Souleymane Faye, Sarah Godar, Carolina Moura, Gabriel Zucman · EU Tax Observatory Working Paper No. 41, December 2025; R&R Journal of International Economics (per CV)

Global Offshore Wealth, 2001-2023

18 problems found: 9 moderate · 9 minor

What the paper claims

Global household offshore financial wealth has remained broadly stable relative to world GDP, and to household financial wealth, since 2001.
Offshore wealth rose sharply relative to world GDP in the 1980s and 1990s before stabilising (back-extension from capitalised missing investment income).
About USD 13 trillion of household financial assets were held offshore in 2023.
The location of offshore wealth shifted away from Switzerland towards Asian financial centres, the United Kingdom and the United States.
A growing share of offshore wealth is owned by residents of middle- and low-income countries.
Rising transparency (FATCA, CRS) had no dramatic effect on offshore holdings, 'indicating that tax evasion cannot be the primary motive for holding financial assets offshore today'.
Country-level estimates are broadly consistent with benchmark country studies and CRS data; notable country levels.

Every problem found

1moderateconfirmed · newly identified here · speculative assumption

To fill a gap in official statistics, the authors assume that three-quarters of the stock-market value of all companies registered in the Cayman Islands (mostly large Chinese tech firms) is owned from abroad in a way that ends up counted as hidden household wealth. This guess is not tested, yet it explains most of the increase in offshore wealth since about 2016 and roughly $1.7-1.9 trillion of the $13 trillion figure for 2023. Without it the series is flatter, which still fits the paper's 'stable' story but makes the 2023 total smaller.

Technical detail and evidence

Why it matters, in detail. The NFC term is a pure addition to Cayman liabilities with no asset-side counterpart, so it enters the global gap one-for-one (and total offshore wealth with the 1.22 deposit multiplier). Because the market value of Cayman-listed companies rose from about USD 0.04tn (2001) to 2-3.8tn (2019-21), it imparts a trend: its marginal effect is about 0.1% of world GDP in 2001-05 versus about 1.5% in 2019-23 (about 1.9 pp of offshore wealth/GDP after deposit scaling). Without it offshore wealth is 9.4% (2001-05) vs 10.2-10.4% (2019-23) of world GDP and USD 11.4-11.6tn in 2023. The 0.75 portfolio share, the cross-border status and the household nature of the unrecorded holders are all unvalidated.

Evidence. Paper §2.2 p.7; App. A.2 pp.38-39; fn.10 p.18. Repo raw-data/dta/KY_liab_nfc.dta (eqliab_nfc = 0.75 x mktcap_usd); 2a_do_full_matrices.do Cayman block (eqliab_KY = augmented Cayman portfolio assets - banks/insurers + eqliab_nfc, then raised to raw CPIS-derived liabilities if lower). Adjudicator scripts a05_cayman_block.py (IMF PIP CPIS current vintage, TIC SLT and Bertaut-Judson, EWN Jan-2025) and a08_combined.py: without the term the 2023 headline is USD 11.4-11.6tn (10.7-10.9% of world GDP) instead of 13.25tn (12.48%); the 2001-05 to 2019-23 change is +0.80 to +0.96 pp instead of +2.72 pp; OLS slope +0.028 to +0.042 pp/yr (HAC t 1.3-1.7) instead of +0.129 (t 3.9). Auditor A's independent bounds (2023: 10.7-10.9%) agree.

Why this severity. Lowers the 2023 level by about 13-14% and accounts for roughly two-thirds of the authors' own upward drift; but removing it makes the series flatter, i.e. closer to the paper's 'broadly stable at about 10%' claim, so the qualitative finding survives.

The authors' response. App. A.2: the 0.75 factor reflects that 'some shareholders own more than 10% of the stock (e.g., SoftBank owned 24% of Alibaba in 2022; Prosus 28% of Tencent)'; fn.15: no double counting with hedge-fund liabilities. No validation or sensitivity is reported.

