Technology shares on Wall Street have fallen sharply in response to the emergence of a lowcost chatbot. Salesforce, Snowflake, few other software names falling sharply to start the week again as AI disruption fears pummel the sector. Agentic AI is basically a robot in a digital form. AI is sucking up incredible amount of capital. About 80% of the gains in the stock market this year have been powered by AI plays. In a few moments, I will sign a historic executive order instituting reciprocal tariffs. Reciprocal. That means they do it to us and we do it to them. Very simple. We begin with developments from Kiev, the ongoing war. Like the Ukrainians, I very much hope that we will have peace in Ukraine. But I am worried that we might uh push this to a cheap ceasefire which will lead to a very expensive peace. You don't have the cards right now. New strikes across Thran tonight as the USIsraeli war on Iran is expanding. They all rely on the same narrow passageway, the Straight of Hormuz. Iran's Revolutionary Guard is sending radio transmissions to ships, warning them that quote, "No ship is allowed to pass the Straight of Hormuz." Oil prices have been on a roller coaster and remain quite a bit higher this morning. We need Greenland working with everybody involved to try and get it. One way or the other, we're going to get it. Says he's serious about something, he means it. Is the United States now at war with Venezuela? There's not a war. I actually think to have a a long-term plan and a long-term mandate in these turbulent times is a very very good idea. The essence of investment management is the management of risks, not the management of returns. Benjamin Graham, the father of value investing, wrote that. And tonight is about exactly that, understanding risk, managing it, and making better decisions because of it. My name is Espen Fistru, and I will be your host for tonight. Welcome to MBIM's first ever risk summit. We live in a world testing risk managers in ways we haven't seen in decades. Wars in Europe and the Middle East, a changing global order, shifting trade relationships, and AI transforming how we work while introducing risks. uh and we while introducing risk we may not yet fully understand. In that world the work done by risk professionals by the people presenting here tonight has never been more important. Tonight you will hear from colleagues at NBIM who integrate risk thinking into investment along with experienced voices from the broader investment community. We have a full program ahead of us. So let's get started. And for your information, photos will be taken during the event and made available to all participants tomorrow. Our CEO needs no introduction. But I will say this, 20,000 billion croner, the shared savings of the Norwegian people. He is responsible for that risk. That responsibility in times like this, I'm not quite sure how he sleeps at night. Nikolai Ten. So um I asked uh I asked the risk department uh whether this was going to be a good evening and they came back with a 95% confidence interval uh and some downside scenarios. I don't need that. I only ask you is this going to be good? and I said,"Well, we cannot rule that out." Now, has your uh job as uh risk professionals become easier over the last 5 years? No, it's become much more difficult because stability has never been more unstable and we have a lot of new risks on the agenda and on the horizon. Just over the last few years I would say that the geopolitical risk picture has totally changed. We have uh AI risks, concentration risks, we have new elements of credit risks and all these risks are intertwined in a different way than they were before. AI risks going into cyber risks, AI risks going into geopolitical risks and so on. So it's just becoming more and more complicated. Now we've been working um very hard at MBIM since the start to uh develop our risk department and it really uh I mean it totally accelerated after the financial crisis because we suddenly discovered that we had some risks we we weren't quite aware of or they certainly were much bigger than we thought. Um, so we work hard. We have tried to learn from the best-in-class. We travel the world. We are lucky in that we can meet the biggest banks and the biggest operators. And they are quite willing to share with us the kind of things we do. And so some of these things we will share with you tonight. I think the next horizon for us is one to integrate more AI into everything we do in risk. and we're already doing a lot of it, but I think there's more to come. And then really to integrate the risk data into our investment decisions in an even better way than we have done in the past. Okay. So you know that we are the most transparent fund in the world and I kind of thought that that was a great thing because then people can see what we do and they will trust us more. But what I have seen also is that it makes us more aware of what's going on outside MBIM. It makes us more able to acquire knowledge from other people and that's why we are here today because we think that the more we share with you the more you are willing to share with us and we do not compete with each other in this field. We're all trying to make the trying to make the world a better and safer place. And so we hope that if there were anything you want more of after this, come and see us. And if you have any ideas for how we can get better, please let us know. In the meantime, have a great evening and have fun.

Next is a question central to risk management today. How do you measure something as complex and unpredictable as geopolitical risk, conflicts, sanctions, and disrupted supply chains? A world that may be generating tail risks beyond what our models were designed to handle. Tensin and Marta Palino have built the frameworks we use to measure and stress test geopolitical risk across the fund. Please welcome and Marta. Hi everyone, my name is Marta and I was born in 1997. When I was preparing this presentation, I thought let me go back in time and see how the world looked like when I was born and I found some very interesting headlines. Let me show you. Let's start with the European Union. In 1997, 11 countries from the European Union met in Amsterdam to discuss and agree on one single currency, the euro. And that was the first move towards a united European Union. 122 countries met here in Oslo to agree and discuss on a anti-personnel landmine pan. And there was a first push towards global disarmament. And this one I found very interesting. NATO and Russia met in Paris to agree on a funding act. They put in writing that they would leave peacefully together. And the optimism was so high that some even thought that Russia could eventually join NATO. Three different pictures, three different countries and three different situations. But one common thing underneath that in front of a challenge countries come together, sit together, discuss and find the solution together. There was a word of cooperation. If we move back in time and we move move forwards to 2016, the United Kingdom decided to leave the European Union. And less than a month ago, Hungary decided to put a veto and block 26 countries from the European Union to make a decision about aid to Ukraine. And the European Union keeps fragmenting. What about disarmament? Well, last year NATO decided to increase the defense spending up to 5% of GDP. And NATO and Russia Russia didn't join NATO after all. But last year we actually experienced Russian drones invading the European airspace. Three different pictures, three different countries and three different situations. But one common thing underneath the countries do not come together anymore. do not sit and find a solution together. We live now in a world of competition. The point I'm trying to make here is that the order we knew is not coming back. At least not in the form and in the shape it was before. And even if this might make make you feel pessimistic, I don't think it is the right read. We live in an historic period, a period of transition. And what is different compared to before is that it's moving faster. It's moving much faster. And what is happening today will have impact on the long-term investment outcomes. So let me bring all of this together. The old playbook no longer works. Geopolitical risk was something considered as a footnote for decades, but it cannot be considered a footnote anymore. It is essential for investment analysis because the assumption that countries come together, sit together, find the solution and there is some sort of stability is gone. The assumptions have changed. The logical power has changed and this that is happening today, it's already impacting the financial returns. Geopolitical risk is not something you read about in the news and you move on. it is affecting asset pricing and it sits in our investment decision. So how do we manage this risk?

