Emmanuel Saez · Contemporary Economic Policy 35(1): 7-25 (and 'Questions and Answers', 26-28)

Income and Wealth Inequality: Evidence and Policy Implications

13 problems found: 1 major · 6 moderate · 6 minor

What the paper claims

US pre-tax fiscal top income shares are U-shaped: about 45% for the top 10% before WWII, a precipitous fall that 'exactly lines up with World War II', about one-third in the post-war decades, then a rise since the late 1970s to over 50% (above the pre-war peak), with 12-13 of the ~17-point gain going to the top 1% and 8 points to the top 0.1%.
A century ago the very top earned mostly capital income; the post-1970s rise was, at least up to 2000, largely a labour/business-income phenomenon, with an apparent resurgence of capital income at the top since 2000.
Distributional national accounts: growth since 1980 has been very unequally shared; bottom-90% pre-tax income grew at half the average rate, bottom-50% pre-tax income stagnated, taxes and transfers closed about a third of the gap, and post-tax concentration is not as high as a century ago.
US wealth concentration has surged: the top 0.1% holds slightly more than 20% of household wealth, almost as much as before the Great Depression and about the same as the bottom 90%; the bottom-90% share doubled to over 35% by the mid-1980s and has since fallen, mostly due to debt; the bottom 90% saves about 0%.
English-speaking countries show a U-shaped top-1% share, whereas France, Sweden and Japan show no comparable rise since the 1970s; hence the rise is not due only to globalization or technology but to country institutions.
There is a strong link between top marginal tax rates and pre-tax top income shares: the US series are 'mirror images' over a century, and across 18 OECD countries the largest top-rate cuts since the early 1960s coincide with the largest top-share increases.
The tax-avoidance explanation is ruled out: top-1% charitable giving grew in tandem with the reported top-1% share, so changes in avoidance are insufficient to explain the tax-rate/top-share relationship for the US over a century and across countries.
There is no clear cross-country correlation between top-rate cuts and GDP growth (with or without convergence adjustment); in the US the bottom 99% grew fast and the top 1% slowly from 1933 to the late 1970s and the reverse after the 1980s rate cuts, consistent with (not proof of) rent-seeking; conclusion: high top tax rates reduce the pre-tax income gap without, so far, a clear negative effect on growth.

Every problem found

1majorlikely · previously known · overclaiming

The lecture argues that the rich cannot have been hiding much income in the 1960s, because their charitable donations, which lowered their taxes, were not unusually high then. But donations also depend on tax incentives and on changing habits of giving, and the data cover only the US after 1962. They cannot settle how much tax avoidance mattered, let alone over a whole century or in other countries. The finding that inequality rose is not affected; the claim that avoidance has been ruled out is.

Technical detail and evidence

Why it matters, in detail. Fig. 18 jointly tests 'no avoidance' and 'stable giving behaviour'. Persistent price elasticities in the published -0.5 to -1.4 range, combined with a 1.4-2x rise in top donors' tax price of giving since the 1960s, mean the near-constant top-1% giving rate (3.6% in 1962-79 vs 3.5% in 2008-14) already requires a 13-80% rise in 'generosity' with no avoidance at all. The 1960s ratio of true to reported income is therefore not identified. Giving rates also drift 25-35% across decades at roughly constant tax rates. The source attributes the 1984-92 gap to TRA86 income shifting, and TRA86 is the episode that carries the post-1960 US rate-share correlation (r = -0.51 for 1960-86, +0.32 for 1988-2015). The evidence thus cannot exclude a reporting component at least as large as the fiscal-vs-national-income gap in the authors' own DINA (10-20% of the post-1978 rise). It provides no evidence on pre-1962 US data or on the cross-country elasticity of 0.47.

