← All papers
Annette Alstadsæter, Matthew Collin, Bluebery Planterose, Gabriel Zucman, Andreas Økland · EU Tax Observatory Note, May 2024
Foreign Investment in the Dubai Housing Market, 2020-2024
14 problems found: 1 major · 4 moderate · 9 minor
What the paper claims
- Foreign nationals hold around 43% of the value of Dubai's existing residential property.
- Foreign-owned residential real estate grew by about $23bn between early 2020 and early 2022, while the foreign share stayed roughly stable.
- In the two years after the February 2022 invasion of Ukraine, Russians bought (in the Summary's words, 'conservatively' and 'up to') $2.4bn of existing and $3.9bn of off-plan residential property. That is about 5-6% of the market, and roughly a 940% and 1,500% increase on the previous two years.
- Dubai's foreign ('offshore') ownership share is higher than in any other city with estimates; only London is larger in dollar terms.
- (Secondary) Country patterns: India, the UK, Pakistan, Saudi Arabia and Iran hold almost half of foreign-owned value. Conflict-affected countries hold 5-10% of their GDP in Dubai housing. Dubai holdings would raise Pakistan's recorded foreign wealth by almost 30%.
- (Secondary/policy) About 70% of foreign-held property belongs to nationals of CRS countries. Real-estate agents filed very few suspicious-transaction reports. Hence: renewed AML pressure on the UAE, extension of the CRS to real estate, and a global asset registry.
Every problem found
1majorconfirmed · newly identified here · methodology
Most people who live in Dubai are foreigners, so many 'foreign-owned' homes are simply homes that expatriates live in. The note compares its figure with other cities where only owners living abroad are counted. Counted the same way in Dubai, as the authors later did, the share is about 30% instead of 43%. That is still far higher than anywhere else, but about a third less 'offshore' wealth than the note implies.
Technical detail and evidence
Why it matters, in detail. The numerator counts all non-Emirati nationals, including UAE-resident expatriates (about 92% of the population), while the comparator cities and the GDP/IIP denominators are residence-based. This biases offshore holdings upward. The authors' 2025 residence split of the identical 2020 country totals shows the size of the bias. Holdings fall from $98.1bn to $67.6bn (-31%) and the share from 43.8% to 30.2-30.4% (-13.5 pp). Dubai drops below Singapore in dollar terms ($67.6bn vs $75.8bn, though Singapore's figure is itself nationality-based). The GDP ratios that the note highlights fall by 30-55%. Pakistan's IIP increment falls from about 28% to about 21%. The split observes residence for only about 60% of owners and treats owners who rent out all their units as non-resident, so it is itself uncertain in both directions. Golden-visa 'residents' who live elsewhere would push the true tax-relevant share above 30%.
Evidence. The authors' revised WP is Alstadsæter, Collin, Planterose, Zucman & Økland, 'Who Owns Offshore Real Estate? Evidence from Dubai', EU Tax Observatory WP 1, 16 Sept 2025. In its Tables H1-H5, the 'Total' column equals the note's Appendix C 2020 city-wide values for all 194 matched countries (max abs. difference 0.0; scripts/compare_wp2025.py). The tables split the $98.11bn into $67.62bn held by non-residents and $30.49bn (31.1%) held by UAE residents. The offshore share is 30.2% of $223.96bn (WP Fig. 2c: 30.4%), against 43.8% in note Fig. 4b, with comparator values unchanged. In WP Fig. 2b, Dubai's $67.6bn falls below Singapore's $75.8bn. Ratios to GDP, 2019, by nationality vs by residence: Syria 9.1% vs 5.0%; Yemen 7.5% vs 4.6%; Jordan 6.9% vs 3.6%; Afghanistan 4.0% vs 2.8%; Palestine 3.2% vs 1.5%. Pakistan's holdings are 72.6% offshore, so the note's 28.3% of 2021 IIP gross foreign assets ex-gold ($8.57bn / $30.25bn; IMF IIP) becomes about 20.6% on a residence basis. India: $17.27bn by nationality, of which $9.69bn is held by non-residents. The 2025 WP abstract headlines the non-resident share ('30% ($68 billion)'), and fn 4 states 'Foreign ownership includes offshore ownership and ownership by foreign nationals residing in Dubai'.
