Camille Landais, Thomas Piketty, Emmanuel Saez · Paris: Le Seuil (La République des idées), 2011 (about 140 pages)

Pour une révolution fiscale : un impôt sur le revenu pour le XXIe siècle

14 problems found: 2 major · 3 moderate · 9 minor

What the paper claims

All compulsory levies combined, the French system is 'légèrement progressif jusqu'au niveau des « classes moyennes », puis devient franchement régressif au sein des 5 % les plus riches – et surtout à l'intérieur des 1 % les plus riches'; 'la réponse est sans appel'. Back cover: 'Il démontre scientifiquement, pour la première fois, le caractère régressif de l'impôt'.
Modest earners pay a higher overall rate than the richest: 'les revenus modestes supportent ... des taux effectifs d'imposition de l'ordre de 45 % à 50 %, alors que les plus riches sont à 30 % ou 35 %'; 'les personnes disposant de 1 700 euros par mois paient 45 % ..., alors que ceux qui ont 63 000 euros par mois paient 35 %'.
The cause is that capital income largely escapes the progressive income tax (plus regressive consumption taxes and capped social contributions on labour).
A revenue-neutral individualised progressive income tax replacing IRPP, CSG, CRDS, PL, PPE and the bouclier fiscal (effective-rate schedule 2% to 60%) restores progressivity at the top and cuts taxes for almost everyone below about 7,000 EUR/month.
Behavioural responses reinforce the reform; negative net effects are implausible.
Per-child family support (family benefits + quotient familial) is flat at ≈175 EUR/month across deciles 1-9 and rises to ≈400 EUR at the top; the top 10% receive 1.5-2 times more per child.
Macro and distributional benchmarks for 2010.

Every problem found

1majorconfirmed · previously known · methodology

The book counts pension and unemployment contributions as taxes. Yet they mostly buy workers their own future pensions and benefits, and the very rich pay almost none of them because they are capped and do not apply to investment income. Counted instead as deferred pay, as the authors' own unpublished tables do, they leave the lowest-paid full-time workers paying about 27% of their income, less than the richest (33%). The fall in rates at the very top is then about half as large.

Technical detail and evidence

Why it matters, in detail. The convention adds the full contributory wedge c to the numerator without adding the corresponding deferred benefits to income. c is about 15.5-18% of primary income for P0-P90, 7.5% at P99-99.9 and 0.4% at P99.999-100, because contributions are capped and apply only to labour. Treating c as deferred pay, τ_sec = (τ−c)/(1−c), or re-ranking on the authors' y_sec/impot_sec, gives these changes: P0-10 falls 40.5→27.4%, P90-95 falls 49.0→40.3%, the top 0.001% barely moves (33.4→33.1%). The peak-to-top gap shrinks 15.6→7.3-7.5 pts, and the profile is progressive up to P95-P98 (≈40.5%). With partial contributivity (half of c treated as tax), P0-10 (34.6%) is about equal to the top 0.001% (33.3%). The bias runs one way: it raises low and middle rates relative to the top.

Evidence. DataG1 (columns B and H) and the authors' micro-files (adjudicator reproduce_profiles.py; table_checks.py). For full-time workers on the secondary concept, re-ranked: P0-10 27.4%, P10-20 31.5, P20-30 34.1, P50-60 37.6, P90-95 40.2, P97-98 40.5 (peak), P99-99.9 39.0, P99.9-99.99 36.5, P99.99-99.999 34.2, P99.999-100 33.2. Book values: 40.5 / 43.5 / 45.5 / 48.1 / 49.0 / 48.1 / 43.7 / 38.1 / 34.8 / 33.4. Peak-to-top decline: 15.6 pts in the book, 7.3-7.5 on the secondary concept. Counting only half the contributions as taxes: P0-10 34.6% vs top 33.3%. Authors' STI7 (all adults): bottom 50% 23.7%, P98-99 39.4%, top 33.5%. IPP successor report (Bozio et al. 2012, TAXIPP 0.1, 2010, all adults, secondary income): about 38-39% from P90 to P99.9, then about 34% at P99.9-100 (Fig. 2.3, p. 30); the text gives 37% to 38.5% from P90 to P99.9 and 32.5% for the top 0.001% (p. 64). CPO (May 2011) excludes these contributions as 'un transfert intertemporel d'une personne à elle-même' (PDF p. 148). It notes 'si les prélèvements sont pris en compte, les dépenses ne le sont pas' (PDF p. 345). It says that without them the decline 'interviendrait alors plus loin dans la distribution' (PDF p. 267, fn 265, citing this model).

