Stock Research from SEC Filings and Earnings Calls: Why Primary Sources Matter

What counts as a primary source

A primary source in stock research is a document produced by the company itself under disclosure obligations, or raw market data recorded as it happened. For US-listed companies, that means SEC filings — the 10-K, 10-Q, 8-K, proxy statement, and their relatives — plus earnings call transcripts, investor presentations the company publishes, and the prices and volumes the market actually prints. Everything else — news articles, analyst notes, forum threads, summaries of summaries — is commentary written about those documents.

The distinction matters because of who bears the consequences of being wrong. A company that materially misstates its filings faces regulatory and legal liability; a columnist who mischaracterizes a quarter faces, at most, a correction. That asymmetry is why the source hierarchy in serious research runs one way: filings and company disclosures establish the facts; commentary explains context around those facts. Commentary is genuinely useful — it can surface what to look for, supply industry background, and challenge a narrative — but when a claim in an article conflicts with a claim in a filing, the filing wins until proven otherwise.

The principle is US-framed here because SEC disclosure is unusually standardized, but it generalizes: most developed markets require comparable periodic reports, and the discipline — go to what the company was obligated to disclose, not what someone wrote about it — travels everywhere.

What the major filings tell you

You do not need to memorize the SEC’s form catalog. Four documents cover most of what an individual investor researches, and each answers a different question.

The 10-K: the annual full picture

The 10-K is the company’s annual report to regulators — the single most complete document a company produces about itself. It describes the business in the company’s own words, lists the risk factors management is obligated to disclose, and contains the audited financial statements with their footnotes. An investor uses the 10-K to answer the foundational questions: what does this company actually sell, to whom, what does it earn, what does it owe, and what does management itself say could go wrong. When you research a company for the first time, the most recent 10-K is where you start.

The 10-Q: the quarterly pulse

The 10-Q is the quarterly version — unaudited, lighter on narrative, but current. Its job in a research workflow is trend detection: is revenue growth accelerating or fading, are margins holding, is the balance sheet drifting, did a new risk disclosure appear that wasn’t in the last 10-K. A single 10-Q rarely settles anything; a sequence of them shows you the direction of travel between annual reports.

The 8-K: material events as they happen

An 8-K is filed when something material happens between scheduled reports — an acquisition, an executive departure, a debt raise, a restatement, a major contract. Investors use 8-Ks as the event stream: if a stock moves and you want to know what the company actually said rather than how headlines characterized it, the 8-K (and its exhibits, often including the press release itself) is the record. Reading the 8-K instead of the headline about the 8-K is the cheapest upgrade available in stock research.

The proxy statement: pay, governance, and incentives

The proxy statement (form DEF 14A) discloses executive compensation, board composition, insider ownership, and shareholder proposals. Its research use is incentives: it tells you what management is actually paid to do, which may differ from what the earnings narrative implies. If executives are compensated on revenue growth while telling investors the priority is profitability, the proxy is where that tension becomes visible. It is the most skipped major filing and often the most revealing.

What earnings calls add

Filings tell you what happened. Earnings calls reveal how management frames it — and framing, tracked over time, is information.

The prepared remarks at the start of a call are largely a narrated press release; the useful signal concentrates later. Guidance is the closest thing to a forward-looking commitment a company makes, and the gap between guidance and subsequent results is a running score on management’s credibility. Tone shifts matter too: a business described as “sold out” one quarter and “demand-normalizing” the next has told you something no single filing states outright.

The Q&A is the least scripted part of any company’s public communication, which is why it repays the closest listening. Watch for questions that get answered with a different question’s answer, precision that appears and disappears — management quoting exact figures when the news is good and gesturing at “trends” when it isn’t — and analyst questions that repeat quarter after quarter because they never get a straight response. An evasive answer is not proof of a problem, but a pattern of evasion on the same topic usually precedes the filing that explains it.

Reading them without drowning

Honesty about the volume problem: a 10-K routinely runs a couple hundred pages, and a thorough research process wants several years of them plus quarterly reports and call transcripts. Almost no individual investor reads all of that, and pretending otherwise leads to reading none of it. The realistic skill is prioritization.

A few sections repay attention far out of proportion to their length:

• Risk factors — not the boilerplate (every company discloses that markets fluctuate) but the specific ones, and especially the new ones. Companies add risk factors when lawyers insist, and lawyers insist when something changed.
• Footnotes to the financial statements — where revenue recognition policies, off-balance-sheet obligations, and accounting changes live. The income statement is the summary; the footnotes are the terms and conditions.
• Share count — diluted shares outstanding, tracked over years. A company can grow net income while shareholders’ per-share claim on it shrinks. The share count is where that shows up.
• Segment trends — which parts of the business are actually growing versus which parts the narrative emphasizes.

The single highest-leverage technique costs nothing extra: read the same section across multiple years and note what changed. Risk factors that appeared, guidance language that softened, a segment quietly reclassified. Any one year’s filing is a snapshot; the diff between years is the story.

Where AI genuinely helps

AI’s real edge in this workflow is reading breadth. Covering five years of 10-Ks, twenty quarters of 10-Qs, and every earnings call transcript is exactly the kind of high-volume, consistency-demanding reading that fatigues a human and doesn’t fatigue a machine. Done properly, that breadth enables the cross-check most investors never have time for: does the story management tells match the numbers the same company reports — does “record demand” show up in receivables and inventory, does “disciplined capital allocation” show up in the share count.

The failure mode is equally real. A language model can produce a fluent, confident summary that is subtly wrong — a stale figure presented as current, a paraphrase that flips a qualifier — and fluency gives a reader no way to catch it. The discipline that keeps AI-assisted research honest is claim-level citations: every material claim traced back to the specific underlying document, so a skeptical reader can open the filing and verify the sentence rather than trusting the summary. What that standard looks like across the whole category is covered in how AI stock research should work; AI reading of filings without that traceability is commentary with better grammar.

How Monsaic applies the source hierarchy

Monsaic is an AI stock research platform, and its reports apply the hierarchy this page describes: filings and company-disclosed data are the baseline for factual claims, commentary is used for context rather than facts, and every report ships with its source list — claim-level citations a reader can follow back to the underlying documents. Each report also carries an explicit analysis date and price-as-of context, because a sourced claim is only as good as its timestamp.

The same primary-source discipline feeds the report’s forensic checks — grades on management and governance, dilution, and revenue quality, and a narrative-versus-substance score comparing what the company says against what it reports. In one Monsaic example analysis of NVIDIA Corporation (NVDA), the kill criteria — the pre-stated conditions that would break the thesis, such as gross margin structurally falling below 68% without a clear mix-transition explanation — are all checkable against future filings and earnings calls, which is the point: a thesis anchored to primary sources can be proven wrong by them. That is an illustration of the method, not investment advice. The full research philosophy, including how sourcing, scenarios, and grades fit together, is laid out in the Monsaic methodology.

FAQ

See filings become a verdict

This page argues that primary sources should carry the weight. A covered stock’s excerpt shows the result — a thesis built from filings, transcripts, and market data, with the full report’s source count behind it.

Browse covered stocksSee pricing

Keep reading

• AI stock research, done rigorously — the Monsaic pillar
• The Monsaic methodology
• The Monsaic source hierarchy
• How AI stock research works
• Generate your first Monsaic