2moderateconfirmed · newly identified here · speculative assumption

The authors do not measure offshore bank deposits directly; they assume deposits always make up the same 18% slice of the total. The deposit statistics they use elsewhere show deposits shrinking sharply after 2008, and earlier versions of this work used different rules. Using the actual deposit data removes most of the recent increase in offshore wealth and trims the 2023 total: this again supports 'stable', but shows the published path is partly an artefact of the assumption.

Technical detail and evidence

Why it matters, in detail. Tying deposits mechanically to the securities gap imposes the gap's growth on the ~18% deposit component, whereas every observed deposit series (SNB fiduciary, BIS non-bank liabilities of havens) is flat or falling after 2008. Re-estimating deposits from observed data raises 2001-2008 levels by about 1-1.7 pp of GDP and lowers 2019-2023 levels by about 0.3-0.9 pp; together with F1 the authors' +2.7 pp rise becomes a -0.5 pp change and the 2023 level about USD 11.0tn. The paper's only justification is that the gap 'provides an acceptable global trend across business cycles'.

Evidence. Paper §2.2 p.8; App. A.9 p.43; 7c-build-countries.do (deposits grown with the portfolio gap); AJZ2017bData.xlsx sheet T.A1 (note and deposit series 532.5bn in 2001, 1,292bn in 2008, 1,500bn in 2013); FGZ (2023) draft p.6; Zucman (2015), The Hidden Wealth of Nations, pp.39-40; Zucman (2013) QJE Table III (deposits 1,388bn, total 5,878bn). Adjudicator a03_swiss_and_deposits.py (exact Swiss rebuild; implied vs observed deposits), a02_location_rebuild.py (BIS pools), a08_combined.py. Alternatives: observed-deposits rule (Swiss fiduciary + scaled household-adjusted BIS pool, anchored at USD 1,500bn in 2013): 2001-05 to 2019-23 change +1.00 pp (slope +0.023 pp/yr, t 0.8), 2023 USD 12.5tn; AJZ-type rule with the raw BIS pool: +2.01 pp; Swiss deposit ratio: +0.95 pp. Combined with removing the NFC term (F1): -0.46 to -0.57 pp (slope -0.05, HAC t -2.1/-2.5), 2023 about USD 11.0tn (10.3% of GDP).

Why this severity. Changes the trend of the headline series by roughly 60-65% (from +2.7 to about +1.0 pp of GDP) and levels by up to about 10% (higher before 2008, lower after 2019); like F1 it biases the series upward, so correcting it reinforces rather than overturns 'broad stability'.

The authors' response. §2.2: 'This constant ratio ignores potential temporary restructuring of assets in high vs. low interest phases but we assume that the portfolio asset-liability gap provides an acceptable global trend across business cycles.'

3moderateconfirmed · previously known · speculative assumption

Apart from Switzerland, the statistics do not reliably show how much of the money held in each financial centre belongs to rich individuals, so the authors assign each centre an assumed 'household share' (20% for the UK, 70% for Hong Kong, and so on). Those guesses largely set how big each centre looks: changing the UK's share from 20% to 5% would shrink its slice from about a fifth to about a twentieth. The direction of change over time is reliable; the exact percentages are not.

Technical detail and evidence

Why it matters, in detail. Shares are proportional to h_c x D_c; with BIS non-bank liabilities dominated by NBFIs (funds, broker-dealers) and fixed or transparency-driven h_c, levels are calibrated rather than measured, and cuts in h_c mechanically reallocate wealth to other centres. The time pattern of the UK, Hong Kong and US shares is data-driven (growth of their BIS liabilities) and robust; their levels are not.

Evidence. Paper §2.3 p.9, §3.2 pp.13-14, §5.1 p.18, App. B.4 pp.49-50, fn.26 p.44; ofc_and_dep_assumptions.xlsx sheet sharehouseholddep; AJZ2017bData.xlsx sheet T.A2 (UK fraction 0.1); BIS LBS downloaded by the adjudicator (a02_location_rebuild.py; sector split lbs_sectors_adj.csv). The rebuild from current-vintage BIS data matches the published centre shares within 1.75 pp; both auditors' rebuilds give the same sensitivities (Auditor B: Europe 33.88%, Asia 32.21% with frozen fractions; Auditor A: UK 6.1-26.2% for fractions of 5-30%).

