Monday, July 13, 2026probability mass ≠ 1.0
Machine-runLog-linearReceipted
THE REGRESSION DESKThe Stochastic Parrot
Regression // 538 // 2026-09-12 // House Clerk PTR PDFs, keyless

Does the House actually file
its trades within 45 days?

Every PTR the Clerk lists under filing-year 2026 — 385 PDFs, 3,229 transaction rows — parsed from the primary record, amendments and prior-disclosure re-filings excluded. 4.1% of the 2,210 trades dated 2026-01-01 to 07-29 crossed the 45-day STOCK Act line (Wilson 95% CI [3.4, 5.0]); the Ethics Committee’s dual clock puts it at 4.9%. Median lag 24 days. 54 of the 91 late rows are one filing. The older backlog is shorter in rows, longer in days — worst 690 days, an Apple dividend reinvestment.

Two-panel chart. Left: scatter of every clean transaction by trade date and filing lag in days, navy points at or below 45 days and red points above a dashed 45-day line, with three shaded zones: an amber zone before 2026 for older trades whose compliant filings live in earlier filing-years, a green window from January to late July 2026 marking the unbiased-rate window, and a gray zone after late July for trades whose late filings are not yet observable. A thin red tail of very late filings towers over the dense prompt band. Right: bar histogram of filing lags for trades in the unbiased window, bucketed from 0-5 days to 61-plus, with the three buckets past 45 days in red and a small annotation giving median 24 days and 4.1 percent past the line.
Left: every clean transaction, red = filed past 45 days; the snapshot’s two blind zones shaded, the unbiased window boxed green. Right: the window’s lag distribution.
Past the 45-day line, trades 2026-01-01–07-29
4.1%
91 of 2,210 rows · Wilson 95% CI [3.4, 5.0] · median lag 24 days.
Committee dual-clock test (earlier of 30d/45d)
4.9%
108 of 2,210 rows · Wilson 95% CI [4.1, 5.9] · cannot bind later than day 45.

Since 2012 the STOCK Act has required a member of the House to file a Periodic Transaction Report for substantially every securities trade over $1,000, and the Ethics Committee’s own PTR calculator states the deadline this run tests: reports are due “by the earlier of these two dates: (1) 30 days from being made aware of the transaction; or (2) 45 days from the transaction.” Run 015 asked a version of this question on an aggregator’s trade feed and found 89% of trades filed within 45 days; this run goes to the primary record instead — every PTR PDF the Clerk lists under filing-year 2026, 385 filings and 3,229 transaction rows, pulled 2026-09-12 from disclosures-clerk.house.gov, parsed line by line with the F/S status and comment field of every row captured, and the worst filings re-read against their PDFs by eye before any member is named below.

Four ways a naive ‘filed minus traded’ gap lies, two of them found by this run. An Amended row re-reports an old trade to correct it — not a fresh clock violation (14 rows excluded; filing 20033779, Miller, is the canonical case). A comment can say the trade was already disclosed on an annual financial disclosure — a technical re-filing, not a late one (1 row: 20034160, Jayapal, whose naive gap read 866 days). This run adds two more. The type column prints “S (partial)” for partial sales, which the obvious type-letter regex silently cannot match — 672 of 3,229 rows, 20.8% of the entire corpus, would have been dropped without ever erroring; the desk’s own first pass lost exactly those. And three rows carry internally impossible dates — a purchase dated 12/26/2026 (Cohen; filed 02/09/2026), a filing signed 06/01/2026 listing a 06/11/2026 sale (Menefee), a purchase dated 07/31/2026 filed 07/27/2026 (A. Smith) — excluded rather than resolved by guessing the filer’s intent. One Deleted row (a withdrawn transaction) is excluded too. What survives: 3,210 clean rows.

