Personal Holding Companies, Tax Progressivity, and Inequality

Why it matters, in detail. The C7 statistic is (dSE[fiscal+PHC] - dSE[fiscal]) / (dUS[fiscal] - dSE[fiscal]) = 3.2/7.3, about 44%, with an implicit US holding-company adjustment of zero. Each +/-0.8 points of omitted US retained income under an RSZ-equivalent definition moves the closure by -/+11 points (33-55%). The IRS filer count measures compliance with a narrower legal test, and the deterrent surtax itself depresses it. It does not measure income retained in RSZ-type structures such as consolidated holding groups or gains-dominated vehicles. The omitted US trend is plausibly zero or negative, because closely held income migrated out of C-corporations after TRA86 (Auten-Splinter 2024; Kopczuk-Zwick 2020). The '40% vs <5%' contrast mostly reflects where gains are realized: personally in the US, inside exempt holdings in the Nordics.

Evidence. RSZ p.11 and fn 15; pp.28-29 and fn 37. 26 USC 542(a)-(b) and 543(a)-(b) (official 2020 text). IRS SOI Pub. 5108 (2020 line items), Form 1120: box A2 'Personal holding co.' = 2,253 and box A1a consolidated = 29,529. Balkir, Saez, Yagan & Zucman (2025, NBER w34170) Table 1 (top 0.0002%): fiscal income $188.7m, of which realized gains $114.9m (61%) and dividends $41.7m; economic income $452.3m, which includes 50% of realized gains; ratio 42%. Excluding gains: 73.8/452.3 = 16.3%, or 73.8/394.9 = 18.7%. Recomputation (scripts/check_endpoints.py): closure = (3.2 - d)/7.3 for a US-side increment d. That gives 71%, 55%, 44%, 33% and 22% for d = -2.0, -0.8, 0, +0.8 and +1.6. The US top-1% share (excluding gains) jumps 4.04 points in 1986-88 (TabFig2024 Table A1). Treating all of that jump as reclassification gives a closure of 98%; treating half of it gives about 60%.

Why this severity. The 'half' is not established on a like-for-like basis because its US premise is untested. But the best-documented US-side correction, the post-1986 move of closely held business income from C- to S-corporations, would shrink the US rise and so increase the closure. There is no evidence of a rising US analogue large enough to overturn the direction. The magnitude is therefore uncertain, but the conclusion is not unsupported. Auditor B rated this major, Auditor A minor.

Why it matters, in detail. Closure = [PHC(2020) - PHC(1980)]/7.32, where PHC is the holding-company part of the Swedish top-1% share. Setting PHC(1980) = PHC(2000) = 1.22 gives 46% at the exact 2020 endpoint. Each 1 point of 1980 holding-company income changes the closure by about 14 points. Early-2000s levels (0.9-1.2) give 42-49%; mid-2000s levels (1.6-2.75) give 22-33%. The Norwegian series cannot substitute because the 2005-06 dividend-tax shock dominates it. The arithmetic slips (6.9 vs 6.8-7.1; 7.8 vs 7.6) matter little.

Evidence. RSZ pp.26, 28-29. Fig 9a, exact vector values: fiscal 5.93 to 7.59; fiscal plus holding-company earnings 7.15 to 12.21. Fig 1a, exact. Roine-Waldenstrom appendix Table A2: 4.05 (1980), 5.97 (2000), last year 2013 (7.24). scripts/gap_sensitivity.py, closure by base period: 2000 = 46% (2020 end) or 43% (2018-20 end); 2000-02 average = 46-49%; 2003-05 = 39-42%; 2004-07 = 29-33%; 2006-07 = 22-26%. A 1980 holding-company component of 2.18 points would cut the closure to one third.

Why this severity. The direction is unaffected: fiscal income understates the rise of Swedish top incomes. The size of the 'half', however, depends on an unobserved 1980 level. Plausible alternative bases give about a quarter to a third, while early-2000s bases reproduce the paper's figure. This is a material uncertainty in one abstract claim, not a reversal.

