Camille Landais, Thomas Piketty, Emmanuel Saez · Revue de l'OFCE n° 122 (2012/3): 145-151

Commentaire de l'article d'Henri Sterdyniak « Lecture critique de l'ouvrage pour une révolution fiscale »

13 problems found: 5 moderate · 8 minor

What the paper claims

Sterdyniak falsely attributes to the authors the aim of 'supprimer la prise en compte des enfants dans le calcul de l'impôt sur le revenu'; the book keeps child-related tax reductions (as a refundable credit of the same total amount) and individualisation does not question horizontal solidarity between families with and without children.
Familialisation is a marginal feature of French income taxation, so individualising the income tax is 'une réforme de continuité', not a 'big bang fiscal'.
Joint taxation heavily taxes second earners, who are the most tax-responsive in their participation decisions, while very high earners are much less so (Blau & Kahn 2007; Chetty et al. 2011); the unitary model of the household is empirically rejected (Attanasio & Lechene 2002; Blundell et al. 2007; Lundberg, Pollak & Wales 1997); individual taxation is neutral between marriage and cohabitation whereas the quotient conjugal creates marriage penalties and bonuses (Amar & Guérin 2007).
'Nos simulations prouvent' that the fear that single-earner (mono-actif) couples will be the big losers of individualisation is unfounded: few such households have incomes high enough for the income tax to matter, quotient-conjugal gains for very unequal couples are relatively small up to very high incomes, and the average quotient-conjugal gain is limited.
The book deliberately covers taxes only; Sterdyniak's addition of a few transfers cannot characterise overall redistribution, because other public spending (education, health) also redistributes; e.g. CERC (2003) found 'une certaine régressivité' of public education spending.
A microsimulation of the whole tax system 'reflète la réalité de la distribution des revenus et prélèvements', whereas Sterdyniak's case types are ad hoc, partial and misleading; his Table 6 is 'méthodologiquement faux' because it uses statutory rather than effective rates (effective IS 'bien inferieur à 33 %').
Book pp. 130-131 show that the authors 'évidemment' see IS as a précompte to be credited once profits are distributed; effective IS rates that differ from the statutory rate and vary across securities rule out a single-rate avoir fiscal until a European IS exists; Sterdyniak caricatures the proposal.

Every problem found

1moderateconfirmed · previously known · external inconsistency

The reply says the tax advantage married one-earner couples get today is small except for the very rich. In fact it is worth about 5% of earnings for a modest earner (about EUR 100 a month at EUR 2,000 gross) and 6-8% higher up. What keeps most of these couples from losing much under the reform is its general tax cut, not a small advantage.

Technical detail and evidence

Why it matters, in detail. The QC gain for a single earner equals T(R,1 part) - 2T(R/2,1 part) (plus the EUR 83 single-earner PPE supplement at low wages). On the 2010-income schedule it is 5-8% of gross earnings from EUR 2,000/month upward, more than the reform's average middle-class gain of 2.3-2.8% of income. Hence single-earner couples gain little or lose under the reform even though the schedule cuts average rates in the middle.

Evidence. scripts/casetypes.py (2010-income parameters of the authors' SimulCasTypes.xls; validated to the cent): QC+PPE-couple gain vs a single person EUR 1,255/yr (5.2% of gross) at EUR 2,000/month, 1,417 (3.9%) at 3,000, 2,853 (5.9%) at 4,000, 4,288 (7.1%) at 5,000, 5,566 (7.7%) at 6,000, 7,642 (6.4%) at 10,000. Sterdyniak (2012) p. 125: EUR 336/month for an earner on EUR 4,000/month (616 vs 280); his figures reproduce exactly from the 2010-income schedule when EUR 4,000 is read as the monthly taxable wage. Amar & Guérin (2007, Économie et Statistique 401, Table 3, 2004 law): single-earner couples' income tax net of PPE EUR 2,260 jointly vs 3,010 separately; 75% of them gain (EUR 1,010 on average among gainers). Book p. 66 ('une machine à subventionner les couples inégaux'); book pp. 91-92 call average reform gains of 2.3-2.8% of income 'loin d'être négligeable'. gagnants.xls: current-law advantage of unequal over equal couples 1.8-3.2% of income at P50-P99 (all married adults). Later evidence: Allègre, Périvier & Pucci (2021, Économie et Statistique 526-527) put the gain from abolishing the QC (individualisation without optimisation, 2016 law) at about EUR 10 bn.

