Gabriel Zucman · Quarterly Journal of Economics 128(3): 1321-1364

The Missing Wealth of Nations: Are Europe and the U.S. net Debtors or net Creditors?

13 problems found: 3 major · 4 moderate · 6 minor

What the paper claims

Around 8% of the global financial wealth of households is held in tax havens.
Three-quarters of that offshore wealth goes unrecorded in countries' international investment positions.
Foreigners hold about $2tn of offshore wealth in Swiss banks, about one-third of the global unrecorded portfolio, with a composition 'strikingly the same' as the global gap (about half mutual fund shares).
Offshore wealth belongs mostly to residents of rich countries, in particular Europeans.
Accounting for unrecorded assets turns the eurozone, officially the world's second-largest net debtor, into a net creditor ('in all plausible scenarios').
It significantly reduces the U.S. net foreign debt.
The rich world is (likely) a net creditor: unrecorded assets are about double its recorded net debt; 'China does not own the world yet'; the rich world was a sizable creditor in the 1980s-1990s.
Flow anomalies (missing investment income, excess sales) are internally consistent with the stock anomaly.
The missing wealth is concentrated in Luxembourg, Irish and Cayman fund shares, mirroring the Swiss data, which supports the household-offshore interpretation.

Every problem found

1majorlikely · newly identified here · speculative assumption

The paper says that once hidden offshore money is counted, the eurozone stops being a net debtor and becomes a net lender to the rest of the world. That only works if eurozone residents own about half of the hidden money held outside Switzerland. The author's own later estimate is about a quarter. With that estimate the eurozone comes out roughly even, and on the paper's own assumptions it is still a net debtor in 2008. The hidden wealth clearly makes the eurozone look much less indebted, but the switch to net lender is not established.

Technical detail and evidence

Why it matters, in detail. Eurozone NFA/GDP (2001-08 mean) = -0.1148 + 0.1333a + 0.2265b, where a and b are the eurozone shares of Swiss-held and non-Swiss Omega. Break-even b = 27% at a = 0.40 (57% in 2008); common break-even share 32%. Benchmark (0.40, 0.50) = +5.2%; AJZ-2018 shares (0.364, 0.265) = -0.6% (2008: -6.9%); BCG-consistent ~28%/28% = -1.4%. The ECB baseline is itself uncertain by several pp (book- vs market-value FDI; national-sum vs aggregate), so the sign is assumption-dependent in both directions; the magnitude of the correction is not.

Evidence. Appendix Tables A27-A28 (inputs L/M/N, rows 10-17), A31; AJZ2017bData.xlsx sheet T.A3 cols D-F (13 euro-area members; Cyprus, Luxembourg, Malta are treated as havens and not allocated); QJE p.1353 (BCG 42%/60%), p.1354 and Table V note. Adjudicator scripts: scripts/nfa_check.py (out_nfa_check.txt) reproduces Table V cell by cell and the year-by-year values; out_a31_ajz.txt. Both auditors obtained the same AJZ shares (36.35%/26.52%) and the same -0.63% mean. Current EWN (Jan-2025) data put the official 2008 eurozone position at -19.2% (vs -16.8%), which makes 2008 more negative. Counter-evidence checked: Appendix fn 118 notes that the ECB series values FDI mostly at book value, and in the paper's own EWN (Aug-2009) source the sum of euro-area members' national positions averages -6.4% of GDP over 2001-07 (-1.3% with market-value FDI) against -10.7% for the ECB aggregate (scripts/ez_baseline_alt.py).

Why this severity. The improvement in the eurozone position (+11 to +16 pp of GDP under any of the shares examined) is robust, but the headline sign reversal depends on an unvalidated ownership share. Within a defensible range (the author's later shares, Europe-wide survey figures) the eurozone is roughly balanced or a small net debtor, and it is a net debtor in the latest year even at the benchmark. Not critical: the paper presents the result as conditional, and the uncertain official baseline cuts the other way.