Evidence. Lecture pp. 21-23. Data: saez_tpe16alone.xlsx, sheet data-charitable (Saez 2017 TPE appendix) [scripts/04_charity.py]. Levels: correlation 0.962; giving 39.2% of mean income in 1962 -> 81.4% in 2014; share 9.95% -> 21.43%. TRA86 gap: 1984->1988 share +29% vs giving +3%; 1984->1990 +20% vs +0%. Source text (Saez, NBER WP 22798, pp. 37-38; TPE 31, pp. 110-111): giving 'does not grow as much as reported income around the Tax Reform Act of 1986 ... a gap arising between the two curves from the early 1980s until about 1992', 'fully consistent with ... income shifting played a large role in the surge of top incomes during the Tax Reform Act episode'. Drift in the top-1% giving rate (giving / own reported income), at roughly constant rates after 1988: 3.72% (1962-68), 3.58% (1962-79), 2.78% (1987-92), 3.46% (2008-14); 1990->2014 share +50% vs giving +111%. Identification: suppose giving responds to its tax price with elasticity -0.5/-1.0/-1.4. Elasticities in this range are reported in Splinter (2018, pp. 3-4): Auten-Sieg-Clotfelter 2002 -1.3; Bakija-Heim 2011 ~-1; Peloza-Steel 2005 -1.4; top 1% -0.8 to -1.4. With an illustrative tax price rising from 0.40-0.45 to 0.62, a world with no avoidance still requires top-1% 'generosity' to rise x1.13-1.20/x1.33-1.50/x1.51-1.78; for the top 0.1% (price 0.30 -> 0.62) the factors are x1.23/x1.77/x2.36. Measurement: giving is itemized deductions 'excluding contributions above the AGI limitations', from public-use files to 2010. In the 1960s the limits were 20/30% of AGI and only half of capital gains entered AGI (Auten & Splinter 2020, Sec. III.D). Splinter (2018, fn. 14, citing Duquette) puts the effect of the 1917-1969 rise in the limits at +10-20% of giving. Against national income including retained earnings, top-0.1% giving in 1970 was ~3% of income vs 8% of fiscal income (Splinter 2018, p. 3).

Why this severity. Fig. 18 is the lecture's only evidence for 'discarding' the avoidance scenario, a pivotal step toward its policy conclusion. It supports the weak claim that much of the reported rise is real. It cannot bound how much of the tax-rate/top-share relationship is reporting, and it says nothing about 1913-1961 or the other 17 countries. C7 as stated is therefore not supported; the descriptive findings are unaffected.

The authors' response. Saez (2017 TPE; NBER WP 22798, pp. 36-38 and fn. 22-24) makes four points. Only a 'minuscule share' of top earners hit the deduction caps (1962-2015). Income accumulated in closely held firms could be realized and given at no tax cost. Rising generosity moving in tandem with reduced avoidance is 'a stretch'. TRA86 shifting inflated top incomes only in the short run, and giving 'catches up' in the 1990s. No reply to Auten & Splinter (2020) on this point was located.

2moderateconfirmed · previously known · methodology

The figure that the richest 0.1% of families own about 22% of all household wealth, nearly as much as just before the 1929 crash, assumes the very rich earn the same interest rate on their savings as everyone else. The authors' own research paper showed that the very rich seem to earn higher rates, which would bring their share below 19%. The lecture did not mention this. Wealth concentration still rose sharply, just somewhat less than stated.

Technical detail and evidence

Why it matters, in detail. The baseline assumes equal within-class yields. Top fixed-income capitalization factors near 100 make the post-2000 rise hostage to that one parameter. With the estate-based premium the 2012 top-0.1% share is 18.7% (-3.3 pp); the 1978-2012 increase is 22% smaller and the 2000-2012 increase 53% smaller. The ratio of top-0.1% to bottom-90% mean wealth falls from about 870 to about 730-740. The authors' 2020 revision puts the 2012 level at 20.1% (tax units). That revision adopted a higher yield for the wealthy since 2008, partly offset by more business wealth at the top. Including Social Security wealth (Catherine, Miller & Sarin 2025) would change the estimand rather than correct it.

Evidence. SZ (2016 QJE, pp. 549-551; Online Appendix Table B41c) [scripts/07_wealth.py]: 18.70% vs 22.01% in 2012 (-15%). The 1978-2012 rise is 11.63 vs 14.94 pp (-22%) and the 2000-2012 rise 2.81 vs 6.02 pp (-53%). Relative to 1929 (24.79%), 2012 stands at 0.75 vs 0.89. Appendix Table B40 (10-year Treasury yield for the top 1%): 19.73%. Bricker, Henriques, Krimmel & Sabelhaus (BPEA Spring 2016, p. 298): 'just under 19 percent' with Treasury-yield capitalization; their preferred estimate is 'about 15 percent'. The authors' later series (SaezZucman2020JEPData.xlsx): Sept-2020 revision 20.10% (tax units) / 18.82% (equal-split adults), with a 1978-2012 rise of 12.60 pp (-16%); WID Oct-2026 19.0% (adults). Smith-Zidar-Zwick (Apr-2020 draft of QJE 2023): 17.05% (tax units), rise 10.07 pp (-33%). Upward sensitivity with offshore wealth (Appendix Table B6): 22.89%.