Why this severity. The literal statistic (foreign nationals, about 43%) is unaffected. As a measure of offshore (non-resident) holdings, however, it overstates by about $30bn: 31% of the level and 13-14 pp of the share. That is the sense in which the note uses it for the cross-city comparison, the GDP/IIP ratios and the tax-enforcement argument. The qualitative conclusion survives, because Dubai's like-for-like share (about 30%) is still roughly three times the highest comparator (11.1%).
The authors' response. No direct response. The successor working paper (16 Sept 2025) adopts a residence-based headline (non-residents 30%, $68bn, end-2019) and reports the nationality share (about 44%, $98bn) separately as 'foreign-held'. It does not cite the note.
2moderateconfirmed · newly identified here · speculative assumption
The note did not observe Russian purchases directly after the invasion. It assumed that Russians bought homes in the same proportion as the rental contracts they signed. That rule was off by about half even before the war, and the war itself sent many Russians to Dubai to rent rather than buy. The authors' later work, using a different method, gives figures of a similar size. So the $6.3 billion is a reasonable ballpark, but it could be noticeably higher or lower and should not be called 'conservative'.
Technical detail and evidence
Why it matters, in detail. The level in C3 is identified only by assuming that purchase shares are proportional to rental-value shares, neighbourhood by neighbourhood, and that this holds through the invasion shock. The pre-war back-test under-predicts by 45-50%. The post-war migration surge could bias the estimates the other way, and the estimated series peaks with that surge. No interval is reported. The reconstructed gross purchases implied by the 2025 broker-based estimates bracket the note's $2.4bn / $3.9bn, with the on-plan figure at the top of its range. The order of magnitude is therefore corroborated, but the claim that the estimates are a lower bound ('conservative') is not established, and 'up to' contradicts it.
Evidence. Note pp.2-3, 18-21, fn 21-22, Figs. 9-11. The adjudicator reconciled the note with the 2025 WP (scripts/russian_reconciliation.py). The 2025 WP (Table 1) uses a synthetic control to estimate ADDITIONAL purchases through Russian brokers, scaled to the city: $2.37-5.91bn in total, of which $1.22-1.47bn on-plan and $1.58-4.59bn off-plan. The WP calls these 'a credible lower-bound'. A counterfactual was added, equal to pre-invasion Russian purchases scaled by DLD market growth between Mar 2020-Feb 2022 and Mar 2022-Feb 2024 (×2.16 ready, ×4.06 off-plan). This gives a counterfactual of $0.50-0.97bn on-plan and $0.99-2.28bn off-plan. The implied gross purchases are then $1.7-2.4bn on-plan, $2.6-6.9bn off-plan and $3.9-9.2bn in total, against the note's $2.4bn / $3.9bn / $6.3bn. The 2025 WP also finds that post-invasion purchases were 'likely owned by Russians residing elsewhere', that is, not by the renters whose contracts drive the note's allocation. If the pre-war under-prediction (450 vs 230-250) had carried over, on-plan purchases would have been $4.3-4.7bn.
Why this severity. The dollar levels depend on the model, and the stated direction of bias ('conservative') is not demonstrated. But the authors' later, independent broker-based method implies gross purchases of the same order (about $4-9bn vs $6.3bn), and several indicators confirm the surge. C3 survives qualitatively; its precision is overstated.
The authors' response. The note itself calls the estimates 'preliminary and tenuous' (p.19) and lists failure modes. The Sept 2025 WP replaces the rental-share method with a broker-based synthetic control ($5.9bn of additional purchases preferred; $2.4-4.7bn in other variants) without citing or reconciling the note's figures.
3moderateconfirmed · newly identified here · selective presentation
The note says Russian buying rose about ten- to sixteen-fold. But Dubai's whole property market roughly tripled over the same period, and the starting point was the note's own estimate, which it admits was too low. Russians' share of purchases rose about four- to five-fold. That is still a big jump, but it is not tenfold.
