Camille Landais, Emmanuel Saez, Gabriel Zucman · VoxEU column, 3 April 2020; reprinted in A. Bénassy-Quéré and B. Weder di Mauro (eds.), Europe in the Time of Covid-19, CEPR Press 2020: 113-118

A Progressive European Wealth Tax to Fund the European COVID Response

7 problems found: 1 major · 2 moderate · 4 minor

What the paper claims

A time-limited, EU-wide progressive wealth tax on the net worth of the richest 1% (1% above ~EUR 2m, a further 1% above ~EUR 8m, a further 1% above EUR 1bn) would raise 1.05% of EU GDP each year, 'accounting for evasion and avoidance responses'.
If fighting COVID-19 requires issuing 10 points of EU GDP in Eurobonds (or a rescue fund of that size), the tax would be enough to repay all this extra debt after ten years.
Tax-base inputs. EU household wealth is about 5x GDP (Piketty & Zucman 2014). The top 1% of EU adults own about 22.5% of wealth and the top 0.1% about 10%. The taxable fraction above each threshold is 60% (Pareto a = 2.5/1.5 = 1.67, 'in line with' Garbinti et al. 2017). The bases are 67.5%, 30% and 7% of GDP. There are 330 EU billionaires with 'almost exactly' EUR 1tn of wealth above EUR 1bn (Forbes 2019). National top-1% shares are 20-25% in France, Germany, Spain and Scandinavia. The shares are said to be 'net of tax evasion' because they are based on reported wealth or reported capital income (VoxEU fn. 5).
Historical rationale. France inflated its debt away immediately (~50% inflation a year in 1945-48). The UK erased its debt only in the 1970s, with double-digit inflation for an entire decade. Post-WWII Germany instead 'put in place progressive wealth taxes' that were time-limited and 'highly progressive' and 'paved the way for the German post-war economic miracle'. Hence 'we would be wise to follow the example Germany set after WWII'.
Design and efficiency claims. (a) A time-limited wealth tax 'operates like a capital levy': it taxes past accumulation and leaves the returns to current investment and innovation unaffected, so it is less likely to harm growth than other consolidation. (b) The rates are 'neither large nor unprecedented'. (c) At EU level, intra-EU migration 'becomes irrelevant' and enforcement is easier.

Every problem found

1majorconfirmed · previously known · overclaiming

The column says its revenue figure already allows for wealthy people dodging the tax. In fact the calculation multiplies the tax rates by today's wealth and assumes nobody changes behaviour. Suppose the rich shield only 15% of their wealth, the figure the same authors assumed for a US proposal a year earlier. The tax then raises about 0.9% of GDP instead of 1.05%, and repaying the debt takes about 11 years rather than 10. With larger but plausible responses it takes 13-16 years. The tax would still raise a lot of money, but the precise promise is not backed by the calculation.

Technical detail and evidence

Why it matters, in detail. R = sum_k tau_k B_k with static bases, i.e. an implicit taxable-wealth elasticity of zero. Revenue under alternatives, computed by bracket slice and by taxpayer at the marginal rate on a distribution calibrated to the column's bases plus the 302 Forbes 2019 EU27-citizen billionaires: - 15% uniform (SZ 2019): 0.89% of GDP - semi-elasticity 8 (SZ's own average): 0.89-0.82% - semi-elasticity 7-17 (Advani-Tarrant): 0.91-0.58% - semi-elasticity 13 (PWBM): 0.80-0.68% - semi-elasticity ~20.8 (Jakobsen et al. DiD): 0.65-0.48% - 30-40% (Kapeller et al.): 0.73-0.63% At r = g the years needed to repay 10% of GDP rise from 9.6 to 11-21 (0.4-2.3 years fewer if responses phase in linearly over 5 years). Scandinavian bunching estimates (semi-elasticity <1) would leave C2 intact; this is the authors' best case, along with possible capture of previously hidden offshore wealth, which the column does not quantify. The 0.26 effective/mechanical ratio for Scandinavian reforms (~0.27% of GDP) is an upper-bound reference only, because an EU-wide tax neutralizes intra-EU migration.

