← All papers
Emmanuel Saez, Gabriel Zucman · Economic Policy 37(110): 213-227
A Wealth Tax on Corporations' Stock
12 problems found: 2 moderate · 10 minor
What the paper claims
- A 0.2% annual tax on the market value of listed (and large private) companies headquartered in G20 countries would raise about 0.2% of world GDP, about $180bn in 2019; for the US alone about 0.3% of GDP (about $60bn).
- The tax would be progressive because stock ownership is concentrated: in the US the top 1% of adults by pre-tax income earn 19% of pre-tax income but own about 30% of equity (including via pensions) and pay about 21% of all taxes, so the levy would raise the progressivity of the US tax system; elsewhere it 'could be even more progressive'.
- The tax is easy to administer and hard to avoid (collected by securities commissions; G20 coordination; remedial sales-pro-rated levies on haven-listed or foreign firms), and payment in kind removes liquidity, valuation and business-operation problems, including for private firms worth over ~$1bn.
- Economic effects are small and borne by current owners: the return on stocks falls by 0.2 points, prices fall ~5% once (0.2/4), the capital-structure bias can be neutralised by taxing bonds, and the tax reaches highly valued, low-profit firms (Amazon, Tesla) sooner than profit taxes, complementing the 15% minimum tax.
- Supporting US context: private wealth doubled from ~260% to ~530% of GDP (1975-2020), driven mainly by listed equity (~30% -> >150% of GDP, about half of the rise); listed equity is ~30% of private wealth vs gross housing 35%.
Every problem found
1moderatelikely · newly identified here · methodology
Governments, especially China's and Saudi Arabia's, own a big slice of the companies that would pay this tax. When a state-owned company pays the tax to its own government, no new money is raised: the government is paying itself. Taking this into account cuts the real yield by about one-seventh, although it stays roughly the same size.
Technical detail and evidence
Why it matters, in detail. Gross receipts are 0.2% x MCap. Net new consolidated public resources are 0.2% x (1 - s) x MCap, where s is the taxing governments' ownership share: about 0.137 at end-2019 including Aramco, or 0.125 excluding public pension funds. For 2019 this lowers revenue from $168bn (0.189% of world GDP) to $145bn (0.163%), or $138bn (0.155%) after the paper's own 5% capitalisation; at 2020-21 valuations the net is about 0.19% of world GDP. Only the privately owned (1 - s) share carries the paper's progressivity argument. Under the pooled allocation keys the paper mentions (population, sales), payments by, say, Chinese state firms would become real transfers between governments. For the G20 as a whole they remain circular.
Evidence. OECD (2019) 'Owners of the World's Listed Companies', Annex Table A.5 (end-2017, market-cap weighted), domestic public-sector ownership: China 37%; Hong Kong 37% (mainland public owners, counted here as domestic for China); Saudi Arabia 45%; Russia 31%; Indonesia 18%; India 16%; Japan 9%; US 2%. Weighted over the G20 universe (19 members + other EU + Hong Kong, $78.2tn), this gives 11.8%, or 10.6% excluding public pension funds (Table A.6 split). Adding Saudi Aramco (about $1.88tn at end-2019, 98.5% state-held after its December-2019 IPO) raises it to 13.7% of the end-2019 WFE base of $84.0tn. Net of these circular payments, 2019 revenue is $145bn (0.163% of world GDP) rather than $168bn gross. At 2020-21 valuations it is about 0.19-0.20% of world GDP. For Aramco alone, the gross levy is $3.8bn but only about $56m moves from private holders. Scripts: state_ownership_net_revenue.py, headline_reconciliation.py. Auditor A's figures are reproduced exactly.
Why this severity. Lowers the economically meaningful (net) revenue by about one-seventh at any valuation, and removes the progressivity rationale for that slice of the base. The order of magnitude survives (0.16-0.20% of world GDP net, 2019-21), well short of the quarter-of-headline threshold for 'major'.