So we think about geopolitical risk analysis along four pillars. Firstly inform making sure that everyone in organization is on the same page when it comes to these risks. That means across different teams and all the way up to the executive board. Secondly, simulate get together over tabletop exercises where we can play out how events might unfold. This helps us with awareness and preparedness about operational risks as well as the financial risks. And then integrate, making sure that when we have geopolitical risk analysis, it flows through our financial risk analysis and our stress tests. And then finally, consult. We're a small team. We don't have all the answers. So we rely on internal as well as external experts because in this field no one has all the answers. Our geopolitical risk scenarios focus on three main things that are important for us. It is war and conflict between great nations. It is financial sanctions and finally investment restrictions that makes it more difficult to trade and invest across nations. So how do we make these scenarios? Uh if you start on the left side, it starts with an annual horizon scanning where we identify the tops risks to the fund and then we focus on making some scenarios where we look at the tail risks of these scenarios and the economic and financial impact of these tail risks. Then we have some early warning signals that we call key indicators that tells us whether we're likely to get towards one scenario or the other scenario. And finally, probabilities. We assign probabilities on a on these based on expert judgment and then we update these probabilities on a quarterly basis. In between the quarters, we use AI to monitor these scenarios. And then we have our annual stress test which we publish on our website. This is an good example of a collaborative team effort between the risk department, the strategy research team and our portfolio managers. Here you see four different stress scenarios and the value of the fund on the left side is 20 million 20 billion uh Norwegian croner 20 billion 20,000 billion Norwegian croners. It was it was much lower when I started here and it does translate to around$2 trillion US dollars and the the scenario in the middle is we call scenario called fragmented world. This is a world where there's a total collapse of trust between countries and countries starts forming their own economic blocks and there's limited trade and investment across these blocks. In such a scenario, the fund can lose 37%. This is not a prediction. It is the tail risk. But it is something that tells us about something about what is at stake. Scenarios is the main tool we have to make sense of the world we're in. Scenarios also help us to communicate the risk across the organization and to external stakeholders. These scenarios also help us to spark the right discussions that we need. We cannot predict the path the world will take. But our job is not to predict. It is to be prepared for the journey. No matter the final destination.

geopolitical risk at the macro level. Now, let's bring it closer to the portfolio. Matthew Brunette is our global head of financing and responsible for our securities lending strategy. Through the VA is our head of credit and counterparty risk. Securities lending generates stable returns by lending out our inventory of equity at fixed income assets in exchange for a fee. In fact, it has contributed to roughly a quarter of our long-term relative performance. It is commonly viewed as a lowrisk strategy. But when something quietly earns billions year after year, you have to ask, are there some big risks here that we should be concerned about? Matr. >> Yes. aspen counterparty uh default do happen securities lending isn't risk-f free. Matt and I know this firsthand. We were both uh here in this uh building when uh Leman defaulted uh and that experience still shapes uh how we think about downside risk today. Yes. So securities lending has been very profitable uh for MBIM. So profitable in fact that the recent uh independent review of the um or active management actually single it out uh as a remarkable uh success success. So why don't we just uh uh lend as much as we can. Uh there's a catch and that's counterparty risk. If a borrower defaults uh we could face significant losses um given the volumes we lend. So the last 12 uh months we've lent securities uh worth almost 2,000 billion Norwegian croner. So a counterart uh default is a low probability event. Uh but when it happens the consequence could be large. So managing that tail risk is our joint is a joint effort between our two teams. >> Thanks Juda. >> Yeah so we we've been pretty successful with the strategy 6 billion croner in 2025 in excess return for the fund but two things it's complicated and the numbers are quite large. So, first of all, we're lending equity and fixed income securities across 35 markets globally. And at any given time, there can be more than 7,000 unique securities out on loan. Look at the chart here. The red line shows you the market value of securities out on loan at any given time as a percentage of the nav of the fund. So, this tells you two things here. one 10% to the NAV of the fund like Truda said is two trillion croner right so massive in exposure the second thing is is the it's not static over time so we're actively making decisions to scale up or scale down risk depending on the opportunity set in the market so when we're lending securities equities and fixed income we're taking other securities as collateral and and sometimes cash as collateral the the primary risk or the principal risk, the legal risk in this transaction is with prime brokers. These tend to be very large international investment banks, but we need to remember that behind them, they're facilitating transactions for their hedge fund clients. So, this becomes super important when we start looking at the asset risk in our securities lending portfolio in a Leman type situation. So in that situation, we're selling collateral in the market to try to buy back our loan portfolio, but we need to remember that the collateral that we're selling is the hedge fund longs and the loans that we're buying back are their shorts. So in other words, we're positioned the same way as very active levered investors in the market. So we we refer to this as tail risk because it is a lowrisk strategy by definition. Um but in my experience in financial markets, tail risks that are meant to happen every hundred years kind of happen every eight or 10 years. So what are the components of securities lending risk? The first is counterparty default. So what is the probability that another leman type event will happen with the counterparty that you're transacting with? Well, it's really small but it does happen, right? Um second is a loss on default. This is where we start looking at the asset risk. So what is your expectation for this portfolio of loans and collateral and how is that going to sorry how is that going to perform in this type of market environment? So think in terms of volatility and liquidity. So again not normal markets but a situation where a major bank is defaulted. Finally insurance most institutional lenders have some type of insurance around this type of risk. In the industry we call it indemnification. But the question we should be asking is again in this type of crisis market environment. What is the insurers's ability to make good on their commitments? All of this wraps up into the trading decision which is the revenue piece on the bottom. So the traders on my team I really want them asking themselves two questions. Is one for negotiating an individual transaction am I getting paid for the risk that I'm taking here? So, is the accumulated revenue over time going to compensate me versus future losses? And two, probably more importantly, how does this individual transaction that I'm executing right now contribute to the total portfolio of risk with that counterparty?

