Top Incomes and the Great Recession: Recent Evolutions and Policy Implications

Why it matters, in detail. A change of x points in a fiscal-income share equals x times (fiscal income/NI) points of NI if the non-fiscal ~37% of NI is distributed unchanged; fiscal/NI was 0.63-0.70 in 2007-10. The +16.2 pp (1977-2007, incl. realized gains, which are not NI) corresponds to 11.3% of NI; the excl.-gains shift (+13.2 pp) to 8.3%; DINA, which distributes all of pre-tax national income, gives +10.2 pp (equal-split adults) or +11.1 pp (tax units). Direction: overstatement by a factor of about 1.35-1.85. Consequences: the Section III ratio to the 2000-10 current-account deficit (5.0% of NI) is about 2x, not 4x, and the debt thought experiment's premise (half the shift, 7.5% of NI a year) becomes roughly 4-5.5% of NI a year.

Evidence. Authors' data (TabFig2010.xls, March-2012 vintage): fiscal income incl. gains $8,050bn in 2010 = 62.7% of the $12,840bn national income in the paper's own Table 3 (70.2% in 2007 using the June-2013 NIPA vintage). 30-year changes in the top-decile share: +15.9 pp (1976-2006) and +16.2 pp (1977-2007) incl. gains, but +13.3 pp (1980-2010); excl. gains +11.3 to +13.5 pp for every window ending 2004-2010. In dollars: 1977-2007 shift $1,406bn = 11.3% of 2007 NI; 1980-2010 $1,068bn = 8.3% of 2010 NI; excl. gains 8.0-8.3% of NI; 15 fiscal points in 2010 = 9.4% of NI. Authors' own later national-income series (WID DINA, Oct-2026 bulk): top-10% share of pre-tax national income +10.2 pp (equal-split adults) / +11.1 pp (tax units) 1980-2007, +10.0 / +11.2 pp 1980-2010; top-1% share 10.4% -> 18.4% (1976-2007), i.e. +77% rather than 'more than doubled'. Current-account deficit 2000-10 averaged 4.3% of GDP = 5.0% of NI (FRED IEABC, GDPA, A032RC1A027NBEA), so the domestic shift is about 1.9-2.2 times the deficit (3.0 times even with 15%). Scripts: us_checks.py, wid_checks.py, macro_checks.py.

Why this severity. The paper presents this magnitude as its key quantitative point for Section III and for the 'truly enormous' claim; in like-for-like national-income terms the 30-year shift is roughly 8-11% of NI (10-11 pp in the authors' own DINA), so the stated figure is about 35-85% too high and the comparison with the external deficit falls from four times to about two. The qualitative conclusion (a very large upward shift) and all fiscal-income share series are unaffected.

The authors' response. Piketty & Saez (20 Dec 2006 reply to Reynolds, https://eml.berkeley.edu/~saez/answer-WSJreynolds.pdf) defend fiscal income as a legitimate object; Piketty, Saez & Zucman (2018, QJE) later distributed all national income and found a smaller but still large rise. No response to this specific claim located.

Why it matters, in detail. Under beta = pi*e1, the supply-side scenario (e1 = 0.5) predicts a growth coefficient of ~0.05 on dlog(1-MTR); the cross-section s.e. is 0.08-0.14 and the clustered panel s.e. 0.036, so the data cannot reject e1 = 0 or e1 = 0.5 (power 6-10%); the adjusted cross-section point estimate (0.105) lies on the supply-side side. Since tau* = (1 + a*e3)/(1 + a*e) moves from 57% to 93% as e1 falls from 0.5 to 0.08, the evidence leaves tau* essentially unbounded within 57-93%; the 83% additionally presumes uniform externality incidence and a fixed Pareto parameter a = 1.5. 'No detectable growth effect' is being read as 'no effect'.

