Warren Buffett advised that you should never invest in a business you can't understand . But that hasn't stopped many investors.
New research from the McCombs School of Business at The University of Texas at Austin might help them better understand the complications of companies they're investing in. The study offers the most precise and comprehensive tool yet for measuring business complexity.
The tool, devised by Sara Toynbee , associate professor of accounting, simplifies the measurement by using artificial intelligence. It also finds that in areas such as structuring debt, complexity can sometimes be a good thing.
She defines complexity from the perspective of an observer: how hard it is to understand a company's financial position and performance based on the information in its financial reports.
A problem with measuring complexity has been that it's, well, complex.
Traditional yardsticks, such as a company's size or number of business segments, miss deeper levels of complexity in operations, risks, and financial structures, she says.
"Business complexity is a very elusive construct, because the source of complexity can vary by firm," Toynbee says. "Our model captures it across a wide variety of dimensions."
She breaks down business complexity into 29 categories, including debt, equity, derivatives and hedging, income tax, revenue, and compensation.
Classifying Complexity
With Darren Bernard, Elizabeth Blankespoor and Ties de Kok from the University of Washington, Toynbee trained a large language model — a version of Meta's Llama 3 — on 200,000 sentences from company financial footnotes.
The sentences included embedded iXBRL tags: labels that human readers don't see but computers do. They describe what a number "means," allowing the model to be trained to predict what a number represents.
After training their model, the researchers had it classify over 8 million individual numbers from more than 50,000 company reports from 2016 to 2024.
"It's like asking a sophisticated human to read through millions of sentences and then give a one-phrase description of what a number means based on the surrounding text," Toynbee explains.
The harder a number was to classify — the less confident the model was about getting it right — the higher its complexity score.
Upsides of Complexity
Complexity turned out to have business implications, both negative and positive.
- Stock price slowdown. Stock prices reacted more slowly to corporate filings with higher average complexity scores, suggesting investors needed more time to digest the information. For the most complex reports, prices took 7.9% longer to fully react than for the least complex.
- Debt stabilization. Complex debt structures are often viewed as risky, but researchers found they could help manage risk. Debt with less standard terms — such as being convertible to stock — is more complex. But it allows companies to better manage risk and provide more predictable interest payments.
"People typically think of complexity as a bad thing, but we show that in some cases, taking on complex debt can actually be helpful for businesses," Toynbee says. "It provides some companies with greater stability and better persistence."
The model could have a wide variety of uses, Toynbee says. Investors could use it to spotlight complicated companies that require closer analysis.
Standard setters and regulators could employ it to identify categories of financial information that are overly complex, she says. "They might consider ways in which they could simplify or augment reporting standards, to help investors better understand a firm's financial position and performance."
Lastly, businesses might use the tool to curb excessive complexity. "While some types of complexity can be beneficial, managers might be unaware of areas of their business that are more complex than their peers," Toynbee says. "Highlighting these areas could allow them to try to find a way to simplify such areas."
" Using GPT to Measure Business Complexity " is published online in The Accounting Review.