Yes. So, within counterparty risk, we assess two things as Mata alluded to is the credit risk. what is the likelihood of a counterparty defaulting and then uh the asset risk uh how much could we lose if if it defaults. So bank defaults are rare uh but they do happen. Uh we had a big scare in March uh 2023 with a collapse uh of credit Swiss and then the subsequent uh forced merger with UBS. So credit risk had been on our watch list uh already since uh March 2021 following the bank's uh over 50 billion uh croner loss uh from the ogo's default. So as we watched credit Swiss uh credit quality uh worsen uh we shifted our focus to asset risk. How can we mitigate losses in case of default? Uh we worked with financing to run down on securities lending exposures. Uh at the same time uh financing asked uh for more collateral and got it. The result we went from 60 billion croner in securities lending exposure uh to just 600 million and we were very well collateralized uh uh collateralized. So by the time credit Swiss uh actually collapsed uh we were quite confident that we wouldn't face uh any losses uh even if uh the bank hadn't been bailed out. But we could also experience a sudden counterparty default. And in that uh case, we don't really have time to run down securities lending exposure or ask for more uh collateral uh while the bank is still operating. So this means that we truly need to understand the asset risk that we are facing for the different investment strategies. So we develop and use a model to estimate losses in bad but realistic uh scenarios. Uh and this allows us to set effective requirements um uh on exposure limits, collateral concentration risk, liquidity risk, etc. So close cooperation with Matt and his team ensures we have all the necessary information and that we truly understand the different securities lending strategies that we are facing. Ultimately the requirements we set shape Matt's investment mandate uh and define the boundaries uh for which financing can operate. So counterparties credit quality as I said they can uh worsen quickly or suddenly it's very hard to know up front how things will develop. Uh so that's why the monitoring of our credit quality is uh so key to what we do. To strengthen that uh monitoring, we have recently developed an AI tool that looks through thousands of news articles all over the world in different languages. Uh and we set the bar high so we actually get the back relevant article uh with negative news. So every morning when we come in uh a news summary is waiting for us. Uh this is a great way to keep uh up to date on events uh that we should really pay attention to and we use uh uh this uh insight in a conversation with uh financing about counterparties. Here you see a younger Matt uh right after the Leman default. is looking uh quite relieved. Uh he didn't lose any money. Uh at the time uh we had no plans in place and we just had to manage the default as we went along. Uh but uh we rather be ready. So now we have detailed response plans in place. Uh we have clear responsibility across teams. uh and uh uh we all tie that together in a in a common run book. Uh this ensures uh that we knew uh we know who does what and that we do the right thing at the right time in a stressful uh situation. To improve and be ready, we practice together. Uh these drills also include uh colleagues from legal uh trading uh and operation. Uh when we find weaknesses uh we uh improve our procedures. We have for example uh uh found weaknesses in our collateral processes uh which we then quickly rectify. Uh we've also updated some of our contracts so that they are more in our favor. Uh and the legal team has taken this uh a step further. They're now using AI to extract uh key terms across our contracts, mapping trends and analyzing what these mean for our access uh to our collateral and our creditor rights. So our drills are cross team uh because when the moment comes uh it's a common uh uh we all need to be ready. So to conclude there is uh tail risk in securities lending uh and that risk need to be actively managed. We do this on three fronts. Uh we monitor the credit quality of our counterparties. We manage our assets exposure to contain uh potential losses and we test our response uh so that when the moment comes we know what to do.

Earlier you heard about our scenario analysis in the presentation on geopolitical risk. Now we will dig deeper into it. Our stress test report has attracted attention from both Norwegian and international media. But publishing the analysis is only the start. The big question is what comes next. That is why this panel will discuss how to channel scenario analysis into real impact. Whether that means portfolio decisions, educating asset owners, or simply building a deeper understanding of the risks we face. Yun Nin, head of market risk measurement here at NBIM will moderate joined by Lash Ksta Cernson from STR and Danny Schlots from PGGM investments. L Danny and Yo, the floor is yours. Okay, thank you, Espen. So in our most recent stress test report, we had four scenarios that we modeled. We had AI correction, fragmented world, regional depth crisis, and extreme weather events. Over the past year, we have actually seen uh elements of the first three in the markets. But every time this the crisis didn't fully materialize and the markets re recovered quite fast. So one one could say that we got it wrong very accurately as as RIA showed you. We we our models estimate the impact of the scenario to the nearest person percentage point but the markets didn't read the report. Now while while this is of course good for most investors, it also uh raises an uncomfortable question. If the if the markets keep defying our carefully defined uh scenario narratives, what should we actually do with the scenarios? Acting acting defensively on any of these would have cost returns last year. not acting leaves us exposed if the next one does escalate or plays out differently than our careful narrative anticipates. So th that is what this panel is about. Not the scenarios themselves but how to use them. So how to balance scenario risk against expected return and how to turn analysis into real insight and action for the organization. Now with me I'm I'm lucky to have here two friends of the fund. So Lars quickstart certensen head of uh fixed head of index and quant strategies at steward plant uh previous employer of employee of the fund. So welcome back. >> Thank you. >> And then I Danny Slots the chief financial and risk officer from PGGM uh investment management which is a large Dutch asset manager with assets of 260 billion euros. So Danny uh Lars both of you run scenario exercises. So in a few words what is the main thing that this exercise delivers? Is it quantifying risk, challenging assumptions, preparing the organization or perhaps managing asset owner expectations for what might come. Um I think that uh for us the the order would be uh organizational preparedness first because uh uh you can sort of uh prepare the organization for what is typically an adverse scenario and then uh when such a scenario hits you can sort of have the feeling that you've seen the film before you know what comes next and you know how to how to address the situation. And then the second we use uh and probably many people here do we use factor models for uh for risk and return and those are typically based on on historical