Can a corporate network and news sentiment improve portfolio optimization using the Black–Litterman model?

Abstract

The Black–Litterman (BL) model for portfolio optimization combines investors’ expectations with the Markowitz framework. The BL model is designed for investors with private information or knowledge of market behaviour. In this paper, I propose a method where investors’ expectations are based on either news sentiment using high-frequency data or on a combination of accounting variables; financial analysts’ recommendations, and corporate social network indicators with quarterly data. The results show promise when compared to a market portfolio. I also provide recommendations for trading strategies using the results of this BL model.

Keywords:

• Link mining
• Social network
• Machine learning
• Boosting
• Text analysis
• Portfolio optimization
• Black–Litterman model

JEL Classifications:

• C44
• C58
• C63
• G11

Acknowledgements

The author thanks Ionut Florescu, Maria Christina Mariani, Frederi G. Viens, two anonymous referees, and participants of the Financial Mathematics session of the American Mathematical Society meeting 2010, the Workshop on Information in Networks (WIN)-NYU 2010, the IEEE CEC 2011 Workshop on Agent-Based Economics and Finance, the Sixth Rutgers-Stevens Workshop Optimization of Stochastic Systems 2011, and the Eastern Economics Association meeting 2012 for their valuable comments. The author also thanks Patrick Jardine for proof-reading the article. This work was supported by the Howe School Alliance for Technology Management.

Notes

No potential conflict of interest was reported by the author.

1 COMPUSTAT is an accounting database managed by Standard & Poor’s.

2 Long or short positions refer to buy a specific asset or to sell a borrowed asset based on the expectation that price of the asset will increase or decrease respectively.