Taxing Extreme Wealth in the United Kingdom: Revenue Estimates for a 2% Minimum Tax on Ultra-High-Net-Worth Households

Why it matters, in detail. Revenue = sum_i max(0, 2% x W_i - T_i). With T/W = 0.2% for all households, gross = 1.8% x GBP642.5bn = GBP11.57bn. Zucman's (2024) group means (0.3% on the GBP417.5bn in billionaire households; 1.2% on the GBP225bn in centimillionaire households) give GBP8.90bn gross and GBP8.01bn net (-23%). Almost all of the effect comes from centimillionaires, whose gross contribution falls from GBP4.05bn to GBP1.80bn. Because the floor is computed household by household, dispersion around these means raises revenue, so -11% to -22% is the plausible band. The effect compounds with F2: jointly, 2026 net revenue is GBP6.9-7.6bn centrally.

Evidence. Zucman (2024) G20 blueprint (June 2024 and May 2025 HAL versions, identical on these points): p.16 'only 0% to 0.6% of their wealth'; p.23 'equal to 0.3% when expressed as a fraction of wealth'; pp.24-25 and fn 17 'I assume an effective tax rate of 1.2% for people with wealth between $100 million and $1 billion', derived from the income-tax rates of the P99.99-P99.9999 groups times an assumed 11% income/wealth ratio; p.24 'Following the billionaire minimum income tax...'; p.10 regressivity onset; fn 15 notes that taxpayer-level heterogeneity raises minimum-tax revenue. Balkir, Saez, Yagan & Zucman (2025, NBER w34170, s.3.3 and Appendix Fig. A1): US top-400 individual income taxes 0.6% of wealth, top-100 0.4%. Recomputed on the paper's Table 2 groups (GBP417.5bn in households above GBP1bn; GBP225.0bn in GBP100m-1bn households): 0.3%/1.2% -> GBP8.01bn net (-23%); centimillionaires alone at 1.2% -> GBP8.38bn (-19%); 0.3%/0.8% -> GBP8.82bn (-15%); 0.2%/0.6% -> GBP9.60bn (-8%); same means with lognormal within-group dispersion (log-sd 0.5-1.5) -> GBP8.13-9.23bn (-22% to -11%). Source: final/scripts/adjudicator_checks.py, output data/adjudicator_results.json.

Why this severity. This is one of the two parameters that set the headline. Moving it to the cited source's own values lowers 2026 revenue by about a fifth to a quarter (-17% to -23%; -11% under wide dispersion), and the bias runs one way, because the paper's 0.2% is at or below the lowest of its sources' values. The qualitative conclusion survives, but the GBP10bn rests on an assumption that the authors' own source and their own text contradict.

The authors' response. No response to this point located. In-paper defence: no UK data link the rich to the taxes they pay (fn 7; s.7.2.5); Table 6 varies a uniform 0.1-0.5% rate and the paper calls revenues robust to it.

Why it matters, in detail. Base erosion is roughly eps x tau. The relevant tau lies between 1.8% (the incremental burden, which drives migration and real responses) and 2% (the marginal rate on reported wealth, which drives valuation and evasion responses). For reporting responses the credited income tax does not fall, so revenue falls by about 1.1 times the erosion. eps = 11.3 implies 20-23% erosion against the assumed 10%, giving GBP8.95-9.21bn net (-11% to -14%); eps = 17 gives GBP7.63-8.03bn (-23% to -27%). The bias relative to the paper's 'conservative' label is one-directional: 10% is below SZ (2019)'s 15%, Zucman (2024)'s 20% low-end, and A&T's lower bound at a 2% rate (14%). Combined with F1, 2026 net revenue is ~GBP6.9-7.6bn centrally and ~GBP5.9bn at the pessimistic end.

