A Multimodal and Sentiment-Based Trading System for Financial Portfolio Optimisation

The keys to accurately predicting future price trends and correlations between assets in a portfolio are not only learning from historical price data, but also considering the current market sentiment among investors. Yet there is a lot of news published by different media everyday, some of which may describe irrelevant or even contradictory information, providing noise signals to estimate market sentiment. The existing portfolio management frameworks usually treat all news in the same way to extract sentiment intention, which unavoidably results in biased trading decisions. Accordingly, a Multimodal and Sentiment-based Adaptive trading framework, namely the MuSA, is proposed in this paper where the sentiment analyser collects relevant news from different media and evaluates the sentiment behind the news by using large language models. Afterwards, an entropy-based confidence learning mechanism measures the confidence of the news by checking the reliability of information sources to reduce the effects of irrelevant messages. Meanwhile, the price analyser learns the future trends of assets from historical price data. Finally, combining sentiment information and price estimation generates a new portfolio to adapt to the ever-changing financial markets. The empirical results in two real-world datasets clearly reveal the proposed framework can help extract useful sentiment information for achieving higher returns and lower risks.