Evidence. Replication from the official PSS (2014) AEA package (AllCountries_update_pre75.dta; pss_regressions.py): Figure 6 elasticity 0.470 (s.e. 0.114); Figure 7(a) slope of dlog(GDP per capita) on dlog(1-MTR) -0.002 (s.e. 0.138), p = 0.99; Figure 7(b) (adjusted) +0.105 (s.e. 0.077), p = 0.19, i.e. +0.23 pp/yr per log point (95% CI -0.10 to +0.57). PSS's mapping beta = pi*e1 with pi ~ 0.1 (PSS 2014, pp. 250, 257): the US net-of-tax rate rose 1.47 log points (top MTR 85.8% -> 38.5%), the UK 1.65, Germany 0.05, so e1 = 0.47-0.5 implies only +0.12-0.16 pp/yr of extra US growth (US-Germany gap 0.12-0.16 pp/yr) against a cross-country s.d. of growth of 0.36-0.62 pp/yr; power of a 5% test against beta = 0.05 is 6-10%, and the adjusted point estimate is about twice the supply-side prediction. PSS's preferred panel (Table 2 panel B, 1981-2010) gives beta = 0.0083 (HC1 s.e. 0.017; country-clustered 0.036), i.e. e1 = 0.08 with a clustered 95% CI of about -0.6 to 0.8. Table 4's formula over e1 in [0.08, 0.5] (e = 0.5, e2 = 0): 93%, 83% (0.2), 74% (0.3), 66% (0.4), 57% (0.5). PSS (2014, pp. 257-258) call the growth regressions 'not precisely estimated', 'not fully robust' and dependent on 'a very strong identifying assumption'; the 83% also requires PSS's uniform lump-sum incidence of bargaining losses (PSS fn. 21: 'admittedly unrealistic'), and if rents come from other top earners the rate is 1/(1+a*e1) = 77% (Rothschild & Scheuer 2016).

Why this severity. The paper explicitly presents Table 4 as scenarios, reports the 57% alternative and disclaims a precise decomposition, so its conditional result stands; but the main empirical argument for favouring the bargaining scenario has negligible power, so the 83% is an illustration rather than an evidence-based estimate. Auditor A rated this major; it is graded moderate here because the paper's own hedging limits how far it undermines the stated (conditional) conclusion, and the descriptive core findings are unaffected.

The authors' response. PSS (2014, AEJ:EP, pp. 257-258) acknowledge that growth regressions are imprecise, not fully robust to the period, and rest on a strong identifying assumption, and (pp. 243-244) that the uniform-externality assumption is strong (citing Rothschild-Scheuer). The IMF text says 'We certainly do not claim that we are able to put a precise number on each channel' (p. 469). No reply to Reynolds (2019) or Mertens & Montiel Olea (2018) located.

Why it matters, in detail. Net effect of the constructions on Germany: the 1995 dip is smoothed (+1.8 pp), the 2001-07 trend is cut from +2.5 to +0.9 pp by freezing the top-1% component, the 2007 level is lowered by 1.8 pp relative to WTID (incl. gains), and the 1926-28 level is raised by 3.2-3.4 pp, turning a WTID increase of ~5.7 pp between the late 1920s and 2007 into ~0.6 pp (an 'L' instead of a partial 'U' since the 1920s). Comparing the gains-inclusive US series with gains-exclusive French/UK series adds about 4 pp to the 2007 US lead (US-France gap 16.4 -> 12.4 pp on a common excl.-gains basis). The qualitative cross-country contrast, and Figure 2's like-for-like top-1% comparison, are unaffected.

