Investor sentiment indices based on k-step PLS algorithm: A group of powerful predictors of stock market returns

Abstract

We construct a group of new investor sentiment indices by applying a new dimension reduction technique called k-step algorithm which adopts partial least squares method recursively. With the purpose of forecasting the aggregate stock market return, the new group of investor sentiment indices performs a greater ability in predicting the market return than existing investor sentiment indices in and out of sample by adequately using the information in residuals and eliminating a common noise component in sentiment proxies. This group of new investor sentiment indices beats five widely used economic variables and still has a strong return predictability after controlling these variables. Moreover, they could also predict cross-sectional stock returns sorted by industry, size, value, and momentum and generate considerable economic value for a mean-variance investor. We find the predictability of this group of investor sentiment indices comes from its forecasting power for discount rates and market illiquidity.