News sentiment and investment risk management: Innovative evidence from the large language models

Abstract

This paper reexamines the significance of news sentiment in explaining stock return volatility persistence and its role in driving underlying volatility states. Our simulation study demonstrates that more accurately measured news sentiment has a greater impact on volatility dynamics. Using data from firms in the Dow Jones Composite Average index spanning 2019–2023, we compare news sentiment classified by GPT-4 with that classified by RavenPack. Our findings show that both negative and positive firm-specific and macroeconomic news significantly affect intraday stock return volatility. The classification accuracy achieved by employing GPT-4 potentially surpasses that of using RavenPack.