A Modern Excess Profit Tax

Why it matters, in detail. The apportioned base for imputed firms is sum(g_i) x 0.10, but the key the design calls for is the gain-weighted EU sales share of those firms. Non-EU domestic utilities make up 33% of their gross, and much of the rest is North American onshore producers and Asian state-owned domestic suppliers. A weighted share of about 2-7% (central ~4%) is therefore more plausible than 10%. At 33%, revenue is overstated by EUR 6-17bn on the note's values (central EUR 11-12bn) and by EUR 10-19bn on stated-window values. The non-EU share of revenue falls from 80% to about 62-70%. The note's validation compares the implied aggregate (12.9%) with the EU's ~10% share of world primary energy. The two figures do not match, and the benchmark does not measure the needed quantity.

Evidence. (1) 62 of the imputed firms are non-EU utilities, power generators, grids, gas distributors or domestic coal-power groups: Adani x4, China Shenhua, Coal India, US/Brazilian/Canadian/Japanese/Indian utilities, Constellation, Sempra. They hold EUR 205.8bn (33%) of the imputed base and EUR 6.8bn (10.5%) of revenue, with essentially no EU sales. Adani Transmission AR FY22 states that 'the business operations of the Group are entirely based in India' (scripts/domestic_utilities.py). (2) The paper's own observed values for US majors are 2% (Chevron, Occidental), against 10% for EOG, Devon, Pioneer and other US producers. (3) The aggregate check fails on its own terms. The implied EU share of non-EU gains is 12.9% (11.0% after F3), against an EU-27 share of world primary energy of 10.1% in 2021 and 9.5% in 2022 (OWID). Either way, that benchmark is not the gain-weighted EU sales share of these firms. (4) Adjudicator reassignment of EU-destination shares for the 70 largest imputed firms (83% of the imputed base) raises exporters (Cheniere 40%, Aker BP 60%, Vermilion 45%, Jinko 25%, Cameco 25%, Glencore 20%) and puts domestic firms near 0. It gives a gain-weighted share of 4.1% (range 1.9-7.0%) and a revenue change of -EUR 12.1bn (-6.1 to -16.6) on published values, -EUR 10.6bn after F3/F4/F5 on the note's values, and -EUR 15.6bn on WP5's stated-window values (scripts/apportion_sensitivity.py, correction_ladder.py). Auditor A's independent classification gave -EUR 7bn to -13bn; Auditor B's uniform-5% scenario gave -EUR 10.2bn.

Why this severity. About a third of the revenue rests on an assumption that is asserted but not validated. A third of that base has essentially no EU sales, which puts a floor of -10.5% on the overstatement. Evidence-based reassignment lowers the headline by roughly a fifth (central -16% to -24% depending on the window; range -9% to -30%). With F3 and F4 it takes the estimate to roughly EUR 35-50bn. 'Tens of billions' survives, so the issue is major rather than critical.

The authors' response. The note itself warns that firm-level results for non-EU firms 'should be interpreted with caution' but asserts aggregate accuracy (p.11). No response to this critique was located.

Why it matters, in detail. The headline is a mechanical score that assumes source-state jurisdiction over non-resident parents' worldwide market-value gains, apportioned by sales, with no permanent-establishment nexus. That jurisdiction is doubtful under treaty provisions such as those cited above (Arts. 2, 7 and 14 of the France-UK treaty) and under Art. 5 of the Capital Duties Directive. If it fails, about EUR 41-52bn of the score is not collectable as designed. What remains is about EUR 13-15bn from EU-headquartered firms, roughly 0.5-0.6 times the solidarity contribution.

Evidence. France-UK DTC 2008 (SI 2009/226): Art. 2(1)(b) covers French 'taxes on capital appreciation'. Under Art. 7(1), a UK enterprise's profits are taxable in France only through a permanent establishment, and then only the profits attributable to it. Art. 14(5) reserves other gains to the residence state. The Commission's own levy applies only to 'energy undertakings ... in the Union' (QANDA/22/5490). Several large non-EU payers are state-owned: Aramco's public float is ~1.7% (2019 IPO), the rest held by the Saudi state and PIF; Equinor is majority state-owned. The November note added a retrospective-announcement argument (p.9), which answers Neidle's snapshot point, but did not engage with the treaty, Directive or GATS points. It does cite a different Neidle piece in Section 5.

Why this severity. If the obstacles bind as Neidle argues, collectable revenue falls toward the EU-firm component (about EUR 13-15bn). That is below the Commission's EUR 25bn: it reverses the 'three times' comparison and removes most of the headline. This is an undisclosed precondition, not a demonstrated error: no court has ruled, and a redesign (treaty renegotiation, a levy outside treaty coverage) is conceivable. Hence confidence is only 'plausible'.