The authors' response. The paper calls the scenarios 'thought experiments ... speculative' (p.1353) and shows the full grid. Appendix F.2 (pp.80-81) argues the result survives a worst-case allocation of other statistical errors (+2% with 50% ownership; +2.85% when correctly averaged). The author's later work (AJZ 2018; Faye, Godar, Moura & Zucman 2025) uses lower eurozone shares but does not revisit the 2013 sign claim. No direct response located.

2majorlikely · newly identified here · overclaiming

The paper suggests that, once hidden offshore money is counted, rich countries as a group are owed more by the rest of the world than they owe. Using the paper's own evidence on who owns the hidden money, rich countries come out roughly even, and slightly in debt in 2008. If the other known errors in world statistics are shared out in proportion to each country's size, as the paper says its check does, they remain net debtors. Rich countries are much less indebted than official figures show, but it is not established that they are net creditors.

Technical detail and evidence

Why it matters, in detail. Rich-world NFA/world GDP = -4.33% + 7.84% x share (2001-08 means); break-even share = 55.2% (68.8% in 2008). Raising the rich-world share of the unexplained residual from 0.5 to a proportional 0.8-0.85 lowers the A32 result by about 0.9-1.0 pp of rich-world GDP: at 60/60, -2.42% (k=0.8) vs -1.54% (k=0.5, correctly averaged) vs -1.35% (published, mis-averaged). The sign of the rich-world position therefore depends on contestable ownership and error-allocation choices.

Evidence. Appendix Table A27 cols C and R; Table A30; Table A32 col L formula (charges 0.5 x unexplained residual P); Appendix pp.79-81; QJE pp.1322, 1354, 1356, 1358-1360. EWN Aug-2009 vintage (data_lanemilesi09.dta in the replication package): rich-world share of world gross external liabilities 83.9-86.6% in 2001-07 (scripts/rich_liab_share.py). Recomputations: scripts/nfa_check.py (break-even 55.2%/68.8%; NFA at 60/62/64.6%), scripts/a31_a32_check.py (A32 60/60: published -1.35%, correct mean -1.54%, k=0.8 -2.42%, k=0.85 -2.57%). Auditor A obtained an 80% liability share from the EWN Jan-2025 vintage and the same -2.42%.

Why this severity. The claim that the rich world is a net creditor does not survive defensible assumptions. On the evidence-based ownership shares it is near balance on average and negative in 2008; under the paper's own worst-case consistency check, implemented as described, it is a net debtor. What survives is that the rich world is much less indebted than official data show (about -6% of its GDP officially, versus roughly -2.5% to +1%).

The authors' response. Appendix F.2 and pp.1358-1359 argue that the FDI and trade discrepancies most likely originate in developing countries (book-value FDI; under-recorded imports, Fisman & Wei 2004), so that the proportional case is a 'worst-case scenario'. No later response located.

3majorlikely · previously known · methodology

The paper counts all the 'missing' securities in world statistics as hidden household money. Its own appendix says part of that hole belongs to Gulf governments' investment funds, and that part is probably holdings of U.S. hedge funds that simply did not report them. Taking these out, the share of households' wealth hidden offshore is more like 6-7% than 8%. That is still a very large sum.

Technical detail and evidence

Why it matters, in detail. Household Omega = Omega - (official Gulf holdings + unreported U.S. fund-of-fund/feeder holdings of Cayman funds + SWF/reserve under-measurement). Central: 4,490 - 650 = 3,840, times (1 + 0.309) = 5,026, which is 6.8% of net (4.5% of gross) household financial wealth. Affects C1-C2 and the household reading of C9; NFA scenarios C5-C7 are affected only through who owns the residual (Gulf official holdings belong to developing countries). Offsets: zero fund liabilities assumed for the Bahamas, Isle of Man and BVI; nominee holdings booked in reporting countries; non-financial assets excluded.