Why this severity. A confirmed factual error in one of the two reasons offered for C4. It does not by itself change the net reform effect, but it misstates the size of the advantage individualisation removes from single-earner couples, which is the crux of the critic's objection.

2moderateconfirmed · previously known · misrepresentation of source

The critic calculated that, under the authors' reform, interest on savings could be taxed at more than 100% of its real value. The authors called his table methodologically wrong because it used the official company-tax rate instead of the lower rate firms actually pay. That objection is fair for dividends, but it has nothing to do with the interest result, which comes from taxing the part of interest that merely makes up for inflation. The reply never answers that point.

Technical detail and evidence

Why it matters, in detail. Effective-rate correction moves only the dividend column (by 6.6-12.3 points at t = 60/25%). The interest result is an inflation wedge, t x (nominal/real) = 1.8t, that a nominal income base with high average rates produces regardless of IS. A statutory marginal wedge answers a conditional incentive question; an effective average rate answers a distributional-burden question. The reply treats the first as an error.

Evidence. Reply p. 150; Sterdyniak (2012) Table 6 p. 117, text pp. 116-118 ('ils refusent de tenir compte de la dépréciation des intérêts induits par l'inflation'), Annex II pp. 141-142. Recomputed (scripts/reply_checks.py): reform at t = 25/50/60%: salaries 42.1/65.1/74.3, rents 32.5/55.0/64.0, interest 45/90/108 (= 1.8t with 4.5% nominal, 2.5% real; 60% at t=60% with zero inflation), dividends 50.8/67.2/73.8 with 34.43% IS. With CPO (2009, p. 158) average implicit IS of 18%: reform dividends 38.5/59.0/67.2 and current-law dividends 41.4-43.7% (vs 53.2-55.0). Critic's own slips: interest at t=2% and 10% should be 3.6% and 18% (annex prints 4.5 and 10.2). Auditor B: the combined burden is IS + (1-IS)t, not a simple addition of rates, and reproduces exactly.

Why this severity. Partly undermines C6's flagship example. The rebuttal is valid for one of four columns and leaves the table's main result (interest taxed above 100% of its real return) untouched, so 'méthodologiquement faux' overstates what was shown.

3moderateconfirmed · previously known · selective presentation

The critic's main objection was that the book counts pension and unemployment contributions as ordinary taxes, although they buy future benefits. The authors' own tables show this choice matters a lot: counted the other way, the tax burden rises steadily with income almost up to the very top. The reply only attacks the critic's illustrative examples and never answers the objection itself. The very richest still pay a somewhat lower rate under either choice.

Technical detail and evidence

Why it matters, in detail. Moving from primary to secondary income (contributory contributions treated as deferred wages rather than taxes) on the same all-adult population cuts the P95-99 -> top-0.001% decline from 13.8 to 5.7 points. It also turns the middle of the distribution from a high plateau (40-48%) into a profile rising from 21.5% to 39.1%. The authors' later DINA framework adopts a pre-tax concept measured after the pension system. The reply's 'reflète la réalité' therefore conceals a contestable convention choice rather than settling it.

Evidence. SimulationsTousImpots.xls (authors' results), all adults 18+: STI3 (primary income) P0-10 29.3, P50-60 40.4, P90-95 48.1, P95-99 47.6, P99.999-100 33.8%; STI7 (secondary income = primary - contributory contributions + contributory replacement income) 24.7, 29.8, 38.3, 39.1, 33.5%; decline P95-99 -> P99.999-100: 13.8 vs 5.7 points; STI7 rises from 21.5% (P20-30) to 39.1% (P95-99). STI5 (book graph p. 50; full-time 18-65; primary): 40.5/48.1/48.0/33.4%, decline 14.7 points (logs/reply_checks.txt). Sterdyniak (2012) pp. 110-112 and Table 3; Annex I case types pp. 139-141. IPP (2012, Bozio et al., authors' institute, same simulator) p. 22 presents both conventions and p. 113 finds rates regressive within the top 1% 'même si on retire de l'analyse les cotisations sociales contributives'. The DINA Guidelines co-authored by Piketty and Saez (Alvaredo et al. 2016, WID.world WP 2016/1, p. 30) define benchmark pre-tax income 'after taking into account the operation of the pension system'. Fairness: Sterdyniak's Table 3 mixes STI5 (full-time workers) with STI7 (all adults), his 'P98-99 à 33 %' should be 36% on his own Table 1, and his 'la prétendue régressivité ... disparaît' is overstated.

