Predicting stock returns with partially observed dividend yield

This paper tests stock return predictability using dividend yield. I estimate two unobservable components of dividend yield using Kalman filter technique, and then regress stock return on these two individual components. I show that one component is more persistent than the other. The less persistent component has more predictive power and accounts for most variation of the stock return. The two-component model performs better than the OLS model both in sample and out-of-sample.