correlations but in such stressed scenarios those correlations might break down and that's uh one of the values I think uh uh that scenarios do have and I know some of the limitations of these models because I I used to work for two of the leading providers of such models bar and and risk metrics. So I know they have limitations and I know that uh scenario analysis can address some of those gaps. And finally the quantities themselves the losses themselves I I don't think when you said for example that the equity portfolio would lose 53% in an AI correction scenario you know you have to take that with a pinch of salt. It's a very uncertain number but maybe the ordering is kind of correct. So so I think those those are are are good uh uh are good use cases. And if I just may add uh when uh a scenario hit us in practice when when Russia invaded Ukraine on February 24 with sort of there was a lot of confusion what to do. Do we exclude the entire country? Do we exclude stateowned securities? uh or do we exclude all Russian stocks? And and by NBIM acted first and then other asset managers followed and we divested but we knew exactly what to do because we had sort of prepared and we had like an a decision authority matrix in place. So so factory regimes are the most important for us but it's probably a little bit different for for you Danny. >> Thank you. Well uh we don't have a uniform way of using scenarios within PGM. Of course, scenarios are used and they're used all across the investment chain. Um, the the main point where we what we use a scenario for is for the strategic long-term ALM and asset allocation. So you we use stochastic uh stoastic scenario set to to mainly determine uh a base scenario and then use deterministic scenarios as the ones you've just described to test the robustness of that base scenario instead of using it as a prediction for the future. So main usage and where it really affects the portfolio is for ALM and strategic asset allocation. But all through the investment chain scenarios are used. So they're used as uh uh they're used in stress test for financial but also uh uh uh climate simulations. They're used for um decision making uh in operational uh portfolio buy uh uh decisions. Um and again those sets are not uniform for the whole organization. So there is not one let's say mandatory set for for the whole investment chain to use. Different teams different managers can use different scenario sets. Uh thank you. So if if we if we go to the tension in in in the scenario analysis. So an AI correction scenario has been on I think everyone who works in in investment management. So they have had it on their list for a few years now. So if someone would have acted or has acted on it, it has actually they have given away significant returns. So, how do you how do you balance the expected return on the one hand and then the scenario risk on the on on the other hand? Well, that's a that's a great question. I uh I recall there was the an expert panel recently by Helandre and Magnus who who sort of gave topics that the fund should be and the ministry of finance should be concerned about and sort of concentration in the US technology stocks was one of the things that they highlighted and um and I think it might even be worse than sort of just looking at the weight of mag 7 because if you if you think about this as sort a production line. Then you have uh one company produces uh the machines ASML. Uh one company produces uh the chips uh TSMC in Taiwan and one company uh designs them Nvidia and then you have the hyperscalers buying them. And so I think that uh in a in a traditional risk model these will be in different countries Netherlands, Taiwan, US, different sectors, IT, communication services, etc. And so in a traditional risk model, it might look like they're quite diversified, but it's one it's really one single bet. And so it I mean it's I think it's super interesting because it's a well worth risk uh that you should highlight. But but the question then becomes what do you do about such a risk? If you if you run an index tracking portfolios, we have both quant funds and index tracking portfolios. In an index tracking portfolios portfolio, it's not within my mandate to say I think this is too risky. I want to derisk and neither I guess can NBIM because if you are underweight communication services and it you'll blow the tracking error limit rapidly. In an active mandate, you have to assess whether you get paid for the concentration risk. So, it depends on the type of mandate. I think what you can do about it otherwise it's educational. >> Um I I agree in all the things you said. I'd like to introduce a different perspective and that's the context of um uh of the asset manager or the investment manager that we are and that is we are investing for a pension fund. So what does that mean? We h we have a very long horizon to look at and then when when we were talking about the AI bubble and and and what to do with it, we have to put it into that perspective. A lot of nervouses nervousness arises around this point when you you look at all the uh all the news we've just seen and that triggers decision makers as well. So what we need to do is to put those decision makers in the perspective of this long-term investment and then and then look at what the scenarios mean and what we should do. So the and then there's two decisions or two again two perspectives to look at is what question are we trying to answer with the scenario but because when we look at the AI bubble it can very easily be mixed with for example a geopolitical uh scenario. So which answer are we trying um to put a solution to? Is it the AI bubble? Is it geopolitical risk? Is is it a mi mix of both? So we have to separate the questions and then answer each uh question separately with the proper scenario. So that's one thing and the other thing is put decision makers back in the chair of a long-term investor. What should we do given this scenario being a long-term uh investor? And we of of course we're constantly looking at this scenario. We're looking at it again both from the AI bubble perspective and from the uh uh geopolitical perspective and trying to make sense of it all. But in balancing uh return and risk from this perspective we also have to look at the opportunity loss that might be at stake here especially in the long long term. So what we do is not only model and look at what the risk might be but also look at what the opportunity loss can be. And for decision makers we use tools like um uh what we call a premortem. The decision went haywire. Why did that happen and use that as well as the downside risk scenario for decision makers to uh to to base their decision on in this scenario. And so far, this scenario has uh the the AI bubble scenario has uh has has led to the decision not to do anything yet, but we're still looking at it because uh the world is changing ever more rapidly. So, we have to stay in a ball at every time. >> Yeah, those are good points, I think. and uh and balancing uh the expected return loss against the we know the answer is that you should hold the market portfolio that's the best trade-off between uh return and risk but then there might be other considerations from a client's uh have approached us and and want to sort of uh diversify a little bit away from AI risk. So >> can I have a question? Does uh does uh reputation