Source: Campbell, J. Y. & Shiller, R. J. (1988) · Journal of Finance 43(3), 661–676 (NBER WP 2511) · DOI: 10.1111/j.1540-6261.1988.tb04598.x
TL;DR
Using aggregate U.S. stock data 1871–1987 (S&P Composite extended back to 1871 via Cowles), the paper introduces a log-linear present-value / VAR framework and shows that valuation ratios — the log dividend-price ratio and long moving averages of earnings-price — forecast multi-year returns but not dividend growth. Simple present-value models are strongly rejected: prices and returns are far more volatile than dividend news warrants, and this excess volatility directly implies the forecastability of long-horizon returns. A long earnings average is the single best predictor of the present value of future dividends.
The idea
Log-linearize the return identity around the mean dividend-price ratio. The one-period log return is approximately
h₁,t+1 ≈ k − ρ·δ_{t+1} + δ_t + Δd_{t+1}, where δ is the log dividend-price ratio, Δd log dividend growth, and ρ = 1/(1+exp(δ̄)); the authors set ρ = 0.936. Iterating forward, the log dividend-price ratio equals a discounted sum of expected future returns minus expected future dividend growth. Since dividend growth is hard to forecast, variation in valuation ratios must reflect variation in expected returns — i.e. return predictability. The VAR estimates how much of returns is forecastable, and how much is justified ex post by news about future dividends.
Evidence
A long earnings average predicts fundamental value. The optimal forecast of the present value of future real dividends is roughly a weighted average of a moving-average of earnings and current real price, with 2/3 to 3/4 of the weight on earnings — even after conditioning on the information in price.
Long-horizon return predictability (Table 1, 1871–1987). The log dividend-price ratio explains 3.9% of the variance of one-year real returns but 26.6% of ten-year real returns; the 30-year moving-average earnings-price ratio explains 54.6% of ten-year real-return variance.
Replication of Fama–French. On the 1927–86 sample, the dividend-price ratio explains 21.9% of exact 4-year real returns, confirming their ~29% estimate; at the 10-year horizon the 30-year earnings average explains 45.5%.
Dividend growth is not forecastable by these ratios at any horizon; lagged dividend growth has no predictive power.
Excess volatility. Log dividend-price ratios are more variable than, and virtually uncorrelated with, their theoretical present-value counterparts; annual returns are correlated with their theoretical counterparts but are two to four times as variable.
Why it matters
Unifies two literatures previously seen as distinct — the 1980s "excess volatility" debate (LeRoy–Porter, Shiller) and the long-horizon return-forecastability findings (Fama–French) — by showing they are the same phenomenon. The log-linear present-value decomposition is the workhorse behind the dividend-yield, CAPE, and book-to-market return-predictability literatures and Cochrane's (2008) "all variation in the price-dividend ratio is expected returns" argument.
Caveats
Valuation-ratio predictability is a low-frequency, long-horizon result driven by a few non-overlapping observations; overlapping multi-year returns make R²s look high and standard errors must be corrected (Wald/White used here).
Highly persistent regressors induce small-sample (Stambaugh) bias; out-of-sample power is far weaker (Welch–Goyal).
Results depend on stationarity of the dividend-price ratio and on accounting earnings, whose economic meaning shifts over the 1871–1987 span.
Key references
Campbell, J. & Shiller, R. (1988) — The Dividend-Price Ratio and Expectations of Future Dividends and Discount Factors — Review of Financial Studies
Fama, E. & French, K. (1988) — Dividend Yields and Expected Stock Returns — Journal of Financial Economics
Cochrane, J. (2008) — The Dog That Did Not Bark: A Defense of Return Predictability — Review of Financial Studies
Provenance: verified/generated from the paper's full text.