Evidence. Appendix A.5 pp.24-31 and Table A8; Appendix B.3.1 pp.41-46 and Table A6; Appendix fn 8 p.9; QJE pp.1338, 1340-1341, 1349-1351, 1357. Treasury, FRBNY & Fed, Report on U.S. Portfolio Holdings of Foreign Securities as of Dec 31, 2011 (2012), pp.5-8 (Table 4) and Table 6. Sensitivity: scripts/headline_check.py (out_headline_check.txt). Central deduction $650bn (75% of the Gulf offshore portfolio public, $250bn of U.S. institutional Cayman holdings, $75bn Singapore). Range $416bn-$1,103bn, the high end adding a 95% China securities share. Household Omega falls 9-25% (central 14.5%) and the headline falls from 8.0% to 6.8% (range 6.0-7.2%) of net financial wealth. Auditor A's central -22% deducted the whole Gulf portfolio and a 95% China share, which the author's own 2025 update does not adopt for 2001-08. Auditor B gave illustrative 75% / two-thirds household shares (6.0% / 5.3%).

Why this severity. The household label is the defining feature of C1, and it rests on an assumption that the paper's own appendix partly contradicts. Documented non-household components plausibly make up about a seventh of Omega (range a tenth to a quarter; up to 30% with F13), so the household headline is about 6-7% rather than 8%. Rated major at the lower edge of that category: the qualitative finding, trillions of dollars of household wealth offshore, survives, and other omissions (TJN critique) bias total offshore wealth downward.

The authors' response. Appendix A.5 argues that the public/private distinction 'is not always clear' and that official use of offshore custodians is 'interesting per se'. The paper calls its $5.9tn 'at the low end of the scale' of existing estimates (p.1344). Faye, Godar, Moura & Zucman (2025) now remove UK residents' under-reported Irish fund shares ('these should not be part of our global offshore wealth estimate') but keep the residual approach. No direct response to Schmitz (2021) located.

4moderateconfirmed · newly identified here · internal inconsistency

The '8%' compares hidden offshore assets with households' financial assets after subtracting all their debts, including mortgages. Measured against what households actually own in financial assets, which is the natural comparison, the hidden share is about 5%. The paper's text explains its definition, but its summary drops the word 'net', and one figure caption describes the base incorrectly.

Technical detail and evidence

Why it matters, in detail. Ratio = gross offshore financial assets / (household financial assets - total household liabilities). On a gross-assets basis (Credit Suisse 'financial wealth'): 5.32% total and 4.06% unrecorded, vs 7.98% and 6.1%. The net denominator also makes the share rise mechanically with household leverage.

Evidence. Credit Suisse Global Wealth Databook 2010, Table 2-4, 2008 world row (financial wealth/adult 25,779; debts/adult 8,603; adults 4,286,507k). QJE abstract; Fig. I caption p.1323; p.1341; Table III note p.1345. WP 2011 (HAL v1) abstract; WP 2012 (HAL v3) abstract. scripts/headline_check.py: 7.98% of net vs 5.32% of gross; net = 120% and gross = 180% of world GDP.

Why this severity. Not an arithmetic error: the concept is defined in the body. But the headline ratio mixes a gross numerator with a net denominator, and the abstract wording plus a factually wrong caption invite readers to take 8% as a share of households' financial assets. On that natural reading the share is overstated by half (8.0% vs 5.3%).

5moderateconfirmed · newly identified here · data error

To show that the 'missing' securities really are households' offshore holdings, the paper says they look just like foreigners' portfolios in Swiss banks, about half of which are mutual funds. But the Swiss 'mutual fund' figure also includes a different product, structured notes. Counted separately, the Swiss portfolios are only about 38% funds and more bond-heavy, so the two look less alike. The total amount of hidden wealth is not affected.

Technical detail and evidence

Why it matters, in detail. Swiss fund share 49.7% -> 37.6%. Swiss debt share 35% -> 47% (2008) against a global-gap debt share of 37% (2013 data) or 30% (2015 revision): a 10-17 pp difference instead of 2 pp. Pre-2006 the composition differs by 9-17 pp on any classification.

Evidence. Zucman2013LivreSuisse.xlsx sheet SNB-OffshorePortfolios rows 102/114/126/138/150, with the SNB notes on category breaks (rows 20-24). Appendix Table A23 cols G-M; Table A3 Panels B-C. SNB, Banks in Switzerland 2008, Table 38b and note 7 (pp.A132-A133): non-resident holdings of collective-investment units (all issuers) CHF 671bn, below the CHF 816bn implied by the paper for foreign issuers alone (Auditor B). Zucman2015MissingWealth.xlsx Table A3 (2008 equity share 69.8%).

