Excess stock returns and news: evidence from European markets[Note 1. The paper benefited substantially from the comments of an ...]

This paper aims at decomposing the forecast error variance of excess returns in five major European stock markets into the variance of news about future excess returns, dividends and real interest rates. Special emphasis is given on the issue of stationarity and structural breaks in the unconditional mean of dividend yields and their implications for variance decompositions. Empirical results indicate that in some markets the dividend yield is subject to structural breaks in the mean. Evidence from Monte Carlo simulations suggests that this kind of structural breaks cause small‐sample bias in variance decompositions of a magnitude comparable to bias introduced by unit roots. Our results constitute a warning about return decompositions that, in particular, use variables in the forecasting equations that may be nonstationary or contain a structural break.