Why this severity. Moves 2023 location shares by several percentage points (UK anywhere between about 6% and 26%; Asian centres 32-45%), but the core directional claims (Switzerland down; Hong Kong, Singapore, UK and US up) survive all variants.

The authors' response. §3.2: 'The frontier between the different financial centers can be fuzzy ... so the results should not be over-interpreted.' §2.3: the BIS household/firm breakdown 'is not useful for our purpose because the use of shell companies blurs the distinction between personal and firm accounts.'

4moderateconfirmed · newly identified here · methodology

To go back to 1980 the authors convert unexplained gaps in world income statistics into a wealth estimate, but they add bank deposits that the source figures already include, inflating the pre-2001 numbers by about 19%. Where the old and new methods overlap, they do not move together. The big rise in the 1980s and 1990s still shows up, but the early numbers are less precise than the chart suggests.

Technical detail and evidence

Why it matters, in detail. A constant 18.6% scaling error affects the capitalised series from 1980 to 2001; the corrected series splices smoothly into the stock-based estimate. Annual movements of the flow-based proxy are dominated by noise in missing-income and yield measures (negative correlation with the stock series), so only the long-run rise is informative.

Evidence. Paper §4 pp.16-17, Fig. 6 note p.31; Zucman2013AppendixTables.xlsx Table A22 (col [11] = -T/S with T = 'Table A21'!T 'Portfolio and other income' and Q = 'Deposits & portfolio, stocks'); 7d-graph-paper.do lines ~613-651; adjudicator a06_historic.py.

Why this severity. Overstates pre-2001 levels by about a fifth and the proxy fails its only overlap test, but the abstract's qualitative 'sharp increase in the 1980s and 1990s' (about sixfold) survives.

5moderateconfirmed · newly identified here · misrepresentation of source

The paper says its estimate for Americans is close to what US tax authorities now see through FATCA reports. But the FATCA figure leaves out the UK, while most of the paper's US number is wealth it assigns to the UK. Comparing like with like, the paper finds only about a third to a half of what FATCA shows in classic tax havens.

Technical detail and evidence

Why it matters, in detail. A large share of 'US-owned' offshore wealth in the model is UK-hosted and derives from UK banks' liabilities to US non-banks (mostly financial institutions), which FATCA's haven aggregate does not cover; outside the UK the model captures roughly a third to a half of FATCA-reported haven wealth, pointing to under-allocation to US residents of wealth in classic havens.

Evidence. Paper §5.2 p.19, Fig. 9 p.34, Fig. 11; Johannesen, Reck, Risch, Slemrod, Guyton & Langetieg, 'The Offshore World According to FATCA', draft of 9 March 2023 (https://www.danreck.com/s/FATCA-descriptive.pdf), fn.15 p.15 (haven list) and Table 4 (USD 1,940bn); the published version (Tax Policy and the Economy 38, 2024) was not checked. Authors' output (USA by host group); BIS LBS UK liabilities to US non-banks (adjudicator download).

Why this severity. Overstates the external validation of the country allocation (C7), the main benchmark the paper offers; it does not affect the global aggregates.

The authors' response. §5.2: FATCA wealth 'might be a little inflated ... However, this is unlikely to explain a difference of 2 percentage points of GDP so our methodology probably underestimates offshore wealth of the U.S.'

6moderateconfirmed · newly identified here · methodology

To decide which country the money belongs to, the authors must guess who is behind shell companies; different reasonable guesses double or halve the figures for countries such as Russia, Greece or Portugal, and some results (Taiwan's offshore wealth equal to its whole annual GDP) look implausible. The broad finding that more offshore wealth now comes from poorer countries holds under all these guesses.

Technical detail and evidence

Why it matters, in detail. Country levels are identified only up to the shell-reallocation key and centre-specific restrictions: the preferred and FDI variants differ by 50-65% for several European countries, and single assumptions (80% Cyprus to Russia) halve or double individual countries.

