PE of S&P 500: Why the Numbers Disagree

What is the P/E of the S&P 500, in one sentence?

The P/E of the S&P 500 is the combined market capitalisation of the index constituents divided by the combined earnings attributable to those same index shares, not the average of the five hundred individual price-to-earnings ratios. It is a constructed statistic, assembled from a constituent list, a share-count convention and an earnings definition, rather than a quantity that exists in the market and can be observed.

That construction is why "the" P/E of the S&P 500 does not exist as a single number. At least four distinct ratios circulate under the same three letters: trailing twelve-month as-reported (GAAP) P/E, trailing twelve-month operating P/E, forward twelve-month P/E on analyst consensus, and the cyclically adjusted (Shiller) P/E on ten years of inflation-adjusted earnings.

Each is internally coherent. Each answers a different question. On any given day in 2026 they can sit several points apart, and none of them is wrong.

The rest of this article walks the arithmetic that produces those differences, shows a five-line worked example where the same data yields five defensible answers, and sets out what a research desk needs in a daily constituent file to reproduce the figure instead of quoting a print it cannot audit. The companion piece on how the S&P 500 P/E number is built covers the index mechanics in more detail.

Why do published S&P 500 P/E ratios disagree on the same day?

Published S&P 500 P/E ratios disagree because the calculation involves at least six independent convention choices, and no two publishers make all six the same way. A spread of three to five points between two reputable sources on the same date is normal, not a sign that one of them is broken.

Here are the choices that move the number, in rough order of how much they move it.

The aggregation point is the one most often missed. Because the aggregate P/E equals the reciprocal of the cap-weighted sum of earnings yields, it is mathematically a cap-weighted harmonic mean of the individual ratios. A cap-weighted arithmetic mean of the same ratios is a different statistic and, by Jensen's inequality, always at least as large.

Loss-making constituents are the second trap. In an aggregate calculation a loss simply subtracts from the numerator's denominator, which is arithmetically clean. In a mean-of-ratios calculation a negative P/E is meaningless, so the publisher must exclude the name, floor it, or cap it, and each of those decisions is a judgement that the headline number does not disclose.

The practical rule: never compare two P/E prints unless you have read both methodology notes. If a source does not publish one, treat its series as a chart, not as an input.

How is the index P/E actually computed from constituent data?

The index P/E is computed by multiplying each constituent's price by its index share count to get index market value, summing that across the index, then dividing by the sum of each constituent's trailing earnings scaled to the same index share count. Both sides of the ratio must use the identical share basis, or the result is not a valid index-level statistic. The step-by-step index P/E calculation sets out the same sequence with the index-level conventions attached.

The steps, in the order a desk runs them:

1. Pull the dated constituent list with index shares (shares outstanding multiplied by the investable weight factor), price and a stable identifier for each line.
2. Compute index market value as the sum of price times index shares across all constituents.
3. Compute index earnings as the sum of trailing four-quarter net income per share times index shares, using the same share count as step two.
4. Divide index market value by index earnings.
5. Record the convention used for losses, share classes and the earnings vintage alongside the result.

Why the divisor does not change the answer, but still matters

The index divisor exists to keep the index level continuous across corporate actions. The published level is index market value divided by the divisor, and index EPS is published on the same divisor, so the divisor cancels in the ratio. You can compute the P/E without ever knowing the divisor.

It matters anyway, for two reasons. If you reconstruct the index level as a sanity check, the divisor is the only thing standing between your market value sum and the printed level. And a divisor change on a reconstitution day is the signal that your share counts have moved, which means your earnings scaling must move with them on the same date.

Share classes, ADRs and the double-counting risk

Companies with multiple listed share classes appear in the index as separate lines. Attributing full company net income to each line double-counts earnings and depresses the index P/E, sometimes noticeably given how much index weight sits in dual-class names. The correct treatment is earnings per share scaled to each line's own index shares.

Foreign domicile is a smaller issue in the S&P 500, which has a US-domicile requirement, and a large one in global indexes where a company may appear as both a local line and a depositary receipt, in different currencies, on different exchange calendars. For those indexes the earnings side needs a currency convention that is documented and applied consistently, because the ratio is not currency-neutral if the numerator and denominator are converted on different dates.

