In this paper, we develop a new investor sentiment index that is aligned to predict the excess returns on U.S. government bonds that have 2–5 years maturities. The new index is constructed by eliminating a common noise component in underlying sentiment proxies using the partial least squares (PLS) approach. The findings show that the new aligned sentiment index has much greater predictive power than the original principal component analysis (PCA)-based sentiment index both in- and out-of-sample. In addition, predictability is statistically significant, especially for bond premia with shorter maturities, even after controlling for a large number of financial and macro factors, as well as investor attention and manager sentiment indexes. Given the role of U.S. Treasury securities in forecasting of output and inflation, as well as in portfolio allocation decisions, our findings have significant implications for investors, policymakers, and researchers interested in accurately the forecasting return dynamics for these assets.