Evidence. Formulas decoded by the adjudicator's own BIFF8 decoder from PikettySaezIMF2013.xls (DataFig3 vs the WTID extract dated 06-04-2012 in DetailsDataFig3): Germany 2001-07 = top-10% incl. gains - top-1% incl. gains + the 1998 top-1% share excl. gains held fixed at 10.88% (2007 plotted 36.28% vs WTID incl. gains 38.11%; 2004 34.32% vs 32.44%; 2001-07 change +0.87 vs +2.49 pp); Germany 2008 = 2007 - 0.5 pp; France 2007 = 2006 + 0.5 pp and 2008 = 2007 - 0.5 pp (no French data after 2006); Germany 1926 = 0.015 + AVERAGE(1919-1928 values, including a 1919 Prussian value of 37.92) = 35.71% vs WTID 32.5%, with 1925, 1927-29 tied to it (1928: 35.61% vs 32.2%); Germany 1995 = AVERAGE(1992-98) = 33.17% vs WTID 31.40%; UK 1908-17 built from the top-0.05% share times its 1918 ratio, 1920-36 from the top-0.1% share plus a constant gap, 1938-48 as 1937 plus twice the change in the top-0.1% share, and an unbridged switch from tax units (1989: 34.15%) to adults (1990: 36.90%); US 1910 = average of imputed 1913-14 values (no US income-tax data exist before 1913). Effects: 2007 US-Germany gap 13.5 pp plotted vs 11.6 pp with both series incl. gains, and about 8.4 pp with both excl. gains (German value approximated by subtracting the 1998 incl./excl. gap of 0.83 pp); 1980-2007 rise US +15.1 pp (incl.) / +12.8 (excl.) vs Germany +4.6 plotted and +6.3 in WTID; Germany 1926-28 to 2007 +0.6 pp plotted vs +5.6 to +5.9 pp in WTID. The authors' current WID German fiscal series rises from 31.8% (1980) to 40.0% (2008). Scripts: xls_formula_decode.py, fig3_check.py.

Why this severity. The US/UK versus Continental divergence survives (on like-for-like WTID data the US rise is still about 2.3-2.4 times Germany's, and France is roughly flat), but the plotted gap is overstated by roughly 15-60% depending on the concept, Germany's 1920s-to-2007 rise is suppressed, and the specific 'Germany stable around or below 35 percent' description is contradicted by the authors' own input data; none of this is disclosed.

Why it matters, in detail. The capture share (s1*r - s0)/(r - 1) is very sensitive to average growth r - 1: r = 1.447 per tax unit (CPI-U-RS) gives 57.6%; r rises to 2.38 for aggregate income (tax units +64.6%), 1.56 with PCE, and 1.71 for DINA per adult, giving 34.3%, 49.5% and 29.7%. The headline thus ranges 30-58% for the same window depending on unit, deflator and income concept (72% if the window ends in 2010).

Evidence. us_checks.py: per tax unit (Table 2 definition) 57.6% / 44.7% / 64.7% for 1976-2007 / 1993-2000 / 2002-07 in both the Aug-2009 and Mar-2012 vintages; share of aggregate real income growth 34.3% / 38.3% / 50.5% (tax units +64.6% over 1976-2007); re-deflating average income with the PCE price index instead of CPI-U-RS: 49.5% / 41.8% / 60.7%; extending to 2010: 72.4% per unit, 30.0% aggregate. Authors' DINA (pre-tax national income per adult, WID): 29.7% (1976-2007), 31.0% (1993-2000), 44.0% (2002-07) (wid_checks.py).

Why this severity. The table note defines the per-family statistic correctly and the arithmetic is right, but the text's 'aggregate' wording overstates the literal statistic by about 70% (58% vs 34%) and the authors' later national-income figure is about half (30%). The qualitative point—the top 1% captured several times its ~9-10% initial share—survives every variant.

The authors' response. Saez's own 'Striking it Richer' update (2013, fn. 4) cautions that share-of-growth percentages are sensitive to the estimated growth in the number of families; no response specific to this paper.

Why it matters, in detail. Log-additive decomposition (1976-2007): dln(GDP pc) - dln(bottom-99% income per unit) = -dln(1 - s_top1) [0.56] + dln(units/population) [0.56] + residual [0.37]. The per-unit vs per-capita mismatch alone inflates the gap by ~0.56 pp/yr; a like-for-like comparison (bottom 99% vs all tax units: 0.59 vs 1.2%/yr) attributes a 0.6 pp/yr gap entirely to concentration. In absolute terms bottom-90% market income per tax unit rose 9-17% over 1980-2007 (CPI-U-RS vs PCE).