2moderateplausible · previously known · speculative assumption
The paper simply multiplies the tax rate by today's stock-market value. It does not ask how many companies might move abroad, stay private or reorganise to pay less. And the 5% cut it calls a 'behavioural' adjustment is really just the expected drop in share prices. The true yield could be somewhat lower; nobody knows by how much.
Technical detail and evidence
Why it matters, in detail. Each 1% of base erosion cuts revenue by 1%. The paper's only adjustment is the capitalisation term tau/(y + tau), about 5%, which is a price effect on an unchanged base. Elasticities estimated for personal wealth taxes (Kopczuk 2019; Advani and Tarrant 2021; Brulhart et al. 2022) do not transfer directly to a 0.2% levy on third-party-observed listed values, so erosion is probably smaller than for individual wealth taxes. But the relocation, delisting, restructuring and debt-substitution margins are real and unestimated.
Evidence. Paper pp. 216, 219, 222, 225, 226. WFE Annual Statistics Guide 2019, sheet 'Equity 1' (g20_base_wfe.py). OECD (2019) Table A.3: 'private corporations' own about 10% of the G20 base, an upper bound on listed cross-holdings since the category includes unlisted holding companies. Authors' CEPR panel slides, 14 April 2021 ('Creates incentives to remain private'). arith_incidence_checks.py.
Why this severity. The estimate is explicitly mechanical, and the design (G20 coordination, a remedial levy on haven-listed firms, a private-firm extension) closes the most obvious channels, so a large erosion is not demonstrated. But nothing in the paper bounds the response, so the net yield is uncertain by an unknown and plausibly non-trivial margin.
The authors' response. The paper says it focuses on 'mechanical revenues' (p. 222). It relies on G20 coordination, remedial pro-rated levies on haven-listed firms (p. 225) and taxation of large private firms (p. 220) to limit avoidance. No direct response to these points was located.
3minorconfirmed · newly identified here · internal inconsistency
The paper puts the size of the G20 economies in 2019 at $100 trillion; the true figure is about $76 trillion. Its estimate of the G20 stock market is also about 7% too high. The two mistakes roughly cancel, so the main claim (about 0.2% of world output) still holds, but readers cannot reconcile the paper's different statements.
Technical detail and evidence
Why it matters, in detail. Base over G20 GDP is 110% in 2019, not 90%, and about 129-131% in 2020-21; revenue over G20 GDP is 0.22-0.26%, not 0.18-0.2%. Base over world GDP is 95-114%, so revenue over world GDP is 0.19-0.23%. The $90tn overstates the 2019 WFE G20 base (including the EU and Hong Kong) by 7%. The +31% error in G20 GDP and the +7% error in the base push the G20-GDP ratio in opposite directions, so the world-GDP headline is unaffected.
Evidence. WFE Annual Statistics Guides 2019, 2020 and 2021, sheet 'Equity 1' (g20_base_wfe.py). World Bank WDI NY.GDP.MKTP.CD, NY.GDP.MKTP.PP.CD and CM.MKT.LCAP.GD.ZS, October-2026 vintage (g20_gdp_wdi.py). WFE FY2019 highlights, p. 3. WFE H1-2021 release ('surpassing the 116 USD [trillion] mark'). Drafts of 11 April 2021 and 18 July 2021. Saez and Zucman (2024), Oxford Open Economics 3(S1): i1130-i1132.
Why this severity. The errors partly offset. Correcting them leaves the summary's headline intact (about 0.2% of world GDP: 0.19% in 2019, 0.22-0.23% in 2020-21). The 2019 dollar figure is only about 7% too high ($168bn vs $180bn), and the share of G20 GDP comes out higher, not lower (0.22%). This is a factual error in a supporting denominator, not in the conclusion.
4minorconfirmed · newly identified here · data error
For the US, the paper's method understates what the tax would raise. Using the actual value of US-listed companies gives about $75 billion for 2019 and over $90 billion for 2020, not $60 billion. This error makes the proposal look less lucrative than it is.
