The predictive power of dividend yields analyzed by methods preserving time-dependent structures

This paper analyzes the predictability of dividend-price-ratio for long horizon stock returns with the help of methods that preserve the time dependence of the underlying problem. Bootstrapping and subsampling approaches are employed to handle serial correlation in the data and to achieve reliable t-statistics. Optimization procedures were used to obtain the optimal moving-block-size and subsampling-size respectively. Furthermore to get valid results, it is necessary to calculate the calibration function and hence the corresponding nominal significance level to a requested level. Both moving-block bootstrapping with the optimal block size and the calibrated significance level as well as the optimized subsampling technique strongly reject the hypothesis of predictability in dividend-price-ratio for long horizon returns of the equities at NYSE.