Why this severity. Under the source's own alternative and the authors' later revision, the 2012 level falls 9-15% and the long-run rise 16-22% (33% in the most critical alternative). The surge in wealth concentration survives, but 'almost as much as 1929' becomes three-quarters to four-fifths. This is an important undisclosed caveat, not a reversal. Auditor B's 'major' rests on the 2000-2012 window, which is not the lecture's headline.

The authors' response. SZ (2016, p. 550) kept the baseline because only a few hundred non-married individuals die with estates above $20m each year, so the premium is noisy, and because decedents may not represent the living. SZ (2020, NBER WP 27921, pp. 2-3) later built a higher interest rate for the wealthy since 2008 into their revised series, while disputing Smith-Zidar-Zwick's larger premium. They also argue that offshore wealth and conservative business valuations bias top shares down.

3moderateconfirmed · previously known · internal inconsistency

To compare how the top 1% and everyone else fared, the lecture uses income reported on tax returns, which it admits leaves out much of ordinary workers' pay, such as employer health and pension benefits. With the authors' own fuller measure, the incomes of the bottom 99% grew about twice as fast after 1980 as the chart suggests. That is still much slower than before 1980 and than the top 1%.

Technical detail and evidence

Why it matters, in detail. Untaxed fringe benefits and pension and health contributions grew, mostly for the bottom 99%, so fiscal income fell from ~67% to ~60% of national income. As a result, fiscal per-adult growth of the bottom 99% (0.49%/yr, 1980-2010) understates the DINA pre-tax figure (1.02%/yr) by half. CPI rather than national-income deflation explains about a third of that gap. The pre-/post-1980 contrast remains large under DINA (3.4% vs 1.0%/yr for the bottom 99%).

Evidence. Fig. 21 logic (PSS usdata.do: CPI-deflated fiscal income incl. capital gains per adult 20+) [scripts/08_fig21.py]. Bottom 99%, 1980-2010: 0.49%/yr (+16%), against 1.02%/yr (+36%) pre-tax and 1.10%/yr post-tax in the PSZ Dec-2016 DINA. Bottom 99%, 1933-79: 3.20% vs 3.38%/yr. Top 1%, 1980-2010: 3.08% vs 3.50%/yr. About 0.17 pp/yr of the bottom-99% gap is the deflator (CPI vs national-income deflator). Top shares (PSZ Dec-2016 comparison columns) [scripts/02_dina.py]: top 10% per tax unit +13.6 pp (national income) vs +16.5 pp (fiscal) over 1978-2014, and +14.0 vs +15.3 over 1980-2014. Top 1% equal-split, 1978-2014: +9.4 vs +11.8 pp. Fiscal income is 66-68% of national income in 1950-90 and ~60% in 2012-14. Auditor A's -18% to -25% mixes unit and concept effects; same-unit comparisons give -9% to -20% depending on the start year.

Why this severity. Two things survive in the authors' national-income data: the Fig. 21 flip (bottom 99% faster before 1980, slower after) and the rise in top shares. But the post-1980 slowdown of the bottom 99% is overstated about twofold, and the top-share rise by 10-20%. This affects the magnitude of secondary evidence for C8 and C6, not its direction.

The authors' response. The lecture itself flags the limits of fiscal income (p. 11) and presents DINA (Figs 5-8) as the remedy. PSZ (2018 QJE) conclude that 'rising inequality is not an illusion of tax data'. Fig. 21 reproduces PSS (2014), which predates DINA.

4moderateconfirmed · newly identified here · methodology

The charts show the top US tax rate at 70% until 1981, but from 1972 the top rate on salaries, which is how executives are paid, was capped by law at 50%. Top salaries were already rising in the 1970s under that unchanged 50% rate. This weakens the neat story that executive pay took off only once top rates were cut in the 1980s. The long-run pattern across countries is not affected.

Technical detail and evidence

Why it matters, in detail. For labour income, the US net-of-tax rate rose x1.44 between 1980 and 1988, not x2.40. Top-1% wage shares rose 38% during 1972-86 while that rate stayed at 50%. The timing behind the 'mirror image' and the rent-seeking reading of executive pay is therefore looser than Figs 16 and 21 suggest. Replacing the TPC rate with the earned-income rate lowers the 1913-2015 elasticity from 0.24 to 0.20 and removes the 1960-86 correlation (-0.51 -> -0.05). This is an upper bound, because the 70% rate still applied to capital income.

