FinSentGPT: A universal financial sentiment engine?

Abstract

We present FinSentGPT, a financial sentiment prediction model based on a fine-tuned version of the artificial intelligence language model, ChatGPT. To assess the model’s effectiveness, we analyze a sample of US media news and a multi-language dataset of European Central Bank Monetary Policy Decisions. Our findings demonstrate that FinSentGPT’s sentiment classification ability aligns well with a prominent English-language finance sentiment model, surpasses an established alternative machine learning model, and is capable of predicting sentiment across various languages. Consequently, we offer preliminary evidence that advanced large-language AI models can facilitate flexible and contextual financial sentiment determination, transcending language barriers.