Skip to main content

The P/E Ratio: Basics, Reliability Band, and When It Lies

Created Sep 15, 2026 at 1:20pm

Research Draft Medium 1,485 words

Price divided by a year of earnings per share: "how many years of today's profit am I paying for." It is the most-quoted valuation number there is, and it is a reciprocal — a small, noisy or temporarily inflated denominator turns into a wildly swinging ratio. A P/E is a valuation read only inside a middle band; outside it, the number is telling you about the earnings, not the price. The other multiples and the comparison workflow live in the parent node Valuation Multiples (P/E, EV/EBITDA, P/B, P/S).

The basics

Formula. P/E = price ÷ EPS. Its inverse is the earnings yield: 20× = a 5% yield (CFI).

Which EPS. Every quoted P/E hides a choice (Damodaran, Investment Valuation, ch. 18): trailing (last twelve months), forward (next-year estimate), run-rate (latest quarter × 4), GAAP vs adjusted, basic vs diluted. The honest default is trailing GAAP diluted — diluted is the conservative share count, and adjusted EPS is usually higher than GAAP so an "adjusted P/E" is usually lower. Never compare two P/Es on different EPS definitions (Non-GAAP Adjustments, Analyst Estimates & Consensus).

What it means. P/E compresses the market's expected growth, risk and payout into one figure: forward P/E = payout ÷ (r − g) in Damodaran's stable-growth form. A "high" P/E is only high relative to what its growth and risk justify. PEG (P/E ÷ growth) is the shortcut; Lynch's "fair at PEG ≈ 1" has no theoretical anchor, and Damodaran shows riskier firms trade at lower PEGs for the same growth — the cheapest-looking PEG is often the riskiest name.

Typical levels. S&P 500 trailing P/E: long-run median about 15, mean about 16, extremes 5 (1917) and 124 (2009, when earnings collapsed); about 26 as of 2026-09-14 (multpl.com, Shiller data). Context, not a target.

The reliability band

P/E is a hyperbola in earnings, not a line. Its information is concentrated in a middle range and destroyed at both ends — for two different reasons.

The P/E surface: flat where earnings are comfortable, a wall as they vanish Height clipped at 60× · drag to rotate
The P/E surface over price and earnings: flat where earnings are comfortable, a wall as EPS approaches zero. Height is P/E clipped at 60×; colour is the reliability zone (usable 8–30×, fading 30–60×, trap below 8×, noise above 60×).

High end — noise. Reported EPS carries an error floor (accruals, one-offs, tax items, a weak quarter) that scales with the size of the business, not with profit. Relative error in P/E equals relative error in EPS, and that explodes as EPS approaches zero. Hypothetical: price $100, EPS noise ±$0.50. At EPS $10 the P/E is 10× with a band of 9.5–10.5×. At EPS $1 it is 100× with a band of 67–200×. The second reading is not "ten times more expensive"; it is "earnings are near zero." Above roughly 50–60× (a judgment line) a trailing P/E mostly reports the denominator — and for a fast grower, trailing E also omits the future earnings the price is discounting.

The same earnings noise, two very different P/E bands Hypothetical · price fixed at $100
Noise amplification: at a fixed $100 price, the same ± noise in EPS produces a narrow P/E band when EPS is fat (9.5–10.5× at EPS $10) and an enormous one when EPS is thin (67–200× at EPS $1).

Low end — earnings about to fall. A very low P/E (roughly under 6–8×) is usually the market expecting E to drop: peak-cycle earnings (commodities, shippers, homebuilders, semis), a one-time gain or tax release, a business in decline, or leverage that makes the equity fragile. The cyclical version is the Molodovsky effect: P/Es are lowest at the earnings peak and highest at the trough, because prices move ahead of earnings. If sustainable earnings are a fraction f of reported, the true multiple is reported P/E ÷ f: 6× on earnings that are 40% sustainable is 15× — ordinary, not cheap (Value Traps, Cyclical vs Defensive Industries).

A cheap headline on earnings that will not last Hypothetical
The low-end trap: a headline P/E on reported earnings versus the same price on sustainable earnings. True multiple = headline ÷ sustainable fraction, so 6× on earnings that are 40% sustainable is 15×.

Negative earnings. P/E is undefined ("N/A"). The only information is one bit: loss-making.

The usable band. Roughly 8–30× (earnings yield 3–12%) is where P/E works as a cross-company comparator, and it works better the more stable the earnings: a utility's 14× and a staple's 22× are comparable; a semiconductor's 6× at peak and a biotech's 90× at first profit are not. The edges are judgment, not measured constants. Do not use them as screens.

Where a P/E reading carries information Reasoning aid · not measured data
Informativeness of a P/E reading across its range, drawn as a reasoning aid: undefined for negative earnings, near zero below 8×, a plateau across roughly 10–25×, fading past 30×, near zero beyond about 100×.

Reading a P/E on a company profile

Four checks, in order:

1. Which EPS? Trailing, run-rate or guided; GAAP or adjusted. A run-rate or guided P/E below the trailing one means the newest quarter is running ahead of the year — a growth signal, not a cheapness signal. 2. Inside the band? 8–30×: read it as valuation, within the sector. Above ~50×: read "earnings near zero or growth not yet in trailing E" and switch to a guided P/E or an enterprise multiple. Below ~8×: ask what is about to happen to the denominator before calling it cheap. N/A: loss-making, nothing more. 3. Is the denominator clean? One-off gains, tax releases, a single blow-out quarter annualised (Quality of Earnings). 4. What is the leverage? P/E sits after interest. Two firms with identical EV/EBIT show different P/Es once one carries debt — the levered one looks cheaper because its equity is the riskier slice. Compare debt first, or use EV/EBIT.

Same business, same operating valuation, two P/Es Hypothetical
P/E is an equity multiple: two firms with the same enterprise value and EBIT (EV/EBIT 10×) show different P/Es once one carries debt — the levered equity looks cheaper because it is the riskier slice.

When the multiple breaks, change the denominator. Guided or forward EPS for growers; a multi-year average for cyclicals (Graham and Dodd: "preferably seven to ten years"); EV/EBIT or EV/Sales when E is near zero. In screens, use earnings yield (linear, tolerates losses — the Fama–French construction) and the median, since P/E is zero-bounded and skewed so its mean runs high.

Adoption, debate & evidence

Universal and mechanically undisputed; what it predicts is contested. Cyclically-adjusted earnings yields carry long-horizon information at the index level (Campbell–Shiller) but little for timing. At the single-stock level the value evidence rests on book-to-market and earnings yield, not raw P/E, and a low P/E is at least as often a correct forecast of falling earnings as a bargain.

Strengths & limitations

Works for mature, profitable, moderately levered businesses with stable earnings, compared within a sector on the same EPS definition. Fails near zero earnings, at cycle peaks, across capital structures and EPS definitions, and for firms whose value lies in future rather than trailing earnings. The most common misuse: reading an out-of-band P/E as a valuation — calling 80× "expensive" or 5× "cheap" — when the number is a statement about the denominator.

Sources