MIDAS and Dividend Growth Predictability: Revisiting the Excess Volatility Puzzle

We examine dividend growth predictability and the excess volatility puzzle across a large sample of international equity markets, using a mixed frequency data sampling (MIDAS) regression approach. We find that accounting for dividend seasonality under the MIDAS framework significantly improves dividend growth predictability, compared to simple regressions with annually aggregated data. Moreover, variance bounds tests that allow for non-stationary dividends consistently fail to reject the hypothesis of market efficiency across all countries. Our findings suggest that the common rejection of market efficiency in the previous literature is most likely driven by the annual aggregation of dividend data as well as by the assumption of stationary dividends.