Why this severity. Omega and the headline are unaffected. What is weakened is one of the paper's main independent validations that the global gap is household offshore wealth: the composition match becomes a loose resemblance.

6moderateconfirmed · newly identified here · overclaiming

The paper checks its estimate of hidden wealth against another clue: the income that is paid out but never shows up as received. The two agree on the rough size, a few trillion dollars, but not on how the hidden wealth changed from year to year. The paper presents this check as stronger confirmation than it is.

Technical detail and evidence

Why it matters, in detail. VAL residual deviates from market-implied revaluation by 7-22 pp in six of seven years. D/r_recorded = $3.2-4.7tn over 2001-08 vs Omega $2.4-5.1tn, with opposite trends in 2004-07. Implies year-to-year measurement noise in Omega of the order of 10-20%.

Evidence. QJE Table IV and p.1352; Appendix Tables A3 (cols 14-15), A21, A22 (cols 7-11); Appendix D.3 and fn 94. scripts/flow_stock_check.py (out_flow_stock_check.txt) reproduces Auditor A's figures.

Why this severity. A secondary consistency claim. The flows corroborate a stock of a few trillion dollars, but not the year-to-year path or the 2001-08 growth of Omega (which feeds the 2001-08 averages in Tables V-VI and the pre-2001 extrapolation). This removes part of the independent support without changing Omega.

7moderateplausible · previously known · speculative assumption

The paper assumes every foreign-owned security in Swiss bank custody belongs to a private household. Swiss statistics classify more than half of them as held by institutions, including banks. The author argues these are mostly families' holding companies, which is plausible but not shown. So the Swiss benchmark of about $2 trillion may be too high.

Technical detail and evidence

Why it matters, in detail. s = $1.545tn could fall to about $0.7-1.1tn depending on how much of the 'institutional' category is personal holding companies versus banks/funds, i.e. Switzerland about 16-25% of Omega instead of 34%; the composition comparison (F5) is equally affected.

Evidence. QJE p.1331 and fn 11 p.1332; SNB, Banks in Switzerland 2008, Table 38a and notes 4 and 6 (pp.A130-A131); Auditor A scripts/snb_custody.py output (SNB cube bawebedomsecwa).

Why this severity. Affects the size of the Swiss benchmark and the 'one-third of the gap is in Switzerland' cross-check. Omega itself is unaffected. In Tables V-VI a smaller Swiss share would, if anything, raise the eurozone correction, because the non-Swiss share assumed is higher.

The authors' response. QJE fn 11: the breakdown 'is misleading' because the SNB counts sham holding companies as institutional investors. Later author-led work (AJZ 2018; FGMZ 2025) continues to treat Swiss custody holdings of non-residents as private wealth.

8minorconfirmed · newly identified here · methodology

One chart meant to show that Europeans hid their Swiss accounts behind shell companies after 2005 includes an adjustment by the author. He re-labelled a sudden jump in Venezuelan deposits as European money in disguise. Leaked bank files later suggested the money belonged to the Venezuelan state. The main pattern, a sharp drop in deposits held in Europeans' own names, does not depend on this.

Technical detail and evidence

Why it matters, in detail. Haven-share gain 2004-05: 8.25 pp reported vs 5.88 pp without the imputation. If the Venezuelan funds were official Venezuelan deposits, post-2005 'haven' shares are also overstated by about 1-2 pp. No effect on the 62% rich-world share for 2004.

Evidence. do_table_A24.do (read as text), section on Venezuela; Appendix Table A24 rows 2004-2005 (havens 166,646/331,766 = 50.2% -> 207,498/354,790 = 58.5%; 56.1% without the transfer); Appendix fn 108 p.75; QJE Fig. III p.1333 and p.1334. Armando.info, 'La Venezuela bolivariana confia su dinero a la banca suiza' (Poliszuk & Diaz-Struck, 8 Feb 2015), https://armando.info/la-venezuela-bolivariana-confia-su-dinero-a-la-banca-suiza/ (cached copy read); ICIJ Swiss Leaks country page for Venezuela (cached copy read).