Why this severity. Under the authors' own alternative concept, C2 reverses: the full-time bottom decile pays 27.4% against 33.2% for the top 0.001%. C1's peak-to-top decline roughly halves (15.6 to 7.5 pts) and starts inside the top 1%, not the top 5%. A decline at the very top survives, so the core diagnosis is weakened rather than eliminated.

The authors' response. Book p. 53 and the companion-site page 'Réformer les cotisations sociales' (Wayback capture of 6 Nov 2011, linked from the FAQ entry 'Cotisations chômage et retraite') argue that including these contributions is legitimate. Their reasons: the pension and unemployment systems are 'loin d'être purement contributifs' (life-expectancy gaps, family credits, transfers between generations and occupations). The authors' 2012 OFCE reply (pp. 145-152) does not address the point. Sterdyniak overstates it: 'la prétendue régressivité disparaît' is wrong, and his 33% for P98-99 should be 35.9% by his own numerator-only method; the consistent method leaves a ≈7-pt decline within the top 1%.

2majorlikely · previously known · speculative assumption

About 45% of the measured income of the richest people in the model is company income, and the tax on it, that never shows up on their own tax returns. The authors assign it in proportion to the dividends people declare, and spread the corporate tax over all savers. Plausible alternatives make the drop in tax rates at the top a third to two-thirds smaller, and it almost disappears once pension contributions are also treated as deferred pay. One such alternative is the choice later made by the authors' own successors, who count part of this income as business owners' self-employment income. The very richest still pay lower rates in most versions, but how much lower is uncertain.

Technical detail and evidence

Why it matters, in detail. Top incomes in the model are built from national-accounts aggregates allocated by keys proportional to reported items. For full-time P99.999-100 about 46% of primary income (dividend gap, retained profits, corporate tax) bears no personal income tax by construction, so the low top rates are largely a ratio of observed taxes to an imputed denominator. Reclassifying D4222 as mixed income raises the top-0.001% rate by about 4-5 pts and cuts the peak-to-top decline from 15.6 to ≈10-11 pts (secondary concept: 7.3 to ≈1.3-2.4). Shareholder-only corporate-tax incidence adds ≈5.7 pts. The two together leave a ≈3-4 pt decline. The successor model (TAXIPP 0.3, 2010) shows the same pattern: a nearly flat secondary-income top decile and a progressive top under declared dividends or shareholder incidence. The direction of the true bias is not certain. Offshore wealth is excluded, and retained earnings in holding companies (IPP 2023) concentrate even more at the extreme top. So what is undermined is the size and onset of the decline, not its existence at the extreme top.

Evidence. Adjudicator capital_imputation_sensitivity.py: partial-equilibrium re-imputation on the authors' micro-files, full-time 18-65, re-ranked; indirect taxes either fixed or moved at each person's average rate; wealth taxes not re-imputed. Top-0.001% rate, against 33.4% in the book: (i) D422 treated as mixed income and allocated by declared self-employment income: 37.4-38.2%, with P99.9-99.99 at 41.9-42.4%; peak-to-top decline 10.0-11.0 pts instead of 15.6, and 1.3-2.4 pts instead of 7.3 on the secondary concept. (ii) Half the dividend gap moved to interest and life-insurance holders: 37.0-37.8% (A's 37.8% reproduced). (iii) Corporate tax on shareholders only: 39.1%. (iv) (i)+(iii): 44.3-45.2%, decline 2.9-3.9 pts; on the secondary concept the top rate (44.0%) exceeds P90-95 (39.4%). (v) Declared dividends only: 41.2-42.7%. Source cells: raw INSEE T_3401 rows 37-39 (D42 70.674, D421 43.211, D4222 27.463 for 2009); CN15 K12-K13; INSEE T_7401 (base 2020) D422 empty for all years. External: Bozio, Guillot & Laffeter (2015), Économie et Statistique 481-482, TAXIPP 0.3 for 2010. They treat D422 as mixed income (p. 47). Fig. V (secondary income, 18+; approximate chart readings): ≈37% at P90-91, ≈40.5% at P99.9-99.99, ≈39% at P99.999-100. Fig. VII: P99.9-100 ≈44.5% with declared dividends only vs ≈40% with national-accounts calibration (primary); the secondary profile with declared dividends is progressive to the top (≈42%). Fig. VI: corporate tax on dividend holders plus property taxes on owners gives ≈47% at P99.9-100 vs ≈38.5% at baseline. The authors note the model's 'manque de robustesse' to corporate-tax and property-tax incidence (p. 48). Bach et al. (2023, Note IPP 92), linked firm-owner data for 2016, direct taxes: progressive up to the top 0.1% (46%), regressive above (26% for the top 0.0002%).

