Annette Alstadsæter, Matthew Collin, Bluebery Planterose, Gabriel Zucman, Andreas Økland · EU Tax Observatory Working Paper No. 1 (May 2022), revised September 2025; R&R AEJ: Economic Policy (per CV)

Who Owns Offshore Real Estate? Evidence from Dubai

14 problems found: 8 moderate · 6 minor

What the paper claims

Non-UAE-resident foreigners ('offshore' owners) held about 30% of the value of Dubai's existing residential real estate at end-2019, a share much larger than in the other large cities measured; foreigners including UAE residents held about 44%.
Foreign ownership grew from close to zero after the 2002-2006 liberalisation to about 35% by 2009 and 40-45% from 2012; the abstract states that the non-resident share grew 'from nearly zero in 2005 to 30% in 2019'.
Country origins of offshore owners follow a strong gravity relationship (distance, size, wealth, language and colonial ties); some poor or conflict-ridden countries hold large amounts relative to GDP and to recorded foreign assets.
In Norway, the probability of owning Dubai real estate rises with wealth, but holdings are much less concentrated at the top than offshore financial wealth.
Tax evasion is an important motive: most offshore Dubai real estate is not declared to home tax authorities.
Sanctions-driven capital flight is an important motive: Russian money poured into Dubai real estate after the 2022 invasion of Ukraine, mostly as long-term (off-plan) investment likely owned by Russians residing elsewhere.

Every problem found

1moderateconfirmed · newly identified here · methodology

To decide whether a foreign owner lives abroad, the paper mostly relies on a rule of thumb: an owner whose address is unknown but who rents out all their Dubai flats is assumed to live abroad, and that pattern is then extended to everyone else. The rule is sensible but untested, so the $68 billion could plausibly be somewhat lower or higher (roughly $50-75 billion). Dubai would still stand far above the other cities compared.

Technical detail and evidence

Why it matters, in detail. The $67.6bn is the foreign stock ($98.1bn) times an imputed non-resident share of 0.689. That share is identified mainly by a classification rule that sets P(non-resident | no listed residence, all units let) = 1, plus a missing-at-random pooled imputation for the 40% not renting out. Under conditional independence and the paper's own renting rates, the rule is close to accurate (implied P(N|R) of 0.97-1.0) and the bias is small. If renting is correlated with residence being unlisted for reasons other than residence (stale addresses, tenant records, investor-type owners), the bias runs downward relative to the headline: -15% to -33% in the scenarios examined. The 30% share therefore carries an unreported uncertainty band of roughly +/-20%. The cross-city ranking (Dubai at about 20-34% vs at most 11.1% elsewhere) is unaffected.

Evidence. Section 3.3.2 p. 16; App. F p. 64 (33%; 27% = 67% x 40%; fn. 60: 35% of all owners rented out everything); G.2 pp. 69-70 and fn. 66. Adjudicator aggregate model (scripts/06_residency_sensitivity.py, logs/06_residency.txt), which treats owner shares as value shares and pools nationalities: - Reproducing the paper's 0.689 offshore/foreign ratio implies a non-resident share of 0.437 among listed owners. - About 21% of the offshore stock then rests on listed residence, 39% on the renting rule and 40% on imputation. - If 25% (50%) of rule-classified landlords are in fact UAE residents, the stock is $56.6bn ($45.6bn), i.e. -16% (-33%), matching Codex. - Imputing the 40% from listed owners only gives $57.7bn (-15%), matching Claude's scenario A. - A model calibrated to the paper's own renting rates (24.8% of listed vs 40% of unlisted owners rent out everything), assuming conditional independence, gives $60-77bn (-11% to +13%); Claude reports -9% to +17%. - Weak external check: for Norwegian nationals the procedure assigns 75.9% of value to offshore owners. Journalists were certain that 66% of Norwegian-linked owners (172 of 259) were Norwegian tax residents; Skatteetaten found 82% (214 of 260) tax-liable in Norway.

Why this severity. The issue concerns how precise the headline level is, not whether it exists. Defensible alternatives move the offshore stock by about -15% to -33% in adverse cases, or by -11% to +13% when calibrated to the paper's own renting rates, so mostly by less than a quarter. Even the most adverse scenario (~$46bn, ~20% of the market) leaves Dubai at about 1.8 times the next-highest place measured (11.1%).

