Stress Testing: Current Issues, Regulatory Analysis, and a Sneak Peek at the Future

Speaker A: Foreign hello everyone and welcome to our latest financial risk podcast. My name is Robert Sales and I am the editorial director at garp. Today we're going to discuss stress testing challenges and trends with an eye on how regulation and recent events are shaping these important exercises. Regulatory stress tests play a key role in ensuring that large banks hold enough capital to withstand extreme recessions, while internal stress tests at banks are used for everything from capital and liquidity planning to risk monitoring, risk identification and operational resilience. Today, though, there are questions about whether regulatory stress tests, particularly the Federal Reserve's CCAR exercise, are transparent enough. Internal stress tests, moreover, are being heavily influenced by heightened geopolitical risk and US Policy changes such as fluctuating tariffs to more effectively manage all of the different scenarios they must consider. Financial institutions are also making greater use of next generation technology like artificial, uh, intelligence and machine learning in their stress testing methodologies. Today we're grateful to be joined by Chris Deridis, the deputy chief economist at Moody's analytics and the author of Risk Intelligence's Modeling Risk column. Across the past two decades, Chris has written many thought provoking articles and insightful reports on stress testing and has well informed perspectives of the issues currently facing regulators and banks and the trends that could shape the future. Thank you for joining us, Chris.

Speaker B: Thank you, Robert. Always a pleasure to be here.

Speaker A: Late last week, the Federal Reserve released the results of this year's so called ccarve stress test for large banks. What do you think were the key findings of this exercise?

Speaker B: Yeah, so, uh, first of all, I'd say based on the overall results, the test was a success. We had 22 large banks taking the test. They all passed. They all have sufficient capital to absorb more than $550 billion in losses and continue lending under a severe economic scenario. So from that standpoint, the core part of the test is just to see how banks would perform if things really went off the rails and they do have sufficient capital. They actually have more than double the capital needed to, uh, meet the regulatory minimum. So from that standpoint, success. One thing I'd note is that with all banks passing their common equity tier 1 capital, that's so called CET1 capital ratios forecast fell only 1.8 percentage points and that's the smallest decline that we've seen in recent years. If you go back to 2000, you've had declines that were in the range of 2.1% to 2.8%. So this year just a 1.8 percentage point decline. A couple of reasons for that are worth noting. Here one is just given the dynamics of the economy and the fact that the stress test is countercyclical, meaning, uh, that we always project an unemployment rate that rises up to 10%. That means that because the unemployment rate actually did tick up this year, we started the test at a little bit of a worse position than we did in 2024. The scenario ended up being relatively less severe. And so that translated into lower losses and the lower decline in the capital ratio that I mentioned. So that was the bulk of the reason why we experienced that decline. Just the nature of being in a somewhat worse economy relative to previous years. On top of that, you did have higher bank profitability last year. So think bank profitability has been improving and that certainly helps in terms of the, uh, capital position as well. You also had a couple of technical factors, I would say, in terms of changes to the test. One is the treatment of private equity. In the stress test itself, the private, uh, equity exposures of banks were removed from the global market shock portion of the test and evaluated instead under the macroeconomic scenario, which were a bit more favorable.

Speaker A: Right.

Speaker B: So that certainly contributed as well. And then another perhaps technical, more one off factor is the fact that you'd have some atypical, uh, trading positions back in October of last year. So banks benefited from some of that trading. So they, again, they went into the test with a somewhat better portfolio than they did in previous years. But nonetheless. Right. That doesn't discount the fact that it's still very severe. Test. Right. You still have an unemployment rate of 10%. You have very significant declines in house prices and CRE prices under the test. So I don't want to discount the test was quite severe, but nonetheless, banks performed quite well under it.

Speaker A: So overall, ah, pretty positive results. But are there any red flags that stand out, particularly with respect to, say, credit card losses and trading losses in the Fed stress testing report?

Speaker B: Yeah. We can always, uh, take a closer look at the losses and it's important to do so to understand what are the weaknesses or relative weaknesses, if you will. Right. You could still have a very strong balance sheet. You could still have a lot of capital, but that doesn't mean you're immune to losses. So one thing that did stand out to me is that the credit card segment represented a disproportionate share of the losses. About 20% of the total projected losses is due to credit cards, which in the grand scheme of things is a small portfolio segment.