play a role in this as well? Reputational risk >> uh in the sense that >> that decision makers want to lower concentration risk yes or no or uh the AI bubble or not. >> Yeah, probably. I mean uh I I we found that for example uh I think clients wants us to have a sort of tailor made solution in place where we follow something uh that is possible to track. So for example, many people want to have an equal weighted index, which I personally find a little bit strange, but but that sort of has um uh the reputational risk for the asset owner is probably less than allocating between regions. Uh if if if that's what you were after, maybe something else. >> No, reputational risk for being in in these stocks or not being in these stocks. Reputational risk from >> taking no decision versus a decision, for example. >> Yeah. No. Uh I haven't thought about it that way. Reputational risk is typically handled in in Norway by by uh following NBIM's exclusion list. >> We do the same. >> Okay. So, so a little bit on this organizational setup. So who owns the scenario exercise inside your organization and who has the mandate to act on them or make sure that that that someone acts on them in a in a good manner and does it matter? So do you see so that for example if risk owns the process the the response comes too uh cautious or if the if the front office owns the scenario analysis so they they tend to overlook the risks and uh related to this so if if we have these big positions say now the AI cluster so should the scenarios have their own risk budget or should they be be on the on the normal active management uh budget I think um it's a really good and really difficult question. So typically what uh what is in place in in asset manager organizations is that you have you take second moments into account when you construct mandates. So you have limits on tracking error and and how much you can deviate from a certain benchmark and that's easy to operationalize because you can uh measure tracking error every day and if it exceeds a certain limit for example 125 basis points you're outside the limit and otherwise you're inside. So, but scenarios are not that easy because what sort of constitutes too much risk in a scenario like for example the 53% loss in an AI correction is that high or low? Well, it depends uh what the scenario entails. So, so I think that um well, I think it's a very hard question. We don't have the answer to that. We have sort of constructed our own scenarios. Uh but we don't have like they're educational and forformational purposes but we don't we don't have a ma govern uh governance mandate in place that if a scenario entails this much loss then you need to derisk something like that. Maybe you have that in place in PGM. >> Well who owns a scenario is very much a question uh about uh what it is used for. So uh who is responsible for the use of the scenario in the actual process? So when when it comes to risk scenarios for example stress risk scenarios and uh manager or the whole of the portfolio having to be within that uh scenario limit uh and it is a risk limit then it is risk who defines the and owns this the scenario but but this is the for formal let's say responsibility or accountability of that scenario when that scenarios when risk scenarios are constructed uh the experts in the market. So the teams are always consulted but risk has the has the final say. If it if it is a risk limiting scenario when it comes to scenario use for asset allocation for example the strategic asset allocation long-term asset allocation then it then then it is ALM and strategy who own the scenarios but again they consult uh the front office functions and also the risk function when defining the scenario. So it depends on who owns it. That function is responsible for the scenario. But we try to always build in the processes that other parties are consulted when constructing the scenarios and also an outside in view. So not just internal uh views on uh the scenario but also getting in an external outside perspective when using the scenarios and getting the scenarios right so to say. Okay. Uh just uh to cap it off, can you just very quickly what is the what do you think is the like what what should we be doing more with scenarios as as an industry? Do you have a punchy line to cap it off for? Um now I think uh well tying it back to to sort of or organizational preparedness being the most important for us as a long only uh equity uh portfolio manager then I think that uh the best scenarios uh would be the ones that makes the organization look boring when it happens. You know what to do. >> We we did we didn't consult each other on the answer but the the the answer is actually quite the same. So uh key is responsiveness of the organization not to a single scenario but to a set of scenarios or two scenarios in general. So when crisis comes or when a scenario comes that you don't need to practice you don't need to be prepared for one scenario or but for when any given scenario would come and that will mean organizational responsiveness. >> Okay I I think we're we're out of uh time. if if I land this. So, I think I think we we very much agree here. So, of course, I was being cheeky in the in the start that we didn't get the exact number correct, but I I I don't think that or I agree. I think we all agree that that's actually not the point, but it's it's really this first getting the the whole organization to think this in in a common platform and then also to to be prepared when when it happens. Um before the before we break so I I I'd like to introduce we have such a good audience here now so I'd like to introduce a networking concept from Finland so so it's called the the world famous Finnish small talk and and the rules rules are very simple like it's what we do in in this type of event. So during every break, so you must happily chat with at least three people. But we we we make we make a local exception here that that you don't need to discuss only ice hockey and Sona, but you're free to do scenario analysis as well. Yeah. Thank you. Great. Thank you uh and great discussion. Now we're going to take a 20 minutes break. There are refreshments outside. So see you in 20 minutes.

That is a question a lot of people are asking these days. uh what if there is a correction in the AI valuations? What would the implications of a sharp correction in the AI valuations be for a fund like us? And that's exactly what we wanted to understand. So we stress tested our fund against this hypothetical scenario and we found out that the fund could lose up to 31% of its value which translates to roughly $650 billion US. It starts with concentration. Investments in AI have been enormous but they have been flowing through to a small group of companies. So there is a lot of money riding on high expectations from them. The trigger is simply disappointment. Uh what if the gains that are expected from AI don't materialize fast enough or what if these companies don't earn enough from AI as the market expects them to? When these high expectations come down, prices follow. And it's not just the eye companies. Their suppliers get hit. Consumers start acting cautiously. And confidence just erodess across the board. Melody Hobson once said, "The biggest risk of all is not taking one." But in real assets, striking that balance between taking risk and managing it can be especially challenging.