Evidence. 26 CFR sec. 1.1348-1: 'for taxable years beginning after December 31, 1971, the maximum tax rate applicable to the earned taxable income ... is not to exceed 50 percent'; 60% for 1971; not available with income averaging or to married individuals filing separately. 26 U.S.C. sec. 1348 was repealed by Pub. L. 97-34 (1981). TPC: exactly 70% in 1971-80 and 69.125% in 1981. PS Table B2: the top-1% wage share rose from 5.32% (1972) to 7.33% (1986), +38%, while the top rate on wages stayed at 50%. Sensitivity using the earned-income rate for all top income (an upper bound) [scripts/06_us_timeseries.py]: time-series elasticity 0.20 vs 0.24 (1913-2015) and 0.32 vs 0.41 (1960-2015); 1960-86 correlation -0.05 vs -0.51. The net-of-tax rate rose x1.44 for wages vs x2.40 for ordinary income between 1980 and 1988. UK: Finance Act 1974 s.7 sets a top higher rate of 83% and a 15% additional rate on investment income above GBP 2,000; Fig. 15 plots 98% for 1974-78.

Why this severity. It modestly affects the US time-series pillar of C6 and the Fig. 21 narrative behind C8: the long-run elasticity falls by at most 17-22%. The cross-country Fig. 17, whose 1960-64 base predates the provision, is unaffected. Capital income, and many top earners (through the preference offset and income averaging), still faced 70%, so the true effect lies between zero and these bounds.

5moderatelikely · previously known · overclaiming

The lecture shows that places and periods with lower top tax rates had a richer top 1%, and that economic growth did not obviously differ. That pattern is real, but it does not prove that high top tax rates cut top incomes at no cost. The growth comparison is too imprecise to detect even large effects, and since 1960 the US link rests mainly on the 1986 tax reform. The lecture is cautious along the way, but its conclusion is worded more strongly than the evidence allows.

Technical detail and evidence

Why it matters, in detail. Over 1913-2015 the log top-share/net-of-tax-rate elasticity (about 0.24) survives a linear trend. After 1960 it is identified by the TRA86 episode and vanishes once a trend is added (-0.08, s.e. 0.07). The 18-country first-difference elasticity (0.47) is robust to dropping any country. For growth, the 95% CI for a 40-point cut runs from -0.8 to +0.7 pp/yr. 'No clear negative effect' is therefore consistent with both zero and economically large effects (a third of mean growth).

Evidence. US series, TPC top rate vs PS top-1% share incl. capital gains [scripts/06_us_timeseries.py]: r = -0.69 (1913-2015), -0.51 (1960-86), +0.32 (1988-2015). Log elasticity with respect to the net-of-tax rate: 0.24 (HAC s.e. 0.04) for 1913-2015, and 0.26 with a linear trend. For 1960-2015 it is 0.41 (0.12) without a trend and -0.08 (0.07) with one. PSS (NBER WP 17616, fn. 26): 'the correlation disappears when additional polynomials in time are added'. Auten & Splinter (2020, Sec. III.D), using their national-income shares: r = -0.58 (1960-2015), +0.61 (1960-86), +0.23 (1988-2015). Cross-country, PSS data [scripts/05_pss.py]: elasticity 0.470 (s.e. 0.114), leave-one-out range 0.42-0.54, 0.49 (0.18) without the US and UK — robust as a correlation. Fig. 19: slope 0.0008 pp/yr per tax-rate point (s.e. 0.0097). The implied effect of a 40-point cut is -0.03 pp/yr, 95% CI [-0.79, +0.73]. The minimum detectable effect at 80% power is 1.08 pp/yr, against mean growth of 2.44%/yr. Convergence-adjusted log slope (Fig. 20): 0.105 (0.077). Mertens & Montiel Olea (2018 QJE) find output effects of top-rate cuts using narrative identification.

Why this severity. The correlations behind C6 reproduce and are robust across countries. But they cannot separate supply, avoidance and bargaining responses, and the growth test cannot rule out effects worth a third of average growth. The lecture hedges the intermediate steps, so the overreach is confined to the causal wording of the conclusion.

The authors' response. The lecture concedes that the growth test is 'not a very compelling test' and that the evidence is 'consistent' with rent-seeking without proving it (pp. 23-24). PSS (2014 AEJ:EP) add panel regressions with country fixed effects and CEO-pay evidence.

6moderatelikely · previously known · external inconsistency

A later study using the same tax records argues that inequality before the Second World War was lower than the lecture says. It also finds that inequality fell more gradually, starting in the 1930s rather than exactly during the war. This changes the history but not the finding that top shares have risen sharply since 1980.

Technical detail and evidence

Why it matters, in detail. On GMMS's corrected series the 1928-44 top-decile decline is 11.2 rather than 16.8 pp, and over half of it occurs by 1939. The 'precipitous' WWII collapse is therefore largely a feature of PS's pre-1944 net-income adjustments and denominator. The top-1% decline shrinks from 12.7 to 8.6 pp. Post-1960 and current levels are not affected.