Technical detail and evidence
Why it matters, in detail. The raw multiples (10.4× on-plan, 16× off-plan) are ratios of model estimates across two windows during which market sales rose ×2.16 and ×4.06. Market growth accounts for about one third (on-plan) to one half (off-plan) of the log increase. Against the observed pre-war on-plan level, the multiple is at most about 5.3 (raw) and at most about 2.5 (share). The broker data imply a Russian-specific factor of about 3.5-5. The 940% is internally consistent only under a proportional bias that would also make the C3 level understated by about half.
Evidence. scripts/dld_checks.py on the DLD sales mirror (sha256 9c8e3c01…): Mar 2020-Feb 2022, ready $19.5bn and off-plan $17.4bn; Mar 2022-Feb 2024, $42.2bn and $70.7bn. scripts/russian_reconciliation.py. In 2025 WP Fig. 9b, purchases through Russian brokers peak at about 10 times their Q4-2021 level, against about 2-3 times for all other brokers (text: 'approximately double'). That implies a Russian-specific surge of about 3.5-5 times.
Why this severity. This is a secondary statistic within C3, but it was widely quoted. Relative to the market, the Russian-specific surge is about 4-5 times rather than 10-16 times. A large surge remains.
4moderateconfirmed · newly identified here · documentation gap
The data behind the note are leaked or confidential, so no one outside the team can check the numbers, and several of the steps that turn raw records into the headline figures are not fully explained. Nothing suggests the numbers are wrong, but readers have to take them on trust, and other reasonable choices could move the 43% figure by a few points either way.
Technical detail and evidence
Why it matters, in detail. No result can be replicated, even in part, from public material. The undocumented choices map into roughly ±3-7 pp on the 43.8% foreign-national share: the Unknown allocation pushes it upward, and valuation of the Emirate-only stock can move it either way. They also leave the Russian allocation rule ambiguous. There is no evidence of error; the uncertainty is simply not quantified.
Evidence. Note pp.8-11, fn 10-14 and 20-22, Fig. 3 notes, Appendix C/E notes. Property counts from the adjudicator's own parse (scripts/analyze_note_tables.py). 2025 WP Section 3.2.2, fn 21, Table G1, fn 64. Adjudicator sensitivities: the foreign share is 43.8% if 'Unknown' is non-foreign, 46.9% if it is allocated pro rata, and 50.4% if it is all foreign. A ±20% valuation error on the $80bn Emirate-only stock moves the share to 40.9-47.2%.
Why this severity. Nothing indicates the numbers are wrong: the tables are internally consistent and were carried unchanged into the 2025 WP. But C1-C3 rest entirely on processing that cannot be checked, and the plausible ranges implied by undocumented choices (about 41-50% for the foreign-national share) are not reported.
The authors' response. The note explains why the data are confidential, describes how they were corroborated (Appendix A) and refers to a Data Protection and Ethics Statement. The 2025 WP adds documentation: valuation of Emirate-only areas, Table G1, and accuracy statistics in its Appendix C.
5moderatelikely · newly identified here · overclaiming
The note argues that because so much foreign and Russian money flowed into Dubai property, the UAE cannot have improved its anti-money-laundering controls. But buying property in Dubai is legal, and the note shows no data on how much of this money was dirty. Its point that agents report very few suspicious deals is fair; the stronger claim is not shown.
Technical detail and evidence
Why it matters, in detail. The inference from large foreign and Russian investment to no AML progress is not identified. Lawful demand, and purchases by nationals of a partner that does not apply sanctions, carry no information on the illicit share or on changes in enforcement. The STR evidence predates the period about which the claim is made, and the cited source's enforcement data are omitted. The policy conclusion is overstated; C1-C4 are unaffected.
Evidence. Note pp.22-23. UAE Financial Intelligence Unit (Dec 2023), 'Real Estate Money Laundering: Typologies and Patterns', Sections 2.1 and 3.2. The 2025 WP (Section 7) restates the AML point more cautiously ('expose significant money laundering risks … suggests minimal effective oversight') and drops the Russian argument.
Why this severity. This does not affect the empirical estimates (C1-C4). But the first policy recommendation in the Summary relies on an inference that the evidence cannot support: the evidence warrants 'high vulnerability and weak reporting', not 'no progress'.
