The CAPE ratio, one of the most closely followed gauges of U.S. stock valuations, spent the decade after the Global Financial Crisis warning of a reckoning that did not arrive. In a working paper dated September 18, 2026, Federal Reserve Board economist Dino Palazzo blames most of that misfire on two accounting changes and builds a corrected measure, CAPE-H, that forecasts five-year price gains far better than the original.
Where it stands: The paper’s data end in December 2025. At that point, Palazzo reports, CAPE and CAPE-H were both above their 97th historical percentiles, and the paper’s model put the probability of a five-year market correction at 61.8% using CAPE and 60.0% using CAPE-H. Those are the paper’s model estimates as of December 2025, not October 2026 readings, and “correction” has a specific statistical meaning in the paper, explained below. A footnote states that the views are the author’s own, not those of the Board of Governors or its staff.
What the CAPE ratio measures
CAPE stands for cyclically adjusted price-to-earnings ratio. It divides the price of the broad U.S. stock market by the average of the previous ten years of company earnings, adjusted for inflation, so a single boom year or a single recession cannot swing the ratio on its own. Robert Shiller, the Yale economist best known for the measure, writes on his data site that he “developed the Cyclically Adjusted Price Earnings ratio with John Campbell.” His public data set holds monthly stock prices, dividends, earnings and consumer prices “to allow conversion to real values,” going back to January 1871.
New to the plain multiple? Start with our price-to-earnings ratio explainer.
The ratio matters beyond stock-picking: Palazzo notes that the European Central Bank uses it directly in a financial conditions index for the United States. The paper’s starting puzzle is that, beginning in the early 1990s, CAPE climbed from a historical mean of about 15 to a new level near 28 and has stayed up for more than thirty years. A break test in the paper dates the shift to March 1995, with the average rising from 14.8 to 28.4, an increase of about 90%.
Shiller has flagged one earnings problem himself: since September 2018 his data page has also offered a total return CAPE, because buybacks can change how fast earnings per share grow. Palazzo’s paper targets two different distortions.
Why CAPE-H exists: two accounting rules
Palazzo argues that today’s CAPE uses earnings recorded under different rules from most of its history, so its long-run average mixes two yardsticks. The paper calls the fix “a measurement-consistency correction rather than an attempt to model economic earnings.” CAPE-H puts old and new profits on one footing; it does not claim to know what companies truly earn.
Rule one: research spending is expensed. Under Financial Accounting Standards Board (FASB) Statement No. 2, issued in 1974, companies charge research and development to expense right away instead of recording it as an asset. A heavy research spender therefore reports lower profit now, even if the work pays off for years. As the economy tilted toward technology, pharmaceuticals and other research-heavy industries, the paper finds research spending rose from 16% of adjusted earnings before 1992 to 22% afterwards. When research runs at roughly a fifth of adjusted earnings, Palazzo writes, “reported earnings persistently understate sustainable profitability,” pushing the measured P/E higher even if prices do not move. We walk through the mechanics in why mandatory R&D expensing makes the stock market look more expensive.
Rule two: special items became routine. Two mid-1990s standards set the accounting for charges usually reported as special items. Emerging Issues Task Force (EITF) Issue 94-3 dealt with severance and other restructuring exit costs, and SFAS No. 121, from 1995, set rules for impairments of long-lived assets. Recognition of these items “rose substantially in the years that followed,” the paper says, from 1.2% of adjusted earnings to 11.7%. Because write-downs tend to cluster in bad years, special items add what Palazzo calls “substantial cyclical volatility concentrated in downturns” to the ten-year average under CAPE. One recession’s wave of write-offs can hold that average down, and the ratio up, for a decade. Our piece on how special-items write-downs distort the CAPE ratio has more.
Reported profits have also become far noisier. Relative to sales, GAAP earnings volatility went from 1.77 in 1957–1979 to 7.02 in 2003–2025, while the adjusted figure moved only from 1.53 to 2.55. CAPE-H uses that steadier number: take reported earnings, remove special items, put research outlays back in, then run the usual ten-year, inflation-adjusted calculation.
CAPE vs CAPE-H: the numbers behind the gap
Table 2 of the paper splits its sample at December 1991, which the paper calls the onset of the high-valuation regime. Before that point, reported earnings averaged 82.8% of adjusted earnings; afterwards, only 66.4%.
| Item (Table 2) | Pre-regime, from 1957 | High-valuation regime, through 2025 |
|---|---|---|
| Research outlays vs. adjusted profit | 16.0% | 21.9% |
| Write-downs and other special items vs. adjusted profit | 1.2% | 11.7% |
| GAAP profit vs. adjusted profit | 82.8% | 66.4% |
| Traditional CAPE, average | 15.2 | 27.6 |
| CAPE-H, average | 14.7 | 19.3 |
The two series tracked each other for decades, began to separate in 1984 and drifted far apart after 1991. Across the high-valuation regime, CAPE averaged 27.6 against 19.3 for CAPE-H, a difference the paper puts at approximately 43%.