Why this severity. It affects illustrative supporting evidence for the shell-company interpretation, not Omega, the 2004 ownership shares (Table A26) or Tables V-VI. The main European decline is unaffected. Auditor B rated it moderate; downgraded here because no core number depends on it.

The authors' response. Appendix fn 108 and the do-file comment explain the adjustment ('under strong suspicion that they correspond to the holdings of sham corporations created to avoid the directive').

9minorconfirmed · newly identified here · arithmetic or transcription

A spreadsheet formula in two appendix tables divides seven years of numbers by eight. This makes the reported averages slightly too small, but it does not change any conclusion.

Technical detail and evidence

Why it matters, in detail. Published = 0.875 x true mean; e.g. A31 50/50 +2.50% -> +2.85%; A32 60/60 -1.35% -> -1.54%.

Evidence. Zucman2013AppendixTables.xlsx sheets 'Table A31' and 'Table A32', cells D49:G52 (formulas and stored values); scripts/a31_a32_check.py.

Why this severity. A mechanical slip confined to appendix robustness tables. It shrinks all averages 12.5% toward zero, in both directions, and changes no sign.

10minorconfirmed · newly identified here · arithmetic or transcription

A handful of small errors in wording, labels and footnotes. None of them changes the paper's results.

Technical detail and evidence

Why it matters, in detail. No effect on Omega, the headline ratios or the scenario tables; (e) slightly loosens the 'one-third in Switzerland' corroboration.

Evidence. Table A3 N15; data_lanemilesi09.dta (Luxembourg/Ireland 2007); Table A26 row 2004; RawDataSources cols EK-EP; MercoPress (18 Sep 2009) and Spear's summaries of BCG Global Wealth 2009; Appendix p.40 vs Table A2 G16.

Why this severity. Presentation-level slips; none changes a conclusion.

11minorconfirmed · newly identified here · speculative assumption

The claim that three-quarters of offshore wealth is invisible in official statistics comes from assuming that the mix of bank deposits and securities everywhere is the same as in Switzerland. Later estimates suggest the figure is about right.

Technical detail and evidence

Why it matters, in detail. Unrecorded share = 1/(1 + d/s) with Swiss d/s = 0.309; d-share 18% -> 82% unrecorded, 7.4%; 40% -> 60%, 10.2%.

Evidence. Main tables workbook Table 3 (deposits = Omega x 477.7/1,544.7); QJE pp.1342-1343, fn 27; FGMZ (2025) section 2.2; scripts/headline_check.py.

Why this severity. The assumption is disclosed and later author estimates corroborate the rough split. The sensitivity of the 8% to it runs in both directions. Auditor A rated this moderate; downgraded here.

The authors' response. Disclosed in the text (pp.1342-1343): 'I assume ... that 25% of it takes the form of deposits and 75% of securities, as is the case in Switzerland.'

12minorplausible · newly identified here · methodology

Where other countries report owning more of a country's securities than that country admits issuing, the paper raises the country's figure. This is standard practice, but it can only make the 'missing' amount bigger. It also makes the claim that most countries' figures match look stronger than it is. The effect on the main result is small.

Technical detail and evidence

Why it matters, in detail. Upper bound of the bias 11.7% of 2008 Omega (headline 8.0% -> 7.1%); a likely bias of a few percent. The country-match statistic is uninformative for the imputed entries.

Evidence. Appendix pp.37-41, Table A2 col [5] (2008: 523.3; 2007: 619.2), Table A12; Tables A13-A14 col I (counts verified here); scripts/headline_check.py: without the floor the headline would be 7.1%.

Why this severity. Removing the floor entirely is not a defensible alternative, given documented under-reporting. The plausible upward bias is a fraction of the $523bn, and the concentration of the gap in fund centres is not constructed. Auditor B rated this moderate; downgraded here.

The authors' response. Appendix B.2.3 argues the cases reflect obvious deficiencies in debtor data (Italy, Canada, Germany, Cyprus), notes that EWN and the IMF made similar corrections, and argues the correction is 'on the low-end'.