The authors' response. No direct response. The paper states the representativeness assumption (fn. 22; G.2) and says assignment is by value (fn. 66); no validation or sensitivity analysis is reported.

2moderateconfirmed · newly identified here · circularity

The raw data support the paper's finding that richer Norwegians are more likely to own Dubai property. But the paper then assumes that every owner who still looks 'not rich' must really be rich and moves them into the top 10%. That assumption, not the data, creates the picture of ownership concentrated at the top. Without it, about 70% of the value (raw) or roughly 30% (after counting the Dubai property itself) belongs to people outside the top 10%.

Technical detail and evidence

Why it matters, in detail. The corrected Panel C shares sum to 100% within P90-P100 by construction: the reallocated residual (about $19-22m of $65-66m) is spread with population weights. Removing the reallocation but keeping the re-ranking leaves about 29% of value below P90, the top 1% at about 30% (vs 34%) and the top 0.1% at about 3-4% (vs 4%). The probability gradient (uncorrected 0.0036% at P0-50 to 0.095% in the top 0.1%) is robust. The 'concentration among the upper end' and the corrected 'zero below P90' are assumption-driven. With about 5 top-0.1% owners and different units (individuals vs households), the top-cell comparison with HSBC carries no statistical information.

Evidence. Paper: Section 5.2 pp. 27-30; Figure 7 p. 28; App. N pp. 88-90. Digitisations: - claude/logs/04_norway.txt: 96 owners and $18.9m (29% of value) reallocated. - codex/figure7_metrics.json: raw vs corrected top-10% 30.25% vs 100%, top-1% 18.1% vs 34.8%, top-0.1% 3.9% vs 4.2%; diagnostic 102 owners / 34% of value. - Adjudicator visual check of Figure 7 rendered from PDF p. 30. Other sources: AJZ (2019) online appendix workbook E sheet 'Stata-NOR' (4/2,729 = 0.147%); E24 SKUP method report pp. 9-10.

Why this severity. The assumption produces the picture of all holdings sitting in the top decile and the wording 'concentrated among the upper end'. The abstract's own claims survive in the raw data: the ownership probability rises roughly 25- to 30-fold from P0-50 to the top 0.1%. Without the reallocation, the top-0.1% and top-1% value shares (~3-4% and ~30%) are close to the paper's 4% and 34%, so the contrast with HSBC (98%) and the amnesty (77%) is, if anything, stronger.

The authors' response. Partly anticipated: the uncorrected series are shown. App. N p. 90 lists three sources of noise and argues that 'the overall picture [is] hardly affected'. fn. 31 explains why taxable wealth understates market wealth.

3moderateconfirmed · newly identified here · version instability

The first version of the paper (2022) said foreigners held at least $146 billion of Dubai property, twice as much as in London. The latest version says people living abroad hold about $68 billion, and that London has about twice as much as Dubai. The new figures are better, but the paper does not explain the change, and the old, larger numbers are still on official websites.

Technical detail and evidence

Why it matters, in detail. Moving from foreign ownership of all property (2022) to non-resident ownership of completed residential property (2025) cut the headline 'offshore' stock by 54% ($146.2bn to $67.6bn). The foreign stock fell 33% ($146.2bn to $98.1bn), partly through reclassifying commercial property, land, off-plan and cancelled projects, per the 2024 note. Intermediate outputs (2024 note Fig. 4; Atlas Table 1) compared Dubai's foreign share (43.8%) with other cities' non-resident shares. No bridge table links the versions.

Evidence. Versions: - APZO 2022 abstract and pp. 1-2. - APZO2022Data.xlsx: rows sum to $532.6bn; excluding UAE and unknown rows, $146.2bn = 27.4% (scripts/04_workbook2022.py). - EUTO 2024 note pp. 3-4 and 12-13 (Fig. 4). - Atlas methodology note (Nov 2024): Table 1 and p. 7. - ACPZO 2025 pp. 1-3, 19. Web pages: taxobservatory.eu repository page (captured 25 Oct 2026); gabriel-zucman.eu/who-owns-offshore-real-estate/ (checked 25 Oct 2026).

Why this severity. The 2025 estimates are conceptually better, and their own validity is not affected. But the work's best-known 2022 headline overstated offshore holdings by more than a factor of two relative to the authors' current estimate, the London comparison reversed, and the correction is not flagged where the old numbers still circulate.

The authors' response. Partial reconciliation in the May 2024 EU Tax Observatory note (pp. 3-4): commercial property, empty land, off-plan and cancelled projects removed; the leak no longer treated as comprehensive.

