Striking gold: Spears finance department head harnesses ChatGPT to analyze gold market
Friday, November 1, 2024
Media Contact: Terry Tush | Director of Marketing & Communications | 405-744-2703 | terry.tush@okstate.edu
Dr. Betty Simkins marveled at the organized, well-written email that suddenly appeared in her inbox.
She knew her oldest son, Luke, rarely had time for quick replies as a busy software engineer who works with artificial intelligence. But in the fall of 2022, when the Spears School of Business Department of Finance head asked her kids to contribute to the annual holiday newsletter they send to friends and loved ones, she barely had time to blink before Luke was the first to finalize his year-in-review paragraph.
Luke shamelessly spoiled his secret. He had prompted a newly launched chatbot to write his segment of the letter, sparking his mother’s amused curiosity about the cutting-edge technology. How did ChatGPT work? What were its capabilities?
After an early introduction to ChatGPT thanks to her son’s tech-savvy shenanigans, Simkins has extended her generative AI knowledge far beyond fun and games, shedding light on its power to drive innovative academic research. The Oklahoma State University Regents Professor teamed up with lead author Dr. Yunchuan Sun and his students, Ying Xu and Buxin Zhang, from Beijing Normal University in China for the study, “Investor sentiment in gold and gold futures market: Evidence from ChatGPT-generated sentiment index.”
The artificial intelligence tool highlights the strong relationship between investor sentiment and gold futures returns, especially in periods of economic turmoil. While the researchers’ conclusions about the gold market bolster previous scholarship in a popular field, the most groundbreaking insight comes from their methodology itself.
“We also confirmed that leveraging AI to analyze market trends is not only feasible, but it’s highly effective,” Simkins said. “This is one of our really important findings because AI is so new, so we show that the ability of ChatGPT to process and take vast amounts of textual data allows us to capture and quantify, in particular, subtle shifts in investor sentiment.”
The fluctuating attitudes of investors can significantly impact returns for gold futures, which, at their core, are agreements to buy gold at a set price on a later date. If investors predict an economic crisis or a future spike in gold prices, then they might rush to buy gold, driving the prices up. Gold futures are intriguing to many investors because of their hedging function, potentially providing a safety net during times of market instability.
Social media is a revealing tool for analyzing these investors’ opinions — which range from logical to emotional — but it isn’t easy to review a mind-boggling amount of data.
Imagine sifting through thousands of social media posts about the gold market, carefully assigning a numerical value to each one after reading every word. This would require painstaking effort from multiple researchers, consuming hours that could be spent in numerous other ways.
In Simkins’ study, no human had to trudge through that tedious process. Instead, the team trained ChatGPT to create a daily investor sentiment index using the gold discussion forum on Guba, a bustling social media platform for investors in China.
The beginning of the prompt, written for a machine, reads like a college assignment.
You are a gold futures investor and I will give you a post title. What emotions does this post title evoke in you? Please rate the mood on a scale of -50 to 50.
Recognizing key words and connotations, ChatGPT went to work, reviewing posts from 2010 to 2022. For example, a post reading “The short sellers are worried sickly now” received a score of -20, clearly indicating negative sentiment.
“This is what our sentiment index aims to achieve, not only differentiating between positive and negative, but also quantifying their intensity,” Simkins said.
The index required fine-tuning. Posts with more comments and engagement received more weight because of their greater potential to influence investors. Any neutral post or a post unrelated to gold received a score of 0, eliminating it from the analysis. Using mathematical equations and 4,459 data entries, the index took shape.
To determine how investor attitudes influence the market, the researchers looked at the historical gold futures daily closing prices from the time periods they analyzed. They also accounted for control variables that could have impacted gold futures returns, including the United States inflation rate. The data indicates positive investor sentiment did, indeed, boost gold futures returns. The findings show when positive investor sentiment grows by 1%, as measured by the index, average gold futures returns will climb by 0.143% if other variables are consistent.
Simkins, Sun, Xu and Zhang also used the index to analyze four market crisis periods: the 2015 stock market crash, the 2018 beginning of the U.S.-China trade war, the 2020 COVID-19 pandemic and the 2022 Russian invasion of Ukraine. Although the stock market suffered in times of turmoil, gold futures stayed afloat in comparison. In these crisis times, investor sentiment aligned more strongly with gold futures price shifts, according to the index.
This data highlighted what Simkins called the “safe haven” power of gold futures amid economic chaos, but the index also offered insight into “normal markets,” she said. Real-world investors could use similar AI-powered tools on an everyday basis to track market trends and make risk-minimizing choices.
“As an academic, what really thrilled me about this line of research is this is something really positive about ChatGPT,” Simkins said.
Relying on a 2-year-old chatbot as a research assistant does come with caveats, as Simkins is well aware. The young technology can make mistakes, and it’s more of an enhancement to human cognition instead of a substitute for it. Simkins emphasized the importance of scholars clearly disclosing the use of AI in their studies, including the name of the specific tool. In her case, it’s in the paper’s title.
If researchers know how to interact with AI, then they can sharpen its functions in incredible ways. In Simkins’ study, the researchers randomly sampled Guba posts from three separate years and manually calculated their positive or negative ratings, comparing them with ChatGPT’s scores to gauge how well the index was working. They then further trained ChatGPT, altering the prompt to achieve a more accurate result.
Although the researchers initially planned to publish in a letters journal, they are now working on expanding their paper for a larger journal, elaborating on their experiences with training ChatGPT. Their novel method opens the door to vast possibilities in various fields of study because of AI’s ability to streamline research processes.
Simkins, a prolific researcher who serves on multiple editorial boards, now not only can teach her son something about AI, but also can use her work to inspire others.
“You want it to influence other researchers,” Simkins said. “We’re hoping what we’ve done here, it could be replicated for many things. In finance, it could be trying to predict lots of other things and in other markets. Not only are we excited about what we find, but also the purpose of research is to help move research forward, and here’s a fairly new tool to do that.”
Story by: Hallie Hart | Discover@Spears Magazine
Photos by: Adam Luther