|
Listen to this story
|
Two finance professors from the University of Florida, Professor Alejandro Lopez-Lira and Professor Yuehua Tang, conducted a study to explore the potential of ChatGPT and other large language models in predicting stock market returns. The professors used ChatGPT to parse news headlines for whether they’re good or bad for a stock, and found that ChatGPT’s ability to predict the direction of the next day’s returns was much better than random.
Lopez-Lira said he was surprised by the results, which seemed to suggest that sophisticated investors aren’t using ChatGPT-style machine learning in their trading strategies yet.
The researchers also found that other basic models such as GPT-1, GPT-2, and BERT were not as accurate in forecasting returns, indicating that the ability to predict returns is an emerging capacity of complex models. These findings suggest that incorporating advanced language models like ChatGPT into investment decision-making processes may lead to more accurate predictions and improved performance of quantitative trading strategies.
Consequently, Bloomberg released a new GPT-based language model called BloombergGPT, which is trained on a dataset called FinPile, consisting of English-language financial documents, news, filings, press releases, and social media. The company claims that this new model will improve existing natural language processing tasks such as sentiment analysis, news classification, headline generation, question-answering, and other query-related tasks.
The model is not trained on multilingual data but is trained on Bloomberg’s vast repository of financial data over the past forty years. It’s worth noting that Bloomberg already has Bloomberg Terminal, which utilizes NLP and ML models to provide financial data. As such, it begs the question of how much BloombergGPT will add in value and how it stacks up against other GPT models.
The use of LLMs, such as ChatGPT, has gained momentum in various domains, but their potential in predicting stock market returns remains relatively unexplored in financial economics. There are arguments on both sides – on the one hand, since LLMs are not explicitly trained for this purpose, they may offer little value in predicting stock market movements; on the other hand, given their capability to understand natural language, they could be a valuable tool for processing textual information to predict stock returns.
Not Bullish on ChatGPT
While businesses, transportation, science, law enforcement, and medicine amongst others have embraced AI, the finance industry has been doing the opposite. For a while now, Wall Street has been using computer programs to handle tasks like trading and risk management. However, investors haven’t made significant advancements in utilizing artificial intelligence to overcome their main hurdle – outperforming the market. Although some view ChatGPT as a means to enhance their sales and research endeavors, the utilization of AI in investing hasn’t produced remarkable outcomes.
“Progress in applying AI to investing has been limited, though innovations in language modeling could change that in the years ahead,” says Jonathan Larkin, a managing director with Columbia Investment Management Co, which manages the $13 billion endowment for Columbia University and invests in various funds.
Decades ago, finance pioneers like Jim Simons of Renaissance Technologies used machine learning to create algorithms that allowed computers to make investment decisions using past data with minimal human input. However, despite their success in building trading models that can identify patterns and generate profitable trades, these firms have not fully transitioned to automated operations using cutting-edge AI methods such as self-learning or reinforcement learning. Instead, they continue to rely on advanced statistics and a “theory-first” approach, where they establish hypotheses and build models around them, according to industry insiders.
“Most quants still take a “theory-first” approach where they first establish a hypothesis of why a certain anomaly might exist, and they form a model around that,” Larkin told The Wall Street Journal.
So, there might be a potential use case for ChatGPT in the stock market and the financial structure at large, but there are apprehensions in the market given the risks it could pose if it doesn’t provide the accuracy and assistance that is expected. Hence, the market at large seems to have chosen to stick with older methods until something truly groundbreaking comes around.