Real assets are different. This is the framing for the discussion we'll have over the next 10 minutes. For NBIM specifically, one of the main differences is that real assets are not in our investment benchmark. We'll discuss what this means for how we assess and manage real assets. But first, let's hear a bit about our real assets portfolio. >> So, our unlisted real assets are a small part of the fund. It's only 1.7% in real estate and 0.4% in renewable infrastructure. It sounds small, but the nominal figures are quite big. It's 372 billion Norwegian croner in real estate and 91 billion Norwegian corona in renewable infrastructure. In real estate, we have more than 1,400 properties. So, it's a big portfolio. Also, in size, it's a very big part of our relative risk because in the benchmark we have from the ministry, we don't have real assets. As a backdrop for this risk, we look at the returns for 2025 for that year. You can see that we had negative 7.5% in real estate and plus 8.9% in renewable infrastructure. This translates to minus5 basis points for real estate and plus four basis points for renewable infrastructure. Over the last years, we have seen uh some negative returns contributions from real estate and positive from renewable. And some of the factors for real estate has been that the equities we've sold to fund it has done very well. But we also have had some headwinds from structural changes to the office sector due to co for renewable infrastructure. On the other hand, that asset class has done better than the fixed income we sold to fund it and also uh there has been high energy prices and the development projects we've had have done according to plan. We will not go through any more about the returns but we will dive a little bit into the funding process which is important both for the returns and for the risk management. We have two highle ways of mitigating risks associated with real assets. The first is related to how we fund our investments as Lars just alluded to and we will discuss this now and the second is related to how we select and manage individual assets which we will touch on in a few minutes. So as we've established real assets aren't in the benchmark we get from the ministry. In simple terms, this means that in order to purchase real assets, we need to sell things that are in our benchmark, specifically equities and fixed income. If equities markets do well, then our real assets have to perform even better in order to outperform what we sold. This is what we mean when we refer to relative return and outperforming the funding benchmark. The question then becomes, how much equities versus fixed income do we sell? The framework we use to make this decision has changed over time. Whereas we used to rely on fixed percentages, we now look at it a bit more dynamically and allow the percentages to vary across assets based on their characteristics. So this funding design also help us to manage structural risks. When we buy real assets, we buy them in the same currency as we sell in fixed income and equities. So we are currency neutral. Also we have an equity portion when we fund our real assets. So the equity beta is somehow mitigated by the equities we sell to fund it. Same with fixed income risk. This ensures that the risk that you see on the right hand side of the slide are the ones driving the long-term returns and those are more asset specific and more specific to the asset class. We will dive a little bit more into those types of risks next. So we will go through five sort of defining risks for this asset class. The first being the ownership share. In equities we have singledigit ownerships most often but in this asset class we often have 50% ownership or even 100% for fully owned assets. The second one are the very high transaction costs that this assets class inherently have. We must do the risk assessments before we buy the assets through due diligence but also the understanding of the assets. It's very expensive to change your mind and roundtrip cost for such assets can be north of 5% easily. A third one which is specific to real assets is that very often they have quite distinct return drivers. uh they are explicit but they are different for the different investments. I'll take one or two examples. Buying a wind park we you would believe that it has a risk to energy prices and it of course has but very often you buy it with fixedterm contracts for the first 10 years. So the energy price risk are much more sort of to the longer end energy prices. It's very hard to get data for that and those are more that risk assessment is much more based on scenario analysis and and researchdriven research. For grids on the other hand, they're not driven by energy prices at all. Those have regulated returns and could have political risks to them. So even in the same asset class, there are quite different return drivers. >> It's hard to compare real assets because no two assets are exactly alike. But we still need to establish common risk metrics to be able to understand the portfolio level risks and to analyze new investments. One such metric is the connectivity score of an office building. This looks at public transport times and local demographics to assess the attractiveness of the location. The fifth point is a natural consequence of the other four. Real assets are private and illquid. There's no observable market price and each asset is unique. So standard comparisons don't work. As a result, fundamental research in the risk assessment is key and the assets must undergo regular valuations to establish a best estimate of fair value.

Turning now to asset management to handle some of the asset specific risks. There are some key metrics we can use to create a qualitative assessment of risk and return. So we have four examples of these key metrics here. First, development covers risks associated with construction or repositioning such as cost overruns or delays. These are actively assessed during due diligence and may require specific partner agreements as part of the risk mitigation. Leasing risk relates to the income side of the asset, so who your tenants are, how long they're locked in for, and the risk of not being able to relet the space at the rents you're counting on. An important part of asset management is to diversify this risk to avoid clustering of expiring leases. For example, on leverage, it's common for real assets to include debt financing. So, we need to assess how much leverage is in the capital structure and how that amplifies both returns and downside risk. In NBIM, we prefer to use low leverage and rather fund with a higher portion of fixed income as fixed income in the funding stack would replace debt at a lower cost and higher nominal returns. Sustainability covers ESG factors that can affect asset value. So things like energy efficiency, physical climate exposure, and regulations around building standards. Sustainability topics are becoming increasingly important in how we assess long-term risk. So that was real estate. What are we doing on the infrastructure side, Lars? >> Yeah, so we Oh, the clicker. Yeah, thank you. So we're doing a lot but I'm going to show an example of a tool that we are quite proud of. Um so we've built this tool to be scalable um and future proof. So imagine you're doing an investment in renewable infrastructure. Of course you do your due diligence. You have quality information legal etc. But you also have quantitive information that you receive in huge Excel sheets. This is the standard of the market. You get enormous Excel models. We've seen models of sort of 80 tabs, thousands of rows. Then we do the next investment. We get a new Excel sheet. It looks different. Then the next one and the next one and you see the pattern. We have we would get enormous amounts of Excel really hard and not tangible to run. So what we did did we do when we got the mandate? We quickly realized we cannot manage that type of complexity. So we started building a tool we call UVM unlisted valuation model. So what we do here is that we put all these types of models that we get from externals into such a tool and do our modeling in there. I'm going to do one or two examples of usage of that tool. What you see on screen here is that we plotted two assets that we have or example assets. One of them have this PPA contract that I alluded to earlier. It has the first 10 years fixed prices, but you are exposed to energy prices on the long tail. So you can see that if energy prices on the x-axis go up, the value doesn't go up that much. So 60% up in energy prices, 20% up in value of the asset. The other asset on their hand other hand is sort of curved. It has a downside protection scheme. typical sort of uh government guaranteed scheme in this business. You can see that on the downside we have protection same slope as the one with the PPA but on the upside we have the upside potential. So why is this important? By having it in one tool we can aggregate this risk. We can understand these risks for the portfolio as such. We can do it within one tool. Of course, I can extend this to all input parameters, production, other inputs to make sort of a unified risk framework. And also lately, we've applied AI in getting these models fast into UVM. So, we've started to see the potential for supporting the investment process so we're quick enough to get in early so we actually can model these as we buy them. >> You can keep it. >> I'm clicking for you at least. There you go. Coming back to what we emphasized at the beginning, real assets are different. They're different for the fund because they're not in our benchmark. And so we have this funding decision and they're also different due to the ownership structures, the high transaction costs, and the fact that all assets are unique. However, it's not all differences. One similarity lies in the fact that we aggregate risks and assess them in a portfolio context, just as is done on the listed side. Real assets investing is also done with the same values and goals as the other investment areas. Lastly, we are one common risk department and we cover everything from liquid listed markets to longdated physical infrastructure. This diversity means our department has had to develop deep expertise across a variety of risk environments. The accumulated knowledge is one of the strengths of how we're organized. Thank you.

