How Have Stock Markets Responded to 35 Years of Analyst Reports? Evidence from Machine Learning and Textual Analysis

Using machine learning (ML) and contrasting with simple textual sentiment score and principal components analysis (PCA) methods, we examine the time series of content within over 700,000 sell-side analyst research reports from 1983 to 2017. We find that analyst reports have significantly changed across a variety of dimensions including length and content of four existing and two new dictionaries related to valuation and alternative metrics. We find that the naive net tone of reports only explains contemporaneous, not future, equity returns. On the other hand, we find that ML methods provide substantially different results from naive sentiment and PCA approaches on determining the impact of analyst reports on financial markets. We also examine the ability of reports to predict changes in firm short interest and volatility. Overall, we find that sell-side analyst reports have stronger impact on smaller cap stocks.