Introduction : Où est l'argent ?

Why it matters, in detail. C1 implies a change of +20 pp (ratio 2.0) in the US top-1% wealth share from the early 1980s to the late 2010s. The authors' then-current capitalized tax-unit series give 1980-84 averages of 23.7-25.0% and endpoints of 38.6% (2016) to 41.8% (2012): +14.9 to +16.9 pp, ratio 1.62-1.68. Zucman (2019) states 25-30% -> ~40%, i.e. +10 to +15 pp. The BPEA 2019 revised column gives +8.1 pp and the current WID +11.9 pp. Against the authors' central contemporaneous estimate the claimed rise is about 5 pp too large, roughly 25% of the headline change, and the proportional increase is overstated by about 60% (a 100% rise against about 62%). The most favourable reading (1978 trough to the superseded 2012 peak in SZ 2016) still gives only +18.9 pp (x1.82). Pairing the flat estate series (~20%) with a capitalization or SCF endpoint would produce a spurious doubling. The ~40% level and the sign of the trend are unaffected; 'quasiment comparable' to 1913-1929 (44-51%) is also overstated.

Evidence. Recomputed in final/scripts/us_top1_series.py. SZ 2016 QJE data (SaezZucman2015MainData.xlsx, sheet DataFig1-6-7b, 'Top 1%'): 1980-84 between 24.3% and 25.7% (average 25.0%), 2012 41.8%; +16.9 pp, x1.68. SZ BPEA 2019 data (SZ2019BPEAfinal.xlsx, DataFig2, column T): 23.3-24.4% (average 23.7%) -> 38.6% in 2016; +14.9 pp, x1.63. Zucman 2019 ARE data (Zucman2019Data.xlsx, DataF2, column C): 23.6-24.7% -> 38.9% (2016); +14.9 pp. WIR 2018 Fig. E8: 22.5% (1980) -> 37.2% (2014). WIR 2018 executive summary (English), p. 12: 'from 22% in 1980 to 39% in 2014'. Zucman (2019), Annual Review of Economics 11, abstract p. 109 and p. 121: 'a top 1% wealth share of approximately 40% in 2016 versus 25-30% in the 1980s'. The authors' alternative and later series rise less: BPEA 2019 'Capitalization revised' (DataFig2, column V) 27.1% -> 35.1%, +8.1 pp (x1.30); WID bulk 2026-10-15 (shwealt992) 25.6% -> 37.4% (2019), +11.9 pp (x1.47). Household survey data on a consistent definition (Wolff 2012, NBER WP 18559, Table 2): 33.8% in 1983 and 35.4% in 2010 (Zucman 2019, p. 121, warns that the 1983 SCF is not directly comparable with post-1989 waves). The only series near 20% is the Kopczuk-Saez estate series (1981-84: 19.1-21.1%). It is flat: 22.0% in 1989, 20.8% in 2000, 20.2% in 2008 (BPEA DataFig2, column W; ARE DataF2, column H). Post-war trough (1978): 21.7-22.9%. Early 20th century in the same series: 1913-16 averages 43.6-47.3%, 1929 48.0-50.6%, against 38.6-38.9% in 2016. WIR 2018 summary, p. 12: wealth inequality 'has not, however, yet reached its early-twentieth-century levels in Europe or in the United States'. SZ 2016's phrase 'almost as high as in 1929' refers to the top 0.1%, not the top 1%. The longer online version (RCE blog, 23 Jan 2020) says '20 % à la fin des années 1970 et au début des années 1980'.

Why this severity. The doubling is the interview's central quantitative claim about inequality. On the authors' own then-current numbers the rise is about a quarter smaller in percentage points (+14.9 vs +20 pp), and the proportional increase is about 60% rather than 100%; on Zucman's own 2019 summary (25-30% -> 40%) it is 25-50% smaller. The rubric threshold for major is about a quarter of the headline trend. The direction and the ~40% level survive, so C1 is weakened in magnitude, not reversed.

Why it matters, in detail. Numerator USD 616bn (2015); the denominator is the USD 1,703bn of profits of foreign-controlled corporations, a pool equal to 14.8% of world net corporate profits. Read as a share of multinationals' worldwide (home plus foreign) profits, the figure would be substantially lower. That share cannot be computed from TWZ's data, because multinational groups' home-country profits are pooled with local firms; shifted profits are 5.4% of all corporate profits. The 36.2% is also rounded up to 'environ 40 %'.

