How EquiReason produces and reviews research
This page states the whole method: what we read, how a market event becomes a measured trend, how a company's exposure is verified against its own reported revenue, what would falsify a conclusion, and the standards the published material is held to — including what the system cannot do.
The inputs
Every conclusion traces back to a named source. These are the inputs, in full — not a summary of them.
SEC filings
- 10-K annual reports
- 10-Q quarterly reports
- 8-K current reports
- Form 4 insider transactions
- 13F institutional holdings
- XBRL structured financial facts
Company disclosure
- Earnings call transcripts
- Reported segment, product and geographic revenue
- Management guidance
Market and positioning
- Daily price history
- Options surface and implied volatility
- Short interest
- Analyst estimate revisions
Macro
- Federal Reserve and FRED series
- Credit spreads
- Labour market indicators
- Sector capital flows
From event to company, in seven steps
01 · Collect
23 live inputs feed one research stream, continuously through the trading day rather than as a weekly snapshot. Filings arrive from SEC EDGAR directly; nothing in the fundamental layer depends on a third party's interpretation of a filing.
02 · Read the filing, not the headline
The extraction layer reads full 10-Ks, 10-Qs, 8-Ks and earnings transcripts — 5,700+ documents to date — looking for changes in demand, margin, guidance, capital expenditure, backlog and risk language. Language models structure the text. Rules, source data and the company's own reported numbers test whether the conclusion survives.
03 · Identify the mechanism
A trend is not a headline or a keyword. It is an economic mechanism — the causal path by which an event reaches an income statement, such as grid capital expenditure or AI compute demand. We track 107 of them. Because a trend is keyed to its mechanism rather than to whatever the press is calling it this month, its identity survives the churn in how it is described.
04 · Verify exposure against reported revenue
This is where most screens fail and where we spend the most effort. A company is not exposed to a trend because it mentioned it on an earnings call. Exposure is measured against segment, product and geographic revenue as reported in XBRL — the same numbers the company files with the SEC and its auditors sign.
- A claim traceable to audited segment revenue carries more weight than one inferred from a sector label.
- Where only a proxy exists, the page says so instead of presenting the proxy as proof.
- Sector membership alone is never treated as evidence of exposure.
05 · Corroborate before confirming
A mechanism moves from forming to confirmed on hard evidence across independent channels, not on narrative volume: how many companies show the same exposure, whether insiders are transacting, whether institutional holdings are shifting, whether analyst estimates are being revised, and whether the evidence comes from genuinely different sources rather than one story repeated.
06 · State what would break it
Every thesis is recorded with its falsifiers — the specific conditions that would invalidate it — before the outcome is known. Falsifiers are then tracked, and a thesis whose falsifiers fire is marked broken rather than quietly dropped. A claim published without a condition that could refute it is a story, not research.
07 · Gate, price and rank
6 hard gates run before anything carries conviction: is the company investable, is it financially sound, is the trend real, is the exposure verified, is the impact direct, and is the opportunity still unpriced. A separate multi-axis check — valuation, institutional positioning, expectations and price behaviour — asks whether the market has already paid for the idea, because being right about a trend that is fully priced is not an edge.
Editorial standards
We do not give investment advice
EquiReason is a research and decision-support system. Nothing it publishes is a recommendation to buy or sell any security, and nothing is tailored to an individual's circumstances. Where the system reports what other parties did — an insider transaction, an institutional position change — that is a statement of fact, not a suggestion to copy it.
Every number carries its period and its source
Financial figures are meaningless without knowing what window they cover. Each metric we display is defined with its period, its unit and how it was derived — trailing twelve months versus last completed fiscal year, reported versus estimated — so a reader can tell what they are looking at rather than inferring it.
Impossible values are treated as defects, not noise
A negative enterprise value, a net margin above 100 percent or a price/earnings ratio computed from negative earnings is a bug in the pipeline, not a finding about the company. Automated invariant checks run against the published data and a value that cannot be true is withheld rather than displayed.
Corrections are made at the source
When we find an error we fix the mechanism that produced it and let the correction propagate, rather than patching the single visible symptom. Where an error changed a published conclusion, the conclusion changes with it.
Machine-generated, human-governed
The analysis is produced by an automated system running continuously; it is not hand-written per company. What is human is the method: which evidence counts, where the thresholds sit, what disqualifies a thesis, and the review of the material published here.
What this method cannot do
A method is only credible if its limits are stated by the people who built it rather than discovered by the people relying on it. These are ours.
- Institutional holdings from 13F filings are disclosed up to 45 days after quarter end, so that channel is structurally backward-looking and is weighted accordingly.
- Segment revenue is only as granular as a company chooses to report it. Where a business does not break out the relevant line, exposure cannot be measured directly and is marked as a proxy.
- Detecting a real trend says nothing about timing. A mechanism can be correct and take several quarters to reach reported results.
- No system forecasts markets. The method is built to make reasoning auditable and to reject weak evidence, not to remove uncertainty.
Frequently asked questions
Where does EquiReason's data come from?
Company fundamentals, insider transactions and institutional holdings come from SEC EDGAR filings directly — 10-K, 10-Q, 8-K, Form 4, 13F and XBRL structured data. These are supplemented by earnings transcripts, macroeconomic series, price history, options positioning, short interest and analyst estimate revisions.
How is a company's exposure to a trend measured?
Against reported revenue, not narrative. EquiReason reads segment, product and geographic revenue as filed in XBRL and measures how much of a company's economics actually sits in the path of the mechanism. Mentioning a theme on an earnings call does not create exposure.
What is the difference between a forming and a confirmed trend?
A forming trend has early evidence across more than one company but has not met the confirmation bar. A confirmed trend is corroborated across independent channels: breadth across companies, insider activity, institutional flow, analyst revisions and source diversity. The two are shown separately so early signal is never presented as settled fact.
Does EquiReason give investment advice?
No. It is a research and decision-support tool. It does not make recommendations, does not account for an individual's circumstances, and does not place trades.
Are the conclusions written by a person or generated?
The analysis is generated by an automated system that runs continuously across the covered universe. The method behind it — what counts as evidence, where the thresholds sit, what disqualifies a thesis — is set and reviewed by people, as is the material published in our guides.