Why this severity. Weakens C6's framing of the dispute as realistic microsimulation versus misleading case types: the disagreement is mostly about a convention on which the authors' own simulator gives materially different profiles. The qualitative decline at the very top survives under both conventions, so the narrower top-end claim is not overturned.

The authors' response. IPP (2012, p. 113), from the authors' institute, reports that top-end regressivity persists when contributory contributions are excluded. No response in the reply.

4moderateconfirmed · newly identified here · selective presentation

To suggest that counting public services might change the picture, the authors cite a 2003 government report as finding that education spending favours the better-off. The table they cite does show slightly more spending per child in richer families. But the same report concludes that, overall, school spending makes the system fairer, cutting the gap between rich and poor families by about a third.

Technical detail and evidence

Why it matters, in detail. Per-schoolable-child spending rising by 19% from D1 to D10 is compatible with strongly progressive incidence, since D10 income is about 6 times D1 income, families with children are concentrated at the bottom, and financing scales with income. Under the source's own conventions, education spending lowers the D10/D1 ratio by about a third, which reinforces rather than offsets the critic's point that the system is redistributive.

Evidence. Reply p. 149. CERC rapport n 3 (2003), Éducation et redistribution (cited URL; Wayback copy md5 7f1213e42ad54d92e584ac432c17d4ee): Table 6 p. 41 per enrolled child flat at EUR 5.4-5.6k, per schoolable child D1 4.3k (-7%) vs D10 5.1k (+10%); Table 5 p. 40 per family D1 11.6k (+32%) vs D10 8.3k (-5%), 'concerne davantage les classes de revenu initial faible'; Table 3 p. 58 D10/D1 ratio 6.1 -> 4.1 with education (34.0/5.6 = 6.07; 37.9/9.1 = 4.16); Table 4 p. 59 net transfer +EUR 2.9k per consumption unit for D1, -0.2k for D10; p. 59 conclusion quoted above. Mitigating: CERC p. 39 notes that higher-education cost differences by track are not captured, which probably understates spending on well-off families.

Why this severity. The reply's scope defence (the book covers taxes only; 4% of transfers cannot characterise total redistribution) survives, but its one illustration points the opposite way to its source's conclusion. That leaves the implication that other spending might offset measured redistribution unsupported.

5moderatelikely · previously known · overclaiming

The authors said their simulations prove that couples living on one income would not be the big losers if spouses were taxed separately. Their own published figures show that such couples gain from the reform much less often than other couples. For most of them, though, the change is small, and those with children and modest pay actually gain. The reassurance is only half true.

Technical detail and evidence

Why it matters, in detail. Removing the quotient conjugal shifts income-tax burden toward single-earner couples; in the LPS package this is offset in the middle by the new schedule's rate cut and, for parents, the flat child credit, but not at the top. Net effects (wage-only case types, book schedule): -EUR 26 to -424/yr between the SMIC and EUR 3,000/month, +191 to +457 at EUR 3,500-5,000, -488 at 6,000, -4,911 at 10,000, while singles at the same wage gain up to EUR 5,078. In the authors' microsimulation, 65.1% of married adults in unequal couples gain vs 89.1% in equal couples (full-time workers: 43.6% vs 88.7%). 'Prouvent' is unsupported: no output is cited, and the conclusion depends on reading 'grands perdants' as large absolute losses rather than as the group bearing the reform's losses.

