The usual bubble debate is a fight between two stories. One side says prices are high because investors expect faster growth. The other says prices are high because investors have lost discipline. A new paper by Andreasen and Bro in the Review of Finance (2026) gives an uncomfortable answer to the bull side. Over monthly U.S. data from January 1971 through December 2024, expected dividend growth contributes little to explaining swings in the S&P 500 price-dividend ratio. The estimated speculative component accounts for 61% of that variation, and expected returns account for 42%.
The intuition the paper overturns is that valuation booms are mostly growth booms with some froth on top. In this model the froth is the main event. It also has a macroeconomic footprint: it lowers unemployment, lifts industrial production, and draws a policy-rate response of roughly 48 basis points.
The Growth Story Explains Almost Nothing
Start with the arithmetic of a present-value model. A stock's price-dividend ratio can rise for three reasons. Investors expect dividends to grow faster. Investors accept lower future returns, which raises the price they will pay for the same cash flows. Or prices contain a component unrelated to either, which the authors call a rational bubble: a non-fundamental piece of value that rational holders will carry as long as they expect it to keep appreciating.
Most practitioners, asked why the market trades at a rich multiple, reach for the first channel, because growth expectations are what earnings calls, strategist notes and AI capex narratives are built around. The paper's full-sample result says that channel is close to silent. Fluctuations in the ratio come from changes in required returns and from the bubble term, not from revisions to the dividend outlook.
The two variance figures, 61% and 42%, add to more than 100%. The authors warn that these contributions are not additive shares of a fixed total. One reading of the overshoot is that the components partly offset each other: periods when the speculative piece is swelling may also be periods when the required-return piece moves the other way. If so, the visible multiple understates how much speculation is underneath it, because part of the bubble is masked by shifts in discount rates.
A second result strengthens the case. Unadjusted price-dividend ratios show excess volatility and statistical evidence of explosive dynamics, two long-standing puzzles in asset pricing. Strip out the estimated bubble and the excess volatility falls substantially and the explosive signature disappears. Whatever the bubble term is capturing, it is the part of the series that standard models could never explain.
How Surveys Pry the Bubble Loose
The identification problem is old. Market data alone cannot separate a rational bubble from a shift in expected returns, because both raise prices relative to dividends without any change in cash flows. Econometricians have spent decades on tests that cannot tell the two apart.
The authors bring in a third input: survey-based expectations of future dividend growth and future stock prices. This is the core of the method. If investors report modest dividend expectations while also expecting prices to keep climbing, the gap has to be closed by something. A lower required return can absorb part of it. The remainder, price appreciation that neither the dividend outlook nor the discount rate can justify, is assigned to the bubble.
So the model measures speculation as the share of price expectations that has come loose from cash-flow expectations. It does not measure how "expensive" the market looks. A high ratio paired with strong dividend expectations would register as fundamental. A high ratio paired with flat dividend expectations and confident price expectations registers as a bubble.
The historical estimates follow a recognizable pattern:
- Early 1970s Nifty Fifty: roughly 10% of the price-dividend ratio.
- 1980s: declining to around 2%.
- Dot-com peak: 27% in January 2001.
- COVID-era boom: about 19% in November 2021.
- December 2024: nearly 25%, which the authors suggest may reflect enthusiasm around artificial intelligence.
The sequence matters as much as the peaks. The dot-com reading tops out in January 2001, after the Nasdaq had already broken in March 2000. The bubble estimate builds and fades gradually rather than collapsing on a single date. That is a feature of the model, and it is why the authors say identification is not a timing signal.
A Bubble That Hires Workers
The paper then moves from markets to the real economy. Using innovations in the estimated bubble as an instrument in a structural vector autoregression, the authors find that positive bubble shocks are followed by lower unemployment, higher industrial production and longer working hours.
Two transmission channels fit. Households whose portfolios rise feel wealthier and spend more. Firms whose equity is richly valued can raise capital more cheaply and loosen financing constraints, which supports investment and hiring. Neither channel requires the bubble to be grounded in anything. A share price that is 25% speculative still pays for the same equipment when a company issues stock against it.
