Enhanced risk management: Leverage new technologies for stress testing
===

Ali Curi: Hi everyone and welcome to Markets ConversatION, I'm Ali Curi. On today's episode, we're discussing how risk management is evolving to meet a new era of threats from climate shocks to deepfake disinformation. Joining me is Marco Martinolli from ION Markets, who will discuss how financial institutions are rethinking how they stress test, simulate, and plan for disruption.

There's a lot to discuss, so let's get started.

Marco Martinolli, welcome to the podcast. 

Marco Martinolli: Hi Ali. Thank you for having me here. 

Ali Curi: Marco, before we get to our conversation, let's learn a little bit more about you. Tell us about your background and what your current role and responsibilities are at ION. 

Marco Martinolli: So I'm based in Triste, Italy, where I work as a quantitative analyst in the risk and compliance department of LIST, subsidiary of ION Markets.

In my role, I manage different projects in risk management designing customized solutions for banks and insurance companies. And actually most of these projects deal with stress testing and therefore, I'm excited to be here to share some of my insights on how technology is shaping this critical area.

Ali Curi: Let's start with the big picture. What are some of the major changes you're seeing in the risk landscape and how are they reshaping how firms approach stress testing? 

Marco Martinolli: We can say that uncertainty has always been a constant in the financial ecosystem, but in the last decades risk factors are evolving very fast and this causes an increase in the frequency that we are observing in the rate of financial shocks and crisis worldwide.

Indeed, there are some new emerging risks that are specific of our times and that need to be addressed into our risk management applications. There are geopolitical risks that are materializing nowadays, quite evidently as form as military conflicts or shifts of economic alliances or increases in trading tariffs.

Then there are the climate related risks. Here we do not refer just to the natural catastrophes or the extreme weather events that cause damages and operational dysfunctions, but specifically for banks and financial institutions. Also, there is a secondary class of transition risks that are very important, that are more related to the changes in policies that favor green markets and sustainability against high emission industries causing new imbalances in the markets.

And then finally, there are risks that come from technological disruptions. Indeed we know we are living an era where there is an extreme digitalization with real time payments, digital assets, cryptocurrencies that are changing structurally the financial systems. And of course, there are also innovations that are very important, like the AI boom, blockchain technology, quantum computing, cloud computing that offer massive opportunities to improve systems performance.

But on the other way, they open to new vulnerabilities. Say first, in cybersecurity. And from the risk management point of view, this represents a big challenge in guaranteeing stability for financial systems. And that's why firms nowadays are looking at stress testings as a strategic tool to test the resilience of their portfolios, also in simulated stress conditions.

Ali Curi: Now firms have traditionally looked to past crises in order to model future risks. How effective is that approach today? 

Marco Martinolli: The study of historical crisis and events can still provide valuable data on market dynamics or systemic vulnerabilities that can raise in stress conditions. But as we said before, risk landscape is changing dramatically fast.

While traditional models often rely too much on past data, and this means that they do not consider drastic extreme tail events or the hidden new risks and they just not reflect the current speed of propagation of financial shocks in the world.

Ali Curi: Marco, so what does a modern forward-looking stress test actually involve? Can you walk us through that process? 

Marco Martinolli: Of course, but first let me make a brief overview of the state of the art in stress testing. So we start from the type of scenario analysis where the risk analyst explores the future possible state of the world making plausible but extreme assumptions. So basically he builds a narrative of events that can be inspired by historical events or can be totally hypothetical and identifies key variables to define the scenario.

Typically they are macro economical variables such as GDP or unemployment rate and the impacted risk factors in the scenario. Then he moves to the impact simulation phase where the specific bank portfolio is projected into the stress conditions of the scenario. And we measure the impact on bank exposures of the stress assumptions to measure and evaluate key metrics that can be used for monitoring phase where the risk strategy of the bank can be adapted to reduce the vulnerabilities highlighted in the stress test. Now moving towards a forward looking approach, I think that we first need to prefer hypothetical scenarios.

Indeed, we said it before, future won't look like the past, so we need to explore more dynamically the scenarios making plausible but extreme assumptions working with imagination. Then we need to incorporate novel risk factors related to the emerging risks, and perhaps also micro economical variable, such as local trends of the markets, sector, profits or costs.