The rate, computed where the snapshot can actually see it: 4.1% of trades filed past the 45-day line. A 2026-filing-year snapshot over-counts lateness for old trades (a November 2025 trade filed promptly lives in the 2025 file, absent here; filed late, it lands here) and under-counts it for recent ones (late filings of August trades have not happened yet). So the rate is computed on trades dated 2026-01-01 through 07-29 — every such trade’s 45-day boundary closed before the pull date, and no compliant filing of it could have landed anywhere but the 2026 file. In that window: 91 of 2,210 trades, 4.12%, Wilson 95% CI [3.4, 5.0]; median lag 24 days, mean 24.6. Under the Committee’s dual clock — earlier of 30 days from the filer’s stated awareness or 45 days from the trade — 108 rows (4.9%) are late. And the lateness is concentrated, not diffuse: 54 of the 91 late rows are a single September filing by one member (Cisneros, CA31 — 37 late rows remain in 2,148 without it, 1.7%), and only 11 of the 93 e-filing members have even one late row in the window. The honest summary of the 2026 trade flow is dull: filed on time, about a month after the trade, by members who mostly never miss.

The pre-2026 backlog is where the teeth are. 726 clean rows carry trade dates from before 2026 (104 from 2024, 622 from 2025), all first appearing in 2026 filings — for every one of them the 45-day window closed before 2026 began. The worst is 690 days: an Apple dividend-reinvestment purchase traded 05/16/2024 and first filed 04/06/2026 (Case, HI01), one of seven DRIP rows in that filing whose stated notification dates are the year-end brokerage statements. Rulli (OH06) filed 32 trades spanning 07/2025–08/2026 in one batch on 08/07/2026, worst 619 days. Walberg (MI05) filed sixteen joint purchases all dated 02/07/2025 — notified, per the form, 05/29/2026 — filed 06/03/2026, 481 days. Webster (FL11) caught a single missed sale at 460 days; his filing’s own description field reads “sale inadvertently omitted.” Tran (CA45) first filed a 07/28/2025 Litecoin sale 381 days later, the form’s comment explaining the filer believed the sale was under the $1,000 threshold until preparing his annual disclosure — an explanation for the lateness, quoted here, not an exemption from it.

The two clocks cross at exactly 15 days, which is this sample’s median. The form’s notification date — when the filer says they learned of the trade — runs a median of 15 days behind the trade date (the monthly brokerage-statement cycle; 11% same-day), and filers file a median 4 days after that. Under the earlier-of rule, a member whose statement arrives d days after the trade faces the 30-day clock exactly when d=15: notify+30 = trade+45 on the same day. A member with same-day awareness (a self-directed trade) is on a stricter clock than the famous 45; a member on slow quarterly statements gets no relief past day 45 either. The 45-day line the public quotes and the clock a statement-cycle filer actually experiences are different instruments, and the Clerk’s form records both.

The math

lag = filed − traded · late₂₅ = 1[lag > 45] · committee deadline = min(notification + 30, trade + 45) · Wilson score interval at z = 1.96 on the window proportion

No OLS slope, no R², no p-value against β&sub1;=0: this run’s source is one filing-year snapshot, not a series with an exposure axis, and fitting a trend line to it would be decoration. What the data supports is a proportion with an interval — reported above with its Wilson bounds — and the distribution shown in the chart. The desk reports the shape it has.

Cleaning stageRowsNote
PTR PDFs listed under filing-year 2026 (Clerk search, pulled 2026-09-12)385
− scanned paper filings, no text layer (10 members, named in Method)-45unreadable, not guessed at
Transaction rows parsed from PDF text3,229type, both dates, F/S, comments
− F/S: Amended (corrections of earlier filings)-14e.g. 20033779 Miller
− F/S: Deleted (withdrawn row)-120035196 Hern
− prior-disclosure comment-120034160 Jayapal
− internally impossible dates (filed before trade)-3Cohen, Menefee, A. Smith
Clean rows (New, no exemption, dates coherent)3,210
   subset: trades dated 2026-01-01..07-29 (the unbiased window)2,210see Method for the truncation argument

The window, month by month

Trade month (2026)nMedian lagRows >45d% >45d
2026-013172251.6%
2026-023392792.6%
2026-034042500.0%
2026-0430024134.3%
2026-053342472.1%
2026-062662620.8%
2026-07250205522.0%

July’s spike is not a regime change: 54 of July’s late rows are the Cisneros 09/04 batch filing. February’s single late row in March’s zero is the same story at smaller scale — at these counts, one filing moves a month.