Evidence. TWZ (2018, NBER WP 24701, Aug. 2018 version), p. 3: 'close to 40% of multinational profits—defined as profits made by multinational companies outside of the country where their parent is located'. TWZ2018.xlsx, Table 1 (2015): shifted USD 616.46bn / profits of foreign-controlled corporations USD 1,703.23bn = 36.2%. Foreign-controlled profits are 14.8% of global net corporate profits (USD 11,515bn); shifted profits are 5.4% of global corporate profits (final/scripts/offshore_profit_checks.py). TWZ (2023 REStud, §5.2): '36% of multinational profit'. The RCE blog version (23 Jan 2020) adds 'Cela correspond aux profits qui sont fait par des sociétés en dehors des pays où elles sont domiciliées, comme ceux d'Apple en dehors des États-Unis ou Renault en dehors de France'; printed p. 15 lacks it.

Why this severity. The absolute amount shifted (about USD 600bn) is reported correctly, and so is the source's headline number. But the share as printed invites general readers to take it as a share of all multinational profits, which would be substantially lower (it cannot be computed from TWZ's data). The qualitative claim that large-scale profit shifting occurs is unaffected.

Why it matters, in detail. For 2016 the top-1% share ranges from 30.5% (DFA households, broader wealth concept) through 33.7% (SZZ heterogeneous returns), 35.1% (BPEA revised) and 36.3% (SZ 2020) to 38.6-38.9% (SZ baseline; SCF families); the 40% sits at the top of this range but is supported by the SCF. The trend is the fragile part: 1989-2016 rises span +6.6 (SZZ) to +11 pp (capitalized). The pre-1989 rise rests on the equal-returns capitalization alone, and the authors' own revised column cuts the 1980-84 to 2016 change from +14.9 to +8.1 pp. None of this is disclosed. The direction of C1 survives every series; the magnitude of the rise is uncertain within roughly a factor of two.

Evidence. Level in 2016: SZ capitalized 38.6-38.9% (tax units). SCF, top 1% of families: 38.55%, recomputed here from the Fed's SCFP2016 public extract over 5 implicates (final/scripts/scf2016_top1.py; Fed Bulletin 2017: 38.6%); SCF 2019: 37.2% (scf2019_top1.py). SZ (2020 JEP 34(4), p. 9): the harmonized DFA and SZ both give 38% for tax units in 2018, +10 pp since 1989. Lower estimates: Fed DFA households 30.5% (2016:Q4, current vintage; includes consumer durables and DB pension entitlements). SZZ (2023 QJE, p. 518 fn. 1): 33.7% in 2016, a rise of +6.6 pp over 1989-2016, against +10.0 pp in PSZ and +7.7 pp to 36.3% in Saez-Zucman's own 2020 heterogeneity-adjusted series. BPEA 2019 'Capitalization revised': 35.1%. Current WID: 37.4% (tax units, 2019) and 34.8% (equal-split adults). Trend: the comparable SCF starts in 1989; Zucman (2019, p. 121) reports +9 pp (raw SCF) or +10 pp (with Forbes 400) since 1989, against +11 pp in capitalized data. Before 1989 only the capitalization method shows a rise; the estate-multiplier series is flat. Unit: SZ (2020 JEP, fn. 3), the Fed's top 1% of households is about the top 0.7% of tax units.

Why this severity. The level is corroborated by independent survey data, so the issue bears mainly on the size of the rise, which F1 already shows is overstated. Depending on method, the post-1980 rise runs from about +8 to +17 pp. This is a known, actively debated critique with substantive author responses; it qualifies C1 rather than undermining it. It should not be added to F1 as a separate deduction.

The authors' response. Saez & Zucman (2020 JEP, pp. 8-9): after converting DFA households to tax units and excluding durables and unfunded pensions, both sources give a top-1% share of 38% in 2018, +10 pp since 1989. Saez & Zucman, 'Comments on Smith, Zidar and Zwick (2020)' (14 May 2020): SZZ's top interest yields are too high and dividend-based capitalization undercounts billionaire equity; once these are fixed, SZZ is close to the SZ benchmark. BPEA 2019 (pp. 451-453) presented a 'modified' series with higher top interest rates and a x1.9 pass-through valuation.

Why it matters, in detail. The 36% rests on the counterfactual that, absent shifting, foreign affiliates in havens would have the profit-to-wage ratio of local haven firms; all excess profitability is attributed to shifting. Within TWZ's own sensitivity the share falls to ~30% under a +45 pp benchmark or without the upward corrections to haven data, and to 21.7% in the combined lower bound. Firm-level estimates are lower still (13-22%; <20% micro-implied), while non-linear CbC estimates are close to TWZ. Calling this 'the estimates' overstates the certainty of a number whose plausible range is roughly 13-43%. Revenue-loss estimates converge more than shifted-profit shares.