The authors' response. Partial. The Nov 2022 note (p.9) adds that the tax could be announced retrospectively in early 2023, which answers Neidle's snapshot and manipulation point. No response to the treaty, Capital Duties Directive or GATS points was located.

Why it matters, in detail. Shell's correct gain is EUR 43-47bn, an overstatement of 2.4-2.7x. Apportioned at the paper's 45.12%, revenue falls by EUR 10.0-10.7bn (Shell EUR 17.0bn to about EUR 7.0bn). The non-EU share of revenue falls from 80% to 76%, and the multiple of the Commission's EUR 25bn from 2.6 to 2.2. The WP5 headline falls by EUR 10.4bn. The base the note calls 'perfectly observable' thus contains a share-class splice in its largest component.

Evidence. Adjudicator reconstruction (scripts/shell_check.py): LSE closes 1621.8p (31 Dec 2021) and 2273.0p (1 Sep 2022), ECB GBP rate 0.86473. With all shares the gain is EUR 46.98bn (Amsterdam EUR line 43.3bn; USD ADR 23.7bn). Counting only the A shares in January gives EUR 114.17bn, within 0.2% of the paper. Under the window the note's other values imply (6 Jan to 2 Sep), the figures are EUR 42.7bn with all shares and EUR 113.9bn with A shares only. companiesmarketcap.com, the list source cited in fn.6, shows Shell at USD 89.77bn on 31 Dec 2021 and USD 193.43bn on 31 Jan 2022. That is +115% in a month in which the price rose ~15%: the same artefact. WP5's EUR 121.1bn carries the same error. Both auditors found it independently.

Why this severity. On its own it lowers the headline from EUR 64.6bn to EUR 54.5bn (-15.5%), below the ~25% guide for major. Under the note's stated window, the undercount of other firms (F6) would roughly offset it. It is still the largest single confirmed error. It falsifies the 'more than EUR 100 billion' statement for Shell and compounds in the same direction with F1 and F4.

Why it matters, in detail. Using the change in market capitalisation without capital-flow adjustment (M1 - M0 - I + D) taxes contributed equity. In this sample that is about EUR 21bn (Woodside), EUR 9bn (Aker BP), EUR 2.4bn (Orlen) and a subsidiary's post-IPO value (Jinko), worth about EUR 1.4bn of revenue at the paper's keys. With issuance, a 100% levy expropriates new investors' contributions, so the shareholder-protection statement fails.

Evidence. scripts/issuance_check.py. Woodside (Table B p.26, EUR 26.12bn): the market-cap rise of EUR 28.9bn is EUR 7.6bn of price gain plus EUR 21.2bn of the 914,768,948 shares issued for BHP Petroleum (HY2022 Note C.1: 969.6m to 1,898.7m shares). Aker BP (EUR 10.50bn): 271,908,701 consideration shares for Lundin's E&P business took the count from 360.1m to 632.0m; the price gain was EUR 2.66bn. PKN Orlen (an EU firm taxed at 100%, EUR 0.80bn): its price fell 23% (PLN 74.34 to 57.20); market cap rose only through 198,738,864 series-E shares issued for the Lotos merger. Jinko Solar (EUR 9.73bn; EUR 0.84bn in WP5): this matches the EUR 10.0bn rise of Jiangxi Jinko (688223.SS) from its 26 Jan 2022 listing (CNY 10.55 to 17.46 on 10bn shares), whereas the NYSE parent rose about EUR 0.6bn. Removing these overstatements lowers revenue by EUR 1.4bn (-2%). Buybacks bias the other way: Shell's price-only gain on its January share count is EUR 57.9bn, against a market-cap rise of EUR 47.0bn.

Why this severity. The headline effect is small (about -2%). But the firm-level misattribution is confirmed: 72-78% of Woodside's and Aker BP's 'gains' are acquisitions, and Orlen is taxed despite a falling share price. The issue also falsifies the claim that a 100% rate leaves shareholders no worse off.

Why it matters, in detail. The two versions' firm values differ by 20-36% for the largest US and Saudi inputs under the same stated window. The note's values are consistent with a later start date (first week of January, after an ~8-10% energy rally), so its base is about 17% below what its own description implies. This partly offsets F3, so under the stated window the confirmed-corrected headline is about EUR 63-66bn. That the published figure is roughly right under the stated window is therefore the result of two undocumented errors cancelling.