Evidence. GMMS replication workbook (DataAppendix-6-7-21.xlsx, sheet TopIncomeShares; column 'PS latest updated file' vs 'Corrected - Cumulative + full K-gains'). Top 10%: 1928 49.29 vs 42.47; 1939 45.52 vs 36.45; 1944 32.51 vs 31.25. The 1928-44 fall is 16.78 vs 11.22 pp (-33%), and the share of it completed by 1939 is 22% vs 54%. Top 1% fall: 12.66 vs 8.63 pp (-32%). In PS (TabFig2015prel), 12.78 of the 16.78-pp fall (76%) occurs in 1940-44 [scripts/01_topshares.py]. Also reproduced by Auditor B.

Why this severity. It changes the depth (by about a third) and the timing of the mid-century compression. That is a secondary part of C1, though it also colours the historical tax narrative. The U-shape, the post-1980 rise and the claim that recent shares exceed pre-war peaks are unaffected; the last is strengthened.

The authors' response. No direct reply by Piketty or Saez was located (targeted searches by both auditors and the adjudicator); the critique postdates the lecture.

7moderateplausible · previously known · external inconsistency

Other researchers redid the calculations with different assumptions about income that never appears on tax forms. They find that the top 1%'s share rose about half as much as the lecture says, and that the poorer half of Americans saw real income gains rather than stagnation. The original authors dispute these re-estimates and the debate is unresolved, so the size of the trends is uncertain, though not their direction.

Technical detail and evidence

Why it matters, in detail. Three contested allocations drive the gap. Underreported business income is allocated in proportion to positive reported business income by PSZ and from audit data by AS (~2.0 pp of the 6.0-pp 2014 level gap). Retirement income accounts for ~1.0 pp, partly accepted by PSZ; corporate taxes for ~0.7 pp. Together they halve the 1979-2019 rise (+4.4 pp in AS vs +8.2 pp in PSZ). For the bottom 50%, AS use per-capita, household-size-adjusted units and count more transfers, which raises growth from about zero to over +33%. The lecture's numbers are internally correct for equal-split adults and PSZ allocations.

Evidence. Auten & Splinter (2024, JPE 132(7), p. 2182 and Table 5): top-1% pre-tax national-income share 9.4% (1979) -> 13.8% (2019), +4.4 pp, 'about half the increase in fiscal income'. PSZ current vintage (WID Oct-2026, sptinc992j): 10.86% -> 19.04%, +8.2 pp [scripts/03_wid_current.py]. Bottom 50%: AS find real pre-tax income up 'more than one-third' over 1979-2019. PSZ show it near-flat: Dec-2016 vintage +0.8% for 1980-2013 [scripts/02_dina.py]; WID-2026 +8.7%; post-tax +19% for 1980-2013. Bricker, Henriques, Krimmel & Sabelhaus (BPEA Spring 2016, p. 301): 2012 top-1% income share 'just under 18 percent' (preferred) vs 'nearly 23 percent' (PS). Piketty, Saez & Zucman (2024 comment): after correcting AS's allocation of excess partnership depreciation and other untaxed income, AS's estimates become 'similar in level and trend' to PSZ. Gale, Sabelhaus & Thorpe (2023) mostly side with PSZ on underreporting but not on government consumption. Iselin & Reck (2025) show that both studies rest on assumed relative misreporting rates.

Why this severity. If the Auten-Splinter allocations were right, the post-1979 rise in the top-1% share would be about half as large, and bottom-half 'stagnation' would become moderate growth. That matters for the magnitudes in C3, not their direction. The alternatives are actively disputed, part of the bottom-50% gap reflects unit and transfer conventions, and the authors' own updated series confirm the lecture's figures. This is a moderate uncertainty, not a demonstrated error.

The authors' response. Saez & Zucman (2020, NBER WP 27921) and Piketty, Saez & Zucman (2024 comment) argue that the Auten-Splinter results stem from allocating untaxed business income, especially partnership income, to less-rich owners, and that after correction AS's estimates become similar to PSZ in level and trend. They accepted a revised allocation of untaxed pension income; their revised 1978-2014 rise in the top-1% pre-tax share is 8.5 pp, vs 9.4 pp originally.

8minorconfirmed · newly identified here · version instability

One remark depends on which version of the authors' data is used: that during the Second World War taxes cut the top 10%'s share by more than the before-tax figures show. It is true in the lecture's version and in today's version, but not in the version published in between.

