Facundo Alvaredo, Anthony B. Atkinson, Thomas Piketty, Emmanuel Saez · Journal of Economic Perspectives 27(3): 3-20

The Top 1 Percent in International and Historical Perspective

16 problems found: 1 major · 5 moderate · 10 minor

What the paper claims

The US top 1 percent share of pre-tax market income (fiscal income per family/tax unit, including realized capital gains, excluding transfers and non-taxable fringes) more than doubled, while the P95-99 share rose only modestly.
Top-share paths diverge across rich countries: sharp U-shaped rises in English-speaking countries but L-shaped paths with much smaller rises in continental Europe and Japan; hence explanations cannot rest solely on forces common to advanced countries (technology, globalization) and institutions/policy play a key role.
US historical pattern: no secular interwar decline (15-20% outside the late-1920s bubble), 1941 about equal to 1918, a sustained fall starting with World War II; explanations must accommodate the early-twentieth-century falls seen in virtually all high-income countries.
Tax policy: top marginal tax rates moved in the opposite direction from top income shares; across 18 OECD countries changes in top rates since the early 1960s are strongly correlated with changes in top 1 percent shares.
Mechanisms: reporting/avoidance cannot account for a significant fraction of the long-run US surge; the absence of a growth-tax correlation and CEO-pay evidence favour bargaining over productive-effort responses; revenue-maximizing top rate 57%, potentially 83% if three-fifths of the response is bargaining.
Capital income: the decline of top capital incomes was the main driver of early-twentieth-century falls in top shares; its share of top incomes has fallen.
Return of inheritance: the French annual inheritance flow followed a U-shape and is back to roughly its level of a century ago; inheritance may be returning in Europe.
Private wealth/national income followed a U-shape, more marked in Europe, accounted for by beta = s/g; US wealth concentration is higher but its measured rise moderate.
The association between top labour and top capital incomes in the US has increased (fourth factor).

Every problem found

1majorconfirmed · previously known · external inconsistency

The paper's figures are correct for the income people report on tax returns, but that measure leaves out a lot of income. When the same authors later counted all national income, the US increase was about a third smaller and France's increase was much bigger. The gap between the US and continental Europe is therefore far less dramatic than the paper's numbers suggest, although the US and other English-speaking countries still saw larger rises.

Technical detail and evidence

Why it matters, in detail. Fiscal income omits corporate retained earnings, imputed rents, employee fringes and much tax-exempt capital income, and includes realized capital gains. DINA therefore raises the US 1970s base (10.4-10.7% vs 8.9-10.0%) and removes the realized-gains spike in the 2007 end year (DINA 18.4-19.9% vs 23.5% fiscal). It also raises France's measured rise, because the French fiscal base excludes much top capital income taxed at flat rates. Net effects: US x1.75-1.86 (1976-2011) instead of x2.24; US 1980-2007 +76-86% instead of +135%; US-France pp gap 3.7 instead of 11.9. Under DINA the US relative rise (+76%) is even smaller than the UK's (+117%), Australia's (+97%) and Canada's (+87%), although it remains the largest in percentage points. Critics' tax-based reconstructions (Auten & Splinter 2024: +4.4 pp of national income 1979-2019) imply smaller US rises still, so the fiscal series sits at the top of the range of credible estimates. The paper's inference that divergent paths require institutional explanations rests on a contrast whose size is materially concept-dependent. Caveat: DINA rests on imputations, and its 1940s and Japanese values are low quality.

Evidence. Fiscal (PS Jan-2013, TabFig2011prel Tables A1/A3; equals JEP DataFig1 within 0.005 pp): 8.86 -> 19.82 (x2.24, +10.96 pp) 1976-2011; +134.5% (+13.48 pp) 1980-2007; 2011 = 1.61 x 1945-49 mean. Authors' DINA (PSZ2018MainData.xlsx, 'Top 1% pre-tax share equal-split'): 10.53 -> 19.60 (x1.86, +9.07 pp) 1976-2011; 1980-2007 +86.2% (+9.19 pp); 1980-2014 10.67 -> 20.20 (+89%); 2011 = 1.33 x 1945-49 mean. Current WID (bulk 15-Oct-2026, sptincj992, p99p100), 1980-2007: US 10.42 -> 18.38 (+76%, +7.96 pp); UK +117%, Canada +87%, Australia +97%; France 7.54 -> 11.85 (+57%, +4.31 pp; WID data quality 5); Germany +38% (+3.77 pp); Japan +29% (+3.08 pp; quality 1); Sweden +73% (+5.27 pp). Mean rise: English-speaking +94% (6.95 pp) vs FR/DE/JP/SE +50% (4.11 pp). US-France gap in the pp rise: 11.86 (fiscal) vs 3.65 (DINA). Fiscal US-minus-mean(FR,DE,JP) gap 11.3 pp vs 4.2 pp under DINA. DINA 'today vs 1945-49' ratios: US 1.20, Germany 1.21, France 1.10, Japan 1.04, Sweden 1.05, but WID grades 1940s values 2-4 on its 1-5 quality scale. Adjudicator scripts: us_headline.py, psz2018_us.py, wid_concept.py (logs/logs_*.txt).

