Wealth Taxation: Lessons from History and Recent Developments

Why it matters, in detail. On recomputed values actual/benchmark becomes ES 25% (printed 19%), NO 95% (printed 50%; 100% with the current OECD figure of 0.451% of GDP), FR 27% (unchanged) and CH 89% (printed 76%). The four-country benchmark mean falls from 1.00% to 0.79% of GDP, which also puts the stylized 'about 1 percent of GDP' at the top of the range. The two countries used to generalize erosion beyond high-threshold taxes show essentially none on the paper's own metric, so the cross-country evidence for C1 reduces to France and Spain. The Spanish and Norwegian discrepancies lie outside all vintage and year variation. The Swiss one could be a vintage effect or a national-for-household wealth substitution.

Evidence. Adjudicator scripts table1_recompute.py, table1_search.py, table1_bulk_vintage.py (run_all_output.txt, sections A1-A3). Snapshot of 25 Oct 2021 (wid-data-25102021.dta inside the WIR2022 data.zip), 2016, revenue = 1% x 1% x adults x (mean above P99 - P99 threshold) / GDP; the share-based and wealth-income-ratio routes agree to four decimals: ES 0.710, NO 0.452, FR 0.827, CH 1.162 (printed 0.96, 0.86, 0.83, 1.36). Same snapshot, every year 1995-2020: maximum ES 0.749 (2007), NO 0.514 (2018), CH 1.329 (2020). Nov-2023 WID bulk vintage (Wayback copy, Last-Modified 21 Nov 2023), 2016: ES 0.75-0.80, NO 0.37-0.45, FR 0.79-0.83, CH 1.11-1.13. Oct-2026 vintage, 2016: ES 0.74, NO 0.46, FR 0.80, CH 1.26 (CH 1.34 for 2017, 1.38 for 2020). Search over household, private and national wealth, over GDP or net national income, 2005-2020, with WID or Pareto-2/3 excess shares: no combination reproduces Norway; Switzerland is matched by national (private plus public) wealth/GDP in 2016 (1.361); Spain only by an inconsistent combination (2012, national wealth/NNI, stylized 2/3). Norway's printed value needs household wealth of 4.93x GDP vs 2.39-2.68x in 2015-17 in all vintages; substituting France's 2016 ratio (4.937) gives 0.862, which suggests (but does not prove) a lookup slip. Both auditors independently obtained the same snapshot values (Claude I1: 0.71/0.45/0.83/1.16; Codex I2: 0.70956/0.45221/0.82650/1.16250).

Why this severity. The error is in the exhibit the paper presents as its 'main finding'. It overstates the theoretical revenue by about 35% (Spain), 90% (Norway) and 17% (Switzerland), which wipes out Norway's measured shortfall (50% -> 0-5%) and halves Switzerland's (24% -> 11%). It does not overturn the narrow-base conclusion, which France, Spain and independent statutory-schedule tests (F2) support.

Why it matters, in detail. R_actual/R_benchmark conflates the statutory schedule and threshold with erosion. A broad low-rate tax can match 1% x sum(w - P99)+ while collecting a small share of its own mechanical yield, and a steep high-threshold tax falls short of its own yield by more than the table shows. Holding schedules fixed, collection ratios are NO 32%, FR 12-18%, ES 15-17% and CH roughly 60-82%, against corrected Table 1 ratios of 95%, 27%, 25% and 89%. Because the bias differs by country, Table 1's ordering of countries does not rank their erosion.

Evidence. Adjudicator script statutory_schedules.py, an independent re-implementation of Claude's statutory_base_erosion.py with identical results: each country's 2016 statutory schedule applied to WID market-value wealth (g-percentile averages, equal-split adults). Norway, 0.85% above NOK 1.4m: 1.35% of GDP (43% of adults above the threshold) vs 0.43 actual = 32%. France ISF: 1.20-1.77% (individual vs couple units) vs 0.22 = 12-18%. Spain state scale: 1.06-1.17% vs 0.18 = 15-17%. A rough Zurich-type Swiss schedule: 1.26-1.72% vs 1.03 = 60-82%. Codex reproduction 8: the same top-1% base taxed at Switzerland's printed rates of 0.2/0.35/0.5% yields 0.27/0.48/0.68% of GDP, all below actual receipts of 1.03%. OECD (2018), p.18: cross-country differences reflect 'the design of the tax (e.g. taxed assets, tax schedule and rates, exemption thresholds ...)'.

Why this severity. Even with correct inputs, Table 1 would show Norway with no erosion (95-100%) although its own schedule implies about 68%, and it understates erosion in France and Spain. That limits what the exhibit can show, but tests that hold each schedule fixed confirm, and strengthen, the narrow-base conclusion.

