Generative Artificial Intelligence and the Pricing of Complex Earnings Information: Evidence from the Release of ChatGPT
This paper provides evidence that the release of Generative AI, specifically ChatGPT, reduced investors' processing costs for complex financial disclosures, thereby attenuating the post-earnings-announcement drift primarily for firms with high information complexity and low institutional ownership, which supports the costly processing explanation over competing risk or arbitrage-limit theories.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
For decades, financial markets have operated on a simple, frustrating truth: information is only valuable if someone can read it. When a company releases its quarterly earnings, the numbers are public, but the accompanying reports are often long, dense, and difficult to navigate. For years, researchers observed a peculiar pattern where the stock price of these companies did not react fully to the news immediately. Instead, the price continued to drift in the same direction for weeks afterward, as if the market was slowly digesting the information. This delay, known as the post-earnings-announcement drift, has long puzzled economists. Some argued it was a reward for taking on extra risk, while others believed it was a friction that prevented smart investors from fixing the price quickly. A third group, however, suspected the delay was simply a matter of human effort: reading complex reports takes time and attention, and investors were just slow to process the details.
The question of why this delay exists has remained unresolved because it was impossible to separate the cost of reading from other factors. Previous studies relied on comparing companies that wrote difficult reports to those that wrote easy ones, but companies choose their own writing styles, making it hard to know if the difficulty caused the delay or if something else was at play. Furthermore, past technological changes usually altered the reports themselves or the tools used by professional analysts, rather than changing how the average investor reads. This meant no one could definitively prove that the delay was caused by the sheer difficulty of processing the text.
In late 2022, a new tool arrived that changed the landscape for every investor simultaneously. OpenAI released ChatGPT, a generative artificial intelligence capable of summarizing, rephrasing, and answering questions about long documents in seconds. This event provided a unique opportunity to test the theory that processing costs drive market delays. Researchers used this moment to ask a straightforward question: if the cost of reading complex reports suddenly drops for everyone, does the market start reacting faster? By treating the release of this tool as a natural experiment, they could observe whether the slow price drift disappeared for companies with the most difficult reports, while remaining unchanged for those with simpler ones.
The researchers built a system to measure the complexity of annual reports filed by companies in the United States and abroad. They analyzed the text of thousands of documents, looking at factors like sentence length, vocabulary difficulty, and the density of numbers, creating a score that represented how hard a report would be for a human to read. Crucially, they calculated these scores using only data from before the AI tool was released, ensuring that the measure of difficulty was fixed and not influenced by the event itself. They then tracked the stock prices of these companies around their earnings announcements, comparing how quickly the market reacted before the tool was available versus how quickly it reacted afterward.
The results showed a clear shift in behavior. After the release of the AI tool, the slow price drift that typically followed earnings announcements for companies with complex reports shrank significantly. For these difficult-to-read firms, the market began to incorporate the information much faster, with a larger share of the price adjustment happening immediately on the day of the announcement. In contrast, companies with simpler reports, which were already easy to process, showed no such change in their speed of reaction. This pattern suggests that the delay was indeed caused by the effort required to read the reports, and that when that effort was reduced by technology, the market corrected itself.
The effect was not uniform across all types of investors. The researchers found that the speed-up was most pronounced in companies with lower institutional ownership, where individual retail investors make up a larger share of the trading activity. These are the investors who are most likely to struggle with long, technical documents and who benefit most from a tool that can summarize them. The study also looked at companies listed on US exchanges but based in other countries, whose reports are often the most difficult to process due to language and formatting differences. These firms showed the strongest reduction in the delay, further supporting the idea that the technology helped overcome the specific barrier of reading difficulty.
To ensure these findings were not just a coincidence, the researchers tested whether the change held up under different conditions. They examined a second event, the release of a mobile app version of the tool, which made the technology even more accessible. This second release led to a further reduction in the delay for complex firms. They also looked at periods when the tool was temporarily unavailable due to technical outages. While the data during these outages was too sparse to draw a definitive conclusion, the trend did not reverse, suggesting the change was structural rather than a fleeting reaction. Additionally, they tested whether the tool simply made investors trade less or if it changed the quality of the information. The evidence pointed to the latter: the total amount of information absorbed by the market remained the same, but the speed at which it was absorbed increased.
The study also addressed whether the change was due to companies writing better reports or analysts using the tool to create better forecasts. The researchers found that the results held even when they excluded companies in the technology sector, which were the most likely to have adopted the tool for their own internal reporting. This indicated that the change was driven by how investors consumed the information, not by changes in how the information was produced. The findings suggest that the "drift" anomaly was not a permanent feature of the market or a reward for risk, but rather a friction caused by the limits of human attention and reading speed.
While the evidence strongly supports the idea that processing costs were the cause of the delay, the researchers note that the results are suggestive rather than a final proof. A statistical check revealed some underlying trends in the data that predated the tool's release, which means the results should be viewed as a strong indication rather than an absolute certainty. However, the consistency of the pattern across different types of companies, different market segments, and different time periods makes a compelling case. The study demonstrates that when the barrier to understanding complex information is lowered, the market becomes more efficient, and the slow, lingering price adjustments that once rewarded those with the patience to read disappear.
This work offers a new perspective on how financial markets function. It suggests that the value of a financial report is not just in the numbers it contains, but in the technology available to read them. A report that was too difficult to use in one year can become instantly useful in the next without a single word changing. For regulators and standard-setters, this implies that improving the readability and machine-consumability of reports is not just a cosmetic exercise, but a direct way to improve market efficiency. For investors, it signals that the profits once available to those who could manually process complex data are being compressed by technology, leveling the playing field but also removing the edge that came from sheer effort. The study concludes that as artificial intelligence continues to evolve, the nature of market anomalies will shift, and researchers will need to look for new ways to measure how quickly information is truly absorbed.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.