Technical detail and evidence
Why it matters, in detail. On the statutory year-end issuer base, US receipts are 0.34-0.36% of GDP in 2019 and 0.42-0.44% in 2020-21, against the paper's roughly 0.3%. Part of the gap is the switch from midyear household holdings to the year-end issuer base. The rest is that 'slightly higher than 150%' understates intercorporate, foreign and institutional holdings, plus the 2019-20 rally.
Evidence. Fed Z.1 release of 23 September 2021, table L.223 line 29 (LM883164115.Q) (z1_equity.py). WFE guides (headline_reconciliation.py). WDI US GDP. BEA international investment position tables (Survey of Current Business, April 2020) confirm near-balanced cross-border portfolio equity at end-2019: $9.51tn of assets vs $9.27tn of liabilities.
Why this severity. The error is conservative: correct figures raise US revenue by 20-55%. It strengthens rather than undermines the revenue claim, and amounts to a calibration and documentation inconsistency.
5minorconfirmed · previously known · speculative assumption
Because the tax is a share of stock-market value, its yield rises and falls with share prices. In a crash like 2008 it would bring in only about half as much. The paper does not mention this.
Technical detail and evidence
Why it matters, in detail. Revenue is proportional to MCap/GDP. Scaling the 2019 gross yield of 0.189% of world GDP by the World Bank world ratio gives 0.156% at the 2000-18 mean (x0.83), 0.099% at the 2008 low (x0.52) and about 0.215% at the 2020-24 mean (x1.14). A yield that halves in crises is less suited to financing fixed commitments than the paper's framing implies.
Evidence. World Bank WDI CM.MKT.LCAP.GD.ZS, world aggregate, 1975-2024 (valuation_history.py). The WFE-based 2019 base was 95% of world GDP (headline_reconciliation.py). The World Bank world aggregate covers reporting countries only (France and the UK are missing in several recent years), so only its relative movements are used.
Why this severity. The headline is explicitly conditional on market cap being about 100% of world GDP, and valuations since 2019 have mostly been at or above that level (2022 slightly below). The omission concerns volatility and representativeness. It moves the long-run average yield by about a fifth, not the order of magnitude.
6minorconfirmed · newly identified here · methodology
The paper says that also taxing corporate bonds would stop firms from switching from shares to borrowing to avoid the tax. But firms can also borrow from banks, so bank loans would have to be taxed too.
Technical detail and evidence
Why it matters, in detail. The levy raises the cost of equity relative to all untaxed debt by about 0.2 percentage points, on top of the interest deductibility the corporate income tax already gives debt. A bond-only extension removes the wedge only for bond finance and leaves loan and private-credit substitution open as an avoidance channel.
Evidence. Paper pp. 221-222. Auditor B's reproduce.py (untaxed-loan example).
Why this severity. This is a conditional design claim about a hypothetical extension. The debt margin is acknowledged, and its revenue effect is unquantified; it is one of the avoidance channels listed under F2.
7minorconfirmed · newly identified here · overclaiming
The paper says the new global minimum tax on company profits would not touch firms like Amazon that make little profit compared with their value. In fact such firms are covered; they just pay little because their profits are small. The paper's broader point still stands.
Technical detail and evidence
Why it matters, in detail. The Pillar Two top-up equals (15% minus the jurisdictional effective tax rate) x excess profit, and is positive in any in-scope low-taxed jurisdiction. It is small when profits are small, which is the paper's substantive point, but it is not zero by design.
Evidence. OECD, Statement on a Two-Pillar Solution, 8 October 2021 (cached text): Pillar One scope 'profitability above 10% (i.e. profit before tax/revenue)'; GloBE EUR 750m threshold and 15% minimum rate. Auditor B's reproduce.py (stylised top-up example).
Why this severity. A wording overstatement in the motivation; the argument that the two taxes complement each other survives.
8minorconfirmed · newly identified here · arithmetic or transcription
A few small calculation and factual errors appear in side remarks; for example, the paper says 73% where 63% is correct. They do not change its conclusions.