Evidence. Recomputed exactly from Box 1 (scripts/revenue_response_debt.py). Saez & Zucman, letter to Sen. Warren, 18 Jan 2019, p. 2 and fn. 2: 'To be on the conservative side, we assume that households subject to the wealth tax are able to reduce their reported net worth by 15%' (from an average semi-elasticity of 8). The letter also taxes only 85% of Forbes values. Comparable EU-wide estimates build responses in. Krenek & Schratzenstaller (2018, WIFO WP 561, pp. 18-19) apply the Bruelhart et al. Swiss response: a 1 pp rate rise cuts the base by 35%. Kapeller, Leitch & Wildauer (2021, FEPS, Table 5, p. 17) cut mechanical revenue by ~30% ('evasion') and ~40% ('strong evasion'). The column cites France, Germany, Denmark and Sweden as precedents for its rates. In their last years those taxes raised 0.24%, 0.11%, 0.06% and 0.19% of GDP; Spain raised 0.11% (K&S 2018, Table 1). French ISF revenue was 'one sixth' of K&S's estimated potential, mostly because of design. CAE Focus no. 118 (July 2025, p. 16) is co-signed by Landais, one of the column's authors. Citing Jakobsen, Kleven, Kolsrud, Landais & Munoz (2024), it states that in the Scandinavian context behavioural responses cut revenue from taxing top wealth or capital income by 74% relative to the mechanical effect: 54 cents at the intensive margin and 20 cents through migration. Forbes 2019: 59 of the 330 EU27-citizen billionaires ($282bn) are not recorded as EU27 residents (Switzerland 20, UK 17, US 7, Monaco 7; 2 with no residence recorded).

Why this severity. The response adjustment the headline claims is absent, and the ten-year horizon tolerates only a 4.3% revenue loss. Under the authors' own 2019 convention (15%), revenue is 0.89% of GDP and repayment takes 11.3 years. Mid-range literature elasticities (13-38% losses) give 0.65-0.91% of GDP and 11-16 years. The qualitative conclusion survives (substantial revenue, repayment within one to two decades), but the precise headline does not.

The authors' response. None specific to this column located. The authors' general position (Saez & Zucman 2019 Warren letter and BPEA 2019) is that avoidance is small when the base is comprehensive, there is systematic third-party reporting and audits ('avoidance is a policy choice'). According to the corpus (offshore-havens-17; a France 24 report, 27 Sep 2025, not read directly), Landais said the Danish evidence 'n'a en pratique rien a voir' with a minimum tax on total wealth at market value.

2moderateconfirmed · previously known · methodology

The authors say a temporary wealth tax only hits wealth people already have, so it does not discourage new investment. That is true only if the tax is based on what people owned on a fixed date in the past. The column's numbers describe a tax charged every year on current wealth, which does tax new savings and new businesses for as long as it lasts. The growth argument is therefore overstated, though the revenue sums are unaffected.

Technical detail and evidence

Why it matters, in detail. Under annual reassessment the tax is a recurrent levy on the stock. That is equivalent to a capital-income tax of tau/r on the marginal return, i.e. 25-75% at r = 4% for tau = 1-3%, not a lump-sum levy on inherited positions. Neutrality requires a credible fixed-date base. The column's revenue path (1.05% of GDP each year) instead assumes a base that moves with current wealth.

Evidence. eBook p. 116 ('a time limited wealth tax operates like a capital levy: you tax past accumulation but the returns to current investment and innovation are unaffected'). Adjudicator check of (1-tau)^10 = 0.904 / 0.817 / 0.737, matching Auditor B. Retrospective valuation dates: O'Donovan (2020, Wealth Tax Commission Evidence Paper 7) pp. 12-15, and Albers, Bartels & Schularick (2020) pp. 22-23. Gros (2020, VoxEU, 22 April, 'A Corona financial solidarity levy') contrasts LSZ's wealth tax with a one-time levy whose base date is set 'at the time of the announcement' so that it 'cannot be anticipated'. Jakobsen, Jakobsen, Kleven & Zucman (2020, QJE) document wealth-accumulation responses to annual wealth taxes.

Why this severity. The issue does not change the static revenue arithmetic. It does invalidate the column's main argument that the instrument is less growth-harmful than alternatives. It also shows that the larger annual-tax responses in F1 are the relevant ones for the design actually scored.

