Filip Novokmet, Thomas Piketty, Gabriel Zucman · Journal of Economic Inequality 16(2): 189-223 (special issue in honor of Tony Atkinson)

From Soviets to Oligarchs: Inequality and Property in Russia, 1905-2016

20 problems found: 2 major · 8 moderate · 10 minor

What the paper claims

Official survey-based measures vastly under-estimate the rise of Russian inequality since 1990; corrected top income shares are very high.
Russian top income shares are now similar to (or higher than) US levels; intro: 'at least as high as in the United States, with a top 1% income share around 20-25%'.
Inequality increased substantially more in Russia than in China and ex-communist Eastern Europe, related to the shock-therapy transition.
Offshore wealth of rich Russians is about three times official net foreign reserves and comparable to total household financial assets held in Russia.
Long-run U-shape: high Tsarist inequality, very low Soviet inequality, high post-Soviet inequality; Tsarist and post-Soviet levels roughly comparable.
Transition from public to private property while the national wealth-income ratio stayed roughly constant.
Wealth concentration rose 1995-2015 to a level comparable to or higher than the US (explicitly hedged).
Per-adult national income grew ~41% over 1989-2016, but growth was concentrated at the top (bottom 50% negative).

Every problem found

1majorlikely · previously known · speculative assumption

Russian tax statistics group people by their gross takings: for a trader, total sales; for someone selling shares or a flat, the full sale price. To turn those takings into income, the authors assume everyone deducts 70% as costs. But most people in the top brackets are business owners and asset sellers who typically deduct 80-95%. With more realistic deductions, the share of income going to the top 1% falls from about 20% to somewhere between 14% and 19%. That is no longer clearly as high as in the US, though still far above what surveys show.

Technical detail and evidence

Why it matters, in detail. Mechanism: ratio1 = (1-r)·10M/thr_RLMS(P99.6), which equals 2.235 at r = 0.70; it is spread piecewise-linearly over P90-P99.6 and beyond. The 2015 PNI top 1% falls by about 0.45 pp per point of r: 20.2% (0.70), 18.8% (0.739), 16.2% (0.80), 13.9% (0.85), 11.9% (0.90). Bracket composition implies filers' r of about 0.8-0.9 at the anchor; imputed non-filers would have r of about 0. External 2023 tax counts (F4) are matched at r of about 0.74-0.80, implying an upward bias of roughly 1.5-4 pp in the benchmark top 1% (about 7-20% of the level). Effects on the comparisons: for C2, at r = 0.75 Russia lies 0.5-2.4 pp below the US (authors' vintage) in 2013-15, and at r = 0.80-0.85 it lies 3-7 pp below, exceeding the US only in 2008-10. For C3, the margin over China (13.9%) shrinks from 6.3 pp to 4.5 pp (r = 0.75), 2.3 pp (0.80) and 0 (0.85). C1 survives: the fiscal top 1% stays at least 1.9x the raw RLMS share for r <= 0.85.

Evidence. Paper p.195 ('In our benchmark series we assume the same deduction rate for all brackets'), pp.195-196; Appendix B.1 pp.18-19. TB13 in the 2017 package and journal ESM2 (identical): O5:V9 deduction rates, B5:F9 declaration profiles. Raw 1ddk2015.xlsx: Р1 codes 1040-1080 (29,988 / 4,383 / 587 / 398 / 26 filers); Р3 codes 3120-3150 (entrepreneurial revenue alone: 12,894 / 2,508 / 328 / 205, i.e. 43%, 57%, 56%, 48% of filers); Р3 3030/3050 expenses = 93.4% of entrepreneurial revenue (95.6% for individual entrepreneurs). 'Справочно к Р1': securities 977bn gross vs 93bn taxable (90.5% deducted), other property 95.5%, real estate 81.7%, declared dividends 8.3%. Adjudicator re-implementation (scripts/npz_taxcorr.py, npz_copula.py, sensitivity2015.py) reproduces the 2015 fiscal top 1% (17.93%) and the PNI top 1% (20.25% vs published 20.24%). Uniform deduction rates of 0.739 / 0.75 / 0.80 / 0.85 / 0.90 give a PNI top 1% of 18.8 / 18.4 / 16.2 / 13.9 / 11.9% (top 10%: 43.4 / 42.8 / 39.7 / 36.3 / 34.2%). A's composition-adjusted rates (0.82/0.85/0.84) give 14.5%; the authors' declaration variants give 18.4-26.3%. B's table (20.24/18.77/16.23/13.87/11.89/10.11%) reproduces exactly.