The authors' response. No direct response. The 2025 WP's conclusion uses more measured language.
6minorconfirmed · newly identified here · internal inconsistency
The note gives three different growth rates for the same number: 20%, nearly 25%, and 22.8% in its own tables. The dollar increase (about $22-23 billion) is right; the percentage is just stated inconsistently.
Technical detail and evidence
Why it matters, in detail. +22.8% on existing property (tables) vs 20% (Summary) vs 'almost 25%' (text); the Summary mixes the dollar change for one universe with the percentage for another. No effect on C2's substance.
Evidence. Note pp.2, 12, 14; Appendix C/E sums (scripts/analyze_note_tables.py).
Why this severity. A difference of about 3 pp in a stated growth rate; the dollar increase and the conclusions are unchanged.
7minorconfirmed · newly identified here · internal inconsistency
One sentence misreports how much Russian-owned property under construction grew before the war ($1.4 billion instead of about $0.56 billion) and refers to a table that is not in the note. It doesn't change any result.
Technical detail and evidence
Why it matters, in detail. Overstates the pre-invasion off-plan stock change by a factor of 2.6 ($1.4bn vs $0.56bn). The tables themselves are correct, and the post-invasion estimates are computed separately.
Evidence. Appendix C, Russia: 2,786.5 → 3,692.8; Appendix E, Russia: 3,368.3 → 4,836.8 (USD m); scripts/analyze_note_tables.py.
Why this severity. A narrative slip in background to C3; no estimate depends on it.
8minorconfirmed · newly identified here · arithmetic or transcription
The note's example of how it splits sales among nationalities contains an arithmetic slip: 5% of $10 million is $0.5 million, not $2 million. It is probably a typo, but it makes the method harder to follow.
Technical detail and evidence
Why it matters, in detail. A factor-of-four inconsistency in the only formula-level description of the allocation rule; it adds to the documentation gap (F4) but probably does not affect the estimates.
Evidence. Note p.19.
Why this severity. Almost certainly a typo; there is no sign that a factor of four enters the estimates, since the back-test shows under-prediction.
9minorconfirmed · newly identified here · internal inconsistency
The paragraph describing one chart gets some details wrong. It names four countries as the 'top five', and says all of them grew faster than their economies when Syria did not. The chart itself is fine.
Technical detail and evidence
Why it matters, in detail. Misdescribes the dynamics of the GDP-scaled ratios for 2 of the 5 countries; the levels in Fig. 6 are unaffected (though see F1 on the residence basis).
Evidence. Fig. 6 (p.15) values read from a page render; Appendix C (scripts/analyze_note_tables.py).
Why this severity. Narrative errors about a secondary figure whose plotted values are correct.
10minorconfirmed · newly identified here · misrepresentation of source
The note says official statistics on a country's foreign wealth leave out property owned abroad. In principle they include it, although in practice it is often missed. For Pakistan this changes little; the bigger issue is that many Pakistani owners actually live in Dubai (see F1).
Technical detail and evidence
Why it matters, in detail. The 'increase … by almost 30 percent' is an upper bound. Double counting is capped by Pakistan's total recorded direct-investment assets ($1.95bn), small next to the residence correction (F1: about 21%).
Evidence. IMF BPM6 ¶4.34 and ¶6.31. IMF IIP for Pakistan, 2021: total assets $34.03bn, monetary gold $3.78bn, direct-investment assets $1.95bn.
Why this severity. A conceptual misstatement with little practical effect on the Pakistan example.
The authors' response. Not corrected; the 2025 WP repeats the statement.
11minorconfirmed · newly identified here · internal inconsistency
Two background statements don't match the note's own charts. Prices rose about 37%, not 25%, and the six-fold jump in Russian utility sign-ups came in late summer 2022, not by mid-2022. Neither affects the main results.
Technical detail and evidence
Why it matters, in detail. The price rise is understated by about 12 pp, and the timing of the DEWA surge is overstated by about 2-3 months; neither enters any estimate.
Evidence. Page renders of Fig. 1a (p.6) and Fig. 9a (p.19); readings accurate to about ±2 index points and ±0.3.
Why this severity. Background statistics only; nothing in C1-C4 depends on them.