Technical detail and evidence

Why it matters, in detail. The 1940s post-tax top-10% series differs by up to about 5-6 pp across PSZ vintages (28% vs 33.7% in 1945), reflecting the allocation of wartime deficits and government spending. The qualitative claim holds in the lecture's vintage and the current one but fails in the Dec-2016/QJE vintage.

Evidence. Lecture Fig. 5: post-tax top-10% share about 28% in 1945 vs about 36% pre-tax (drops of about 16 vs 12 pp). PSZ Dec-2016 and Nov-2017 data, 1940 -> 1945: pre-tax 47.7 -> 35.8 (-11.9 pp), post-tax 45.6 -> 33.7 (-11.9 pp); in 1944, post-tax 35.9 vs pre-tax 36.2. WID Oct-2026: pre-tax -12.8, post-tax disposable -16.9 (scainc -15.7) [scripts/02_dina.py; 03_wid_current.py].

Why this severity. This is a secondary descriptive remark. Whether it holds depends on how each vintage allocates wartime deficits and public spending, and no core result rests on it.

The authors' response. The lecture describes the DINA results as 'preliminary findings' (p. 11).

9minorconfirmed · newly identified here · arithmetic or transcription

A few numbers are slightly off. For example, the rich gave about 39%, not 35%, of average income to charity in the 1960s. None of these slips changes any conclusion.

Technical detail and evidence

Why it matters, in detail. The deviations are 0.1-4 percentage points on descriptive magnitudes. None enters an estimate or alters a comparison; the corrected 1960s giving level makes the 1962-2014 giving and income-share factors closer (2.10 vs 2.05).

Evidence. (a) Top-1% giving in the 1960s is given as 'about 35%' of mean income (p. 21). The data show 38.8% (1962-68 mean), and the source's own note gives 39% for 1962. Correcting it makes the parallel tighter (x2.10 for giving vs x2.05 for the share). (b) 'Above 70% from 1936 to ... 1980' (p. 19): the TPC rate was exactly 70% in 1965-67 and 1971-80, and 69.125% in 1981. (c) 'The United States and Japan tax about 30% of national income' (p. 8): Japan collected 30.0% of GDP in 2014 (OECD Revenue Statistics), about 37% of national income (Auditor A; WID NI/GDP 0.806); the US figure is 25.8% of GDP, about 30% of national income. (d) Bottom-90% pre-tax growth is given as '0.7%' a year (p. 12); the data show 0.76%/yr. The '30%' cumulative figure is 28.6%.

Why this severity. These are rounding or wording slips of a few points that leave every conclusion unchanged; (a) even strengthens the author's argument.

10minorconfirmed · newly identified here · documentation gap

The charts comparing countries say they cover 1960-64 to 2005-09, but six countries use different years, and the growth chart uses yet another set of years. Fixing this does not change the results.

Technical detail and evidence

Why it matters, in detail. Six of the 18 countries enter Fig. 17 with windows other than 1960-64 -> 2005-09, and the growth figures use different windows (Portugal's x-value moves 49 points). Neither the elasticity (0.495 on the 12 correctly windowed countries) nor the null growth result depends on this.

Evidence. PSS international.do (read as text) and AllCountries_update_pre75.dta [scripts/05_pss.py]. Income windows that differ from 1960-64 -> 2005-09: Denmark 1980-84 -> 2001-05; Ireland 1975-79; Italy 1974-78; Portugal 1976-80 -> 2001-05; Spain 1981-85; Switzerland ends 1991-95; several others end 2004-08. The GDP and tax-rate windows of Figs 19-20 are 1960-64 -> 2006-10, so Portugal's tax change is -39 points in Fig. 17 but +9.8 in Fig. 19. Growth is annualized over 45 rather than 46 years (Auditor B). PSS (NBER WP 17616, text at fn. 36) disclose the selection rule; the lecture does not. Restricting Fig. 17 to the 12 countries with true ~1960-64 -> 2005-09 windows gives 0.495 (s.e. 0.134). Matching GDP to income windows keeps the growth slope insignificant (Auditor B: 0.168, s.e. 0.151).

Why this severity. This is a documentation problem with no effect on the results: the elasticity and the absence of a growth correlation survive correct windows.

The authors' response. The selection rule is disclosed in PSS (2014) and its replication code.

11minorconfirmed · newly identified here · other

Some figure numbers, axis labels and one reference are wrong or confusing, but none of this affects the results.

Technical detail and evidence

Why it matters, in detail. Labelling and citation errors; the projection revisions to 2015 shares are 0.3-0.5 pp and leave all comparisons unchanged.