Why this severity. Within a defensible range of the anchor deduction rate (about 0.74-0.85), the 2015 PNI top 1% moves from 20.2% to 18.8-13.9%. That is enough to put Russia below the US (20.8%) in 2015. At the upper end Russia would be level with China (13.9%) and, on a fiscal basis, with Poland. The qualitative finding (corrected shares well above the raw-survey 5.8%) survives every variant up to an 85% deduction rate (fiscal top 1% at least 10.8%).

The authors' response. No published reply located. The paper pre-emptively states that the tabulations are imperfect and that deduction and declaration rates must be assumed. It also states that 'in all possible variants' corrected shares are much larger than raw survey shares (pp.195-196, 216; Appendix B.1).

2majorlikely · newly identified here · speculative assumption

To estimate wealth hidden abroad, the authors add up money that left Russia unrecorded each year. They assume it grew as fast as Russia's economy measured in dollars, which boomed in the 2000s. With ordinary investment returns instead, the estimate drops from about 85% to roughly 45-75% of national income. That is still a very large amount, comparable to all the financial assets Russian households hold at home, but not clearly 'three times' official foreign reserves.

Technical detail and evidence

Why it matters, in detail. Because r_t equals per-capita dollar NI growth, W_T/Y_T is approximately the sum of F_t/Y_t, so the roughly $106bn of 1992-2001 flight is scaled up by the 1999-2008 dollar boom in Russian NI. Central offshore stock: 86.4% vs 42-75% under fixed 0-6% net returns and 47% under T-bill compounding. Multiples: vs NFA (25.0%), from 3.5x to 1.7-3.0x; vs reserves (30.6%), from 2.8x to 1.4-2.5x; vs household financial assets (63.1%), from 1.37x to 0.67-1.2x. The g plus or minus 2 pp band does not span these conventions. Household attribution of all residuals and the inclusion of emigrants add uncertainty in both directions. The authors' round-tripping interpretation (assets reinvested in Russian firms) would justify returns nearer g.

Evidence. Paper p.205 and fn 23; journal ESM1 A33 rows 10-33 (reproduced exactly: 67.22/86.41/111.57%); DataBOP rows 57, 69, 71. Adjudicator offshore.py; B's fixed-rate figures reproduce exactly; FRED TB3MS. Comparators: NFA 24.97% (ESM1 A29b), household financial assets 63.15% (A21), reserves $368.4bn = 30.6% of NI (CBR 2015 BoP/IIP release, IIP Table 5.1, reserve assets at 1 Jan 2016). AJZ (2018) JPubE, Fig. 5 and text (China at 2.3% of GDP is '20 times less than in Russia' using NEO and '6 times less' using BIS data).

Why this severity. Defensible return assumptions change the headline level by -13% to -51% (-45% under the T-bill convention), turning 'about three times net foreign reserves' into roughly 1.5-2.5 times. The qualitative claims survive every variant: offshore wealth is large, a multiple of official net foreign assets, and of the same order as onshore household financial assets.

The authors' response. No reply located. Pre-emptive text: the estimates are 'by construction relatively conservative' because return differentials and valuation effects are taken as given (p.205); much offshore wealth may be reinvested in Russian companies (p.206).

3moderateconfirmed · newly identified here · internal inconsistency

The appendix says the authors used the deduction rate actually observed in the tax statistics (about 74% in 2015), but the computer code uses 70%. That small difference adds about 1.5 percentage points to the headline top-1% share.

Technical detail and evidence

Why it matters, in detail. Following the documented rule would give r = 0.739 in 2015, lowering the PNI top 1% from 20.2% to 18.8% and the fiscal top 1% from 17.9% to 16.3%. With observed rates Russia is below the US in 2014-15 (19.2/18.8% vs 20.2/20.8%). In 2009-10 the observed rates would cut the fiscal top 1% by 4.6-6.4 pp; the authors smooth this away with 0.80. The unexplained 0.70 moves the 2013-15 endpoint in the inequality-increasing direction.

