Global Inequality Dynamics: New Findings from WID.world

Why it matters, in detail. Mechanism: about half of national income is not on individual returns; its allocation drives the top-share trend. PSZ allocate underreported pass-through income proportionally to positive reported income (about 50% to the top 1% vs about 15% after re-ranking in AS), allocate partnership excess depreciation and S-corporation exempt income to their owners, and treat consumption taxes as not affecting pretax inequality; AS do the opposite and use individuals (including children) ranked by size-adjusted income. Direction: PSZ give a higher and faster-rising top share and lower bottom-half growth. Magnitude: top-1% rise +4.4 pp (AS, 1979-2019) vs about +8.5 pp (PSZ/WID, 1978-2015); 2014 level gap 6.0 pp; bottom-50% real pretax growth about +40% (AS) vs about 0 (PSZ). Comparisons are not fully like-for-like (units, deflator, income concept). Independent assessments split: Gale, Sabelhaus & Thorpe (2023) side with PSZ on underreporting; Iselin & Reck (2025) judge AS's implied decline in relative misreporting implausible and PSZ's assumption at the high end of plausible values.

Evidence. AS (2024) p. 2182: 'Our pretax top 1% share of national income increased 4.4 percentage points (pp) between 1979 and 2019, about half the increase in fiscal income'; Table 5 baseline 9.4% (1979) -> 13.8% (2019) (p. 2216); top-1% pretax share 6 pp below PSZ in 2014 (2.0 pp from underreported business income, 1.0 pp retirement income, 0.7 pp corporate taxes); 'PSZ estimated that average real pretax incomes of the bottom 50% remained virtually unchanged between 1979 and 2019. In contrast, our analysis shows that real pretax incomes increased by more than one-third' (p. 2182; +40% since 1979, PCE-deflated per capita, pp. 2204-2205); bottom-50% pretax share -5.1 pp. Authors' current series (WID.world, 15 Oct 2026 release): US top-1% pretax share 10.35% (1978) -> 18.89% (2015), +8.5 pp; bottom-50% real growth +7% (outputs/wid2026_table1.txt). PCE vs national-income deflator changes the US bottom-50% growth by only 1.7 pp (outputs/us_deflator_sensitivity.txt), so most of the gap is allocation and units, not prices.

Why this severity. If AS are right, the rise of the US top-1% share is roughly halved and 'no growth left at all for the bottom 50 percent' becomes growth of about 40%, a change far above a quarter of the headline trend; the directions (rising top share, falling bottom share) survive in every series. The dispute is unresolved, so confidence is only 'plausible'.

The authors' response. Saez & Zucman (2020, NBER WP 27921, Section 3) accepted the retirement-income point (F4) but rejected AS's underreporting allocation; Piketty, Saez & Zucman (2026, 'Income Inequality in the United States: A Comment', 19 Aug 2026) argue that eight AS methodological choices (partnership excess depreciation, S-corporation exempt income, consumption taxes, retirement-account income, retained earnings, underreporting, government consumption, deficits) explain about three-quarters of the gap in the level and rise of the top-1% share; AS (2025, 2026 replies) maintain their allocations.

Why it matters, in detail. Group growth = (share ratio) x (real average growth); the price index enters only the second factor. Average growth factor: 9.11 (paper), 11.81 (January 2017 draft), 13.85 (official deflator), 6.38 (WID 2026 / Maddison-Wu). Bottom 50%: +401% (paper), +550% (Jan), +662% (official), +251% (paper's shares with the WID 2026 deflator) or +264% (WID 2026 shares). The ratio of top-1% to bottom-50% growth factors (3.99) and the cross-country ordering are invariant. The undisclosed January-to-February change cut the growth factor by 23%, so anyone citing the circulated January draft quotes China numbers 31-37% higher.