Inflation responds only modestly, and the response is generally not statistically significant. That makes bubble-driven expansions an awkward case for an inflation-targeting central bank. Activity strengthens and labor markets tighten, but the price signal that would normally justify tightening is weak. A bubble expansion looks more like a supply-friendly boom than a classic overheating, which helps explain why such episodes can run for years without an obvious policy trigger to end them.
Why the Fed's Response Leaves Conditions Looser
Policy does respond. The estimated policy-rate increase peaks at about 48 basis points eight months after a one-standard-deviation positive bubble shock. At the same time, the spread between long-term Treasury yields and short rates narrows, and corporate bond spreads tighten.
Put those together. Short rates rise, long rates rise by less, and credit becomes cheaper. The tightening lands on the front end while the parts of the financial system that fund long-duration projects and corporate balance sheets get easier. The speculative component pulls financial conditions in two directions at once, and the net effect on the cost of capital for a firm issuing long-dated debt or equity can be accommodative even as the policy rate climbs.
The authors are careful about causation here, and the caution is warranted. The analysis does not show that policymakers respond to the bubble itself rather than to the stronger employment and production that follow it. For a portfolio, the distinction matters less than the sequence: a speculative equity rally tends to arrive alongside a firmer labor market, a modestly higher policy rate, a flatter curve and tighter credit spreads. A reversal would plausibly unwind that package. The paper tests positive shocks and does not establish that negative shocks produce effects of equal size in the other direction.
Where the Decomposition Could Be Wrong
The strongest objection concerns the choice of yardstick. The price-dividend ratio was a cleaner valuation measure in 1971 than it is today. Over the sample, U.S. companies shifted a large share of shareholder payouts from dividends to buybacks. A market that returns cash through repurchases will show a structurally rising price-dividend ratio even if total payout expectations are unchanged. If dividend-growth expectations in the surveys do not fully reflect that shift, part of what the model labels a bubble could be a change in payout policy. The summary does not say how the authors handle this, so the concern is open rather than settled.
The second objection is about the surveys. Survey expectations of stock prices are known in the broader literature to extrapolate recent returns. The paper's bubble is "rational" in the sense of being consistent with its model's rational-expectations assumptions. If the survey respondents are simply chasing momentum, the estimated component could be measuring belief error instead of a rational bubble. For practical purposes the label matters less than the behavior, since either version is a valuation layer that depends on continued optimism. For the paper's theoretical framing, it matters a great deal.
Third, the estimates carry uncertainty bands. The 25% reading for December 2024 sits just below the 27% dot-com peak, and the gap between the two is small compared with the estimation error typical of models of this kind. Ranking the AI episode against the dot-com era on point estimates claims more precision than the method supports.
None of these objections reverses the central finding that dividend-growth expectations explain little of valuation variation. They do weaken any reading of the bubble series as an exact measure.
What the December 2024 Reading Does and Does Not Say
The sample ends in December 2024. The model says nothing about what has happened since, and the authors state explicitly that the estimate is not evidence that AI-related stocks were overvalued or that a correction was inevitable. Readers looking for a sell signal will not find one.
What the paper provides is a different decision variable from the multiple. The useful question about any valuation boom is whether investors' price expectations are running ahead of their cash-flow expectations, and whether that gap is widening. When it widens, the market's level depends more on continued belief than on earnings, and the real economy borrows from that belief through hiring and cheap financing. A concentrated equity book, a capital-market assumption built on current employment, and a credit portfolio enjoying narrow spreads are then exposed to the same variable.
The read changes if survey dividend expectations start rising to meet price expectations, which would shift the gap back toward fundamentals, or if a payout-adjusted version of the model shrinks the recent bubble estimate substantially. Until one of those happens, the paper's evidence points one way: in the U.S. market since 1971, the growth story has done the least of the work it usually gets credit for.
