Price series cross-correlation analysis to enhance the diversification of itemset-based stock portfolios

Abstract

Planning buy-and-hold strategies for stock trading is a challenging financial task. It entails building a portfolio of stocks maximizing the expected return in the medium- or long-term while minimizing investments' risk. Diversification is the most common strategy to manage risk in financial investments. It entails spreading bets across multiple assets, typically by picking stocks from different financial sectors. This paper presents a time series clustering-based strategy to improve the effectiveness of stock diversification across sectors. It analyzes the cross-correlation among price series in order to identify groups of stocks belonging to different sectors that unexpectedly show similar trends as well as dissimilarities among stocks of the same sector. The diversification strategy has been integrated into a state-of-the-art itemset-based approach to stock portfolio generation. The performance achieved on the U.S. stock market show relevant improvements in portfolio returns and drawdown control.