Evidence. us_checks.py, 1976-2007: real GDP per capita +2.09%/yr (FRED A939RX0Q048SBEA) vs PS bottom-99% real income per tax unit +0.59%/yr, a gap of 1.50 pp/yr = falling bottom-99% share 0.56 (38%) + tax units per person 0.56 (38%; tax units +64.6% vs population +38.3%) + residual deflator/income concept 0.37 (25%). Bottom-90% average real income per tax unit (incl. gains, CPI-U-RS): +7.9% (1976-2007), +8.8% (1980-2007), -4.8% (1980-2010, owing to -12.5% over 2007-10); with the PCE deflator +18.0%, +17.2% and +3.2%.

Why this severity. Concentration explains only about 38% of the gap the paper attributes to it, because the comparison mixes per-tax-unit and per-capita measures; the direction survives (concentration alone halved bottom-99% income growth relative to average income per tax unit, 0.59 vs 1.2%/yr). 'Poorer' is accurate only relative to the top, or for 1980-2010 after the recession, not for the boom years the paragraph describes.

The authors' response. Piketty, Saez & Zucman (2018, QJE, fn. 53) later switched to equal-split adults partly to neutralise falling household size; no response specific to this paper.

Why it matters, in detail. Footnote 6 treats the TRA86 response as transitory, but A1 shows a level shift of +4.0 pp that persists (and +2.8 pp in A3), consistent with C-to-S conversions and owner-salary payments moving income onto individual returns permanently. Part of the measured long-run response of reported top income to the net-of-tax rate is therefore e2-type shifting (Auten & Splinter 2024: one-third of the 1962-2019 fiscal top-1% rise occurs around TRA86). With e2 = 0.1-0.15, tau* falls from 83% to 73-76%; PSS's counter-argument (series incl. gains show a similar elasticity, so e2 < 0.1) only partly addresses C-to-S shifting, which raises both series.

Evidence. TabFig2010 Table A1 (top 1% excl. gains): 9.13% (1986), 10.75% (1987), 13.17% (1988), then 12.17-13.53% in 1989-95; the 4.04 pp step equals 39% of the 1976-2007 rise (7.89% -> 18.33%). Incl. gains (Table A3): 12.67% (1985), 15.92% (1986 gains surge), 12.66% (1987), 15.49% (1988), 14.3-14.5% (1989-90). With e1 = 0.2 and a long-run e2 of 0.10-0.15 instead of 0, Table 4's formula gives 73-76% instead of 83%. Dropping the US entirely from Figure 6 gives 0.42 (0.13) (pss_regressions.py).

Why this severity. A known and partly accepted critique: reclassification inflates the fiscal-series rise (up to ~40% of the excl.-gains top-1% rise if the whole step were reclassification, which is an upper bound) and implies a non-zero long-run e2, but the direction of C1 survives (DINA, which imputes retained earnings, still shows the top-1% share up ~77% over 1976-2007) and the cross-country elasticity is little affected.

The authors' response. Piketty & Saez (2003) acknowledged TRA86 income shifting; PSS (2014, pp. 248-249 and fn. 29) argue the long-run shifting elasticity is small (e2 < 0.1) because top shares including realized gains show the same tax correlation; PSZ (2018) DINA adds retained earnings.

Why it matters, in detail. GMMS revise the numerator harmonisation (net income to AGI) and the pre-1944 denominator; the denominator mechanism alone moves the 1928 top-1% share by ~0.7 pp per 2.7-point change in the 80% assumption (Codex's sensitivity). Their preferred series lowers the 1928 top-decile share by 6.8 pp and the top-1% share by 4.5 pp, so 2000s top shares exceed rather than merely match the pre-Depression peak, and the 'two highest years were 1928 and 2007' coincidence (C6) no longer holds. Status in the corpus: unresolved; the alternative also rests on assumptions.

Evidence. GMMS replication output (DataAppendix-6-7-21.xlsx, sheet Table2, Zenodo 6248701): 1928 top 10% / top 1% = 42.47% / 19.47% versus 49.29% / 23.94% in the paper's vintage; 1944 = 31.25% / 10.84% versus 32.51% / 11.28%. The 1928-44 top-1% decline shrinks from 12.7 to 8.6 pp (-32%) and the top-decile decline from 16.8 to 11.2 pp (-33%). Under GMMS every year from 1996 on (PS incl. gains, e.g. 1996: 43.5%) exceeds the 1928 level. The adjudicator verified the critic's preferred numbers but did not rebuild the historical series.