Evidence. p. 19 refers to 'Figure 19' for the top-share/tax-rate scatter; it is Figure 17. Fig. 12's left-axis labels are truncated ('0,000,000', '2,000,000', '4,000,000' for $10m, $12m and $14m), so the axis looks non-monotone. Fig. 3's note says 'including or excluding realized capital gains', but only one series (including gains) is plotted. David Card's introduction (p. 7) cites Alvaredo et al. (2013), a JEP article on top incomes, for IRS-based work on the geographic variation of intergenerational mobility. The DINA groups are called 'families' in the text but 'adults' in the figures (pp. 12-13). The 2015 fiscal shares are June-2016 projections, later revised: top 10% 50.47 -> 49.99, top 1% 22.03 -> 21.59, top 0.1% 10.90 -> 10.61 [scripts/01_topshares.py].

Why this severity. Presentation only; no number or inference depends on these errors.

12minorplausible · newly identified here · overclaiming

The lecture says the middle class lost its share of wealth mainly because it took on more debt. Depending on how the accounting is done, rising debt explains between a quarter and about half of the loss by 2012. Falling house prices and faster wealth growth at the top explain much of the rest.

Technical detail and evidence

Why it matters, in detail. Debt explains 24-36% of the 1986/82-2012 decline under an additive share decomposition and 46-55% under a leverage counterfactual, and a majority at 2008-09 end points. 'Mostly' is therefore convention- and endpoint-dependent. Neither decomposition identifies causation, since debt also finances housing and consumption.

Evidence. SZ (2016) Appendix Tables B1, B5 and B13 [scripts/07_wealth.py]. Additive share decomposition (Auditor B): debt as a share of total net wealth rose 3.30 pp, 24% of the 13.63-pp loss (36% from 1982; 54% to 2008; 69% to 2009). Leverage counterfactual: hold the bottom 90%'s debt-to-gross-assets ratio at its 1986 level (29.2%, vs 44.5% in 2012); debt then accounts for 46% of the loss (55% from 1982; 67-78% to 2008-09). The 2012 end point is the housing-price trough, which the lecture also cites (p. 16). SZ (2016, Sec. V.B, p. 555) make the same attribution, citing household debt rising from 75% to 110-135% of national income.

Why this severity. This is a loose statement about mechanism in an oral lecture, and it follows the source. The descriptive facts are correct: the bottom-90% share fell and its real wealth has been flat since 1986. Debt accounts for roughly a quarter to a half of the decline, depending on the convention.

The authors' response. SZ (2016, p. 555): the upsurge in debt has had 'a dramatic effect on middle-class wealth, more than offsetting the rise of pensions'.

13minorplausible · previously known · external inconsistency

Answering a question, Saez says US taxes are less progressive than in the 1960s. That is true on the authors' own measures, which emphasize falling corporate and estate taxes on the rich. Other measures covering the whole population show federal taxes becoming more progressive since 1979.

Technical detail and evidence

Why it matters, in detail. On PSZ conventions, top-end progressivity fell because corporate and estate tax burdens on the top 1% fell. Whole-distribution indices of federal taxes (CBO, from 1979) rose, largely through refundable credits. The claim is therefore measure-dependent.

Evidence. PSZ Dec-2016 DINA: the top-1% average tax rate (all taxes) fell from 43.2% (1960) to 36.4% (2014). Corporate taxes (17.0 -> 10.4 pp) and estate taxes drove the fall, while the top-1% individual income-tax rate rose (16.4 -> 18.9 pp). The bottom-50% rate rose from 18.2% to 24.4%. Splinter (2020, National Tax Journal 73(4)): the CBO-based federal Kakwani index rose 46% over 1979-2016, and 1960s top rates are overstated by sheltering.

Why this severity. This passing Q&A remark is consistent with the authors' published estimates. It depends on incidence conventions and on the progressivity measure, and no core finding rests on it.

The authors' response. Piketty & Saez (2007 JEP) and Saez & Zucman (2019) argue that progressivity at the top has declined; they treat refundable credits as transfers rather than negative taxes.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