Speaker A: Right.

Speaker B: Relative to, say, commercial and industrial loans or commercial real estate lending.

Speaker A: Right.

Speaker B: Credit cards are Relatively smaller portfolio. So the fact that they account for 28% of the losses is something that we need to pay attention to. The loss rates on credit cards are projected to be relatively high 16.9% right now. That doesn't mean it's a bad business. There's a reason why credit card interest rates are higher. It's in part to cover these higher losses, loss rates. But the one thing that sticks out to me is that if you look at recent credit card performance, say credit cards that originated back in 2022 or 2023, their performance is not that great. It's actually been deteriorating. So that's still in a relatively strong economic environment with unemployment rates still relatively low. If we were to experience more of a downturn, if unemployment rates were to rise. Right. Certainly you'd expect those delinquency and default rates to rise. And so certainly that's something we want to pay attention to now. I don't think that's a financial system risk here that we're talking about. It's a relatively small part of the portfolio. Banks have a lot of mitigation strategies to deal with credit cards. It's a well established, mature business, but still something we want to pay attention to. Perhaps. The other thing, uh, that really stuck out to me was the fact that you had quite a wide variation in individual bank performance. So even though overall, on average, things look quite, quite strong, you do have some banks that are performing much worse than others under the stress test. Right. Where their capital ratios do fall much more substantially, uh, than what we saw for the average overall. So if the average was 1.8%, as I mentioned, you had some banks that fell maybe by 4 percentage points. Right. So that does speak to some differences in risk management practices. So that's worth a look. If we're thinking about specific banks. On the one hand, you could think of that as a positive in the sense that we don't want banks to be monolithic entities. We don't want, uh, the crowd thinking out there. So some variation certainly is welcome to understand that banks are doing their own thing, they're doing their own analysis and making their own strategic, uh, decisions. But on the other hand, it does suggest that, hey, maybe some banks are worth taking a closer look at just to understand why are their loss projections so much higher than others? Are there things that they're doing, some decisions that they're making that are inherently riskier than others? So that's something I would point out as well.

Speaker A: What about with respect to trading losses? Can you talk about that a little bit, yeah.

Speaker B: So trading losses are part of the equation as well. What kind of stuck out here is banks did have a reported $44 billion loss on, uh, trading portfolios, and it was concentrated in a few of the larger, most complex institutions. Right. So it's not so much the number, I would say $44 billion relative to more than trillion dollars in total aggregate capital is not a big deal. It's something that could certainly be absorbed, but the fact that it is fairly concentrated in complex institutions, that always makes my ears prick up. There could be interconnections between institutions that we want to pay attention to. So that's certainly something to take a closer look at. Again, not suggesting that this is a red flag that's going to take down the financial system, but certainly something that sticks out as a, uh, potential factor that we should investigate a bit further.

Speaker A: I appreciate that kind of look under the hood, you know, when we step back and take a look at the bigger picture. One additional issue is transparency. Earlier this year, in a lawsuit that the American Bankers association and other banking groups brought against the Fed, questions were raised about the opaqueness of ccar. And in short, the lawsuit argued that the stress testing exercise was too opaque and that the Fed's models produced, quote, unquote, vacillating and unexplained requirements on bank capital. Interestingly though, before the lawsuit was officially filed, the Fed promised to, quote, unquote, improve the transparency of its models and to reduce the volatility of the capital buffer CCAR imposes upon banks. Do you think CCAR needs to be more transparent and less rigid? And what steps can the Fed take to achieve these objectives?