Evidence. Paper §2.4 pp.10-12, §3.3 pp.15-16, App. B.1 p.44, B.3 pp.46-49; 6a-build-fiduciary.do (80% Cyprus to Russia); authors' output (RUS 'total' vs 'total_russia_adjustment'; 'total_corrected'); FGZ (2023) draft pp.9-10; adjudicator a01_load_and_headline.py and direct queries of the output file.

Why this severity. Country estimates (C7) move by factors of two or more under defensible alternatives; the aggregate and the income-group trend (C5) are not affected.

The authors' response. §3.3 pp.15-16: comparisons 'illustrate the sensitivity of individual country results to the underlying assumptions'; the FDI-corrected variant is provided; §2.4 motivates the Russia adjustment (sanctions; symmetric Russia/Cyprus movement); App. B.3 explains the Greece freeze.

7moderatelikely · previously known · overclaiming

The paper finds that money held offshore did not shrink after countries began swapping bank information, and concludes that dodging taxes can no longer be the main reason people keep money there. But the data cannot tell taxed from untaxed money, and the same pattern could arise if evaders moved to places that share less information, or if previously hidden money was declared but left abroad. The conclusion goes further than the evidence.

Technical detail and evidence

Why it matters, in detail. Identifying motives requires compliance data (declared vs undeclared holdings) or a counterfactual; the asset-liability gap is invariant to whether holdings are declared. Under plausible constructions offshore wealth fell 19-30% relative to household financial wealth between 2001-05 and 2019-23, a magnitude compatible with a sizeable transparency effect, so 'no dramatic effect' depends on the metric and on assumptions F1-F2.

Evidence. Paper §1 p.3, §3.1 p.13, §6 p.21; adjudicator a09_finwealth_share.py (WID WO-MER, net financial wealth as coded in 1d_import_auxiliary_data.do; Oct-2026 vintage); Boas, Johannesen, Kreiner, Larsen & Zucman, NBER WP 32714 (Oct 2024), pp.2-3, 29, 31; Tax Justice Network, State of Tax Justice 2024 (corpus offshore-havens-07).

Why this severity. Leaves the measured findings C1-C5 intact, but the motive claim (C6) is not supported as stated; it is an interpretive extension, hedged in the conclusion and absent from the abstract.

The authors' response. Conclusion p.21: 'This might suggest that tax evasion is not the primary reason for holding financial assets offshore today or that the transparency measures are still patchy globally.'

8moderatelikely · previously known · methodology

Swiss statistics say who holds the securities in Swiss bank accounts. Today only about a quarter belongs to private customers; most belongs to institutions such as funds. The paper counts all of it as rich individuals' wealth, so Switzerland's current share may be well below the 22% reported, which would make the shift away from Switzerland even bigger than the paper says.

Technical detail and evidence

Why it matters, in detail. The SNB custody statistics are the only direct measurement in the system; counting institutional custody (e.g., foreign funds using Swiss custodians) as household wealth inflates Swiss offshore wealth, and because Swiss wealth is subtracted from the global total before the remainder is allocated, it understates other centres. With household shares of 24-55% Switzerland hosts 7-14% rather than 22% of the 2023 total.

Evidence. Paper App. B.2 p.45; 7b-build-offshore_switzerland.do (reproduced exactly in a03_swiss_and_deposits.py); SNB data portal cube bawebedomsecwa (publishing date 2026-10-21; non-resident holders, all currencies, foreign issuers; sectors P, T1, T2), adjudicator computation; Schmitz (2021) §3.2.

Why this severity. Makes the level of Switzerland's share (22% in 2023) uncertain by a factor of up to three, but strengthens rather than weakens the direction of C4 and C5.

The authors' response. None in the paper (Schmitz 2021 is not cited); App. B.2 treats all non-resident custody holdings as offshore household wealth, as in Zucman (2013).

9moderateplausible · previously known · speculative assumption

The method counts every dollar of 'missing' securities in world statistics as belonging to wealthy households. Some may belong to companies or funds, or reflect plain statistical errors. If, say, a quarter did, household offshore wealth would be about a fifth lower than reported, although the trends would look similar.