I12: the 'U vs L' international dichotomy is level-based. Sweden's top-1% share roughly doubled from its 1981 trough (x2.18 in WID-2026), like the UK, so Sweden is a weak member of the 'L' group; the UK and Canada series also switch units without disclosure. — The lecture claims only that France, Sweden and Japan 'do not experience nearly as large an increase in income concentration since the 1970s compared to the United States' (p. 18). That holds in percentage points: about +3 pp for Sweden vs +10-11 for the US in the lecture's own Fig. 14; +4.8 vs +12.8 in WID-2026; 4.1 -> 6.8% vs 8.1 -> 17.6% in the PSS data (Auditor B). It also holds in ratios in the lecture's own figure (about x1.75 vs x2.3). Top-share changes are conventionally compared in points, and 'L-shaped' is a stylization. The UK and Canada unit switch is a documentation detail of the source database and does not affect the contrast.
I1 severity: omitting the interest-yield sensitivity is a major problem because it halves the 2000-2012 rise in the top-0.1% wealth share (-53%). — Partly rejected (severity). The sensitivity is confirmed and retained as F2, but the lecture's headline is the long-run surge and the current level. These fall 16-22% and 9-15% under the source's alternative and the authors' later revision; only the most critical alternative (Smith-Zidar-Zwick, -33%) exceeds a quarter. Graded moderate.
I4: the statement that the bottom 90% lost its gains 'mostly due to an increase in debt' is a confirmed overclaim, because debt accounts for only 24% of the 1986-2012 decline. — Partly rejected. The 24% comes from one additive convention (debt divided by total net wealth). A leverage counterfactual that holds the bottom 90%'s debt/asset ratio at its 1986 level attributes 46% of the decline to debt (55% from 1982), and end points in 2008-09 give a majority. The statement is convention-dependent rather than shown to be wrong; retained as a minor, plausible issue (F12).
I3 magnitude: the fiscal series overstate the rise in top shares by about 18-25% relative to the authors' DINA. — Partly rejected (magnitude). The -25% figure compares tax-unit fiscal shares with equal-split national-income shares, mixing unit and concept effects. Same-unit comparisons give -18% to -20% for 1978-2014 and -9% to -14% for 1980-2014. The issue is retained (F3) with the corrected range; its main content is the Fig. 21 growth comparison.
I1 magnitude: the charitable-giving test cannot exclude reporting effects 'of the order of a quarter to a third of the fiscal top-share rise'. — The specific magnitude is not demonstrated: the price-elasticity calculation is illustrative, with assumed tax prices. The adjudication retains the qualitative conclusion that the test cannot bound avoidance (F1). The lower reference point used is the 10-20% fiscal-vs-national-income gap in the authors' own data.

Audit notes

What reproduced. Reproduced with the adjudicator's own scripts (final/scripts, logs in final/logs). Fiscal top shares, thresholds and composition (C1-C2) reproduce exactly from TabFig2015prel. DINA levels and growth rates (C3) reproduce within rounding from the public Dec-2016 PSZ release; the lecture's own spring-2016 pre-release vintage is not public. That vintage differs mainly in the 1940s post-tax series: the 'even bigger' post-tax WWII drop does not reproduce in the Dec-2016 or QJE data but does in WID-2026. Wealth shares, means, thresholds, saving and the source's own alternatives (C4) reproduce exactly from the SZ (2016) appendix. The PSS cross-country elasticity and growth regressions (C6, C8) replicate exactly from the AEA package (Stata logic re-coded in Python, do-files read as text only). Fig. 21 reproduces to 2010 from usdata.dta; the 2011-13 extension was not rebuilt here (Auditor A: 2013 indices 419/367). Fig. 18 data reproduce from the TPE appendix. The Geloso et al. comparison reproduces from their released workbook, but their historical reconstruction was not independently rebuilt. The Auten-Splinter, Smith-Zidar-Zwick and Bricker et al. estimates are quoted from their texts or the authors' compilation file and were not re-estimated. The international WID comparison (C5), the IRS AGI-share cross-check and the SCF tabulations were run by the auditors, not re-run here.

Limits of this audit. Several sources could not be fully examined: - The 2020 OUP adapted chapter (Saez, 'Income and Wealth Inequality: Evidence and Policy Implications', pp. 38-60) could not be obtained, so changes in that version were not checked. Auten & Splinter (2020) quote its giving passage ('35 percent to 80 percent'; 'their ability to give has grown almost parallel...'), so at least that argument carried over. - The lecture's own pre-release DINA vintage is not public. - The critics' re-estimates (Auten-Splinter, Smith-Zidar-Zwick, Geloso et al., Bricker et al.) were not re-estimated. - Confidential IRS giving data are inaccessible, so the claim that few top donors hit the deduction caps could not be tested. - The tax prices of giving used in F1 are illustrative, not computed from microdata. - The legal check of the maximum tax relies on the CFR/USC text; its effective incidence (preference offset, eligibility) was not quantified. - Values read from figures (Figs 5, 14, 21) carry about +/-0.5 pp reading error. - The international comparison, the IRS AGI-share cross-check and the SCF checks rely on the auditors' runs.