S&P 500 Equity Risk Premium Turns Negative Again: A 36-Year History Shows This Has Happened Three Times Before, With Opposite Results

Correction October 31, 2026: the headline and opening line were corrected from two prior negative episodes to three. The article itself documents 1990, 2000 and 2009, so the current reading is the fourth, and the 2004 to 2023 positive stretch now notes the 2009 exception. Source: U.S. Treasury daily par yield curve and multpl.com S and P 500 series, as cited at the end of this article.

As of the close on Monday, September 21, 2026, the S&P 500 traded at a trailing price-to-earnings ratio of 26.48, according to data compiled by Robert Shiller and S&P and published by multpl.com. Invert that and you get an earnings yield of 3.78%. On the same day, the 10-year U.S. Treasury note yielded 4.96%, confirmed directly from the U.S. Treasury's daily par yield curve rates. All index levels, yields and dollar figures in this article are quoted in U.S. dollars and reflect the composite close of U.S. exchange trading for the S&P 500 index (not any single constituent stock), as reported by the sources cited throughout. Subtract one from the other and stocks are paying investors 1.18 percentage points less than a 10-year government bond, before any consideration of growth or risk. That gap is called the equity risk premium, and a negative reading is rare enough that it has only shown up in a handful of distinct stretches since 1990. This is the fourth one. The three before it ended in very different ways.

The New York Stock Exchange, Broad Street, Manhattan (Photo: Arild Vågen, Wikimedia Commons, CC BY-SA 4.0, original file page)

What the gap actually measures

The equity risk premium used here is the simplest version of the concept: the S&P 500's trailing earnings yield (net income per share divided by index price) minus the 10-year Treasury yield. It is not a forecasting model and it says nothing about next quarter's returns. What it does capture is a valuation fact: when it is positive, a dollar of index earnings currently costs less, relative to a risk-free bond, than average; when it is negative, that dollar of earnings is priced above what a 10-year Treasury pays, and the market is betting that earnings growth (or multiple expansion) will make up the difference. As of this week, the market is making that bet by 1.18 points. It is a different lens than the cyclically-adjusted Shiller CAPE ratio covered on this site earlier this month, which smooths earnings over ten years rather than using the latest trailing twelve months, so the two indicators can and sometimes do disagree about how expensive the market looks.

Thirty-six years of the spread, year by year

Using the same Shiller/S&P earnings-yield series and Treasury-yield series, both taken as of January 1 of each year (with September 21, 2026 added as the current reading), the spread has moved through three distinct regimes rather than sitting near zero:

Date	S&P 500 earnings yield	10-year Treasury yield	Spread (equity risk premium)
Jan 1, 1990	6.61%	8.21%	-1.60
Jan 1, 2000	3.44%	6.66%	-3.22
Jan 1, 2004	4.40%	4.15%	+0.25
Jan 1, 2009	1.41%	2.52%	-1.11
Jan 1, 2023	4.38%	3.53%	+0.85
Jan 1, 2024	4.00%	4.06%	-0.06
Jan 1, 2026	3.57%	4.21%	-0.64
Sep 21, 2026 (current)	3.78%	4.96%	-1.18

Two things stand out. First, the spread was negative for most of 1990 through 2002 -- a full decade-plus -- and then flipped positive for a long, mostly stable stretch from roughly 2004 through 2023, the one exception being the brief negative reading at the 2009 crisis low covered below. Second, the current negative reading isn't a one-week accident: it has been negative or flat since 2024, meaning the market has now gone two full years without a comfortable equity risk premium after two decades of having one.

Case one, 1990: the false alarm

At the start of 1990 the spread was -1.60, driven mostly by a Treasury yield above 8% rather than by expensive stocks. Over the following ten years, the S&P 500 went from 339.97 to 1,425.59 (both Jan 1 index levels, Shiller/S&P data via multpl.com) -- a cumulative price gain of 319%, or roughly 15.4% a year in price terms alone, before dividends. A hypothetical $10,000 invested on that basis would have grown to roughly $41,900 in index price value ten years later. Anyone who avoided U.S. equities in 1990 because the earnings yield sat below the bond yield missed the best decade of the entire post-1960 sample. The reason: Treasury yields spent the 1990s falling from above 8% toward 5%, and falling rates lifted valuations even as the starting spread looked unfavorable.

Case two, 2000: the one that mattered

At the start of 2000, the spread was -3.22 -- the most negative reading in the modern series and roughly triple the depth of today's gap. This time the mechanism was different: earnings yields were low because dot-com-era prices had become extreme, not because bond yields had merely drifted up. The S&P 500 fell from 1,425.59 on Jan 1, 2000 to 895.84 by Jan 1, 2003, a cumulative price decline of 37% over three years (roughly -14.5% a year). Stretched to ten years, the index stood at 1,123.58 on Jan 1, 2010 -- a cumulative price loss of 21% from the 2000 starting point, or about -2.4% a year in price terms. A $10,000 investment made on that basis was worth roughly $7,880 in index price value a decade later, before counting the roughly 1.5-2% annual dividend yield the S&P 500 paid across the 2000s, which narrows but does not erase the shortfall. This is the "lost decade" that appears in almost every long-run market history, and it started from a spread reading not far from where we are today.