Evidence. Reply p. 148. gagnants.xls (www.revolution-fiscale.fr/xls/gagnants.xls; md5 0e1ffc41ce8c9f2e3cf246725678a471, identical to the Wayback capture of 5 May 2012). Sheet 18+_couples_inegaux (unequal = one spouse earns at least 2/3 of couple income): winners 89.1% in equal vs 65.1% in unequal couples; losing more than 1% of income 5.5% vs 22.1%; average rate unequal 13.87% -> 15.58% vs equal 13.96% -> 13.24%; sheet note '90% des couples egaux gagnent ... seulement 65% des couples inegaux'. Sheet 18-65_workers_couples_inegaux: 88.7% vs 43.6%; P30-40/P40-50/P50-60: 92.1/89.7/92.4% vs 33.9/24.9/31.2%; 60.4% of unequal-couple workers at P40-50 lose more than 1% of income (logs/reply_checks.txt). FAQ (Wayback 17 Jul 2011 and 28 Feb 2012): lower winner share in unequal couples above P30, but also 'un couple totalement inégalitaire ... voit ses impôts augmenter ... uniquement si le salaire brut du mari dépasse environ 5 000€ par mois'. Adjudicator case types (scripts/casetypes.py; reproduces SimulCasTypes.xls sheet v0 to the cent), childless single-earner married couple, book schedule (p. 79): -198 (SMIC EUR 1,365), -272 (EUR 2,000), -424 (2,500), -26 (3,000), +191 (3,500), +344 (4,000), +457 (5,000), -22 (5,500), -488 (6,000), -1,548 (8,000), -4,911 (10,000), -12,171 (20,000) EUR/yr; single person at the same wage +9 to +5,078; break-even about EUR 5,480. With two children and the workbook's EUR 750/child credit: +505 to +1,462 EUR/yr from SMIC to EUR 5,000, +139 at 5,500, -484 at 6,000, -4,414 at 8,000. Under SimulCasTypes 'v0' (8% at EUR 2,200, labelled LPS 2011 version 0), the childless single-earner couple gains EUR 111-687 between EUR 2,000 and 5,000, matching the FAQ. Auditor A's household-level microdata estimate (approximate; file vintage differs from the published tables): 58-69% of 1.78 m working-age single-earner couples lose vs 13-19% of two-earner couples. Auditor B's narrow diagnostic: 96.7% of childless full-time earners with a zero-income spouse in the EUR 2,000-3,000 band pay more (mean +EUR 39/month, 1.6% of income).

Why this severity. Substantially weakens C4 without reversing it outright: in its natural relative reading (who bears the losses) the claim is contradicted by the authors' own tabulation, but in the narrower absolute sense the reply's reasons address (small losses for most single-earner couples below about EUR 5,500/month, gains for those with children) it largely holds. C1-C3 and C5-C7 are unaffected.

The authors' response. The authors' FAQ (online July 2011 and February 2012) already reported a 'nettement plus faible' winner share in unequal couples above P30 and stated that a fully single-earner couple loses only above about EUR 5,000/month gross; the reply notes that a dependent-spouse allowance could correct the effect but does not recommend it. No later response located.

6minorconfirmed · previously known · methodology

The authors' model spreads company tax evenly over all savers' financial income, including interest on savings and bonds. If instead it is charged mainly to shareholders, the very richest appear to pay up to about 6 points more tax. The book's finding that the very top pays less than the upper middle class then shrinks but does not disappear. The reply praises its model's realism without mentioning this.

Technical detail and evidence

Why it matters, in detail. With the reply's own emphasis on effective rates, the relevant uncertainty is the incidence of a disclosed EUR 35 bn aggregate. Plausible allocations move the top 0.001% rate by 1.4-5.7 points and shrink the P95-99 -> top decline by 9-36%. The reply's statement that per-security heterogeneity makes a single-rate avoir fiscal impossible is in tension with a model that itself assigns one rate to every security.

Evidence. do_revcap.do l. 126 (read as text); book p. 38 ('l'impôt sur les sociétés est attribué aux résidents français possédant des placements financiers, en proportion de leurs revenus financiers'); IPP (2012) section 1.3.4 p. 18 ('notre hypothèse d'incidence de l'IS ne prend pas du tout en compte les variations du taux effectif d'IS'). Adjudicator re-implementation of STI5 (scripts/is_incidence_check.py; reproduces published STI5 within 0.1 point): implicit rate 18.7% of the financial-income base. Same EUR 35 bn on dividends + retained profits only: top 0.001% 33.4 -> 39.1%, P99.9-99.99 38.3 -> 42.2%, decline P95-99 -> top 14.6 -> 9.3 points, but this variant implies a 36.6% IS rate on household equity income, above the 34.4% statutory rate (an upper bound). Including life-insurance income with equity: top 0.001% 34.8%, decline 13.4 points. Auditor A reported 38.3% and 10.1 points for the shareholder variant with a slightly different population filter.

Why this severity. The convention is disclosed and the qualitative top-end decline survives every variant. The variant producing a large shift is an upper bound, so this qualifies C6's 'reflète la réalité' rather than undermining it. Downgraded from auditor A's 'moderate'.