Evidence. TB13 O5:V9 vs A20:H20 (both vintages); Appendix Table B12 (PDF pp.67-68) and B10 (p.61). Adjudicator observed_rates.py, fiscal top 1% with operative vs observed rates: 2015 17.93 -> 16.29; 2014 18.72 -> 17.38; 2013 19.64 -> 18.32; 2010 18.61 -> 12.24; 2009 20.15 -> 15.58; 2008 23.67 -> 25.36. series_alt.py, PNI top 1% with observed rates: 2015 20.2 -> 18.8. These match A's figures.

Why this severity. This undisclosed implementation choice raises the headline 2015 top 1% by about 1.5 pp (about 7%) and the 2013-14 values by 1.2-1.3 pp. The 2009-10 smoothing is defensible but shows how sensitive the series is to this single parameter.

The authors' response. Spreadsheet note TB13!A17 (2009-10 smoothing only); no published discussion.

4moderateconfirmed · newly identified here · selective presentation

The paper's sensitivity checks never changed the assumption that matters most for the top-1% figure, so the reassuring statement that all variants give similar results is not a real test.

Technical detail and evidence

Why it matters, in detail. The published deduction variants span 16.5-19.3% (fiscal top 1%, 2015). Varying r at the anchor over 0.739-0.85 spans 16.3-10.8% fiscal (18.8-13.9% PNI). The published uncertainty band therefore omits most of the parameter uncertainty relevant to C2-C3. The declaration-rate variants (15.8-24.6% fiscal) mostly explore the inequality-raising side.

Evidence. TB13 X5:AA9; do_gpinter_RussiaRLMS.do (ratio1 uses rded`year'[128]*rded`variant'[128]). Authors' output CorrectedSurveySeriesRLMSTax.xlsx v31-v34 (2015 fiscal top 1% 16.5-19.3%, per A) vs the adjudicator anchor-rate range (fiscal 16.3-10.8% for r = 0.739-0.85). Paper pp.196, 216.

Why this severity. This does not change the estimates but misstates the robustness of the headline level. Readers cannot see that the most influential parameter was never varied in the direction the bracket composition suggests.

5moderateconfirmed · newly identified here · version instability

Between the draft and the published version the authors changed how they estimate hidden wealth, roughly doubling the money counted as leaving the country. The article doesn't say so and still quotes some numbers from the old method. This makes the estimate hard to check, though the published central figure can be recovered from the journal's supplementary files.

Technical detail and evidence

Why it matters, in detail. Between versions both inputs changed: flows doubled (fictitious transactions $377bn replaced capital transfers) and the benchmark return fell by 2 pp. With r = g the WP flows imply 56% and the journal flows 86%. The flow definition alone moves the estimate by about 30 pp, a dimension of uncertainty that appears nowhere in the published range.

Evidence. NPZ2017 WP (PDF pp.21-23) vs NPZ2018 pp.204-206, fn 23-24. A17!A33 vs ESM1!A33 row 33 (NI $1,137bn vs $1,204bn; stock $844bn vs $1,040bn); ESM1!A35 note; author page https://gabriel-zucman.eu/russia (links the NPZ2017 files). Adjudicator offshore.py: flows of $250bn vs $511bn; with r = g the WP flows give 56.1%.

Why this severity. The central journal figure (85%) can be recovered from the journal ESM, so the conclusion is unaffected. But two undisclosed flow definitions differ by a factor of two, and the published text is internally inconsistent; this impedes replication and overstates the estimate's stability.

6moderateconfirmed · previously known · speculative assumption

There are no tax data before 2008, so for 1980-2007 the authors chose the size of the correction by hand and joined it smoothly to the 2008 value. The reported steady rise of the top 1% in the 2000s therefore reflects that choice; the raw surveys show the opposite trend.

Technical detail and evidence

Why it matters, in detail. The corrected b(0.9) rises from 2.05 to 3.01 over 1998-2007 while the survey b falls from 1.89 to 1.47. Alternative interpolations give a 2007 fiscal top 1% of 21-29% (vs 26.4%), and a fixed tail reverses the subperiod trend. The 1999-2008 dynamics in C1 and Fig. 8 are model output, not measurement.