Why this severity. Under the authors' own later concept the US relative rise falls by 31-40% and the US-continental gap in the 1980-2007 rise by about two-thirds, well above the rubric's one-quarter threshold for a core magnitude. The qualitative direction survives (US and other English-speaking countries still rise much more than Germany and Japan), so the problem is not critical.

The authors' response. The JEP defines its concept explicitly (p.4; Figure 1 notes). The authors' later DINA work (Piketty, Saez & Zucman 2018, QJE) was designed to address the narrowness of fiscal income and still reports a near-doubling of the US top-1% share since 1980; PSZ dispute Auten & Splinter's smaller rise (corpus us-income-08). No response specific to this article located.

2moderateconfirmed · newly identified here · methodology

Across 18 countries, those that cut top tax rates most since the early 1960s did see the biggest rises in top income shares, and that pattern is solid. But looking only at the years since 1980, when the countries actually drifted apart, the link is weak and statistically uncertain. Japan and Sweden, for example, cut rates sharply but saw small rises. Using the authors' later, fuller income measure roughly halves the strength of the link.

Technical detail and evidence

Why it matters, in detail. Over long differences anchored in 1960-64, R2 = 0.52. With 1980-84 or later anchors, cross-country variance in rate cuts falls (s.d. 9 vs 15 pp) and the slope drops to roughly 0-0.3 with s.e. 0.22-0.29. Large cutters with small rises (Japan, Sweden, France, Netherlands) dominate, so the post-1980 cross-section offers little discriminating support for taxes as the driver of the divergence. The two-way FE 1981-2010 estimate (0.15) is significant with PSS's HC1 s.e. but not with country clustering. The 1980-84 anchor post-dates the UK's 1979 cut and part of the US 1982 cut, which partly explains the drop. On DINA shares the elasticity is 0.21. Plugged into footnote 2 this would raise the revenue-maximizing rate (tau = 1/(1+1.5*0.21) = 76%), so the issue weakens the explanatory claim, not the policy direction.

Evidence. PSS NBER-WP data (AllCountries_update_pre75.dta), MTR construction from international.do (read as text), adjudicator fig4_replicate.py: published 0.470 (0.114), R2 0.52, N 18 (exact); leave-one-out 0.42-0.54; excluding US and UK 0.49 (0.18); 13 countries with 1960-64 data 0.52 (0.12). PSS first/last-5-years rule, all 18 countries: start 1975 -> 0.35 (0.15) [end <=2004], 0.49 (0.15) [end <=2010]; start 1980 -> 0.15 (0.23), 0.28 (0.23); start 1985 -> 0.04 (0.28), 0.20 (0.29). Same 13 countries: 0.52 (0.12) from 1960-64 vs 0.15 (0.29) from 1980-84. Auditor A's strict complete-window rule reproduced exactly: 1980-84 -> 2000-04 -0.09 (0.26) N 15; 1975-79 -> 2000-04 0.25 (0.22) N 14; 1980-84 -> 2004-08 -0.20 (0.22) N 10. 1980-84 -> last five years of data (<=2010) detail: Japan -31.9 pp tax / +1.8 pp share, Sweden -29.1 / +2.8, Netherlands -20.2 / +1.8, France -17.5 / +1.4. Annual panel (PSS Table 2A reproduced exactly; pss_panel.py): 1981-2010 country FE 0.626 (HC1 0.044; country-cluster 0.131); adding year FE gives 0.154 (HC1 0.043; cluster 0.106); 1960-2010 two-way FE 0.269 (cluster 0.058). Current WID DINA shares with PSS MTRs (fig4_dina.py): 0.213 (0.089), R2 0.26; excluding US and UK 0.211 (0.153).

Why this severity. The descriptive long-run correlation stated in the paper survives every check. What is weakened is its use as evidence that taxes explain the post-1980 cross-country divergence, and the elasticity's magnitude (0.15-0.5). Post-1980 estimates are too imprecise to reject 0.47 either. PSS disclose the smaller two-way FE estimates and report an all-positive period/country bootstrap. Auditor A rated this major; downgraded because no core claim is reversed.

The authors' response. PSS (2014, p.254, fn.37): 'Estimates using both country and time fixed effects generate smaller elasticities as they rely on year-to-year variation for identification. Our analysis focuses instead on long-run effects of top tax rates.' PSS Table 2 cols 4-6 report a bootstrap over random periods of at least 17 years and random 6-18-country subsets with positive 5th-95th percentile estimates. The JEP (p.10) cautions that the correlation 'may of course reflect in part coincidence rather than causality'.