The correction does not erase the rise in valuations, and Palazzo says so. The median CAPE-H climbs from 14.1 before December 1991 to 19.1 afterwards, an increase of 36%, which the paper calls economically meaningful but much smaller than the jump in CAPE. With measurement fixed, Palazzo writes, current valuations look “cyclically elevated within historical ranges, rather than representing either a permanent regime shift or unprecedented overvaluation.”
The cycle model behind that reading covers 145 years. In it, the expected length of an expensive spell is 10.3 years and of a cheap one 8.1 years, and four expensive spells lasted more than a decade, each coinciding with a technology wave. The latest began in 2013, and the paper ties it to digital platforms.
Why CAPE cried wolf from 2011 to 2020
From 2011 through 2020, CAPE sat between the mid-70s and the high-90s percentile of its own history, slipping just under the 80th percentile only briefly in late 2011 and in a few months of 2012. Throughout, Palazzo writes, it “continuously predicted elevated probabilities of both market corrections and crashes.” Those outcomes did not come within the forecast windows: the S&P 500 more than doubled between January 2011 and its pre-pandemic peak in February 2020. In the paper’s words, “This sustained period of false alarms materially weakened CAPE’s credibility as a forward-looking valuation indicator.”
The corrected measure tells a quieter story for the same decade. Over 2011–2020, CAPE ranked in the 75th–98th percentile range, while CAPE-H ran lower and moved around more, anywhere between the high-30s and the low-90s, its median sitting near the 69th percentile. In January 2015, CAPE implied a 47.1% chance of a five-year correction; CAPE-H implied 29.4%, a gap of 17.7 percentage points. On Palazzo’s evidence, much of CAPE’s early-2010s alarm came from accounting, not from genuinely high valuations.
Does CAPE-H forecast returns better?
On the paper’s tests, yes, though only for one part of returns. Over the full sample, CAPE-H explains 25% of the variation in returns, against 8% for CAPE. In 2002–2025, when the two diverge most, CAPE’s explanatory power for excess returns drops to 10% with an insignificant coefficient, while CAPE-H’s reaches 37%.
The tougher test is out-of-sample forecasting, which scores predictions against data that were not used to fit the model. The paper reports an out-of-sample R² (R²OOS), and a negative value means the forecast did worse than the plain historical mean. At five years, CAPE falls below that benchmark for both excess returns and capital gains. Expressed in percent, the five-year capital-gain R²OOS is −60.23 for CAPE and a positive, statistically significant 14.31 for CAPE-H, which Palazzo describes as “a swing of roughly 75 percentage points relative to CAPE.”
Which adjustment does the work? Scoring each fix on its own, the paper credits approximately 73% of the improvement to adding back R&D and 27% to removing special items. A research-only version still finishes the sample with a negative R²OOS; it takes both adjustments to turn the result positive.
What CAPE-H cannot do is forecast dividends. Neither measure predicts dividend growth reliably out of sample, at any horizon, so the gain comes through price appreciation. Why that split matters for income-focused investors is the subject of our explainer on dividend growth versus price gains.
Corrections and crashes, as the paper defines them
The tail-risk results rest on two definitions that differ from everyday market talk. A correction is a cumulative S&P 500 change, over the forecast horizon, that ranks in the bottom 25% of historical outcomes for that horizon. A crash is one that ranks in the bottom 10%. Both are measured over the same one-, three- or five-year window, so a five-year correction means a five-year market path among the worst quarter on record, not a single sharp drop.
With those definitions, horizon matters. At one and three years, CAPE-H helps predict both corrections and crashes, which the paper reads as high valuations raising near-term fragility. At five years, its record on corrections improves further, while its record on crashes fades. Against a fixed-rate benchmark, CAPE-H’s log-predictive score for corrections improves by “+4.0 at one year, +33.3 at three years, and +54.6 at five years.” For five-year crashes it scores −36.6, below the null model. Palazzo’s explanation is the difference between slow valuation normalization and sudden shocks “whose timing remains inherently unpredictable.”
Sorting history into valuation fifths agrees. At the five-year horizon, starting points in the highest CAPE-H quintile were followed by average S&P 500 changes roughly 54 percentage points below those in the lowest quintile, and the correction rate in the top quintile reached 46%, compared with an unconditional frequency of 25%. Our shorter pieces cover what CAPE-H signals for corrections at one, three and five years and why five-year crash forecasts lose their power.
Where CAPE and CAPE-H stood at the end of 2025
The paper’s final section turns its title around: “CAPE-H is crying wolf.” By late 2025, CAPE and CAPE-H were both above their 97th historical percentiles. Palazzo stresses that this differs from the 2011–2020 episode. Back then, the gap between the two measures came from accounting; by late 2025 they had converged because CAPE-H rose to meet CAPE “through a genuine valuation cycle rather than CAPE correcting downward.”
