The negative equity risk premium is not the warning. The warning is the route by which it closes. When the S&P 500's earnings yield sits below the 10-year Treasury yield, the gap must eventually close through lower yields, lower prices, or higher earnings. At today's starting valuation, the most comfortable of those routes, falling bond yields, is also the one most likely to bring the earnings disappointment that undoes the rescue.
Lance Roberts' rebuild of the premium from raw data supplies the inputs. At the October 6 record close of 7,818.93, the index traded near 27 times trailing as-reported earnings on Robert Shiller's figures, or 26.4 on an independent count. That is an earnings yield of roughly 3.7%. The 10-year closed at 5.31%. Stocks therefore "earn" about 1.6 percentage points less than a government note. In Shiller's monthly series since 1950, the average spread is about +1.0 point. Outside the 2009 earnings collapse, the last time the spread sat this far below zero was 2002.
The headline makes this sound like a sell signal. Roberts' own sort of the history shows that it is not one. The mechanism underneath explains why.
The Sign Matters Less Than the Starting Price
Roberts sorted every month since 1950 by its starting premium, then measured the S&P 500's real total return over the following ten years. A negative premium alone did not predict poor returns. The early 1980s are the counterexample. The 10-year yielded in double digits, stocks earned less than bonds month after month, and buyers then collected one of the strongest decades on record.
Starting price is what separated those cases. In the early 1980s the earnings yield ran above 5%, at times above 12%. Stocks were cheap and bonds were cheaper still. On Roberts' numbers, negative-premium months with an earnings yield above 5% averaged a 10.3% real annual return over the next decade, and only about one in twenty lost money. Negative-premium months with an earnings yield below 5% averaged 3.7% real, and more than one in four ten-year stretches finished negative. Those months cluster in the late 1990s and around the dot-com peak.
The spread is a difference of two yields, and a difference carries no information about levels. A -1.6 point spread built from a 12% earnings yield against a 13.6% bond yield describes a cheap market priced off a bond bubble in reverse. The same spread built from 3.7% against 5.3% describes an expensive market priced as though it shares the bond's safety. The number is identical and the structure is opposite.
So the useful question to ask of a negative premium is not "how negative?" It is "which leg is out of line?" In 1982 the bond yield was the outlier, and when it normalized, both assets rallied. In 2026 the earnings yield is the outlier, sitting well below its own history. Normalization has to come from somewhere else.
Three Ways a Spread Closes
Roberts frames the resolution as binary: bond yields fall, or stock prices do. A third path exists, and the bull case lives on it. Earnings can grow into the price.
The arithmetic is direct. If trailing earnings grow 30% while the index stays flat, the trailing earnings yield moves from 3.7% to about 4.8%. That is real progress, but it is not enough. The forward numbers already encode the growth: FactSet's forward P/E of 19.0 implies an earnings yield of about 5.3%, which lands on top of the 10-year. Even if Wall Street's optimistic earnings path is delivered exactly, flat prices and flat yields only bring the premium back to roughly zero. Restoring the post-1950 average of +1.0 point needs some combination of the other two routes on top.
That changes how the bull case should be read. Delivered earnings growth does not create a margin of safety. It closes most of a deficit. Any shortfall against a near-30% growth rate leaves the gap where it is now, and the remaining routes are the ones that hurt equity holders or require macro help.
Why the Bond-Side Rescue Is Not Free
Roberts notes the irony himself. He expects yields to decline over time because of the disinflationary drag of debt on growth, and if they do, "the premium rebuilds from the bond side, and stocks get rescued by the same math that hurts them today."
The mechanism behind that rescue matters. In the 1980s, yields fell because inflation was being broken by very high real rates, while nominal growth stayed healthy enough for profits to expand. Falling yields lowered the discount rate on a cheap and growing earnings stream, so both legs of the valuation moved in the equity holder's favor at once.