3moderateconfirmed · previously known · overclaiming

The paper suggests that when taxes fell, top earners got richer mainly by bargaining for a bigger slice rather than by producing more, so top tax rates could be as high as 83%. But the growth statistics it relies on are too noisy to tell these stories apart. The CEO comparison rests on just 14 countries in a single year, and other studies point the other way. The 83% figure follows from an assumed split, not from evidence.

Technical detail and evidence

Why it matters, in detail. With beta = pi*e1 and pi = 0.1, the HC1 interval bounds e1 <= 0.42 (s >= 0.16), while the clustered interval (18 clusters, serially correlated log-GDP levels) leaves e1 in [0, 0.80], i.e. s anywhere in [0, 1] and tau anywhere in about [57%, 100%]. The CEO tax coefficient is identified from 14 country-level tax rates, so it cannot be separated from other US-specific factors (firm scale, institutional ownership, equity-pay practice) that FFMM and Kaplan-Rauh emphasize. Time-series evidence of real effects (Mertens & Montiel Olea) directly contradicts the inference that the response is mostly zero-sum, although it estimates a short-run rather than a long-run elasticity. Footnote 2 also feeds e = 0.5 'as estimated from Figure 4' into a causal formula, after the text notes that the correlation may be coincidental.

Evidence. PSS Table 2B reproduced exactly (adjudicator pss_panel.py). Preferred 1981-2010 specification beta = 0.0083: HC1 s.e. 0.0173, 95% CI [-0.026, 0.042]; country-cluster s.e. 0.0364, CI [-0.063, 0.080]. Pure supply-side benchmark beta = pi*e1 = 0.1*0.5 = 0.05 (PSS 2014, p.257 mapping) is rejected only narrowly with HC1 (implying s >= ~0.16) and not with clustering. Long-difference growth on net-of-tax change: -0.002 (0.138). Formula: s = 0 -> 57.1%, 0.2 -> 65.7%, 0.6 -> 82.9%, 1 -> 100%. PSS (2014, p.257): 'e1 is at most 40 percent of the total elasticity. Note that the standard error around this estimate is large'. CEO evidence, PSS Table 4: 14 OECD countries, 2006; 'Identification is based on cross-sectional variation in top tax rates across countries'; s.e. clustered by country; elasticity 1.90-1.97, governance interaction -0.13 (0.057). Fernandes, Ferreira, Matos & Murphy (2013, RFS, the data source; abstract): the US premium 'is economically modest and primarily reflects the performance-based pay demanded by institutional shareholders and independent boards'. Kaplan & Rauh (2013, JEP 27(3):35-56; pre-publication text krtop.pdf, Conclusions): pay rose across private-company executives, hedge-fund/PE managers, lawyers and athletes; 'poor corporate governance cannot be more than a small part of the picture'; technology founders in the Forbes 400 are 'not directly consistent' with tax-induced bargaining. Mertens & Montiel Olea (2018, QJE; as summarized in corpus entry optimal-tax-behavior-11, not re-read here): short-run top-1% ETI about 1.5 with positive spillovers to real GDP and to the bottom 99%.

Why this severity. Bargaining is the paper's second 'factor', but it is presented as a contrast or possibility and the 83% is explicitly conditional ('could potentially'); the arithmetic is correct. The issue removes empirical support for the specific 3/5 split and for reading the growth null as robust evidence against supply-side responses. It does not show that bargaining is absent.

The authors' response. PSS (2014, pp.257-258) concede the growth estimates 'are not fully robust to the choice of time period' and that 'the standard error around this estimate is large', concluding only that low top rates do not have 'any detectable positive impact'. The JEP hedges ('Naturally, cross-country comparisons are bound to be fragile', p.11; 'could potentially', p.12). No direct reply to Kaplan & Rauh or to Mertens & Montiel Olea located.

4moderateconfirmed · newly identified here · internal inconsistency

The chart labelled as an average of France, Germany and the UK quietly adds Italy, where wealth is highest relative to income, from 1970 onward. This makes Europe's recovery in private wealth look about a sixth larger than for the three countries named. The paper's written conclusions still hold.

Technical detail and evidence

Why it matters, in detail. A changing-composition aggregate is presented as a fixed three-country average. Italy (6.71 in 2010) enters exactly when the rebound occurs, which mechanically steepens the plotted European recovery and widens the post-1990 transatlantic gap.

Evidence. AlvaredoetalJEP2013.xls, DataFig6_JEP: the 'Europe' column equals mean(Germany, France, UK) for 1870-1960 and mean(Germany, France, UK, Italy) for 1970-2010. 2010: plotted 5.448 vs stated-basis 5.027. 1950-2010 rebound +3.12 vs +2.70 (+16%). 2010 Europe-US gap 1.35 vs 0.93 (+45%). Europe first exceeds the US in 1990 as plotted vs 2000 on the stated basis. Text: 1910 6.59 ('around six'), 1950 2.32 ('less than two and a half'), 2010 5.03 on the stated basis ('more than five'), so all hold.