Why this severity. Defensible alternative imputations, including the convention later adopted by the authors' successors, cut the book's top-end decline by about a third to two-thirds, and nearly eliminate it on the secondary concept. The magnitude and onset of C1's regressivity are therefore materially uncertain. A fall at the extreme top survives most single variants and is supported by later linked data, so the finding is not eliminated.

The authors' response. Simulator annex C.2.2 (p. 24): proportional imputation of retained profits 'aboutit presque certainement à sous-estimer fortement les très hauts revenus du capital' (simulated top fortunes ≈1 bn EUR vs 5-15 bn). Book p. 75 fn 1 concedes that some NA dividends may be received through holding companies or be taxable as self-employment income; p. 75 argues that NA probably underestimate dividends because of holdings in Switzerland. No published response on D422 or the incidence sensitivity was found.

3moderateconfirmed · previously known · overclaiming

The book says at least 90-95% of people in every income group below the top would pay less tax under its reform. The authors' own later tables show that about three people in four would pay less overall, and only a little over half of the lowest-paid full-time workers. This is mainly because the reform scraps an existing tax credit for low earners. The average gains the book reports are correct, but many individuals would lose, usually small amounts.

Technical detail and evidence

Why it matters, in detail. The text conflates centile-average tax changes (SIR2: lower mean tax for 97 centiles) with individual incidence. Heterogeneity in PPE receipt, exempt capital income, niches and spousal income gaps produces about 25% net losers overall and 44% among full-time P0-10. The '15 bn' transfer is understated: summing disjoint SIR2 group changes gives +20.4/−21.2 bn, a lower bound on gross individual transfers.

Evidence. gagnants.xls (first archived 5 May 2012). Full-time workers: 74.7% strict winners (85.6% counting losses below 1% of income or 10 EUR/month; 72.2% including the child reform). By group: P0-10 56%, P10-20 82%, P20-30 89%, P30-P90 76-83%, P90-95 66%, P95-96 42%, P96-97 23%. All adults: 73.4% winners; P10-20 62%, P20-30 50%. FAQ 'Gagnants et perdants' (added between the 29 Jan and 17 Jul 2011 Wayback captures): '75% de gagnants nets', '80%-90% ... du 10e au 90e percentile', 'à peine 60 % dans le 1er décile'. SIR2 group-level tax increases sum to 20.4 bn, against the book's '≈15 milliards' transfer. Among full-time workers (SIR3), average increases start at P95-96, not with the top 3%.

Why this severity. The reform's revenue neutrality and its average tax cuts below ≈7,000 EUR/month hold. But the individual-level claim is overstated by roughly 5-50 points in most groups, and most in the lowest-paid full-time decile. This is a material misstatement about the reform's distributional effect, not about the regressivity diagnosis.

The authors' response. The FAQ (Wayback capture of 17 Jul 2011) reports 75% net winners and explains the PPE mechanism. It says most losses are small and that lowering the 2% bottom rate to 1% or 0% (cost 1-3 bn) would compensate. The book text was not corrected.