Yong Shen Fu will now walk us through how AI is being used to drive performance analysis and how it integrates into our investment processes. A good example of how we are using technology to work smarter and more efficient. Yong Shen is joining us live from Singapore tonight. So, Yong Shen, thank you for staying up very late for us. Over to you. >> Thank you, Espen. So, my name is Yong Shen and I'm from the performance measurement team in NBIM. So, we have team members in Singapore and Oslo covering both Asian and European time zones. And in just one sentence, our team's main responsibility is to measure and explain the investment performance of the fund. And we also have the important job of communicating this performance to the rest of the fund on a daily basis. So today's presentation is about AI and I'm here to offer you three reasons why we're so eager to use AI to help us in our daily work. So the first reason is scale. The fund has over 300 portfolios ranging from fundamental equity portfolios that are more concentrated in nature to the very broad equity enhanced indexing portfolios. Positions in these portfolios change and global market events happen unexpectedly. So we need to always remain responsive and keep ahead of the performance at all times. The second reason is that we need the ability to do a deep dive analysis on any portfolio at any given time. Every now and then portfolios can exhibit unusual performance that exceeds our bands of expectations either on the upside or the downside. In such circumstances, we need to have a tool set that can readily explain the key drivers of this unusual performance. So we can communicate this quickly within risk and to the investment teams. And the third reason is that we have an AI team in AmbIM that is dedicated to bringing the latest in AI technology and tools to the fund. Hence, we see this is a very strong opportunity to work with them to use AI to increase our work efficiency and to help us gain additional insights that we as human analysts might miss. So now I'll jump into a use case for AI which is to identify key performance drivers in a portfolio. Now, one of the big challenges behind this use case is there can be many angles to look at in order to explain performance. Take an equity portfolio that focuses on technology stocks as an example. It could be the portfolio's industry tilts, for example, being overweight in semiconductor companies versus traditional software service companies that's driving the performance. Or it could be down to the individual companies such as Nvidia or Microsoft that the portfolio manager has selected that drives the performance. Or it could be more broad market factors, non-active positions that suddenly steer performance behind the scenes. So the main point is there are many angles to look at across many time periods across many portfolios. So it's a very time conssuming process for us analysts. So the big question here is can we use AI as a risk analyst just like one of us to help us do all this analysis. So now I'll walk you through a prototype that we have developed in close collaboration with the AI team. So at the top of the diagram we start with an AI agent that functions just like a risk analyst like one of us. And just like with any regular risk analyst, we assign it tasks. Tasks that we would normally do ourselves. For instance, to look at performance from various angles or to examine the portfolio now and look at its risk profile in order to say something about the expected performance of the portfolio in the future. The key advantage of an AI agent over a human analyst here is that it can be configured to do all these tasks at once, thus saving a lot of time. Now that the agent has task to do, it still needs to know how how to go about doing its task. It can't simply use its own performance calculations or risk calculations though. After all, it has to use our definitions and follow our methodologies. So, we provide the agent with a set of tools. Tools that are developed by us, tools that we know would give the correct results. So, although the agent can only use this set of tools, it can choose when to use them and how to use them. So, the agent uses these tools to perform its task and produce its first sets of outputs. uh as an analysis. However, it always helps to have an extra pair of eyes to look at this output. Hence, we introduce another AI agent known as the review analyst whose sole job is to evaluate this output based on a set of broad guidelines that we provided. So if it's not satisfied by the output, it writes some feedback and it goes back to the risk analyst and says, "Nice try, but here are some tips for improvements. Please try again." And the key part of the design is that the feedback loop does not happen just once, but it can happen many times with the risk analyst or agent improving its output each time until the review analyst is finally satisfied. So now I'd like to show you some sample output from the agent. So we find that on its own the agent can uncover significant performance contributors from various angles like from industries, countries, individual companies or even systematic risk factors. It also has the freedom of coming up with its own visualizations to express its key findings. for instance, time series graphs, bar charts, or even summary tables that highlight main key takeaways. So with some guidance, we find that the agent can even identify trending periods in the performance profile of the portfolio. So as an important next step, we would like the agent to automatically drill down into these trending uh periods and with all the analysis is done, choose the main important driver and summarize that concisely in a short report. So I've mentioned that the agent can uncover and highlight individual companies that uh contribute significantly to the performance and even quantify how much they have contributed. However, there is still one important analysis left to be done and that is to explain in this particular case why the company's return or stock price went up. So without AI you can imagine it is a time consuming process to scroll through all the news events of this company let alone many companies in order to uncover the reasons. So we have a solution in place that combines news with AI agents and we have been using this solution for almost a year now and we are very keen to integrate that with our agentic prototype. So the solution essentially uses two agents. The first agent takes all company news and scores them based on how it thinks uh how relevant it thinks the news is in explaining the company's returns. It then extracts highlights from this news, sends the scores and the highlights to the second agent who then looks at all the highlights and then on its own it picks the most relevant events on the company return graphs and creates a written timeline of highlights. So what we do next is to scale this process up in a back in a process that runs in the background. So suddenly we have a library of company explanations that we can readily use. This is a huge timesaver for us. So to end off, I'll like to say that the prototype shows potential and it already adds value uh the way it is to our current work processes. We have important next steps which is to improve the quality of the agent output as well as to scale the solution to handle different kinds of portfolios. At the same time, we also note that AI technology is developing very rapidly. Today's best practices may not be the same tomorrow. Hence, it's very important to keep Colos's collaboration with the AI team to make sure that our solution is up to date. And last, but definitely not least, we will maintain a strong human oversight over all aspects of our solution. With that, thank you and back to you, Espen.

