What the Largest US Companies Actually Earn, and What You Pay For It
Built entirely from our own EDGAR extraction. Every share count comes from a filing cover page where one exists — multi-class issuers tag it per class where the JSON API cannot read it, so a minority fall back to the weighted-average diluted count — and every earnings figure from the statement that reported it, with each quarter entering the series on the date it was first filed, not the date it ended. No index vendor, no smoothed estimates, no forward numbers.
Latest read
At October 9, 2026 the 326 largest US companies in our financials base trade at 28.1× trailing earnings: $66.0T of market value against $2.35T of net income over the last four reported quarters. That is elevated, and short of extreme. It sits at the 79th percentile of the 11.6×–48.2× range this basket has recorded since 2011, well below the 48.2× reached against the pandemic earnings trough. Against the recent past it is not elevated at all: the five-year median is 27.9×, so the basket is priced where it has been priced since 2022, and the percentile is a statement about the 2010s. The move is broad. The median profitable company sits at 25.5×, close to the cap-weighted figure, so a few mega-caps are not carrying it — though the median is taken over the 308 companies with positive earnings, since the 18 loss-makers have no finite multiple to rank. Read the series as a cross-section of today's large caps, not as the market's valuation history: the basket is built from today's winners, which flatters every earlier reading. One reading here is not merely elevated. Inverted, the basket yields 3.56% against a 10-year note at 5.22%, so a dollar of market value now earns 1.66 points less than a Treasury pays, against +3.46 points in 2011. The basket covers 77% of the market value of US filers, so it is most of the reachable market rather than all of it.
Sources, methodology & freshnessSEC EDGAR XBRL filings for the covered companies (net income and revenue from the statements, share counts from filing cover pages), joined to our daily price data and the Treasury 10-year yield · Rebuilt every day in the pipeline against that day's closes; the earnings side changes as each 10-Q and 10-K landsData as of 2026-10-09 · Open ↓Close ↑
$66.0T of market value against $2.35T of net income over the last four filed quarters. The median company reads 25.5×, so the level is broad. The peak is an earnings trough, not a valuation extreme.
Aggregate P/E since 2011
Why this differs from a published P/E
- It is not the S&P 500. It is the 334 companies we hold full filing histories for. That basket is larger and more concentrated than the index, so the multiple runs higher.
- Trailing, never forward. Only quarters that have actually been filed. No analyst estimates and no operating-earnings adjustments, so write-downs stay in.
- Loss-makers subtract. Earnings are summed across the basket, so a large loss lifts the multiple, exactly as it does in an index-level figure.
How much of the re-rating is composition
The median company, equal weight
Concentration
Earnings yield against the 10-year
Largest holdings at October 9, 2026
Method
$66.0T of $85.6T across 5,513 issuers. The denominator starts from $103.3T of US-listed value and makes two corrections. Share classes are counted once: the universe file carries the whole issuer's value on each class row, so GOOG and GOOGL both read about $4.0T and summing them would double-count Alphabet. And $13.3T of foreign filers is removed, because companies reporting under IFRS on a 20-F or 40-F can never enter a base this extractor builds from US GAAP. That list is hand-maintained and thin in the small-cap tail, which understates coverage slightly.
Each quarter enters on first_filed — when the market first saw it. Values are latest-vintage (restated) placed at their original reporting date.
Cover-page count (dei:EntityCommonStockSharesOutstanding) where tagged, falling back to weighted-average diluted for multi-class issuers.
Trailing four quarters of net income, summed across the basket. Loss-makers subtract, as they do in an index-level multiple.
The basket is today's largest US companies, so earlier readings omit both the failures of that era and the high-multiple companies that listed later. Both omissions bias the historical multiple down and overstate the re-rating. The fixed cohort isolates that effect.
$66.0T of market value against $2.35T of net income over the last four filed quarters. The median company reads 25.5×, so the level is broad. The peak is an earnings trough, not a valuation extreme.
Why this differs from a published P/E
- It is not the S&P 500. It is the 334 companies we hold full filing histories for. That basket is larger and more concentrated than the index, so the multiple runs higher.
- Trailing, never forward. Only quarters that have actually been filed. No analyst estimates and no operating-earnings adjustments, so write-downs stay in.
- Loss-makers subtract. Earnings are summed across the basket, so a large loss lifts the multiple, exactly as it does in an index-level figure.
How Market P/E Ratio Works
- 1Read earnings from the filings, not from estimatesNet income comes from each company's own 10-Q and 10-K as tagged in XBRL, summed across the basket over the trailing four reported quarters. There are no analyst forecasts and no operating-earnings adjustments, so write-downs and one-off charges stay in exactly as filed.
- 2Price the basket with the shares actually outstandingMarket value uses the cover-page share count from each filing, which is a point-in-time count of shares in issue and the only share series that exists for all four quarters. Multi-class issuers tag it per class, so those fall back to the weighted-average diluted count.
- 3Date every quarter by when it was first filedA quarter enters the series on the day it was first reported, never on the day it ended, so the timing carries no look-ahead. The values are latest-vintage, so a figure later restated appears here as restated at its original filing date — a restated history dated by first disclosure rather than a point-in-time series. Dating by period end would pretend the market knew a December quarter on December 31 when it was filed five weeks later, and would put a bend in every history that crosses an earnings season.
- 4Divide the whole by the wholeThe ratio is total market value over total trailing earnings, so a large loss lifts the multiple exactly as it does in any index-level figure. The median company is published beside it as the check on whether a few names are carrying the aggregate — taken over the profitable companies only, because a loss-maker has no finite multiple to rank.