4moderateconfirmed · newly identified here · internal inconsistency

The book claims its reform would create about 2% more jobs and leave the state 1-2 billion euros better off once people react to it. The authors' own technical appendix says about 1% more jobs and under 1 billion euros, and that small gain turns into a loss under slightly less favourable assumptions about how people respond. The behavioural case for the reform is therefore much weaker than the book says, although the reform remains roughly self-financing either way.

Technical detail and evidence

Why it matters, in detail. Each headline number on p. 100 is rounded in the reform's favour relative to the authors' own output. The net fiscal cushion (+0.93 bn on a 147 bn tax) is within the range of modelling error. The intensive top elasticity (0.15 on labour income only) is low relative to the taxable-income elasticities the book cites (Saez-Slemrod-Giertz best estimates 0.12-0.40). Excluding capital-income responses biases the net effect upward. If the published runs used ir1, the bias would favour the reform somewhat (higher new marginal and participation rates for homeowners).

Evidence. SimulationsMTR_PTR.xls: intensive response −2.976 bn in total, of which −5.017 bn for the top 1%. Participation: +303,314 workers (1.08% of 28.17 m; women 82.6%), +3.906 bn. Net +0.930 bn. Using the annex's linearity, the net turns negative at a top-1% elasticity of 0.178, against the 0.15 assumed and called 'une estimation haute'. With participation elasticities scaled by 0.75 it is −0.05 bn (linear; −0.11 bn nonlinear per auditor B). The posted do_mtr_ptr.do (line 153) computes new marginal rates from ir1, the imputed-rent variant, not the version-0 reform, so the lineage of the published MTR/PTR tables is unclear.

Why this severity. This is a secondary claim. The static budget neutrality and the regressivity diagnosis are untouched. But the claimed employment gain is overstated about 1.9-fold, and the 'robust' +1-2 bn net gain is in fact ≈+0.9 bn and changes sign under modest elasticity changes.

The authors' response. The authors' 2012 OFCE reply (pp. 146-148) defends the responsiveness of second earners. No reconciliation of the p. 100 figures was found.

5moderateconfirmed · overclaiming

The book presents its finding as proven beyond appeal and says every assumption it made understates the problem. In fact several of its most important choices make the rich look less taxed, and the book never shows how much the result depends on them. The numbers are model estimates built from 2006 data, not a measurement.

Technical detail and evidence

Why it matters, in detail. The statement that all assumptions minimise regressivity is false for the conventions that matter most: the contributory-contribution wedge and the capital-income keys together move the peak-to-top gap from 15.6 pts to ≈1-4 pts, or reverse it, in combined variants. Neither the book nor the annex presents sensitivity bands. The profile also mixes 2006 joint distributions with 2010 income totals and legislation, and the top 0.001% cell is an extrapolated synthetic tail.

Evidence. Book pp. 48-49, 54, back cover. Simulator annex B.2.2. Adjudicator logs: each alternative listed in F1 and F2 lowers measured top-end regressivity by 2-8 pts at the top or halves the middle-to-top gap. Observation count from reproduce_profiles.py: 174 records at P99.999-100. Some choices are conservative and run the other way: uniform effective corporate-tax rate (p. 54 fn 1), DMTG on the top 5% of wealth, ISF calibrated on tabulations, proportional retained-profit imputation (annex C.2.2), exclusion of offshore wealth, and inclusion of realised gains (removing them with re-ranking lowers the top-0.001% rate to 32.4-32.7%).

Why this severity. This is not itself a change in level. But the book tells readers that its central result is beyond dispute and conservatively measured, when several of its most consequential conventions push the other way and alter the result's magnitude (F1, F2).

The authors' response. The authors published all code, parameters and synthetic data, invited users to change assumptions (pp. 15-16), and listed simplifications in the technical annex. Some assumptions are genuinely conservative (see evidence).

6minorconfirmed · previously known · selective presentation

The book's comparison covers only full-time workers aged 18-65. It says other population choices give similar results, which is true for the drop at the very top but not for the claim that people with modest incomes pay more than the rich.

Technical detail and evidence

Why it matters, in detail. The 'modest pay more' comparison is specific to full-time workers. In the all-adult fields the bottom 40% pays 26-30% (primary) or about 22-26% (secondary), below the top 0.001%; only the top-1% decline generalises.