Evidence. Advani & Tarrant (2020, WTC evidence paper; abstract, s.1 and s.4): 'a well-designed UK wealth tax could achieve an elasticity ... in the region of 7-17 after a period of 4-8 years', the spread depending on international migration. The 'very wealthy' Danish estimate without international migration is 11.3 (Jakobsen, Jakobsen, Kleven & Zucman 2020); adding an inter-regional migration effect gives an upper bound of ~17.1. WTC Final Report (2020): '7%-17% of the initial tax base would be lost ... at a tax rate of 1%', and responses 'may be larger at higher echelons of wealth'. Zucman (2024, p.24): low-end scenario 20%, described as 'higher ... than the 15% typically used in the literature'. Saez & Zucman Sanders memo (2 Mar 2026, p.1): 10%, because 'migration responses are likely to be very small as American billionaires would continue to be liable even if they move [abroad] and renouncing US citizenship ... triggers a large exit tax'. HMRC IHTM47001: the inheritance-tax tail (3-10 years) applies to long-term residents at death or transfer only. STRL 2026: 111 UK-citizen entries live off the British mainland, and at least 15 foreign nationals were removed because they moved away. Recomputed 2026 net revenue (gross GBP11.57bn), applying A&T elasticities at the 1.8% incremental / 2% marginal rate: eps 7 -> GBP10.11/9.95bn; eps 11.3 -> GBP9.21/8.95bn; eps 17 -> GBP8.03/7.63bn (-3% to -27%). Flat haircuts: 15% -> GBP9.83bn; 20% -> GBP9.25bn (final/scripts/adjudicator_checks.py).

Why this severity. The haircut is the other parameter that sets the headline, and it is unvalidated for this design. Within the authors' own cited UK evidence it lowers 2026 revenue by about 11-14% at the very-wealthy, no-migration estimate and by up to 23-27% at the upper end. The cited responses build over 4-8 years, so the 2030s path is affected more. The qualitative conclusion survives.

The authors' response. In-paper: the design (no exemptions, household unit, 10-year post-emigration liability, immediate effect, HMRC focus on ~1,000 cases) makes avoidance 'costly, difficult and highly visible' (s.6.3), and the 10% is called conservative (s.7.2.5). Zucman's broader position is that the level of avoidance is a policy choice (Saez & Zucman 2019; corpus offshore-havens-17 author response). No response to these audits located.

Why it matters, in detail. With 1.8% deducted each year, taxed wealth relative to GDP changes at about (1+g)(0.982)/(1+g_Y) - 1 a year. The paper's g = 6.48% against g_Y = 3.51% gives about +1.0% a year; g = g_Y + 1.85pp (the 2016-26 premium) gives about zero. The '22% vs 17.5%' and '0.33% -> 0.39% of GDP' paths therefore embed a ~1.1pp-a-year larger premium than the paper's own reference window, justified by inflation that the GDP denominator does not contain. The 37-year premium (3.3pp) would support the paper's numbers, so the projections are window-dependent rather than wrong.

Evidence. A household simulation matched to the Table 2 bins reproduces Table 1 (gross GBP11.57bn -> 19.58bn vs 19.56bn; 0.39% of GDP net in 2036). 2036 net revenue, no-tax top-300 wealth and with-tax top-300 wealth, all as % of GDP: paper's 6.48% -> 0.39 / 21.9 / 17.9; constant 2016-26 premium over GDP (5.36%) -> 0.35 / 19.7 / 16.1; with-tax growth = GDP, as in the logic of Saez & Zucman's Sanders memo (5.41%) -> 0.35 / 19.8 / 16.2; constant 1989-2026 premium (6.8%) -> 0.41 / 22.6 / 18.5; 2021-26 top-300 trend (4.35%) -> 0.31 / 17.9 / 14.6; pre-tax growth = GDP -> 0.29 / 16.5 / 13.5; zero growth -> 0.19 (final/scripts/adjudicator_checks.py; GDP from OBR March 2026 Table A.3 extrapolated as in the paper).

