TY - JOUR T1 - Machine Learning Algorithms to Classify Future Returns Using Structured and Unstructured Data JF - The Journal of Investing DO - 10.3905/joi.2021.1.169 SP - joi.2021.1.169 AU - Joshua Livnat AU - Jyoti Singh Y1 - 2020/02/12 UR - https://pm-research.com/content/early/2021/02/12/joi.2021.1.169.abstract N2 - This study employs a machine learning (ML) algorithm to classify future stock returns into three categories: large positive, large negative, and all others (zero). We provide evidence of marginal classification improvement after using ML tools, which employ analysts’ earnings estimate revisions and natural language processing to extract the tone and events of interest from news stories written about a firm. However, this improvement is statistically very significant. An application of this study is in portfolio rebalancing and short-term trading decisions, especially in quantitative portfolio management. With broad portfolios, during a normal rebalancing, one can delay the sale of a few stocks with prices predicted to increase significantly in the short term and substitute stocks with similar characteristics and alpha expectations. Likewise, one can delay the purchase of stocks that the quantitative model wishes to buy, but the ML model predicts to decrease significantly in the short term.TOPICS: Big data/machine learning, statistical methods, portfolio construction, performance measurementKey Findings▪ We use a time series machine learning algorithm to classify firms into three potential outcome categories: those whose short-term prices are likely to increase significantly, those expected to decline significantly, and all others.▪ We find that machine learning improves the accuracy of predictions in the three classes to a degree that is statistically very significant when judged against random classifications. Careful design of the problem and its parameters can yield even more desirable results.▪ We show a potential application of the model to rebalancing a broad portfolio in a quantitative asset management or in short-term trading. ER -