Evidence. STI3 (all adults, primary income): P0-10 29.3%, P10-20 27.5%, P20-30 25.9%, P30-40 29.8%, against 33.8% for P99.999-100. STI7 (all adults, secondary income): bottom 50% 23.7%. Both are reproduced by the adjudicator within 0.2 pts. Book pp. 48-49, 55; graph notes pp. 50-51 point to the all-adult figures in the online annex.

Why this severity. The restriction is disclosed and defended as a proxy for permanent income. The all-adult primary-income rates of people with almost no primary income are not meaningful, and the all-adult secondary result largely reflects F1. The flaw is an inaccurate robustness statement, not a hidden change in the main result.

The authors' response. Book p. 49 fn 1 and the annex justify the field as the most suitable for measuring permanent-income inequality and progressivity. The all-adult tables are published online (STI3-STI8).

7minorconfirmed · newly identified here · arithmetic or transcription

In the book's own chart, people earning 1,700 euros a month pay about 43.5% and people earning 63,000 euros pay about 38%. The book's sentence says 45% versus 35%, doubling the difference between them.

Technical detail and evidence

Why it matters, in detail. The sentence pairs the income label of one fractile (P99.9-99.99) with the rate of the next (P99.99-99.999) and rounds the P10-20 rate up by 1.5 pts. This inflates the illustrated modest-to-rich gap from 5.4 to 10 pts.

Evidence. Rendered graph on p. 50 (label arrows); DataG1 rates; SIR3 group mean monthly gross (CSG-base) income: P10-20 1,654 EUR, P80-90 4,229, P97-98 9,371, P99.9-99.99 62,845, P99.99-99.999 253,636, P99.999-100 1,091,667 (adjudicator table_checks.py).

Why this severity. This is a rhetorical sentence. The graph itself is correct, and groups further up do pay about 33-35%. But the sentence doubles the gap between the two income levels it names (5.4 pts in the data, 10 in the text), always in the direction of the thesis.

8minorconfirmed · newly identified here · internal inconsistency

The book counts the income-tax reductions parents get for children as if parents had paid that tax. This slightly inflates middle-income tax rates and makes the simulated total of taxes about 14 billion euros higher than the official figure the book says it matches exactly.

Technical detail and evidence

Why it matters, in detail. Adding back the family-quotient reduction raises the measured rates of parents in the middle and upper-middle of the distribution, where the reductions are concentrated below the cap. It barely moves the top 0.001%, slightly widening the measured gap. The simulated aggregate exceeds macro levies by about 14 bn (0.8% of national income).

Evidence. SIR1: IRPP before QF 68.35 bn (= 52.0 net + 11.89 QF + 4.46 PPE added back), QF to deduct 11.89 bn. STI1: levies of 755.2 bn on primary income + 66.0 on replacement income + 9.2 on pure transfers = 830.4 bn. Auditor A estimates the effect, using the earlier-version IRPP archive, at +0.6 to +1.1 pts on P0-P99 rates and ≈0 at P99.99+.

Why this severity. The convention is internally coherent with chapter 3 but undisclosed in chapter 1. It widens the measured middle-to-top gap by about one point.

9minorconfirmed · speculative assumption

The model assumes the richest save more than half their income, more than the authors' own description says, which lowers the sales taxes attributed to them. The book's text also exaggerates how much more sales tax the poor pay than the rich, compared with its own figures.

Technical detail and evidence

Why it matters, in detail. The doubling pushes realised top saving rates above 50%, beyond the annex's own description, mechanically lowering consumption-tax incidence at the top. The text then rounds the bottom and top shares in the direction of the thesis (13.5→'près de 15', 7.2→'à peine plus de 5').

Evidence. do_autres_impots.do phases 5-1/6-1; simulator annex C.8.1 (p. 46-47). Adjudicator saving_rates.py: realised saving rates are 26.9% (P95-96), 38.1% (P98-99), 45.6% (P99-99.9), 52.0% (P99.9-99.99) and 54.3% (P99.999-100). Indirect taxes are 13.5% of primary income at P0-10 and 7.2% at P99.999-100 (DataG1: 13.46% and 7.20%). Auditor A reports that removing the doubling (aggregate saving restored) raises the top-0.001% rate by ≈1.7 pts, and lifetime incidence of the consumption part by ≈2.2 pts (not re-run here).

