Ask any finance professor what drives stock prices, and the standard answer is discount rates—the expected return investors demand for holding risky assets. When investors become more fearful, they demand a higher return, and prices fall. When optimism returns, the required return drops, and prices rise. This is the backbone of modern asset pricing, most famously articulated by John Cochrane in his 2011 presidential address to the American Finance Association.
If that story is right, then valuation ratios—measures like the price/dividend ratio or the Shiller CAPE (cyclically adjusted price-to-earnings)—should be useful for predicting future returns. When stocks are expensive relative to their dividends or earnings, expected returns should be lower; when they’re cheap, expected returns should be higher.
The reality is that, while these ratios are the best predictors we have had, they are not strong predictors in practice. Sebastian Hillenbrand and Odhrain McCarthy, authors of the March 2026 study “Expected Returns with Cash Flow Trends and Cycles,” set out to understand why traditional valuation ratios have uniformly failed to beat the historical mean in US markets.
The key finding: A signal buried in noise
The authors’ central insight is deceptively simple. The price/dividend ratio reflects not one thing, but three: the market’s expectation of future returns, the long-run trend in cash flow growth, and short-term cycles in cash flows. Standard predictive regressions treat the entire ratio as a return signal—but two-thirds of it is actually noise from the cash flow side.
Think of it this way. If a company’s dividend has permanently grown faster—say, because the whole economy is more productive—its stock price will be structurally higher relative to today’s dividend. A naive observer looking only at the elevated price/dividend ratio would conclude that future returns must be low. But they’d be wrong: The high ratio reflects higher growth expectations, not lower returns.
This is precisely what has happened over the past century and a half. The authors show that long-run trend dividend growth drifted from roughly 2% per year in the late 1800s to approximately 6% per year by the early 2000s. As the growth trend crept upward, so did valuation ratios—persistently and structurally, not because expected returns were falling, but because the economy’s fundamental earning power was rising.
The model: Filtering out the noise
Hillenbrand and McCarthy built a state-space model that jointly estimates three hidden components simultaneously: the permanent trend in cash flow growth, the short-run cyclical component of cash flows, and the true expected return signal. They estimated this using an extended Kalman filter—a sophisticated statistical technique that updates estimates in real time as new data arrives—applied to 150 years of S&P 500 data from 1871 to 2022.
The payoff is dramatic. Their ‘purified’ expected return measure—the price/dividend ratio stripped of cash flow trend and cycle contamination—predicts future returns. Out of sample, using only data available at the time of each forecast, the model was significantly better at forecasting future returns (22% R-squared for five-year returns) than traditional valuation ratios. By the rigorous standards of financial forecasting, these are substantial numbers, especially when compared with the negative out-of-sample R-squared for traditional valuation ratios (price/dividend, price/earnings, CAPE).
Critically, the model passes a series of quantitative tests. It not only says returns should be more predictable after purging cash flow noise, but it also predicts exactly how much the predictive coefficient should increase and by how much the out-of-sample performance should improve. The data match these predictions closely, lending confidence that this is a genuine signal rather than data mining.
Rewriting the excess volatility debate
The paper’s findings also speak to one of the oldest debates in financial economics: Robert Shiller’s 1981 claim that stock prices are “excessively volatile” relative to fundamentals. Shiller argued that dividends simply don’t move around enough to justify the swings we see in stock prices, which implies irrational behaviour or time-varying discount rates must be doing the heavy lifting.
Hillenbrand and McCarthy offer a nuanced reconciliation. Both sides of the old debate turn out to be correct—just at different time horizons. At short and medium horizons (up to roughly a decade), discount rate variation does dominate: Roughly 75% of the transitory movements in the price/dividend ratio reflect changing expected returns. Over this horizon, Shiller’s excess volatility critique holds.
But zoom out to the full 150-year sample, and the picture inverts. Permanent shifts in trend growth account for 74% of the long-run variation in the price/dividend ratio. The steady secular rise in valuations over the 20th century was not irrational exuberance—it was the rational repricing of an economy whose long-run cash flow growth rate had genuinely and permanently accelerated.
