We find that investor attention proxies proposed in the literature collectively have a common component that has significant power in predicting stock market risk premium, both in-sample and out-of-sample. This common component is well extracted by using partial least squares, scaled principal...