Valuation multiples
Shiller CAPE vs P/E ratio: 26 and 24
On 2026-09-13, 26 US listings carried a trailing P/E of 30 or more with a Shiller CAPE below 12, and 24 ran the other way. The ratio is inflation adjusted now.
By Javier SanzPublished Updated : Rewritten after the cyclically adjusted ratio became inflation adjusted on 2026-09-11 and began averaging loss-making years. The one-sided result this page reported was an artefact of the older method, and every count has been re-measured.
Founder of ValueScreener. Built Ninety Nine, a securities services provider, and sold it to a tier-one global bank; ran operations and internationalization at Alpaca, the brokerage API.
Who writes thisEditorial processLinkedIn (opens in a new tab)X (opens in a new tab)
On 2026-09-13, 26 US listings carried a trailing price to earnings ratio of 30 or more alongside a cyclically adjusted ratio below 12, and 24 ran the other way. Six days earlier the same two counts were 41 and 3.
The market did not become symmetrical in a week. The measure changed. Until 11 September the ten-year average behind the ratio was taken at face value and only the profitable years were counted. It is now restated into today's money with the consumer price index, and every filed year is averaged, losses included.
The two conditions, before and after
| Condition | 2026-09-07 | 2026-09-13 |
|---|---|---|
| Trailing P/E at or above 30, cyclically adjusted below 12 | 41 | 26 |
| Trailing P/E below 12, cyclically adjusted at or above 30 | 3 | 24 |
Loosen the thresholds and the same thing happens. Sixty-two listings sit at a trailing ratio below 15 with a ten-year ratio at or above 25, and 56 sit at a trailing ratio at or above 25 with a ten-year ratio below 15.
The first row is the trough case: last year's earnings were well below the decade's, so the one-year multiple is large and the ten-year multiple small. The second row is the peak case, and it is the one the old method could not see.
United Airlines is the whole story in one company
United Airlines trades at 10.64 times its last filed year. Its cyclically adjusted ratio is 30.19. It is in the second row of the table above: cheap on one year, expensive against ten.
Under the old method it read 13.45. Two things moved it. The losses of 2020 and 2021 are now in the average, which pulls the mean down, and the earnings of 2016 are now restated into 2026 dollars, which pulls them up relative to a face-value decade. Both push the ratio in the same direction.
Carnival is the other half of the correction. It used to publish a cyclically adjusted ratio of 6.83, which read as one of the cheapest companies in the market on a ten-year view. Its real ten-year average earnings are a loss. A price over a loss is not a multiple, so it now publishes nothing, with "the decade's real earnings average to a loss" recorded as the reason.
A measure that exists to catch a company at the top of its cycle was doing the opposite: dropping the loss years made every cyclical look cheap, which is precisely the case the ratio was invented for.
Where the 26 and the 24 sit
By sector, the 26 in the first row: consumer cyclical 7, consumer defensive 5, industrials 4, financial services 3, energy 3, and one each in technology, communication services, real estate and basic materials. Twelve are worth a billion dollars or more and one more than ten billion.
The 24 in the second row: energy 8, financial services 4, real estate 3, technology 3, industrials 3, and one each in utilities, consumer cyclical and communication services. Fifteen are worth a billion dollars or more.
Energy is the largest group in the peak-earnings row and the third largest in the trough row, which is what a cycle-sensitive measure applied to a cycle-sensitive industry should look like.
The measure is thinner than the market it describes
A ten-year average needs ten years of earnings. On 2026-09-13 the pool held verified filing history of ten years or more for 2,882 of its listings, and a cyclically adjusted ratio for 1,903, which is 34.5 percent of the 5,518 in the pool.
The three refusals are counted and named:
| Reason recorded | Listings |
|---|---|
| The decade's real earnings average to a loss or to nothing | 1,386 |
| The company reports in another currency | 8 |
| No consumer price index month to restate a year with | 9 |
The middle row is a limit worth stating. US consumer prices do not measure inflation in another currency, and restating a foreign filer's earnings with them would be a mixture of two measurements rather than one adjustment.
The bottom row used to be much larger. The index behind it is now read from the US Bureau of Labor Statistics and runs 151 months, from January 2014 to August 2026, so almost every ten-year window has a month for every year in it.
For comparison, the trailing ratio is held for 2,360 listings and the five-year normalised version for 1,962. All three thin out for the same underlying reason: averaging earnings requires the earnings to have been positive often enough to average. The trailing ratio has no average to hide a bad year inside, so one loss-making year removes it altogether rather than turning it negative, and even at the widest coverage of the three it is held for fewer than half the pool.
For comparison, the trailing ratio is held for 2,078 listings and the five-year normalised version for 1,771. All three thin out for the same underlying reason: a ten-year mean needs ten years on file.
What a ten-year per-share average is still sensitive to
One hazard survives the correction, because these two numbers cannot detect it.
A ten-year mean of earnings per share is a mean of a per-share figure, and the number of shares can change a great deal in ten years. Anything that alters the share count without altering the business alters every historical year in the average unless the history has been restated. A company that issued heavily, bought back heavily, or split its shares will show a divergence between its trailing and its cyclically adjusted ratio for a reason that has nothing to do with a cycle.
Nothing on this page separates that case from a genuine cyclical trough. The two figures look identical. Distinguishing them means opening the ten filings and reading the share count, which is the work the ratio was supposed to save and does not.
How to use the pair
Read them as a question, not a verdict. A large gap between the two says the last year was unlike the decade. Whether the decade or the last year is the better guide to the next one is the whole question and neither ratio answers it.
Check the share count before the cycle. It is the cheaper of the two checks and it rules out the artefact.
Do not screen on the ratio alone. It exists for 34.5 percent of the pool, and a quarter of the pool is refused because its real decade averages to a loss, which is information rather than absence.
What this leaves out
Every count is one reading of the production database on 2026-09-13, after the recompute that finished at 03:03 UTC, over the 5,518 active equity listings in the pool. The 2026-09-07 column is the reading this page carried before it was rewritten, taken on a pool of 5,257 listings against prices stamped 2026-08-26.
The inflation adjustment uses the US consumer price index, all items, seasonally adjusted, published by the US Bureau of Labor Statistics. The stored series begins in January 2014, so a fiscal year ending before that has no month to restate it with and the ratio is refused rather than approximated.
The counts are over stored values. A published figure can still be withheld at display time where the metric is not meaningful for that filer, and such a listing still answers a numeric filter.
The 2,882 figure is the count of listings with ten years or more of verified filing history on that morning. It is not a claim that all ten years contain a filed earnings per share line.
Counts are per listing. A company with two share classes appears twice.
Nothing here is a view on any company or on the market, United Airlines and Carnival included. A gap between two ratios is an arithmetic observation about two accounting periods.