Why this severity. No textual statement changes, but the figure that carries the abstract's 'particularly in Europe' claim overstates the European rebound by about a sixth and the recent Europe-US gap by almost half. That is a noticeable but not transformative distortion of a core-figure magnitude; it sits at the low end of moderate.

5moderatelikely · previously known · overclaiming

The paper argues that the rise in top incomes is not mainly an artefact of the rich reporting more income to the tax authorities after tax cuts. But its test, adding capital gains, does not check the biggest such effect. After 1986, many businesses switched to a legal form whose profits appear on their owners' personal tax returns. How much of the measured rise this explains is disputed, from a modest share to about half, so the paper's dismissal is too strong.

Technical detail and evidence

Why it matters, in detail. Realized gains capture conversion of wages into gains but not the migration of business profits from the corporate to the individual base. Pass-through growth is partly real human-capital income (Smith et al. 2019), so not all of it is artefact. A reclassification-neutral benchmark (DINA with retained earnings) still shows +2.7 pp in 1986-88 and about 83% of the fiscal 1976-2011 pp rise. AS-type adjustments imply far more. The paper's statement implicitly bounds the reporting share near zero, which neither its evidence nor the later literature supports. The defensible range is a modest to substantial fraction.

Evidence. PS Jan-2013: 1976-2011 +10.96 pp incl. KG vs +9.54 pp excl. KG, so the paper's premise is true. Excl.-KG share: 1986 9.13, 1987 10.75, 1988 13.17, 1989-95 12.2-13.5; incl.-KG 1986 15.92 -> 1988 15.49 (realization timing). PSZ 2018 DINA: 1986 12.21 -> 1988 14.88 (+2.67 pp); 1976-2011 +9.07 pp vs fiscal +10.96. Kopczuk & Zwick (2020, JEP, pp.36-37, citing Cooper et al. 2016; quoted from corpus entry us-income-17): 'nearly half of the rise since 1980 in the fiscal income share of the top 1 percent comes from pass-through business'. Smith, Yagan, Zidar & Zwick (2019), as summarized by Kopczuk & Zwick (per Codex): about 30% of post-1986 pass-through growth reflects reorganization. Auten & Splinter (2024, p.2181; corpus us-income-08): one-third of the 1962-2019 fiscal-income increase occurred around TRA86. These three sources were not re-read by the adjudicator. Adjudicator us_headline.py, psz2018_us.py.

Why this severity. The logical gap is real. The magnitude at stake ranges from roughly 15-20% of the US rise (authors' DINA vs fiscal, a gap that bundles all concept differences) to about half (critics' reconstructions). This qualifies the US headline and the 'real rise' reading without overturning either.

The authors' response. Piketty & Saez's 2006 reply to Reynolds (corpus us-income-01) argues that the gains-inclusive series and the top-1% W-2 wage share (6.4% -> 11.6%, 1980-2004) both rose strongly. PS (2003) acknowledged TRA86 shifting, PSZ (2018) DINA adds retained earnings, and the JEP itself notes the 1986 jump (p.9).

6moderatelikely · previously known · external inconsistency

How unequal America was before World War II depends on how old tax records are adjusted. A 2022 re-analysis finds that top incomes were lower in the 1920s and 1930s than the paper shows and declined more gradually. On that reading the famous wartime drop was about a third smaller; the overall 'U' shape over the century remains.

Technical detail and evidence

Why it matters, in detail. Numerator adjustments that turn pre-1944 net income into AGI, missing-filer imputations, and a fixed 80%-of-personal-income denominator inflate pre-war top shares; their effect fades as filing becomes near-universal in the 1940s. The measured wartime fall is therefore exaggerated and misdated. The 1928-44 drop is 8.6 rather than 12.7 pp, and much of the decline occurs in the 1930s rather than 1941-44. This weakens inferences tying the compression to wartime progressive taxation. It does not affect any post-1960 statement.

Evidence. Geloso et al. replication package (Zenodo doi:10.5281/zenodo.6248701), DataAppendix-6-7-21.xlsx, sheet TopIncomeShares. Column V ('PS latest updated file (with full K-gains)') equals the JEP values: 1918 15.96, 1928 23.94, 1941 15.79, 1944 11.28. Column AA ('Corrected - Cumulative + full K-gains'): 14.44, 19.47, 11.98, 10.84. Interwar range (1919-40 excl. 1925-29): PS 14.83-19.29 vs GMMS 12.51-14.80. 1945-49 mean: 12.34 vs 11.24, so the 2011/late-1940s ratio rises from 1.61 to 1.76. Adjudicator geloso_compare.py; Codex reproduce.py gives the same values.

