The Untold Story of Alex Kim – Financial Statement Analysis with Large Language Models

In this video, I tell the story of Alex Kim, a PhD student at the University of Chicago.

Alex has published many interesting papers, but this video discusses one in particular: Financial Statement Analysis with Large Language Models. The paper is over 50 pages long, so I summarized the first major question that Alex posed: are large language models like ChatGPT effective at performing financial analysis?

To study this, Alex compared using Large Language Models to 2 alternative approaches.

The naive approach took the change in EPS on year t-1, and predicted the change in EPS on year t to be the exact same.

The analysts approach took the consensus rating of financial analysts on year t-1, and measured how accurate their predictions were on year t.

The LLM-approach took financial statements on a stock on year t-1, and asked the LLM to perform Chain of Thought prompting to predict the change in earnings on year t.

He found that the naive approach was 49% accurate in predicting the change in a stock's EPS. The analysts approach was 53% accurate (even though it had much more contextual information than the large language model).

The LLM-approach was by far the most accurate, at over 60% accuracy using solely financial statement data.

Imagine how accurate the LLM-approach would be if you gave it more information? For example, imagine if you gave it the mean change for companies in the same industry? Or, you input backtesting results from past analysis?

AI will change how people approach financial markets. Are you going to standby as this opportunity passes you by? Or are you going to take action?