Measuring Wealth Inequality in Europe: A Quest for the Missing Wealthy

Transcription

1 Measuring Wealth Inequality in Europe: A Quest for the Missing Wealthy 1 partly based on joint work with Robin Chakraborty 2 1 LISER - Luxembourg Institute of Socio-Economic Research 2 Deutsche Bundesbank Oesterreichische Nationalbank March 15, 2018 This project has mainly been carried out during the authors employment at the European Central Bank and within the work of the ECB Expert Group on Linking Macro and Micro Data for the Household Sector. The usual disclaimers apply.

2 Google Books Ngram Viewer (English)

3 Wealth and wealth distributions are back! Thomas Piketty (2013): Capital in the Twenty-First Century Stiglitz-Sen-Fitoussi report Recommendation 4: Give more prominence to the distribution of income, consumption and wealth G-20 Data Gaps Initiative Recommendation 16: [... ] statistical experts to seek to compile distributional information (such as ranges and quartile information) alongside aggregate figures, wherever this is relevant [... ]

4 Wealth and Income database: [... ] access to the most extensive available database on the historical evolution of the world distribution of income and wealth, both within countries and between countries

5 Wealth and Income database: [... ] access to the most extensive available database on the historical evolution of the world distribution of income and wealth, both within countries and between countries Expert Groups OECD-Eurostat Expert Group (mainly: income and consumption) ECB Expert Group on Linking Macro and Micro Data for the Household Sector (wealth) - launched in December exclusive focus on wealth - long-term goal: DINA for wealth - assessment of the use of the HFCS for distributional breakdowns

6 The particular importance of wealth inequality Wealth inequality through generation of capital gains can deepen income inequality, which is known to potentially disrupt social cohesion and may harm economic growth. High degrees of wealth inequality are associated with - high economic vulnerability - unequal political capital - limited social mobility (particularly due to inheritances) - limited equality of opportunity (particularly due to inheritances) (Program of the 2018 OECD Conference on wealth inequalities: Measurement and policies )

7 We currently experience rising wealth inequality and, in particular, rising wealth-to-income ratios: The inequality of wealth, and potentially the inequality of inherited wealth, is likely to play a bigger role for the overall structure of inequality in the twenty-first century than it did in the postwar period. (Piketty and Zucman, 2014, page 1261)

8 Why linking micro and macro data? Macro data embedded in the System of National Accounts links all sectors of the economy Linked data leads to a better understanding of macro phenomena (e.g., where does growth come from? who is affected by certain macro policies?) NA data are quite harmonized globally (SNA). In Europe, the ESA 2010 standard is legally binding for EU member states. Break-downs (by income/wealth quintiles, socio-economic indicators) should add up to macro totals to avoid confusions. Nudge users. Linking micro and macro data means exploiting the strengths of each source: - NA are good for aggregates - HFCS is good for distributional structure

9 How to Measure Wealth Distributions? Tax data Income Capitalization Method Estate Multiplier Method Direct Taxes on Wealth Household Survey Data HFCS (Household Finance and Consumption Survey) US Survey of consumer finances, UK Wealth and Assets survey, etc.

10 How to Measure Wealth Distributions? Tax data Income Capitalization Method - Data: Income tax files - Use capital income flows (mainly interest and dividends) and rates of return by asset class to impute wealth components - Not feasible for asset classes not generating taxable capital income flows (e.g., OOH) Estate Multiplier Method Direct Taxes on Wealth

11 How to Measure Wealth Distributions? Tax data Income Capitalization Method Estate Multiplier Method - Data: Inheritance taxes - Use wealth of decedents together with mortality rates (specific for certain demographic groups) to infer wealth of the living - Major drawbacks: mortality rates not random within groups, tax avoidance and evasion, tax exemptions, important components (pension/insurance entitlements) missing Direct Taxes on Wealth

12 How to Measure Wealth Distributions? Tax data Income Capitalization Method Estate Multiplier Method Direct Taxes on Wealth Household Survey Data HFCS (Household Finance and Consumption Survey) US Survey of consumer finances, UK Wealth and Assets survey, etc.

13 In theory, the ideal data source to study the distribution of wealth would be high-quality annual administrative data on wealth, based upon automatic transmission of information from financial institutions and real estate transactions to tax authorities. The perfect data source on wealth does not exist and will never do: one needs to be pragmatic [... ] (Alvaredo et al., 2016) Is the survey approach the flawed approach?

14 Pros and Cons of using Survey Data I + All asset classes and liabilities obtained from one data source + Asset classes that do not generate income flows are included (owner occupied housing) + Direct link to socio-economic characteristics potentially leading to detailed break-downs + Harmonization of data source across countries + Possibility to cross-check and correct answers based on register data

15 Pros and Cons of using Survey Data II Reporting errors (intentional and unintentional) Sampling errors Bad representation of top tail (Vermeulen, 2016, 2017) oversampling increases precision but cannot remove non-participation bias Long time spans between survey waves Self-valuation (households have lots of information but they are no experts questionable results for real estate and equity) Consistent definitions in survey and NA (challenging but possible)

16 The Household Finance and Consumption Survey (HFCS) Coordinated by the European Central Bank and carried out by National Central Banks and some National Statistical Institutes in Europe (in Austria: OeNB) Harmonized methodology leaving room for country-specific implementation General purpose survey: focus on assets and liabilities plus - Demographic information and Employment status - Income - (some information about) Consumption - Inheritances and gifts 1st wave in 2010, covering 15 countries 2nd wave in 2014, covering 20 countries 3rd wave fieldwork partly finalized / in progress

17 Oversampling Strategically sample more households from the top of the wealth distribution No over-sampling: x% of survey observations belong to the population s top x% of the wealth distribution. Successful over-sampling: More than x% of survey observations belong to the population s top x% of the wealth distribution (effective oversampling rate greater than 0). Over-sampling increases precision at the top but decreases precision everywhere else (BUT: no bias correction!). Wealth is heavily concentrated at the very top precision there is essential.