Why this severity. Affects a secondary descriptive claim (pre-war levels and the dating of the compression) and shrinks by about a third the early-century fall that the tax and capital-income narratives must explain. The U-shape, the post-1976 rise and the US-vs-late-1940s comparison are unaffected; the last is strengthened. Codex rated this major; downgraded because the affected statements are not the paper's headline findings and the reconstruction is itself contested.

The authors' response. No response by Piketty or Saez located (corpus us-income-07; none of the 29 OpenAlex works citing Geloso et al., checked 2026-10-25, is by them). Codex notes that PS's original construction notes (Comp1398.xls, CorrecDeducAdjust) document more provenance than the critics' 'educated guesses' characterization.

7minorconfirmed · newly identified here · misrepresentation of source

The paper says a US study it disagrees with ignored the growth of inherited assets after they were received. In fact that study did add a normal rate of growth. The slip affects a side argument about US data, not the French evidence on inheritance.

Technical detail and evidence

Why it matters, in detail. The 'additional bias' alleged is smaller than stated: Wolff-Gittleman capitalize at a normal return, so the residual bias concerns only above-normal returns and asset-specific gains. This modestly weakens the paper's explanation of why US survey studies find no inheritance boom; it does not bear on the French administrative flows.

Evidence. Wolff & Gittleman, BLS Working Paper 445 (Jan 2011), p.9: 'we compute the present value of all inheritances as of the survey year ... by accumulating them at a real interest rate of 3.0 percent ... then converted to 2007 dollars'; p.10: 'A normal rate of return on assets received from wealth transfers is assigned and this part of the return is counted in the inheritance portion'; p.25: 'three-percent capitalization rule'. With 3% real accumulation, a transfer is multiplied by 1.81 after 20 years and 2.43 after 30 (Codex).

Why this severity. A footnote-level misdescription used to discount contrary US evidence. The French inheritance series and the footnote's main explanation (survey under-reporting, which Wolff & Gittleman themselves acknowledge) are unaffected. Codex rated this moderate; downgraded because no finding depends on it.

The authors' response. Alvaredo, Garbinti & Piketty (2017, Economica 84:239-260, pp.254-255) later argue that self-reported US inheritance flows are too low, without repeating the zero-return characterization (Codex); no explicit correction located.

8minorconfirmed · newly identified here · misrepresentation of source

The paper cites a Danish study as support for the idea that tax cuts let top earners bargain for higher pay. That study actually found that tax cuts led to lower pre-tax pay, so it supports the general idea that pay is negotiated but not the specific mechanism the paper proposes.

Technical detail and evidence

Why it matters, in detail. KLSS identify incidence under bargaining (dw/d(1-t) < 0 for pre-tax pay), whereas the PSS e3 channel requires d(pre-tax pay)/d(1-t) > 0 through bargaining effort. Both can coexist, but the cited result is not evidence for the latter.

Evidence. JEP p.10, fn.3. KLSS (QJE 2014; NBER WP 18885) abstract: 'we find compelling evidence of a negative effect of the scheme-induced reduction in the average tax rate on pretax earnings ... rationalized by a matching frictions model with wage bargaining' (OpenAlex abstract retrieved 2026-10-25).

Why this severity. Supporting citation only; no number depends on it. It somewhat overstates the empirical backing for the specific tax-bargaining channel.

9minorconfirmed · newly identified here · internal inconsistency

Measured against household disposable income, French inheritances are almost back to their level of a century ago. Measured against total national income, the denominator the original study preferred, they are only about two-thirds of the way back. The paper uses the first without saying that it makes the recovery look more complete.

Technical detail and evidence

Why it matters, in detail. Disposable income fell from 90-95% to about 70% of national income, which mechanically raises the recent flow ratio by about 30%. On the source author's preferred denominator, the recovery to 2008 is about two-thirds complete.

Evidence. DataFig5_JEP (= Piketty 2011 Tables A4/B2). Economic/Yd: 23.96% (1910) -> 5.69% (1950), factor 4.21. Fiscal/Yd: 21.51 -> 3.83, factor 5.61. Economic/Y: 22.66 -> 4.35, factor 5.21. Fiscal/Y: factor 6.95. 2008: economic/Yd 20.88%, economic/Y 14.53%. Piketty (2011, QJE, pp.1073-74): 'Though we prefer to use the national income denominator (both for conceptual and empirical reasons)'. NBER WP 19075 rev0: '20-25 percent of national income'.

Why this severity. The figure title and note disclose the denominator, and the U-shape is unaffected; this is a wording and selective-emphasis issue about how complete the 'return' is.

10minorconfirmed · newly identified here · documentation gap

The paper's numbers for how much of the top 1%'s income came from capital can be reproduced only by mixing two slightly different ways of defining 'the top 1%'. Either way, the share of capital income fell, which is the point the paper makes.