Why this severity. If correct, the 1920s peak is about 7 points lower and the wartime compression about a third smaller, which weakens the 1928-vs-2007 comparisons and the policy-shock narrative, but the U-shape survives and no post-1945 or 2008-10 result is affected.

The authors' response. No response by Piketty or Saez located (corpus and web search).

Why it matters, in detail. The 'larger than' result is PSS's US time-series finding (significant at p = 0.05-0.06 for the period difference), not B&M's; the cross-country claims extrapolate from a 2006 level cross-section with country-level tax variation and few clusters. Effect on C5: removes the international/UK support for e3 > 0 claimed in the text; no effect on the estimates.

Evidence. Bertrand & Mullainathan (2001, QJE 116(3), abstract): pay 'responds as much to a lucky dollar as to a general dollar'. PSS (2014, Table 3) find beta_IV > beta_OLS for 1987-2010 for both performance measures (e.g. 0.95 vs 0.37 for market value) but in 1970-86 only for net income, using US data only; no UK or cross-country pay-for-luck evidence is reported. PSS's international evidence (Table 4) is a single 2006 cross-section of pay levels in 13-14 countries with sales and industry dummies as controls, reporting no comparison of their importance with tax rates and no changes over time. Fernandes, Ferreira, Matos & Murphy (2013, RFS; the source of PSS's data) find the US pay premium 'economically modest' and that US and non-US CEO pay 'has largely converged in the 2000s'.

Why this severity. Narrows one strand of the micro evidence for bargaining (it is US-only and partly PSS's own result) without changing any number; the paper's conditional policy result does not depend on it.

Why it matters, in detail. Closely-held-business owners' income is largely pass-through profit (partly the post-TRA86 reclassification, F6) and is not set by compensation committees; the salaried-executive channel invoked in the text accounts for about 8% of the rise, finance for 25%.

Evidence. BCH Table 7 (top 0.1%, excl. gains): share 2.83% (1979) -> 7.34% (2005), +4.51 pp; the four groups +3.15 pp (70%; 65% non-finance, 35% finance as the paper's fn. 5 says). Of the +4.51 pp, salaried non-finance executives/managers/supervisors contribute +0.38 pp (8%), the same occupations in closely held businesses +1.67 pp (37%), and finance +1.11 pp (25%). BCH Table 1: arts, media and sports 3.1% of top-0.1% taxpayers (as quoted).

Why this severity. The quotation is accurate; only its relevance to the compensation-committee bargaining channel is overstated, which does not affect any estimate.

Why it matters, in detail. Capital-income reporting ratio 22.8% (aggregate) vs 21.5% (unweighted mean of components); wage ratio 83% (fiscal wages incl. pensions over NIPA compensation) vs 91% (wages over wage accruals). The gap between wage and capital reporting (~60-70 points) is robust.

Evidence. Table3!B7 = D7/E7, where D7 is the IRS sum (376.95bn) and E7 = (L9+L10+L12)/3 is the unweighted mean of three component ratios from an external workbook. In the June-2012 NIPA vintage (SCB Table 1.12) rental income 350.2 + net interest 564.3 + net dividends 737.3 = 1,651.8bn, giving 22.8% rather than 21.5%. National income 12,840.1, compensation 7,971.4, proprietors' income 1,036.4 and undistributed profits 651.7 match NIPA exactly. IRS 'wage income' (6,592bn) includes taxable pensions and IRA distributions while NIPA compensation includes 1,563bn of employer supplements (wages to wages 5,837/6,408 = 91%); IRS 'total income' (8,210bn) includes 361bn of realized gains and about 311bn of taxable transfers that are not in national income; the 'realized capital gains vs undistributed profits' row compares conceptually different items. The text's 'downward trend' in capital-income reporting is shown only as a 2010 value and a 2000-10 average.

Why this severity. The ratios move by 1-8 points under consistent definitions, but the qualitative message—capital income is far less completely reported on tax returns than wages—holds under any of them.