Speaker B: Yeah, in a nutshell, I'd say yes. This is not a new criticism, actually. I'd say from, uh, the very beginning of stress testing, we go back to 2009, banks have complained about the lack of transparency and the volatility of the tests. And the Fed, to its credit, has taken certain steps over the years to be more transparent in terms of disclosing more and more about the model. But still, uh, I think there's more work to be done here. I'm certainly a big advocate for transparency. I think the better informed the banks are in terms of the test requirements and what the major factors or risk factors are, the better they are able to plan and make decisions going forward. The counter argument that has been put forward is that too much transparency could lead to gaming of the system. Right. If you know all the answers to the test, then you're going to adjust your portfolio in such a way that you'll pass the test, no problem, or you'll take risks that perhaps the test doesn't fully uncover. Uh, it can be somewhat sympathetic to that, especially if we were talking about a test that was very rigid, if it was really just a scorecard. If you had so many assets in this bucket, you get this type of grade. If you have so many assets in another bucket, a different type of grade. But I don't think that's what we have here. I think the test can still be dynamic. The Fed can still update its models. Nobody's saying that they need to set these models in stone forever and never make a change. But, um, having more transparency in terms of what are the drivers, how are the models developed, what's the validation of the models? Right. I think that all would help not only in terms of the quality of test, but then just in terms of the confidence in the test. And that's a big, big component of performing the test. Overall, the results are important, but it's actually the confidence generated by taking the test that really leads to a lot of the value of the test. So I think we can be more transparent here without the danger of gaming. We should certainly pay attention to that. Um, so I think the Fed is taking steps in the right direction. They were already doing so. Perhaps the lawsuit only accelerates, uh, that process. And then in terms of the volatility there too, I think there are reasons to agree with the fact that there has been some volatility in the results from year to year. So that, again, makes it somewhat difficult to plan if you think that they particular loan has a certain loss profile as assessed by the Fed in one year, and then suddenly it changes the next year, that can make it difficult to plan going forward. So taking steps to reduce some of that volatility, or at least to explain it, would be helpful. A, uh, proposal that has been put forward now is to average the test results over two years. Right. So that certainly can help to smooth things out a bit. Although I would argue if, based on this year specifically, it actually might in the other direction. Right. You'd have to end up with more capital versus what the test, the single year test result would say. But nonetheless, I think that's a reasonable step to take to just smooth out the path here. And again, the banks have more than doubled the required amount of capital. So it's not really a concern about, you know, just barely passing the test. They're passing the test with flying colors. It's just, um, Pointing to more of the transparency and better ability to manage through her plan for the future.

Speaker A: Right. So it's a complex issue, more transparency without allowing banks to gain the system. Right. Right now I'd like to talk a little bit about current events and I wanted to get your take on the Trump administration's constantly shifting policy on tariffs and how that has impacted the stress testing landscape this year. We've seen on, um, again, off again tariffs for Canada and Mexico increased, but fluctuating tariffs for China and the evolving tariffs situation with Europe. I imagine that this is all difficult for financial institutions to track because there are so many different potential scenarios. What challenge does this present to modelers? And do you think that firms have significantly increased the frequency of their stress tests for tariffs?

Speaker B: Yeah, great question. Certainly this is the hottest question in the modeling and risk management space these days. The fact is that for many years, decades, we haven't worried about tariffs.

Speaker A: Right.

Speaker B: Uh, um, that hasn't been even an issue to any large degree in terms of financial market or credit loss impact. Right. We have tariffs here and there along the way, some increases in steel tariffs perhaps, or in other particular industries that uh, banks have had to manage. But this is certainly a much larger proposal in terms of broad based types of tariffs. And then it's just the announcement of the tariffs in such a short period of time. There wasn't really an expectation certainly of this type of tariff action occurring simultaneously across multiple countries, multiple products. So that's certainly shaken up the modeling community in terms of trying to understand how to uh, first of all incorporate tariff effects at all, because as I mentioned, they had basically been excluded from the model wasn't viewed as a key risk at all. Right. So this wasn't on the radar screen to a large degree. I think the greater challenge now is how do we deal with all the volatility in the day to day announcements. And that presents a significant challenge from a quantitative modeling perspective because it's not as though you flip a switch and everything runs instantaneously. If you want to understand a specific tariff proposal, you have to dissect it a bit. You have to understand what products, which industries, which industries within your specific footprint or region may be impacted. It becomes a very almost company specific impact. So difficult to generalize. So that introduces some complexity in just developing a scenario. So it could take several days just to develop a new scenario. And by the time you develop it, the rules on the ground may have changed once again. There may have been an extension or a new announcement. So that's just Made it very difficult from a quantitative standpoint. So what I've seen is a lot of qualitative analysis. So you do have a lot of banks and other firms certainly looking at the tariffs and not taking any immediate actions, perhaps taking more of a wait and see approach. But they are gaming through kind of broad strokes. What could this potential set of tariffs mean for portfolios or individual businesses? That's kind of been the first approach here. But there's certainly a lot of work being done to consider, uh, extreme kind of tail risks as well, if the tariffs were to go in a much more negative direction, or if they were to go in a much more positive direction, just to understand what the range of possible outcomes are here. But as you mentioned, it's just a very volatile time in terms of trying to understand what the impact would be on, um, portfolios. Not so much from a regulatory perspective, but more from a business management or portfolio management perspective, if indeed the tariffs go in a different direction than what the consensus thinks they will currently.