Technical detail and evidence

Why it matters, in detail. Capital share including gains = c(1-g)+g, with c taken from A7. Using g from the gains-exclusive ranking gives 26.8% for 1987-2010; using the gains-inclusive ranking gives 33.6%. The level depends on ranking and income concept; the direction does not.

Evidence. TabFig2011prel Table A7 (P99-100 dividends + interest + rents) and Table A8 col D (panel A) / col S (panel B); adjudicator capital_share.py: mixed 50.4% -> 33.6%; consistent 47.9% -> 26.8%. PSZ2018MainData 'Share of capital in top 1% pre-tax income': 63.8% -> 55.2%.

Why this severity. The qualitative decline holds under every construction, and the paper itself warns that 'these calculations depend on the definition of taxable incomes'.

11minorconfirmed · newly identified here · documentation gap

The spreadsheet posted with the article contains some different numbers from those actually plotted in Figure 4. The figure itself is correct and can be rebuilt from the original source files.

Technical detail and evidence

Why it matters, in detail. For late-starting countries the worksheet's tax windows are misaligned with the share windows. A reader relying only on the JEP workbook would obtain an elasticity about 10% lower.

Evidence. DataFig4_JEP col I vs PSS-based values (adjudicator comparison): Portugal -13.18 vs -39.00 pp; Ireland -16.48 vs -25.40; Italy -22.38 vs -30.12; Japan -28.73 vs -31.85; Denmark -2.66 vs -6.42; Spain -1.47 vs 0.00. Worksheet log columns: 0.426 (0.132); rebuilt from endpoints: 0.419 (0.133). The sheet pairs 1960-64 tax rates with later share windows for late-starting countries.

Why this severity. Documentation only; the published figure and coefficient are correct and reproducible from the cited source. Codex rated this moderate; downgraded because nothing published is affected.

12minorconfirmed · newly identified here · arithmetic or transcription

A note under one table mixes up which column is which year; the numbers and the text are right.

Technical detail and evidence

Why it matters, in detail. A literal reading of the Panel B note would reverse the time direction of the association; the values (17% -> 27% etc.) and the p.17 discussion are correct.

Evidence. JEP p.17 Table 1 header and notes; NBER WP 19075 rev0 and v2 Table 1 notes (both '2000 (column 1) and 1980 (column 2)' for both panels).

Why this severity. Labelling only; the header, values and text read the table correctly.

13minorconfirmed · newly identified here · selective presentation

The paper quotes growth figures only for the countries that fit its main story. Computed the same way, Sweden's top share grew faster than any of them, and Japan's is well above its late-1940s level. The main contrast with France and Germany still holds.

Technical detail and evidence

Why it matters, in detail. Relative-rise statistics favour low-base countries (Sweden 1980: 4.1%). In pp terms the US (+13.5) still dominates. Japan's late-1940s base is depressed by the post-war collapse, so the 'not far from the late 1940s' benchmark is weakest for Japan.

Evidence. DataFig2AB_JEP: Sweden incl. KG 4.13 (1980) -> 9.95 (2007), +140.9% (+5.82 pp); excl. KG +70.6%; Australia +5.05 pp. Japan 1945-49 mean 7.40 -> 2010 10.44 (x1.41) vs US x1.61, France x0.92, Sweden x1.02, Germany x1.08 (vs 1950). Figure 4 data: Spain +1.10 pp; Japan -31.9 pp tax / +0.40 pp share; France -17.2 / -0.93; Netherlands -18.5 / -2.81. Adjudicator intl_fig2.py, fig4_replicate.py.

Why this severity. Descriptive nuance: the text flags Sweden and Norway as 'intermediate cases' and Figure 4 plots every country; the contrast for France and Germany is unaffected.

14minorconfirmed · newly identified here · documentation gap

The finding that top earners increasingly also hold top capital income is confirmed for 1980-2006. But a fuller later study shows the overlap was just as high in the 1960s, so 1980 was a low point rather than a typical baseline.

Technical detail and evidence

Why it matters, in detail. Because the 2000s association is only back to its 1960s level, when top shares were low, the association cannot by itself account for the level difference between the 1960s and 2000s. Its contribution is to the post-1980 rise.

Evidence. JEP references p.19 (Aaberge et al. 'Unpublished paper, not completed'); Table 1 equals Aaberge et al. (2013) draft Table 3 cell by cell. Atkinson & Lakner (2017, World Bank WPS 8268, p.14): 'in the 1960s tax units that were among the top 1% earners were among the top quintile of capitalists with an 80% probability. This number fell to less than 60% by the early-1980s, before again approaching 80% in the 2000s'; the association 'became weaker from 1966 to 1985 and then started increasing'.

