CAPE explained: calculation methodology, 10-year real earnings average and Shiller inflation adjustment

Behind the acronym CAPE, Cyclically Adjusted Price-Earnings, sits a calculation of remarkable simplicity: real S&P 500 price as numerator, arithmetic mean of the previous ten years’ real earnings as denominator, double CPI deflation on both terms, series back-calculated to 1881.

Reconstructing the formula step by step and making explicit Shiller’s methodological choices (why ten years, why CPI, why the S&P 500) is the prerequisite to any serious reading of the ratio and to any distinction with its cousins forward P/E and trailing P/E.

The Formula Step by Step

The CAPE is computed in four mechanical steps. Step 1: retrieve the monthly S&P 500 closing price at date t. Step 2: deflate that price by the U.S. consumer price index (CPI All Urban Consumers, FRED series CPIAUCSL) rebased to the current period, to obtain a real price. Step 3: retrieve the S&P 500 GAAP earnings over the ten years preceding date t (40 quarterly releases of trailing four-quarter earnings, which Shiller interpolates month by month into 120 monthly observations) and deflate each value by the corresponding period CPI. Step 4: compute the arithmetic mean of these 120 monthly real earnings and divide the real price from step 2 by that mean. For the full framework, see our sub-pillar on equity market valuation.

This construction yields a compact formula: CAPE(t) = P_real(t) / [10-year mean of E_real]. An equivalent variant, more intuitive, uses annual earnings: CAPE(t) = P_real(t) / [arithmetic mean of annual real earnings over the ten years preceding t]. Both versions yield statistically indistinguishable results. Robert Shiller publishes the official monthly series on his Yale academic site, the reference Shiller Online Data, updated to the current month based on BLS data for CPI and S&P Dow Jones Indices for earnings.

The apparent simplicity masks a few technical subtleties. The earnings used are the GAAP reported earnings as they appear in the audited accounts of S&P 500 companies, aggregated by S&P Dow Jones Indices. This scope excludes pro forma adjustments or “adjusted” earnings published by certain companies to neutralize non-recurring charges: a distinction that becomes important after 2001 in the critical debate on the FASB 142 break. The general properties of the CAPE as the academic reference measure rest precisely on this GAAP discipline that ensures historical comparability before arbitration of the critiques.

The result is a unit-free multiple, published once a month: 37.0 in March 2026, 41.1 in August and 40.6 in September according to Robert Shiller's data. Expressed in constant dollars in both numerator and denominator, it compares directly with the readings of 1881, 1929 or 1999, which makes it a regime thermometer across 145 years.

Why Ten Years: the Cyclical Choice

The choice of a decadal average is not arbitrary. Shiller inherited the idea from Benjamin Graham and David Dodd, whose Security Analysis published in 1934 already advocated a seven-to-ten-year average to neutralize the cycle. The justification is cyclical: per the FRED USREC series, the 34 U.S. cycles since December 1854 lasted 4.9 years on average from trough to trough, and 6.2 years since 1945. A ten-year average therefore usually spans at least one complete cycle, without guarantee: the 1991-2001 and 2009-2020 cycles both ran longer than ten years. A five-year average would remain sensitive to the cyclical phase of observation; a twenty-year average would dilute information across a sample whose sectoral composition would have meaningfully shifted.

The fundamental property sought is the stationarity of the denominator. Earnings of a broad index like the S&P 500 can collapse in deep recession then double in vigorous expansion: in Shiller's data, trailing twelve-month earnings fell sharply from 2000 to 2002, from 1929 to 1932 and between January 2008 and March 2009. This produces explosive annual P/Es at cycle troughs and artificially low P/Es at peaks. In real terms, earnings went from 136 in June 2007 to 11 in March 2009 while their ten-year average only moved from 88 to 89 and then to 84 by end-2009, as the chart at the top of this article shows. Smoothing over ten years neutralizes these oscillations without introducing structural bias: the ratio between 10-year mean real earnings and long-run normalized real earnings stays approximately constant independently of the observation moment.

The ten-year window is not a statistical optimum. Recomputed by Eco3min on Shiller's data (starting points 1881-2016), the correlation between the log of the ratio and the annualized real total return over the following ten years is −0.50 with a five-year average, −0.51 with ten years, −0.52 with fifteen and −0.53 with twenty. Predictive power comes from normalizing earnings far more than from the exact window length: ten years is a convention inherited from Graham and Dodd, long enough to smooth a cycle, short enough to track the index's composition. Related explainer: what CAPE captures that forward P/E misses.

