Trump Trades Data: Methodology & Disclosure Brief

TREMOR's Trump Trades tool makes every transaction in President Donald J. Trump's recent federal financial-disclosure trade reports searchable. This page documents the methodology - the source filing, how it was extracted, how amounts and companies are handled, and the known caveats - plus a pointer to the companion analysis brief.

Open the interactive Trump Trades tool on TREMOR to search all 8,374 transactions from five filings.

Source

US Office of Government Ethics (OGE) Form 278-T (Periodic Transaction Report) filed by Donald J. Trump (President of the United States of America). Each filing discloses purchases, sales, or exchanges of securities greater than $1,000 made on behalf of the filer, the filer's spouse, or dependent children. The transaction-level pages of the tool cover five filings, each a scanned PDF on the OGE site:

The cover page of the first June 25 filing carries the OGE note "Filer paid late fee". The 14 earlier 278-T filings (August 2025 to May 2026, mostly bonds) are summarised on the tool's Filings tab and are not part of the transaction-level pages.

Reporting period

Trades dated 2026-01-06 to 2026-07-31. Total transactions extracted: 8,374 across 263 pages - 4,518 purchases and 3,856 sales.

A note on filing timing - this is NOT a quarterly disclosure

The OGE Form 278-T is a Periodic Transaction Report required by the STOCK Act of 2012. It is NOT a calendar-quarter disclosure. Each individual trade must be reported within 30 days of the filer becoming aware of the trade, or within 45 days of the trade itself, whichever is later.

As a result, a single filing typically bundles many trades that happened over the preceding weeks or months. The "reporting period" of a filing is simply the date range of the trades inside it - not a clean quarterly cut. The May 8 filing, for example, holds trades from January to March 2026; each later filing covers roughly one calendar month of trades.

Bottom line: filings are inherently lagged. By the time the public sees these trades, the market has already moved. The tool's S&P 500 view shows the index movement during the trade period and marks the date OGE posted each filing, so you can see what the market did before vs. after each disclosure went public.

Extraction methodology

May 8, 2026 filing

Extraction used a four-layer multi-method approach: LLM vision models, specialized document extractors, and sub-agent orchestration for vision extraction. All four independent readings of each row (description, type, date, amount range) were compared against one another.

Where the four agreed, the value was accepted as-is. Where they disagreed, a judgment call was taken either on a majority basis (at least three methods read the same way) or case-by-case for the harder cases. Approximately 50 rows were identified for human visual review against the source PDF, alongside spot-check sampling across the remaining rows.

June to September 2026 filings

Every page of the four later filings (4,732 transactions) was read by three independent methods: a vision LLM, a specialized document extractor, and vision sub-agents working from full-resolution crops. Each cell takes the value at least two of the three agree on. All three agreed on 28,329 of the 28,392 cells; one cell with no majority was read by hand from the page.

Further checks: row counts match the last row number printed in each filing; every amount is a standard OGE bucket; no trade dates fall on weekends or market holidays; no trade is dated after the filing was signed; no overlap between filings; and twelve random pages (396 rows) were re-read in full, four by hand and eight by a separate vision model, with no differences.

Value field

The OGE form discloses an amount range, not a single value (e.g. "$1,000,001 - $5,000,000"). We expose three sortable numeric forms: lower bound, upper bound, and integer midpoint (the default for sort and rollup). Tables always display the raw range as the primary form so the underlying disclosure stays visible.

Amount buckets used in the form

• $1,001 - $15,000
• $15,001 - $50,000
• $50,001 - $100,000
• $100,001 - $250,000
• $250,001 - $500,000
• $500,001 - $1,000,000
• $1,000,001 - $5,000,000
• $5,000,001 - $25,000,000
• $25,000,001 - $50,000,000
• $50,000,001 - $100,000,000

Company rollup & ticker standardization

Companies are derived by normalising the security description field: stripping broker-disclosure boilerplate ("UNSOLICITED", "Average Unit Price Transaction", "Your Broker Acted As Agent"), terminal token suffixes ("COM", "NEW", "SHS", "ADR", "REIT"), and trailing share-class indicators ("CL A", "CLASS B"). Multiple share classes of the same issuer collapse to a single company key.

The standardized view maps each normalized company key to a stock ticker so clerical variants of the same security (e.g. "AMAZON COM INC" and "AMAZON COM INC COM") roll up to one row under ticker AMZN, while real share-class distinctions (GOOGL vs GOOG) are preserved. The mapping is a judgmental ticker-mapping pass - each unique company key assigned a most-likely ticker, a confidence level (high / medium / low / no_ticker), and a note where the call was non-obvious. A raw view bypasses the mapping for one row per distinct description.

Known caveat

The OGE Form 278-T as submitted is a scanned PDF with imperfect OCR. Some security descriptions may be marginally wrong. The cross-section multi-layer validation should have caught most of these, but a small residual is possible. The numeric fields (date, transaction type, amount range) are inherently more robust because they live in a small bounded vocabulary with additional internal-consistency checks. The description column is where any residual noise would surface.

Analysis brief

A companion 10-slide brief on this filing - what is in it, what the press flagged, and the headline numbers - is published on the main TIGZIG site as a free downloadable PDF plus a blog write-up: Trump Trades: from mostly bonds to 3,642 equity trades in one filing. For nine months the filings showed mostly bonds; then one filing came in with 3,642 equity trades ($220M to $750M by bucket-midpoint range), with heavy activity in Nvidia, Apple, Microsoft, Amazon, Meta, Oracle and other big-tech names. The deck covers the interactive tool, the bonds-to-equities pivot, top buys, top sells, the volume-from-bucket-midpoints nuance, and the press cross-validation.