Why this severity. The claimed direction (rising association since about 1980) is confirmed by the completed work; the qualification concerns the association's role as a driver of long-run levels.

The authors' response. Atkinson & Lakner (2017), the completed version co-authored by an author of this paper, reports the U-shape and attributes the post-1985 rise possibly to accumulation by high earners after top-rate cuts.

15minorconfirmed · previously known · overclaiming

The paper says a simple formula linking saving and growth explains why European wealth fell and then recovered. The source it relies on shows that swings in asset prices and wartime destruction mattered a great deal too, and that the formula fits individual countries poorly.

Technical detail and evidence

Why it matters, in detail. beta = s/g is a steady-state relation with constant net saving. Over 1910-2010 relative asset prices and war shocks dominate the swings, and with a constant gross saving rate falling g raises beta far less (Krusell & Smith 2015). The level gap between the US and Europe fits only on the three-country average.

Evidence. PZ 2013 Tables, Table 10 (shares of the 1910-50 fall): capital losses DE 40%, FR 35%, UK 50%; war destruction DE 29%, FR 27%, UK 4%. Table 9 (1980-2010 capital-gains share of real national-wealth growth): US 42%, UK 72%, FR 35%, DE about 0%. Table 2 (1970-2010 net private saving s and growth g) gives s/g: US 2.73 vs observed 2010 private wealth/NI 4.10; UK 3.34 vs 5.22; DE 6.25 vs 4.12; FR 5.04 vs 5.75 (Europe-3 mean 4.88 vs 5.03).

Why this severity. Concerns the explanatory gloss, not the descriptive U-shape; PZ's 1870-2010 decomposition does find saving responsible for the bulk of long-run accumulation (76-128%).

The authors' response. Piketty (2015, JEP 29(1), pp.81-82) acknowledges that net saving need not be constant (Codex); PZ (2014) argue that saving flows account for most long-run accumulation.

16minorconfirmed · newly identified here · arithmetic or transcription

A few quoted percentages (for the UK and New Zealand, and some dates) do not exactly match the underlying data, but none of the differences changes the story.

Technical detail and evidence

Why it matters, in detail. The UK and NZ figures cannot be traced to a specific vintage (the WTID May-2013 extract was not recovered); the UK splice adds up to about 1.1 pp to the measured rise.

Evidence. Adjudicator intl_fig2.py: UK families 1979 5.93, 1981 6.67, adults 2007 15.44; families 1989 8.67 -> adults 1990 9.80. NZ (PSS panel) 5.65 -> 8.54 (+51.2%); WID sfiinc 5.65 -> 7.76 (+37.3%). us_headline.py: trough incl. 1976 (8.86), excl. 1973 (7.74); 2011/1981 incl. 1.979. fig4_replicate.py window table.

Why this severity. None changes a conclusion; these are rounding, vintage and wording imprecisions typical of a survey.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

I2: with windows starting in 1975-79, 1980-84 or 1985-89 the Figure 4 cross-country elasticity is 'about zero or negative', so Figure 4 'gives no cross-country support for the post-1980 divergence' (rated major). — Partially rejected and downgraded (the robustness point is kept as F2, moderate). Claude's numbers reproduce exactly under its strict rule: all 5 years present in both windows, with a 2000-04 end window that catches the US 2001-02 dip. Under the PSS window rule applied to all 18 countries, 1980 and 1985 starts give +0.04 to +0.28 (insignificant) and a 1975 start gives 0.35-0.49 (significant). Post-1980 estimates cannot reject 0.47; two-way FE over 1960-2010 stay significant (0.27, cluster s.e. 0.058); and PSS disclose the smaller two-way-FE estimates (fn.37). The signs are rule-dependent, and 'about zero or negative' does not hold for the 1975-79 start.
I1: under DINA 'the US no longer stands out from Germany' relative to the late 1940s. — Narrowed, not counted separately. The computation is correct (US 1.20-1.33 vs Germany 1.21). But it rests on 1940s DINA values that WID itself grades low quality (2-4 on a 1-5 scale) and on contested allocations of 1940s retained earnings. Under the alternative pre-1960 fiscal correction of Geloso et al. the US ratio rises to 1.76. Retained only as a caveat inside F1.
I3: the clustered growth regressions leave the bargaining share s unrestricted on [0,1]. — Narrowed. True with country-clustered s.e.; with PSS's own HC1 s.e. the 95% CI [-0.026, 0.042] excludes the pure supply-side value 0.05, implying s >= about 0.16. Either way s = 0.6 is not identified; incorporated in F3 with this nuance.
I8: CEO-pay evidence is presented without its data source's (Fernandes et al. 2013) contrary interpretation (standalone minor issue). — Not retained as a standalone issue. FFMM do not study tax rates, and PSS control for firm performance and a governance index, so FFMM's governance reading of the US premium is an alternative interpretation rather than a contradiction. The underlying weakness (14-country, single-year cross-country identification) is folded into F3.
I10: the capital-income shares 'reproduce only by mixing two rankings', implying an error. — Narrowed. The published tables do not report non-gain composition under the gains-inclusive ranking, so the authors' combination is the closest available approximation to a gains-inclusive ranking (as Codex notes); the decline holds under any construction. Kept only as a minor documentation point (F10).
I16 sub-items (b) US '10 percentage point increase' vs 9.38 pp plotted; (e) France 'only 10' pp vs 11 pp; (i) US 2001 rate 38.6% vs 39.1%; (j) Europe 'around six times' vs 6.59; (k) Figure 2 note wording on Sweden's capital gains; (l) garbled sentence in the AEA web abstract. — Rounding- or wording-level deviations with no bearing on any claim; verified but not counted (the substantive sub-items are in F16).
I1 (historical compression, Geloso et al. 2022) rated major. — Claim upheld but severity downgraded to moderate (F5). The affected statements (pre-war levels, 1918 vs 1941, dating of the decline) are secondary to the paper's core findings; the U-shape and all post-1960 results are unaffected; the reconstruction is itself contested and was not re-derived from SOI scans.
I3 (Wolff-Gittleman footnote) and I6 (Figure 4 worksheet) rated moderate. — Both claims upheld but downgraded to minor (F7, F11): neither affects any published number or finding. The WG footnote is a side argument about US survey data; the worksheet mismatch leaves the published figure correct and reproducible.
Implicit in Codex's critique-coverage table: later DINA/national-income series should not be used to assess the 2013 fiscal-income magnitudes ('not misrepresented as like-for-like refutations'). — Not accepted as a reason to exclude the comparison. The DINA series are the same research group's later estimates of the same phenomenon, and the comparison tests the concept-dependence of the paper's magnitudes rather than asserting an error in the 2013 fiscal numbers. See F1.