Case three, 2009: same sign, opposite cause

The spread also went negative heading into 2009, at -1.11 -- almost identical in size to today's -1.18. But the earnings yield collapsed that year because corporate earnings themselves had collapsed during the financial crisis, pushing the P/E ratio's denominator down while the index price was already near a multi-year low, not a high. The S&P 500 stood at 865.58 on Jan 1, 2009 and 2,607.39 by Jan 1, 2019: a ten-year cumulative price gain of 201%, or about 11.7% a year, one of the strongest decades on record. A $10,000 investment on that basis would have reached roughly $30,100 in index price value. The lesson isn't that negative spreads are bullish -- it's that the same number can appear because prices got too expensive (2000) or because earnings got temporarily crushed at a market bottom (2009), and those two situations call for opposite conclusions.

Where September 2026 actually sits

Today's setup does not map cleanly onto any of the three precedents. Unlike 1990, the 10-year Treasury yield is not falling from a double-digit peak -- it sits at 4.96%, close to its highest level of the past two years, which means today's negative spread is not a rates tailwind waiting to happen. Unlike 2009, index earnings have not collapsed; S&P 500 earnings have been reported as growing through 2026, which means the negative spread here is closer in mechanism to 2000: prices, not earnings, are doing the work. A companion factor also matters here: as covered in this site's September 16 piece on S&P 500 concentration, index-level earnings and price action are increasingly dominated by a small number of mega-cap technology companies, which can push the aggregate P/E ratio higher than the typical constituent's valuation would suggest. A negative aggregate spread driven by concentrated mega-cap pricing is a different animal from a broad-based valuation bubble across hundreds of names, even though both would show up as the same negative number in the table above.

What a negative spread does and doesn't tell an investor

In our own judgment, the most useful way to read this indicator is as a valuation thermometer, not a trading signal. It correctly flagged elevated risk ahead of the 2000-2002 decline. It said nothing useful in 1990, when falling rates rescued a similarly cheap-looking bond alternative into a historic stock rally. And in 2009 it fired for an entirely unrelated reason -- a temporary earnings collapse -- right as the market bottomed. Three data points cannot support a firm base rate, but they do support one conclusion with some confidence: the spread's sign alone is not enough information to act on. What separates 1990 from 2000 and 2009 is not the number itself but the reason behind it -- whether Treasury yields are falling, earnings are temporarily depressed, or index prices are simply expensive relative to trailing profits. On that test, 2026 shares more with 2000 than with either alternative, which is a reason for caution, not a reason for a forecast.

Why the starting point of the measurement matters

One methodological point is worth stating plainly: the earnings-yield and Treasury-yield figures compiled above are snapshotted on January 1 of each year, except for the current September 21, 2026 reading, which is a mid-year data point. That means the historical rows are not perfectly comparable calendar positions to today's reading, and a spread that looks similar in size can still reflect a different point in that year's earnings or rate cycle. It also means the true trough or peak of the spread in any given year could have occurred at a different date than January 1 and gone unrecorded in this table. None of the three case studies above change direction because of this, since each one persisted for multiple years rather than a single data point, but a reader building a more precise model should pull monthly, not annual, data before drawing sharper conclusions than the ones offered here.

A reader's checklist, not a trade idea

• Check whether a negative equity risk premium reading is being driven by rising bond yields, falling earnings, or rising prices -- the same number means different things depending on which one moved.
• Look at how concentrated the index's valuation is. A handful of large stocks trading at extreme multiples can make the aggregate S&P 500 earnings yield look worse than the median company's valuation.
• Remember that the "lost decade" and the "best decade" in this article both started from a negative spread -- the indicator alone did not distinguish between them in advance.
• Treat dividends, fees, and taxes as missing from every price-only calculation in this article; total returns for the 2000s decade were less negative, and for the 1990s and 2010s modestly higher, once dividends are reinvested.
• Diversification across sectors and market capitalizations, not a single valuation ratio, is what actually addresses concentration risk in a portfolio.

Tip: Episodes like 1990, 2000 and 2009 are easier to keep in perspective when you have read how earlier investors lived through them — a stock market investing book or two covers most of that ground. Writing down what you believed at the time in an investment journal is what lets you check yourself against it later. (These are Amazon Associate links — we may earn a small commission on qualifying purchases.)

Sources, as-of date, and disclaimer

S&P 500 trailing P/E ratio, earnings yield, and historical index price levels: Robert Shiller and Standard & Poor's data, compiled and published by multpl.com, accessed September 22, 2026, reflecting the market close of Monday, September 21, 2026. 10-year Treasury yield: U.S. Department of the Treasury, Daily Treasury Par Yield Curve Rates, September 2026. All dollar and percentage figures in this article are price-index calculations only, exclude dividends, transaction costs, and taxes, and are provided for general investment education. They are not a recommendation to buy, sell, or hold any security, and past index behavior does not predict future returns. Individual investors should consider consulting a licensed financial professional before acting on valuation data.