Why it matters, in detail. Mixing Table A8 Panel B gains shares (ranked incl. gains) with Table A7 composition (ranked excl. gains) assigns capital gains of a different, richer group to each fractile; with Panel A the 1929 crossing point moves from P99 to P99.5. No consistent incl.-gains composition is published, so the authors' choice is an approximation that should have been disclosed.

Evidence. DataFig4AB formulas: composition = DataFig4CD share x (1 - KG share) [+ KG share for capital], with KG shares from columns X-AC of the external table, i.e. TabFig2010 Table A8 Panel B (fractiles defined by income including gains); verified from cached values (9.85% used for 1929 P99-99.5, vs 7.11% in Panel A). DataFig4CD comes from Table A7 (fractiles ranked excluding gains). With the consistent Panel A share, 1929 P99-99.5 has wages 39.0% vs capital 37.6%, so capital dominates only from P99.5; the 2007 threshold (P99.9) holds under either ranking.

Why this severity. Changes the 1929 threshold from the top 1% to the top 0.5% under a consistent ranking; the qualitative contrast with 2007 (top 0.1%) is unchanged.

Why it matters, in detail. Matched-concept CPS estimates track the tax series for P90-99 in level and trend; the comparative advantage of tax data is the top 1% (top-coding, realized gains, under-reporting) and the pre-survey era.

Evidence. APS (2011, JEL, p. 29): Burkhauser et al.'s CPS-based series 'match the Piketty and Saez (2003) series very closely for the second vingtile and the next 4 percent (i.e., the top decile excluding the top percentile)'; the CPS comparison is APS Figure 5, while APS Figure 11 is 'Top 1 Percent Share: Developing Countries'. Burkhauser, Feng, Jenkins & Larrimore (2012, REStat, pp. 377-381) locate the divergence in the top 1%. (The matching CPS series uses less top-coded internal data.)

Why this severity. An overbroad methodological statement and a wrong cross-reference; the case for tax data at the top 1% and for long historical series is unaffected, and no estimate changes.

The authors' response. The authors' own APS (2011, p. 29) already states the P90-99 agreement; no other response located.

Why it matters, in detail. (a)-(d) leave the coefficients unchanged (0.47 vs 0.495 full-window); (e) distinguishes the revenue-maximising rate (57% at e = 0.5) from the welfare optimum with a zero-sum externality (83%); (f) understates German top rates by about 2.3-2.9 points after 1998.

Evidence. (a) Figure 6 note: 'from 1960-64 to 2005-09' — six of 18 countries use other windows (Denmark 1980-2005, Ireland 1975-2009, Italy 1974-2009, Portugal 1976-2005, Spain 1981-2009, Switzerland 1960-95), so Portugal's top-rate change is -39.0 points in Figure 6 but +9.8 in Figure 7(a); the elasticity with the 12 full-window countries is 0.495 (0.134). (b) 'based on Figure 2 data' refers to PSS Figure 2; Figure 7's 'Table 2 reports estimates' refers to PSS Table 2 (this paper's Table 2 is US income growth); Figure 8's 'as in Figure 5, Panel B' should be Figure 7(b). (c) Italy's dot in Figure 6 is unlabelled. (d) PSS code annualises 1962-to-2008 growth over 45 rather than 46 years (span = year_final - year_initial - 5), overstating plotted growth rates by ~2% in relative terms (log-change regressions unaffected). (e) Table 4 note: 'In all cases, top tax rates are set to maximize tax revenue raised from top bracket earners' — in PSS's proof (p. 242) the 83% maximises top-bracket revenue net of the bargaining externality on everyone else; revenue from top earners alone peaks at 1/(1+1.5*0.5) = 57%. (f) Adjudicator coverage check of Figure 5: German top rates (e.g. 45% in 2010, 42% in 2005) exclude the solidarity surcharge (5.5% of tax since 1998) although the series note says 'general income tax supplements' are included (2010: ~47.5% with the surcharge).

Why this severity. Documentation and labelling errors; none changes an estimate, though (e) matters for interpretation because it invites reading 83% as a revenue-maximising rate.