4moderatelikely · previously known · overclaiming

Norway's tax office confirms that about four in five Norwegian owners of Dubai property did not report it, so the paper's main tax point is sound. But the paper presents that Norwegian figure as a general '80% evasion rate'. Its comparisons for Britain, Spain and Italy set figures against each other that do not quite match, which makes compliance there look worse than the numbers show. The paper also cannot show that avoiding tax is why people buy.

Technical detail and evidence

Why it matters, in detail. Norway: compliance is 18.2-22.7% (39/214 to 39/172), an upper bound because declarations cover all seven emirates. UK: replacing properties by owners (dividing by 1.29) and excluding UAE-resident Britons (multiplying by 0.79) gives 23.9% instead of 14.6%. Spain: Modelo 720 totals after the first filing year are flows of (re)filings, so 10% is a lower bound on compliance. Italy: cost basis vs market value, and nationality vs tax residence, bias the rate in an unknown net direction. All sources still imply compliance well below 50%. 'Motive' is interpretation: no design isolates evasion as a reason to buy, and fn. 26 disclaims a causal link to the CRS.

Evidence. Paper: Abstract; pp. 2-3, 30-33; Figure 8 p. 31; Conclusion p. 40. Sources: - E24 SKUP method report (2023) pp. 9-11 (39 declarers; 214 of 260 tax-liable; TV 2 Denmark: 38 of 507). - Mosaic Chambers (1 July 2024) summary of The Times. - InfoLibre (B. P. Ramirez, 15 May 2024). - AEAT Modelo 720 FAQ. - IrpiMedia (14 May 2024). - Agenzia delle Entrate, REDDITI PF 2023 instructions p. 46 (via Codex). Recomputation: scripts/05_compliance_income.py (logs/05_compliance.txt).

Why this severity. Every check supports the finding that most Dubai property owned by European taxpayers goes undeclared (Norway 77-82% and Denmark 92.5% non-reporting), so the qualitative claim stands. What is weakened is the presentation of a single 80% as a general evasion rate, and the precision of the non-Norwegian rates, which concept mismatches push toward low compliance.

The authors' response. Partly anticipated: the paper calls the non-Norwegian evidence less systematic, gives an upper bound for Norway (30-35%), and notes in fn. 38 that Italian residence is measured less accurately.

5moderatelikely · previously known · speculative assumption

The paper's $2.7 billion a year of hidden income assumes that owners everywhere comply as poorly as Norwegians, including Gulf owners whose home countries do not tax income at all, and it counts rents before costs. Fixing just those two points lowers the figure to about $2 billion a year; the true number could be somewhat lower or higher.

Technical detail and evidence

Why it matters, in detail. Undeclared income = (R + G)(1 - c)/2, with c = 0.227 imposed on every jurisdiction. Excluding no-income-tax Gulf nationals (16.4% of value) and using net rents (70% of gross) gives $1.98bn (-27%). Unknown netting of capital losses and remittance or territorial regimes would push the figure further down. Lower compliance elsewhere (Denmark 7.5%) would push it up (+20%). The number is a scenario conditional on c, not an estimate of income undeclared to 'the relevant tax administrations'.

Evidence. Paper: Section 6.1 pp. 33-34; Introduction p. 3; Conclusion p. 40; Table H shares. Recomputation (scripts/05_compliance_income.py), $bn per year: - paper: 2.706 - excluding no-income-tax Gulf nationals: 2.26 (-16%) - net rents at 70% of gross: 2.37 (-12%) - both: 1.98 (-27%) - 50% compliance (Codex sensitivity): 1.75 (-35%) - Danish 7.5% compliance: 3.24 (+20%)

Why this severity. This is a secondary, derived number, not in the abstract. Documented definitional corrections lower it by roughly a sixth to a quarter. The uncertainty from transferring Norway's compliance rate cuts both ways, so the order of magnitude ($2-3bn a year of income, not of tax) survives; but 'globally ... not declared' overstates what is measured.

The authors' response. Partly anticipated: the paper introduces the estimate conditionally ('If we apply the same compliance rate observed for Norwegian taxpayers') and limits it to the residential market before the post-2022 boom.