Why it matters, in detail. In the Scandinavian cases, erosion came mainly through legal valuation (Norwegian taxable net wealth is about 23% of WID market-value household wealth, largely because of discounted housing values and debt deductibility) and exemptions. Behavioural responses persist even with third-party reporting: JJKZ estimate taxable-wealth effects of 19-31% after 8 years. Third-party reporting is therefore necessary but not sufficient. Enforcement bites on the self-reported residual (cash, unlisted equity, offshore assets), consistent with AJZ.

Evidence. JJKZ (2020, QJE 135(1)), pp.332-333: 'Most assets were third-party reported under the Danish wealth tax—thus limiting the scope for evasion—but some were self-reported'; p.336: deposits reported by banks, listed stocks and bonds by financial intermediaries, real estate from registries. Advani & Tarrant (2021, Fiscal Studies 42(3-4)), Table 2, pp.514-515: third-party reporting 'Yes' for Denmark, Norway and Sweden; partial for Spain (dwellings, bank-held financial assets); 'None' for Switzerland and the Netherlands. OECD (2018), p.89: filing 'generally based on self-assessment'; in Norway and Switzerland the wealth return is consolidated with the income tax return. Narrow bases despite reporting: Norway collected 32% of its schedule's mechanical yield (F2); Denmark (1996) and Sweden (2006) raised 24% and 41% of the WID benchmark (repealed_and_count.py); SSB table 08815 shows taxable net wealth of NOK 1.82tn in 2016, against WID market-value household wealth of about NOK 8.0tn.

Why this severity. Weakens the paper's third explanation and the implied promise that information reporting fixes erosion. The Scandinavian taxes eroded despite third-party reporting, through statutory valuation, exemptions and behavioural responses. Measured narrowness is unaffected.

Why it matters, in detail. The paper infers political causation from the existence of exemptions, with no counterfactual. For France, the two mechanisms it highlights are an exemption present since 1982 and a ceiling that the Conseil constitutionnel required for an ISF schedule like the pre-2012 one, and whose income base may not include retained corporate earnings. That retained-earnings channel is the one the paper blames for the 0.3% effective rate. Under OECD MTC Art. 22(4), a resident's other capital is taxable only in the residence state, so taxing emigrants requires treaty change; inside the EU, exit taxes are constrained by freedom of establishment (C-9/02). Responses are documented even under third-party reporting. These are design constraints, not proof that a well-designed tax cannot work.

Evidence. Paper p.61. OECD (2018), p.17: repeal justified by 'efficiency costs and the risks of capital flight', narrow bases with avoidance and evasion, and 'high administrative and compliance costs'. Bach et al. slides (7 June 2022), slide 13: 'Professional wealth never included in the tax base'; 'Tax ceiling as a legal requirement of Constitutional Court'. Conseil constitutionnel 2012-654 DC (9 Aug 2012, cons. 33), quoted in Sénat Rapport n° 689 (2024-25), part IV.A: the legislator 'ne saurait établir un barème de l'impôt sur la fortune tel que celui qui était en vigueur avant l'année 2012 sans l'assortir d'un dispositif de plafonnement ou produisant des effets équivalents'; confirmed in 2012-662 DC (cons. 93). Bach, Bozio, Guillouzouic & Malgouyres (2023, IPP Note 92): 2012-662 bars extending the ceiling's income concept to income the taxpayer does not freely dispose of. OECD Model Tax Convention (2010), Art. 22(4): 'All other elements of capital of a resident of a Contracting State shall be taxable only in that State'. CJEU C-9/02, de Lasteyrie du Saillant (11 Mar 2004): French exit taxation of unrealised gains infringes freedom of establishment. Behavioural responses: JJKZ (2020), 19-31% after 8 years; Brülhart, Gruber, Krapf & Schmidheiny (2022, AEJ:EP), semi-elasticity of at least 43.

Why this severity. This interpretive conclusion frames the paper's policy lesson. It is overstated rather than refuted: constitutional, treaty, EU-law and behavioural constraints make 'other choices' costlier than the paper implies. The measured narrowness of the bases is unaffected.

The authors' response. Within the paper: 'While these choices were sometimes supported by economic arguments, there is no fundamental economic reason why billionaires should be exempted from wealth taxes, nonresidents untaxed, and enforcement weak' (p.61). No direct reply to these specific points located.

Why it matters, in detail. Revenue scales with (1 - avoidance) times a DINA/SCF/Forbes base. The simulator gives $203bn (about 0.98% of 2018 GDP) at 15% avoidance; at 40% it gives $143bn (-29%) and a top-400 rate of 39.2%. BPEA's own bases with its 17.7% haircut give 0.62-0.93% of 2019 GDP, with the estate-multiplier base at the low end. PWBM-type rate-scaled elasticities cut the ten-year score by about a third. The 23% baseline is corroborated by later administrative data co-authored by Saez and Zucman (Balkir, Saez, Yagan & Zucman 2025: 23.8% for 2018-20), so the dispute is about the increment, not the baseline.

