Learning From Investor Attention: Examining the Predictive Power of Investor Attention on Market Returns with Machine Learning

Sammanfattning: We study the predictive properties of investor attention on time series market returns. Extending an earlier proposed index of investor attention aggregated from twelve popularly studied attention proxies, we show that it strongly predicts excess returns on the stock market. Adding to an inconclusive body of literature, our results suggest that when attention is high, the market earns higher returns in the subsequent months. Uniquely, we examine whether a deep learning method from the machine learning catalog, Long-Short Term Memory, can enhance the predictability. Our results show that the nonlinear patterns in the data studied in this paper are not strong enough to yield economic gains.