What this means for today’s elevated market
One of the most practically important implications of this work concerns the persistent complaint that markets look expensive by historical standards. CAPE ratios, price/dividend ratios, and similar measures have been elevated for decades in the US, leading forecasters to predict poor future returns—predictions that have failed to materialise.
The trend-cycle framework offers an explanation. Traditional predictive models, unable to distinguish higher growth expectations from lower discount rates, systematically underpredicted returns throughout the postwar period. As trend growth drifted higher, the historical average valuation ratio became an increasingly misleading benchmark. A ratio that looks expensive relative to 1950s norms may simply reflect a structurally higher-growth economy—not an overvalued one.
This doesn’t mean markets can never be overvalued. It means the right benchmark for so-called normal valuations is itself a moving target, one that needs to be estimated rather than assumed fixed. The authors’ purified expected-return signal is stationary by construction—it fluctuates around a stable long-run mean, even as raw valuation ratios drift ever higher—and it is this signal that carries genuine forecasting power.
Five takeaways for investors
1. Don’t use raw price/dividend or CAPE ratios as your return forecast.
These ratios embed permanent cash flow trend changes that have nothing to do with future returns. Using them naively as forecasting tools has consistently produced negative out-of-sample accuracy—worse than simply using the historical average return.
2. High valuations don’t automatically mean low future returns.
If elevated valuations reflect structurally higher trend growth in the economy—as this paper argues has been the case for the past century—then those high valuations are fundamentally justified. The relevant question is always whether the current valuation is high relative to its growth-adjusted fair value, not its raw historical average.
3. Expected returns are countercyclical—and meaningfully so.
The model’s expected return signal spikes during genuine crises (the Great Depression, the stagflation of the 1970s–80s, the 2008 global financial crisis) and compresses during booms. This cyclicality is real and substantial. The annualised volatility of expected one-year returns is around 6%, far larger than standard regression approaches imply. Patient investors who can act on this signal have an edge.
4. The longer your horizon, the more cash flows matter.
For most investors thinking about the next year or two, discount rate variation is the dominant force. But for long-horizon investors—pension funds, endowments, individuals saving for retirement—the permanent growth component becomes increasingly important. Forecasting five-year returns is significantly more tractable than one-year returns under this framework, with 22% out-of-sample R-squared.
5. The stock market is more predictable than the consensus suggests.
The apparent weakness of traditional return predictability is an artifact of measurement error, not evidence that markets are perfectly efficient. Once cash flow noise is properly filtered, the market is genuinely and substantially predictable—particularly over multiyear horizons.
The bottom line: Price/dividend ratio is a noisy signal
Hillenbrand and McCarthy have produced a paper that simultaneously resolves long-standing theoretical puzzles, passes stringent empirical tests, and delivers practically useful results. The core message is elegant: The price/dividend ratio is a noisy signal for expected returns because it conflates discount rate variation with cash flow trend and cycle dynamics. Filter out the noise, and return predictability is not only restored but is also quantitatively substantial.
For investors, the practical implication is that building better return forecasts requires disentangling the sources of valuation changes. A rising price/dividend ratio driven by accelerating trend growth is a fundamentally different signal from one driven by falling risk premium. Treating them identically—as nearly all standard models do—is a mistake with real consequences for portfolio construction and capital allocation.
The authors are careful not to claim this model is a crystal ball. Out-of-sample R-squareds of 9% and 22%, while impressive by academic standards, still leave the vast majority of return variations unexplained. However, their work does establish that the conventional wisdom—that future returns are essentially unforeseeable from valuation ratios—is incorrect. The signal is there. You just have to know how to listen for it.
Larry Swedroe is a freelance writer and author. The views expressed here are the author’s. For informational and educational purposes only and should not be construed as specific investment, accounting, legal, or tax advice. The author does not own shares in any of the securities mentioned in this article.
This article was first published by Morningstar.com, and has been edited slightly for an Australian audience.