A refined asymptotic framework for dividend yield in predictive regressions

•Predictability by dividend yield is modeled using a local-to-zero signal-to-noise ratio refinement.•I study the asymptotic properties of predictive regressions across forecast horizons and sample sizes.•The model explains many previous simulation-based results in the finance.•Dividend yield carries a weak signal but the results stand in contrast to the spurious regression literature.

I model predictability by dividend yield using a local-to-zero signal-to-noise ratio refinement. Under the local-to-unity assumption, I study the limiting behavior of the R2statistic and the slope estimate as functions of forecast horizon and sample size. The new asymptotic framework provides a theoretical explanation for many previous simulation-based results in the finance literature.