Shiller P/E (CAPE): how expensive is the stock market?

In depth9 min read · 8 sources

The Shiller P/E, also known as CAPE (cyclically adjusted price-earnings ratio), divides the price of a stock market by the inflation-adjusted average earnings of the last ten years. The long average smooths out recessions and booms. Robert Shiller popularized the metric; for the US market, data go back to 1881. Hardly any valuation metric has been researched so thoroughly or misunderstood so often.

How to calculate the Shiller P/E

The ordinary P/E ratio divides by the earnings of the last twelve months. In recessions, earnings collapse and the P/E jumps, even though prices have fallen. It then sends the wrong signal. The ten-year average covers roughly one full business cycle. Using US data from 1871 to 1986, Campbell and Shiller (1988) showed that long-term average earnings predict the value of future dividends well: two thirds to three quarters of the weight fell on this earnings measure.

Earnings yield: what a CAPE means for returns

The reciprocal 1 / CAPE is called the earnings yield. If companies pay out their earnings or reinvest them at market rates of return, it is a rough estimate of the long-term real equity return, i.e. after inflation. A CAPE of 17 corresponds to an earnings yield of 5.88%, a CAPE of 40 to only 2.50%.

For the US, the CAPE averaged 17.8 from 1881 to October 2026 (median 16.6) and was between 10 and 25 in 71% of months. Values above 30 occurred in 1929, from 1997 to 2002 and then again almost continuously since 2017. In October 2026 it stood at a provisional 40.8. It was only higher around the turn of the millennium.

Excess CAPE Yield: equities versus bonds

A high CAPE is less worrying when bonds also yield little. Shiller et al. (2020) therefore compare the earnings yield with the real interest rate on long-term government bonds, and so estimate the expected real excess return of equities over bonds. This difference is called the Excess CAPE Yield.

What the research shows

The CAPE became famous through Shiller's book on speculative excesses (Shiller 2000), published shortly before the tech bubble burst. Campbell and Shiller (1998) showed that high valuations in the US preceded low returns over the following ten years. Even critics regard the metric as one of the better long-term indicators:

At the same time, Siegel (2016) argues that changed accounting rules (such as mark-to-market write-downs) have depressed reported earnings and biased the CAPE upwards; with corporate profits from the national accounts, the forecasts are less gloomy. Welch and Goyal (2008) tested many proposed predictors of the equity premium and found that most perform poorly both in and out of sample and would not have helped an investor with timing.

For advanced readers: how reliable is the forecast?

The usual analysis is a regression of the real return over the following ten years on the earnings yield 1 / CAPE. With monthly US data from 1881, our own analysis gives roughly: subsequent return ≈ 0.21% + 0.91 × (1 / CAPE), with an R² of 0.28.

The number of observations is deceptive: ten-year returns from monthly data overlap almost completely. 1,590 monthly values contain only 1,590 / 120 ≈ 13 independent ten-year periods. Boudoukh et al. (2008) show why regressions over long, overlapping horizons therefore produce seemingly high predictive power: the estimators for different horizons are almost perfectly correlated. Standard errors following Newey and West (1987) with a correspondingly large number of lags help, but with so few independent periods they are, if anything, still too optimistic.

• Earnings yield or logarithm: instead of 1 / CAPE, log(CAPE) can also serve as the explanatory variable; both forms produce similar rankings, but different forecasts at the extremes.
• Sample: calculated from 1950 or 1990, the mean and the coefficients shift considerably. This points to structural changes.
• Percentile instead of level: how high the CAPE is relative to its own history can be expressed as a percentile or z-score; both depend on the chosen comparison period.

What the CAPE cannot do

• Short term: the CAPE says almost nothing about the next one to two years. Markets can stay expensive for a long time, as the years 1997 to 2000 show.
• Structural change: different accounting, higher margins and lower interest rates have shifted the level; the long-term average of around 17 is not a fixed target.
• Only one market: the long data series apply to the US; for Europe or emerging markets, values are usually considerably lower, with a shorter history.
• Composition: an index with many fast-growing technology companies has, for good reasons, a higher P/E than one made up of banks and utilities.

Putting the CAPE in context

Three questions help you avoid overrating a CAPE figure. First: compared with what? A percentile of 95% relative to all months since 1881 may be only a middling value relative to the last 30 years. Second: how high are interest rates? With real interest rates close to zero, a low earnings yield is less unusual than with real rates of 3%. Third: which market? The US CAPE says little about European equities or about a global portfolio in which the US dominates but does not stand alone.

A useful exercise is to take the earnings yield as the starting point for your own planning and deliberately calculate a range: what happens to your retirement plan if equities return only 2% instead of 5% in real terms over the next ten years? If you have a good answer to that, you need not fear the CAPE. The historical equity premium over long periods is the subject of the article equity risk premium.

In practice

The CAPE is suitable for slow, rule-based adjustments to the equity allocation and for setting expectations in planning, for example a more cautious return in the Monte Carlo simulation. It is no use as a signal for short-term moves in and out of the market (see market timing).

One example of such a rule is the valuation model in Tactical: if the CAPE is at or below the 50th percentile of its history to date, the risk portfolio is fully invested; from the 90th percentile onwards, only 30%; in between, the share falls linearly. At a percentile of 70%, it is therefore 1 − (0.7 − 0.5) / 0.4 × 0.7 = 65%, with the rest in the money market.

How Portfolio Lab calculates it

The Valuation page shows the CAPE of the S&P 500 since 1881: from Shiller's monthly data, extended to the present with a current monthly series (the current month counts as provisional, and deviations in the overlap period are reported). The real total return for each month is derived from price and dividend, adjusted for the consumer price index. The expected return is a regression of the subsequent return over 5, 10 or 15 years on 1 / CAPE or log(CAPE), starting from 1881, 1950 or 1990 as you choose, with Newey-West standard errors (horizon minus 1 lags); the t-statistic and R² only appear once there are about five independent periods. The band around the estimate is the 10th to 90th percentile of historical deviations, not a confidence interval. The Excess CAPE Yield uses the yield on 10-year US Treasuries minus average inflation over the last ten years. More in the valuation methodology.

This article explains a method; it is not investment or tax advice. Worked examples are simplified and rounded. All terms in the glossary.