Evidence. ESM2/2017 TB0 I85:I112 formulas and values; do_gpinter_RussiaFinal.do. Alternatives: B finds that holding the 2008 tax-corrected b(0.9) = 3.558 fixed turns the 1998-2007 fiscal top 10% from 41.8% -> 47.8% into 55.5% -> 52.0%. A finds a 2007 fiscal top 1% of 21.4% with a /10 divisor, 28.8% with the full anchor, and 7.7% if the 1998 correction ratio is held.

Why this severity. The endpoints (late Soviet; 2008-15) are unaffected, so the 1990-2015 rise survives. But the 1999-2007 timing (continued concentration up to a 2007-08 peak 'over 26%') is assumed, not observed.

The authors' response. Fn 12: the 1990-2007 profile 'appears to be relatively robust', but tax tabulations before 2008 would be preferable; no reply.

7moderateconfirmed · newly identified here · data error

A data-entry error made the 2015 estimate of the richest 1%'s share of wealth use the previous year's billionaire list. Corrected, the share is 40% rather than 43%. Russia's wealth concentration remains very high and comparable to the US.

Technical detail and evidence

Why it matters, in detail. Top 1% wealth: 42.58% -> 40.02%; top 10%: 71.32% -> 70.04% (B). The 2014 -> 2015 change shrinks from +5.7 pp to +3.1 pp. The effect on income shares through the copula is about -0.26 pp.

Evidence. GpinterWealth/WealthParameters.dta (year2 = 2015) vs TB1 B34:C35; adjudicator wealth2015.py: stale inputs 42.5818% (= published WealthSeriesRussiaBenchmark 2015); TB1 inputs 40.0205% (top 0.1%: 26.3% -> 23.9%). B obtains 40.0205% independently and estimates the PNI top 1% falls by about 0.26 pp.

Why this severity. A confirmed data error in a secondary, explicitly hedged result: the 2015 top-1% wealth share falls 2.6 pp (about 6%), and most of the 2014-15 jump in Fig. 11 is an artifact. The comparison with the US (38.6%) survives.

8moderatelikely · previously known · methodology

The paper's dramatic finding that the poorer half's share of income fell from 30% to under 10% between 1990 and 1996 stitches together two data sources that disagree strongly about the mid-1990s. The fall was real, but how deep it went depends on which source you trust.

Technical detail and evidence

Why it matters, in detail. About 5-6 pp of the roughly 20 pp fall over 1990-96 coincides with the switch from Rosstat to RLMS. RLMS shows a further 5.7 pp fall over 1994-96 that Rosstat does not (+3.3 pp). The 0.9 adjustment lowers the 1990-91 starting point by about 4 pp (raw Rosstat 1991: 35.4%). Monthly x 12 annualization during arrears likely inflates measured annual dispersion.

Evidence. Gpinter/Output/RawSurveySeriesHBS.xlsx and RawSurveySeriesRLMS.xlsx ('series'); FinalSeriesCopula.xlsx 'series' (bottom 50%: 29.5% in 1990, 20.1% in 1993, 15.3% in 1994, 9.6% in 1996). RawSurveyTabulationsHBS.xlsx sheets 1990RU = 1991RU; do_gpinter_RussiaFinal.do ('replace thr=0.9*thr if p<0.5' for 1990-91). Paper p.197 fn 11, p.215.

Why this severity. The 1996 trough and the depth of the 1990s collapse, a prominently reported result, are source-dependent by about 10-15 pp. The long-run decline in the bottom-50% share (30% in 1989 to 17% in 2015) and the 1989-2016 growth incidence are not affected.

The authors' response. Fn 11 notes that HBS income is measured indirectly and that only 'rudimentary and relatively opaque' tabulations exist; no reply.

9moderatelikely · previously known · misrepresentation of source

The 1905 starting point is not a historical table but a construction: a 1928 Soviet wage distribution stretched to resemble 1914, with the top set by assumption. Detailed independent studies find pre-revolution inequality lower than the paper does. The claim that Tsarist and modern Russia were about equally unequal is therefore uncertain, though both were clearly far more unequal than the Soviet period.

Technical detail and evidence

Why it matters, in detail. If 1905 inequality is nearer L&N (Gini ~0.36-0.40; top 1% ~13.5-15%), the Tsarist peak of the U is 5-10 pp lower in the top-10% share, and 1905-56 growth was less tilted to the bottom 90%. The asymmetric treatment of Soviet-era non-monetary privileges further exaggerates the amplitude of the U.