Why this severity. This affects secondary, explicitly conditional findings, not the 2026 headline. Under an equally defensible assumption (a constant 2016-26 growth premium), revenue/GDP rises only to ~0.35%, and top wealth/GDP with the tax falls slightly rather than rising, which reverses one executive-summary claim.

The authors' response. In-paper: the projections are conditional on 'the trends of the wealth of the super-rich and GDP projections over the last 10 years' continuing (Executive Summary); fn 5 notes that the decade 'included both periods of rising and falling wealth'; s.7.2.4 defends nominal growth by expected high inflation.

Why it matters, in detail. The evidence supports 'a high threshold makes administration tractable', not the categorical dismissal. The unanalysed margins (low-yield, high-valuation private holdings; legal and collection mechanics for non-residents) are the ones that decide whether a 10% haircut is achievable. The 'one-third of the deficit' comparison holds only for FY2026-27: from FY2027-28 the same sum exceeds the projected current deficit (GBP4.6bn), and from FY2028-29 OBR projects a current surplus. The '<1% administrative cost' figure excludes the WTC's one-off set-up cost (~GBP0.6bn) and private compliance costs, which are small but not negligible relative to one year's revenue.

Evidence. Paper: Abstract (PDF p.2); Executive Summary p.3; s.2 p.4; s.6 pp.9-13; text search: 'illiquid' appears only in the abstract. WTC Final Report (2020) executive summary: 'we recommend that instead of an annual wealth tax, the government should reform existing taxes on wealth'; '44,000 people (7% of those paying) would be liquidity constrained'; 'The timeline for delivering an annual wealth tax would likely be longer than for a one-off'; s.5: 'These may be larger at higher echelons of wealth'. OBR EFO March 2026 Table A.9: current budget deficit GBP33.9bn, 4.6bn, -3.3bn, -23.6bn and -30.3bn for 2026-27 to 2030-31. Neidle (2025): first receipts no earlier than January 2030 if announced in the Autumn 2026 Budget. HMRC IHTM47001. The Conseil d'Etat opinion read in the French Assemblee on 31 Oct 2025 (corpus offshore-havens-21) found a 10-year post-emigration tail incompatible with EU freedom of establishment; this does not bind the UK but shows the design is legally untested.

Why this severity. This does not change the revenue arithmetic, but it materially overstates how far the paper establishes feasibility and fiscal relevance: readers are told objections 'do not hold' when several are unexamined.

The authors' response. In-paper: s.6 offers qualitative answers to each objection; taxpayers could pay in shares if they dispute valuations (s.6.4); immediate effect from the announcement date to prevent forestalling (s.6.3).

Why it matters, in detail. On AHS's own estimates, the very top's EATR is 23% (single year, including gains), 33% (5-year average) or 42% (income only). At 2018-19 statutory rates (20% above an GBP11,850 allowance; 12% employee NICs above GBP8,424), a median earner paid ~20-22% of earnings in income tax and NICs. So 'much below the average person' does not follow on a like-for-like personal-tax basis; it may hold on an all-taxes basis, which AHS do not estimate. Demonstrating parity would require UK data linking wealth, economic income and all taxes, which the paper says does not exist (fn 7; s.7.2.5).

Evidence. Advani, Hughson & Summers (2023), Oxford Review of Economic Policy 39(3), s.IV p.413 and Figure 1 p.414 (rendered and read): income-only EATR peaks at 43% (GBP1.3m) and is flat above; EATR on total remuneration peaks at 38% (GBP0.5m) and declines ~7pp between GBP2m and GBP10m in a single year; over 5 years it 'rise[s] by around 4pp between GBP2 million and GBP5 million and remain[s] largely flat'. Zucman (2024) pp.28-29. Paper s.2, fn 2, s.3, fn 7.

Why this severity. No effect on the revenue estimates. But the abstract's lead claim is presented as established UK evidence when the cited source supports only a narrower statement (single-year, gains-inclusive regressivity above GBP0.5m), and the parity result is asserted rather than demonstrated.