12minorconfirmed · newly identified here · misrepresentation of source
The note's estimate that agents flagged only about 0.1% of property deals mixes UAE-wide reports with Dubai-only deal counts. Done consistently, the rate is still tiny, so the point holds.
Technical detail and evidence
Why it matters, in detail. The numerator (UAE-wide) and denominator (Dubai-only, including mortgages and gifts) are mismatched. Correcting it changes the rate from 0.135% to 0.12-0.17%.
Evidence. UAE FIU (Dec 2023) Section 3.2; scripts/dld_checks.py.
Why this severity. The rate is 0.12-0.17% under any combination of numerator and denominator, so the point stands.
The authors' response. Repeated unchanged in the 2025 WP (fn 50).
13minorconfirmed · newly identified here · other
There are a few labelling and dating slips, such as inconsistent descriptions of when the 2022 leaked data were collected. None changes the results.
Technical detail and evidence
Why it matters, in detail. None affects an estimate; the inconsistent dating of the snapshot makes the pre/post-invasion split of the validation window ambiguous.
Evidence. Note pp.2-3, 6-8, 12-13, 18, 29.
Why this severity. Cosmetic, apart from the ambiguous snapshot date, which slightly weakens the back-test described in F2.
14minorlikely · newly identified here · data error
The chart comparing owners' Dubai property with average incomes puts North Korea far above every other country. That is because it uses a North Korean GDP about eight times lower than standard UN figures; with standard data, North Korea would sit in the middle of the chart.
Technical detail and evidence
Why it matters, in detail. An input error in the custom GDP series (about $2bn vs about $16-17bn) makes North Korea the dominant outlier in Fig. 7. Fig. 7 and the related 2025 WP ratios are otherwise unaffected.
Evidence. scripts/figure7_check.py and scripts/figure7_un_check.py; UN AMA GDP in current USD for D.P.R. of Korea; 2025 WP Table H4.
Why this severity. The error inflates one bar about 8 times in a secondary figure; the text does not rely on it.
The authors' response. Not corrected; the 2025 WP Table H implies the same GDP of about $2bn.
Claims examined and rejected
Problems one auditor raised that did not survive checking.
- The authors' 2025 re-estimate puts Russian purchases at $2.4-5.9bn in total and only $1.2-1.5bn on-plan. This brackets the note's $6.3bn from below, falls below its $2.4bn on-plan figure, and shows the level is uncertain by at least a factor of two (I2 'Version check'; the auditor itself called the comparison indicative). — The comparison sets gross purchases against ADDITIONAL purchases through Russian brokers above a synthetic counterfactual, which the 2025 WP calls a lower bound. The adjudicator added a counterfactual equal to pre-invasion Russian purchases scaled by DLD market growth ($0.5-1.0bn on-plan, $1.0-2.3bn off-plan). The implied gross purchases ($1.7-2.4bn on-plan, $2.6-6.9bn off-plan, $3.9-9.2bn in total) are consistent with the note's figures. The 2025 estimates therefore corroborate the order of magnitude rather than suggest overstatement. The factor-of-two uncertainty is real, but it comes from the note's own back-test, not from this comparison.
- The rental-to-purchase issue is 'major'. The magnitude is illustrated by scenarios in which post-war conversion is 25% or 50% lower ($1.8bn/$2.9bn or $1.2bn/$1.95bn). — The issue is retained (F2) but graded moderate. Nothing supports the downward scenarios: the note's back-test shows the proxy under-predicting, and the authors' later broker-based estimates imply gross purchases of the same order as the note's. The magnitude is uncertain in both directions, not shown to be overstated, and the qualitative surge is corroborated.
- Overclaiming: p.12 speculates that the 2021-23 boom 'may have increased the foreign share substantially', whereas the 2025 revision finds only a rise of 'a couple of percentage points'. — An explicitly hedged conjecture about a period the note's data do not cover; no estimate or conclusion rests on it.
- Using 'on-plan' for completed property is non-standard usage (part of I15). — The note defines the term explicitly (pp.3, 5; Fig. 1 notes) and uses it consistently. This is a matter of terminology, not an error.