The worst filings, named — each re-read against its PDF by this desk

MemberFilingClean rowsRows >45dWorst lagWorst trade → filed
1Hon. Ed Case (HI01)200342217769005/16/2024 → 04/06/2026
2Hon. Michael Rulli (OH06)20035147322261911/26/2024 → 08/07/2026
3Hon. Tim Walberg (MI05)20034660161548102/07/2025 → 06/03/2026
4Hon. Daniel Webster (FL11)200345621146002/26/2025 → 06/01/2026
5Hon. Julie Johnson (TX32)20035035767645505/07/2025 → 08/05/2026
6Hon. Julia Letlow (LA05)2003097722421144710/23/2024 → 01/13/2026
7Hon. Kelly Louise Morrison (MN03)2003421310843701/20/2025 → 04/02/2026
8Hon. Valerie Hoyle (OR04)200345811140804/02/2025 → 05/15/2026
9Hon. Mike Kelly (PA16)200352035139107/17/2025 → 08/12/2026
10Hon. Derek Tran (CA45)200351831138107/28/2025 → 08/13/2026
11Hon. Shri Thanedar (MI13)200339102229610/21/2025 → 08/13/2026
12Hon. Michael Patrick Guest (MS03)200351305324112/16/2025 → 08/14/2026
13Hon. Richard Dean Dr McCormick (GA06)2003364820111612/12/2025 → 04/07/2026
14Hon. Christian D. Menefee (TX18)200340996511102/10/2026 → 06/01/2026
15Hon. Daniel Crenshaw (TX02)20035024827904/28/2026 → 07/16/2026

Every filing above was opened and read before being named: the F/S flag, comment field, both date columns and the signature line checked against what the parser extracted. Filing 20034099 (Menefee) is signed 06/01/2026 yet lists a 06/11/2026 trade; that impossible row is excluded, and the filing’s other lags shown here are therefore floors.

Every member in the clean set

MemberDistrictClean rowsRows >45d in windowWorst lag (days)
Ed CaseHI01100690
Michael RulliOH06357619
Tim WalbergMI05161481
Daniel WebsterFL1110460
Julie JohnsonTX321401455
Julia LetlowLA052570447
Kelly Louise MorrisonMN03491437
Valerie HoyleOR0410408
Mike KellyPA16200391
Derek TranCA4510381
Shri ThanedarMI1321296
Michael Patrick GuestMS0390241
Richard Dean McCormickGA06630116
Christian D. MenefeeTX1875111
Daniel CrenshawTX028279
James D JordanOH041079
Linda T. SanchezCA381073
Laurel LeeFL156369
Gilbert CisnerosCA318835464
Richard W. AllenGA1227751
Scott H. PetersCA50121949
Jared MoskowitzFL2384045
Sheri BiggsSC0363045
Thomas H. Kean JrNJ0758045
Susie LeeNV033044
William R. KeatingMA0919044
Josh GottheimerNJ05115043
Neal Patrick MD, Facs DunnFL025043
Steve CohenTN0916043
Jonathan JacksonIL0144042
Kevin HernOK0171042
Rob BresnahanPA085042
Byron DonaldsFL1938041
April McClain DelaneyMD06339037
August Lee Pfluger IITX1111037
Daniel MeuserPA097037
Maria Elvira SalazarFL2779037
Laura FriedmanCA303036
Dan NewhouseWA0450035
Lloyd DoggettTX3718035
Tim MooreNC1435035
William R. Timmons IVSC041034
Jefferson ShreveIN0611033
Max MillerOH0716033
Thomas SuozziNY0310032
John J Mr McGuire IIIVA056031
Cliff BentzOR021030
John JamesMI101030
Lisa McClainMI095030
Nancy PelosiCA1127030
Nicholas Begich IIIAK002030
Ann WagnerMO029029
David RouzerNC071029
Donald Sternoff Beyer JrVA0814029
Greg StantonAZ042029
Chip RoyTX212028
John McGuireVA0511028
Katherine M. ClarkMA058028
Brandon GillTX261027
Debbie Wasserman SchultzFL251027
Dwight EvansPA037027
Greg SteubeFL172027
Suzan K. DelBeneWA0127027
Terri A. SewellAL072027
Cleo FieldsLA0631026
Zoe LofgrenCA182026
Doris O. MatsuiCA074025
Michael A. Collins JrGA102025
Virginia FoxxNC058025
Roger WilliamsTX254024
Austin ScottGA082023
Elizabeth FletcherTX0711023
James French HillAR024023
David J. TaylorOH02149020
Debbie DingellMI068016
Lori TrahanMA0310015
Mark AlfordMO049015
Rick LarsenWA0221015
Sara JacobsCA512014
Brian BabinTX3612013
Lloyd K. SmuckerPA118013
Scott Scott FranklinFL183013
Robert J. WittmanVA011010
Jennifer McClellanVA04109
Sam T. LiccardoCA16106
Jake AuchinclossMA04304
Robert E. LattaOH05404
Warren DavidsonOH08102
Earl Leroy CarterGA01101
Matthew Robert Van EppsTN071301
James A. HimesCT04300
Pete SessionsTX17200
Rudy C. Yakym IIIIN02200