6moderatelikely · newly identified here · overclaiming

After Russia invaded Ukraine, purchases through Russian-named estate agents in Dubai jumped about tenfold, which clearly shows more Russia-linked buying. But the paper cannot tell how much was bought by Russians (the number of Russian agents itself roughly tripled), whether the buyers lived abroad or were moving to Dubai, or whether sanctions rather than the war, conscription or capital controls drove it. Its own estimates range from about $2.4 billion to $5.9 billion.

Technical detail and evidence

Why it matters, in detail. Russian-broker volume is V_R = s_R R + s_N N. The synthetic control (Germany 56%, Iran 21%, Iraq 14%, Saudi Arabia 9%) estimates the change in V_R, but identifies the change in R only if s_N is constant and there is no broker-supply shock; Fig. O4a (3.5x vs 2x broker growth) contradicts the latter. An off-plan share of about 75% in the first year (and $4.59bn of $5.91bn in Table 1's separate fits), against a market-wide off-plan share of roughly 60-70% in 2022-23, is weak evidence of non-resident investment. Across the paper's own scaling and donor variants the estimates range from $2.37bn to $5.91bn. The 8% discrepancy between figure and text ($5.47bn vs $5.91bn) for the same window is unexplained.

Evidence. Paper: Section 6.2 pp. 34-39; Fig. 9 p. 36; Fig. 10 p. 38; Table 1 p. 39; App. B pp. 50-51; App. L p. 84; App. O Figs. O2-O5 pp. 92-94. Figure 10 vector paths (scripts/02_fig10_vectors.py, logs/02_fig10.txt): quarterly gaps 2022q1-2024q1 sum to $5.47bn; 2022q1-2023q4 sums to $4.63bn; pre-period RMSPE is about $30m a quarter. EUTO (2024) note Fig. 1(b) p. 6: on-plan vs off-plan residential sales values.

Why this severity. The Russian-linked surge itself is large relative to the pre-period fit (about 20x the RMSPE) and is corroborated by DLD broker data and population indicators, so the qualitative finding stands. What is not established is its size (the paper's own variants span $2.4-5.9bn), its attribution to Russian buyers, its 'offshore' nature and its sanctions cause. That is an important qualification of the abstract's 'sanctions-driven capital flight' motive.

The authors' response. Partly anticipated: the assumptions are listed (p. 37), switching is acknowledged (fn. 48), spillovers are discussed (fn. 46-47), and conservative variants are reported (Table 1).

7moderatelikely · newly identified here · internal inconsistency

The paper's country figures count citizens of a country who live outside the UAE, but the text treats them as people living in that country. A Pakistani living in Saudi Arabia or a Syrian living in Europe is counted as 'Pakistani' or 'Syrian'. Statements such as 'Dubai property is worth 5% of Syria's economy' or 'would raise Pakistan's foreign assets by 25%' are therefore likely overstated.

Technical detail and evidence

Why it matters, in detail. %GDP_i and %GFA_i use holdings of nationality i by non-UAE residents. If a fraction theta_i of those nationals live in third countries, the home-country ratios are overstated by a factor 1/(1 - theta_i). For diaspora-heavy nationalities (33.7% of offshore value) theta may be substantial: Pakistanis in Saudi Arabia or the UK, Syrians in Lebanon, Turkey or Europe. Pakistan's 24.4% ($5,814.6m / $23,779.6m, SBP IIP excluding gold) is thus an upper bound for Pakistani residents. The Italian compliance denominator ($640m at end-2019, Table H's Italian-national offshore value) likewise includes Italians resident in third countries.

Evidence. Paper: Table H notes p. 76; Section 3.3.2 p. 16; App. F p. 64; Section 5.1 pp. 23-26; Figs. 4-6 pp. 22-25; Section 6.1 p. 32. Earlier versions: APZO 2022 pp. 7-8; EUTO 2024 note p. 10. Adjudicator Table H parse: diaspora-13 share 33.7%; Italy offshore $638.7m.

Why this severity. The total offshore stock and the gravity result (confirmed by our re-estimation) are unaffected. But the secondary claims about holdings as shares of origin-country GDP and recorded foreign assets, and the use of nationality-based values as denominators for Italian and Spanish 'tax residents', are upper bounds of unknown size.

The authors' response. Not addressed in the 2025 version; the caveat appears in the 2022 version (pp. 7-8) and the 2024 note (p. 10).

8moderateplausible · newly identified here · data error

Journalists who worked with the same leaked files say the files sometimes list the tenant rather than the owner. Because the paper treats people whose flat is rented out as living abroad, a tenant wrongly listed as an owner would be counted as a foreign investor living abroad, which pushes the estimate up. How often this happens is unknown.