Why this severity. The effect is 1.7-2.2 pts at the top, and the qualitative result is unchanged.

10minorconfirmed · newly identified here · data error

The chart of child support by income inflates the tax break the richest parents get per child beyond what the law allows, and adds an unexplained 20 euros to the poorest group's bar. The main message, that top earners get much more per child through the tax system, still holds.

Technical detail and evidence

Why it matters, in detail. The QF per child at the top is overstated by at least ≈11-50 EUR/month (≥4% against the generous bound, ≥21% against the minors-only bound). The corrected bottom-decile total (124.6 EUR) contradicts the description of a profile 'remarquablement stable, autour de 175 euros ... du 1er au 9e décile'.

Evidence. Adjudicator qf_cap_ratio_of_sums.py (2010 caps: 2,336 EUR per half-part, 4,040 EUR for a single parent's first child). The legal ceiling per minor child at P99-100 is 240.6 EUR/month (ratio of sums, as the graph aggregates), or 280.1 EUR even crediting adult dependants' half-parts to minors. Auditor A's 226.6 EUR is a mean of per-family ratios and understates the ceiling. table_checks.py: DataG3-1 C5 = 141.56 vs SimulationsEnfants.xls R8 = 121.56; every other row matches exactly. Auditor B found the cell formula adds the integer 20.

Why this severity. The top QF bar is overstated by at least 4-21% and the bottom bar by 16%. The steep top-decile rise, which is the point of the graph, survives.

11minorconfirmed · newly identified here · internal inconsistency

Two background tables on wealth and income shares do not match the model results the book cites as their source. The differences do not help the book's argument and do not affect its conclusions.

Technical detail and evidence

Why it matters, in detail. About 6 pts of wealth share are shifted from the top 1% to P90-99 relative to the model, and the income top-1% share is 1.2 pts higher than the stated method gives. The source of the printed synthesis is undocumented.

Evidence. Wealth (adjudicator table_checks.py): STI9/10/11 give top-1% shares of 25.5/36.4/28.6% and P90-99 shares of 33.6/33.1/32.8%. The book prints 24% (mean 4.37 m EUR) and 38%. No nonnegative average of the three variants gives 24%. Income (auditors A and B): STI5 shares give a top 1% of 9.8% vs 11% printed, and 28.3/41.3/30.4 vs 27/42/31.

Why this severity. These are descriptive background tables. The errors do not favour the book's thesis, and the printed values happen to be close to later WID estimates.

12minorconfirmed · newly identified here · documentation gap

The simple tax schedule printed in the book is not exactly the one the authors simulated, and the stated top marginal rates are averages; the true rate on an extra euro reaches about 77% in some income ranges. This does not change how much tax the reform raises.

Technical detail and evidence

Why it matters, in detail. A linear effective-rate schedule implies a saw-tooth marginal rate, with drops at each knot (76.7%→56.7% at 40 k; 76.7%→60% at 100 k EUR/month). The annex's single interval rates misdescribe local incentives.

Evidence. BAR-V0. At 25,000 EUR/month the simulated rate is 40% against 37.5% on the printed straight line. Local marginal rates T'(y) = t(y) + y·t'(y) run 35→65% on 10-25 k EUR, 56.7→76.7% on 25-40 k and 40-100 k, and 60% above 100 k EUR/month. Auditor A's statement that the simulated marginal rate just below 40,000 EUR is 66.7% is incorrect: that is the interval average, and the local rate is 76.7%. Sterdyniak's ≈83% correctly reflects the printed six-point schedule.

Why this severity. Tax liabilities are continuous and revenue is unaffected. The issue is that readers cannot reproduce the schedule or its incentive structure from the book.

The authors' response. The chapter annex (pp. 7-8) states that the printed schedule is illustrative, points to the complete version in the Barèmes file, and justifies the 50%/66.7% pattern.

13minorconfirmed · newly identified here · documentation gap

The authors posted an unusually complete set of files, and the main tax-rate chart can be rebuilt exactly from them. But some files were later replaced by other versions, so a few of the book's reform numbers cannot be recomputed from scratch.