Evidence. Independent window fit (scripts/window_fit.py; SEC cover-page share counts x daily closes for 19 US firms). WP5's values best fit 31 Dec 2021 to 2 Sep 2022 (median firm error 2.9%). The note's values are a median 0.83 of the stated-window gain and best fit about 6 Jan to 2 Sep (4.9%); Auditor A found ~5 Jan to 7 Sep (3.5%). Exxon: EUR 109.6bn in the note, 128.9bn for the stated window, 138.3bn in WP5. Aramco: EUR 172.0bn in the note, ~267bn for the stated window, 269.0bn in WP5. Petrobras (+EUR 14.3bn at 20% in WP5) was dropped without explanation, and WP5's duplicate Cheniere row and its Brookfield Renewable Partners value (EUR 13.87bn) were silently changed. After the confirmed fixes plus Adani at 0%, the estimate is EUR 50.2bn on the note's values and EUR 63.5bn on WP5's stated-window values (scripts/correction_ladder.py).

Why this severity. The headline moved by about 18% between versions without reconciliation. The firm-level tables cannot be reproduced from the window described, and the window ambiguity alone shifts the corrected headline by about EUR 13bn. The authors do disclose window sensitivity in general (Appendix B: EUR 51.2bn with monthly means), so readers are partly warned.

Why it matters, in detail. Country-by-country reports allocate revenue by the tax residence of group entities. Destination-based apportionment of non-resident parents' gains would need new customer-location reporting, and the appropriate-use rules for CbC data bar formulary apportionment of income.

Evidence. OECD (2015) BEPS Action 13 Final Report defines CbC revenues as 'the sum of revenues of all the Constituent Entities of the MNE group in the relevant tax jurisdiction'. Para. 25 says CbC data 'should not be used by tax administrations to propose transfer pricing adjustments based on a global formulary apportionment of income'. The appropriate-use commitment adds: 'Jurisdictions should not propose adjustments to the income of any taxpayer on the basis of an income allocation formula based on the data from the Country-by-Country Report'. Shell's CbC report books USD 38.1bn of third-party revenue in Singapore and USD 7.0bn in the Bahamas, and its annual report labels its geographic split 'by origin'. Laffitte & Toubal (2022) document sales booked in havens (corpus offshore-havens-15).

Why this severity. There is no direct effect on the number. But the feasibility claim for the destination key on which about 75-80% of the revenue depends is unsupported, and the same confusion probably produced the mislabelled shares in F4.

Why it matters, in detail. Under the paper's assumption that the profit levy is capitalised, the combined liability is W + t(G - W), not max(W, tG). Overlapping rents therefore face cumulative taxation, and the valuation tax is not a minimum tax.

Evidence. Take the authors' own full-capitalisation assumption. A windfall of 100 taxed at 33% by a profit levy leaves a valuation gain of 67, and a 33% valuation levy then takes 22.1, for a total of 55.1 (Auditor B). The profit levy is in effect deducted from the valuation base, not credited against the valuation tax. Saez & Zucman (2022, Economic Policy 37(110), section 3.4) say that to make such a tax a minimum tax, 'any wealth tax paid could become a non-refundable tax credit against the corporate profits tax'. The note provides a credit only for cross-country claims (p.8). It also follows that if the Commission's levy (announced 14 Sep 2022, after the note's window) were capitalised into end-2022 values, the valuation tax's own yield would fall by up to a third of the capitalised levy, so the two revenues are not additive.

Why this severity. The EUR 64.6bn score is unchanged. But a stated design conclusion is wrong: the combined burden on overlapping windfalls is about 55% at 33% + 33%, and the 'minimum tax' label requires a credit the note does not provide.

Why it matters, in detail. The apportionment keys labelled 'EU' appear in at least two cases to be Europe or EMENA aggregates, or figures not present in the cited documents. For Shell, the largest contributor, the key is about 1.5-2.1 times Shell's own EU figures. The two Shell errors (F3 and this one) multiply: EUR 114.41bn x 45.12% = EUR 51.6bn of apportioned base, against EUR 47.0bn x 29.4% = EUR 13.8bn.

Evidence. Shell's 2021 country-by-country report (Tax Contribution Report 2021) puts the EU-27 at 21.17% of third-party revenue (USD 278.3bn) and 21.21% of total revenue; EU-27 plus UK is 29.42% (scripts/shell_cbc.py). Shell AR 2021 Note 7 gives third-party revenue 'by origin' for Europe of USD 78,549m out of 261,504m (30.0%), of which the UK is 21,846m. Europe's share was 28.5% in 2019, 27.8% in 2020 and 35.7% in 2022: nothing near 45%. ConocoPhillips' 10-K 2021 EMENA segment is USD 5,902m of 45,828m (12.9%) and covers Norway, Qatar, Libya and UK terminalling, with no EU operation; the note uses 13.1%. Chevron's 10-K splits revenue only into United States and International, so a 2% EU share cannot be read from it. ExxonMobil's 10.5% is consistent with its 10-K (France+Italy+Belgium = 11.7%).