That can be very useful to represent specific local scenarios in case of banks with exposure concentration in certain sectors. And finally, of course, we are talking about technology here. It is essential today to integrate modern technologies, to optimize processes and automated tasks in stress testing operations.

Ali Curi: So how do firms handle the challenge of making sure stress tests meet regulatory rules, but still reflect the firm's specific exposures? 

Marco Martinolli: It is a delicate balance for firms, but in the last years actually, there have been some changes into the approach for these regulated stress testing exercises.

Indeed we are moving away from the old approach, check the box approach, let's say. Because this is not anymore sufficient to meet regulatory requirements, but authorities actually encourage risk managers from each bank to personalize, customize the stress scenario to the specific portfolios, geographies, and business models of each bank.

Each risk manager should make its own assumptions on the credit risk models, liquidity management, and so on, and explain later the deviation from the standard methodologies. Another novelty is that regulators are asking more and more for thematic stress tests that target specific risks, novel risks typically, to examine how banks cope with the target shocks.

And one of the latest examples is the 2025 EU wide stress test mandated by EBA, which starts specifically from a geopolitical event.

Ali Curi: Now, can you share with us an overview of how AI and automation are being used in risk management today? 

Marco Martinolli: AI is a game changer that is transforming all sorts of fields, including risk management, of course. And indeed we are seeing banks and firms moving towards the enhanced risk management framework where modern technologies like AI are integrated into many operative and management processes.

And these kinds of enhancements come from techniques like machine learning and neuro networks to learn from data, training data, scan high volumes of data with low latency, almost real time, and ultimately find, identify, hidden factors inside the data. So from the point of view of risk management, this opens to big advancements into higher prediction, accuracy and efficiency, upgrading pricing models and trading strategies, real time monitoring of systems and operations especially in terms of fraud detection or preventing market abuse. And finally, thanks to automation, specifically for repetitive tasks, we can reduce costs, manual error, and improve compliance.

I want to give you an example to explain the potentiality of AI integration into risk assessment, specifically for credit risk assessment. Where the competition of credit scoring is much faster, thanks to AI, because we can use also some alternative data sources like history of transactions or credit history of each client to make credit scoring faster, and of course reduce time to delivery of loans.

And also specifically for the context of, stress testing, AI can be used to automate the generation of stress scenarios based on latest events and with extreme velocity.

Ali Curi: Before the AI boom, big data modeling were the go-to form of analysis. Will that continue, especially unstructured inputs like news or social media. How useful is it still for understanding market sentiment or emerging risks?

Marco Martinolli: Oh, it is still very useful. And actually we can say that AI definitely unlocked the full potential of big data just because using machine learning combined with natural language processing, we can navigate and analyze the streaming data from social media, news articles, forums, scientific publications, whatever, and detect the weak signals that are very important in this data. By the way, I recently read in a publication from "Statistics in Technology and Telecommunications," that they estimated that the daily generation of data to be about 400 of million of terabyte. This is a huge amount of information that properly employed can really upgrade the prediction models and market simulations perhaps by forecasting the nonlinear market behaviors that were not captured by the traditional models or by having a sort of social listening to have the sentiment analysis to improve the strategy review.

Ali Curi: You mentioned that firms might explore using AI-generated stress scenarios to account for events not found in historical data. Why are these scenarios important and can you share an example? 

Marco Martinolli: Building scenarios is about storytelling with data. We define coherent narrative of events and assumptions, combining data from macroeconomic variables, regulatory frameworks, and mitigation strategies.

So AI has access to all this information and furthermore, it can have an eye always pointed to the latest emerging risks. Compared to other providers of stress scenarios like regulators or risk managers, AI can be seen as an alternative, valuable tool to dynamically explore scenarios with high velocity of generation being always up to date and with extreme adaptability of the risk strategies of the bank.

So you asked me for an example of a AI-generated scenario. And I would like to share one that is particularly stuck with me, which is about an AI disruption, specifically deep fake technology, that is thought to improve so much in the next future to be hyper realistic. Let's say three to five years, there will be a sort of a crisis of truth where deep fake videos can be almost undistinguishable from reality. And in this scenario specifically there is a assumed, let's say that there is a deep fake video that gets immediately viral, of the ECB president, which announces an emergency raise of the rates of 500 basis points, causing of course, sudden crash of the markets and a crisis in the trading system.