3moderateconfirmed · newly identified here · misrepresentation of source

To justify the plan, the column says post-war Germany paid off its war debts with highly progressive, temporary wealth taxes instead of inflation, and that this launched its 'economic miracle'. In fact a 1948 currency reform wiped out most of Germany's war debt and left savers with a few percent of their balances. The famous 1952 levy was a flat 50% charge on 1948 wealth, paid off over about thirty years, and it compensated war victims rather than repaying debt. France also taxed wealth after the war, and two of the three works cited conclude that such levies usually failed. The history does not support the lesson drawn from it, although this does not change the revenue sums.

Technical detail and evidence

Why it matters, in detail. The relevant precedent is a flat, fixed-date capital levy used for burden-sharing after a currency-reform write-off of public debt. It is not an annual progressive wealth tax used to amortize an intact debt stock. The precedent therefore supports neither the repayment mechanism nor the progressive annual design, and the evidence on capital levies the column cites (Eichengreen) is negative except under occupation (Japan) or capital controls. No revenue number changes.

Evidence. Albers, Bartels & Schularick (2020, the column's own German source) pp. 22-23. O'Donovan (2020) pp. 12-16 (ISN rates and yield; levy 'charged at a flat rate of 50%'). Deutsche Bundesbank Monthly Report, March 2002, p. 21. Galofre-Vila, Meissner, McKee & Stuckler (2019, ERHE 23(1)), abstract. IMF Global Debt Database (DEU 1950: 18.8% of GDP; GBR: 179.9% in 1950, 73.4% in 1970, 43.7% in 1980). ONS RPI series CZBH, 1970-79. Hicks, Hicks & Rostas (2nd ed. 1942), ch. XXIII-XXIV and preface. Eichengreen (1989) pp. 33-38. INSEE price index: 47.5%, 52.9%, 49.7% and 58.5% for 1945-48. Hughes (1999) was not read.

Why this severity. There is no effect on the revenue arithmetic. But the column's stated reason for the policy ('We would be wise to follow the example Germany set after WWII. This is why we propose...') rests on an inaccurate account and on citations that conclude otherwise. This is a secondary conclusion.

4minorconfirmed · newly identified here · documentation gap

The estimate is built from a few round numbers, such as how large European wealth is relative to the economy and what share the top 1% hold. Some are a little off. European wealth is closer to 4.5 than 5 times annual output, while the share held by the very rich is understated. The errors roughly cancel: with today's best European data the authors' tax base comes out almost exactly right, so this does not change the conclusions.

Technical detail and evidence

Why it matters, in detail. B = (W/Y) x s(top 1%) x (1 - 1/b) = 5 x 0.225 x 0.6. Each factor deviates from its cited source or from current data: W/Y ~4.2-4.7 rather than 5; b ~2.2-2.3 for France but ~2.7-3.1 for the pooled EU; s ~25-26% pooled rather than 22.5%. Because the errors offset, the mechanical base is well supported by independent 2018-2020 data, even though its construction cannot be reproduced from the column.

Evidence. Taken one at a time, the corrections change revenue by -8.8% (W/Y 4.53), -14.9% (W/Y 4.2) and -6.2% (France excess fractions) (scripts/revenue_response_debt.py, scripts/pz2014_europe_gdp.py, scripts/wealth_gdp_check.py). They are offset by concentration: the current WID pooled EU27 2019 top-1% share is 25.9% (vs 22.5%), with b(p99) = 2.70. An independent reconstruction from current WID g-percentiles for the EU27 (scripts/wid_eu_bases.py) gives wealth above EUR 2m of 67.6% of 2019 GDP and above EUR 8m of 32.0% (column: 67.5% and 30%). The mechanical yield is 1.03% of GDP (1.05% with Forbes residents for the top layer); it is 1.01% in 2018 and 1.13% in 2020.

Why this severity. Individual inputs are off by 6-15%, but they err in offsetting directions. An independent reconstruction reproduces the bases within a few percent, so the mechanical revenue estimate stands. What is missing is documentation and an uncertainty band of roughly +/-10%.