For any of this to be auditable, the constituent file has to carry three things per line: the index share count, the weight, and an identifier stable enough to join to a fundamentals source across corporate actions. That is the reason Bloomberg FIGIs are carried in the constituent holdings files rather than tickers alone: tickers are reused and reassigned, and a broken join silently drops earnings from the denominator, a failure mode discussed further in the mechanics of the S&P 500 P/E.

A worked example: five lines, five defensible answers

Take a five-name miniature index built the way the real one is built. All figures below are illustrative and chosen to be checkable by hand.

Aggregate market value is 280,000 and aggregate earnings are 13,700, so the index P/E including the loss-maker is 20.4. Drop line E, as many mean-of-ratios methodologies must, and the aggregate becomes 276,000 over 14,200, or 19.4.

Now switch aggregation method on the same data. A cap-weighted arithmetic mean of the four positive ratios gives 22.0. An equal-weighted arithmetic mean of those ratios gives 21.3. An equal-weighted harmonic mean gives 17.9.

Five numbers between 17.9 and 22.0, from one set of five rows, with no disagreement about any price or any earnings figure. That is the entire explanation for why the published prints disagree, scaled up by ninety-nine.

Trailing, forward and Shiller: which P/E answers which question?

Each of the four ratios is fit for a different decision, and using one to answer another's question is the most common analytical error in index valuation work. The table below sets out what each measures and where it misleads.

The forward P/E has a property the others do not: it moves on estimate revisions as well as on prices. A flat market with falling estimates produces a rising forward P/E, which looks like multiple expansion and is nothing of the sort. Anyone using the forward figure as a valuation signal needs to decompose the change into a price component and a revision component, or the signal is uninterpretable.

The Shiller P/E reads structurally higher now than its own long-run average, and the reasons are mechanical rather than judgemental: the ten-year window still carries depressed earnings from earlier cycles, the index-level payout mix has shifted toward buybacks, and sector composition has moved toward businesses with different capital intensity. That does not make the ratio useless. It makes the comparison to its own full-history mean a comparison across accounting and composition regimes.

As for who publishes what, the convention split is real but you should verify it at the source rather than take our word for it. Free chart pages tend to lead with a trailing as-reported series because that data is public and requires no estimate feed, while institutional market-guide decks tend to lead with forward P/E because that is what their client conversations are about. In every case the methodology note attached to the series is the authority, not the headline.

What does the historical P/E of the S&P 500 actually tell you?

The long P/E series tells you how the multiple has moved within each accounting and composition regime, and much less than it appears to tell you across regimes. A chart running from 1926 to 2026 contains several distinct definitions of earnings, several eras of index membership rules, and an index whose sector composition today bears little resemblance to its composition even twenty years ago.

Three specific distortions are worth naming.

Earnings collapses spike the trailing ratio without signalling expensiveness. In 2009 and again in 2020, denominator collapse drove the trailing as-reported P/E to levels that, read naively, implied extreme overvaluation at what turned out to be favourable entry points. A trailing ratio measures the past against the present, and when the past is anomalous the ratio is a statement about accounting, not about price.

Composition drift makes the average a moving target. The index-level margin, the capital intensity and the share of earnings coming from asset-light businesses have all shifted materially. Comparing today's multiple to a hundred-year mean compares two different indexes wearing the same name.

Membership rules changed too. Float adjustment, profitability requirements at entry and the treatment of multiple share classes have all been revised over time, and each revision changed the population being measured.

The practical response is to test rather than assume. With archived constituent list datafiles going back to July 2009, a desk can rebuild the index P/E on the members who were actually in the index on each historical date, with the share counts that applied then, and compare that to the headline series, following the same construction used for the current-date ratio. That comparison is impossible from a published chart, because the chart gives you one number per date and no way to see what went into it.

Is the S&P 500 overvalued? What the ratio can and cannot settle

The P/E ratio bounds the valuation question, it does not close it. A single multiple compared to a single historical average is the weakest available use of the statistic, because both terms of that comparison are unstable for the reasons set out above.

Four conditional comparisons carry more information than the level alone.