Evidence. TWZ (2018), p. 23: estimates 'should probably be viewed as conservative'; no range given. TWZ (2023 REStud, §5.2): preferred 36%, 'lower bound of 22% and an upper bound of 43%'. TWZ2022.xlsx Table 2, recomputed on the 2015 foreign-profit denominator: +45 pp benchmark profitability 29.7%; removing corrections to haven data 29.8% (USD 507bn); lower and upper bounds 21.7% and 43.4% (USD 369bn and 740bn); excess-transactions alternative 37.9% (USD 646bn). Dharmapala (2019, AEA P&P 109: 488-492): extrapolating micro semi-elasticities 'implies that less than 20% of MNCs' foreign profits are shifted to tax havens'. Fuest, Greil, Hugger & Neumeier (2022, CESifo WP 9757, pp. 25-26), using country-by-country reports (post-dating the interview): 13% of foreign profits (22% with a payroll-based counterfactual). Forstater (2018, CGD) and Hodge (2018, Tax Foundation) question labelling returns to intangibles as artificial. On the other side, Garcia-Bernardo & Janský (2024), with a log specification on CbC data, find more than USD 850bn shifted in 2017 (per corpus offshore-havens-10).

Why this severity. The interview reports the authors' central estimate correctly, so this is a robustness and disclosure concern, not an error. Across defensible counterfactuals the share runs from about 13% to 43% of foreign profits, and 40% sits at the top. The qualitative conclusion that shifting is large survives every estimate.

The authors' response. TWZ (2023 REStud): firm-level data miss most haven profits, since in 2012 only 17% of multinationals' consolidated profits could be traced in Orbis (§2.1); the preferred estimate is reported with 22-43% bounds and an excess-transactions estimate of USD 646bn (§5.2); BEA double counting does not affect their profit-type-return data (§4.2.1).

Why it matters, in detail. Under Ramsey's assumptions (independent demands, small taxes), optimal rates are inversely proportional to compensated elasticities, so any finite elasticity implies a positive rate. A zero source tax on perfectly mobile capital is a different, small-open-economy result. The misstatement makes the opposing position look more extreme than the textbook rule; no quantitative implication.

Evidence. Ramsey (1927), 'A Contribution to the Theory of Taxation', Economic Journal 37(145): 47-61, p. 59(b): 'that should be taxed most for which the elasticity of demand is least'. In the same passage (p. 59, (e)), even with an infinitely elastic demand for saving, Ramsey concludes that 'income-tax should be partially but not wholly remitted on savings'. Page image read: /mnt/zdata/downloads/ou-est-argent-2019/ramsey-page-14.jpg.

Why this severity. This is a pedagogical misstatement in a passage that sets up the opposing view. It affects no empirical claim or number.

Why it matters, in detail. Using the weighted rate on shifted profits (6.6%) instead of the 0-5% band (midpoint 2.5%) lowers the implied rate wedge against 25-30% statutory rates from about 25 pp to about 21 pp, i.e. the implied revenue loss per euro shifted by roughly a sixth. The USD 616bn shifted-profit figure is unaffected.

Evidence. TWZ2018.xlsx, sheet 'Data F8b' ('Tax rate on shifted profits'): Luxembourg 2.8%, Puerto Rico 3.3%, Ireland 4.4%, Malta 4.8%, Caribbean 4.9%, Singapore 5.8%, Switzerland 8.0%, Netherlands 10.5%, Cyprus 12.5%, Hong Kong 14.2%, Belgium 18.7%. Shifted-profit-weighted mean 6.6% (Netherlands counted once; the sheet lists it twice); havens taxing above 5% receive about 45% of these shifted profits (final/scripts/offshore_profit_checks.py). TWZ (2018), p. 3: havens tax paper profits 'at very low rates (less than 5%)'. Zucman, Le Monde (7 Nov 2017): 'généralement entre 0 % et 5 %'.

Why this severity. This is a small simplification that follows the source's own text. It widens the implied rate gap by a few points but does not affect the shifted-profit estimate.

Why it matters, in detail. The relevant event is the acceleration of an existing excise component (EUR 30.5 -> 44.6/tCO2 in 2018, then +10.40 a year, i.e. EUR 55 scheduled for 2019), legislated simultaneously with the ISF->IFI and PFU reforms. The tax was not newly introduced. No quantitative claim depends on this.

