Time-Varying Predictive Power of the Dividend Yield for Stock Returns: Evaluation Based on Bootstrapping
38 Pages Posted: 1 Dec 2014 Last revised: 17 Feb 2017
Date Written: November 30, 2014
This paper evaluates the predictive ability of dividend yield for stock return using a new bootstrap test for the significance of predictive coefficients. The predictive model is expressed as a restricted vector autoregressive model, and the bootstrap is conducted with resampling based on generalized least-squares estimation. Our Monte Carlo study shows that the bootstrap test has no size distortion in small samples with highly satisfactory power properties. We employ a range of bivariate predictive models using monthly U.S. data from 1926 to 2012. We find that the dividend yield does show predictive ability for stock return from time to time, but its effect size and out-of-sample forecasting performance have been weak except for the 1990s.
Keywords: Dividend-yield, EGLS estimation, Monte Carlo experiment, Return predictability
JEL Classification: C32, G12, G14
Suggested Citation: Suggested Citation