Stock Price Predictability and the Business Cycle via Machine Learning

It is an issue of great importance for policy and investment decision makers (Schwert, 1989; Fama, 1990; Corradi et al., 2013; Chauvet et al., 2013). Empirical studies have been used to examine whether stock market volatility, which behaves differently in expansion and recession periods, can be predicted by macroeconomic variables (Schwert, 1989; Hamilton and Lin, 1996). Research has also established a link between stock market volatility and macroeconomic fundamentals (Engle and Rangel, 2008; Diebold and Yilmaz, 2008; Corradi et al., 2013; Choudhry et al., 2016). However, despite recent successes in developing machine learning (ML) models for predicting financial prices of different assets (see e.g., Gu et al. (2020); Heaton et al. (2017); Gu et al. (2021); Bianchi et al. (2021)), there is little literature discussing the impact of business cycles and market volatility on stock price forecasting with ML models. This paper fills this gap and explores the data-shifting effects of market volatility resulted from recessions on ML models. Specifically, we focus on answering the following three research questions in this work: 1. Do ML models perform differently during the recession compared to non-recession? 2. Does including recession data in the in-sample (training) period improve ML performance?