Technical detail and evidence

Why it matters, in detail. Household attribution is the identifying restriction of the asset-liability-gap method. A constant non-household share h lowers every year's total by about 0.82h and leaves trends intact; a rising non-household share (e.g., from Cayman-listed corporate equity, F1, or, if the shift toward institutional holders visible in the Swiss custody data (F6) is representative, from institutional custody) would bias the trend upward. An official replication (Schmitz 2021) puts euro-area household offshore assets at roughly half the research estimates.

Evidence. Paper §2.1 p.5, §2.2 pp.6-8, App. A.2-A.8 pp.38-43; Schmitz, M. (2021), 'An assessment of euro area households' missing foreign assets', Proceedings of the 63rd ISI World Statistics Congress, pp.1-3 (https://isi-web.org/sites/default/files/import/pdf/144-day3-ips214-an-assessment-of-euro-area-hou.pdf); Gervais & Bui Quang (2019), IFC Bulletin 49 (read by Auditor B). Stress tests (Auditor B; reproduced in a08_combined.py): assigning 10/25/33% of the securities gap to other sectors gives 2023 USD 12.2/10.5/9.6tn (-8/-21/-27%) with the time pattern unchanged.

Why this severity. Could plausibly lower the level in all years by a tenth to a quarter, but there is no evidence the non-household share trended, so the stability finding and the location/origin directions are unaffected; the impact is not demonstrated.

The authors' response. No direct response. The corrections for reserves (China, Middle East, other), Cayman banks and insurers and UK-owned Irish fund shares (App. A.2-A.7) address part of the critique, but the paper does not cite Schmitz (2021) or bound the remaining non-household share.

10minorconfirmed · newly identified here · internal inconsistency

The paper says offshore wealth stayed around 10% of world output, but its own numbers climb to about 12% in recent years. The word 'stable' is fair compared with the huge rise before 2001, but the text makes the series sound flatter than it is.

Technical detail and evidence

Why it matters, in detail. The authors' series has a statistically significant positive trend in GDP terms (about +29% between 5-year end-period means) and a mild negative one relative to net household financial wealth; the qualitative 'no collapse' reading holds, but the text overstates flatness and reports no uncertainty.

Evidence. Paper §3.1 pp.12-13, §6 p.21, Fig. 1-2; FGZ (2023) draft p.8; authors' output; adjudicator a01_load_and_headline.py and a09_finwealth_share.py (both auditors obtain the same statistics).

Why this severity. A description problem: 'broadly stable' is defensible relative to the earlier fivefold rise, and the drift is largely due to F1-F2 and asset prices, but 'no clear trend' is contradicted by the paper's own numbers.

The authors' response. §3.1 attributes year-to-year variation to asset prices ('in years of strong stock market growth ... offshore wealth in particular tend to grow faster than GDP').

11minorconfirmed · newly identified here · internal inconsistency

The paper counts the Bahamas, Bermuda and Curacao as 'Asian' centres. Counted where they are on the map, the share held in the Americas fell rather than rose, while the rise of Asia is even bigger than reported.

Technical detail and evidence

Why it matters, in detail. Group shares are sums of centre shares; relabelling three Caribbean centres moves 5.8 pp (2001) and 0.8 pp (2023) between groups, flipping the sign of the 'American' change; individual-centre results (US +8.3 pp) are unaffected.

Evidence. Paper §3.2 p.13, Fig. 3 note p.28, App. B.5 p.51; 7c-build-countries.do (bis_asia includes AN, BM, BS, CW); adjudicator a01_load_and_headline.py; Auditor B obtains identical numbers.

Why this severity. A disclosed labelling choice; it reverses one secondary sentence in §3.2 but not the abstract's claims, and strengthens the Asian result.

The authors' response. App. B.5 explains the grouping by the lack of bilateral BIS data for most of the centres concerned.

12minorconfirmed · newly identified here · version instability

Earlier versions of the same estimates, used in the EU Tax Observatory's 2024 report, put offshore wealth in 2022 about 14% higher than this paper does, and the share of Asian centres several points higher. The paper does not explain why the numbers changed.

