Robust Portfolio Optimization using Sentiment-based Uncertainty Sets

Abstract

Portfolio optimization is among the most important problems in quantitative finance, with a large audience from individuals to institutional investors, traditionally relying on static or normally distributed returns. However, these approaches often fail to capture the dynamic nature of market conditions and the impact of social sentiment on asset prices. This thesis explores the application of Robust Optimization (RO) in portfolio management by introducing sentiment analysis of social data to adjust the uncertainty set dynamically. Our proposed methodology employs sentiment analysis techniques to extract valuable insights from a significant body of tweets related to financial assets in the portfolio. By quantifying the sentiment expressed in these tweets, our goal is to capture the collective mood of market participants, a factor known to impact individual asset performance. Our proposed model, sentiment-based robust optimization (SRO), utilizes this sentiment information to modify the uncertainty set, enhancing the decision-making process under market uncertainty. We conducted empirical experiments to validate the effectiveness of this novel approach using actual trading data. By integrating sentiment analysis into the robust optimization framework, our proposed model demonstrates outstanding performance in risk-adjusted return measures compared to benchmark models.