Evidence. WID.world WP 'This version: January 8, 2017' (https://wid.world/www-site/uploads/2017/01/ACPSZ2017AER.pdf; Last-Modified 8 Jan 2017; md5 a3876947...) p. 6 and Table 1: 1081/550/1040/1709/2491/2962/3513/4065. AER Table 1: 811/401/779/1294/1898/2261/2685/3111. Both columns reproduce 8/8 from ACPSZ2017.xlsx with the same shares (Data4!A8:B14) times either Data1!B ('2013 yuan (GDP deflator)', PYZ Table C1, 2014-15 projected as 1.065 x previous year; 2015/1978 ratio 11.81) or Data4!C7:D7 (6,349 -> 57,807 'yuans 2015'; ratio 9.105) (outputs/table1_repro.txt). World Bank WDI GDP deflator 14.39 -> 89.78 (x6.24) with WID nominal NI per adult (x86.4) gives x13.85 = +1,285%; WID 2026 (metadata inyixxi999: '1980-2014: GDP deflator from Maddison & Wu (2017)') gives x6.38 = +538% and bottom 50% +264% (outputs/china_growth_deflators.txt, wid2026_table1.txt). PYZ (2019, AER 109(7), p. 2474) state that their corrected real growth 'lies in between the official GDP growth rate and the conservative estimate of Maddison and Wu (2008); it is closer to Maddison and Wu's', and their Table 2 prints +811%.

Why this severity. Every China cell of Table 1 moves by a quarter or more across defensible deflators, but the deflator scales all China rows by one common factor, so no distributional or cross-country conclusion changes (China's bottom half still 'grew markedly'; the top still grew fastest). The choice is documented in the source paper and lies between the official and Maddison-Wu bounds; what is missing is disclosure in this paper and a note on the January revision.

The authors' response. None located specific to ACPSZ. PYZ (2019, p. 2474 and online Appendix A21) justify the Young-type deflator as lying between the official and Maddison-Wu estimates.

Why it matters, in detail. Mechanism: non-taxable distributions are mostly rollovers (asset stocks moved between accounts, not income flows; 79% of pension distributions of filers with AGI of $1m or more were non-taxable in 2014, AS fn. 33); using them as an allocation key ties retirement income and wealth to executives' rollovers. Direction: overstates top shares, mainly in recent years (the old weight 'could be higher than 10%, especially in recent years'), hence the trend. Magnitude: -0.6 pp on the top-1% pretax share (2016), about 6% of the 9.4 pp 1978-2015 rise; -1 to -1.5 pp on the top-1% wealth share, about 6-10% of its rise. The residual dispute over the 10% weight is about 0.3 pp or less.

Evidence. Saez & Zucman (2020, NBER WP 27921, Section 3.3, p. 31): 'Piketty, Saez and Zucman (2018) put too high a weight on non-taxable distributions, especially in recent years, because non-taxable pension income is inflated by rollovers. This led to allocating too much retirement income to the top ... This revision closes about 10% of the gap between AS and Piketty, Saez and Zucman (2018).' PSZ 'US Distributional National Accounts: Updates' (Feb 2022 file), Sections 2.1.7 and 2.2.1 (September 2020 update): allocation now 60% taxable distributions / 30% wages / 10% non-taxable; 'reduces the top 1% wealth share by 1-1.5 points in recent years'; 'reduces the top 1% pre-tax income share by about 0.6 point in 2016'. AS (2024, pp. 2209-2211) put retirement income at 1.0 pp of the 6.0 pp 2014 gap.

Why this severity. A confirmed error in a core input, but its effect (about 0.6 pp on a 20% income share and 1-1.5 pp on a 37% wealth share, roughly 5-10% of the documented rises) is modest and changes no conclusion.

The authors' response. Accepted: 'We are grateful to AS for bringing this issue to our attention' (Saez & Zucman 2020, p. 31); corrected in the September 2020 series. SZ add that AS now allocate too little pension wealth to the top.

Why it matters, in detail. Sources of change: NIPA/INSEE benchmark revisions (France's per-adult growth +39% -> +47%), the 2020 PSZ methodological update (pensions, business income, interest), PYZ's re-estimation of China's top income tail between the 2016 working paper and the 2019 publication, and WID's 2025-26 macro and price-index rebuild. Effects: US bottom-50% growth -1% -> +3..+7% (still stagnation, 0.1-0.2%/year); US top-1% 2015 share -1.3 pp; France bottom-50% growth relative to average growth (ratio of growth rates) 1.02 -> 0.85-0.93; China B50-minus-T1 gap +2.4 -> +0.1..+0.8 pp; China very-top shares differ by a factor of up to 8 across versions, so the top-0.1/0.01/0.001% rows of Table 1 carry little information beyond 'very high growth'.