Why it matters, in detail. The extrapolation from the GBP340m list floor with alpha = 0.91 gives 873 entries and 946 households. Reconciling this with ~2,400 individuals needs a steeper tail below GBP340m (alpha ~1.2-1.5), more than one qualifying person per household plus pooled couples, or residents excluded by rule (i). The extra households would sit mostly at GBP100-250m and would add roughly GBP0.5-1.8bn of net revenue (+5% to +17%) if HMRC's count is accurate. This partly offsets F1-F3, but HMRC's wealth estimates are themselves imprecise.

Evidence. NAO (Davies 2025) para 1.3 p.14: 'HMRC estimates around 5,000 wealthy individuals have wealth above GBP50 million, with around half of these having wealth above GBP100 million'. NAO Figure 5: ~2,300 individuals with GBP100m-1bn; billionaires 'between approximately 50 and 165' (public estimates); 'HMRC does not require taxpayers to report their assets'. The correct conversion gives ~1,250-2,500 households, 32% to 164% above 946. Pareto sensitivity: alpha = 1.2-1.5 (vs 0.91) gives 1,160-1,460 households and +5% to +12% revenue. Fitting the full 2026 list gives alpha ~0.84-0.98 above GBP340m, so any gap must arise below the list floor (final/scripts/adjudicator_checks.py). 5,000/2.3 = 2,174, not 2,171.

Why this severity. This invalidates the paper's only external check on the taxpayer count. If HMRC's estimate is right, the abstract's 'fewer than 1,000 households' is too low by roughly a quarter to well over half. Feasibility is little affected, and revenue would be higher rather than lower.

Why it matters, in detail. If wrongly included emigrants hold 4-15% of the GBP642.5bn base, net revenue falls by GBP0.4-1.6bn. Around 45-50 excess entries correspond to ~90-100 households, about 10% of C2's 946. The net sign is uncertain: excluding even a few GBP5bn-class foreign-born UK residents under rule (i) would understate the base by a similar order. Fixed flags also blur the historical series (C3), but the rise survives dropping the residency filter altogether (raw top-199 entries: 6.0% -> 22.3% of GDP).

Evidence. Paper Tables 8-10: non-resident counts of 69, 65, 64 and 61 in 2023-2026. WP2 s.2.10: time-invariant flags; rule (i) 'born in a foreign country'; '287 out of 350' resident in 2025. WP2 s.3.2: non-resident share falls to 25% in 2025. STRL 2026 as syndicated (Irish edition of The Sunday Times and BusinessCloud, 15 May 2026; eprivateclient, 18 May 2026): 'Nearly a third (111) of the UK citizens who appear in the main list of 350 individuals no longer live on the British mainland. At least 15 foreign nationals who appeared in last year's Rich List have been removed because they now live elsewhere'; compiler Robert Watts: 'As UK nationals these people remain on our Rich List - wherever they now live.'

Why this severity. The 40-50 marginal misclassifications are most plausibly sub-billionaire fortunes, since the paper already excludes 42 billionaire entries. That implies roughly -4% to -10% of 2026 revenue (up to -15%), partly or fully offset by any UK-resident foreign nationals wrongly excluded. The misclassification also inflates the taxpayer count and makes the 'conservative' label unsupported.

Why it matters, in detail. Revenue is linear in the assessed base, and the base is dominated by a few dozen dynasties' private-company and listed holdings valued by journalists. No allowance is made for valuation shortfall or volatility beyond the generic 10% haircut. Concentration also makes a flat haircut a poor stand-in for migration risk: one Hinduja-size departure removes ~6% of revenue.