5minorconfirmed · newly identified here · documentation gap

In the printed book, the footnote that should explain how tax evasion was handled is blank, so book readers get the claim with no explanation. The numbers are the same in all versions.

Technical detail and evidence

Why it matters, in detail. The citable book version drops the only argument offered for the response claim in C1 (that the shares are net of pre-existing evasion). The numbers are identical across versions.

Evidence. eBook PDF page 133 (printed p. 116), rendered and inspected; VoxEU text of 3 April 2020 (Wayback captures).

Why this severity. This is a documentation and version defect with no numerical effect. It is probably an editorial omission.

6minorconfirmed · newly identified here · misrepresentation of source

The authors say their wealth figures come from tax records, so wealth already hidden from the tax authorities is left out. That is true for France and Scandinavia, but not for Germany, whose figures come from surveys and rich lists, or for the billionaire figures. It hardly changes the totals, but it weakens the only explanation given for the claim that evasion is accounted for.

Technical detail and evidence

Why it matters, in detail. Footnote 5 is literally true only for the French and Scandinavian inputs. The survey-plus-rich-list German shares and the 7%-of-GDP Forbes layer are not 'net of evasion', so this argument cannot be relied on even as a bound on pre-existing evasion.

Evidence. ABS (2020) Table 1 and pp. 2-3; WIR 2018 Part IV, section 4.5; AJZ (2019) data workbook, 'onshore' shares.

Why this severity. The components affected are a minority of the base, so the effect on revenue is small. The issue further weakens the only justification given for the evasion claim.

7minorlikely · newly identified here · methodology

The count of 330 'EU billionaires', with about EUR 1 trillion above the EUR 1 billion threshold, includes EU citizens living in Switzerland, the UK, Monaco and the US. A European tax would normally apply to residents, and counting only residents gives about a fifth less. Because this top bracket is a small part of the total, overall revenue falls by only about 1%.

Technical detail and evidence

Why it matters, in detail. Moving the EUR 1bn layer from 7.0% to 5.6% of GDP lowers revenue from 1.045% to 1.031% of GDP. The choice is undisclosed, and it is inconsistent with a residence-based tax and with the claim that only intra-EU migration matters. 18% of EU27-citizen billionaires (19% of their wealth) are not recorded as EU27 residents.

Evidence. scripts/forbes_base.py on the Forbes 2019 API list (2,153 entries), with ECB rates (8 Feb 2019: 1.1346; 2019 average: 1.1195 USD/EUR) and Eurostat EU27 2019 GDP of EUR 14.138tn. Both auditors obtained the same 330 / 272 counts independently. Taking 85% of Forbes values gives a layer of 5.64% of GDP; 50% (estate-tax evidence, us-wealth-taxes-11) gives 2.68%.

Why this severity. The 3% bracket provides 0.07 of the 1.045 pp. Residence instead of citizenship lowers total revenue by 1.3%; other adjustments are of similar or smaller size (up to ~4% for a 50% Forbes discount).

Claims examined and rejected

Problems one auditor raised that did not survive checking.