Evidence. Gpinter/Input/RawTabulationsEmplSurvey.xlsx sheets '1905', '1914', '1928' (formulas) and ReadMe item (iii). Lindert & Nafziger, NBER WP 18383 (2012), pp.22-23 and Table 7. Didenko D.V. (2022), Ural Historical Journal 1(74): 6-15, doi:10.30759/1728-9718-2022-1(74)-6-15, pp.9-10. NPZ fn 31 says only that L&N argue the official estimate 'might be somewhat overestimated'.

Why this severity. The U-shape survives, since Soviet levels are far below either end. But 'Tsarist and post-Soviet inequality roughly comparable' and the 1905-1956 growth incidence rest on a construct. Independent estimates put 1904-10 inequality between Soviet and post-Soviet levels. The paper itself warns that the 1905 estimate is imprecise.

The authors' response. Pp.212-214 and fn 31: the 1905 estimate is 'at least as imprecise' as recent ones, and tax-administration projections tend to underestimate top incomes; no reply to L&N or Didenko.

10moderateplausible · newly identified here · external inconsistency

Official Russian tax data for 2023 show fewer high earners than the paper's 2015 estimates imply, even after allowing for income growth: about 20% less income at the top. That suggests the headline top-1% figure is somewhat too high, by a few points rather than by a lot.

Technical detail and evidence

Why it matters, in detail. At tax level, adults above 2.31M rubles (5M in 2023 rubles): NPZ 0.656% vs MoF 0.473%; above 4.63M: 0.254% vs 0.149%. Income above them: 8.55% vs 6.99% and 6.51% vs 5.38% of NI. If top incomes grew faster than average after 2015 (WID's own series has the top 1% at 23.8% by 2020), the consistent r is about 0.77-0.80. Caveats work in both directions: emigration and sanctions after 2022; de-offshorization raising 2023 declared income; entrepreneurs under special regimes outside PIT; and uncertainty over whether the reported groups use gross income or the tax base.

Evidence. Interfax/Finmarket 31 May 2024 (https://www.finmarket.ru/main/article/6186649), citing Vedomosti 31 May 2024 (MoF material for the Russian Tripartite Commission, 24 May 2024): 167k >10M, 1.2 trn PIT (19%); 364k 5-10M, 323bn PIT; 53.6M <1M and 11.2M 1-5M, each 2.5 trn (38%); 2-pp surcharge above 5M yielded 159.5bn in 2023. WID RU bulk (anninci992 x inyixxi999, npopuli992). Adjudicator minfin_triangulation.py: income-share ratios 1.22 (r=0.70), 0.99 (0.739), 0.93 (0.75), 0.68 (0.80), 0.47 (0.85), 0.29 (0.90). A's external_minfin_check.py gives the same tax-level figures.

Why this severity. Points to a benchmark overstatement of roughly 1.5-3 pp in the PNI top 1%, consistent with F2. That is a noticeable but not transformative change, and the evidence is indirect (eight years apart, with large shocks in between).

11minorconfirmed · newly identified here · misrepresentation of source

The paper says a certain type of payment is under 10% of the relevant flows, and uses this to argue that the hidden wealth belongs mostly to people still living in Russia. In fact it is about half, and mostly reflects the Russian government writing off other countries' debts (Cuba's in 2014, for example). It doesn't change the published estimate, but the argument is wrong.

Technical detail and evidence

Why it matters, in detail. Capital-transfer payments are 53% of |NEO| + capital transfers, and 24% of |NEO| + fictitious transactions. The residency inference therefore rests on a false statistic, and the WP and 2017 files embed roughly 20 pp of NI of sovereign write-offs as private capital flight.

Evidence. ESM1 DataBOP rows 34-36 and 69 (1997-2015 sums, adjudicator offshore.py); CBR, 'Balance of payments, IIP and external debt of the Russian Federation in 2015' (Vestnik, June 2016), p.10 text and capital-account table (2014 payments $42,147m); A's CEIC summary (record low -$31.8bn in Q4 2014).

Why this severity. Affects an interpretive side argument (who holds the wealth) and the superseded WP number, not the published estimate.

12minorconfirmed · newly identified here · internal inconsistency

Two small spreadsheet inconsistencies in the hidden-wealth calculation; correcting them changes the estimate by about one percentage point.