Why it matters, in detail. Under the Saez-Zucman (2026) all-taxes convention, payroll taxes on labor and VAT on consumption mechanically vanish for households whose income is retained capital income. Part of the gradient therefore reflects income composition and high saving, not avoidance. Holding companies act on the personal income tax margin. In Norway, corporate tax absorbs most of that collapse, so the net direct-tax drop is about 7 points. The mechanical holding-company counterfactual adds about 17.6 points at the top (22.0 to 39.6 micro rate), i.e. 51-55% of the gap to the flat benchmark in the top-0.01% bins.

Evidence. Fig 5c-d (scripts/fig5cd.py); Fig A14 (scripts/figA14.py): the holding-company counterfactual gives 39.6-41.9% in the top-0.01% bins against the 56.5% benchmark; abstract; Sec. 5.1 pp.18-20; Sec. 5.2.

Why this severity. The measured all-in rates are correct under the paper's stated convention, and Sweden's top-end regressivity survives on direct taxes alone. But attributing the collapse to holding companies is overstated, especially for Norway, where direct-tax regressivity at the top is modest. This qualifies the reading of a core result without overturning it.

The authors' response. The paper states the all-taxes convention (following Saez & Zucman 2026) and shows the decomposition (Fig 5c-d). Answering the analogous French critique, Zucman (Le Grand Continent, 2 Jul 2025) argued that the gap survives excluding pension contributions and VAT. Here it survives for Sweden but largely not for Norway.

Why it matters, in detail. Ranking each year by economic income selects household-years with transitory spikes in retained income (e.g. exempt gains inside holdings). That biases annual top-cell rates downward relative to persistent-rank measures. Part of the A9 gap also reflects higher top rates in 2007-17 (Fig 8). Norway: ranked in 2020, 24.7% against 22.3% annual; ranked in 2007, 34.8%. Sweden: 22.6% against 13.6%. The P95-99 comparator is stable across measures (51-56%).

Evidence. Fig 5a (scripts/fig5.py); Fig A9 (scripts/figA9.py); Fig A12 (scripts/figA12.py); Fig 8; RSZ abstract, pp.2, 20, 23; Statistics Norway table 06070; Realtid, 'Duell om miljardarsskatt: Zucman mot Waldenstrom' (1 Apr 2026); Balkir et al. (2025) Table 2.

Why this severity. Regressivity at the top survives every measure: top rates of 14-35% against 51-56% at P95-99. But the headline level is the low end of the paper's own estimates, it is computed for a few dozen households with no uncertainty, and the multi-year rate is 1.1-1.7 times higher.

Why it matters, in detail. A cumulative flow ratio over a finite window stays below 1 for growing profit streams, and for firms that were young at the window's start, even when everything is eventually paid out. Conversely, for pre-existing steady-state firms a fixed-delay payout process would push the ratio toward 1. So the low ratios do indicate accumulation, but not that it is permanent. With carryover basis in both countries, tax eventually falls due on distribution or sale. A lifetime rate would add the present value of that liability, e.g. the 31.68% Norwegian dividend rate discounted over the deferral period.

Evidence. RSZ pp.2-3, 16-17, 23, fn 23. Fig 4 (scripts/fig4.py). Auditor B's synthetic cohorts: constant profits from 2002 with a 15- or 16-year payout delay give 21% or 16% observed payout despite 100% eventual payout. J. Fall, 'Skatt pa agande hotar hela tillvaxten', DN Debatt reply (summarized by TN, 11 Aug 2026). D. Waldenstrom, 'Skattedebatten bor starta i ratt ande', Svenska Dagbladet, 8 Aug 2026 (IFN PDF).

Why this severity. The descriptive payout facts are right, and conservative in that the numerator is broad. But permanence is not established, and the annual-rate concept ignores deferred tax. This qualifies the welfare and policy reading of C5 without changing the measured annual rates.