Evidence. US: top-1% pretax share in 2015 18.9% (PSZ Feb-2022 and Sep-2020 TB3; WID 2026) vs 20.2% printed; bottom-50% growth 1978-2015 +3.3% (Feb-2022), +3.7% (Sep-2020), +7% (WID 2026) vs -1% (outputs/psz_vintages_tb3.txt, wid2026_table1.txt). France (WID 2026): average +47% (1978-2015; +46.5% for 1978-2014) vs +39%; bottom 50% +40% (1978-2015) or +43% (1978-2014), i.e. somewhat below the average; top 0.001% +47% vs +158%. China: 2015 top-1% share 13.9% (PYZ posted gperc file), 14.5% (PYZ 2019 AER Table 1), 14.0% (WID 2026) vs 12.7%; 2015 bottom-50% minus top-1% gap +2.4 pp (paper) vs +0.8 pp (PYZ 2019) vs +0.08 pp (WID 2026), with the top 1% above the bottom 50% in 2008-2011 and 2013 in WID 2026; top-0.001% share in 2015 0.33% (ACPSZ workbook) vs 1.36% (PYZ posted gperc file) vs 2.8% (PYZ 2019 Table 1); PYZ 2019 Table 2's very-top growth is non-monotone (+2,645% / +2,766% / +2,529% for top 0.1 / 0.01 / 0.001%) vs monotone in ACPSZ (+2,261 / +2,685 / +3,111) (outputs/pyz_gperc_check.txt). Section III: Japan's net public wealth share -0.9% (workbook, 2015) vs +6.6% (WID 2026); China's 1978 public share 82% and private wealth 73% of NI in WID 2026 vs about 70% and 115% (outputs/wid2026_wealth_ratios.txt).

Why this severity. No main US/China/France contrast reverses, but several printed magnitudes move by 10-60% and three textual claims are weakened or (Japan) reversed in the authors' own later data: modest effects on core results, consistent with the paper's own warning that its series are provisional.

The authors' response. The paper states its series are 'imperfect, provisional, and subject to revision' (p. 407); revisions are documented in PSZ 'Updates' notes and WID metadata.

Why it matters, in detail. Outside 2006-2010 the corrected top-1% series equals the survey series times a fixed upgrade plus the growing national-accounts imputation (corrected/raw is about 1.9 in both 1978, 5.78% vs 3.08%, and 2015, 12.68% vs 6.57%); no year-specific tax information enters. Direction of bias: unknown. If survey under-coverage of top incomes grew over time, the rise is understated; if the 2006-10 market-economy correction overstates 1978 inequality, the rise is also understated; the individual-vs-equal-split mismatch biases levels up. Magnitude not quantifiable without PYZ's tabulations.

Evidence. PYZ (2019, AER p. 2477, fn. 19): 'we choose to apply the average profile of upgrade factors f(p) ... estimated for the 2006-2010 period to all years of the period 1978-2005 ... We also choose to apply this same average profile ... for all years of the period 2011-2015'; fn. 20: 'our estimates tend to overestimate inequality as compared to the equal-split benchmark'; p. 2471: tax data released from 2006 for taxable income above 120,000 yuan. ACPSZ p. 405 and fn. 2. Workbook Data3: raw survey top-1% share 3.08% (1978) -> 6.57% (2015) vs corrected 5.78% -> 12.68%; the raw 2015 value (6.5653%) equals 2014 (6.5651%) to four digits. Undistributed/tax-exempt income rises from under 5% to 15% of national income (PYZ p. 2478).

Why this severity. An important undisclosed caveat on a headline number rather than a demonstrated error: the uncorrected survey share also rises and later vintages moved the 2015 share up (about 14%), so direction and order of magnitude survive, but the 2015 level and the corrected trend outside 2006-2010 rest on an untested constant-correction assumption and the 'lower bound' label is one-sided.

