A1248
Title: Analyzing market response to SFCR in European insurance with topic modeling and deep learning methods
Authors: Ling-Jing Kao - National Taipei University of Technology (Taiwan)
Ting Kang Liu - BenQ Corporation (Taiwan)
Chih-Chou Chiu - National Taipei University of Technology (Taiwan) [presenting]
Abstract: A BERTopic-LSTM model was proposed to investigate the market response to Solvency and Financial Condition Reports (SFCRs) published by 27 publicly-traded insurance companies in the European Union in English between 2018 and 2021. The BERTopic method, a natural language processing (NLP) technique, was utilized to analyze the content of the SFCRs. A deep learning LSTM model was then developed to predict the changes in the stock yields of life insurance companies following the release of the SFCRs. The essential financial metrics and textual features of interest to investors were also examined. The results demonstrate that the proposed model surpasses competing machine learning models (MARS, Random Forest, and Gradient Boosting) in terms of prediction accuracy. Furthermore, the findings reveal that the corporate governance attributes of financial performance and risk management (topic 1) and operational management and resilience (topic 2) had the most significant influence on changes in stock yield prediction.