The authors' response. IPP (2012, p. 18), from the authors' institute, acknowledges that the IS incidence assumption ignores variation in effective IS rates.

7minorconfirmed · newly identified here · internal inconsistency

The authors accused their critic of caricaturing their view on whether shareholders should get a refund for company tax, and pointed to two pages of their book. Another passage of the same book says there is no good economic reason to give such a refund. The critic's description of what the reform does was accurate.

Technical detail and evidence

Why it matters, in detail. The book offers two rationales for no dividend credit: a neutrality argument against refunding one tax to one income category (presented as primary) and a conditional integration argument (feasible only with a uniform, non-evaded, European IS). The reply cites only the second and recasts the first as an implementation constraint, so its rebuttal of an accurate description is selective.

Evidence. Book pp. 83-84 and p. 84 fn 1; book pp. 130-131 (IS 'essentiellement un précompte', legitimate to re-credit if IS is not evaded and is the same for all firms; restore credits at European level); reply p. 150; Sterdyniak (2012) pp. 117-118.

Why this severity. An internal inconsistency and an unfounded 'caricature' charge. No number is affected, and the précompte principle is indeed stated on pp. 130-131.

8minorconfirmed · newly identified here · overclaiming

The shares the authors quote are correct. But part of the supposedly individual tax (the levy on pensions and unemployment benefits) actually depends on the household's situation. And the book itself presents taxing spouses separately as a historic break, not the small step the reply describes.

Technical detail and evidence

Why it matters, in detail. Household-dependent income taxation is about 4% of national income (IRPP 3.1% + replacement-income CSG 0.9%, plus the PPE), and the policy choice concerns a merged base of about 8.7% of national income. 'Continuity' is a rhetorical framing, not an empirical finding.

Evidence. Book table p. 43 / book-graphs.xls T1-5: IRPP 51.6, CSG (broad) 93.9, non-contributory contributions 164.2 bn of 1,680 bn = 3.07/5.59/9.77%. CCSS (Sept 2011) Tableau 1 p. 39: CSG 84.006 bn, of which 15.101 bn on replacement incomes (0.90% of national income); CSG rate table p. 420. Book pp. 67, 89; reply pp. 145-146.

Why this severity. The figures are right; the framing understates the familialised base by about 1 point of national income and is rhetorically inconsistent with the book. Conclusions are not affected.

9minorconfirmed · previously known · internal inconsistency

The reply says the book is only about taxes. In fact it also proposes changes to family and youth benefits, and the critic's complaint that one of its charts left out some benefits for poor families goes unanswered.

Technical detail and evidence

Why it matters, in detail. The book's family-policy analysis (p. 104) mixes cash benefits with tax reductions, so the critic's objection that RSA and housing benefits are omitted is relevant to it, whatever the scope of the tax analysis.

Evidence. Book pp. 101-115 (pp. 102-107, 109 read directly); p. 104 graph note ('montant moyen par enfant ... de toutes les prestations familiales enfants ... et la réduction d'impôt sur le revenu dû au quotient familial'); Sterdyniak (2012) pp. 130-131, 137; reply p. 149.

Why this severity. Presentation: the book's progressivity graphs (pp. 50, 91) do cover taxes only, so the scope defence survives, but the statement is literally inaccurate and leaves a specific criticism unanswered.

10minorconfirmed · previously known · overclaiming

The authors say all studies show married women are very sensitive to tax when deciding whether to work. The studies they cite do find women respond more than men, but one finds the response has roughly halved, and the other looks at hours, not at whether people work at all.

Technical detail and evidence

Why it matters, in detail. US participation elasticities of about 0.3 by 2000 and Danish intensive-margin bunching evidence support 'second earners respond more' but do not quantify a French participation effect of individualisation. No French estimate is cited, although Blundell, Bozio & Laroque (2011) appears in the reference list.

Evidence. Reply pp. 146-147. Blau & Kahn (Jan 2007 manuscript of JOLE 25: 393-438): own-wage elasticity fell 50-56%; participation elasticities 0.53-0.61 (1980) -> 0.27-0.30 (2000); 'the potential for marginal tax rate cuts to increase labor supply is much smaller now than 20 years ago'. Chetty, Friedman, Olsen & Pistaferri (NBER w15617; QJE 2011): 'we only consider intensive-margin labor supply choices and do not model labor force participation'; bunching 1.79 for married women vs 0.25 for single men, 'consistent with existing evidence that married women exhibit the highest labor supply elasticities'.

