A new study freely published in the SSRN journal has found that large language models (LLMs) with large context windows, such as GPT-4 and Claude 3, can perform successful financial statement analyses to help determine the direction of future earnings. The study found that AI exhibited “a relative advantage over human analysts in situations when the analysts tend to struggle.”
In addition, the researchers found that general LLMs were able to perform just as well as narrowly trained state-of-the-art machine learning models that had been trained to read financial statements. They also found that LLMs were able to make predictions that yield a higher Sharpe ratio which compared the return of an investment with its risk – the higher the Sharpe ratio, the better.
Given these results, the researcher found that LLMs may be able to have a central role in decision-making when it comes to choosing investments. Interestingly, GPT-4 was able to outdo human investors just based on the financial statements alone, devoid of any context, which the human investors would have been aware of.
The researchers also said that the LLM predictions do not come from its training memory but rather “generates useful narrative insights about a company’s future performance.”
For investors out there who think this work sounds interesting, the researchers have created a ChatGPT bot which is available to ChatGPT Plus subscribers so they can submit financial statements and get an analysis.
For anyone interested in analysing financial statements for free, Claude 3 allows you to upload documents and you can ask questions about the data too. While the researchers may be wondering if LLMs can replace human investors, another qualitative change with the availability of these tools is that hobby investors who may not know the ins and outs of financial documents, can get a better idea of what’s going on with various companies.
Let us know in the comments if you buy stocks and whether you’ve used LLMs to help you make stock picks.
Source: SSRN via Venture Beat
_small.jpg)
2 Comments - Add comment