And this of course sounds like fiction and at some level it is. But it highlights specific vulnerabilities in this case the risk of digital misinformation or the weakness of algorithmic trading that are after all grounded in real economic indicators from today.

So the insight of scenarios like this one, can suggest to anticipate crisis, invest for perhaps on technologies like blockchain that today is thought to be the most promising solution to check authenticity of digital information and reduce potential losses like this one.

Ali Curi: Now Marco, from a solution standpoint, what are some of the most important areas where firms need to upgrade or rethink their stress testing capabilities, besides the scenario generation? 

Marco Martinolli: In my opinion, AI integration is the main step to take to enhance stress test capabilities.

But you must be aware that it requires large investments in infrastructure, modern implementations and maintenance. With that being said, the first advice I would like to give is to address specific risks that come from AI integration, and I refer to biases from training data, lack of transparency, or the difficulties in guaranteeing data privacy.

To do that, we need to have high data governance protocols. Pipelines for data quality and extreme processes for validations of the AI models thanks to expert judgment. So risk managers working as supervisors or maybe using some explainability techniques called X AI that are specifically designed to interpret results from AI models.

Another guideline is to invest on cloud computing. Indeed, the migration to scalable cloud native platforms allows to enable big data storage and large scale simulations across scenarios and portfolios. And additionally, this kind of services usually offers some AI tools for model retraining or data quality or DevOps operations.

And finally, I observed that many times stress testing is perceived as a verticals activity in the bank. But actually we should foster cross-functional cooperation between the different departments; market, credit, liquidity to incorporate also cross risk interactions and link together different features like capital planning, liquidity management, and perhaps ESG risks. 

Ali Curi: Now continuing with solutions, how should teams start incorporating things like climate risk, geopolitical shocks, or say cyber threats into their models?

Marco Martinolli: Well, in order to incorporate these emerging risk factors in stress testing, we need to first extend the data model. So we need to use external sources of information like ESG disclosures or insights from geopolitical analysts. We need to enrich anagraphic tables of assets and counterparties using also data from the special analysis or emission data.

And in agreement with DORA, we should classify all the cyber incidents, operational dysfunctions, digital risks to measure the loss events and results. Once you have collect all this information and you keep updated this kind of information in your data model, you make the step forward and you can build a library of stress scenarios, picking events from these emerging features, so having stress scenarios from geopolitical conditions, cyber specific loss events, or climate related disasters. Specifically for climate stress testing, there are some reference scenarios provided by NGFS that can be customized to specific bank portfolios.

And also there are some recommendations. One of these, for instance, to extend the time horizons of stress testing up to 10 to 30 years to capture the longer term degenerations that are linked to climate impact.

Ali Curi: Marco, you've shared a lot on how the risk landscape is shifting fast. What's the one big thing you hope listeners will take away from the episode? 

Marco Martinolli: In a period of, increasing uncertainty, like the one that we are living today, I think that modern technologies should be used not just to react, but also to anticipate crisis.

And in this context, AI, as I said, is a game changer. But at the same time, there are no game rules. So the challenge for us technology providers is really to find a clear way to integrate these kind of solutions, even if there is not a standardized way to do that. And therefore my personal opinion is that in this transition phase, we should adopt a hybrid approach where risk managers are still essential to monitor and supervise AI models using their experience and expert judgment.

Ali Curi: Now one bonus question, a bit of a sidebar. What's an important lesson you've learned in your career so far that others can learn from?

Marco Martinolli: Oh, let's say that my motto is, "Do Not Fear Crisis." Today we have talked a lot about anticipating crisis or preventing crisis, but in my personal experience, they do not have just negative outcomes.

They actually can be opportunities to rearrange systems and open way to personal growth. So in my case specifically the key is to work on my stress management skills, to survive the time of the difficulty and enjoy the learning process.

Ali Curi: I think that is a very unique perspective and some great advice. Marco Martinolli, thank you for joining me today and sharing your insights. Let's do it again soon. 

Marco Martinolli: Thank you, Ali. It has been a real pleasure. 

Ali Curi: And that's our episode for today. You can follow ION Markets on X and on LinkedIn. Thank you for joining us.