Technical detail and evidence

Why it matters, in detail. Applying the stated definition to 1992-96 lowers the 2015 central stock by 1.38 pp of NI; the cell-reference slip is negligible (+0.03 pp).

Evidence. ESM1 DataBOP BI71:BM71 vs rows 53 and 62; A33!K11 formula '=F11'. Effects (B, consistent with adjudicator recursion): 85.03% instead of 86.41% (-1.38 pp); 86.44% after fixing K11 (+0.03 pp).

Why this severity. Small spreadsheet inconsistencies that change the central estimate by about 1.4 pp.

13minorconfirmed · newly identified here · misrepresentation of source

The paper compares hidden wealth with 'official foreign reserves' but actually uses a different statistic, the country's net foreign assets. Against actual reserves, hidden wealth is about 2.8 times larger rather than more than three times.

Technical detail and evidence

Why it matters, in detail. Offshore/reserves = 2.82x (central), not 'more than three'; offshore/NFA = 3.46x.

Evidence. ESM1 A29b S74 (NFA 24.97% of NI); CBR 2015 BoP/IIP release, IIP Table 5.1 (reserve assets $368,399m at 1 Jan 2016; B cites Table 3.3 p.78); ESM1 A33 NI $1,204bn.

Why this severity. Terminology and a slightly overstated multiple; the order of magnitude is unchanged.

14minorconfirmed · newly identified here · documentation gap

The files needed to rebuild the survey part of the analysis from the raw survey data are not included, so that step cannot be checked independently.

Technical detail and evidence

Why it matters, in detail. The first of three pipeline steps cannot be reproduced; the downstream correction and copula reproduce exactly from the supplied tabulations.

Evidence. ReadMeNPZ2017DistributionSeries.pdf; Gpinter/ReadMeNPZ2017Gpinter.pdf; package inventory (B's additional_results.json; adjudicator directory listing). Paper p.192 says the appendix includes 'all our raw data sources and computer codes'.

Why this severity. No demonstrated numerical error. It limits verification of the stage behind all incomes below P90, and contributed to a critic misreading the gross-up.

15minorconfirmed · newly identified here · overclaiming

The introduction says Russia's top income shares are 'at least as high' as America's, but in the paper's own latest-year numbers they are slightly lower; 'similar' would be accurate.

Technical detail and evidence

Why it matters, in detail. Russia is 0.6-2.0 pp below the US in 2015; under F1/F2 alternatives the gap widens to 2-7 pp.

Evidence. NPZ2017MainFiguresTables.xlsx DataSeriesInequality columns I-J, AR, AU (adjudicator maincompare.py); WID bulk 2026 vintage (RU, US sptincj992); paper p.191.

Why this severity. Wording that the paper's own latest-year data do not support; the abstract is accurate.

16minorconfirmed · newly identified here · methodology

When comparing with Poland, Hungary and the Czech Republic, Russia is measured with a broader income definition, which makes its numbers look a bit higher. The difference is disclosed in a figure note but not in the text.

Technical detail and evidence

Why it matters, in detail. On a fiscal-to-fiscal basis the Russia-Poland gap is 4.5 pp, not 6.8 pp, in 2015. At r = 0.80 it is about 0.

Evidence. Fig. 10a note p.219; DataSeriesInequality 2015: Russia fiscal 17.9%, Poland 13.4%, Czech Rep. 9.6%; adjudicator sensitivity: Russia fiscal 13.5% at r = 0.80.

Why this severity. Disclosed; the benchmark gap remains, but it narrows on a like-for-like basis and closes against Poland under F1 alternatives.

17minorconfirmed · newly identified here · arithmetic or transcription

Several small mismatches between text and data (a peak dated 2008 instead of 2007, '18%' instead of 17%, a few ranges); none changes the conclusions.

Technical detail and evidence

Why it matters, in detail. Each discrepancy is at most 1-2 pp, or is a dating or labelling error. Replacing the 2012 counts leaves the snapped percentiles and the benchmark unchanged.

Evidence. Main workbook DataSeriesInequality/DataSeriesIncome; TB0 E-F; TB10; TB13 K5:L9 vs raw 1ddk2012.xlsx (26,516/4,016/553/383/10); B's chart XML check (FB11 chart8.xml).

