Could AI Trigger the Next Financial Crisis?

Hello, and welcome to this 10th Reskill Masterclass. I'm the head of Research Communication. My name is Daniel Brown, and today it's our great pleasure to welcome Thierry Foucault, who is one of the most experienced professors in HEC's finance department. It's really a pleasure to welcome you to this reskill masterclass, where you're going to answer the question: Could AI trigger the next financial crisis? A very topical subject, I believe. After your 20-minute presentation, you'll be answering questions sent in by viewers from all over the world. Thierry, the floor is yours.

Thank you Daniel, and welcome everybody. I'm Thierry Foucault. I am professor of finance at HEC since 1998. Before I start discussing the question for today, let me first explain how I got interested in this question. A large part of my research is about the effect of information on financial markets. As I'm going to discuss, artificial intelligence is a technology to produce information. So it was natural for me to start thinking about the impact of artificial intelligence on financial markets.

The second reason is that from 2020 until 2022, I became the scientific director of I Paris. I Paris is the joint center between HEC Paris and Ude Polytechnique on artificial intelligence. One of the goals of the center is to develop research on the impact of AI on society. This collaboration with the institute stimulated my interest in this topic. The last reason is that two years ago I got a grant from the EU, a so-called ERC grant, to work on the effect of AI and big data on financial markets. My talk today is based on the findings of this research.

The question is coming from Guy Gensler. For those of you who don't know him, he is the current chairman of the SEC, the watchdog of US securities markets. In many reports and interviews in the Financial Times, he has expressed concern that maybe the next financial crisis could be due to the increasing use of AI tools in financial markets. We should take this seriously for at least two reasons: he is the chairman of the SEC, and he has lectured about AI at MIT and produced interesting papers on the potential and risks of AI in financial markets.

Among the people attending the class today, 66% think AI will be the cause of the next financial crisis, and 44% think it will not. Opinions are divided. To start thinking about whether AI could trigger the next financial crisis, we need to understand what artificial intelligence has to do with financial markets and why this technology is important for finance.

Artificial intelligence is a technology to extract information from vast amounts of data and transform it into predictions and decisions. In finance, this means predicting stock returns or corporate earnings. Big data is the raw material, and AI is the technology that transforms it into something useful for decision-making. AI uses machine learning algorithms to mine large amounts of data and extract predictions about the future.

This is important because a core function of the financial industry is to produce information. A security analyst uses financial statements to forecast corporate earnings and predict stock prices. Credit rating agencies use predictions to grade the credit risk of a firm. Banks assess default risk to decide whether to make a loan. Many financial occupations involve forecasting, so AI is naturally impactful because it changes how forecasts are made.

Traditionally, security analysts relied on financial statements and expertise. Increasingly, they use alternative data, which is different from firm-controlled data. Alternative data includes social media posts, satellite images, and transcripts of earnings calls. Every digital footprint online is potential data. Specialized firms collect, clean, and sell this data to the financial industry, acting like a refinery. The number of firms selling alternative data has grown exponentially since 1996. One provider, RS Metrics, has more than 2,000 data vendors and 20,000 datasets.

The question is whether this data can predict corporate earnings or stock returns. Research shows it can, but mainly for the short term. Analysts rely on alternative data to improve short-term forecasts but become worse at long-term forecasting. Our research with colleagues at EA Oli and the University of Lugano shows that over the long run, analysts are better at short-term forecasts but worse at long-term forecasts, especially in industries well covered by alternative data. Alternative data pushes analysts’ attention to the short term.

Another concern is job displacement. Machines are replacing humans in finance. Larry Fink, CEO of BlackRock, explained that traditional fund managers are being replaced by quants, skilled in processing vast amounts of data using AI. Traditional expertise may still have value in some cases, but the trend is towards AI-driven decision-making.

We also studied alternative data from satellite images of retailer parking lots in the US. Quants use this data to predict sales. When RS Metrics starts covering a retailer like Walmart, active fund managers relying on traditional expertise perform worse in that stock, whereas quants are unaffected. This shows that AI shifts the advantage to those able to process large data, and traditional skills may be displaced if they cannot adapt.

Traders are also affected. Algorithmic trading has been replacing humans for 20 years. New algorithms can learn independently, such as reinforcement learning models like AlphaGo. In finance, similar AI could behave in unexpected ways, potentially destabilizing markets. Experiments show algorithms may learn to set non-competitive prices, creating risks for investors.

Will AI trigger the next financial crisis? I don’t know. There are risks: attention may shift to the short term, humans may lack the skills to work with AI, and algorithms may behave unpredictably.

Questions from viewers:

Cedric from Montreal asks if machines could destabilize financial markets. Yes, algorithms have caused erratic price movements, such as the 2010 flash crash. Algorithms learning independently may behave as black boxes, making it difficult to regulate.

Mauricio from Rome asks about AI’s impact on the real economy. Financial markets influence corporate investment. If short-term forecasting dominates, firms may reduce long-term investments, such as climate transition projects, because markets fail to value them.

Dr. Armin Pern asks if models are poorly trained. Some competitive outcomes in experiments were limited by insufficient training. Training AI in real markets is costly, and feedback may involve financial loss.

Paul from Washington asks about the role of quants. Quants currently represent about 8–10% of active assets under management, but the trend is increasing. Human judgment may still be valuable for long-term or novel predictions.

Fel, an executive MBA candidate, asks if high-frequency trading exacerbates volatility. Algorithms react faster than humans, potentially responding to noise rather than signal. Circuit breakers exist, but risks remain.

Bruno DuPont from France asks if AI can outperform humans at forecasting. Studies show machine learning can outperform humans, but hybrid models combining human insight and AI predictions are often superior.

Samantha Pierre from Toronto asks about artificial general intelligence (AGI). AGI can program and train itself. Self-learning algorithms are black boxes, creating challenges for regulation.

Thank you very much, Thierry Foucault. Final words? No, thank you. Hello to former alumni and viewers. Tune in next year for the next masterclass.