18 Oversampling in Austria and Germany Effective oversampling rates for the top 1%: AT: -33% DE: 131%

19 DINA and the HFCS: Strategy of the EG-LMM 1 Bottom-up approach: from instrument to net worth 2 Bridging instruments between NA and HFCS identify conceptual discrepancies and work on improvements 3 micro-macro gap: the most common case for highly comparable instruments is under-coverage Coverage Ratio = HFCS total NA total < 1 4 Understanding the sources of the gap and develop solutions

20 The micro-macro gap Discrepancies in definitions Differences in valuation concepts Population scope Reporting errors in HFCS Sampling errors in HFCS Recording errors in NA Rounding errors Missing wealthy (unit-non-response, lack of precision at the top) Important: Distinguish between errors affecting the shape of the distribution or the total only.

21 The top of the wealth distribution is not well reflected in any household survey. Replace top tail by parametric model (here: Pareto distribution)

22 The top of the wealth distribution is not well reflected in any household survey. Replace top tail by parametric model (here: Pareto distribution) Estimate Pareto model by combining HFCS observations and rich list data (accepting all its flaws) Rich list data: Forbes list (billionaires only), national lists ( obs., national currency) Break down inflated wealth distribution by instruments (relying on HFCS portfolio structure heavy assumption!)

23 Portfolio break-down Portfolio structures change over the wealth distribution

24 Portfolio break-down Portfolio structures change over the wealth distribution Variation also large within the tail (here: 1 meur) Split tail Y into 4 quartiles Q i and apply average portfolio structure per quartile to total wealth per quartile 0.25 N tail E Pareto( ˆϑ) [Y Y Q i ] share instrument i Analytical approach and Monte Carlo/bootstrap approach (totals plus variation)

25 Empirical results Large increases for equity (unfortunately not well comparable to NA definition; HFCS totals exceed NA totals) Also increases for other instruments but less pronounced (usually up to 10 pp) Still substantial gap in coverage ratios Increases are larger for countries that do not oversample the wealthy lacking oversampling limits comparability Interesting: Austria and Germany are more similar after adjustment Robustness checks: General results not very sensitive towards modelling choices

26 Lorenz curves

27 Share of total assets held by millionaires

28 Coverage Ratios in AT

29

30 How to use results for DINA

31 How to use results for DINA

32 How to use results for DINA

33

34 Robustness Check: France Wealth shares computed via a (mixed) income capitalisation method: (i.e., we now how much of total wealth the top x% possess) Assumptions: - HFCS is correct for the entire distribution except the top 3% (millionaires) true wealth = observed wealth + ε - Tail follows a Generalized Pareto Distribution - True wealth share held by the top 3% is correctly measured by WID - Portfolio shares observed in the HFCS are representative for the top (again: quartile-specific approach)

35 Adjusted coverage ratios Before Top 3% Forbes CAPITAL adjustment share list list Liabilities Deposits Bonds Mutual Funds Equity Shares

36 Conclusions I There truly seems to be something missing at the top... and it s mainly equity. Making equity more comparable between NA and HFCS is essential. Harmonization of over-sampling strategies would be great. Introducing oversampling in Austria is even more important! Adjustment to account for the missing wealthy works well, as long as there is one piece of external information (rich list, average wealth of the richest x households, etc.) More information about portfolio structures at the very top would be helpful.

37 Conclusions II Once we ve understood the shortcomings in the HFCS / NA and found methods to adjust for it, the HFCS seems to be well suited for the compilation of DINA. Even better: Register-data enriched HFCS. Very high degree of harmonization possible. Wealth inequality in Europe is striking!

38 DINA HFCS DINA and the HFCS Empirics Conclusions References

39 References Alvaredo, F., Atkinson, A., Chancel, L., Piketty, T., Saez, E., and Zucman, G. (2016). Distributional National Accounts (DINA) guidelines: Concepts and methods used in WID.world. WID.world Working Paper Series, No. 2016/1. Piketty, T. and Zucman, G. (2014). Capital is back: Wealth-income ratios in rich countries The Quarterly Journal of Economics, 129(3): Vermeulen, P. (2016). Estimating the top tail of the wealth distribution. American Economic Review, 106(5): Vermeulen, P. (2017). How fat is the top tail of the wealth distribution? Review of Income and Wealth, forthcoming.

40 More results Consistency of DINA

41 More results Illustration: Semi-parametric approach

42 More results Observed tail portfolios in Austria

43 More results Shares held by millionaires in Germany

44 More results Robustness: Pareto threshold parameter

45 More results Alternative estimation of the Pareto shape parameter