Speaker A: Yeah, all strong points. I think the volatility and the uncertainty has to be just a huge headache, uh, for stress testing modelers. I can't imagine. Speaking of uncertainty, what role do you think geopolitical risk currently plays in stress testing? I mean, we have the Iran Israel military conflict, which is of course grabbed headlines, complicated most recently by US involvement, and we also have ongoing wars between Ukraine and Russia and Israel and Hamas. And have these events forced financial institutions to perform more frequent stress tests across areas like market risk, portfolio management, and operational resilience?

Speaker B: Again, the short answer is yes. Uh, there's just a lot of events going on simultaneously and very difficult to understand the interactions between them as well. Right. So not only do we have these tariff effects going on, they're also going on in a world where you do have these geopolitical factors. And that just adds to the complexity and the potential for exponential risk, that the combination of risk factors, which on their own may seem manageable and something that could be dealt with or mitigated relatively easily, once you combine these factors, suddenly you're in a whole different world and the risks kind of multiply with themselves exponentially. So that just creates a lot of difficulty as well. I'd say the other thing about geopolitical risks specifically is that it is so unique or one off. Right. It's not like modeling credit risk where we have a nice long history through multiple cycles and you can get a sense or you can estimate what the behavioral response would be to certain economic factors. Here we're talking about really unique new situations that they're evolving very quickly. Potential impacts on oil prices, commodity prices, other economic factors or interest rates are occurring in a very short period of time. And by the time you finish again, one type of analysis or one analysis of an impact, suddenly the world has changed again and you have other players coming in that might affect some of these factors as well. So oil prices are perhaps, uh, one example. That's a, uh, primary channel, I would say, for the US Economy to respond to a lot of these geopolitical events. And you just look at the price of oil day to day and you can see that not even the most expert traders have a great understanding of the future. Things are changing very, very rapidly. And the models, the stress testing that we do needs to change as well or be ready to adapt.

Speaker A: Right.

Speaker B: We don't want to overreact every little movement, but we need to be ready when the more sustained type of changes might occur.

Speaker A: Yeah. Again, the uncertainty, the volatility. Just don't know what's going to happen next. All that's got to be a big challenge for stress testing modelers. In addition to geopolitical risk, I think another stress testing trend worthy of discussion is the evolving role of AI and ML. Over the past two years, the GARP Benchmarking Initiative has conducted global stress testing surveys of financial risk managers. Interestingly, this year, 1/3 of respondents said that their firm uses AL for some aspects of internal stress tests, up from 25% in 2024. Additionally, 13% described AI ML as integral to their stress testing process, nearly double the number from 2024. Have you seen any uptick in the use of this technology for stress testing over the past 12 months? And what do you see as the primary stress testing role of AI and ML?

Speaker B: Yes, I'd say absolutely. I don't know if you asked the question in your survey, but I'd say that 100% are interested in AI ML. Right. They may not be using it quite yet, but exploring it or thinking about ways to integrate this new technology. I've seen it being used perhaps most effectively so far, just around documentation and analysis of results.

Speaker A: Right.

