Title: Machine learning utility-maximising market regime classifications
Authors: Richard McGee - University College Dublin (Ireland) [presenting]
Abstract: An unsupervised machine learning methodology is proposed to classify market periods into Bull or Bear classifications, conditional on a set of widely adopted market and macroeconomic bellwether variables. The classification of regimes is determined by decision trees that optimise the long-term investor utility of a market timing strategy, switching between Bull and Bear portfolios, whose optimal portfolio weights are themselves estimated over the sets of returns associated with each classification. Analysis covers Bull and Bear regimes across a range of thematic investment factors such as the market index, size, value and momentum.