Technical detail and evidence

Why it matters, in detail. Tenant records fall into the rule-classified group by construction (residence missing plus an active Ejari lease). They therefore bias both the foreign stock (the occupant's nationality is not the owner's) and the non-resident share: they are counted as non-residents and also raise the pooled share imputed to the remaining 40%. With contamination c of foreign records, the offshore stock falls by roughly 2-5% at c = 3% and 7-14% at c = 8%. They also contaminate the renting rates (40% vs 24.8%) used to argue the residency rule is accurate, so F1's calibrated range would understate the downside.

Evidence. Sources: - OCCRP, 'What Is Dubai Unlocked? Everything You Need To Know' (2024), sections 4-5. - E. Anziano, 'L'inchiesta Dubai Unlocked', IrpiMedia, 14 May 2024. - K. Sawyer, 'Verifying Who Owns Property in Dubai Takes Lots of Data...', OCCRP Unreported (Medium), 17 July 2024 (Wayback copy). - E24 SKUP method report (2023) p. 6: according to C4ADS most of the leak came from the property register, 'but it could not be documented'. Paper: Section 3.1.1 pp. 11-12; App. D pp. 57-58; App. F p. 64. Illustrative adjudicator calculation (scripts/06): if tenant records were 3% (8%) of foreign-owner records, the offshore stock would fall by 2-5% (7-14%). The larger figure in each range applies if the true owners are Emirati.

Why this severity. This is an undisclosed caveat whose direction is known (it biases the foreign and offshore stocks upward) but whose frequency is unknown. At small contamination rates (up to ~8% of records) the headline falls by up to ~15%, which does not overturn the qualitative result; larger contamination cannot be ruled out from published information.

9minorconfirmed · newly identified here · misrepresentation of source

International statistics are in principle supposed to include homes owned abroad, but in practice they often miss them. The paper says such homes are excluded by design, which is not right, though it makes little difference to its Pakistan example.

Technical detail and evidence

Why it matters, in detail. The correct framing is that the ratio measures probably unrecorded direct-investment assets under BPM6. Overlap with recorded positions is bounded by Pakistan's $486.6m of direct investment in the UAE (at most 8.4% of $5.81bn).

Evidence. Paper: p. 26; Fig. 4c p. 22. IMF BPM6 paras 4.34-4.39. State Bank of Pakistan, IIP 2019 data tables (Table 1; direct investment by country). codex/iip_check.json; adjudicator check.

Why this severity. This is a conceptual misstatement with a small quantitative effect on the paper's showcase example: the 24.4% would still be at least about 22% even if all recorded UAE direct investment were real estate. The larger caveat for this ratio is nationality versus residence (F7).

10minorconfirmed · newly identified here · internal inconsistency

The paper sometimes quotes slightly different numbers for the same thing, for example 30% in the summary but 33% in the conclusion. These are slips a careful edit would fix; they do not change the findings.

Technical detail and evidence

Why it matters, in detail. The largest is the conclusion's 33%, 2.8 points above the 30.2% implied by the tables; the others are a few points or less, or are presentation errors.

Evidence. Pages as cited. Adjudicator renderings of Figs. 2, 3, J1 and M1, and vector extraction of Fig. 10 (logs/02_fig10.txt).

Why this severity. Editing and rounding slips; none changes a conclusion.

11minorconfirmed · newly identified here · misrepresentation of source

A handful of references and labels are wrong: a statistic attributed to the wrong group, a date off by most of a year, figure labels mixed up. They are worth correcting but do not change the results.

Technical detail and evidence

Why it matters, in detail. Fn. 32 overstates top-1% amnesty participation threefold (6.5% vs 2.2%). The remaining items are labelling, dating and citation errors.

Evidence. AJZ (2019) online appendix G, sheet 'discloser(NOR2)'; InfoLibre; AEAT; European Commission news item of 20 Dec 2022; UAE FIU, Real Estate Typology Report (2023) section 3. Paper pp. 7, 19, 22-25, 29, 32, 36, 39, 40, 47, 65, 83, 84, 87; APZO 2022 pp. 25, 27.

Why this severity. These errors do not affect the estimates; they affect the accuracy of supporting statements.

12minorconfirmed · newly identified here · speculative assumption

The figures for years before 2019 and after 2021 are partly estimated by assumption, which the appendix says but the summary does not. The overall story of rapid growth from almost nothing is not in doubt.

