Large language models ( LLMs ) such as ChatGPT and Google Gemini are good at training on large datasets to produce information-rich responses to prompts. Cao Yi, Assistant Professor of Accounting at George Mason University’s Donald G. Costello School of Business, and Long Chen, Associate Professor of Accounting and Regional Chairman at Costello School of Business, are actively exploring how individual investors can use LLMs to dazzle the vast amount of data available about companies.
Their new working paper, published in the SSRN Electronic Journal and co-written by Jennifer Wu Tucker of the University of Florida and Chi Wan of the University of Massachusetts at Boston, examines the ability of artificial intelligence to identify “peer companies” or product market competitors in the industry.
Cao linked the process to the real estate market, explaining the importance of choosing a peer. “Capital markets are similar to real estate markets, and the value of a company depends in part on the value of its peers. In the real estate market, we set prices to homes based on the value of comparable properties nearby, or so-called ‘compensation.’ In our paper, our goal was to leverage the power of LLMs to determine comparisons used to assess company value.
This task is at least difficult and necessary. Collecting, summarizing and managing data for your peers requires a lot of time, skills and energy. However, the researchers concluded that LLMs could do the heavy lifting of aggregating and analyzing large amounts of data for individual investors and produce a peer list with comparable validity as determined by human experts.
“The advantage is being able to leverage all potential information so that it can at least work like other traditional methods, helping us investors and researchers,” Cao said.
For the study, Chen and Cao hired Bard (now known as “Gemini”) from Google as their LLM of choice because “Bard has a greater ability to utilize its pre-training data, which is arguably larger and has more parameters than ChatGPT,” Cao said.
After defining “product market competition” and forming hints for Bard, the researchers instructed Bard to limit his knowledge base to specific years between 1981 and 2023 to avoid “forward-looking bias,” the result of future information confusion.
They limit focus companies to large publicly traded companies because there is less data for small or private companies. In summary, this dataset contains more than 300,000 focus company years.
On average, LLM can generate about seven peer companies for a focus company, a number similar to the Securities and Exchange Commission’s recommendations on how companies should disclose their divisions.
The researchers then compared LLM’s performance to a list generated by three human experts for 40 leading computer software companies. The average overlap rate was slightly higher than 40%, which was higher than expected.
They also compared the AI-identified peer list with two alternative systems for identifying peers: the federal government’s Standard Industry Classification (SIC) code and the Text-Based Internet Industry Classification (TNIC), which compares companies based on language similarities in 10-K. There is a large overlap between LLM and TNIC’s output. In addition, the peers identified by LLM are generally more suitable than their counterparts in SIC and TNIC because their monthly equity returns are closer to focus companies.
But TNIC was superior to LLM in identifying peers from mid-sized companies in the sample, which suggests this is not a clear case of LLM’s general superiority.
“We need to understand that LLMs are actually a very powerful new tool whose efficiency, ability to process large amounts of information at low cost, and accessibility to the public are unmatched,” Cao noted.
“This is particularly beneficial to individual investors because all the cost issues we talk about are particularly relevant to them,” Chen added.
Regarding the future of LLM, Chen said,”There are always costs and benefits to using generative artificial intelligence. It’s unclear whether the current system will soon become obsolete.” When asked about the U.S. Securities and Exchange Commission’s adoption of artificial intelligence tools for investors, Chen emphasized that users need to understand the pros and cons of using artificial intelligence to make informed judgments,”because artificial intelligence cannot be responsible for the information it provides or the way it is used.”
Chen concluded: “We need to embrace this new technology, but we must realize that it is not yet in perfect condition. Competition to improve technology is fierce. Our findings may just represent the lower limit of the effectiveness of the technology.”
If you want to learn more, you can click on the link below the video.
Thank you for watching this video. If you like it, please subscribe and like it. thank
Original text:https://techxplore.com/news/2024-08-generative-ai-closer-automating-investment.html
More information: Yi Cao et al, Can generative AI assist investors? An evaluation of machine-generated peer firms, SSRN Electronic Journal (2024). DOI: 10.2139/ssrn.4761624
More information: Yi Cao and others, can generative artificial intelligence help investors? An evaluation of machine-generated peer companies, SSRN Electronic Magazine (2024). DOI：10.2139/ssrn.4761624
Oil tubing: