← All papers
Emmanuel Saez, Danny Yagan, Gabriel Zucman · Policy memo / working paper, October 21, 2020
Improving Tax Information Infrastructure
11 problems found: 4 moderate · 7 minor
What the paper claims
- Important holes remain in U.S. tax information infrastructure; the ten proposed reporting changes would 'substantially improve tax enforcement on high earners and large businesses' and 'substantially improve measurement of income, wealth, savings, and capital formation at the micro-level'.
- Ownership reporting should apply to all sizeable businesses and trusts. Balance-sheet reporting 'already appl[ies] to most business entities' and should be extended to all sizeable businesses, including foreign businesses with US owners, and to trusts.
- 'The wealth of the bottom 99% and racial inequality could be measured well with some modest additional information reporting': account balances, real-estate values, debt, and SSA race data shared with the IRS.
- Partnership ownership is opaque: 'Cooper et al. (2016) show that 30% of partnership income cannot be uniquely attributed to owners'. Of this, 20% accrues to taxpayers 'whom the IRS cannot actually identify' and 'another 10%' is earned in circular ownership structures. Proposed remedies: EIN matching, country and entity type for foreign owners, automatic audits of circular structures, or a small tax on distributions to partnership owners.
- Descriptions of current reporting law that define the 'holes': no balance sheets for US-owned foreign businesses ('Cayman Islands hedge funds currently provide no entity-level reporting'); DTC registration of 'all public companies and most large private companies'; officer ownership on 'Form 1120, Schedule E'; seller-only Form 1099-B; balance sheets reporting 'market values of securities'; no property-tax information reporting; no IRS race data; SSA race information available to share.
Every problem found
1moderateconfirmed · newly identified here · external inconsistency
The memo says the IRS gets no balance sheets from foreign companies owned by Americans. In fact, Americans who control a foreign company or partnership must already send the IRS its balance sheet. The real gap is foreign firms that Americans own without controlling them, so the problem is smaller than the memo says, though it is real.
Technical detail and evidence
Why it matters, in detail. The memo's baseline is wrong for the economically dominant class of US-owned foreign businesses. Controlled foreign corporations file Schedule F through Category 4 filers; SOI TY2016 counts about 95,000 CFCs of US corporations with about $23tn of assets. US-controlled foreign partnerships file Form 8865 Schedule L, and foreign partnerships with US-source income file Form 1065. Item 3's incremental coverage is therefore the non-controlled segment (dispersed >=10% US ownership, PFIC feeders, partnerships without US-source income), which the memo does not size. The error overstates the information gap, but no estimate depends on it.
Evidence. Instructions for Form 5471 (Rev. Feb 2020), Category 4 definition (control = more than 50% of vote or value) and p.4 'Table of Required Information': 'Schedules C and F' is checked for Categories 3 and 4 (rendered page inspected). Instructions for Form 8865 (2019), p.3 chart: 'Schedule L - Balance Sheets per Books' is required of Category 1 (control) filers. Instructions for Form 1065 (2019), p.4 'Foreign Partnerships': a foreign partnership with ECI or 'gross income derived from sources in the United States must file Form 1065, even if ... all its members are foreign persons'; the exceptions require US-source income <= $20,000 and <1% of items allocable to direct US partners, or no US partners at all. SOI Table 1, U.S. Corporations and Their Controlled Foreign Corporations, TY2016 (published Sept 2019): 95,374 CFCs of 16,225 US corporations, with about $23.2 trillion of end-of-year assets (the header says '$ thousands' but the values are only plausible in dollars). U.S. Treasury press release, 29 Nov 2013: the Cayman Islands IGA is Model 1B and Cayman FFIs report US account holders. Memo p.2, item 3. scripts/doc_checks.py.
Why this severity. A categorical misstatement of current law in one of the ten items. It overstates that item's 'hole', but the proposed extension (to foreign entities >=10% US-owned without a US controller) still targets a real gap, so the thesis survives.
2moderateconfirmed · newly identified here · external inconsistency
The memo assumes that the central stock depository knows who owns company shares. It only knows which banks and brokers hold them, and it holds almost no private-company stock. That part of the plan would not work as described, although the memo's other ways of collecting ownership data would still help.
Technical detail and evidence
Why it matters, in detail. A DTC return would identify about 400 participant positions, not taxpayers. For public companies, beneficial owners would come from the separately proposed broker year-end holdings, so the DTC leg adds little. Private C-corporation stock is generally restricted and recorded on issuer or transfer-agent books, so for private firms the DTC leg adds essentially nothing. Private C-corp ownership would then rest on issuer self-reporting with no third-party cross-check. The share of large private companies with DTC-eligible stock is not published, but DTC's eligibility rules exclude ordinary restricted equity.
Evidence. DTCC Issuer Services FAQ (29 Nov 2021): 'DTC is able to provide position information on a security at the DTC participant level ... DTC does NOT have beneficial owner information.' SEC, Concept Release on the U.S. Proxy System, 75 FR 42982 (22 Jul 2010), at 42985-86: Cede & Co. 'appears in an issuer's stock records as the sole registered owner'; securities are held in 'fungible bulk'; 'approximately 400 DTC participants'. The same release estimates that '70% to 80% of all public issuers' shares are held in street name'. DTC Operational Arrangements s.I.A.2 and fn.3: eligibility rules, and securities subject to transfer restrictions are custody-only. Memo p.1, item 1. scripts/doc_checks.py.
Why this severity. Removes the third-party verification leg of the C-corporation ownership proposal for private companies, the main concern for top wealth. The self-reported ownership schedule and the broker year-end-holdings proposal are unaffected.
3moderateconfirmed · newly identified here · overclaiming
The memo suggests the IRS could measure racial gaps by borrowing race information from Social Security. But Social Security has not recorded race for most people who got their numbers at birth since the late 1980s, or for most immigrants, and its older records lump many groups together. On its own, this would not measure racial inequality well.
Technical detail and evidence
Why it matters, in detail. SSA race fields are absent for nearly all SSNs issued through EAB since 1987 (96-99% of new SSNs) and through EAE since 2002. Pre-1980 SS-5 records use three categories with no Hispanic origin or Asian category. SSA-IRS sharing would therefore yield race mainly for older native-born cohorts, adequate chiefly for Black-white contrasts. It would leave systematic gaps for adults enumerated at birth since the late 1980s and for most immigrants, and would miss Hispanic and Asian identification among pre-1980 enumerations. Closing those gaps needs imputation (BISG-type) or Census linkage, which is not 'modest additional information'. Missingness by cohort is not published, so the bias cannot be sized.
Evidence. P. P. Martin (2016), 'Why Researchers Now Rely on Surveys for Race Data on OASDI and SSI Programs: A Comparison of Four Major Surveys', SSA Research and Statistics Note No. 2016-01 (Jan 2016), https://www.ssa.gov/policy/docs/rsnotes/rsn2016-01.html, read from a Wayback capture of 23 Aug 2019 because ssa.gov returned HTTP 403; the note gives the author's views. SSA, RECS OMB supporting statement (Oct 2007): 'has no reliable, statistically valid means of capturing race/ethnicity data as part of our core business process'. POMS RM 10205.505 (eff. 15 Jan 2010): 'About 99% of SSNs for infants are assigned through the EAB process'. 88 FR, doc. 2023-17727 (17 Aug 2023): 'SSA plans to expand the EAB process to include ... race and ethnicity'. Treasury OTA TP-11 (Fisher 2023) imputes race instead. Memo p.1 para. 2 and p.3, item 10. scripts/doc_checks.py.
Why this severity. Substantially weakens the race half of C3, one of the memo's three summary claims, because the proposed data source is inadequate by its custodian's own account. It does not affect the enforcement thesis (C1) or the ownership and balance-sheet proposals.
4moderatelikely · newly identified here · overclaiming
Reporting bank, brokerage and 401(k) balances, home values and debts would capture most of what ordinary families own. It would miss traditional employer pensions, which are nearly a fifth of the wealth of the bottom 99%. Those would still have to be estimated, so the memo's promise is somewhat too strong.
Technical detail and evidence
Why it matters, in detail. On DFA 2019Q4 shares, the proposals directly observe about 64% of bottom-99% gross assets (about 68% of non-durable assets). They partially observe 6.4% (non-corporate equity; small entities file no balance sheets) and leave about 25% of non-durable assets unobserved, of which DB entitlements are 19.2 pp (20.2% of P50-90 assets). DB wealth can be imputed from plan formulas, earnings histories and 1099-R flows, as in DINA. Measurement would therefore improve greatly on capitalization, but the claim needs a DB reporting or imputation step the memo omits. The problem is the omission of a large asset class, not a numerical error.
Evidence. Fed DFA, 2026Q2 vintage, dfa-networth-levels-detail.csv, 2019Q4, Next9+Next40+Bottom50: assets $89.45tn; DB $16.24tn; annuities $2.28tn; life insurance $1.26tn; durables $4.95tn; loans and miscellaneous $1.44tn; miscellaneous other equity $5.74tn. Sep-2020 vintage, available at the memo date: pension entitlements (DB+DC) are 28.7% of bottom-99% assets. scripts/dfa_bottom99.py reproduces Auditor A's figures exactly.
Why this severity. Qualifies the bottom-99% half of C3. Direct observation would cover most but not all of middle-class wealth, and the largest omitted item must still be imputed, so 'measured well with modest reporting' holds only with an imputation step the memo does not mention.
5minorconfirmed · newly identified here · selective presentation
The memo says the IRS lacks information on who owns ordinary corporations. But corporations already have to list anyone holding 20% or more of their voting shares. The plan to collect all owners would still add a lot; it just starts from a less empty baseline than described.
Technical detail and evidence
Why it matters, in detail. Schedule G already captures direct >=20% and direct-or-indirect >=50% voting owners, with s.267(c) constructive ownership. With q.5 and Form 851 it also captures parent and subsidiary links: the main nodes of corporate ownership chains and the principal owners of most closely held private C-corps. Item 1's incremental content is sub-20% stakes, non-voting classes and public-company holdings. Whether the IRS transcribes Schedule G into research files is unknown; Cooper et al. (p.97) say C-corps 'cannot be comprehensively matched' to owners, which may explain the memo's framing. The Schedule E reference is a citation slip.
Evidence. Form 1120 (2019), Schedule K q.4a, 4b, 5, 7, 10. Schedule G (Form 1120) (Rev. 12-2011), Parts I-II. Form 1120 (2010) line 12: 'Compensation of officers (Schedule E, line 4)'. Form 1120 (2011) line 12: 'from Form 1125-E'. Form 1125-E (Rev. 10-2016): 'Complete Form 1125-E only if total receipts are $500,000 or more'. Cooper et al. (2016) p.97: 'C-corporations and their owners cannot be comprehensively matched'. Memo p.1, item 1. scripts/doc_checks.py.
Why this severity. The omission overstates the hole for closely held C-corporations, whose major owners are already reported. The proposal (all owners; 1%+ public stakes) remains a large extension, and no conclusion depends on the misdescription.
6minorconfirmed · newly identified here · misrepresentation of source
The memo says a fifth of partnership income goes to owners the IRS cannot identify; that is about right for roughly two-thirds of that fifth, while the rest goes to owners with valid IRS numbers whose type of entity could not be determined. The proposed fix would reach only the first group. The tax form also already asks for the owner details the memo wants added.
Technical detail and evidence
Why it matters, in detail. Of the 20.1 pp, 7.1 pp (35%) are valid EINs without a matched return, which an EIN-validity rejection rule cannot reach. The remedy therefore addresses at most the 13.0 pp of unclassifiable TINs, plus foreign partners (9.3 pp are already identified separately through Forms 1042/1042-S/8805). Because the K-1 already has entity-type and foreign-partner fields, the effective fix is mandatory, validated transcription of those fields (perhaps plus a country field), not new requirements.