Technical detail and evidence

Why it matters, in detail. Successive vintages differ by methodology (deposit rule, UK-Irish fund correction, UAE) and data revisions; the earlier vintage had a stronger rise (9.4% to 12-15% of GDP in 2019-22).

Evidence. GTER 2024 figure workbook 1-offshore-wealth.xlsx (sheets DataFig1.1-1.2, dataFig1.4); FGZ (2023) draft pp.6, 8-10; authors' 2025 output; adjudicator comparison.

Why this severity. Revisions are normal in working papers and the qualitative story is unchanged, but revisions of 13-14% in levels signal level uncertainty and are not explained to users of the earlier figures.

13minorconfirmed · newly identified here · misrepresentation of source

The paper says only 17% of Danes' foreign wealth is still hidden; the study it cites puts the hidden share at about 27-30%, and its 17% refers to something narrower.

Technical detail and evidence

Why it matters, in detail. Using the lowest figure (bank non-reporting in one sample) rather than the source's headline residual non-compliance (27-30%) overstates CRS effectiveness by roughly half.

Evidence. Paper §1 p.3; Boas, Johannesen, Kreiner, Larsen & Zucman (2024) NBER WP 32714 pp.2, 29, 31; FGZ (2023) draft p.2.

Why this severity. Interpretive framing in the introduction; it supports the motive claim (F4) but no estimate depends on it.

14minorconfirmed · newly identified here · internal inconsistency

In 2023 the Cayman Islands began reporting the investments of many more of its funds. This does not change the paper's global total, but it makes the Cayman Islands look like the place where nearly half of the 'missing' wealth sits, when about a third would be the like-for-like figure.

Technical detail and evidence

Why it matters, in detail. Cayman's reported holdings of foreign securities raise both global recorded assets and FGMZ's Cayman liabilities, so by-debtor discrepancies shift from issuers held by the newly covered funds to Cayman; host-level decompositions of the gap (and the Cayman share's rise from 6% to 46%) are therefore not evidence of a change in the nature of offshore wealth.

Evidence. IMF (2025), 'Cayman Islands: Technical Assistance Report - Report on External Sector Statistics Mission (September 30-October 11, 2024)', prepared by T. Razin, paras 2 and 42 (https://www.imf.org/-/media/files/publications/tar/2025/english/tarea2025024-print-pdf.pdf); repo IMF_20241229_Table_15 CPIS files (Cayman equity 835bn Jun-2023 to 3,141bn Dec-2023; debt 1,724bn to 2,549bn); paper §5.1 p.18, Fig. A.5 p.58, App. A.2; adjudicator a04_digitize_figA5.py and a05_cayman_block.py.

Why this severity. Affects only a descriptive decomposition of where the discrepancy arises (and arguments built on it), not the aggregate; the paper's NAV-based treatment is in fact robust to the break.

15minorconfirmed · newly identified here · internal inconsistency

A number of small mistakes in the text, labels and footnotes; they do not change the results but make the paper harder to check.

Technical detail and evidence

Why it matters, in detail. None of the slips alters the implemented calculations; several (Qatar, Fig. 2 denominator, missing Tables A1-A3) matter for readers' interpretation and replication.

Evidence. Paper pp.13-20, 41-44, 51-52, Fig. 2/4/5/12 notes; repo crs_all.dta, ofc_and_dep_assumptions.xlsx, 7b-build-offshore_switzerland.do, 2a_do_full_matrices.do; both auditors' lists and adjudicator checks (a01, direct queries).

Why this severity. Presentation and documentation slips that do not change any estimate or conclusion.

16minorconfirmed · newly identified here · documentation gap

Most of the paper's materials are public, but the key intermediate tables and the exact data versions needed to rebuild the central estimate are not, so outsiders cannot fully check it.

Technical detail and evidence

Why it matters, in detail. Without the component tables the contribution of each correction to the level and trend of the gap must be inferred from the total, a digitised figure and partial public-data reconstructions.

Evidence. Paper title-page footnote p.1; repo read_me.txt; repo tree at commit 8b0e45c7 (15 Dec 2025) vs HEAD 4d9c2c66 (30 Sep 2026); both auditors' material inventories.

