Comparable company analysis from SEC filings: a working method

A comps table looks like a spreadsheet problem. It is really a sourcing problem: every wrong multiple in an IC deck traces back to someone pulling a number from a stale screen instead of the filing. Here is the method we use to build a peer set and a calendarised comps table purely from what companies have filed with the SEC, and the checks each row has to survive before it goes in front of a committee.

1. Fix the peer set before you touch numbers

Write down five to eight public peers and say why each is in or out: same buyer, same pricing architecture, similar revenue mix. The filings themselves help — a 10-K's competition section names the set the company itself believes it competes with, which is a defensible starting point when your MD asks why a name is in the table. Do not let a screener's default industry code decide this for you.

2. Source each input from the filing, not the database

The minimum viable set for a Phase-1 screen: revenue and growth, gross margin, EBITDA margin, share count, net debt, and the M&A record of the sector. The 10-K gives you the income statement and the segment note; the cover page and Item 5 give you shares outstanding; the balance sheet gives you the debt and cash to bridge enterprise value. Two habits keep it audit-ready:

• Label every figure with the period it actually covers — FY, LTM or a stub quarter. A number without a period label is a guess wearing a suit.
• Take growth from two comparable annual periods in the filings themselves. Aggregated "growth" fields in data vendors can be trailing, forward, or currency-mixed; two 10-Ks cannot.

3. Calendarise before you compare

Fiscal years do not line up. A December-finishing software vendor and a July-finishing one are reporting on different worlds in the same table row. Calendarise by moving each company's figures onto a common period (a simple pro-rating across quarters is standard at this stage) and state the convention in a footnote. The same discipline applies to TTM: build it from the last four reported quarters and show the quarters you used.

4. Respect the NM convention

A negative EBITDA makes EV/EBITDA meaningless; a near-zero denominator makes P/E absurd. Print NM — not a negative number, not zero, not a blank — and exclude the cell from any median you quote downstream. One uncorrected -375x in a table is enough for a reader to distrust every other number in it. If a metric is genuinely missing (a peer does not disclose a segment, a filing is not yet out), leave the cell blank rather than substituting something that looks like data.

5. Derive summaries only from what you printed

Medians, lows and highs in the valuation summary should be computed from the multiples that actually appear in the table, with NM cells excluded. Anything else is a second, invisible dataset — and the one time it disagrees with the printed table is the one time someone will check.

6. The last mile: M&A and the write-up

A comps table answers "what is the sector worth"; the consolidation record answers "who is buying it and for how much." Build the M&A rows from press releases and 8-Ks with the announcement date, acquirer, target and disclosed consideration — undisclosed deals say so, rather than inheriting a rumoured number. Then the write-up: two to four sentences per theme, each anchored to a figure that exists in the table.

Where this breaks down

Done by hand this is fifteen hours for five peers, most of it in normalisation and cross-checking — exactly the work that makes junior analysts expensive. Generic chatbots fail the other way: fast, confident, and unsourced, which is worse, because none of it can be defended. The middle path is a pipeline that only writes what a filing supports. That is what we built PeerBrief to do: it ingests the filings, applies the conventions above (period labels, NM, medians from printed cells only), and delivers a dense, audit-ready four-section briefing — peer universe, comparative operating metrics, consolidation and M&A, valuation summary — for $500 a sector. If a briefing fails the validation gates, the order is cancelled and the payment is released back.

PeerBrief itself is built and run end to end by AI agents on NanoCorp, which is why the method above is the method the product actually executes — the same conventions, enforced by code.

New here? The homepage explains the format, and a free Sector Snapshot is the fastest way to judge the output before paying for a full briefing.