Why this severity. The qualitative ranking that underpins the incentive argument survives; only its strength and generality are overstated.

11minorconfirmed · newly identified here · documentation gap

Much of the reply's first section was copied from a reply the authors wrote for another journal. As a result it quotes the critic's wording from that other article and mentions a critic who does not appear in this issue. This does not change any conclusion.

Technical detail and evidence

Why it matters, in detail. Version and citation hygiene: the quoted passage and a cited critic come from a different exchange, and two references are mismatched.

Evidence. Reply pp. 145-151; TGS reply (doi:10.3917/tgs.027.0177) paragraphs 4-13 and note 1; Sterdyniak TGS article para. 11 vs OFCE article pp. 126-127; OFCE issue 122 table of contents.

Why this severity. Documentation only; no conclusion depends on it.

12minorlikely · previously known · selective presentation

The book's chart overstates how much tax relief rich families get per child, because relief for grown-up children still living at home is divided only among the younger ones. Rich families still get more per child after the correction, but the critic was partly right and the reply does not say so.

Technical detail and evidence

Why it matters, in detail. Numerator-denominator mismatch (QF for all dependants divided by minor children) inflates the top-percentile per-child figure by roughly a quarter; the ratio of the top to the middle falls from about 2.2 to 1.8.

Evidence. do_irpp.do l. 86 (QF parts include nenfmaj) and l. 201 (qf = reduc_qf/(1+marie)); do_prix_enfant.do l. 48-49 (child count = nenf, minors), l. 52 ('calage ... sauf qf'), l. 71 (prixenfant = qf + benefits). Auditor A's re-implementation (not re-run here): per-child QF at P99-100 EUR 3,532/yr (published 3,494); counting adult dependants (21.9% of dependants at P99-100) lowers it by about 24%, i.e. about EUR 224/month, with the total per-child transfer about EUR 318 rather than 385/month. Sterdyniak (2012) p. 130.

Why this severity. Concerns a book graph rather than a reply number; the concentration of QF benefits at the top survives the correction.

13minorplausible · previously known · methodology

The authors argue that couples do not pool all their money, so measuring living standards per household is misleading. But couples do pool a good part of it, so measuring each person only by their own earnings is not obviously better, and the reply does not test whether its conclusions survive the other approach.

Technical detail and evidence

Why it matters, in detail. Non-unitary (collective) models imply sharing rules between 0 and 1; the reply uses the rejection of full pooling to justify an analysis that implicitly assumes no pooling of labour income, without robustness across units (equivalised household income).

Evidence. Reply p. 147 and fn 2 p. 148; book p. 49 fn 1 ('Pour les couples mariés, le revenu individuel du capital est obtenu en divisant par deux le revenu du couple ... Nous obtenons des résultats qualitativement similaires avec d'autres champs'); Sterdyniak (2012) pp. 108-109, 129, 138.

Why this severity. A gap in the argument with no demonstrated effect on any number; the direction of the effect of alternative units on the measured profile was not established.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