• Against its own recent history, expressed in standard deviations from a rolling mean over a period with consistent accounting. This is the framing used on most institutional valuation slides, and it is defensible because it holds the regime roughly fixed.
• Against real rates, via the earnings yield spread over the real yield on long Treasuries. An equity multiple is a discount rate statement, and comparing it to nothing is comparing half an equation.
• Against the equal-weighted index, which strips out concentration. When the cap-weighted and equal-weighted P/E diverge, the divergence itself is the signal: it says the aggregate multiple is being set by a small number of very large names.
• Against the index excluding its largest constituents, which quantifies the previous point directly.

The concentration effect deserves emphasis because it changes what the headline ratio means. In a highly concentrated index, the aggregate P/E is close to the weighted multiple of the top decile, and the median constituent may be trading at a materially different level. Reporting the aggregate as "the market's multiple" is then a statement about a handful of companies dressed as a statement about five hundred.

None of this produces a verdict. It produces a set of conditional statements, each of which a desk can defend to a risk committee, which is a better outcome than a single number nobody can reproduce.

Building your own index P/E series in-house

A defensible in-house index P/E series needs four things: dated constituent snapshots with identifiers and share counts, a fundamentals source keyed to those identifiers, a written and frozen convention for the edge cases, and point-in-time storage so history is not silently rewritten.

Start with the constituent snapshots, because everything else joins to them. Using today's membership for historical dates introduces survivorship bias directly into the earnings denominator: the companies that were removed from the index were removed for reasons correlated with earnings. A series built that way looks cheaper in the past than the index actually was.

Freeze the conventions in writing before you compute anything. Specifically: whether loss-makers are included, how multiple share classes are handled, which earnings definition is canonical, and how currency is converted for global index variants. These decisions are not interesting, but changing them silently mid-series destroys comparability, and undocumented conventions are the reason most internal series cannot be reconciled to anything.

Point-in-time storage is the discipline that separates a research series from a chart. Store what you knew on each date, and store restatements as a separate vintage rather than overwriting. Then, when a published series disagrees with yours, you can decompose the gap into definition, aggregation, membership and vintage components and explain it, which is the only outcome that survives a model review.

What the data layer has to provide

The delivery requirements are mundane and they are where most builds stall. You need a snapshot per index per day, with constituent-level shares, weights and identifiers, in a format a pipeline can ingest without a scraper. FIGIs matter here because they survive ticker changes across corporate actions, which is exactly when a naive join breaks and earnings quietly vanish from the denominator.

Update timing determines what your series can claim. AmericanETP refreshes constituent files twice daily, with a primary update at 6pm EST and a secondary run around noon EST, which supports an end-of-day series with an intraday check but not a tick-level one. That is the right resolution for a valuation series, since the earnings denominator only changes on reporting dates anyway, and the extended-session pricing problem makes anything finer harder to defend rather than easier.

Coverage and access shape the rest of the build. AmericanETP covers 3,878 US and global indexes and US-traded ETFs, archives constituent files back to July 2009, and delivers by CSV with FTP access available, which is enough to build the historical series and the daily update from the same source. Pricing is $1,500 per year or $150 per month for an individual, and $2,500 per year or $250 per month firm-wide. Knowing which dates carry membership changes is what tells you which dates need special handling before you compute anything on them.

Where P/E series break in practice

Index P/E series fail in a small number of recurring ways, and each has a symptom a research team notices first. Knowing the symptom shortens the diagnosis from days to minutes.

Reconstitution days. Symptom: a one-day step change in the ratio with no corresponding market move. Cause: members changed but the earnings denominator was carried forward on the old membership, or the new members' earnings were joined a day late.

Spin-offs. Symptom: aggregate earnings drop while aggregate market value holds. Cause: the earnings left with the spun-off entity, but the parent's trailing four quarters still include them, and the spin-off may enter the index as a separate line carrying the same earnings again.

Restatements. Symptom: a historical value changes between two runs of the same script. Cause: the fundamentals source restated a prior quarter. This is correct behaviour from the vendor and destructive to a series stored without vintages.

Currency treatment on global indexes. Symptom: the ratio moves on days with no equity price action. Cause: numerator and denominator converted at different FX dates, or one converted and the other not.