The authors' response. PYZ (2019, fn. 19) justify the constant profile by the absence of a trend in Piketty-Qian (2009) observed/simulated income-tax revenue ratios (1986-2003) and by provincial high-income-taxpayer data after 2011; PYZ (2019, p. 2471) argue evasion makes their estimates lower bounds.

Why it matters, in detail. Mechanism: capitalizing taxable interest with one economy-wide yield assigns too much fixed-income wealth to top interest recipients when their yields exceed the average, and the bias grows as rates fall (implied capitalization factor near 100 in 2012-13 per BHKS); later SZ vintages added a top interest premium after 2008. Direction: overstates the post-1990 and post-2000 rise. Magnitude: 1992-2013 rise +9.5 pp (plotted) vs +6.1 pp (WID 2026) vs +5.5 pp (SCF households); 1978-2012 rise -6.5% under SZ's own sensitivity; BHH 2001-2011 +3.5 pp vs +9 pp; including Social Security wealth lowers the SCF 1989-2019 rise by about 70%. The estate-multiplier series is flat, though its downward bias from rising mortality differentials is itself disputed.

Evidence. (i) SZ (2016) appendix Table B40 (higher fixed-income yield for the top 1%) vs Table B1: 2012 top-1% share 39.97% vs 41.82% (tax units); 1978-2012 rise 17.65 vs 18.88 pp (-6.5%) (outputs/sz2016_sensitivity.txt). (ii) Bricker, Henriques & Hansen (2018, FEDS 2018-024): heterogeneous returns cut the 2011 share from 40.5% to 33.9% and the 2001-2011 rise from about 9 pp to 3.5 pp. (iii) SCF households: 30% (1992) -> 36% (2013), 'less than half the increase' in SZ (BHKS 2016, p. 262); auditor B's own SCF computation 30.05% -> 35.50%. (iv) DFA: top 1% 'around 31 percent ... versus about 37 percent in the WID in 2014' (Batty et al. 2019, FEDS 2019-017). (v) Authors' current WID series (shwealj992): 29.5% (1992) -> 35.6% (2013), +6.1 pp, vs +9.5 pp (27.57% -> 37.03%) in the plotted series (Data3!AL); 1978-2015 +14.0 pp (21.8% -> 35.8%) vs +15.6 pp plotted (outputs/wid2026_wealth_shares.txt, workbook_checks.txt). (vi) Workbook Data5!J (estate method): 19.3% (1976), 21.0% (1984), 20.9% (1990), 20.8% (2000), 20.2% (2008). (vii) Catherine, Miller & Sarin (2025, JF 80(3), Table III): SCF top-1% share 1989-2019 +6.4 pp for marketable wealth vs +1.9 pp including Social Security wealth (range -0.2 to +3.7 pp).

Why this severity. Every marketable-wealth source shows a sizeable US rise and the authors' revised series still rises about 14 pp since 1978, so 'large rise' survives; but the rise over the last two decades is about a third smaller in the authors' own current series than in the line plotted, and smaller still under alternative methods or a broader wealth concept. A noticeable, not transformative, qualification.

The authors' response. SZ (2016 QJE, Section IV.F) reported the higher-yield sensitivity but kept equal returns as the baseline; SZ (BPEA 2019; NBER WP 27921, 2020) built a higher top interest rate since 2008 into the revised series and argue SCF yields are inflated by pass-through interest; SZ (2020 JEP, pp. 8-9) report that DFA and their series agree (38% in 2018) once units and concepts are harmonized; SZ argue Social Security promises are not wealth under SNA conventions (no direct reply to Catherine-Miller-Sarin located); SZ (2016) attribute the flat estate-multiplier series to rising mortality differentials and estate-tax avoidance (disputed by Kopczuk).