A recurring critique of the ten-year choice comes from the duration of secular cycles themselves. If a secular cycle lasts ten to twenty-five years (Great Moderation 1982-2008 for example), a decadal average can capture an atypical portion of the long cycle and produce a biased reading. This critique is valid for extreme phases but does not disqualify the ratio: it simply implies that the CAPE must be read against its historical distribution rather than against a fixed theoretical equilibrium, a point that structures the historical reading of the ratio over 145 years.

Why CPI: the Double Deflation

Inflation adjustment is the second founding statistical discipline of the CAPE. Its purpose is not to make a multiple "real" (a P/E is already unit-free), but to add up earnings from ten different years in a single currency. Without deflation, older earnings, expressed in pre-inflation dollars, drag the average down and inflate the ratio all the more when the decade's inflation was high: in December 1981, the ratio computed on nominal earnings would read 11.9 against 7.8 once earnings are deflated; in June 2026, 47.1 against 40.2. The U.S. CPI multiplied prices by 19.5 between the 1929 average and August 2026 per the BLS, which would make any long nominal average misleading.

Shiller applies deflation to the numerator (S&P 500 price) and denominator (earnings) simultaneously. This double deflation does not only make levels historically comparable: it also preserves the ratio’s dimensionality. A P/E is a unitless number (dollars over dollars), and deflation by a single index at numerator and denominator ensures this dimensionless property is preserved. If only the numerator were deflated, the ratio would mix real and nominal dollars, making it incomparable across periods.

The choice of CPI rather than PCE (Personal Consumption Expenditures, the price index officially retained by the Federal Reserve for its 2% inflation targeting) is conventional. CPI offers a longer and more methodologically stable series across the century. PCE has a broader scope, including spending paid on households' behalf (healthcare in particular), and tends to print slightly below CPI: 0.46 point a year on average from 1960 to 2024, 0.38 point since 2000 (BLS, BEA). But over the CAPE’s decadal horizon, the cumulative gap remains marginal and does not modify the regime reading. Shiller stays with CPI by tradition and methodological continuity of his series since 1988.

A methodological variant occasionally invokes the use of operating earnings rather than GAAP. Operating earnings exclude non-recurring exceptional charges and provide a smoother but also more optimistic reading of underlying profitability. Shiller explicitly maintains GAAP as the convention because operating earnings carry a structural upward bias, “non-recurring” charges often proving recurring over the long run, and their definition varies from company to company, degrading historical comparability.

CAPE vs Trailing P/E vs Forward P/E: Differentiation

The CAPE must be distinguished from two other equity valuation measures that circulate continuously in financial commentary. Trailing P/E, often abbreviated to P/E without qualifier, is the current price divided by the GAAP earnings of the last twelve months (TTM, Trailing Twelve Months). It is volatile by construction: at cycle troughs, TTM earnings collapse and the P/E mechanically explodes (a P/E of 124 on the S&P 500 in May 2009 on reported earnings, per Shiller's data); at cycle peaks, TTM earnings inflate and the P/E appears low. This cyclical property makes the trailing P/E poorly readable as a valuation-regime thermometer. For the corresponding cross-read, see the equity market expressed in ounces of gold.

Forward P/E is the current price divided by sell-side consensus earnings projected over the next twelve months. It is less volatile than trailing P/E but carries a documented bias: over 25 years, McKinsey (April 2010) found earnings-growth forecasts well above realized growth of 6% a year. This analytical bias makes forward P/E structurally lower than the P/E that will actually be observed in retrospect.

The CAPE solves both problems through decadal smoothing of real earnings. It is not cyclical like trailing P/E and does not depend on sell-side projections like forward P/E. In return, it is less reactive to recent inflections: a structural shift in profitability, S&P 500 rotation toward high-margin services for example, takes ten years to fully incorporate into the decadal mean. Jeremy Siegel's main critique targets a different point, the earnings measure: in "The Shiller CAPE Ratio: A New Look" (Financial Analysts Journal, 2016), he argues that accounting changes, including mark-to-market rules, depress recent GAAP earnings, and that a ratio built on national-accounts (NIPA) profits forecasts returns better and less pessimistically. The detailed comparison of the three measures quantifies historical gaps and examines when each ratio is analytically preferable.

For an investor readership, the typical usage of the three measures splits as follows: CAPE illuminates the long-term regime and the expected forward 10-year return, trailing P/E illuminates the raw cyclical reading, forward P/E illuminates sell-side anticipations. None of the three measures is universally superior; they answer distinct analytical questions. Conflating the three or privileging one as “the” valuation measure is precisely the methodological trap Shiller sought to avoid by formalizing the CAPE as an explicit construction. This positioning fits naturally within the reading framework of equity market valuation, real rates, multiples and earnings that structures the sub-pillar dedicated to equity multiples, with the raw historical data published openly through the canonical S&P 500 CAPE dataset.

Last updated — 30 October 2026