Extended and overnight session pricing. Symptom: the ratio is unstable outside regular hours. Cause: a live price against an earnings denominator that only updates on reporting dates, compounded by thin overnight liquidity making the numerator itself unreliable. The data problems created by 24×5 trading apply directly here, and the practical answer is to compute valuation series on official closes only.

Stale share counts. Symptom: your reconstructed index market value drifts from the published level over weeks. Cause: buyback disclosures changed shares outstanding and the index share count in your file was not refreshed, which biases both weights and the scaled earnings figure.

Every one of these is a data-freshness or a join-integrity problem rather than a modelling problem. That is why the constituent file, not the formula, is where index valuation work actually lives. If your file is dated, identified and updated on a known schedule, the arithmetic in the index-level P/E walkthrough is straightforward.

FAQ

What is the current PE for the S&P 500?

There is no single current figure, and any source quoting one without naming its earnings definition and aggregation method is giving you a number you cannot audit. As of 2026, published S&P 500 P/E ratios on the same trading day routinely differ by several points depending on whether they use as-reported or operating earnings, whether they aggregate or average, and how they treat loss-making constituents.

The reproducible answer is to compute it: take a dated constituent file with index shares and prices, sum index market value, sum trailing four-quarter earnings scaled to the same shares, and divide. Then record which convention you used, so the number means something a month later.

Is the S&P 500 currently overvalued?

The P/E ratio cannot settle that question on its own, because "overvalued" requires a comparison and every available comparison has a caveat. Against its own long-run average, the comparison spans multiple accounting and composition regimes. Against real rates, the answer depends on where you think real rates are heading.

The more informative test is the spread between the cap-weighted and equal-weighted index multiples. When they diverge widely, the headline ratio is describing the largest constituents rather than the index, and a valuation judgement made on the aggregate is being made about a small subset of it.

What does Warren Buffett say about the PE ratio?

Buffett has consistently framed valuation in terms of discounted future cash flows rather than a single multiple, and has argued in shareholder letters that a P/E ratio without context on capital intensity, growth and interest rates says very little. He is not a critic of the ratio so much as a critic of using it alone.

He has also pointed to other aggregate measures. In a 2001 Fortune article he described the ratio of total US market value to GNP as probably the best single measure of where valuations stand at any given moment, a metric now commonly called the Buffett indicator. That measure has the same regime-comparability limits as the long P/E series does.

What is a good PE ratio for the S&P 500?

There is no threshold that separates good from bad, because the index multiple is a discount rate statement and the discount rate is not constant. A multiple that is reasonable at low real yields is demanding at high ones.

A more useful framing is distance from a rolling mean computed over a period with consistent accounting standards, expressed in standard deviations, and read alongside the earnings yield spread over real Treasury yields. That gives you a conditional statement rather than a verdict, which is the most the ratio can honestly support.

What is the difference between the forward PE and the trailing PE of the S&P 500?

The trailing P/E divides current index market value by the last four reported quarters of earnings; the forward P/E divides it by analyst consensus estimates for the next four quarters. The trailing figure is backward-looking and lags reporting by roughly four to ten weeks; the forward figure has no lag but is an estimate that revises continuously.

The forward P/E is typically the lower of the two when earnings are expected to grow, and it moves on estimate revisions as well as on prices. That second property matters: a rising forward P/E in a flat market means estimates fell, not that the market got more expensive.

How is the Shiller PE ratio of the S&P 500 calculated?

The cyclically adjusted P/E divides the inflation-adjusted price of the index by the average of the previous ten years of inflation-adjusted as-reported earnings. Both the price and the earnings series are deflated to a common currency date using CPI, and the ten-year window is intended to smooth out the effect of the business cycle on the denominator.

Its known limitations follow directly from that construction. The ten-year window responds slowly to structural change in index composition or accounting standards, and it carries earnings from earlier cycles well past the point where they describe the current index. It is a long-horizon context measure, not a timing signal.

If you are building or auditing an index P/E series, the constraint is almost never the arithmetic. It is having dated constituent files with shares, weights and stable identifiers, delivered on a known schedule, with an archive deep enough to test the history. See what AmericanETP delivers.