Evidence. Saez & Zucman letter to Sen. Warren (18 Jan 2019), p.1: 'approximately 1.0% of GDP per year ($210 billion relative to a $21.1 trillion GDP in 2019)'. BPEA (2019a), pp.438-439: a 1% tax on the top 0.1% would raise about $120bn ('about 0.6 percent-0.7 percent of national income'); Table 2, p.461: base above $50m of $10.9/9.4/9.7/6.8tn (capitalized/revised/SCF/estates) and a 17.7% base reduction under strong enforcement, implying about 0.62-0.93% of 2019 GDP including the billionaire surtax (Claude R17); p.468: top four hundred '23 percent to 46 percent' with 15% hidden. Simulator (taxsimulator.xlsx, sheet data, row gperc=400): 23.04% + 0.85 x (2% x $4.25bn + 1% x $3.77bn) / $456.6m = 45.9% (figure_checks.py). Codex sensitivity on the same simulator: avoidance of 0/15/25/40/50% gives $238.7/202.9/179.0/143.2/119.3bn and top-400 rates of 49.9/45.9/43.2/39.2/36.5%. Alternative evidence: Penn Wharton Budget Model (2019), semi-elasticity -13 vs -8; Kopczuk (2019), estate-tax-implied elasticity of about 20; Advani & Tarrant (2021), UK range 7-17; Brülhart et al. (2022), at least 43 in Switzerland; Summers & Sarin (2019), about $25-75bn vs $212bn.

Why this severity. The US numbers are secondary to the European analysis and explicitly conditional ('If well-enforced'). But the conditioning assumption is not disclosed, the stated revenue is at the top of the authors' own range, and credible alternative elasticities would cut it by roughly a third or more.

The authors' response. Saez & Zucman, 'Response to Summers and Sarin' (25 June 2019), and BPEA (2019a): avoidance is a policy choice; the Warren design (no exemptions, third-party reporting, minimum audit rates, 40% exit tax) differs from the estate tax and from past European taxes; Swiss responses reflect bank secrecy and cantonal mobility. The P&P conditions the result on the tax being 'well-enforced'.

Why it matters, in detail. A Spanish adult at P99 (about EUR 1.0m net wealth) faces a marginal rate of about 0.3% on the state scale (zero if the EUR 300k primary-home allowance applies), not 1%. Madrid's 100% rebate means part of Spain's measured erosion is a statutory zero rate produced by regional tax competition (Agrawal, Foremny & Martínez-Toledano 2024, AEJ: Applied). Confirmed for Spain's rate range; the Norwegian and Spanish coverage entries cannot be checked on the household unit the column claims, so they are undocumented rather than shown to be wrong.

Evidence. AEAT, Manual Práctico Renta y Patrimonio 2016, ch.20: p.1053 (state scale 0.2, 0.3, 0.5, 0.9, 1.3, 1.7, 2.1, 2.5%); pp.1054-1056 (regional scales, e.g. Catalonia 0.21-2.75%); p.1065 ('Comunidad de Madrid: Bonificación general del 100 por 100'). OECD (2018): p.87 (Swiss top rates 0.13-1.0% in 2012; Spain top 2.5%); pp.81-82 (France 351,152 tax households; Spain 188,680 taxpayers in 2015; Norway 'about 11% of taxpayers in 2016'). SSB table 08815 (2016: 513,532 payers of 4,448,444 persons with net wealth).

Why this severity. Descriptive entries that do not enter the revenue comparison. The Spanish range does, however, overstate the rates most of the Spanish top 1% faced and hides the zero rate in Madrid.

Why it matters, in detail. The French potential is EUR 17.9bn (0.83% of GDP) and the shortfall factor is 3.5, not 4. With 2016 wealth/GDP ratios (4.9-5.3), the stylized calculation gives 0.81-0.88% of GDP rather than 1%. The Pareto algebra (a = 1.5 implies mean = 3 x threshold, so 2/3 of top-1% wealth lies above the threshold) is correct.

Evidence. france_and_us_checks.py (WID 25-Oct-2021 snapshot; nominal 2016 euros via WID's deflator): household wealth EUR 10,684bn; top-1% share 25.41%; mean/threshold 2.931; 65.9% of top-1% wealth above P99; a 1% tax yields EUR 17.9bn = 0.827% of GDP, 3.54 times the ISF yield of EUR 5.051bn (DGFiP; OECD 2018 p.81). EUR 20bn / INSEE 2016 GDP of EUR 2,228.9bn = 0.90%. Claude R6 reports the same figures. Codex checked only the internal arithmetic (EUR 3tn x 2/3 x 1% = EUR 20bn; 20/5.1 = 3.92).