Why it matters, in detail. Mechanism: CHIP-to-CFPS comparisons show much larger rises in concentration than CHIP-only comparisons, and CFPS 2010 contains outlying housing values that alone move its Gini by about 10 points (KLW 2016), so part of the measured rise may reflect the change of survey instrument rather than of the distribution; the series breaks in the switch years (bottom 50% -4.2 pp in 2010; top 10% +4.6 pp in 2010 and +3.9 pp in 2011). Magnitude: top-10% rise 2002-2013 +17.6 pp (PYZ) vs +11 pp (CHIP-only), about 37% smaller; one-year bottom-50% drop of 4.2 pp at the switch. Top-1% effect unquantified.

Evidence. PYZ (2019, p. 2479, fns. 26-28); fn. 27: CHIP 2002 -> CFPS 2010 top 10% 39% -> 64% (Li & Wan 2015) vs CHIP 2002 -> CHIP 2013 37% -> 48% (Knight, Li & Wan 2016). Knight, Li & Wan (2016, Oxford DP 816, fn. 1): the CFPS 2010 Gini of 0.76 'is sensitive to the inclusion of some 300 implausibly or suspiciously large housing values. If they are excluded, the Gini falls by some ten percentage points.' Workbook Data3!R:U: bottom-50% wealth share 11.2% (2009) -> 7.0% (2010); top 10% 58.2% (2009) -> 62.8% (2010) -> 66.7% (2011); PYZ top 10% 49.0% (2002) -> 66.6% (2013), +17.6 pp vs +11 pp in CHIP (outputs/workbook_checks.txt).

Why this severity. All sources show rising Chinese wealth concentration, so the direction survives; the speed of the rise in the top-10% share may be about a third smaller with comparable surveys, and the effect on the plotted top-1% series (driven by the Hurun anchor and Pareto interpolation) is not demonstrated.

The authors' response. PYZ (2019, fn. 28) vary the P90 upgrade factor between 1 and 2 with 'second order' effects on the top-10% share; CHIP 2013 housing values were not publicly released (Knight, Li & Wan 2022, RoIW 68(1), fn. 3).

Why it matters, in detail. For the US the 2015 macro average is genuine and only shares are lagged one year (effect on bottom-50% or top-1% growth about 1 pp in a single vintage). For France both shares (two years) and the average (one year) are lagged; very-top French shares are volatile year to year (2013 is a tabulation-based, non-micro-file year). Replacing the carried-forward 2013 shares with GGP's actual 2014 data moves top-0.01% growth from +93% to +69% and top-0.001% growth from +158% to +95%; in WID 2026, true vs carried-forward 2015 gives +71% vs +34% and +47% vs +5%.

Evidence. Data3!AA43 '=AA42', AD43 '=AD42' (US); W42 '=W41', W43 '=W42' and likewise X-Z (France); Data4!F8 = 0.12531 (US 2014), J8 = 0.22509 (France 2013); Data4!L7 EUR 34,578 vs GGP (2017/18) Table 1 2014 average EUR 34,580; Data2 2014-15 copy formulas (outputs/workbook_checks.txt). GGP (2017/18, fn. 7): 2013-2014 (like 1976-1978) are years without income-tax micro-files. France with GGP's actual 2014 distribution and the same 1978 inputs: top 1% +73% (printed +67%), top 0.01% +69% (+93%), top 0.001% +95% (+158%); average and bottom 50% unchanged (+39%) (outputs/table1_repro.txt). In WID 2026 the same construction changes US bottom-50% and top-1% growth by about 1 pp but gives +5% instead of +47% for France's top 0.001% (outputs/wid2026_table1.txt).

Why this severity. Carrying forward the last observed distribution is standard WID practice and leaves the headline rows essentially unchanged; it matters only for France's very-top rows (which the paper itself calls numerically negligible), in a direction that cannot be determined in general. A documentation problem that does not affect conclusions.

Why it matters, in detail. Published vs posted China public share differs by up to 13 pp in the mid-1980s (69.6% vs 56.8% in 1985) and by about 4 pp in 2015; the decline 1978-2015 is 33 pp (published) vs 37.5 pp (posted), so the qualitative 'spectacular' transfer to private wealth holds in both. The US Figure 3 discrepancy is a stale chart reference (2012: 41.8% tax units vs 38.8% equal-split); the published figure uses the conceptually correct series.