Why this severity. The issue concerns a few large inputs. Replacing Shell's 45.12% with its own EU-27+UK (29.4%) or EU-27 (21.2%) country-by-country share lowers revenue by EUR 2.4-3.7bn on the corrected Shell base (-4% to -6%). Destination shares are unobservable, and entity-based data can understate EU destination sales for exporters such as Equinor. The net bias across all observed firms is therefore less certain than for Shell.

Why it matters, in detail. A positive-part snapshot base, E[max(dV, 0)], is noisy at firm level: whether Aramco pays EUR 3bn or nothing depends on the end date. For firms with a free float of 0.3-1.7% the 'market value' is a marginal price applied to closely held shares, and part of the yield is an intra-government transfer.

Evidence. Over the proposal's full-year window, Aramco's gain turns negative (adjusted close 29.59 to 29.18; -EUR 24.5bn, untaxed). Exxon's rises to about EUR 179bn and Chevron's to about EUR 109bn (adjudicator check). Auditor A's CMC-based full-year aggregate for 273 firms is similar (EUR 63.5bn vs EUR 62.8bn on the same firms' Jan-Sep values, Shell corrected); it was not re-run here. EnBW (EUR 4.92bn base at 100%, EUR 1.6bn of revenue, the 7th-largest contributor): the State of Baden-Wuerttemberg and OEW hold 94% of the capital (EnBW investor page), and median Xetra volume was about 318 shares a day in Jan-Aug 2022 (Auditor A). Aramco's public float is about 1.7%. The Czech state holds 69.8% of CEZ (EUR 1.3bn of revenue). Ownership-weighted, about EUR 2.5bn of the EU-firm revenue from EnBW and CEZ alone would be paid by companies owned by member-state governments.

Why this severity. The aggregate effect is small (the full-year total is similar, and the authors disclose date sensitivity). But firm-level incidence is fragile, and some large bases rest on very thin floats or public ownership. This qualifies the observability claim and the per-household redistribution arithmetic.

The authors' response. The Nov 2022 note proposes a retrospective announcement (p.9) and reports a monthly-mean alternative (Appendix B, EUR 51.2bn).

Why it matters, in detail. Pre-invasion gains reflect the 2021-22 energy-price recovery and pre-war risk. Attributing the whole calendar-year rise to the war overstates the war-specific rent captured by the base by a factor of about two for the main contributors.

Evidence. For 29 large contributors covering EUR 101bn of the EUR 195.7bn apportioned base, 52% of the 31 Dec 2021 to 1 Sep 2022 price gain had accrued by the 23 Feb 2022 close. Examples: Aramco 95%, TotalEnergies 88%, Shell 49%, Exxon 48%, Equinor 29%. The calculation is price-only, weighted by apportioned base (scripts/pre_invasion_share.py). The note acknowledges that energy prices had been rising for over a year before the invasion (p.6).

Why this severity. There is no effect on the revenue score for the base as defined, since a calendar-2022 base is a legitimate design choice. But the causal framing is qualified: a base starting on 24 Feb 2022 would roughly halve the revenue from these firms.

Why it matters, in detail. Different bases (a stock revaluation, including the present value of future rents, against one year's flow of surplus profit) and different jurisdictions (worldwide apportioned against EU-resident) make the 2.6x ratio a comparison of scopes, not of instrument efficiency.

Evidence. 64.6/25 = 2.6. The ratio is 2.2 after F3 and about 1.4-2.0 after F1, F3, F4 and F5 (scripts/correction_ladder.py). The Commission's levy taxes 2022 surplus profits of energy undertakings 'in the Union' (QANDA/22/5490). The multiple comes from taxing non-EU firms (80% of revenue) and from capitalised expectations of future rents, not from a better instrument for the same base. The EU-firm component is EUR 13.0bn, about 0.5 times the Commission's figure, and the note's own worldwide-profit alternative raises more (EUR 75.1bn, Section 5).

Why this severity. The note explains where the difference comes from (upstream rents and non-EU firms), so this is a framing issue. Its substantive content is covered by F2 and F1.

Why it matters, in detail. The GDP ratio understates the published revenue by about a quarter. Ironically, the corrected revenue range (EUR 35-50bn) is 0.2-0.3% of 2022 GDP.

