Large Language Models (LLMs) are sophisticated artificial neural networks with millions or even billions of parameters, trained on massive datasets using self-supervised or semi-supervised learning techniques. These models are designed to understand and generate human-like information. The financial industry has begun utilizing these tools for various purposes, such as predicting the stock market, providing financial education, offering economic advice, creating trading strategies, analyzing market sentiment, and managing risks. With the advancements brought by models like ChatGPT, BloombergGPT, and FinGPT, there’s significant potential for these LLMs to impact the financial sector.
ChatGPT
Two finance professors from the University of Florida argue that advanced LLMs like ChatGPT can improve stock market predictions and enhance trading strategies. In their study, they used ChatGPT to predict stock market returns by analyzing sentiment from news headlines. They found that ChatGPT outperformed other models such as BERT, GPT-1, and GPT-2. Only more advanced models like ChatGPT are capable of analyzing vast amounts of data to make accurate market predictions.
ChatGPT, based on the generative pre-trained transformer (GPT) architecture, was introduced by OpenAI in November 2022. It uses a multi-layer neural network to capture patterns and structures in natural language, trained with unsupervised learning on large text corpora, like Wikipedia articles or web pages. For this particular study, the professors used data from the Center for Research in Security Prices, daily stock returns, news headlines, and RavenPack.
The study highlights the potential of ChatGPT as a useful tool in the financial sector for predicting stock market movements based on sentiment analysis, although the authors acknowledge that further research is needed.
BloombergGPT
In March of this year, Bloomberg introduced its own LLM, BloombergGPT, designed specifically for the financial industry. This 50-billion-parameter model is built on a dataset of 363 billion tokens from Bloomberg’s proprietary data, combined with an additional 345 billion tokens from general-purpose sources.
BloombergGPT was validated on finance-specific natural language processing (NLP) benchmarks, as well as Bloomberg’s own internal benchmarks. The results showed that BloombergGPT outperformed other models like GPT-NeoX, OPT66B, BLOOM176B, and GPT-3, especially in financial applications. Bloomberg’s ability to curate a large, high-quality domain-specific dataset over the past four decades gave it a distinct advantage in training an LLM suited for financial use.
Gideon Mann, head of Bloomberg’s ML Product and Research team, emphasized the importance of high-quality data, noting that Bloomberg’s extensive collection of financial documents allowed them to create a highly specialized model. BloombergGPT is designed to improve existing workflows and open new opportunities to enhance financial services for customers.
FinGPT
Unlike BloombergGPT, which relies on proprietary data, FinGPT is an open-source LLM developed for the finance industry. Released in March 2023 by Finblox, a crypto trading platform supported by Dragonfly and Sequoia, FinGPT is aimed at democratizing AI-powered financial tools. The goal is to make financial literacy and resources accessible to a wider audience.
Peter Hoang, CEO of Finblox, explained that FinGPT’s mission is to empower users by providing personalized financial guidance and recommendations. With its easy-to-use interface, FinGPT represents a step toward creating a more inclusive financial ecosystem that promotes financial literacy and accessibility for all.