Why this severity. Presentation and transcription slips; none changes a conclusion.

18minorconfirmed · newly identified here · speculative assumption

The finding that average income per adult rose 41% since 1989 depends on the authors' own revision of 1990s price statistics; figures closer to the official ones give 9-19%. This affects how much better off Russians are on average, not the inequality findings.

Technical detail and evidence

Why it matters, in detail. Average growth ranges from +8.5% to +41% (up to +70% with welfare adjustments, per the authors). Group shares are unaffected, so the distributional findings in C8 hold.

Evidence. NPZ2017MainFiguresTables.xlsx DataSeriesIncome columns B and I (adjudicator check: 16,431.5 -> 23,181.1 = +41.08%; 19,510.7 -> 23,181.1 = +18.81%); paper p.190, fn 6, p.216.

Why this severity. Disclosed. Lower average growth would make bottom-50% growth more negative, so the revision does not favour the inequality conclusion.

The authors' response. Fn 6 and p.216 justify the revision: official deflators would imply implausibly high Soviet living standards.

19minorconfirmed · newly identified here · misrepresentation of source

The paper slightly overstates how closely an earlier study of leaked Moscow tax records matches its own results, but that study does support the general point that surveys miss the rich.

Technical detail and evidence

Why it matters, in detail. The external validation is weaker and less specific than stated. NPZ's own 2015 fiscal correction (about +12 Gini points) is within G&R's range.

Evidence. Guriev & Rachinsky (2006, UNU-WIDER RP 2006/120), pp.15-18 and Table 4; NPZ p.197 and fn 12.

Why this severity. Embellished citation. G&R's qualitative conclusion (surveys understate the Gini by 10-15 points, or about 25 points in their conclusion) does support the direction of NPZ's correction.

20minorplausible · previously known · methodology

The surveys split a married couple's income equally between the two partners, but the tax data are for individuals. Mixing the two slightly exaggerates top incomes, probably by about one percentage point of the top-1% share.

Technical detail and evidence

Why it matters, in detail. A threshold scale factor of 0.85-1.0 lowers the 2015 PNI top 1% by 0-1.7 pp (0.90 gives -1.1 pp). This adds to F1/F2 in the same direction.

Evidence. do_gpinter_RussiaRLMS.do (ratio = tax threshold / survey threshold at the same p); Table 1 note (equal-split adults). Adjudicator sensitivity2015.py: scaling tax thresholds by 0.90/0.85/0.80 gives a PNI top 1% of 19.1/18.6/18.0%. Pareto-tail approximation: equivalent equal-split threshold is about (1+phi)/2 x 2^(1/alpha) of the individual one, i.e. about 0.85-1.1 for spouse income share phi = 0.2-0.4 and alpha = 1.5-2 (factors above 1 imply no upward bias).

Why this severity. The bias is upward but plausibly no more than about 1-1.5 pp of the top 1%.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