Evidence. Eurostat nama_10_gdp (EU27_2020, current prices): EUR 16,192bn in 2022 gives 0.40%; EUR 14,819bn in 2021 gives 0.44%. The note's own p.12 statement that EUR 195bn is 'about 1.2% of EU GDP' implies 0.4% at 33%. WP5's '0.4%' for EUR 79.2bn was itself low (0.49-0.53%).

Why this severity. A labelling error in the conservative direction. The euro figure is unaffected.

Why it matters, in detail. The discrepancies (up to 15% between text and table totals) signal weak version control but carry no independent revenue bias.

Evidence. p.11 says capitalisation 'had grown by EUR 1.1 trillion', against Table 1's EUR 1,250.4bn. Appendix B (p.18) says EUR 0.9 trillion, against Table A's EUR 1,063.4bn. Appendix B's September-mean capitalisation of EUR 8.1tn exceeds the EUR 7.5tn on 1 Sep, although the firms' capitalisation fell during September (Auditor A: -8.3% in EUR). '11 firms account for about 55%' (p.11) is 48.8% in Table B, where only 8 of the 11 appear (WP5: 57% vs 49.7%). p.7 says 18 Stoxx 600 energy firms rose and 18 fell, plus 283 firms outside the Stoxx 600: 301, not the 289 in Table B. The 9 Nov 2022 copy at gabriel-zucman.eu says 6 rose and 30 fell, which sums to 289 but is implausible, and uses 'Appendix B' for Table B. Numerical tables are identical in the two November copies.

Why this severity. Editing errors that do not change the headline arithmetic, which reproduces exactly from Table B.

Why it matters, in detail. Mislabelled columns (apportioned values presented as gross growth) can mislead readers comparing Table 2 with Table B. The country-coding errors show the appendix's headquarters field was not checked.

Evidence. Table 2 (p.13) Panel B 'Growth in Market Capitalization' shows apportioned values (Shell 51.6 = 114.41 x 45.12%). Table 4 (p.15) uses the same header for excess profits. Both table notes mention '4 different potential tax rates' where three are shown. '11 large non-US energy multinationals' (p.11) includes four US firms. The Fig. 2 note says 'rose or fell in 2020' for 2022. Table B codes CEZ (Czech) and Tsakos (Greek) as DEU and Suncor, Enbridge (Canada), Eletrobras, Cosan (Brazil), YPF and Pampa (Argentina) as USA; it treats Allego, a Dutch company, as a non-EU 'USA' firm at 10%. WP5 had most of these right. The UK levy is the 'Energy Profits Levy', not 'Excess Profits Levy' (p.9). Italy's 2022 levy (DL 21/2022 art. 37) is assessed on the increase in the VAT balance of sales and purchases, not on profits with a margin test (Auditor A). On 'in force until 1950' (p.4): the WWII excess profits tax was repealed by the Revenue Act of 1945, effective 1946, and a separate Korean-War tax ran 1950-53; this was checked against statute titles only.

Why this severity. Presentation errors. Only the Allego coding has any revenue effect (+EUR 0.15bn).

Why it matters, in detail. Without the extract, C5 cannot be checked. Doubling H1 2022 overstates calendar-2022 profits for firms whose earnings peaked in H1 (Shell -17%).

Evidence. Table 3/4 arithmetic is consistent (227.7 x 33% = 75.1; Shell 53.2 x 45% x 20% = 4.8). Auditor B's Shell check (H1 2022 pre-tax USD 36.9bn, 2018-21 average) gives about EUR 55bn, against 53.2 in Table 4. Shell's actual 2022 pre-tax profit (USD 64.8bn) gives excess profit 16.6% below the doubled-H1 figure. Table 4's first column is mislabelled (F14).

Why this severity. A secondary result that cannot be reproduced and uses a disclosed forecasting shortcut. It is plausibly overstated by the second-half fall in prices.

Why it matters, in detail. A tax on each listed entity's capitalisation cascades through cross-holdings. The overlap is small here because most affected subsidiaries carry 10% keys.

Evidence. TotalEnergies (EU, 100%) held roughly 37% of Adani Total Gas and 20% of Adani Green (both at 10% in Table B), per Auditor A; not re-verified here. Other pairs: Marathon Petroleum-MPLX, Hess-Hess Midstream, BP-Aker BP (~16%), Aramco-S-OIL, Plains GP-Plains All American, Delek US-Delek Logistics. Auditor A bounds the overlap at under EUR 1bn of revenue; TotalEnergies-Adani is the largest (about EUR 0.4bn at the 10% key).

Why this severity. Under 1-2% of the headline.