Evidence. Synthetic Pareto households supply ~9% of the base (adjudicator and claude). STRL 2026 full ranking: the top ten entries hold GBP216bn (28% of the list). Hinduja alone holds GBP38bn, 5.9% of the base (GBP0.68bn gross revenue). The eight largest entries not explicitly based abroad hold GBP182.9bn (~28% of the base if all are classified resident). Single-year revaluations: Dyson -GBP8.8bn; Ineos marked down for debt and losses. Top-300 wealth/GDP fell from 18.7% to 16.5% in three years (Fig. 1). Neidle (2025): the list is 'a series of highly educated guesses'. Balkir et al. (2025): estate values of Forbes decedents are 'only about 30% of the Forbes wealth estimates on average', which Saez and Zucman read as estate-tax undervaluation (us-wealth-taxes-11). Skar et al. (2025): Norwegian statutory values of unlisted shares have a median of 61% of sale prices (offshore-havens-23).

Why this severity. The direction is ambiguous, since lists can also miss wealth. But each 10% gap between list values and enforceable values moves revenue by ~GBP1bn, and with revenue this concentrated, a few revaluations or departures shift it by GBP0.1-0.7bn each. This is an important undisclosed caveat on C1 rather than a demonstrated bias.

The authors' response. In-paper: assets would be valued at open-market value as for IHT and CGT; HMRC can concentrate expertise on fewer than 1,000 cases; payment in shares can reveal values (s.6.4).

Why it matters, in detail. Reconstructing the base needs the entry-level residency flags (which decide ~GBP201bn of exclusions), household assignments and the top-181 growth panel. Approximate reconstruction from public data reproduces alpha (0.89 vs 0.91), the household-multiplicity mix (50/9/37/3% vs 49/9/35/5%) and Table 1, so the gap matters chiefly for F3.

Evidence. Drive probe on 25 Oct 2026 returned a Google sign-in page; WP2 pp.8-9 and 13-15; paper Tables 8-11 pp.25-28 and fn 14.

Why this severity. The headline can be closely approximated from the printed parameters and the public 2026 list (alpha, counts and base within ~3%). The gap mainly prevents checking the residency flags (F3) and the historical classification. The undisclosed revisions move historical levels by a few per cent at most.

Why it matters, in detail. The most consequential slip is the 2036 'GBP20 billion', which overstates net 2036 revenue by ~GBP2.4bn (14%). The stale Figure 6 note and the gross-vs-net deduction affect only the illustrative wealth paths, by ~0.4pp of GDP in 2036.

Evidence. Page references as listed. Recomputations in final/scripts/adjudicator_checks.py (33.9 - 10.41 = 23.49; 522 x 1.0648^10 = 978.0; 522 x 0.982^11 x 1.0648^10 = 800.9). Figure 6 and Figure 2 rendered and inspected. King's College London release (21 Jul 2026): 'almost GBP18 billion annually by 2036'.

Why this severity. These are editorial and transcription errors. The tables generally carry the correct values, and none changes the 2026 headline.

The authors' response. The KCL press release (21 Jul 2026) gives 'almost GBP18 billion' for 2036, consistent with the net value in Table 1.

Why it matters, in detail. Annual open-market valuation of private holdings would need new reporting infrastructure that the NAO says does not exist. That bears on the feasibility of the timeline (F6) and on the plausibility of a low haircut for valuation-based avoidance (F2, F5).

Evidence. NAO para 1.3: 'Calculating wealth is less precise than income, as HMRC does not require taxpayers to report their assets, only those aspects on which there is a taxable charge'. Summary para 10: 'HMRC does not collate data on an individual's liabilities across all types of personal tax'. Para 2.15: data on UK accounts 'are currently limited to savings interest', and about one in five accounts cannot be matched. On data gaps, HMRC 'has set no firm plans or timescales' to collect more information from the wealthy. Figure 5 note 5: HMRC 'could not provide us with the number of billionaires living in the UK'. The NAO does note that HMRC 'has been using more data sources to identify the wealthy population' (para 1.2).

Why this severity. This supports a qualitative feasibility argument only and has no direct effect on the revenue figures, but it removes one of the paper's stated grounds for dismissing valuation problems.