Shiller P/E and Future Returns: A Calmer Look at the Data

The Shiller P/E ratio — cyclically adjusted price-to-earnings, or CAPE — is simultaneously the most useful long-run valuation benchmark available to equity investors and one of the most frequently misapplied. Not because the underlying data is unreliable. The statistical relationship between starting valuation and ten-year forward real returns has held with genuine predictive power across 140 years of US market history. The misapplication comes in how the signal gets deployed: as a market-timing trigger, a sell-now alarm, or evidence that equity investment at elevated readings is categorically imprudent. The data doesn't support those uses. It supports something narrower and, I'd argue, more practically valuable.

As of June 2026, the Shiller P/E sits near 41 — one of the two or three highest readings in the full dataset going back to 1881, per Advisor Perspectives' monthly P/E10 analysis. Only December 1999's peak of 44.19 stands clearly above it in the modern record. That context is real. The implications for ten-year forward returns are real. But treating those implications as a countdown timer misreads what the historical data actually says — and what it doesn't.

The argument I want to make — and that the historical record supports, with important caveats — is this: CAPE is a reliable expectation-setter for long-horizon returns, not a reliable timer. High readings consistently predict lower median forward returns than low ones. They don't predict when any compression arrives, whether it arrives via crash or slow grind, or whether sitting in cash is actually the better choice. The track record of acting on high-CAPE sell signals in real time is worse than most commentary suggests.

What the Data Shows — and What It Doesn't

Robert Shiller's dataset, updated monthly through his Yale website, makes the core relationship visible. Sorted into CAPE deciles from cheapest to most expensive, ten-year real total returns have followed the expected pattern with reasonable consistency across distinct market regimes. The cheapest decile — CAPE roughly 5 to 12 — produced median ten-year real returns near 10–11% annually. The middle range, CAPE 15 to 20, produced approximately 6–8% real. Above CAPE 25, median ten-year real returns fall to roughly 3–4%. Above 30, the median approaches 0–2% real. The direction of the signal is genuine and durable.

But the distribution within each bucket complicates the story considerably. From a starting CAPE above 25, historical ten-year real outcomes have ranged from roughly -4% annually to above 10% annually. The central tendency is clearly lower than at cheap starting points. The range still encompasses plenty of positive scenarios. An investor who sold equities in 1996 — when CAPE crossed 25 for the first time in the modern era — watched the S&P 500 roughly double over the following four years before the 2000 peak arrived. The ten-year return from that 1996 entry point, holding through 2006, remained meaningfully positive in real terms despite including the worst bear market since the 1970s. The signal was pointing toward below-average forward returns. And from 1996, it was right. But the path to that outcome rewarded patience rather than tactical exit.

The accurate statement: high starting CAPE shifts the probability distribution toward lower ten-year returns. It does not make negative outcomes certain, and it provides no information about when any correction arrives — or whether it arrives within the holding period most investors actually maintain.

Three Prior Peaks, Three Lessons

Today's reading sits near three prior episodes of extreme CAPE elevation. Each is instructive in a different way.

September 1929, CAPE near 33, preceded the worst equity market collapse in US history. The ten-year real return from that entry point was deeply negative — the S&P 500 in 1939 remained well below its 1929 level in inflation-adjusted terms, even including dividends reinvested. The CAPE signal worked on the horizon it was calibrated for. What it couldn't provide — and what no valuation metric can provide — was any indication that the collapse would arrive in October 1929 rather than 1932, or that the mechanism would be financial panic rather than gradual earnings mean reversion. Direction correct. Timing wholly absent.

December 1999, CAPE at 44.19, is the closest historical analogue to today. Investors who held through the subsequent decade experienced precisely the forward-return deterioration the model predicted: S&P 500 real total returns for the decade ending December 2009 came in at roughly -3% annually — a genuine lost decade. Cisco Systems, trading at a price-to-earnings multiple above 100 at its March 2000 peak, had not recovered its late-1990s highs two decades later. The CAPE was right about what the aggregate would deliver over ten years. But an investor who sold in 1997 — when CAPE crossed 30 for the first time — missed an additional 80%+ S&P 500 gain over the following three years before the actual peak. The signal worked on its ten-year horizon. Its utility as a timing mechanism did not.

The third analogue is 1966, CAPE near 24, less often cited but in some ways the most relevant to the current environment. What followed wasn't a crash. It was fifteen years of real-return disappointment, with inflation eroding purchasing power that rising nominal prices partially obscured. Real S&P 500 cumulative returns from 1966 through 1981 were negative. No dramatic collapse. Chronic underperformance. The CAPE identified poor forward returns, and poor returns arrived — on a timeline that would have exhausted most tactical investors long before any vindication.

