Can machine learning uncover abnormal returns in uncharted financial territories?

Abstract

We examine whether advanced technical analysis, capable of detecting highly complex relationships between past (up to one year ago) and future returns using machine learning, can generate abnormal returns in the US and Korean equity markets. First, we replicate the main results of Murray et al. (2024) in the US market, showing that the trading strategy based on the machine learning-based return forecast yields abnormal returns. Next, we evaluate the validity of this trading strategy in Korea, where traditional technical analyses, such as momentum and reversal strategies, have been notably unprofitable, as documented in prior studies. In the Korean market, we also find that the trading strategy based on the machine learning-based return forecast yields abnormal returns, providing a critical out-of-sample evaluation of machine learning-based charting. This further supports the argument that the violation of the weak-form efficient market hypothesis is more complex than previously acknowledged.