Speaker B: If you're running a stress test and you're doing it properly, you're going to produce a, uh, mountain of output. Right. A lot of information about the portfolio, about different segments of your portfolio, about different regions or different cuts of the portfolio. Very difficult as a human to process all that type of information. And AI ML is just very well suited to consider all the forecasts that come out of a stress testing process and help us to identify or summarize not only the overall results, but the specific combinations or parts of a portfolio that might be particularly vulnerable. And this can be quite useful. Right. So if we use the credit card example, uh, once again, right. Perhaps we're worried about credit cards overall, but if we use our tools and our analysis, we may be able to drill down and identify what specific segments or combination of characteristics are particularly vulnerable. So which customer should we really be paying very close attention to? And that might drive our origination and servicing strategy. So that's how I see a primary use of, uh, AI ML it's not so much to develop the risk model itself. I think there's still a bit of trepidation here, certainly from the regulatory side, about just letting the models go wild or AI go wild on data set and produce a model. So I don't think we're quite there yet. Maybe we'll eventually get to that point. But where I do see it being used quite effectively is in assisting a modeler to develop a mathematical set of equations. So you have a programming tool, um, as a co pilot that might be sitting next to a modeler that can really make them much more efficient, suggest, um, new variables or new procedures that they might want to explore. So still under the hand or under the guidance, the primary guidance of the human modeler, but continuously or additionally providing some assistance, like a research assistance to that model just to make them more productive and ultimately allowing them to develop more and more accurate models, more timely. Right. So I think that's a big benefit of these tools as well, is just the ability to turn things around much more quickly in a very uncertain, dynamic type of world. So even if modeling might be human driven, the ability to use these tools can then automate the models that are produced and turn around results in a much more efficient, faster fashion. That's where I see a lot of the value, at least so far. I think eventually we may get to the point where we hand over more and more of the tasks to AI, But I think again in a regulatory environment you want to be a bit cautious. The transparency, the confidence are really tantamount as well. And if you can't explain the model or understand how it was developed, uh, that certainly is a risk factor as well.

Speaker A: Yeah, that would be a problem. You have to explain it to management, to the board. And you talked about a little bit of trepidation in terms of AI. Uh, there's still concerns about hallucinations and bias and things of that nature. But I think overall it seems like it's trending in a positive direction. As you say, it'll be interesting to see where we're at maybe 12 months from now. Speaking of trends that could impact the stress testing landscape, we've seen some interesting developments in climate risk over the past 12 months. US regulators like the Fed and the OCC have either scaled back or de emphasized climate risk regulation. That's quite different of course, from the proactive approach of the ECB and other European regulators to climate change. Do you think these differences have altered the way financial institutions are stress testing for climate risk? Has this led to, or will this lead to a change in, let's say, the frequency of tests? I realize this depends in part on um, where a firm is based and whether it has a global reach.

Speaker B: Yes, certainly there's a bit more of a bifurcation in terms of climate risk analysis that's going on, I would say, particularly in terms of transition risk. Right. So the um, risk that changes in policies might have on certain financial assets or lending programs. Right. So if certain countries adopt a carbon tax, for example, that's going to have certain implications for industries that use a lot of carbon, use a lot of petrochemicals, for example. And so that's something clearly a lending institution, a bank still needs to consider. But different countries around the globe are going in different directions when it comes to transition risk. So in the US there certainly is less movement in that direction. There's m much more movement in terms of trans transitioning away from carbon usage in Europe and Asia. So there are different stress testing programs going on there that are certainly geographic specific. One area though, I think isn't really changed all that much based on policies is the physical risk aspect.

Speaker A: Right?

Speaker B: Physical risk from natural disasters like a hurricane or a wildfire, a flood. Right. If you are holding mortgage assets, real property assets, loans that are backed by real estate, for example, that is exposed to these physical risks, you still need to care about that. Right. Regardless of what the policies are. And to the extent that you do see more powerful storms or more frequent storms in certain areas, that certainly is something that stakeholders are taking a closer look at, wanting to understand better what the implications might be for their portfolio. We also of course have the shifts in the insurance industry that are having a tremendous impact. Impact in terms of the availability of insurance and the cost of insurance, the annual premiums charged to say, homeowners or other property owners. That again, financial institutions need to be aware of, need to incorporate into their model if those insurance costs continue to rise. That certainly is going to impact the ability to pay the value of the assets. And that's something that a bank or other financial institution needs to be concerned with. So again, I see that aspect as continuing. That's not really going to be affected by policies to any significant degree. And I see that as universal. I see banks around the world globally paying a lot of attention to those physical risks and the uh, changing environment and of course the uh, insurance implications. I think those are increasingly becoming important in terms of a real time reaction to climate risk. Right. You're seeing those insurance increases happening now. Right. And that's having a very direct impact in certain markets.

Speaker A: Yeah, great points on the differences in the approaches to transition risk and physical risk and the uh, impact of changes we're seeing in the insurance industry on banks. Now I'd like to ask you to look into your crystal ball a bit and talk about the future. How do you think current events could potentially impact stress tests in 2026? Will there be more of an emphasis, for example on the impact of geopolitical shocks and tariff, or is it too early to say?

