We present a method for determining whether errors in expectations explain asset pricing puzzles without imposing assumptions about the error mechanism. Using accounting identities and survey forecasts, we find that errors in expected long-term inflation explain price variation, return predictability, and the rejection of the expectations hypothesis for aggregate stock and bond markets. Errors in short-term (long-term) nominal earnings growth expectations explain (do not explain) stock price variation and return predictability. The relevant errors are consistent with mistakes about the persistence of forecasted variables and the response to surprises. A simple framework based on fundamental extrapolation successfully replicates these findings. (JEL G40, G12, G14, E71)

ABSTRACT Why do stock prices vary? Using survey forecasts, we find that cash flow growth expectations explain most movements in the S&P 500 price‐dividend and price‐earnings ratios, accounting for at least 93% and 63% of their variation. These expectations comove strongly with price ratios, even when price ratios do not predict future cash flow growth. In comparison, return expectations have low volatility and small comovement with price ratios. Short‐term, rather than long‐term, expectations account for most price ratio variation. We propose an asset pricing model with beliefs about earnings growth reversal that accurately replicates these cash flow growth expectations and dynamics.

What explains cross-sectional dispersion in stock valuation ratios? We find that 75% of dispersion in price–earnings ratios is reflected in differences in future returns, while only 25% is reflected in differences in future earnings growth. This holds at both the portfolio-level and the firm-level. We reconcile these conclusions with previous literature which has found a strong relation between prices and future profitability. Our results support models in which the cross-section of price–earnings ratios is driven mainly by discount rates or mispricing rather than future earnings growth. Evaluating six models of the value premium, we find that most models struggle to match our results; however, models with long-lived differences in risk exposure or gradual learning about parameters perform the best. The lack of earnings growth differences at long horizons provides new evidence in favor of long-run return predictability. We also show a similar dominance of predicted returns for explaining the dispersion in return surprises.