Why this severity. Presentation and rounding. The intended arithmetic is correct, and Table 1's France entry (0.83) is exact.

Why it matters, in detail. The benchmark is 25% of tax liability for the top 0.01%, in a 20-30% range; 20% is the lower bound, or the share of wealth hidden. The external validity of the Scandinavian concentration has been questioned (Leenders et al. 2023: about 8% for the Dutch top 0.01%), but that bears on AJZ, not on this citation.

Evidence. Alstadsæter, Johannesen & Zucman (2019, AER 109(6)): abstract (p.2073), 'the 0.01 percent richest households evade about 25 percent of their taxes'; p.2093, 'the top 0.01 percent evades 25 percent of its true tax liability'; p.2094, 'For all plausible scenarios, it is in a range of 20 percent to 30 percent'; fn.32, 'the top 0.01 percent concealed 20 percent of its total wealth offshore'. Saez & Zucman (2019a, BPEA) also cite 'about 25 percent'.

Why this severity. A conservative slip that cuts against the authors' own argument and inflates nothing.

Why it matters, in detail. The tenfold claim refers to the top 0.00001%: 0.1325% in 1982 (authors' JEP workbook) vs about 1.43% in 2021 (figure), about 10.8x. For the income-ranked top 400 the Warren increment is 22.8 pp; a wealth-ranked group with a lower income-to-wealth ratio would show a larger increment, so the label does not flatter the result.

Evidence. Paper p.58 text vs Figure 1 title and notes; 18 / about 180m x 100 = 0.00001%. Figure 2 notes ('groups of the pretax income distributions'); taxsimulator.xlsx, sheet data, row gperc=400 (thres = 416,000,000; hweal = 4.27bn; figure_checks.py). BPEA (2019a, p.440; see also p.467) uses 'top four hundred richest Americans' for the same income-ranked group. Typos on p.61.

Why this severity. Labelling only. The figure, its notes and the count of 18 agree with one another, and the Figure 2 result is unchanged by the label.

Why it matters, in detail. The capitalized-income (SZ/DINA) series is the high end of US estimates. SCF and DFA increases since 1989 are about half as large, estate-multiplier series barely move from the 1970s to the 1990s, and the trend depends on the wealth concept. The Forbes-based Figure 1 does show roughly a tenfold rise at the very top.

Evidence. Kopczuk (2015, JEP 29(1)): methods agree until the mid-1980s, after which only capitalization shows a dramatic rise. Bricker, Henriques, Krimmel & Sabelhaus (2016, BPEA): the SCF rise in the top-1% share is less than half of Saez-Zucman's. Batty et al. (2019, FEDS 2019-017): DFA top 1% about 31% vs about 37% in WID (2014), with a more gradual rise. Smith, Zidar & Zwick (2023, QJE). Catherine, Miller & Sarin (2025, JF): including Social Security wealth, the 1989-2019 rise in the top-1% share is +1.9 pp instead of +6.4 pp.

Why this severity. A motivational sentence; the paper's analysis does not depend on the size of the US trend.

The authors' response. Saez & Zucman (2020, JEP 34(4)) harmonize DFA and SZ concepts and report a similar rise since 1989; the P&P points readers to that reconciliation.

Why it matters, in detail. If business assets are about 65% of top wealth, the exemption alone scales the base by about 1/(1-0.65) = 2.9; the further fall to 0.1-0.15% effective comes from the ceiling and other reliefs. The '0.3% even if taxed' counterfactual cannot be checked from the slides, which report post-ceiling rates as a share of comprehensive income, not of wealth. Codex likewise could not reproduce the factor of 10 but declined to infer a contradiction across group definitions and versions.

Evidence. Bach, Bozio, Guillouzouic & Malgouyres, 'Do billionaires pay taxes?', slides dated 7 June 2022 (https://eml.berkeley.edu/~saez/course/bachetal22slides.pdf): slide 4 ('For the top 0.0001%, 65% of wealth is business assets'; wealth tax 0.02% of wealth for the top 0.0001%); slide 8 (business vs gross taxable wealth by g-percentile); slide 15 (wealth tax as % of wealth: about 0.07-0.20 in the top 0.001% bins, 0.02 at the very top); slide 13 (ceiling at 75% of taxable income). IPP Note 92 (2023): personal taxes are about 2% of economic income for the 378 richest households, consistent with the paper's '2-3 percent'.

Why this severity. The combined effect the paper describes (1.5% statutory -> about 0.1-0.15% effective; personal taxes 2-3% of economic income) is consistent with the available evidence. Only the split between exemption and ceiling is in question, and the cited version cannot be checked.