13minorplausible · newly identified here · methodology

For one tax haven, the paper compares debts recorded at their original value with holdings recorded at their much lower crisis-time market value. Part of the resulting 'missing' money in 2008 may simply reflect the price fall, not hidden wealth.

Technical detail and evidence

Why it matters, in detail. Up to $241bn of the 2008 gap (headline 8.0% -> 7.6% if entirely an artefact); together with F3 the central household headline is about 6.6%.

Evidence. Appendix Table A6 lines [19]-[21] and Table A3 col G (Cayman debt gap 0 in 2003-06, 58.6 in 2007, 241.2 in 2008); Appendix p.41-42; BIS Quarterly Review, September 2025, 'Annex: Introduction to the BIS debt securities statistics' (IDS outstanding 'at face value (principal)'; holdings statistics at market value).

Why this severity. At most about 5% of 2008 Omega, and only in the benchmark year; it adds to F3's uncertainty but does not change the qualitative finding.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

Version instability (Auditor A I10): the 2008 gap was revised by -5% in the author's 2015 update, the scenario year changed from a 2005 snapshot (2011 WP) to the 2001-08 average, 'net' and the U.S. 'balanced' claim were dropped, and Switzerland's pre-crisis share is 'one-third' in the QJE but 'almost half' in FGMZ (2025). — Not undisclosed instability: the 2015 workbook and FGMZ Appendix A.9 openly document data revisions of about +/-5% (2001: +1.4%; 2008: -5.0%), and the WP-to-QJE edits are ordinary revisions; Auditor B reached the same conclusion. 'Almost half' refers to total offshore wealth including deposits: AJZ (2018) Table A1 gives Switzerland 45-49% of the total in 2004-07 and 38% in 2008, which is compatible with the QJE's 31-43% securities-only share (Table A23 col 12). The substantive parts are kept elsewhere: the dropped 'net' in F4 and the revised 2008 composition in F5.
The paper understates the China reserve uncertainty ('$100-200 billion' vs the $292bn that 15% of reserves represents) (part of A's I11). — The $100-200bn range corresponds to a 90-95% securities share; a 100% share (no deposits at all) is implausible, and the author's 2025 update keeps 85% for 2001-08. Not an error.
Central estimate of non-household components = 22% of Omega (whole $433bn Gulf offshore portfolio plus a 95% China securities share). — Partly rejected on magnitude: the appendix states that wealthy private Gulf families also hold offshore, and FGMZ (2025) retain 85% for China in 2001-08. The adjudicated central estimate is about 14.5% (range 9-25%) - see F3. The direction of the claim is upheld.
The 1980s-1990s 'large positive net position' of the rich world is an unsupported extrapolation. — Partly rejected: the author's capitalised flow anomalies (Appendix Table A22 col 11: $1.4tn in 1990 to $2.7-3.3tn in 1999-2000, roughly 6-9% of world GDP) give some support; it is retained only as a weakly supported statement within F2.
Corpus critique offshore-havens-01 (Henry 2012; TJN 2014; Murphy 2015): Zucman's estimate is far too low ($21-32tn) because nominee and shell structures and non-financial assets are invisible to the discrepancy method. — Not an error in this paper: the lower-bound and financial-only scope is acknowledged ('at the low end of the scale', p.1344 and fn 28); shells in non-reporting havens do create the gap; the nominee point applies only to structures booked in reporting jurisdictions; independent mirror-statistics estimates (Pellegrini, Sanelli & Tosti 2016: $6.1-7.1tn for 2013) are near Zucman, not Henry. Recorded as a limitation that cuts opposite to F3.
Corpus critique offshore-havens-02 (Pellegrini, Sanelli & Tosti 2016): an 8% rate of return inflates offshore tax-revenue losses. — Not applicable: the QJE paper makes no revenue estimate (it only cites Gravelle 2009 in fn 34); the critique targets Zucman (2014, 2015).
Corpus lead (offshore-havens-03): reserve assets missing from the IMF SEFER survey inflate the gap. — Does not apply generically: identifiable assets include $6.1tn of reserve and international-organisation securities in 2008 (Table A1 col 14, vs $6.2tn reserves minus deposits), with explicit estimates for China, Taiwan and Gulf reporters. The remaining China uncertainty is the paper's stated $100-200bn.
Corpus lead (offshore-havens-03 summary): the SNB holder-sector breakdown has only been published since 2016. — Factually wrong as a statement about the evidence available: SNB 'Banks in Switzerland 2008' Table 38a prints the sector split, and QJE fn 11 discusses it. The substantive point is kept as F7.