Why it matters, in detail. Edits of 1-3 points in single country-years and a ~8-point vintage difference in one wealth curve; they matter for transparency and for anyone re-using the plotted data, not for the cross-country patterns.

Evidence. DataFig2ABC(1)!R128 '=13.77-3' (New Zealand 1999 top-1% share lowered by 3 points, removing a pre-2000 retiming spike); DataFig2ABC(2)!D133 '=9+1' (Germany 2004, incl. gains, raised by 1 point); Germany's 2001-07 top-1% in Figure 2 is scaled by the 1998 excl./incl. ratio (2007: 11.59%) whereas Figure 3 freezes it at 10.88%. Digitising the PDF vector paths (fig9_11_check.py): all curves of Figures 9 and 11 match Piketty-Zucman (2013) Table A5 within ~1 point except Figure 9's German curve (up to ~8 points too high, e.g. 1992), while Figure 11's German curve matches.

Why this severity. Small, undisclosed adjustments with no effect on any conclusion (the German wealth ratio's 1970-2010 rise is similar in both versions).

Why it matters, in detail. Revisions of 0.09-0.15 pp; the 2011 top-1% share was flat in the preliminary data, while the top decile edged up, so 'all the preliminary tabulations ... suggest' the rebound continuing is only partly accurate for 2011.

Evidence. TabFig2010 'Explanations' sheet: '2010 data come from preliminary statistics published by IRS on March 2012 and will be revised when final data are available'; IRS preliminary tables show AGI brackets only 'up to bracket $250,000+', below the top-1% threshold ($352,055). Final-data revisions (Jan-2013 vintage): top 1% 19.77% -> 19.86%, top 10% 47.90% -> 48.04%, top 10% excl. gains 46.26% -> 46.35%. The same Jan-2013 vintage, available before the June-2013 revision, put 2011 at 19.82% for the top 1% (flat) and 48.20% for the top 10% (+0.16); the Sept-2013 vintage later put 2012 at 22.46% (partly retiming before the 2013 tax increase).

Why this severity. An undisclosed caveat on the headline 2010 numbers, but the revisions were tiny and in the paper's favour, and the hedged statement about 2011-12 was broadly borne out by 2012.

Why it matters, in detail. (b) changes the ranking by 0.03 pp; (d) the excl.-gains rise to 2010 is +0.6 pp, entirely in the final year; others are typographic or bibliographic.

Evidence. (a) Text says 'as large as 82 percent' (p. 472) while Table 4 shows 83% (82.86%). (b) 'the two highest top decile income shares occurred in 1928 and 2007' (p. 472): in the March-2012 vintage 2007 49.74% and 2006 49.32% both exceed 1928 (49.29%). (c) 'Bakija, Cole, and Heim (2010)' in the text vs 2012 in footnote 5 and the references; 'Kopczuk and Saez (2010)' (p. 473) is Kopczuk, Saez & Song (2010), a study of earnings (not income) mobility. (d) 'excluding capital gains ... the upward trend has continued since 2007' (p. 458): the excl.-gains top decile is 45.67 (2007), 45.96, 45.47, 46.26 (2010), i.e. flat until the 2010 preliminary value. (e) Figure 1(a) omits the 1956 point and the P90-95 curve of Figure 1(c) omits 1997 (PDF vector paths have 93 instead of 94 points).

Why this severity. Presentation slips with no effect on any conclusion.

Why it matters, in detail. In the DINA series that implements the full-income correction the paper envisaged, recent top shares are 1-5 points below 1928-29, not 'significantly higher'; the comparison also embeds DINA's other adjustments, so it is a strong but not decisive test.

Evidence. WID DINA pre-tax national income (Oct-2026 bulk), tax units: top 1% 19.7% (2010) vs 23.2% (1929); top 10% 48.6% (2010) vs 49.6% (1929), and 48.5% (2007) vs 50.8% (1928); equal-split adults: top 1% 17.9% (2010) vs 22.2% (1929). DINA allocates all capital income, including tax-exempt income, but also adds items (e.g. fringe benefits) that lower top shares.

Why this severity. A hedged conjecture rather than a result; the later evidence reverses its sign but no core finding depends on it.