on income, consumption and wealth at the micro-level. However, such a procedure relies on strong assumptions and may perform very poorly for reproducing the expected marginal distribution of income, consumption and wealth. [2] have described the various caveats that have to come along with this kind of exercise. In particular, the Conditional Independence Assumption (CIA) may jeopardize the entire exercise in the case it is not verified. From this point of view, it is essential to assess uncertainty when performing statistical matching and disseminating the results.
Statistical matching has already been explored for EU-SILC by Eurostat (see [3]). The matching between EU-SILC and HBS or between EU-SILC and HFCS have been experienced, essentially in order to design the 2017 module on Wealth, Over-indebtedness and Consumption. We rely on the results of these first investigations to produce further estimates on the link between income, consumption and wealth. We also try to rely on the existing variables and the already integrated surveys to assess the quality of the methods we are implementing: for the countries that collect at the same time income and wealth, or income and consumption we are able to replicate a natural experiment in order to assess how well the statistical matching is able to reproduce observable links between these dimensions. Also the HFCS contains several variables on income and consumption (following for this last dimension [4]), which may also be used for the statistical matching. We compare the results that we are able to produce according to the different implemented methods (hot-deck, rank hot-deck, distance hot-deck, conditional mean, mixed methods, or even methods described by [5] or [6]). We finally try to relax the CIA, either by producing estimates of the Frechet bounds (for contingency tables) or through multiply-imputed data (following [7]).
2.2.Linking the micro data with National Accounts aggregates
When working on the issue of inequalities in terms of ICW, one natural benchmark is the figures from the Household Accounts, since the distributional information carried out by surveys should enable to go further than the aggregates from the National Accounts. This being said, there are several caveats when comparing macro-statistics with micro-data. It is indeed necessary to compare figures that are conceptually similar, since for instance disposable income in the National Accounts does not designate the same concept than in the surveys. Once things have been made conceptually equal, there are still in general quite strong discrepancies that may be explained thanks to underreporting phenomenon, sampling errors and so on (on possible explanations for discrepancies for instance, see [8]).
One way to produce data that are consistent between micro and macro sources is to upscale the components of income, consumption and wealth measured at the micro-level and mapped with the National Accounts according to their aggregate counterpart. This implicitly makes the assumption that underreporting is uniform across the population for one given component; all the source of heterogeneity is then the difference in terms of portfolio composition, income or consumption structure among households (see [9] and [10] for concrete examples). We follow this framework; also assess the quality of the mapping between micro and macro data. We also experiment techniques for upscaling the amounts in a non-uniform way and evaluate the sensitivity of the results to the uniform assumption. Finally we are able to produce experimental data that provide information on the ICW joint distributions and that are as much as possible coherent with aggregates.
3.Results
3.1.Statistical matching
The exercise of statistical enables to compute joint distribution between income and consumption. The Figure 1 shows the repartition of the population according to deciles of income and consumption for Belgium; the matching has been performed for all EU countries having transmitted micro-data for HBS 2010.
Figure 1. Heatmap for the joint deciles of income and consumption in Belgium, 2010, SILC matched with HBS (hot-deck matching)
The produced data reproduce expected stylized facts such as the "twin peaks" in the bottom and in the top of the distribution (as in [11]). Saving rates computed from this exercise show the distribution of saving flows among the European households, thereby complementing the already existing measures of poverty based on SILC data.
3.2.Comparison between micro and macro sources
Income components have been mapped and compared across the different sources: if employee income (as in Figure 2) shows good coverage rate, other component of income (e.g. mixed income, property income) are less well covered, conceptually and practically.
Figure 2. Coverage rate for employee income between SILC and NA
4.Conclusions
The produced data are at this stage experimental; however they can be used to link saving rates with wealth inequalities, and help in better understanding phenomena of vulnerability. These data have to be used carefully and any conclusion drawn out of them should come along with sensitivity checks. Disseminating data along with tools to better comprehend uncertainty constitutes another challenge; in the end all this should support also the need for better integrated statistics.
References
[1] Stiglitz, J., Sen, A. and Fitoussi, J.-P. "The measurement of economic performance and social progress revisited." Reflections and overview. Commission on the Measurement of Economic Performance and Social Progress (2009).
[2] D'Orazio, M., Zio, M. D. and Scanu, M. "Statistical matching: Theory and practice." John Wiley & Sons (2006).
[3] Leulescu, A. and Agafitei, M. "Statistical matching: a model-based approach for data integration" Eurostat Working Papers (2013).
[4] Browning, M., Crossley, T. F. and Weber, G. "Asking consumption questions in general purpose surveys". The Economic Journal, F540-F567 (2003).
[5] Baldini, M., Pacifico, D. and Termini, F. "Imputation of missing expenditure information in standard household income surveys", Modena: DEMB Working Paper Series (2015).
[6] Conti, P. L., Marella, D. and Neri, A. "Statistical matching and uncertainty analysis in combining household income and expenditure data." Bank of Italy, Economic Research and International Relations Area, Temi di Discussione, Issue 1018 (2015).
[7] Rässler, S. "Data fusion: identification problems, validity, and multiple imputation." Austrian Journal of Statistics, 33(1-2), pp. 153-171 (2004).
[8] HFCN. "The Eurosystem Household Finance and Consumption Survey. Methodological Report for the First Wave", pp. 89-98 (2013).
[9] Fesseau, M. and Mattonetti M. L. "Distributional Measures Across Household Groups in a National Accounts Framework: Results from an Experimental Cross-country Exercise on Household Income, Consumption and Saving", OECD Statistics Working Papers, No. 2013/04, (2013).
[10] Mattonetti, M. L. "European household income by groups of households", Eurostat Statistical Working Papers 13-023, (2013).
[11] Fisher J., Johnson D. S., Smeeding T. M., Thompson J. P., "Inequality in 3-D: Income, Consumption and Wealth", 34th IARIW Conference (2016).
1
[1]Eurostat