Technical detail and evidence

Why it matters, in detail. The 2023 level (45.4%, $214bn) and the 2021 composition are imputations. Fn. 23 flags the direction of the backcast bias: upward if Emiratis were net sellers.

Evidence. Section 3.4 pp. 17-18; G.1.2-G.1.3 pp. 67-69; fn. 23 p. 17; p. 20; Fig. 3 p. 20.

Why this severity. The qualitative growth claim is robust, since foreign ownership was legally restricted before 2006; only the precise path and the 2021 and 2023 levels are model-based.

The authors' response. Disclosed in the appendix (G.1.2, 'even more tentative'; fn. 23).

13minorconfirmed · newly identified here · documentation gap

Because the data are leaked and confidential, outsiders cannot redo the calculations. The authors could still publish their code and summary tables, which would let others check the main steps.

Technical detail and evidence

Why it matters, in detail. Only Table H and the printed figures are machine-readable. Every check of the $224bn stock, the residency split, the Norwegian distribution and the synthetic control therefore relies on digitisation.

Evidence. gabriel-zucman.eu/who-owns-offshore-real-estate/ (checked 25 Oct 2026); App. E; reference list.

Why this severity. This limits independent verification of every core number but is not itself evidence of error.

The authors' response. App. E and the 2022 Data Protection and Ethics Statement explain why the leaked microdata cannot be shared.

14minorlikely · newly identified here · overclaiming

Dubai's share of homes owned by people living abroad is clearly much higher than in the six other places the authors measured, but that is not enough to call it the most international housing market in the world.

Technical detail and evidence

Why it matters, in detail. The definitional differences mostly inflate the comparators: London's numerator includes commercial property and property held by residents through offshore companies. That makes Dubai's lead conservative. But with seven places measured (Dubai and six comparators), no world ranking is identified.

Evidence. Fig. 2 notes p. 18; p. 3; fn. 3; Atlas methodology note (Nov 2024) section 2 and Table 1.

Why this severity. The differences in definition are unlikely to close a 2.7-fold gap (30.2% vs at most 11.1%), so 'much larger than in the other large cities measured' stands; only the global superlative overreaches.

The authors' response. fn. 3 acknowledges 'likely some major U.S. cities, for which the necessary data is not currently existing'.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

The gap between Figure 10's $5.47bn and Table 1's $5.91bn is not an error, because the figure and the text use different windows (the figure note says 'between 2018 and end of 2023'). — Native PDF vector paths show that the plotted series run to 2024q1, and the quarterly gaps over 2022q1-2024q1 sum to exactly $5.47bn (2022q1-2023q4 would give $4.63bn). The figure therefore covers the same window as the text's '$5.9 billion' (Q1-2022 to Q1-2024); the discrepancy is real, and the figure note itself is wrong (kept as part of F6/F10).
No undisclosed version-instability issue is sustained, because the May 2024 note reconciles the move from the 2022 to the newer market definition. — Rejected in part. The 2024 note does explain the $533bn to $224bn redefinition. But the 2025 paper never reconciles its $67.6bn offshore figure with the 2022 'at least $146 billion ... offshore', 'twice London' headline. The 2024 note and the November 2024 Atlas note presented Dubai's 43.8% foreign share as an offshore share. And the superseded 2022 claims remain on the EU Tax Observatory repository page and the authors' project page (kept as F8, moderate).
The forced reallocation of bottom-90 owners (Norway) and the worldwide $2.7bn undeclared-income transfer are major issues. — Both issues are sustained, but at moderate rather than major severity. The abstract's wealth-gradient claim holds in the uncorrected data, and the top-0.1% and top-1% shares the paper quotes change little without the reallocation (F3). The $2.7bn is a secondary derived number whose documented corrections are about -16% to -27%, with two-sided uncertainty from the compliance transfer (F5).
Missing replication materials and the IIP accounting-boundary misstatement are moderate issues. — Both sustained, but as minor (F14, F9). The documentation gap limits verification without indicating error. The IIP misstatement changes Pakistan's 24.4% by at most about 8%, because recorded Pakistani direct investment in the UAE is only $486.6m.
No placebo or permutation inference is reported for the synthetic control, so the 'credible lower bound' is not established. — Rejected as a reason. The post-invasion gap (about $600m a quarter) is about 20 times the pre-period RMSPE (about $30m, from the figure's vector data), so statistical significance is not in doubt. The lower-bound claim fails for identification reasons (broker supply, client mix, bundled shocks), which are retained in F6.
Most Norwegian owners sit below P90 of taxable wealth, so the wealth-tax consequences of non-declaration are often nil, which weakens the 'evasion' label. — Rejected in part. Norwegian tax residents must report foreign property whatever their wealth-tax liability, and rental income and gains are taxable in Norway (there is no tax treaty with the UAE). Non-declaration is therefore non-compliance with a legal obligation; the point bears only on how much tax is at stake. The genuine overreach, generalising the rate and calling it a 'motive', is retained in F4.
The text says the data cover 'all the 172 Norwegian resident owners' although Skatteetaten counted 214 tax-liable owners. — Not a separate issue. Appendix N explicitly describes the 172 as the subset the journalists are certain are tax residents and as 'a conservative estimate'; the loose wording on p. 27 is noted within F3.
The 2009 foreign stock is 'to 22 billion' in the text but $20.5bn in Fig. 3b. — Immaterial: Fig. 3b reads about $21bn at 2009 on our rendering, a rounding difference of about $1bn.
The 2024 note's Russian estimate ($2.4bn existing + $3.9bn off-plan) contradicts the 2025 synthetic-control estimate and is not reconciled. — Rejected as an inconsistency. The two measure different things: gross Russian purchases allocated by assumption vs purchases through Russian brokers in excess of a synthetic control. Non-equality is expected; the absence of a reconciliation is a minor documentation point covered by F6 and F8.