Evidence. SNBC indicator 'Prix réel du carbone' (indicateurs-snbc.developpement-durable.gouv.fr/prix-reel-du-carbone-a132.html): the carbon component 'est gelée à 44,6 €/tCO2 depuis fin 2018 suite à la crise des gilets jaunes. Elle était passée de 7 €/tCO2 en 2014 à 44,6 €/tCO2 en 2018.' Sénat, PLF 2018 presentation note (senat.fr/commission/fin/pjlf2018/np/np10a/np10a3.html): the 2018 rate rose from 30.5 to 44.6 EUR/tCO2, with a trajectory to 86.20 EUR in 2022. The same finance law (loi n° 2017-1837 du 30 décembre 2017 de finances pour 2018) abolished the ISF and created the IFI (art. 31) and created the PFU flat tax on capital income (art. 28) (BOFiP, ACTU-2018-00087 and ACTU-2019-00223, bofip.impots.gouv.fr).

Why this severity. This is a factual imprecision in a political-economy argument. The argument, that a regressive carbon-tax increase coinciding with tax cuts at the top was politically untenable, is unaffected.

Why it matters, in detail. Formulary apportionment by destination sales neutralises intra-group profit location but moves the avoidance margin to the sales factor: booking location, interposed unrelated distributors, digital sales. Group boundaries and consolidation rules are also choice variables, as in US state combined reporting. No estimate of the leakage under a French unilateral rule is available; the existing-system sales-shifting estimate (about USD 80bn, 2013) should not be read as one. The categorical 'cannot be manipulated' overstates robustness; the qualitative feasibility claim stands.

Evidence. Laffitte & Toubal (2022), AEJ: Economic Policy 14(4): 371-396, abstract: US multinationals 'record sales and the profit from these sales in tax havens' (sales shifting about USD 80bn of profit shifting in 2013); sales-based rules 'may not efficiently address profit shifting if the policy designs are unable to identify sales by destination'. Author manuscript, pp. 24-25, citing Fleming, Peroni & Shay (2014): 'sales manipulation is still possible under destination-based taxation by the mean of third-party distributors'; look-through rules may help but 'their administrative cost may be very high'. L&T 'do not argue that sales-based policies should be discarded'.

Why this severity. This is a policy-design caveat on a secondary, non-empirical claim. The interview's conclusion, that some unilateral options exist, survives, and Zucman's fuller op-ed specifies final customers identified through VAT data.

The authors' response. Zucman, 'Pour une vraie taxation des GAFA', Les Echos (30 May 2018): the key is sales in each country, and final customers 'sont bien identifiés, car cette information est utilisée pour appliquer la TVA'. Zucman (2014 JEP, pp. 134-136) presented sales-based apportionment with caveats (per corpus offshore-havens-15).

Why it matters, in detail. The global stock (8%) is identified from cross-border asset-liability discrepancies; its distribution across wealth groups is observed only in Scandinavian leak and amnesty samples, and extrapolating even to all Scandinavian offshore wealth is an assumption in AJZ. External evidence is mixed: the Dutch amnesty data analysed with AJZ's method give a top-0.01% share of hidden wealth of about 7%, while Colombian evidence points to high concentration. The qualitative 'extrêmement concentré' for the world is therefore plausible but untested. No core number changes.

Evidence. AJZ (2019, AER 109(6)), p. 2090: 'we assume that 77 percent of it belongs to the top 0.1 percent richest households, 52 percent belongs to the 0.01 percent', the fractions observed in the HSBC and amnesty samples. AJZ2017.xlsx Data-F6a: top 0.01% holds 55.3% (HSBC) and 47.8% (amnesty), average 51.6%. AJZ (2019, p. 2075): 'We certainly do not claim that our estimates ... hold everywhere as a universal law.' Leenders, Lejour, Rabaté & van 't Riet (2023, J. Public Econ. 217: 104785), abstract: in the Netherlands the 'merely rich' (P90-P99.9) 'own 67% of hidden wealth, while the super rich account for only 7%'. Zucman (2013) Table III: USD 5,878bn / 73,625bn = 7.98% (end-2008).

Why this severity. The numerical claims are correctly sourced and restricted; only the global qualitative generalisation, and the loose 'l'essentiel', go beyond the evidence. AJZ argue that nothing specific to Scandinavia drives top-end evasion.

The authors' response. AJZ (2019 AER, p. 2075): they 'certainly do not claim' universality, but argue that 'there is nothing unique to Scandinavia that could explain the high evasion rates we find at the top' and that evasion among the wealthy 'may be even higher elsewhere'.