Evidence. Cooper et al. (2016, Tax Policy and the Economy 30): p.94; pp.99-100 (partner-type definitions and table); p.102, Fig. 3B (13.0 and 7.1; foreign persons separately 9.3), read from the rendered page; p.103; p.117 (20.1%); notes 3 and 5, p.125. Schedule K-1 (Form 1065) 2011, items H and I; 2019, items H1, H2, I1, I2. Memo p.2, item 2. scripts/cooper_and_capacity.py, scripts/doc_checks.py.
Why this severity. The 20% figure is correct and the remedies are sensible. The issue limits how much of the 20% the fixes could reach (at most the 13.0-pp group) and recasts part of the remedy as enforcement of existing fields; no conclusion changes.
7minorconfirmed · newly identified here · external inconsistency
The memo says company and trust balance sheets already show the market value of investments and lack it only for buildings and similar assets. In fact, company balance sheets use bookkeeping values for everything, while the few trust balance sheets the IRS receives already use market values. The memo's fix is still sensible, but it should cover more assets.
Technical detail and evidence
Why it matters, in detail. Schedule L values follow the filer's books. Securities are at fair value only where GAAP requires it (trading and AFS debt; equity securities after ASU 2016-01) and at cost on tax-basis books; equity-method and subsidiary holdings are not marked to market. Measuring wealth from business balance sheets would therefore need valuation-basis fields for all asset classes, not only real and intangible assets. The trust balance sheets that exist are already FMV-based, so the trust gap is coverage (Form 1041 has none), not valuation.
Evidence. Form 1120 (2019) Schedule L; Instructions for Form 1120 (2019), Schedule L. Instructions for Form 3520-A (2019), Part III; Form 3520-A line 9 'Real property'. Form 5227 (2019) Part IV, column (c) FMV. Form 990-PF (2019) Part II, column (c) 'Fair Market Value'. Form 1041 (2019): no balance sheet. Memo p.3, item 5. scripts/doc_checks.py.
Why this severity. Does not affect the recommendation to add market-value lines, which should if anything also cover business securities. The misdescription understates one gap and overstates another.
8minorconfirmed · newly identified here · selective presentation
The memo proposes that share buyers get a form recording what they paid, to help enforce capital gains tax. For most shares bought through brokers since 2011, brokers already report this when the shares are sold, so the gain for tax enforcement is smaller than implied.
Technical detail and evidence
Why it matters, in detail. For publicly traded securities acquired after 2010 and held at brokers, basis is already third-party reported at sale. A purchase-date return therefore adds enforcement value only for legacy (pre-2011) lots, transfers outside brokers and private-company shares. The memo's other rationale, micro-level saving and consumption from purchases plus balances, is distinct and unaffected.
Evidence. Instructions for Form 1099-B (2019): covered-security definition, box 1e, transfer statements (s.6045A). Saez, Yagan & Zucman (Jan 2021), 'Capital Gains Withholding', printed p.9 (PDF p.10), 'Administration' paragraph. Memo p.3, item 7. scripts/doc_checks.py.
Why this severity. Limits item 7's incremental enforcement gain to non-covered lots and to assets outside brokers (e.g. private shares). Its saving-measurement rationale is unaffected.
9minorconfirmed · newly identified here · external inconsistency
A few smaller descriptions of current rules are wrong, for example which foreign trusts file balance sheets and whether all charities do. These slips mostly make current reporting look more complete than it is, and they do not change any recommendation.
Technical detail and evidence
Why it matters, in detail. These errors run against the memo's own argument. They overstate existing trust and non-profit reporting, and understate the tracing problem for S-corporations owned through trusts. That pattern suggests imprecision about current law rather than one-directional advocacy; no proposal changes.
Evidence. Instructions for Form 3520-A (2019), 'Purpose of Form' and 'Who Must File'. Instructions for Form 990 (2019), General Instructions A-B (Form 990-N for gross receipts normally <= $50,000; organizations not required to file). Instructions for Form 2553 (Rev. Dec 2017), conditions 3-4. Memo p.2, items 2-4. scripts/doc_checks.py.
