The Revolution of AI in Financial Analysis: Danger or Opportunity?

Information industry professionals, from journalists to software developers to financial analysts, are challenged by one question as artificial intelligence progresses: When will a computer take their job? In the finance sector, a recent study at Chicago University’s business school by Alex Kim, Maximilian Muhn and Valeri Nikolaev (KMN) provides some sobering insights. Their article titled “Financial Statement Analysis with Large Language Models” reveals that ChatGPT can analyze financial statements and predict earnings more accurately than human analysts if given minimal prompting.

To feed ChatGPT the researchers removed dates and company names from thousands of balance sheets and income statements obtained from 1968-2021 database that covered over 15,000 companies. Each statement contained two-year standard data only. Thereafter, this AI was asked to perform basic financial analyses including writing economic narratives explaining the results. Lastly, it had to predict whether each firm’s profits would rise or fall next year, how much they would change and its degree of certainty about these forecasts. The results were dramatic when analyzed halfway through the previous year; chatgpt improved accuracy up to 60% after being prompted.

According to researchers this performance “comfortably dominates the performance of a median financial analyst.” More interestingly were the model portfolios built on predictions made by ChatGPT. On a capitalization-weighted basis during back tests these portfolios outperformed broader stock market by an average of 37 basis points per month while they recorded equal-weighted excess returns of 84 basis points per month on similar basis. This means that smaller stocks are better predicted for earnings by our model.

These findings must be taken into account as preliminary while providing important proof-of-concept information rather than definitive stock-picking strategies. However it does raise several significant questions and matters:

1. The power of narrative: The authors showed that asking an AI to compose narratives explaining what the financial statements mean was essential for improving forecast accuracy. This aspect of analysis seems to be human-like.
2. Confirmation of previous findings: The study supports earlier research indicating that computer models or linear regressions can beat the average analyst. These models may perform better since they are rule-based and have no biases as opposed to human analysts who might be influenced by corporate reports or executive statements.
3. Out-of-the-box performance: It is interesting to note that an LLN purchased off-the-shelf was able to outperform humans substantially with simple prompts. It also outperformed simple statistical regression and compared favorably with specialized “neural net” programs trained specifically at forecasting earnings.
4. Limitations of the study: Like all social science researches, these results need to be approached cautiously. Replication and further tests are necessary in order to confirm their relevance.
5. The role of qualitative analysis: Some of the most successful stock pickers focus less on short-term earnings predictions and more on businesses’ structural advantages and long-term trends. Whether AI can make such “big calls” as effectively as it does with short-term earnings still remains unclear.
6. The ever-changing role of financial analysts: what value do these professionals provide if AI can consistently predict higher revenues than their human counterparts? Will they become the ones to explain company’s businesses to their fund managers, interface between companies and the market or support computer-driven analysis?
7. What does this mean for financial services? The ability of AI to beat median analysts or stock pickers may not be a game changer in an industry where index funds have already proved that mediocre performance is worthless. The big question is whether AI can compete with or enhance the performance of the top quartile — many of which are already making use of significant computational power.

It’s obvious that as AI advances in other areas, the finance sector would need to change. There could be some jobs at stake but also new opportunities emerging for those who can work with AI tools effectively and alongside them. In analyzing financial data, it might be that machine learning will work closely with human judgment.

This development could enable investors to make more accurate predictions about earnings and, potentially, better investment decisions too. However, it should not be forgotten that past results cannot guarantee future outcomes while markets remain unpredictable by nature.

Financial professionals on the cusp of this revolution would be wise to stay current on these advancements and think about how they may employ such tools in conjunction with their work. They might just turn out as man’s greatest friends than replacing him; although one thing for sure is that robots are here now – right now – so we’re all going digital sooner or later.”

Acknowledgment: This article was inspired by and includes information from "AI Can Pick Stocks Better Than You Can" published on FT.com. For more detailed insights, you can read the full article here.