Stock valuations · Report analysis
Fed economist: why the CAPE ratio cried wolf for a decade
A Fed economist’s working paper says accounting rules inflated the Shiller CAPE after the 1990s, and that a corrected version now shows stocks are expensive.
A decade of false alarms
The cyclically adjusted price-to-earnings ratio divides the S&P 500 by the average of the previous ten years of inflation-adjusted earnings. Robert Shiller popularized it at the peak of the dot-com bubble, and central banks still use it to gauge equity valuations. From January 2011 it sat mostly above its 80th historical percentile and implied high odds of a market correction. Instead, the S&P 500 more than doubled between January 2011 and its February 2020 peak. The ratio had also shifted: from a long-run mean of about 15 it climbed to a new plateau near 28 in the early 1990s and never came back. A structural-break test in the paper dates the shift to March 1995, with the average moving from 14.8 before to 28.4 after.
Two accounting rules moved the denominator
Palazzo, an economist at the Federal Reserve Board writing in a personal capacity, argues the shift is largely a measurement problem. Since FASB Statement No. 2 was issued in 1974, U.S. companies have had to expense research and development as they spend it, while a factory is capitalized and depreciated over its life. As R&D-heavy firms gained weight in the index, that rule took a growing bite out of reported earnings. Then mid-1990s standards, including EITF Issue 94-3 on restructuring costs and SFAS No. 121 (1995) on asset impairments, standardized how companies record large one-time charges, and those “special items” became far more common. Because CAPE averages ten years of earnings, both effects worked into the ratio gradually.
What changed in the earnings data
| Measure | Before Dec. 1991 | After Dec. 1991 |
|---|---|---|
| R&D expense | 16.0% | 21.9% |
| Special items (write-downs) | 1.2% | 11.7% |
| GAAP earnings as a share of adjusted earnings | 82.8% | 66.4% |
| Average traditional CAPE | 15.2 | 27.6 |
| Average CAPE-H | 14.7 | 19.3 |
Adjusted earnings here means GAAP net income before special items with R&D added back. When reported earnings capture only two-thirds of that figure instead of four-fifths, a price-to-earnings ratio rises even if prices and businesses have not changed at all.
CAPE-H: the same ratio with comparable earnings
The corrected measure keeps everything about CAPE except the earnings definition: same prices, same ten-year average, same inflation adjustment, but earnings measured as GAAP income minus special items plus R&D, built from Compustat data for S&P 500 members. It equals CAPE before 1967 by construction, because ten years of adjusted data are not yet available. The paper is explicit that this is a consistency correction, not an estimate of “true” economic profit; it does not try to build an R&D capital stock or amortize past research spending. The correction does not erase the rise in valuations. Median CAPE-H climbed from 14.1 before December 1991 to 19.1 afterward, a 36% increase, compared with a much larger jump in the traditional ratio.
The forecasting record flips
| Outcome | CAPE | CAPE-H |
|---|---|---|
| Price change (capital gain) | −60.2% | +14.3% |
| Excess return over T-bills | −70.4% | +10.4% |
| Dividend growth | −45.2% | −7.3% |
The statistic is out-of-sample R², the standard test from Welch and Goyal (2008): each forecast uses only data that existed at the time and is scored against simply assuming the historical average return. Negative means the predictor did worse than that naive average. Traditional CAPE failed badly after 1991; CAPE-H beat the benchmark, and the improvement comes through prices rather than dividends. Splitting the correction apart, the paper attributes about 73% of the gain to the R&D adjustment and 27% to excluding special items, and only the combination stays positive through the end of the sample.
Were stocks overvalued in the 2010s?
Mostly not by as much as CAPE said. Measured with only the data available at each date, traditional CAPE ranked between the 75th and 98th percentiles of its history from 2011 through 2020. CAPE-H ranged from the high 30s to the low 90s, with a median near the 69th percentile. In January 2015, for example, CAPE implied a 47.1% chance that the following five years would land in the worst quarter of historical five-year S&P 500 outcomes; CAPE-H put that probability at 29.4%. An investor who trusted CAPE spent the decade braced for losses that the corrected measure never strongly predicted.
What the corrected measure says now
The convergence is the paper’s headline warning. CAPE-H rose sharply after 2020 while traditional CAPE stayed high, and by late 2025 both sat above their 97th historical percentiles. As of December 2025 the paper’s models put the probability of a five-year “correction” at 61.8% using CAPE and 60.0% using CAPE-H, against 25% in an average period. A correction here is a statistical definition, not a set percentage drop: a cumulative five-year S&P 500 change in the bottom quarter of its historical distribution. The quantile model’s median five-year price-change forecast is roughly flat, about −4 log points, with a wide range below it. The paper’s concluding line reverses its title: “CAPE-H is crying wolf.”
How to read it without overreacting
This is a working paper posted on SSRN, not peer-reviewed research or an official Federal Reserve publication, and the author notes the views are his own rather than the Board’s. Figures here come from the September 18, 2026 revision; earlier drafts reported somewhat different numbers. Valuation measures forecast multi-year odds, not dates, and the paper finds its crash signal weakest exactly at the five-year horizon, where rare events give the statistics little to work with. CAPE-H’s high-valuation periods have also run long: 1885 to 1907, 1958 to 1973, 1995 to 2008, and 2013 onward, each lining up with a wave of new technology, and high CAPE-H readings were followed by faster productivity growth in industries that make capital goods. Use the finding as context for expected returns and risk, measured after inflation, rather than as a signal to trade. Track the traditional ratio on our Shiller CAPE ratio page, compare long-run asset returns on the market returns page, and read how R&D accounting distorts P/E ratios.