Three Forces That Distort the Raw Reading

The structural earnings-composition argument is the one I find most analytically compelling — and the one I'm least confident about quantifying precisely. In 1999, the S&P 500's top constituents included a mix of manufacturers, financial institutions, energy companies, and early technology hardware businesses. Today, roughly 28–30% of the index by weight consists of software and platform businesses: high-recurring-revenue models with gross margins often above 70%, near-zero marginal distribution costs, and capital intensity a fraction of the industrials and commodity businesses that dominated the index in the 1960s and 1970s. A dollar of durable software subscription earnings is not structurally equivalent to a dollar of cyclical manufacturing earnings at peak margins. If the quality of the earnings in the CAPE denominator has genuinely improved — and I believe it has, modestly — then today's reading is somewhat less alarming than a naive comparison to 1929 implies. Somewhat. Not dramatically enough to dismiss the concern.

Interest rates are the second distorting force. In December 1999, the 10-year Treasury yielded 6.4%. Today it sits near 4.5%. The discount-rate channel and how it runs through equity multiples connects directly to CAPE interpretation: the same earnings stream justifies a higher multiple when the opportunity cost of capital is lower. This doesn't make today's reading comfortable. At CAPE 41, the cyclically adjusted earnings yield is approximately 2.4%. Against the current 10-year Treasury at 4.5%, the implied equity risk premium is roughly -2.1 percentage points — deeply negative by historical standards. Estimating the equity risk premium for valuation purposes makes clear why this combination of high CAPE and a negative ERP is doubly unfavorable: neither the starting multiple nor the risk-premium buffer supports current valuations in any conventional framework.

Accounting rule changes are the third factor, less commonly discussed but worth naming. Shiller's CAPE uses ten years of GAAP earnings, and the current backward-looking window includes periods of anomalously depressed reported profits — from goodwill impairments, purchase-price accounting write-downs under post-2009 SFAS 141R rules, and pandemic-year charges. These reduce the normalized earnings figure in the denominator, elevating measured CAPE above what an economic-earnings calculation would show. Shiller has acknowledged in published work that alternative earnings adjustments shift the number modestly downward. The distortion probably contributes a few CAPE points to the current reading. Enough to note. Not enough to change the primary interpretation.

How to Use the Signal Without Overreading It

The honest upshot: starting the next decade from CAPE 41 with a negative equity risk premium is a materially different environment than starting from CAPE 12 with an ERP of 6%. The central scenario for ten-year real equity returns is lower than the long-run average of approximately 7% real that investors trained on the 1982–2021 bull market have come to treat as a baseline expectation. Calibrating to something in the range of 2–5% real from this starting point is more defensible than extrapolating historical averages.

Use the CAPE alongside the yield curve's recession-probability signal, not as a substitute for it. The yield curve identifies credit-cycle regimes. The CAPE calibrates long-run return expectations. Neither provides timing information. Both shift the probability distribution in ways worth incorporating into planning — particularly for investors in or near distribution mode who have less time horizon to absorb a prolonged period of below-average returns.

And recognize what CAPE cannot see: individual business quality. The aggregate multiple flattens the enormous variation in competitive durability, pricing power, and capital returns across the 500 businesses that make up the index. When expected market-level returns compress, what you own matters more, not less. The CAPE ratio mechanics, its calculation, and statistical limits are covered in depth here; the point here is translating the signal into useful return expectations — not portfolio-timing instructions.

Key Takeaways

The Shiller P/E is worth tracking and, at the current reading near 41, worth incorporating into forward return expectations. Here is the framework for doing that without overreading it:

• CAPE is a ten-year forward-return expectation tool, not a timing instrument. The signal has consistently identified periods of poor subsequent decade-long real returns. It provides no information about when, within that decade, the compression arrives — or what mechanism delivers it.
• The three prior peak analogues — 1929 (CAPE ~33), 1966 (CAPE ~24), and 1999 (CAPE 44.19) — each validated the ten-year prediction while demonstrating that acting on the first alarming reading, rather than waiting for the actual peak, typically meant missing years of additional gains before the prediction resolved.
• Three regime adjustments modestly weaken the raw signal at today's levels: the index's improved earnings quality from software-heavy composition, lower absolute rates than the 1999 analogue, and accounting-related CAPE inflation from post-SFAS 141R write-downs. None of them erases the concern. They shift the calibration slightly.
• The equity risk premium framing is the more complete measure: at CAPE 41, the cyclically adjusted earnings yield of ~2.4% minus the current 10-year Treasury yield of ~4.5% produces a negative ERP of roughly -2 percentage points — one of the least favorable starting conditions for equity returns in the modern dataset.