The authors' response. Fn 23 acknowledges the carryover basis and leaves heirs' realizations to future work. In their DN Debatt rejoinder ('Bara en del av de rikastes inkomster syns i deklarationen', 18 Aug 2026; paywalled, only the lead was read) the authors defend counting retained profits as income: 'our task as researchers is to show how the system actually works'.

Why it matters, in detail. Pre-trends at t=-3 are significant for the first stage and for both dividend outcomes. The alternative baseline raises the blocked share from about 51% to about 73%. Top-5% passthrough of 0.38 (CI -0.06 to 0.82) might not be significantly below 1 under owner-level clustering; this cannot be tested without the data. Selection into holding ownership is not addressed.

Evidence. RSZ pp.5, 13-16; Fig 3, A3, A4 (scripts/fig3.py); comparison with the identification in Lamadon et al. (2022), pp.187-192, from Auditor B.

Why this severity. The mechanism survives: holdings receive dividends and pass on roughly nothing, and the main bias runs toward more blocking, not less. But the causal language and the precision of 'around half' are overstated, and inference may be too optimistic if standard errors are not clustered.

The authors' response. p.14 argues that the pre-event differences 'work against us' because treated firms converge from above; this is true for the first stage. Fn 20 notes that the tax computation is mechanical, not a revenue estimate.

Why it matters, in detail. Without the endpoint rule, a looser classification in 2019-20 could raise the holding-company share at exactly the endpoint that sets the 3.2-3.4 point Swedish increment. Without group counts and uncertainty, the 13.6%/22.3% rates for groups of about 12-24 households cannot be judged against sampling and outlier variability. Without the alternative definitions, the sensitivity of C1/C6 to the 60% and 3-year thresholds is unknown.

Evidence. RSZ pp.6, 8-11, 19, 47; Fig 11 notes; Fig A17 legends; the NBER and author pages carry no replication link. Both auditors searched; none was found.

Why this severity. The undisclosed choices bear directly on headline endpoints and top groups, and they prevent outside verification; none is shown to bias the results.

Why it matters, in detail. Write consolidated holding-company retained earnings as R_H + R_S. R_H is holding-level retention (upstream dividends and portfolio income kept in the holding). R_S is the retained earnings of operating subsidiaries consolidated under a holding parent. Interposing a holding blocks only R_H; R_S would be retained under direct ownership too. RSZ's statistics and counterfactual use R_H + R_S. Since the holding-company counterfactual is 76% of the all-retained-earnings counterfactual in Norway, a sizeable R_S would push the holding-specific share of the top-end gap below the reported 51-55%. The flow-based event study (C2) independently supports substantial R_H for owners of smaller operating firms, but says nothing about its share at the very top.

Evidence. RSZ pp.7-12, fn 11-12 and 15. Fig 2a-b (exact): the holding-company share is 43-51% (Sweden) and 56-58% (Norway) of income in the top-0.01% bins, out of 77-98% undistributed in total. Fig A14 (exact): the holding-company counterfactual raises 3.93% of revenue, against 5.18% for taxing all retained earnings, i.e. 76%. The top-bin rate of 39.6% equals the observed 22.0% plus 31.68% times 56.1%. This mechanism was first identified by Auditor B; it is confirmed from the method text.

Why this severity. Total economic income, and hence the regressivity finding, is unaffected. But the share credited to holding companies as such may be overstated: C1's 'half', C6's 50-60% and the holding-company increment in C7. The amount is unknown and could be large at the very top, where fortunes sit in consolidated groups.

Why it matters, in detail. Asymmetry: gains realized personally are excluded from economic income, but gains realized inside holdings (and Norwegian fair-value changes) are included and untaxed. For a holding H owning stake s of a firm L whose retained earnings A are also allocated, a gain reflecting A counts s*A twice. If the gross-up applies to domestic dividends, it imputes about 28% (tau = 22%) of dividend income as corporate tax already counted at the subsidiary, raising both the numerator and the denominator of the top rate. Without disclosure, the net effect on the 13.6%/22.3% top rates cannot be signed.