Technical detail and evidence
Why it matters, in detail. The dilution figure for a 1% rate is overstated by 9.6 percentage points and the fee figure by about $8bn. Neither enters the proposal's revenue or incidence calculations.
Evidence. arith_incidence_checks.py. Morningstar press release, 30 April 2019: asset-weighted fee 0.48%; $5.5bn saved from a 3-basis-point decline. DIAN Normograma, Ley 2010 de 2019, art. 43 (downloaded 25 October 2026). Paper pp. 216, 218-220. Drafts of 11 April and 18 July 2021.
Why this severity. Illustrative side calculations and a footnote; no effect on the revenue or progressivity findings.
9minorlikely · previously known · methodology
The paper assumes the whole cost falls on today's shareholders. In practice some of it could pass to workers if firms invest less, and share prices could fall by more than the 5% the paper uses. Even so, under standard assumptions the tax would still fall mainly on the rich; it would just be somewhat less tilted toward them.
Technical detail and evidence
Why it matters, in detail. If labour bears a fraction lambda of the burden, the top-1% share is (1 - lambda) x 0.311 + lambda x 0.094: 25.7% at lambda = 0.25 and 20.3% at lambda = 0.5, against an 18.7% income share and a roughly 21% current tax share. The conclusion that the levy raises US tax progressivity fails only if lambda exceeds about 0.46. The capitalisation loss to enactment-date owners ranges from about 4.8% (y = 4%, total payouts including buybacks) to about 9-13% (dividend-only yields of 1.3-2%), i.e. about $4-11tn on the 2019 G20 base. Revenue falls by the same percentage.
Evidence. Paper Summary p. 213; pp. 219, 221, 224-225. arith_incidence_checks.py: capitalisation sensitivity, and incidence mixtures built from DINA September-2020 cells for 2019: TB15 top-1% equity share 31.1%; TB1 top-1% income share 18.7%; TB2b top-1% labour income 6.6% of national income, i.e. about 9.4% of all labour income if labour is 70% of national income. Auditor B's reproduce.py (net-present-value example).
Why this severity. Under the conventional alternative assumption (25% of the burden on labour, as CBO and JCT assume for the corporate tax), the top-1% burden share falls from about 31% to about 26%. That is still well above their 19% income share and their roughly 21% share of current taxes. The tax remains progressive and its economic effects small; only the degree is affected.
The authors' response. Saez and Zucman's general position (corpus us-wealth-taxes-13; 'Clarifying Distributional Tax Incidence', 2019; NBER w31912) is that tables of current tax burdens should assign each tax to the factor taxed, and that general-equilibrium effects belong in reform analysis. This paper is a reform proposal, yet it relies only on capitalisation.
10minorlikely · previously known · overclaiming
Letting private companies pay in shares avoids arguing over their exact value. But someone still has to decide whether a company is worth more than $1 billion, and small slices of private firms are hard to sell for full value. So taxing private firms is harder than the paper suggests, though this does not affect its main revenue figure.
Technical detail and evidence
Why it matters, in detail. Valuation still enters at two points: at the extensive margin (the threshold) and on realisation (resale of minority stakes at marketability and minority discounts). The in-kind option therefore shifts the valuation problem rather than removing it. Any 'additional revenue' from private firms (p. 222) should carry a discount and an avoidance margin around the threshold.
Evidence. Paper pp. 216, 218, 220. Kopczuk (2019), comment on Saez and Zucman, Brookings Papers on Economic Activity, Fall 2019, pp. 519-520: 'the notion that small stakes in a closely held company are cheaply marketable and correctly valued seems dubious'. Corpus assessment of offshore-havens-20 (payment in kind 'solves liquidity at the cost of revenue'). offshore-havens-23 (Skar et al. 2025: Norwegian statutory values have a median of 61% of sale prices). Auditor B's reproduce.py (threshold notch).