I4: 'Five times GDP (Piketty and Zucman 2014)' relabels a national-income ratio as a GDP ratio (misrepresentation of source). — The mechanism is not established. PZ's French workbook gives private wealth of 5.06x GDP in 2010 (4.9x with WID's NNI/GDP), and their four-country European average converts to ~4.5-4.7x GDP. 'About five' is therefore a ~7-10% rounding up, not a demonstrable relabeling. The substance is retained in F5: 2019 EU data give 4.2-4.5x, and the input is offset by conservative concentration. Auditor B's rejection of the confusion allegation is upheld. B's implication that 5x is supported for the EU, based on France alone, is not.
I5(c): because the column's own inputs imply a p99 of EUR 1.79m, a statutory exemption of exactly EUR 2m would cut the first base from 67.5% to 62.7% of GDP (-4.6% revenue). — This is an artefact of the column's simplified Pareto inputs. Independent data give wealth above exactly EUR 2m of 67.6% of GDP (WID EU27 2019), matching the column's 67.5%, so no revenue correction follows.
I5(b): the 10% top-0.1% share is 'more generous than the cited sources' and overstates the second base (revenue -3% with a Pareto-consistent 8.96%). — The inconsistency is real, but its direction is ambiguous. The two shares jointly imply a thicker tail (b ~2.84), which would raise the excess fraction. Pooled EU data give a higher top-0.1% share (11.4%) and a larger base above EUR 8m (32% of GDP). The point is kept only as an undocumented inconsistency in F5.
R11 / assumption A8: even with zero behavioural response, the tax's own mechanical erosion of top wealth (~1.5% a year) pushes repayment to 10.3 years. — Not upheld as a defect. The calculation assumes that, absent the tax, top wealth grows no faster than GDP. The authors explicitly assume the opposite (Warren letter, point 6), and EU27-citizen Forbes wealth above EUR 1bn rose from EUR 0.99tn (2019) to EUR 1.41tn (2021) and EUR 1.40tn (2022). It is reported only as a sensitivity.
I2: the cited sources (Hicks et al. 1941 and Eichengreen 1990) 'conclude the opposite' of the column. — Upheld for Eichengreen, partly rejected for Hicks et al. Hicks et al. found earlier levies mostly failures but wrote that the cases 'do not prove anything about the feasibility of a levy under appropriate conditions'. Their book is better described as predating and not addressing the post-war German taxes. F3 is worded accordingly.

Audit notes

What reproduced. Reproduced: - Box 1 arithmetic: exact (1.045% of GDP, printed 1.05%; nested and disjoint-slice formulations agree; the Pareto algebra in fn. 4 is correct). - Billionaire layer: EUR 1tn = 7.07% of EU27 2019 GDP. Forbes 2019 has exactly 330 EU27-citizen billionaires with EUR 0.97-0.99tn above EUR 1bn (matched by both auditors and the adjudicator). - Mechanical base, from an independent reconstruction with current WID EU27 g-percentiles (2019): wealth above EUR 2m = 67.6% of GDP, above EUR 8m = 32.0%, above EUR 1bn = 3.1% (Forbes residents 5.6%). Mechanical yield 1.03-1.05% of GDP (1.01% in 2018, 1.13% in 2020). Pooled EU top-1% share 25.9%, threshold EUR 1.71m; top-0.1% share 11.4%, threshold EUR 6.62m. - National top-1% shares are consistent with the cited sources (France ~23%, Spain ~21% excluding offshore, Germany <25%, Scandinavia 21.8%). French inflation of ~50% a year in 1945-48 is correct. - Ten-year repayment: 9.6 years at r = g with zero response. Not reproduced or contradicted: - The claimed allowance for evasion and avoidance: none exists. Under defensible responses revenue is 0.48-0.91% of GDP and repayment takes 11-21 years; the ten-year claim fails beyond a 4.3% loss. - Individual calibration inputs differ from their cited sources and current data (wealth/GDP 4.2-4.7 vs 5; French b 2.2-2.3 vs 2.5; Pareto-consistent top-0.1% share 9.0% vs 10%). The errors offset. - The 330 count is citizenship-based; residence gives 272 billionaires and EUR 0.80tn. - The UK and German historical claims and the cited support are contradicted by the documents checked. - The construction of the EU-wide inputs cannot be reproduced from the column, which has no replication files.

Limits of this audit. The authors published no replication files, so their exact inputs (data years, FX rate, GDP vintage, adult definition) are inferred. The citizenship inference rests on an exact numerical match. The independent base check uses the October 2026 WID release, not the 2020 vintage available to the authors; WID's series for several EU countries are partly imputed. The ECB DWA aggregate excludes non-profit institutions and may differ in its pension coverage. Behavioural-response scenarios borrow elasticities estimated for other countries and designs. They illustrate a defensible range and are not a forecast for an EU-wide tax. The taxpayer-level model uses a stylized two-segment Pareto distribution calibrated to the column's own bases. Hughes (1999) and the published 1990 Eichengreen chapter were not read (the 1989 NBER working-paper version was). Hicks et al. was read in its 1942 second edition. The Landais quote about the Danish evidence is second-hand. Legal feasibility, administrative and valuation costs, and liquidity were not assessed. Several source files came from the shared download cache; they were checked for content but not re-downloaded.