RLMS income is net of the 13% PIT and was not grossed up; grossing the survey up by 1/0.87 lowers the 2015 fiscal top 1% from 17.9% to 16.3% (A's I4, 'tax status'; echoing Kapeliushnikov 2020). — The RLMS tabulations are already grossed up. Their thresholds are round net values divided by 0.87: 2015 206,897 = 180,000/0.87 and 620,690 = 540,000/0.87; 2008 282,759 = 246,000/0.87; 2012 413,793 = 360,000/0.87. Neef (2022, WID Technical Note 2022/03, p.1) also states that NPZ 'simulated the income tax to convert net incomes to gross incomes'. A further gross-up would double count. B reached the same conclusion.
A 'level-consistent' comparison (scaling the survey to the national-accounts total before comparing with tax data) gives a 2015 fiscal top 1% of 12.4%, showing the method in effect assumes tax-reported top incomes are under-reported like everything else. — Not a demonstrated error. Proportional NA scaling leaves shares unchanged. Comparing tax-reported and survey-reported incomes at the same percentile is the standard approach. Pre-scaling the survey by the full 1.68 NA gap assumes the survey under-reports at P99.6 by the same 40% that the tax correction is meant to measure, which double counts. How the residual NA gap is allocated remains a generic, unquantified caveat (international-global-01).
No updated appendix or data files document the journal's offshore method. — The journal's electronic supplementary material (Springer MOESM1/MOESM2; ESM1 internally dated 9 May 2018), obtained by B, contains the revised computation. It reproduces exactly 86.4/67.2/111.6%. The valid residual (the author site distributes the 2017 package, and the stale A33 note) is retained in F6.
After national-accounts scaling, NPZ has 2.1-2.3x as many adults (1.7-1.9x the income) above the MoF/FNS 2023-equivalent thresholds. — Not adopted as the measure of discrepancy: it compares NA-scaled incomes with raw tax data. The like-for-like tax-level comparison (1.4-1.7x in counts, 1.2x in income) is retained in F4.
Check passed: 'The Tsarist 1905 distribution is correctly described as a tax projection, not realized tax-return data.' — Overturned in part. The 1905 tabulation is built from the 1928 Soviet wage distribution, hand-set ratios and a typed top Pareto coefficient, with only the top calibrated to Gregory's tax-projection-based share (F12).
Corpus critique examined by both auditors (Kapeliushnikov 2020): realistic Russian inequality is about one-third lower than NPZ (Gini 0.3-0.4), and the bottom-bracket declaration rate is far too low (moving ~32M people up the distribution). — Not supported. Only the 10M+ brackets enter the correction (rows 128-132 in do_gpinter_RussiaRLMS.do), so the bottom-bracket argument has no effect on NPZ's numbers. The deduction-rate critique is upheld in part (F1), but external 2023 tax counts imply a modest overstatement (about 1.5-4 pp of the top 1%), not a one-third lower Gini. Both auditors reached the same conclusion.
The offshore 'return = g' is the growth of Russian NI in current US dollars. — Minor correction: ESM1 A33 column F is per-capita dollar NI growth (NI growth divided by population growth), as B states. Immaterial to the findings.

Audit notes

What reproduced. Reproduced (independent Python, adjudicator): - The 2008-2015 tax correction of RLMS g-percentiles, exactly (all years). - 2015 pre-tax national income shares via a discretized Gumbel copula (within 0.02 pp), and the 2008-2015 PNI top 1% within 0.1 pp. - The journal and working-paper offshore stocks, exactly. - The 2010/2014/2015 wealth corrections; 2015 only with the stale 2014 Forbes inputs (40.02% with correct inputs). - The 1999-2007 Pareto-coefficient formulas, inspected in TB0. Both auditors' key numbers reproduced exactly: B's deduction-sensitivity table and fixed-return offshore table; A's observed-rate figures and MoF tax-level comparison. New adjudicator computations: T-bill compounding of offshore flows (47.4% of NI); triangulation of the 2015 tail against 2023 MoF/FNS counts across deduction rates; the 2008-2015 PNI series under alternative deduction rates against the US; quantification of the Rosstat/RLMS splice; checks of debt forgiveness and the IIP treatment of fictitious transactions in a CBR primary release. Not reproduced: - The RLMS microdata-to-tabulation stage (code and data absent). - gpinter itself (shares computed directly from the 127 corrected brackets; exact for the benchmark). - The Soviet/Tsarist archival tabulations and Appendix A national-accounts constructions (B reproduced the housing arithmetic).

Limits of this audit. Not used or not re-run: - No RLMS microdata were used, so annualization, weighting, equal-splitting and wage-arrears effects were inferred, not tested. - gpinter was not re-run: alternative distributions were computed directly from the 127 corrected brackets (exact for the benchmark), and the copula is discretized. Sources: - The MoF/FNS 2023 figures come from press reports of a ministry presentation (Interfax/Finmarket; Vedomosti, partly paywalled), not from a primary FNS table; the 2015-2023 comparison rests on growth deflation across major shocks. - The Springer ESM workbooks and the CBR 2015 release were used from the shared cache (downloaded by Auditor B). - Kapeliushnikov was read in the HSE preprint (WP3/2020/01), not the Voprosy Ekonomiki layout. - Didenko (2022) was read in the published PDF; van Zanden et al. and Morrisson were not read directly (cited via Didenko). - Debt-forgiveness attribution for years other than 2014 rests on A's CEIC summary. Not done: - No published author reply to any critique was located. - The Soviet-era tabulations, the Appendix A balance-sheet constructions (housing, land, deflator) and the foreign comparator series were not independently rebuilt. - No confidence intervals can be derived; ranges reported here are scenario ranges.