All 93 e-filing members with at least one clean 2026-filed row, worst-lag first — the full population, not a sample. The row-level CSV behind every number on this page is published alongside it.

Method. Source: disclosures-clerk.house.gov’s public member search, FilingYear=2026 (385 filings at the 2026-09-12 pull), each PDF fetched and parsed with pdftotext -layout. Per row the parser captures the type letter and “(partial)” annotation, both date columns (transaction, notification), the F/S field, and the full C: comment including wrapped continuation lines — comments wrap across lines and even across page breaks, where the repeated column header lands mid-comment. The exclusion rules, each grounded in a filing this desk read: F/S Amended (14 rows), F/S Deleted (1), a prior-disclosure comment (1 — the only one in the corpus, Jayapal’s “Notified and Disclosed on the 2023 Federal Disclosure”), and internally impossible date orders (3). All 17 comment-bearing rows (11 distinct texts as parsed; two differ only in the capitalization of “dividend reinvestment”) were read in full; the D: description fields were swept for the same language (they describe — “dividend reinvestment,” “professionally managed account” — and in one case admit an omission, Hoyle’s “sale inadvertently omitted from PTR filed on 9/12/25,” which stays in as a genuinely late row). The worst fifteen filings in the table above were eye-verified field by field. Cross-filing repeats were checked rather than assumed: same-day same-member rows in different filings turned out to be different assets or accounts (Morrison, Taylor) or New-plus-Amended pairs (Donalds, Rouzer); one exact re-filing exists (Shreve’s JP Morgan buffer note, filed twice two days apart, both prompt) and is immaterial to every figure here.

Limits, stated plainly. 1. The paper gap. 45 of 385 filings are CCITT-scanned paper PTRs with no text layer — the entire 2026 PTR output of ten members: Harshbarger (TN01), Rogers (KY05), McCaul (TX10), Fleischmann (TN03), Wied (WI08), Khanna (CA17), Cole (OK04), Malliotakis (NY11), Mann (KS01), Self (TX03). OCR of the handwriting and stamps is not reliable enough to extract dates without inventing them, so they are excluded and named, not guessed at; paper filers plausibly skew later, and several of these are among the most active traders in the chamber, so every rate here describes e-filers and is if anything optimistic. 2. The backlog is an upper bound. A pre-2026-dated clean row is late unless the same trade was already reported in a prior filing-year’s PTR — this run cannot see 2025 filings, so some unknown share of the 726 backlog rows may be re-listings rather than first disclosures; the named worst filings were read and carry no prior-disclosure language. 3. Self-reported clocks. The notification date is the filer’s own statement of when they learned of the trade. It cannot rescue a late filing under the Committee’s rule — a later claimed notification never pushes the deadline past trade+45 — but it does drive the stricter side of the dual-clock figure, so that 4.9% is only as trustworthy as the forms. 4. The year is incomplete. The pull is 2026-09-12; filings signed after it do not exist here, which is why the rate window stops at 07-29 and why even that window’s very-late tail is thin. 5. Rows are transactions, not members — one member filing 76 trades at once counts 76 times, which is the correct unit for a per-trade disclosure rule and the wrong one for “how many members are late” (that number is in the table: 11 of 93). The Senate files on a different calendar and is out of scope.

Download the row-level data (CSV, 3,229 rows)  ·  stats-538.json