Evidence. RSZ pp.6-11, fn 6, 10 and 15; Alstadsaeter et al. (JEEA forthcoming) fn 14 (double counting called 'empirically minor' in their sample). Corporate tax is 20.4% of income for Norway's top 0.0005% (Fig 5d), but how much of it is imputed is not reported.

Why this severity. The direction is mixed. Untaxed gains inside holdings lower top rates and raise the holding-company share, while an over-broad gross-up raises top rates. Magnitudes are unknown, but in groups of 12-24 households a single exempt exit inside a holding could dominate the annual top-group figures.

Why it matters, in detail. The operational definition is broader than US law (it includes gains and rents, applies to standalone entities and has no attribution rules), so it is an 'inspired by' rule, not an equivalence. The historical rate error does not enter any computation.

Evidence. 26 USC 541-543 (official 2020 text); 48 Stat. 680, section 351(a)-(b); SFS 1999:1229 ch. 24 ss. 32-33 (riksdagen.se text).

Why this severity. These are errors in legal and historical description. The analysis uses its own operational definition, and the substantive consequence for the US comparison is covered under F1.

Why it matters, in detail. Using 0.85% instead of 1.1% for 2018-20 changes nothing computed (Fig 5d is observed). A 10% liquidity discount would scale the counterfactual market-value base, and hence the 8-9% wealth-tax shares, by about 0.9.

Evidence. Skatteetaten wealth-tax rates 2018-2020; RSZ p.3, p.21, App. A; Fig 6, A15, A17 (scripts/fig6_a15.py, figA17.py); Smith, Zidar & Zwick (2023, QJE 138(1)) on the 0.9 factor; offshore-havens-23.

Why this severity. The qualitative conclusion, that the wealth tax takes about 1% of top income against more than 8% on market values, holds across valuation methods.

Why it matters, in detail. The fractile is off by a factor of 10 (about 7-8 vs 70-80 individuals). The reconciliation also mixes household (RSZ) and individual (Hammar) units.

Evidence. Hammar, Roth, Stokke & Waldenstrom (2026), 'Rising Gaps, Falling Inequality', abstract, fn 27, Fig B.3 note; RSZ p.27-28, fn 36, Fig A13.

Why this severity. The citation is supporting evidence only. The rising extreme-tail pattern is not reversed.

Why it matters, in detail. A mean slope below 1 can arise from measurement error in the external source or from sample composition. It does not justify a 12% correction, but 'unbiased' is not demonstrated.

Evidence. RSZ p.10; Fig A2 (PDF p.55), axis titles vs panel titles and notes.

Why this severity. A labelling error and an overstated description of a validation exercise. The median slope near 1 is reassuring.

Why it matters, in detail. Mostly rounding or mislabelled groups. (d) slightly weakens the narrative that rising regressivity came only from the top, since the bottom 99.99% rate also fell 10 points in Sweden while the top 0.01% fell 18.

Evidence. Exact vector reads in scripts/outputs (fig2, fig5, fig8, fig9, figA9, figA12, figA14); RSZ pp.2, 20, 22-24, 26, 50-51.

Why this severity. These are small numeric and labelling discrepancies. The figures are internally consistent with each other, and none changes a qualitative conclusion.

Why it matters, in detail. The difference-in-differences combines exposure to the reform with the cost of creating holdings. Pre-period deviations of about 1 point are of the same order as part of the tax-rate effect. Large samples make even small pre-trends significant.

Evidence. RSZ Sec. 5.3 pp.24-25, fn 32-33; Fig A10-A11 (scripts/figA10.py).

Why this severity. This affects the causal reading of a secondary result. The sharp 2005 jump in indirect ownership does support a role for transfer costs.