This paper investigates the stock return predictability in relation to the Management's Discussion and Analysis (MD&A) section of 10-K filings of US firms from January 1994 to December 2018. Based on innovations in Natural Language Processing and statistical learning, we introduce a novel method to extract the sentiment embedded in the MD&A section. We find that our method outperforms traditional approaches in terms of sentiment classification accuracy. Utilizing this method, the MD&A sentiment is found to be a strong negative predictor of future stock returns, demonstrating consistency in both in-sample and out-of-sample settings. Notably, with traditional sentiment extraction methods, the MD&A sentiment exhibits no predictive ability to stock markets. This finding underlines the MD&A section as an important source for stock return prediction, providing an accurate sentiment analysis method. Additionally, we examine the stock return predictability of the MD\&A sentiment in conjunction with macroeconomic variables. This examination reveals that the MD&A sentiment is associated with dividend-related macroeconomic channels regarding future stock return prediction.