We will close our program tonight on one of the most important developments in responsible investing. how AI is transforming the way we screen investments and manage ESG risk. This is not just about compliance. It's about using AI to spot risks earlier, screen investments more carefully and make better decisions about when to divest. Lisa McCarti will be joined by Christina Shisler. Between them, they'll take us through exactly how we are applying this here at MBI. Please welcome Lisa and Christina. When your portfolio holds thousands of companies, how confident can you be that none of them are engaged in human rights violations? That's one of the questions our team is responsible for answering. And today we're going to show you how we do that. So, as Christina mentioned, our team is responsible for monitoring risk across all of our holdings. So, that's over 7,000 companies across 60 countries at all times. And as Espen mentioned, this is a super important job, if I do say so myself, because the unfortunate truth is that there are some companies hidden amongst these thousands that rely on super unsustainable and problematic business practices. They'll do things like cut corners, exploit natural resources, or even violate human rights just to add a short-term boost to their bottom line. The companies I'm talking about are often poorly run with weak governance and subpar risk management. And ultimately, we believe that these companies will not be profitable in the long term. So when we identify a company that presents significant environmental, social, or even political risk and we do not believe they're equipped to manage those risks. In some cases, we can divest. So when I talk about an unsustainable business model, what do I actually mean? What are we actually looking for? It can take many forms. It could be a tech company quietly discharging toxic waste into local rivers because proper treatment is really expensive. Or it could be your favorite designer label relying on modern slavery in their supply chains. But here's the catch. Eventually, these problematic short these problematic shortcuts turn into real risk exposure with real financial consequences. This can surface through legal liability, operational breakdowns, or even reputational damage that erodess the stock price and our credibility as an investor. We treat sustainability risk as financial risk because that's what it is. It's our job to find it, assess it, and act on it. So, as I mentioned, we're watching thousands of companies all across the world, and the risk is never static. Cir circumstances on the ground change, conflicts escalate, supply chains can shift and new information surfaces. So as this happens, we have some options. We can engage directly with company management pushing for changes in conduct. We can escalate with shareholder resolutions or public statements of our expectations. We can work with our portfolio managers to reduce our exposure if we're struggling to find long-term value. And in the most serious cases, for small companies where we can't find any path forward, we can divest entirely. Since 2012, we've made 633 of these riskbased divestment decisions. So then the natural or obvious question is, what does this mean for our returns? Is our hypothesis correct that reducing sustainability risk reduces financial risk? In short, luckily, yes. Over time, our divestment decisions have generated measurable outperformance. On average, reducing sustainability risk has increased our returns. These 633 decisions have added 68 basis points to cumulative equity returns. That is 12 billion croner. But all of this only works if we can actually identify the risks in the first place. And that's our real challenge. In case you missed it the first 12 times I said it, we invest in over 7,000 companies in 60 countries. We need to monitor the entire list continuously for dozens of risks. Let me put the scale of that into perspective. Let's say we want to screen our entire portfolio for just one risk. Take human rights as our example. It would take 3,000 analysts an entire weekend for just one risk. Our team is only eight people. So, how do we close that gap? This is where we use AI. And you might be wondering, can't we just get this information from traditional data providers? Well, the issue is that they don't cover everything. We see that the most serious risks are often hidden in local language sources and in jurisdictions where information just doesn't flow through the usual channels. So, we built our own AI system to close that gap. Our system takes a company name, searches across every publicly available source. This can range from news to financial data, government records, and local media, and cross references it with our vendor data. Critically, it does this in any language in real time. This means that a news article in Mandarin gets the exact same attention as one published in English. And the output from our system is a structured risk assessment for every single company in our portfolio. So our system was built as an NBIMwide effort and it was created in collaboration with our AI team, our active ownership team and our business and investment units. The key to a successful system like this is that our internal expertise is involved in every step of the process. We spent months designing and optimizing the prompts so that the AI models know exactly what to look for and how we wanted to interpret their findings. And this step is really crucial as the output is directly correlated with how well we actually instruct it. So what does this really really look like? So we designed our screening process to run in two phases. partly to optimize for cost and processing time but also because we see that not every company requires a deep investigation. So we designed it so phase one handles the large volume and phase 2 handles the complexity. In phase one we run it using a lighter AI model designed for speed and scale. The AI searches for any indication of a company's involvement in human rights violations. And as expected, most companies clear this stage. But when anything comes up, even the slightest thread, it triggers phase two. In phase two, we run it using a larger and more complex AI model. And this is where it really goes deep. We deploy multiple AI agents, each responsible for researching the company from a different angle. This means that we can have one AI agent tracing supply chain links, another AI agent looking at direct operations, and a third one could be looking at financial relationships. Once all of our AI agents are done working, they summarize their findings and give each company a risk score. They also critically provide a summary of exactly what they found, why the score was set, and where they found the information. So that is step one. With our thorough guidance, the AI returns a list of the highest risk companies. In step two, our analysts step back in to make the decisions. And we do this in our ESG risk hub. In our ESG risk hub, our analysts in collaboration with our active ownership team review every flagged company. We verify the sources. We see if we agree with the AI models reasoning and we make the final decision. And in our ESG risk hub here, we track all of these decisions and their impact to the fund. Finally, we can move on to step three. In step three, if the risk is confirmed, so if a company that the model flagged as high risk and we agree, the company is flagged in our internal systems. Firstly, it's flagged in Polaris, which is our internal portfolio management system. And secondly, it's flagged in our investment simulator, which is an internal tool we have here to help invest to help enhance our investment decisions and give feedback to our portfolio managers. This means that every relevant investment team will have direct line of sight to a newly identified risk in their holdings. From there, a portfolio manager can choose to take immediate action such as reducing their exposure or we can choose to engage with a company directly. This means that the full chain from AI screening to human verification to portfolio action can happen within hours. So, we want to take it back to the question we started today with. How confident can you be that none of the companies in your portfolio are engaged in human rights violations? A process that was previously practically impossible now runs while we sleep. And the really cool thing is that this is not just theoretical. We've been able to catch risks and act before the broader market, protecting the fund from losses that were invisible to our traditional processes. So, we can't stand here and promise that we catch everything every time. But we can say that we have access to more information than ever before and we can analyze faster with more precision systematically every single company we hold in every language. And when we can see a risk, we can act on it. Thank you, Lisa and Christina. And thank you to every single speaker tonight. We started this evening by talking about the world we live in. Wars, shifting trade relationships, AI changing everything. The work you have heard about tonight, managing risk across all of this has never mattered more. As Nikolai has said a lot lately, we are so lucky to live in interesting times and to work in this profession. A few thank yous to Lash and Danny for joining us to all our speakers for MBIM and from everyone and to everyone in this room. Thank you for being here. Now, the canteen area upstairs is open. So, please stay connect and enjoy the evening. Don't be shy. Please ask questions to the presenters and if you have any ideas for how we can improve and be even better next time, we would love to hear them. So, thank you again and enjoy the evening.