Enhancing portfolio return based on market-sentiment linked topics

While time-series analysis techniques are commonly used in financial forecasting, a key source of market volatility is omitted from these models. Financial news is known to be making persuasive impact to the markets. Without considering these additional signals, only sub-optimal predictions can be made. This paper proposes a supervised topic learning approach to improve portfolio return. It is achieved by considering market-sentiment linked topics retrieved from financial news. Using this approach, we successfully improve the prediction accuracy of a proprietary trade recommendation platform. Different from traditional sentiment analysis and unsupervised topic modeling methods, topics specific to different sentiment levels are identified by our proposed model to quantify market conditions. The topics are learned from historical market performances and commentaries instead of using subjective interpretation of sentiments from human expressions. By capturing the knowledge specific to respective industries and markets, an impressive double-digit improvement in portfolio return is obtained as shown in our experiments.