Why this severity. Limits verification of the gap's components but is not evidence of any error; extensive code, inputs and final outputs are public.

The authors' response. Repository read_me.txt names the data vintages and provides a contact for the non-public inputs.

17minorplausible · previously known · external inconsistency

A central-bank economist using the same basic method found the 'missing wealth' falling after 2014, while this paper finds it flat. Most of the difference comes from changes in how tax havens report their statistics, which this paper corrects for, but the paper does not explain the difference to readers.

Technical detail and evidence

Why it matters, in detail. Raw creditor-based discrepancies fall when previously unreported holders (Cayman funds, reserve managers) start reporting assets while their liabilities remain creditor-derived; FGMZ's NAV-based Cayman treatment offsets this. By rough accounting the unexplained residual is of the order of USD 0.3-0.8tn over 2014-19, within the level uncertainty of F3.

Evidence. Schmitz (2021) p.1 and fn.2; EWN January 2025 dataset; adjudicator a07_ewn_raw_gap.py; Auditor A's EWN-only cross-check (same direction).

Why this severity. The divergence is largely explained by sound homogenisation for reporting breaks; what remains is a reconciliation and disclosure gap rather than evidence that the stability finding is wrong.

18minorplausible · newly identified here · speculative assumption

To estimate how much Gulf state funds own abroad, the authors scale up their US holdings by an assumed fraction that they freeze after 2011; if Gulf funds now invest less in the US than assumed, part of what the paper calls household offshore wealth would really be government money.

Technical detail and evidence

Why it matters, in detail. Total ME assets = TIC/US share, so the gap falls one-for-one with the implied non-US holdings; a 48% to 30% change raises inferred ME assets by 60% (about USD 1.3-1.5tn in 2023).

Evidence. Paper App. A.4 p.40; 2a_do_full_matrices.do lines ~600-604; TIC SLT table 1 (repo ticdata/tic_slt_table1.xlsx); Auditor A me_us_share.py.

Why this severity. An unvalidated but plausible central value; at the extreme it lowers recent levels by about 10%, with no effect on directions.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

The whole rise of the gap relative to GDP sits in the Cayman block (Cayman discrepancy from 6% to 46% of the gap; +2.75 pp vs -0.39 pp elsewhere), showing that the Cayman treatment as a whole drives the trend and that the composition of offshore wealth shifted, which weakens the motive inference (Auditor A, I2/I4). — The accounting is correct (the adjudicator's independent digitisation of Fig. A.5 is identical), but the inference is not: the Cayman NAV block mirrors Cayman's own reported assets, which also enter the global asset side and cancel in the global gap (verified: 2022-23 Cayman assets +USD 2.10tn, NAV proxy +2.10tn, net -4bn); its growth mirrors Cayman's reported assets, including reporting expansions (hedge funds in 2015, private funds in Dec-2023), and moves discrepancy between debtors rather than adding to the total (F16). Only the listed-NFC term is a net addition, and its effect is confirmed separately (F1).
The 1990s rise in the back-extension 'depends on falling yields' and is flat at a constant 2001 yield, a weakness of the extension (Auditor A, I7). — Capitalising missing income at the contemporaneous yield is the method itself; a constant 2001 yield is not a more valid benchmark. The double counting of deposits and the failed overlap test are retained (F7).
The ECB replication (Schmitz 2021: gap down 30% in 2014-19) is a likely external inconsistency of moderate severity (Auditor A, I16). — The divergence is real (adjudicator EWN-only: -29%) but mostly explained by reporting breaks that FGMZ's method correctly neutralises (Cayman's 2015 hedge-fund coverage added USD 1.64tn of reported assets, coinciding with a 1.16tn fall in the EWN-only gap; an apparent further jump in Cayman's reported debt holdings in Dec-2018 and the documented Dec-2023 expansion) plus the NFC term counted in F1; Schmitz himself attributes the decline partly to better coverage. Retained only as a minor reconciliation issue (F11).
The UK 'validation' against Beck et al. (2024) and Milesi-Ferretti (2024) is circular (Auditor A, I5). — Only partly: the level of the UK share rests on a household fraction doubled with reference to those studies, so citing them as corroboration of the level is circular; but the UK's rise over time comes from BIS data and is not imposed by the calibration. Retained within F5 for the level only.
The constant deposit ratio is not a material concern because it does not mechanically fix wealth relative to GDP (Auditor B, checks passed / assumptions table). — True but beside the point: the rule imposes a deposit path contradicted by the observed SNB and BIS deposit series and by the rules used in AJZ (2018) and the 2023 draft, and it accounts for about 1.7 pp of the 2.7 pp rise in the authors' series (F2).
Differences between the GTER 2024 vintage and the 2025 results do not establish version instability (Auditor B, checks passed). — The revisions are large (2022 level -14%, 2015 -13%, Asian share of 2022 -5.9 pp), stem partly from undisclosed rule changes (deposits) and are not reconciled in the 2025 paper; retained as a minor issue (F14), while agreeing that they are not evidence of error.
The regional-grouping problem and the replication-package gaps are each of moderate severity (Auditor B, I3 and I6). — Accepted as real but re-graded minor: the grouping is disclosed and reverses only a secondary sentence while strengthening the Asian result (F12); the documentation gap limits verification but shows no error (F18).