Why this severity. Simplifications that change no recommendation. Items (1)-(2) overstate existing coverage, so they understate the gaps the memo wants closed.
10minorlikely · newly identified here · methodology
The memo says partnerships that own each other in a loop make it impossible to work out who ultimately owns them. In fact the loop can usually be untangled with simple algebra if each partner's share is known; the cited study used a step-by-step method that could not handle loops. These structures still deserve scrutiny because they pay unusually little tax.
Technical detail and evidence
Why it matters, in detail. Under Cooper's own proportional-allocation assumption, the 9.7-pp circular-only component of the 29.8% union can in principle be attributed with a Leontief-type inverse. 'Cannot be uniquely attributed' is therefore specific to the algorithm. The defensible statement is about 20% unclassifiable, plus circular income that recursive tracing did not resolve. Data problems may still block resolution in practice, untested here: K-1 totals that do not reconcile with Form 1065, multiple DLNs per TIN, fiscal-year mismatches, and item-specific special allocations.
Evidence. Cooper et al. (2016) p.94 ('15% of income is in circular structures and cannot be uniquely linked to an originating partnership'), p.98 (14.9%), p.118 ($100bn of K-1 income to non-partnership owners; 22,417 circular partnerships), p.121 ('either the ultimate owner or the originating partnership'), p.122 (the proportionality assumption), pp.123-124 (recursion; 'These partnerships cannot be assigned an effective rate'). scripts/cycle_attribution.py: at 50% cross-ownership det(I-M)=0.875 and outside owners receive 57.14/14.29/28.57 of 100; a closed 100% cycle is singular; a special-allocation example gives identical K-1 totals with different origins.
Why this severity. Weakens one rationale (non-identifiability) for the automatic-audit and distribution-tax remedies, and in principle up to a third of the 30% opacity statistic. But the memo faithfully reports its source, and the practical concern stands: low taxation (8.8% vs 17.1%) and tracing difficulty in real K-1 data.
11minorplausible · newly identified here · speculative assumption
Auditing every partnership caught up in an ownership loop would mean about three times as many partnership audits as the IRS did in a whole year. The memo does not say how this would be resourced, though the threat alone might make many such loops disappear.
Technical detail and evidence
Why it matters, in detail. 22,417/7,478 = 3.0x FY2019 partnership examinations, concentrated on the most complex structures (about 428 K-1s per circular partnership). Unless the trigger is a light-touch check, or deterrence quickly unwinds the structures, the remedy implies a large reallocation of examination resources that the memo does not discuss.
Evidence. Cooper et al. (2016) pp.118, 123. IRS Data Book FY2019 (Pub. 55-B), Table 17b, 'Partnership returns': 4,223,801 filed; 7,478 examined. scripts/cooper_and_capacity.py.
Why this severity. A feasibility caveat on one remedy. Deterrence could shrink the count, and the memo offers an alternative (a small tax on distributions to partnership owners).
Claims examined and rejected
Problems one auditor raised that did not survive checking.
- I5: the memo's 'another 10%' circular share misstates Cooper et al.'s 15%. — The memo partitions Cooper's 29.8% union into non-overlapping parts: 20.1% unclassifiable plus a 9.7-pp circular-only increment. The circular total is 14.9%, with a 5.2-pp overlap. 'Another' signals the increment, the arithmetic is exact, and no conclusion depends on it; Auditor B reached the same view. That readers may infer a 10% circular total is noted under F7, but it is not a misstatement.
- I11(5): required minimum distributions 'for those above age 70' is outdated after the SECURE Act (age 72). — Loose but not false at the memo date. Everyone who reached 70 1/2 before 2020 remained subject to RMDs; the age-72 start applies only to those reaching 70 1/2 after 2019, and the 2020 waiver was temporary (Notice 2020-51). The proposal is prospective and nothing depends on the exact age; Auditor B reached the same view.