Audit notes

What reproduced. Reproduced exactly from the authors' own sources: the US Figure 1 series and the 9% -> 20% / P95-99 +3 pp headline (PS Jan-2013 file); the 1980-2007 rises for the US, Australia and Canada; all Figure 3 statements; the Figure 4 elasticity 0.470 (0.114) and every PSS Table 2 coefficient (re-implemented in Python from the .dta, with the do-file read as text); the 57%/83% arithmetic; the Figure 5 and 6 data; and Table 1 (against the Aaberge et al. draft). The 2011 US level is consistent with IRS SOI tables (19.5% with a crude AGI interpolation vs 19.82%). Not reproduced exactly: the UK +135% (131-160% depending on base year; no 1980 observation) and the New Zealand +39% (+51% in 2012-vintage data, +37% in current WID); the capital-income shares (only with a mixed ranking: 50.4% -> 33.6%; consistent ranking 47.9% -> 26.8%); the Figure 4 companion worksheet (0.42-0.43); and the 'factor of about 5 to 6' on the plotted economic series (4.2). Robustness reproductions: Figure 4 post-1980 windows (-0.20 to +0.28, insignificant; Auditor A's strict-rule values reproduced exactly) and two-way FE (0.154 for 1981-2010); DINA-based Figure 4 0.213 (0.089), matching Auditor A, though A's WID-fiscal variant (0.387) depends on splicing WID variables (0.17-0.23 without splicing); US DINA headline x1.86 (PSZ 2018) / x1.75 (WID 2026) for 1976-2011; international DINA rises (US +76%, FR +57%, DE +38%, JP +29%, SE +73%, UK +117%, 1980-2007); and Geloso et al.'s corrected pre-1960 series (1928-44 fall 8.6 vs 12.7 pp), matching Codex.

Limits of this audit. Geloso et al.'s corrections were compared from their published outputs, not re-derived from SOI scans. DINA comparisons use the WID October-2026 bulk files and PSZ (2018) main data; DINA rests on imputations, its 1940s values and Japan's series are graded low quality by WID, and the comparison tests concept dependence rather than establishing which concept is true. PSS tax-rate inputs (e.g. Japan's 80% cap, Spain's rates) and the statutory-vs-effective top-rate gap faced by the top 1% were taken as given. The UK capital-income share (Atkinson 2007, fig. 4.11), the Earl of Derby, Kay-King and MacLeod-Malcomson quotations, and Lemieux-MacLeod-Parent were not verified. Table 1 was not recomputed from IRS public-use microdata; CEO microdata were unavailable. Kaplan & Rauh (2013) was read in its pre-publication 8-page version (archived krtop.pdf), and KLSS (2014) and FFMM (2013) at abstract level. The May-2013 WTID vintage for non-US countries was not recovered. The 834 works citing the article were not all screened, so a targeted critique outside the corpus cannot be excluded (none of the 131 corpus critiques targets this article directly).