Lesson 2 of 6
What is a good P/E ratio?
No P/E is good on its own. Turn it into a yield, place it in the market's own distribution, and check which earnings it divides by.
By Javier Sanz
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)
In one minute
- A P/E is the price of one dollar of last year's profit. No level of it is good on its own.
- Turned over, it is a yield: 20x is 5 percent of the price, 40x is 2.5 percent.
- The E is a stack of choices, and each one gives the same company a different ratio.
No P/E is good in general. The ratio tells you what you pay for one dollar of last year's profit. Whether that is a lot depends on what the profit does next, and the ratio does not contain that.
Where Graham's 15x sits among US P/E ratios
Graham's ceiling, 15x: between the 25th percentile (13.61x) and the median (21.16x).As a ceiling it admits between a quarter and a half of the cohort.
- 10th percentile: 9.07x
- 25th percentile: 13.61x
- median: 21.16x
- 75th percentile: 36.63x
- 90th percentile: 72.49x
Turn it over
P/E equals Share price divided by Earnings per share.
Example$50 ÷ $2.50 = 20x
Now flip it. One divided by the P/E is the earnings yield: last year's profit as a percentage of today's price. In that form you can set it beside a bond yield, another company, or the same company ten years ago.
The same multiple, read as a yield
- 10x10.0%
- 15x6.7%
- 20x5.0%
- 25x4.0%
- 40x2.5%
A company on 40 times earnings offers 2.5 percent on last year's profit. That is not a verdict. It says what has to happen for the price to work: the profit has to grow a great deal, and the multiple is the market's estimate of how likely that is.
High against what
Across the 1,994 US companies with a computed P/E on 2026-09-07, the median was 21.16x. Most of the spread around it is a sector fact, not a price fact.
Median P/E by sector
- Financial Services · 479 companies14.19x
- Energy · 115 companies17.71x
- Consumer Cyclical · 249 companies20.15x
- Industrials · 321 companies29.75x
- Technology · 281 companies33.07x
Median, all US companies: 21.16x
So "25x is high" is only true once you say high against what. It is above the market median and below the Technology median. The top of the range is mostly small earnings, not big prices: on the same date the 90th percentile was 72.49x and the 95th was 127.52x, and a multiple that size usually means a denominator close to zero.
Is a high or a low P/E better? Neither. A low one says the market expects profit to fall, or that it is wrong. A high one says it expects profit to rise, or that it is wrong. The ratio cannot tell an expectation from an error.
The E is a choice
The price is one number. The earnings are four choices, and each one moves the ratio.
| Choice | One way | The other way |
|---|---|---|
| Period | Trailing: profit already filed | Forward: a forecast nobody filed |
| Basis | Reported under GAAP | Adjusted: management's exclusions |
| Shares | Basic | Diluted: counts options and convertibles |
| Window | One year | Several years, averaged |
Adjusted earnings must be reconciled to GAAP (Generally accepted accounting principles: the US rulebook every filed statement follows.) under Regulation G, and that reconciliation is the useful document, because it lists what was removed. Graham argued in chapter 12 of The Intelligent Investor that one year's per-share earnings carry special charges and accounting choices, which is why he averaged. ValueScreener prefers the vendor's trailing-twelve-month ratio when one is published, and the source trail shows it beside a recompute from the latest annual statements.
Graham's 15x, read properly
Graham's defensive test in chapter 14 was a price of at most 15 times the average earnings of the past three years, and it was one of seven tests. Applied to a single year, it is a near neighbour of his test, not the test. Shiller's cyclically adjusted P/E, built with John Campbell, averages ten years of inflation-adjusted earnings, so a cyclical company does not look cheapest at the top of its cycle.
Do it on a real 10-K
- Open Item 8 and take diluted earnings per share from the face of the income statement.
- Open the company's non-GAAP reconciliation and compute the P/E again on the adjusted figure.
- Check whether each add-back also appears in the two prior years. If it does, it is a recurring cost.
Check yourself
A P/E of 25 is an earnings yield of what?
- 25 percent
- 4 percent
- 2.5 percent
Show the answer
Answer: B.
One divided by 25 is 0.04. The yield is the multiple turned over.A company made a loss last year. What is its P/E?
- A very high number
- Zero
- It has no meaningful P/E
Show the answer
Answer: C.
A loss gives no meaningful ratio, so it is shown as n/a with the reason. A screen on P/E leaves these companies out without saying so.Graham's 15x test divides the price by which earnings?
- Next year's forecast
- Last year's reported profit
- The average of the past three years
Show the answer
Answer: C.
Chapter 14 of The Intelligent Investor uses three-year average earnings. The averaging is the part most screens drop.Sources
- Graham, The Intelligent Investor, revised edition 1973, chapter 12 (per-share earnings) and chapter 14 (the defensive investor's criteria).
- Shiller, Irrational Exuberance, 2000, and the cyclically adjusted P/E developed with John Campbell.
- Regulation G and Item 10(e) of Regulation S-K, on reconciling non-GAAP measures.
- The P/E distribution published on the open research pages of ValueScreener, built 2026-09-07.
What this leaves out
- The distribution was built on 2026-09-07 from prices last stored on 2026-08-26, and it is rebuilt on a schedule, so a later reader will see different breakpoints and cohort sizes.
- Equity REITs are left out of the cohort because they are read on funds from operations. Banks are in it, but a bank is read against its book value first.
- The yield chart and the worked example use round invented numbers.
- Nothing here says what any company is worth or what anyone ought to pay. It is not advice about any security.