Technical detail and evidence

Why it matters, in detail. Temporary and source files were overwritten by later runs: the site's directory dates are a 2022 server migration, and Wayback captures span 2011-2025. Reconciliation must therefore rely on the archived result tables rather than the micro-files for the reform (SIR, gagnants) and behavioural results.

Evidence. Book-consistent: Wayback captures of indiv_demo, indiv_aut_impots and indiv_revcap reproduce DataG1/STI5. The archived TableSTI5_2010.dta matches them to about 0.001 pt, while the book's Excel differs from that .dta by up to 0.15 pt at the top. Inconsistent: (i) the indiv_cotsoc and indiv_irpp archives (dated 12 Jan 2011) are an earlier run (auditor A: cs+ts 367 vs 382.9 bn). (ii) The current-site indiv_nouvel_ir_2010.dta gives a new-IR total of 165.6 bn, old IR 181.9 bn and base 1,261.5 bn, against SIR1's 158.7/159.5/1,223.1 bn (B; verified here). (iii) In temp files, impota_prim does not equal the sum of its components (B). (iv) do_mtr_ptr.do uses the ir1 variant (B; see F4).

Why this severity. The core profile reproduces exactly from posted files, and the reform tables are internally consistent. Only end-to-end regeneration of secondary aggregates is impossible.

The authors' response. The site's 2 April 2012 notice says data and legislation remain those for 2010 and directs later work to IPP. The behavioural program warns that it alters source files (annex C.10).

14minorconfirmed · newly identified here · arithmetic or transcription

A handful of illustrative figures in the text are slightly off compared with the authors' own data. None of them changes the book's conclusions.

Technical detail and evidence

Why it matters, in detail. These are small slips, but their direction is mostly favourable to the thesis (e.g. the dividends-vs-unemployment-benefits contrast uses barème-only dividends and an outdated year).

Evidence. (a) p. 72: declared dividends '13 à 14 milliards' against unemployment benefits '28 à 29 milliards ... légèrement plus que les revenus fonciers'. Parametres.txt gives barème dividends of 12.5-16.6 bn in 2005-10 (17.5-19.3 bn including the flat-rate levy from 2008), unemployment benefits of 25.7-29.9 bn, and rental income of 28.4-31.3 bn, above unemployment benefits from 2006 on. The statement holds only for 2005. (b) p. 59: '41 % ... le plus faible appliqué depuis 1935' conflicts with the 40% top rate of 2006-09 stated on the same page. (c) p. 49: 'ne dépassent guère les 35 %' for the top 0.1%; the authors' figure is 36.8% (DataG2-3). (d) p. 45: taxe foncière '≈15 %' of property income vs 12.5% (auditor A). (e) p. 21: 'près de six années' of income = 5.5. (f) p. 34: the US top-1% share '9 % à 24 % ... dans le revenu national' is a share of fiscal income.

Why this severity. These are rhetorical illustrations and secondary numbers. None affects the conclusions.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