A decline in yields driven by debt-induced disinflation works differently. Yields fall because nominal growth slows. Slower nominal growth reduces revenue growth, and with it the earnings path that makes the forward 5.3% yield possible. The bond leg falls toward the earnings yield while the earnings yield is being revised down toward the bond leg. The premium can rebuild on paper while the index goes nowhere or falls, because the denominator of the P/E is shrinking at the same time the discount rate eases.
This is where intuition goes wrong. Investors tend to treat "yields fall, stocks re-rate" as a single favorable event. It is only favorable when the cause of the yield decline leaves earnings intact. A rescue that arrives through slowing growth, when the index is priced for near-30% profit growth, is the late-1990s-into-2002 sequence Roberts names as the scenario where the analogy "stops being an analogy."
The near-term rate setup points the other way in any case. By Roberts' account, markets price roughly a 20% chance of an October hike and about 80% odds of a move by December, with August headline CPI at 3.4% and Brent above $100. A bond-side rescue needs inflation to cooperate first. As of mid-October, futures are pricing the opposite direction for policy.
The Case Against the Read
The strongest counterargument is the inflation-adjusted comparison, and Roberts presents it fairly. Measured against inflation-protected Treasuries instead of nominal ones, the equity premium stays positive, at roughly 0.8 to 2.3 points depending on whose earnings estimate is used. Equities are real assets over long horizons. If 5.3% nominal yields mostly compensate for expected inflation that corporate pricing power can pass through, then the nominal comparison overstates how unattractive stocks are.
That objection carries weight, and it narrows the claim. The real-yield premium is thin, though, and its upper end depends on the same near-30% earnings growth arriving on schedule. On the trailing earnings that have actually been reported, the cushion is close to the bottom of that range. A premium of under one point against a risk-free real asset is small compensation for owning an asset that has historically suffered drawdowns large enough to need a 111% gain to recover from a 52.6% fall.
The second objection is horizon. The premium is a ten-year signal, and Roberts says plainly that the calendar, earnings, and buybacks still favor stocks into year-end. Both statements can be true together. A poor ten-year starting point says little about the next ten weeks. It says a great deal about how much of the next decade's return has already been pulled forward into today's price.
The third objection is the 1980s themselves. If yields were to fall sharply for benign reasons, such as a supply-side productivity gain that lowers inflation without slowing real growth, the bond-side rescue would arrive without the earnings damage. That path exists. It requires the AI capex cycle to show up as broad productivity rather than concentrated profits in a few names, and Roberts' own warning that "a record standing on a few names is NOT the same as a broad advance" suggests that evidence is not yet in.
What This Week's Data Actually Tests
Because the premium is a level argument, it will not resolve in a week. This week's calendar does test which closing route is gaining ground, and each release maps onto one leg of the spread.
Wednesday's September CPI is the bond leg. A soft print that pulls the 10-year back toward 5% narrows the gap from the rate side. A hot print revives December hike odds and widens the deficit. With the premium already negative, that move matters more than it would at a normal spread.
Bank earnings on Tuesday and Wednesday test whether growth can coexist with 5% yields. Banks are expected to grow earnings 15.2%, against 4.4% for financials overall. Loan growth and steady credit quality at these rates would support the benign version of events, in which the economy absorbs high yields without breaking. Rising provisions would signal that the disinflationary path is starting from the credit side.
TSMC on Thursday and ASML on Wednesday test the earnings leg directly. The forward earnings yield of 5.3% depends on AI-linked profit growth continuing. After the recent chip selloff, a soft read on demand from either company would hit the one route by which the premium closes without lower prices or lower yields.
The read changes on a specific combination. If the 10-year falls while forward earnings estimates hold or rise, the bond-side rescue is working as the bulls hope, and the negative premium resolves the way it did in the 1980s, through the rate leg, with equity holders compensated. If the 10-year falls while forward estimates are cut, the premium is rebuilding through the 1997-to-2002 route, and the falling yield that looks like relief is evidence of the problem. That divergence between yields and earnings revisions is what decides it, not the sign of the spread.