The probabilities converged too. By December 2025, the modeled five-year correction probability was 61.8% from CAPE and 60.0% from CAPE-H, against the 17.7-point gap of January 2015. At the end-2025 CAPE-H reading of 26.22, the model put the median predicted five-year return at −4.1 log points and the 10th percentile at −59.5 log points. A separate passage on post-2020 predictions puts the latest five-year 10th-percentile prediction, at the end of the sample, near −46% and the median near −4%.
| Indicator | CAPE | CAPE-H |
|---|---|---|
| Earnings in the denominator | Reported GAAP | GAAP with R&D added back, special items removed |
| Average after Dec. 1991 | 27.6 | 19.3 |
| Percentile range, 2011–2020 | 75th–98th | High-30s to low-90s |
| Five-year capital-gain R²OOS, in percent | −60.23 | 14.31 |
| Modeled five-year correction odds, Jan. 2015 | 47.1% | 29.4% |
| Modeled five-year correction odds, Dec. 2025 | 61.8% | 60.0% |
| Standing in late 2025 | Above 97th percentile | Above 97th percentile |
The conclusion says current conditions “signal an elevated probability of a market correction in the next 5-years.” Read that with the paper’s definitions in mind. It is a model probability, estimated on data ending in December 2025, that the S&P 500’s five-year path lands in the bottom quarter of history. It is not an October 2026 reading, not a date and not a crash call, since the same paper finds five-year crash timing unpredictable. Life Stonks makes no forecast of its own.
What changed from the earlier draft
If you read about this paper over the summer, you may have seen different numbers. The Idea Farm’s July 2026 summary quoted an earlier draft, which we could not read ourselves, so this comparison relies on that summary:
- Reported versus adjusted earnings: the summary cited 82.3% before 1992 and 66.6% afterwards; the September version reports 82.8% and 66.4%.
- Average CAPE after 1991: 27.8 against 19.3 for CAPE-H in the summary; 27.6 against 19.3 in the September version.
- Size of the gap: the summary described a 44% valuation overstatement; the September version says CAPE averaged approximately 43% above CAPE-H.
- Forecasting: the summary said CAPE-H lifts out-of-sample price-appreciation R² from −29% to +18%. The September version’s five-year capital-gain figures are −60.23 for CAPE and 14.31 for CAPE-H, in percent. The summary does not name a horizon, so the two pairs may not measure the same thing.
As the summary describes it, the core argument is the same. Where figures differ, this article uses the version dated September 18, 2026, posted on SSRN; Palazzo’s research profile is on the Federal Reserve Board’s website.
Frequently Asked Questions
What is CAPE-H?
CAPE-H, short for Historically-comparable CAPE, is a version of the cyclically adjusted price-to-earnings ratio built by Federal Reserve Board economist Dino Palazzo. It keeps CAPE’s prices and ten-year inflation-adjusted averaging but adds R&D spending back to earnings and removes special items such as write-downs, so profits recorded before and after the accounting changes of 1974 and the mid-1990s can be compared.
Why did the CAPE ratio give false signals after 2011?
According to the paper, from 2011 through 2020 CAPE stayed in roughly the 75th to 98th percentile of its history and kept implying high odds of corrections and crashes, yet the S&P 500 more than doubled from January 2011 to February 2020. Palazzo attributes much of CAPE’s elevation in that period to R&D expensing and special items rather than to genuinely high valuations.
What did CAPE-H show at the end of 2025?
In data through December 2025, CAPE and CAPE-H were both above their 97th historical percentiles, and the paper’s model put the five-year correction probability at 61.8% with CAPE and 60.0% with CAPE-H. Correction, in the paper, means the index’s cumulative move over five years ranks among the worst 25% on record. These are end-2025 model estimates, not current readings or a forecast.
Can CAPE-H predict a stock market crash?
Only over shorter horizons, on the paper’s tests. At one and three years CAPE-H helps predict both corrections and crashes, with a crash meaning a cumulative change in the bottom 10%. At five years its crash forecasts score below a simple benchmark, and the paper describes the timing of such shocks as inherently unpredictable.
Is CAPE-H the Federal Reserve’s view?
No. It comes from a working paper written by Dino Palazzo, who works in the Capital Markets Section of the Federal Reserve Board’s Research and Statistics division. The paper states: “The views expressed in this paper are solely those of the author and do not necessarily reflect the views of the Board of Governors of the Federal Reserve System or its staff.”
Sources
- The CAPE That Cried Wolf (version dated September 18, 2026) — Dino Palazzo, Federal Reserve Board (working paper, SSRN), September 18, 2026
- Dino Palazzo: Meet the Researchers — Board of Governors of the Federal Reserve System, page last updated February 26, 2026
- Online Data — Robert J. Shiller, Yale University, accessed October 10, 2026
- Shiller Data — Robert J. Shiller, accessed October 10, 2026
- The CAPE that Cried Wolf — The Idea Farm (summary of an earlier draft), July 2026
This article is for general information only and is not investment, tax or legal advice. Life Stonks does not recommend buying or selling any security, and nothing here predicts what any market will do. It summarizes a working paper by a Federal Reserve Board economist; the paper states that its views are the author’s own and not those of the Board of Governors or its staff. Talk to a licensed professional before making a financial decision.