Financial stability is crucial, but crises like the 2008 housing collapse and the COVID-19 pandemic have shown just how vulnerable the market can be. These disasters have also demonstrated a clear need for robust risk management systems capable of preparing for and mitigating such scenarios.
Financial engineer Mathieu Tancrez is looking to enhance these systems by incorporating artificial intelligence tools that can effectively anticipate and plan for such risks. To this end, he’s developing StressGen, a tool that generates scenarios based on past global crises and future hypothetical possibilities, helping companies prepare for a wide range of financial shocks.
Keep reading to learn more about Tancrez’s solution for enhancing market risk management operations.
Building a Career as a Tech Advocate
Tancrez is an engineer with a degree in financial mathematics from École Centrale de Lyon. Shortly after graduating, he began working as a risk manager at major consulting firms like Deloitte and Nexialog. At the same time, he became intrigued by artificial intelligence as companies worldwide began incorporating this emerging technology to analyze large datasets, automate manual processes, and gain insights into their operations.
Specifically, Tancrez explored large language models, which process large datasets to comprehend human text, generate new content, and serve as the backbone for platforms like ChatGPT. Once he understood the technology, Tancrez started looking for ways to successfully implement LLMs into risk management.
“I started asking myself,” he recalls, “what financial workflows could be automated, enhanced, or replaced by generative AI?”
Eventually, he found a compelling use case for LLMs: generating stress test scenarios.
The Role of Stress Tests in Risk Evaluations
Financial institutions use stress testing to run simulations against their portfolios to see how they would perform under extreme market conditions — such as recessions, deflationary spirals, hyperinflation, liquidity crises, etc. Stress testing is an essential and often mandatory regulation for major institutions since it tests financial resilience, identifies existing vulnerabilities and risk factors, ensures the sufficient allocation of capital to withstand crises, and helps risk managers set up effective contingency plans.
However, most traditional stress testing metrics rely heavily on historical data, and institutions often delay the updating or implementation of forward-looking hypothetical scenarios due to the significant resources required. This presents a major limitation since, as Tancrez points out, “any major change in the current market cycle, regulation policies, or a portfolio’s main risk factors can make those scenarios instantly outdated or meaningless.”
This overreliance on historical data also fails to account for what is known as “black swan events” — rare or statistically improbable scenarios that have no historical precedents and therefore can’t be easily predicted (such as the 2008 financial crisis or the COVID-19 pandemic). This leaves companies unprepared for unprecedented market shocks.
Enter StressGen.
StressGen: Tancrez’s Solution to Financial Stress Testing
To alleviate these shortcomings and help financial institutions better prepare for the unexpected, Tancrez is developing an API that he calls StressGen to help institutions prepare for both historical and hypothetical scenarios. Its functionality is straightforward: After a company inputs potential risk factors from its current portfolio, StressGen retrieves and replicates applicable historical crises in order to determine potential financial shocks.
StressGen utilizes an advanced technique known as retrieval-augmented generation to enhance accuracy. By being trained on historical crises and continuously retrieving real-time information from external sources, StressGen ensures more reliable and up-to-date data when generating scenarios. For example, it may pull in data from recent news articles, stock trends, and official financial reports. With this information, it assesses the current economic context, identifies similarities with previous historical periods, and ideates potential situations and even extreme scenarios called “black swan events”.
Each generated scenario, whether historical or hypothetical, provides a comprehensive analysis that outlines risk factors and estimates financial shocks across various terms, such as the term structure of key U.S. interest rates.
By comparing current market conditions with historical periods of economic turmoil, Tancrez believes StressGen can offer a more accurate and comprehensive financial analysis than traditional methods. “This model provides a deeper understanding of how past crises and current market conditions can relate to each other. With that perspective, it can help us identify potential scenarios that we’re currently not accounting for,” he says.
Tancrez also notes that he and his team are exploring the potential to calculate the probability of each scenario, which would represent a significant advancement.
Enhancing Market Risk Management with AI
While continuing to develop StressGen, Tancrez is also in charge of developing similar Python-based tools at Crédit Agricole CIB in New York, where he’s served as a resident risk manager since 2021. Additionally, he’s currently scheduled to speak at the Risk.net conference in October, where he’ll discuss his accomplishments in risk management assessment and new technologies like generative AI in the market risk framework.
Thanks to the efforts of professionals like Tancrez, financial institutions can integrate AI into their risk management operations to gain a better perspective of potential risks and develop effective strategies to mitigate them.
Copyright © 2024 California Business Journal. All Rights Reserved.