Why this severity. Private firms are excluded from the paper's revenue estimate (p. 222), so the headline is unaffected. The overstatement concerns secondary design claims.
The authors' response. Saez and Zucman (BPEA 2019, pp. 482-483) argue that financial-industry valuations of large private firms should be made reportable, and that where valuation is disputed owners should pay in stock so the government can 'create the missing valuation market' ('the best solution'). For small firms they propose 409A- or Swiss-style formulas. This paper repeats the in-kind idea and adds notional government stakes (p. 220). No reply to the threshold point was located.
11minorplausible · newly identified here · overclaiming
The paper shows the tax would hit rich Americans hardest, but it simply assumes the same holds in the other G20 countries. In countries like China and Saudi Arabia much of the stock market is owned by the government, so 'taxing the rich' does not describe that part.
Technical detail and evidence
Why it matters, in detail. C2 is demonstrated for 43% of the base: in the US the top-1% equity share is 31% against a 19% income share in DINA, and 30.9% against 19.1% in the SCF. For the other 57%, about a quarter is state-held and pension and corporate-ownership structures differ, so the G20-wide distribution of the burden is unknown.
Evidence. Paper Summary and pp. 224-225. OECD (2019) Annex Table A.3 (end-2017). WFE 2019: the US was $36.3tn of the $84.0tn base (43%). Auditor A put the non-US share at 64%; it is 57% on WFE end-2019 data and 61% on OECD end-2017 data.
Why this severity. Equity ownership is concentrated in most countries, so the direction is plausible for privately held equity, and the US conclusion is robust. The problem is an undemonstrated extrapolation, not evidence that the tax would be regressive.
12minorplausible · previously known · overclaiming
If not every big country joins, the paper's backup plan taxes foreign companies according to their sales in the countries that do. Companies can shift where they book sales, and tax treaties may block such levies, so the backup plan is weaker than the paper suggests.
Technical detail and evidence
Why it matters, in detail. Under partial coalitions the pro-rated base (sales share x market value) can be manipulated by booking sales in non-taxing jurisdictions. Its legal form (a capital tax or an income-tax surcharge) determines its exposure to treaties. The paper analyses neither, so its claim that remedial levies make escape impossible is unproven.
Evidence. Paper pp. 225-226. Laffitte & Toubal (2022), AEJ: Economic Policy 14(4): 371-396 (corpus offshore-havens-15): about $80bn of US multinationals' profit associated with sales booked in havens in 2013. OECD Model Tax Convention, Arts. 2, 7 and 22 (cached text). The treaty point is the auditors' and adjudicator's own reading of the Model; it was not checked against specific bilateral treaties.
Why this severity. Concerns the fallback mechanisms for incomplete coordination, not the coordinated base on which the revenue estimate rests.
Claims examined and rejected
Problems one auditor raised that did not survive checking.
- Intercorporate holdings (about 10% of the G20 base) mean net new revenue is overstated (part of Auditor A's I1). — Payments on listed-to-listed cross-holdings are real payments ultimately borne by private owners: a cascade or double tax, which the paper acknowledges for the US (p. 222), not a circular payment. OECD's 'private corporations' category (10%) also includes unlisted holding companies, so it overstates double counting within the listed base. The point is kept only as an avoidance and distortion margin (F2).
- The 2019 net revenue ($138bn) is 23% below the paper's $180bn, which measures the state-ownership problem. — Accepted as arithmetic, rejected as a measure of F1. The 23% combines three things: the 7% base error (F3), the paper's own disclosed 5% capitalisation adjustment (p. 222), and the 13.7% state-owned share. Only the last is a new correction, and at 2020-21 valuations net revenue is about 0.19% of world GDP.
- Passive index funds charge 0.15% (Morningstar), not 'less than 0.1 percentage points' (footnote 4). — The paper refers to 'low-cost passive index funds'. Morningstar's 0.15% is the asset-weighted average of all passive funds, and the same Morningstar release reports Vanguard's asset-weighted average at 0.09% in 2018.