Speaker B: Oh gosh, my, uh, crystal ball might be a little cloudy on this one there. A lot can change between now and February when the next round of stress tests will be uh, kicked off. I think what we can say though more from a policy or procedural standpoint in terms of some of the statements coming out of the Fed is that we seem to be coalescing around a process or a set of scenarios which consist of a baseline severely adverse scenario which is consistent from year to year. Right. Again, unemployment, uh, rate that rises to 10% regardless of where we start. So just a very nasty, very severe type of scenario that's been around since the beginning of the process and I think that continues. And then a third scenario or set of scenarios that might focus on some area of interest. And this year's stress test we looked at nine financial bank institutions or the exposures of banks to these non financial banking institutions which are not regulated.

Speaker A: Right.

Speaker B: Certainly an area of interest from year to year. I think the Fed might reserve that type of test to just investigate certain risks that seem to be emerging. And again not so much to pick on individual banks because the individual results of these particular tests are not released. They're just looking at things from a more systematic perspective and I think that's a, uh, reasonable type of approach. I would suggest the process is well established by now. All these institutions know how to run a baseline in a severely adverse scenario. So there's no real surprise there. And then this third scenario, if you ask the institutions themselves, I think they mostly agree that what the Fed selects is top of mind. Right. Fed has looked at commercial real estate in the past NFBIs this year. So I think that's the process we'll be heading towards. What specifically might be in that third bucket or in that third scenario? That's going to depend on trends between now and the end of the year. But in terms of that process, I think that's where we're landing. So no more, no less, no fewer scenarios.

Speaker A: I would say it'll be interesting to see what's top of mind early next year when the fed develops its CCAR exercise for 2026. Lastly, are there any stress testing trends that we should keep an eye on in the remainder of 2025 and going into next year?

Speaker B: Yes, I think in general in the U.S. we've already kind of alluded to some of the industry pushback or concerns about transparency, whatnot. I think that type of regulatory evolution is on the horizon here in terms of disclosures, in terms of some of the requirements of the stress test, just making it a little bit easier for banks. We certainly could see changes in ratios as well. That's certainly being discussed in terms of leverage ratios or potentially even capital ratios here. There's some discussion there in terms of healthy debate in terms of what is the appropriate level, uh, of capital to ensure that banks are safe and sound at the same time they are providing the credit that is necessary for the economy to function. So I think that debate will continue and I think that's certainly a trend we'll see globally. I think, uh, you might see some different trends. The US remains in a fairly favorable position here from an economic standpoint. Lots of risks obviously, as we've discussed, but still underlying fundamentals, at least up until now, remain relatively, uh, strong and certainly stronger than some other parts of the globe. So regulators, banks in other areas, other parts of Asia or Europe, may be a little bit more, more concerned perhaps about stress testing their institutions, just given the fragility of their economy. So we might see some differing trends there with those other regulators continuing to be very scrupulous, looking very closely at institutions, worry about certain risks. Whereas the US because of the strong capital positions that we have among banks, we might see more of a, uh, lighter touch in terms of regulations going forward. But let's see what happens here. I think there are other areas of interest that continue to arise here. One I, uh, continue to worry about is cybersecurity risk. I know a lot of financial institutions certainly think about cybersecurity risk on a daily basis, and so that certainly could be an area where the Fed might want to focus a little bit more on the operational risk side of things versus the credit risk. So we might see some movement in that direction. Just to make sure, as there is a trend towards more cyber assets or, uh, cryptocurrencies, there might be more emphasis on ensuring that financial institutions have all the measurements in place to mitigate against potential cybersecurity threats.

Speaker A: Yeah. So cybersecurity, digital assets, uh, among the risk factors to keep an eye on going forward. Thank you very much, Chris, for sharing your enlightening viewpoints as well as to all of our listeners for joining us. Uh, I'm afraid we've now reached the end of this podcast. If you've enjoyed the show and would like to stay up to date with new episodes, we'd encourage you to subscribe to, uh, our Financial Risk podcast series, wherever you get your podcasts. If you'd like to peruse more of Chris's views on stress testing, credit risk modeling, and other timely risk management issues, Please go to garp.org risk intelligence modeling risk that's all for today. We look forward to you joining us in the future.