On 27 March 2023, the 33rd Xi Xian Forum on Income Distribution and Public Finance, jointly organised by the School of Public Finance and Taxation of ZUEL and IIDPF, was successfully held in Conference Room 119, Wenqin Building. The forum was chaired by Professor Yang Guochao, Director of the Big Data Research Centre of IIDPF, with over 20 participants including Professor Lu Yuanping, Executive Director of IIDPF, Associate Professor Zhao Ying and Dr Zhang Ziyao from the School of Public Finance and Taxation, and IIDPF's full-time researchers Dr Zou Wei, Dr Zou Jianwen and Dr Zhang Fan.
Textual big data has become more available in the digital economy. A growing number of studies have attempted to extract sentiment information through text analysis from various aspects such as media reports, company news, social networks, etc., and apply it to behavioral finance, asset pricing, risk management, monetary policy and other related studies. Based on financial text big data, Dr Jiang synthesised BERT sentiment variables by using a deep learning BERT model, introducing real market returns as labels to train the model and portraying the asymmetric non-linear theoretical features of financial text sentiment. It was found that compared with the financial dictionary method sentiment variables, the BERT sentiment variables have the following four characteristics: firstly, the BERT sentiment variables have stronger sample predictive power; secondly, they have stronger predictive power for macroeconomic variables in different economic periods while having irrational trading characteristics; thirdly, they have stronger predictive power during sudden extreme events such as stock market crashes and new crown epidemics; fourthly, they remain significant in short text analysis and robust.
Dr Jiang gave a fascinating talk to the audience and also stimulated in-depth thinking on the research topic. During the discussion session, faculty and students spoke enthusiastically and had an interactive exchange of views on the context of words in text analysis, how to conduct text analysis for practical problems, and how the BERT model compares with other methods.