Audit notes

What reproduced. Reproduced exactly from the authors' published country-by-country output (sheet ctrybyctry01-23): the Fig. 1 series and USD 13.25tn (12.48% of world GDP) in 2023; location shares; income-group shares; Fig. 5 country ratios. Reproduced exactly from raw SNB files: offshore wealth in Switzerland (all years). Reproduced closely from public current-vintage BIS data: the allocation of non-Swiss wealth across centres (within 1.75 pp; Auditor A reports an exact match with additional vintage handling). Reproduced: the Fig. 6 historic formula; the Fig. A.5 decomposition (independent digitisation, identical to Auditor A's); the Cayman block's 2022 excess over EWN (1.36tn vs 'USD 1.3 tn' in fn.10); the WID-based Fig. 2 ratio (current WID vintage; Auditor A reports agreement within about 0.1 pp with the plotted points). Not reproducible: the global portfolio gap itself and its component corrections, because the CPIS (Sep-2024), BIS (Dec-2024) and CDIS (Apr-2025) vintages are available only on request, Compustat is proprietary and the intermediate Tables A1-A3 are unpublished; the gap was assessed through the implied gap (total x 0.8195), the digitised Fig. A.5 and public-data reconstruction of the Cayman block. Sensitivities (adjudicator): removing the listed-NFC imputation gives 2023 USD 11.4-11.6tn and a 2001-05 to 2019-23 change of +0.8 to +1.0 pp (authors +2.7 pp); observed deposits give +1.0 pp; both together -0.5 pp and USD 11.0tn in 2023; household-attribution haircuts of 10-33% lower levels by 8-27%.

Limits of this audit. No Stata; the authors' pipeline was not run, and the exact input vintages (CPIS Sep-2024, BIS Dec-2024, CDIS Apr-2025) and Compustat data were not available, so the global gap and its corrections were assessed indirectly. Current-vintage BIS, CPIS and WID data were used for rebuilds and sensitivities; the location rebuild matches published shares within 1.75 pp, not exactly. The Cayman reconstruction uses raw creditor-derived holdings as in the code's floor but not the gravity-augmented holdings of Cayman securities, and US equity held by Cayman is taken from TIC SLT (2020-23) and Bertaut-Judson (2015-19). The Middle-East share sensitivity and the SNB private-customer bound for location shares were taken from Auditor A, and the household-attribution stress tests from Auditor B, after checking their order of magnitude. Milesi-Ferretti (2024) and Beck et al. (2024) were not read directly (Auditor B read Beck et al.); the published 2024 version of the FATCA paper and ECORYS (2021) were not checked; the gravity model, UAE construction and FDI-corrected variant were not audited in depth. The IMF CARTAC report was obtained via WebFetch because the IMF site refuses curl downloads.