- Even in the US about 70% of the burden falls outside the top 1%, so 'highly progressive' (p. 217) should be read in that light (part of Auditor A's I6). — Burden shares exceed income shares at the top and fall below them at the bottom (SCF 2019 equity vs income shares: top 1% 30.9% vs 19.1%; P80-99 51.4% vs 41.9%; P0-80 17.7% vs 39.0%), so the burden is progressive throughout. The share falling outside the top 1% is not a measure of progressivity.
- The article dropped caveats from the authors' April-2021 slides ('creates incentives to remain private'; 'less progressive than a well-enforced progressive individual wealth tax'). — Section 3.3 explicitly addresses the incentive to stay private, by extending the tax to large private firms. The article never claims parity with a progressive individual wealth tax; it presents the levy as an 'alternative form of wealth taxation' that 'remains a progressive tax' (pp. 217-218).
- The part of the G20 base outside the US is 64%. — A numerical slip: the non-US share is 57% on WFE end-2019 data (US $36.3tn of $84.0tn) and 61% on OECD end-2017 data. Corrected in F6; the substance of the point is unaffected.
Audit notes
What reproduced. Reproduced. (1) G20 base from WFE exchange-level data: $84.0tn (2019), $98.3tn (2020), $110.1tn (2021); WFE world total $91.7tn (2019). (2) G20 GDP 2019: $76.3tn ($76.6tn with Hong Kong); world $88.8tn; the paper's $100tn matches no market-exchange-rate aggregate but is close to G20 PPP GDP ($101.5tn). (3) Gross revenue: $168bn = 0.189% of world GDP and 0.22% of G20 GDP in 2019; 0.227% and 0.223% of world GDP in 2020-21; the paper's $180bn is 7% high for 2019. (4) Net of domestic state ownership (OECD end-2017: 11.8%; 13.7% with Aramco): $145bn (0.163% of world GDP) in 2019, about 0.19-0.20% in 2020-21; $138bn with the paper's 5% capitalisation. (5) US: year-end public equity $38.5tn (Fed) / $36.3tn (WFE) in 2019, giving $73-77bn (0.34-0.36% of GDP) against the paper's ~$60bn; $91-94bn in 2020. (6) DINA cells for income, equity and tax shares match the paper exactly; the SCF 2019 corroborates them (30.9% vs 19.1%). (7) Figure 1 levels and the 'about half' decomposition match the authors' aggregates. (8) Arithmetic: 63.4% not 73%; $81.6bn not $90bn; 18.1% not 20%; the other examples are exact. (9) Capitalisation of 4.8% at a 4% payout yield (9.1% at 2%); incidence sensitivity puts the top-1% burden share at 25.7% if 25% falls on labour. Not reproduced: an exact headquarters-based G20 base (exchange-location data are a proxy); the authors' exact Figure 2 workbook (none published); any behavioural response (none estimable); the July-2021 Forbes snapshot. All code is in final/scripts (run_all.sh); output is in final/work/run_all_output.txt.
Limits of this audit. No replication package exists. The WFE exchange-level data only approximate a headquarters-based base: dual and inward listings (e.g. the JSE), exchange groups (Euronext, LSE Group) and non-G20 listings of G20 firms are not netted. The OECD ownership shares are for end-2017 and the 10,000 largest firms; the 2019-21 net figures extrapolate them with a single Aramco adjustment, and treating public pension funds as circular is a judgement (worth about 1 point). The World Bank world market-cap ratio covers reporting countries only, so only its relative movements were used. The incidence sensitivity uses stylised labour shares of the burden and an assumed 70% labour share of national income, not an incidence model. Behavioural responses could not be quantified. Treaty and inversion points were not checked against specific treaties or transactions. The Colombia finding rests on the 2020-21 statute (Ley 2010 de 2019) only. Most data files were downloaded earlier by the auditors and re-read here. New downloads were the World Bank long-run series, the ITEP 2019 report and the Colombian statute.