- The FIU source's other reporting streams (976 real-estate activity reports; a sample of 612 suspicious reports from banks and other entities) make the note's portrayal of reporting misleading (part of I5). — Those streams belong to different obligations or other sectors. The note's specific claim, that real-estate agents filed few suspicious reports, is accurate in substance. The enforcement-action figures from the same source are retained under F5.
- The IIP misdescription (I4), the Russian off-plan misstatement (I2) and the North Korea bar (I6) are each moderate. — The problems are confirmed but graded minor (F11, F7, F10). Each concerns a secondary statement or figure that no estimate depends on. For the IIP, double counting is capped by Pakistan's $1.95bn of recorded direct-investment assets. Codex's North Korea check also used city-wide averages, whereas the figure evidently uses leak-based averages, which reproduce the other bars; the conclusion is unchanged.
Audit notes
What reproduced. No core number can be reproduced from primary data. The ownership leaks, the journalists' transaction, rental and identifier files, and the DEWA series are confidential or withdrawn, and no code is published. The adjudicator wrote an independent parser for Appendix C/E and reproduced the following exactly: Fig. 3 (2020: 224.0 / 111.0 / 14.8 / 98.1; 2022: 286.1 / 145.5 / 20.1 / 120.5, USD bn); Fig. A2 (308.2 / 153.2 / 21.6 / 133.4; 386.4 / 201.8 / 24.8 / 159.8); foreign shares of 43.8% and 42.1%; growth of +$22.41bn (+22.8%); a top-5 share of 48.4% / 48.1%; Saudi +39.9%; Pakistan +6.9%; and Russian existing holdings +$906m. Both auditors' arithmetic matches to the decimal. Cross-version: the 2025 WP's Table H totals equal the note's 2020 country values for all 194 matched countries, and split them into $67.6bn non-resident and $30.5bn UAE-resident holdings. The HAL deposit (Wayback copy, 27 May 2025) is the release PDF with a HAL cover page. Against public data: the Russian estimates are 5.7% / 5.5% of DLD residential sales from March 2022 to February 2024; market sales grew ×2.16 (ready) and ×4.06 (off-plan) between the two-year windows; Pakistan's holdings are 28.3% of its 2021 IIP gross foreign assets ex-gold; St Kitts and Nevis holdings are 98% of its GDP; the CRS share is 72-77%; and 8 of the 13 checkable Figure 7 bars other than North Korea's match current UN GDP data within about 10% (the deviations of the rest are explained by rebasings or exchange-rate sensitivity). Did not reproduce: the Summary's 20% growth (the tables give 22.8%); the p.18 off-plan change ($1.4bn stated vs $0.56bn); the North Korea bar (about 1,054 with UN GDP vs 8,753); the 25% price rise (Fig. 1a shows about +37%); and 'six times by mid-2022' (Fig. 9a shows about 5). The Russian $2.4bn / $3.9bn cannot be re-estimated; the authors' later broker-based estimates, reconciled to gross terms, are consistent with them in order of magnitude.
Limits of this audit. No access to the confidential ownership, transaction, rental (Ejari) or DEWA data, so no core estimate was re-derived; all checks are internal-consistency, cross-version or plausibility checks. DLD figures come from a third-party CC-BY mirror of Dubai Pulse sales, which covers sales only and classifies hotel units as 'Other'; aggregates are therefore close rather than exact. Fig. 1a, 9a and 11 values were read from page renders. World Bank and UN GDP data are 2026 vintages, whereas the note used 2024 vintages. The residence-based benchmark used for F1 is the authors' own model-based split (residence observed for about 60% of owners; owners who rent out all their units classified as non-resident), not an independent measure. The gross-vs-additional reconciliation in F2/F3 assumes a constant Russian market share as the counterfactual and is a rough bracket, not an estimate. Industry comparators (Betterhomes rankings; CI Capital's 12% of Emaar buyers, seen only through search-index summaries of a Bloomberg report) were not verified. The Dubai Statistics Center housing-stock and population figures were not checked; the 2025 WP gives a different housing-unit series (450k → 1.02m, 2011-21). The comparator-city estimates (Atlas) were taken as published. No published critique of, or reply concerning, the note was found in the corpus, OpenAlex or a web search.