- I4(b) sub-claim: an EIN-database matching rule 'targets at most' the 7.1-pp 'Unidentified EIN' group and leaves the 13.0-pp 'Unidentified TIN type' group untouched. — This reverses Cooper's categories. By Cooper's definitions (p.100), 'Unidentified EIN' TINs are already classified as EINs by the tax database, so they would pass an EIN-database match. 'Unidentified TIN type' TINs are recognized neither as an individual nor as an EIN, so any reported as US owner EINs would fail it. The rule therefore bites on the 13.0-pp group, with the foreign-owner requirement covering foreign partners. The correct qualification, kept as F6, is that the remedy cannot reach the 7.1 pp.
- I2 (and Auditor A's I4(a)): the memo recasts a research classification failure as 'taxpayers whom the IRS cannot actually identify' for the whole 20%. — Overstated as a misrepresentation of the source. Cooper et al. themselves write that 'even with administrative data, it is not possible to identify all of the partners' and that 'we are not able to find a tax return associated with the partner's taxpayer identification number' (p.103). For the 13.0-pp majority, the TIN is not recognized by the tax database at all. The point survives only for the 7.1 pp of recognized EINs, and is kept at minor severity as F6.
- Saving/consumption measurement (unscored by B): a $100 wealth increase can reflect $0 or $100 of saving, so purchase and balance data do not identify saving. — Not an error in the memo. Transaction-level purchase and sale records combined with balances separate net purchases from valuation changes for covered assets, which is all the memo claims ('would also allow to estimate savings and consumption'). Gifts, transfers between one's own accounts and non-financial assets are measurement challenges the memo does not deny.
Audit notes
What reproduced. The memo has no data, code or estimates of its own. Reproduced exactly: (1) Cooper et al. (2016): union $200bn/$671bn = 29.8%, unclassifiable 20.1% (Fig. 3B, 13.0 + 7.1, read from the rendered figure), circular 14.9%, overlap 5.2 pp, so 'another 10%' = 9.7 pp. (2) Auditor A's DFA bottom-99% composition, 2019Q4: DB 18.2%, unobserved 29.2%, or 25.1% of non-durable assets; the Sep-2020 vintage gives pension entitlements of 28.7%. (3) Auditor B's circular-ownership counterexample: det(I-M)=0.875; outside owners 57.14/14.29/28.57. (4) Balance-sheet coverage of at least 56.1% of corporations and partnerships (partnerships 75.8%, TY2017). (5) FY2019 partnership exams of 7,478 vs 22,417 circular partnerships (3.0x). Approximately reproduced: Form 1041 returns with income above $20k are 20.2-22.9% (2014) and 25.9-29.6% (TY2019) under Pareto interpolation, consistent with 'roughly 25%'; about 4,400 exchange-listed companies (CRS R46424). Not reproducible: the trust asset leg (Form 1041 has no balance sheet), the size of any enforcement or measurement gain, the DTC-eligible share of private-company stock, Schedule G coverage, and SSA race missingness by cohort. All 57 documentary checks in scripts/doc_checks.py found the quoted text in the cached primary sources.
Limits of this audit. The IRS microdata behind Cooper et al. (2016) were not accessible, so their published aggregates were taken as given and only the memo's use of them was tested. Not established: whether the IRS transcribes Schedule G or K-1 items H and I; the share of large private companies with DTC-eligible stock; SSA race missingness by birth cohort; and whether real K-1 networks meet the proportional-allocation conditions needed to resolve cycles. The DFA classification of assets as 'observed' under the proposals is approximate: some deposits and directly registered securities may escape the proposed returns, and DB wealth could be imputed. Legal checks use the TY2019 forms and instructions in force at the memo date. Later documents (DTCC FAQ 2021, DTC Operational Arrangements 2026, Treasury TP-11 2023, 88 FR 2023) were used only to confirm long-standing practice. ssa.gov returned HTTP 403, so SSA's 2016 research note was read from an August 2019 Wayback capture; it states the author's views. The memo is a three-page policy brief without estimates, so severity is judged against its qualitative claims.