Evidence. Vector extraction of AER p. 407 (outputs/published_fig2_vectors.txt): China public share 68.9% (1978), 69.6% (1985), 35.9% (2015). January 2017 WID WP Fig. 2b: 69.5% (1985), 35.8% (2015); Feb and Apr 2017 NBER Fig. 2b: 57.4% (1985), 32.2% (2015) (outputs/versions_fig2b_series.txt); workbook Data2!W: 69.2% (1978), 56.8% (1985), 31.7% (2015); no Data2 column lies within 10 pp of the published path. The NBER versions print '35%' next to a figure ending near 32%. PYZ (2019, p. 2480): '30 percent of national wealth is public'; WID 2026: 33.8%. Figure 3: published US line = Data3!AL (2012 38.85%, 2014 37.24%; outputs/pub_fig3_series.txt); workbook chart references Data5!P (SZ tax units, 41.82% in 2012), as in the January and February versions (about 42.3% in 2012; outputs/nber_fig3a_series.txt).

Why this severity. Reproducibility of two figures and one secondary number (China's 2015 public share: 35% vs 31.7% in the posted data and about 30-34% in the authors' later work) is affected; no conclusion changes.

Why it matters, in detail. Bottom-50% real pretax income -0.5% (level) vs share -7.5 pp (-37% proportionally); post-tax bottom-50% growth +21% (1980-2014, mostly Medicare/Medicaid; post-tax disposable income flat at $15,000-17,000 per PSZ 2016, p. 21).

Evidence. AER p. 406; Data4!G8 $16,485.64 -> H8 $16,402.10 (-0.51%); share 19.99% -> 12.53% (Data3!AA6, AA43); PSZ (2016, NBER WP 22945, p. 19): 'After taxes and transfers, the bottom 50% only grew +21% since 1980'; p. 20: 'down to 12.5% in 2014'. Bozio et al. (2020, WIL WP 2020/22) find the US-France gap 'entirely due to differences in pretax inequality', consistent with footnote 3's US-France prediction.

Why this severity. Presentation slips; the table and the explicit pretax labelling are correct and no conclusion changes.

Why it matters, in detail. 'Rising' is start-date dependent: the UK top-1% wealth share is +1.1 pp over 1980-2012 and +4.7 pp from the 1984 trough. In WID 2026, four of seven rich or large economies show changes within +/-2 pp over 1980-2015; only three show large rises (US +12.6, China +13.8, France +8.9 pp). On bottom-50% vs average growth (ratio of growth factors), France (about 0.95) is an outlier relative to Germany (0.79) and the UK (0.83).

Evidence. Plotted UK series (AAM; workbook Data5!D): 21.2% (1978), 18.75% (1980), 15.2% (1984), 19.9% (2012). WID 2026 top-1% wealth shares 1980-2015: GB +0.2 pp, DE +0.8 pp, JP +1.8 pp, NO -1.9 pp vs US +12.6, FR +8.9, CN +13.8; top-1% income shares rise in all 7 (outputs/wid2026_generalizations.txt). Bottom-50% vs average real growth: DE +21% vs +54% (1980-2015), GB +50% vs +81% (1978-2015), FR +40% vs +47% (1978-2015). AER p. 405 ('broadly representative of the West European pattern'); p. 408.

Why this severity. The income generalization holds; the wealth generalization rests on four countries and holds strongly in later data only for the US, France and China. Presentation-level overreach that does not affect the documented country results.

The authors' response. None located. In the authors' favour: Alvaredo, Atkinson & Morelli's abstract reports 'evidence of an increase in top shares' after the UK decline ended around 1980, and in early 2017 WID held wealth distributions for few countries.

Why it matters, in detail. Cross-country differences in measured top-share trends confound policy, structure (sectoral composition, globalization exposure, demography) and measurement (data sources, imputations, deflators). With three to four countries and no identifying variation, the evidence is consistent with, but does not establish, a causal role for policies and institutions.

Evidence. AER p. 408 (Section V); abstract of NBER WP 23119. No decomposition or test is presented in the paper.

Why this severity. Hedged ('suggesting') and plausible, but it is an interpretation not shown by the evidence presented; it affects no reported number.