(Considered and dropped by auditor B) The reply misrepresents its labour-supply sources: Blau & Kahn and Chetty et al. do not support greater second-earner responsiveness. — Both sources support the ranking: Blau & Kahn's participation elasticities fall but stay positive (0.27-0.30 in 2000), and Chetty et al. report bunching of 1.79 for married women vs 0.25 for single men, 'consistent with existing evidence that married women exhibit the highest labor supply elasticities'. Only the strength and generality of the reply's wording are overstated (kept as minor F10).
(Considered and dropped by auditor B) temp_nouvel_ir_2010.dta fails to reconcile with the published all-adult results, an additional replication failure. — It is the full-time-worker sample (ages 18-65, at least 80% of full time) and reconciles exactly (auditor B, R9).
Sterdyniak's 55.0% dividend rate in the 41%-bracket row of Table 6, versus 57.3% in Annex II.d, is an arithmetic slip. — 55.0% is the burden under the 19% prélèvement libératoire (PL) option, which is lower than the 57.1-57.3% barème burden and is what a taxpayer in the 41% bracket would choose. Annex II.d reports both, so the figure is not a slip.
Both versions of Sterdyniak's text say he agrees 'à juste titre' that high rates on top earners do not affect their work effort. — Only the TGS article contains 'à juste titre'; the OFCE article (pp. 126-127) does not. The quotation-source problem itself is confirmed (F13).
Even with statutory IS, dividends under the LPS reform are not taxed more heavily than salaries (so the reply's effective-rate point is beside the point). — True only at the 60% rate (73.8% vs 74.1-74.3%). At the 25% and 50% rates Table 6 shows dividends at 50.8% and 67.2% vs salaries at 42.1% and 65.1%. With an 18% effective IS the dividend burdens fall to 38.5% and 59.0%, below salaries, so the effective-rate argument does rebut the dividend part of the critique. It leaves the interest result untouched (F3).
About one third of the book's measured top-end decline in tax rates depends on the IS incidence convention. — An upper bound, not a central estimate: the shareholder-only variant implies a 36.6% IS rate on household dividends plus retained profits, above the 34.4% statutory rate. Treating life-insurance income as equity-like shrinks the decline only from 14.6 to 13.4 points (9%). The adjudicator range is a 9-36% shrinkage; the issue is retained as minor (F6).
Single-earner couples are the only household type in the case types that loses across broad income ranges. — Holds for childless case types only. With two children and the authors' EUR 750/child credit, single-earner couples gain EUR 505-1,462/yr from the SMIC to EUR 5,000/month and lose only from about EUR 6,000. The finding is retained as F1 with this qualification.
Because the published percentages concern individuals in 'unequal couples' rather than single-earner households, no reliable evidence on single-earner couples' losses is established. — Partly rejected. Single-earner couples are the extreme case of the 'unequal couples' tabulated in gagnants.xls. The case types reproduce the authors' own case-type workbook to the cent, and auditor B's own diagnostic (96.7% of childless single-earner full-time earners at EUR 2,000-3,000 paying more) points the same way. Household-level counts remain approximate because the public microfile vintage differs from the published tables.
gagnants.xls puts the current-law advantage of unequal over equal couples at 1.7-3.2% of income across P40-P99. — Minor correction only: 0.9% at P40-50 for all married adults (1.9% for full-time workers); 1.8-3.2% at P50-P99. Not material to F2.

Audit notes

What reproduced. Reproduced by the adjudicator: - the reply's aggregates: IRPP 3.07%, broad CSG 5.59%, non-contributory contributions 9.77% of EUR 1,680 bn (literal CSG 5.0%); pure transfers 4.0% (CN6; 3.5% from CN7 components); public spending above 50%; - CPO's implicit IS rate of about 18%; - the book's STI5 profile from the authors' microfile (within 0.1 point); - Sterdyniak's Table 6 from his Annex II formulas; - the authors' case-type workbook SimulCasTypes.xls to the cent, and auditor A's case-type results exactly; - every gagnants.xls and STI3/STI5/STI7 figure quoted; - the CERC table ratios. Computed: an IS-incidence sensitivity (top 0.001% rate 33.4% -> 34.8-39.1%; P95-99 -> top decline 14.6 -> 13.4-9.3 points). Not reproduced: - the 'simulations' the reply invokes, which are neither identified nor published; - household-level single-earner results from the public microfiles, whose new-IR file totals (EUR 165.6 bn / 181.9 bn) do not match the published SIR2 table (EUR 158.7 bn / 159.5 bn); auditor A's 58-69% loser share is indicative only; - auditor A's QF-per-child re-implementation (-24% at P99-100), which was checked against the code logic but not re-run.

Limits of this audit. - The reply's 'simulations' are unidentified, so F1 is tested against the authors' published outputs and case types. - The public microfiles are partly a different vintage from the 2011 published tables, so household-level single-earner results are approximate and were not re-run by the adjudicator. - No Stata was used; the authors' do-files were read as text and their logic re-implemented in Python. - Case types are wage-only, full-time, without niches, and use the authors' simplified 12% employee-contribution rate. - The IS-incidence variants are illustrative bounds, not estimates of true incidence. - Attanasio & Lechene (2002), Lundberg, Pollak & Wales (1997), Blundell et al. (2007), Hourriez & Olier (1997) and Blundell, Bozio & Laroque (2011) were not read in full by the adjudicator. - Allègre, Périvier & Pucci (2021), IPP (2012) and the DINA Guidelines (2016) are used only as context; the 2021 study simulates a revenue-raising individualisation, not the LPS package. - The Cairn/DOI version was not accessed; the OFCE issue PDF (identical copies held by both auditors, same pagination) was used. - This is not a re-audit of the 2011 book.