Within issue I2: removing realised capital gains, to avoid double counting with retained profits, raises the top-0.001% rate by ≈1.2 pts. — This holds only if fractile membership is kept fixed (adjudicator check: 34.6%). When the profile is re-ranked on the new income concept, as an alternative profile requires, the top-0.001% rate falls to 32.4-32.7% (capital_gains_sensitivity.py), because gains are taxed at 30.1% plus indirect taxes. Including gains is therefore slightly conservative for the book's thesis, not a source of regressivity.
I11: the taxes-only scope and individual ranking are a defect that colours the claims about low-wage workers. — Not retained as an issue. The book is explicitly about levies (pp. 38-41), and the authors' OFCE reply (p. 149) validly makes the scope argument; auditor B reached the same conclusion. Auditor A's own computation shows the full-time bottom-decile rate net of own pure cash transfers is still ≈35.7%, so the 'heavily taxed' characterisation survives. The household and consumption-unit objection (Sterdyniak) is normative, and its effect was not quantified.
I10 detail: the simulated schedule's marginal rate just below 40,000 EUR/month is 66.7%, so Sterdyniak's ≈83% reflects only the printed schedule. — 66.7% is the interval-average marginal rate. The local marginal rate of the simulated schedule just below 40,000 EUR is 76.7%, and Sterdyniak's 83.3% is the correct local rate of the printed six-point schedule. The underlying issue (F12) is retained.
I8 magnitude: the legal ceiling is 226.6 EUR per minor child at P99-100, so the top QF bar is overstated by ≥28% (≥64 EUR/month). — Magnitude corrected, issue retained (F10). The graph aggregates as a ratio of sums, for which the ceiling is 240.6 EUR (minors only) or 280.1 EUR (crediting adult dependants). The overstatement is therefore ≥4-21% (≥11-50 EUR/month), not ≥28%.
I1 and I2 rated major: the winner-share overstatement and the behavioural headline undermine core claims. — Both are confirmed (F3, F4) but downgraded to moderate. They concern the reform's secondary claims (individual incidence, behavioural add-on), while the regressivity diagnosis, revenue neutrality and average gains by income level are unaffected.
I4 (part): the stored impota_prim in the all-tax temp file not equalling the sum of its components (15.6 bn signed), and the new-IR micro-file not reproducing SIR1, indicate integrity problems in the results. — These files are later or different runs. The book-consistent Wayback micro-files reproduce the published profile and the archived TableSTI5 .dta almost exactly. The point is retained only as a documentation and replication-package gap (F13, minor), not as evidence about the published numbers.
I3 as a separate, unquantified moderate issue (non-dividend distributed income in the dividend benchmark). — Not rejected on substance: merged into F2 and quantified. Reclassifying the 27.7 bn as mixed income raises the top-0.001% rate by about 4-5 pts and removes about a third of the primary-income decline (most of it on the secondary concept). It is part of a major issue, though its economic interpretation remains uncertain.

Audit notes

What reproduced. Reproduced: the headline profile (graph p. 50; DataG1/STI5) from the authors' 2010 synthetic micro-files, within 0.023 pt up to P98-99 and 0.154 pt at the top, with the same population (20.222 m) and mean (46.87%). Also reproduced: the all-adult tables STI3/STI7 (≤0.2 pt), the published secondary-income construction, the behavioural-response totals (−2.976/+3.906/+0.930 bn; 303,314 workers), the winner shares (74.7% full-time workers), the revenue tables (SIR1/SIR2), the wealth-share variants (STI9-11), and the child-transfer source table. The legal QF ceilings were recomputed. Not reproduced: the book's p. 100 behavioural figures and the p. 93 winner shares (both contradicted by the authors' own outputs); the wealth table's 24% top-1% share; the top QF bar (above the legal ceiling); the book's '1,700 EUR 45% vs 63,000 EUR 35%' sentence (the graph gives 43.5% vs 38.1%). The reform aggregates cannot be regenerated end to end from the posted micro-files (later runs). The Stata pipeline was not re-run; sensitivity analyses are partial-equilibrium re-imputations on the authors' output files.

Limits of this audit. The authors' Stata pipeline was not re-run; all alternative profiles are partial-equilibrium re-imputations on their output micro-files, with wealth-based taxes (ISF, DMTG, allocated on capitalised income) held fixed and indirect taxes adjusted only approximately. The synthetic micro-files cannot be validated against the confidential sources (ERFS 2006, tax and estate files), so the realism of the synthetic top tail (174 records in the top 0.001%) is untested. The economic nature of D4222 (self-employment income vs corporate saving) is uncertain. Under a corporate-saving reading the model's allocation key would be unchanged, so the D422 scenario is one defensible alternative, not a demonstrated correction. Figures from Bozio-Guillot-Laffeter (2015) and IPP (2012) are read from charts and are approximate; those models differ from the book's in other respects too. Later national-accounts vintages (base 2020: household dividends 41.3 bn, interest 20.5 bn for 2010) are used only to document reclassification, not to re-run the model. Household or consumption-unit re-ranking, lifetime incidence and re-estimation of elasticities were not performed. The Travail, genre et sociétés debate on individualisation was not read. Normative questions (individual vs family taxation, whether contributions are taxes) are flagged as convention-dependent rather than adjudicated.