Methodology
This page documents — in full — where the data comes from, how it is extracted and validated, and the editorial rules that govern how it is presented. The methodology is public and versioned; material changes are recorded here.
What this is
Trump Stock Tracker is a civic-transparency project. It compiles the President’s securities transactions as disclosed under the Ethics in Government Act of 1978 and the STOCK Act of 2012, and correlates them with his public statements and official government actions. It is a transparency tool — not financial advice, and not a legal determination of wrongdoing.
The OGE Form 278-T
Federal officials disclose securities transactions on OGE Form 278-T (Periodic Transaction Report), filed within 30 days of notification and no later than 45 days after the transaction. Each filing lists transactions with a description, a type (Purchase, Sale, Sale (Partial), or Exchange), a date, a late-notification flag, and an amount — reported as a statutory range, never an exact figure.
Amount bands
The form never reports a precise dollar amount. It uses these statutory value ranges. This site always displays the range; aggregate “estimated value” figures use band midpoints and are themselves shown as ranges.
| Band | Range |
|---|---|
| 1 | $1,001 – $15,000 |
| 2 | $15,001 – $50,000 |
| 3 | $50,001 – $100,000 |
| 4 | $100,001 – $250,000 |
| 5 | $250,001 – $500,000 |
| 6 | $500,001 – $1,000,000 |
| 7 | $1,000,001 – $5,000,000 |
| 8 | $5,000,001 – $25,000,000 |
| 9 | $25,000,001 – $50,000,000 |
| 10 | Over $50,000,000 |
Data sources
- Trades: the U.S. Office of Government Ethics (OGE) presidential disclosure system, and the White House disclosures pages. OGE is the source of record.
- Statements: the American Presidency Project and the federal Compilation of Presidential Documents (single-speaker official transcripts), plus Truth Social posts.
- Official actions: the Federal Register API (executive orders, proclamations, rules) and White House fact sheets, articles, and statements scraped from whitehouse.gov.
- Share prices: weekly end-of-day price history from Alpha Vantage and current quotes from Finnhub, used to chart each company and mark the President’s disclosed trades. Prices are non-blocking enrichment — never part of the disclosure record itself.
Extraction & validation
The 278-T filings are scanned documents; some carry a poor embedded OCR text layer and the newest carry none at all. A content-anchored extractor locates each transaction by its recognizable amount band and date rather than by fragile column geometry. Every filing then passes a validation gate — page-count reconciliation, transaction-type and amount-band enum checks, and date sanity. Each filing receives a parse-confidence score; filings below threshold are withheld from the public view and held for human review rather than published unverified. Where a filing scores low, an LLM adjudicator re-reads the source PDF under a strict schema.
Timing correlations — scoring model v1.2
Each trade is scored against contemporaneous statements and official actions found in an asymmetric window (45 days before to 30 days after the transaction date). An event qualifies as a candidate when it names the traded company directly, or when it addresses the company’s sub-industry (e.g. “semiconductors” for a chip maker). The composite 0–100 signal is a weighted sum of the components below — the exact weights the engine uses, rendered from the same source file. Every correlation displays its full component breakdown. The score is an analytical index, never a verdict.
| Component | Weight | Definition |
|---|---|---|
| Temporal proximity | 0.30 | How close the statement/action is to the transaction date: exp(−|days|/14). Same-day = 1.0, two weeks ≈ 0.37. |
| Entity specificity | 0.25 | 1.0 when the event names the traded company directly; 0.6 when it addresses the company's sub-industry (e.g. “semiconductors” for an NVIDIA trade). |
| Authority | 0.10 | The filer's policy power over the traded company — 1.0 for the President; recorded per filer so future filers (Cabinet, Congress) score lower. |
| Directional consistency | 0.10 | Whether the trade's direction aligns with the event's expected price impact: a purchase before favorable words/acts (or a sale before unfavorable ones) scores 1.0; the opposite pairing 0.0; alignment unknown 0.5. Statement direction comes from verified sentiment; a federal contract award counts as favorable. |
| Trade magnitude | 0.10 | Log-normalized midpoint of the disclosed amount band, 0..1 — a $5M trade signals more than a $15K one. |
| Corroboration | 0.15 | Bonus when several independent events cluster in the same window: (events − 1) × 0.25, capped at 1. |
Where direction comes from: statement direction is the speaker’s verified sentiment toward the company (extracted by the language-model mention layer and only stored when its verbatim quote is programmatically confirmed present in the source text); a federal contract award counts as favorable. Pairs without a verified direction score 0.5 — “unknown”, by definition, never a hidden default. Each correlation is additionally passed through a reasoning check that asks whether the matched text is genuinely about the company (and not a string coincidence like “Southern Co” vs. “the southern border”); pairs judged coincidental are removed from public view.
Scoring changelog
- v1.2 (June 2026) — Directional consistency reinstated as a real component (weight 0.10): trade direction vs. the event's expected price impact, derived from verified statement sentiment (LLM mention layer) and action type (contract awards count as favorable). Alignment unknown scores 0.5 by definition. Temporal proximity 0.35 → 0.30, authority 0.15 → 0.10.
- v1.1 (June 2026) — Directional consistency removed from the active model (it was a constant 0.5 placeholder, not real analysis — it returns when price-impact direction modelling lands). Entity specificity now varies: direct issuer mention 1.0, sub-industry topic match 0.6. Authority is read from the filer record instead of being hardcoded. Weights renormalized.
- v1.0 (May 2026) — Initial six-component model.
Editorial discipline
This project publishes verifiable facts and transparently-scored correlations. It does not assert that any trade was illegal, nor that it used nonpublic information. No evidence presented here establishes that any trade used nonpublic information; correlation is not causation; lawful trading is lawful. Amount ranges are statutory bands, not exact figures. Transaction dates may precede filing dates by 30–45 days, and filings may be amended.
Permitted use
Disclosure data is used solely for news and transparency dissemination to the general public — the use expressly permitted under 5 U.S.C. § 13107(c). It is not used for credit-rating, solicitation, or any other commercial purpose.
Attribution
Project by Benjamin Life (@omniharmonic). Scoring methodology version 1.2 (see changelog above).