Audit notes

What reproduced. Reproduced (adjudicator scripts in final/scripts, reading the author's workbooks and replication files): (1) Omega 2001-08 from Table A3 (2,532; 2,392; 2,858; 3,316; 3,676; 3,760; 5,131; 4,490 $bn); Auditor B independently re-summed the distributed CPIS/EWN/IIP source files (2008: assets 35,607, liabilities 40,097). (2) Headline: 4,490 + 1,388 = 5,878 = 7.98% of 73,625; 76.4% unrecorded; Omega = 7.3% of world GDP. The denominator matches Credit Suisse (2010) Table 2-4 exactly (net); gross financial assets 110.5tn. (3) Tables V and VI cell by cell from A28/A29; year-by-year values; break-even shares (eurozone 27% non-Swiss at 40% Swiss; rich world 55.2%, 68.8% in 2008). (4) Alternative ownership: AJZ (2018) T.A3 euro-area shares 36.35%/26.52% (both auditors and adjudicator) -> -0.63%; U.S. -11.7%; rich world 64.6%. (5) A31/A32 averaging error (32 cells = 7/8 of true means) and A32 with proportional residual allocation (rich-world liability share 84-87% in EWN Aug-2009) -> -2.42% at 60/60. (6) Swiss composition: Table A23 'mutual fund shares' = SNB collective investments + structured products + other, exactly, in 2004-2008. (7) Flow-stock residuals and capitalised flows from Tables A3/A21/A22. (8) Venezuela reclassification effect on Fig. III (8.25 -> 5.88 pp). (9) Liability floor ($523bn, 11.7%) and zero-gap counts (53 and 123 of 238). (10) Official NFA baselines vintage-stable (EWN Jan-2025: eurozone -11.5% vs -11.5%, U.S. -18.3% vs -17.9%; Auditor A's script re-run). Not reproduced or not attempted: the Stata gravity-model imputations and 238x238 bilateral matrices (below 5% of assets per the author; neither auditor re-ran them); the Swiss fiduciary country panel (input data_fiduciary_accounts.dta is 'available upon request', not in the package); a like-for-like current-vintage world gap (Auditor A's simple EWN-2025 sum, $2.96tn for 2008, is not comparable and was not used as evidence); the BCG (2009) primary report (press summaries only).

Limits of this audit. The gravity-model imputations and bilateral matrices were not re-run, and the Swiss country-level fiduciary panel is not in the public package. The magnitudes in F3 rest on extrapolations: the public share of Gulf offshore portfolios is unknown, and U.S. institutional Cayman holdings in 2008 are scaled from the 2011 survey. The SNB holder-sector split by issuer domicile (F7) was taken from Auditor A's data-portal extraction; only the all-issuer Table 38a was verified here. The Venezuela evidence (F9) is journalistic, based on HSBC accounts; whether those funds were held as fiduciary deposits is unknown. The F13 artefact is inferred from the valuation conventions and the timing, not from security-level data. The difference between the ECB eurozone aggregate and the sum of national positions (F1 counter-evidence) was not reconciled. Later evidence (Beck et al. 2024 on UK holders of Irish/Luxembourg funds; FATCA data; Schmitz 2021) refers to post-2008 years and was given limited weight. The BCG (2009) report and the 2012 HAL v2 were not read in primary form (Auditor B read HAL v3). Two corpus summaries were found inaccurate in passing (SNB sector data date; FGMZ 2025 add rather than remove a share of Cayman-listed firms' liabilities); they do not bear on this paper's verdict.