Audit notes

What reproduced. Reproduced exactly from published material: - Table H aggregates and shares ($67.6bn offshore, $98.1bn foreign; 30.2% and 43.8% of $224bn). - Agreement between Table H and the 2024 note's city-wide 2020 totals (8 countries checked). - The 2022 workbook headline ($532.6bn; $146.2bn = 27.4%). - Pakistan's 24.4% of recorded gross foreign assets (SBP IIP). - The $2.7bn/yr arithmetic. - Figure 10's $5.47bn over 2022q1-2024q1, from vector paths. Consistent with an independent source: Norway's compliance baselines (Fig. 8: owners about 23%, value 22.7%) match Skatteetaten's counts via E24 (39 declarers: 22.7% of 172; 18.2% of 214 tax-liable owners). Reproduced approximately: the gravity relationship, with PPML distance elasticity -1.2 to -1.3 on World Bank covariates against the paper's -0.9 to -1.0; robust to dropping the largest origin countries or the GCC. Re-derived as sensitivities (aggregate approximations): - The offshore stock under alternative residency rules: -33% to +13%, or -11% to +13% when calibrated to the paper's own renting rates. - The offshore stock under tenant-record contamination: -2% to -14% at 3-8% contamination. - The $2.7bn under definitional corrections: $1.98-2.37bn. - UK compliance under matched denominators: about 19-24%. - The Norwegian distribution without forced reallocation: auditors' digitisations of Fig. 7 agree. Not reproducible: the $224bn valuation, the nationality allocation, the offshore/resident split, the 2005-2023 series, the Norwegian micro-results and the Russian synthetic control. These rest on confidential leaked ownership, transaction and rental data, and no code or intermediate aggregates are published.

Limits of this audit. Main limits of this adjudication: - The confidential leaks (C4ADS 2020/2022), the leaked transaction and Ejari rental data, the Norwegian matched data and any code are unavailable. None of the core estimation could be re-run, and all residency and contamination sensitivities are aggregate approximations: owner shares treated as value shares, nationalities pooled. - Figure-based numbers (Figs. 3, 7, J1, M1) rest on digitisation or visual reading with small reading error. Only Fig. 10 was extracted exactly from vector paths. For Fig. 7 we relied on the two auditors' mutually consistent digitisations plus a visual check. - The companion 'atlas' paper (2025a) was available only as the November 2024 methodology note. - The Times article was checked only through a secondary summary (Mosaic Chambers, an interested party). The Indian press reports and the Spanish and Italian administrative tabulations were not obtained directly. - The UAE's Financial Secrecy Index rank (13th, per the paper) was not verified. - No published critique of this paper and no author response were found (web search and the critique corpus). - The gravity check uses World Bank proxies (GDP per capita, capital-city distances, approximate language dummies) rather than the paper's CEPII and WID covariates.