2026 State Street Markets Research Retreat

Speaker 1: Okay, we are at our final session, but it is one you will be very pleased you stayed for. Um, it's an absolute honor to present our keynote speaker, Professor Daron Acemoglu. Professor Acemoglu is an institute professor at MIT. He is faculty co-director of the Stone Center on Inequality and Shaping the Future of Work. He's a research affiliate at MIT's newly established Blueprint Labs. He is also an elected fellow of the National Academy of Sciences, the American Academy of Arts and Sciences, the American Philosophical Society, the British Academy of Sciences, the Turkish Academy of Sciences, the Econometric Society, the European Economic Association, and the Society of Labor Economists. Quite the list. He's also a member of the Group of 30 and somehow has found the time to write co-write six books, including the New York Times bestseller Why Nations Fail Power, Prosperity and Poverty, The Narrow Corridor states societies in the Face of liberty and power and progress, a thousand year struggle over technology and prosperity. His academic work covers a range of areas, including the political economy, economic growth and development, technological change, inequality, labour economics and the economics of networks. He holds honorary doctorates from Strap In. It Takes a while.

Speaker 1: The University of Utrecht, the Bosphorus University, the University of Athens, Bilkent University, Bath University, Ecole Normale Supérieure, French sur Saclay, Paris, and the London Business School. Wow. Omg, am I right? Um. If I listed all of his awards, we would seriously be here all night, and I really want to get to that gin and tonic that I mentioned earlier. So I'm going to mention just one award. In 2024, he was awarded the Riksbank Prize in Economic Sciences in memory of Alfred Nobel, or as we know it, the Nobel Prize in Economics, where little known fact he actually beat me into second place. So there's always next year. Darren. Um, Darren's going to be taking part in a fireside chat with Sarah. Sally, who is executive vice president and head of State Street Investor Services in North America. And I am reliably informed that Sarah is a huge fan of the professor and so very excited about doing this. Now, after Sarah's chat with Darren, we will have plenty of time for audience questions, so get your questions ready. This is a really rare opportunity to pick the brain of one of the greatest living economists, Sarah professor Acemoglu. Over to you.

Speaker 2: Sir. Thank you. Lee. Um, my introduction seemed a little bit short compared to professor. That's a really hard one to follow.

Speaker 3: Well, Lee had training from New York, so. Okay. For me.

Speaker 2: That's great. But most importantly, you're right. I am a huge, huge fan of Professor Acemoglu. I feel a little bit like. Ron, please. Okay. Thank you Darren. I feel a little bit like a groupie today. I think, um, I've been looking forward to this for a long time. Your book, Why Nations Fail is one of my favorite books of all time. Like top five ever. I've made a few of my kids read it. They were really happy about it. If you haven't read it, um, you know, obviously. Um, Darren. Darren and his work have been noted and recognized for the premise that it's really institutions, right? And these the concept of inclusive institutions versus extractive institutions, which really, under my long term prosperity and economic prosperity in countries, um, countries and regimes. But what I find extremely fascinating about the book, Why Nations Fail. And like, there's a copy of it there. It's, it's a big book, but it's not a tome. Um, it is like a tour de force of everything you've ever wanted to know about anything in the modern era. So it's, it is everything from ancient Rome to medieval Venice to Mayan city states to conflict in the Congo and Sierra Leone. I mean, North Korea, South Korea, creation, it just is on and on and on. Um, and so just the level of writing and editing in that book is really something incredible. I highly recommend it. But we're here to talk not just about why nations fail, but more broadly about a number of the things that you write and where you've been quoted. I think these are all really very, very current themes and very relevant for for what we're seeing today. But you have written most recently that we are in a period right now of a real rise in shifts, be it technology, geopolitics, competition. What is it about the current environment that seems to you structurally different than past cycles that we've seen?

Speaker 3: I don't know exactly why all of these things are happening at the same time, although I can give you my theory, but certainly some of the shifts we are seeing are once in a generation or once two, general. Once every two generation kind of shifts and all of them simultaneously. And some of them may be more major as well. Ai obviously, which is in everybody's mind, whether you want it or not, and you're going to hear even more about it, promises to be truly transformative beyond any of the technologies that have defined the modern age beyond computer, the internet, electricity, even if half of that is realized. That's big. It will change the way we communicate. It will change the way we educate and get information. It will definitely change the way we organize production. It will change our institutions, our social lives. At the same time, not entirely unrelated. We are seeing a crisis of liberal democracy around the world. So since I am a believer that institutions matter, They matter both in the short run and in the medium to long run. Both for the economy and our social lives. What's going on with democracy is also once a generation kind of shift, and we're just at the beginning of it. And in the meantime, we are also seeing all of the global balances geopolitically, economically, in terms of alliances. All of these are shifting in ways that create a lot of tumultuous uncertainty. You can add to that list climate change demographic shifts, especially with rapid aging in a number of countries, but pretty much everywhere where the world population will stop growing and will start declining at first since the Black Death. So those are really big shifts and you can see them everywhere except in investment and stock prices. So perhaps something for you guys to think about.

Speaker 2: Yeah. So maybe we drill into that a little bit, especially AI, which is what you kicked off with. I mean, the markets are pricing AI as a huge productivity boon, right? You seem to be a little bit more skeptical about AI and what the implications are going to be for society. So maybe talk to us a little bit more about that.

Speaker 3: Yeah. I mean, I am definitely not such a skeptic that I would question the potential of AI to bring significant productivity gains. The problem is. What are the exact magnitudes of these productivity gains, where they're going to be realized, how quickly they're going to be realized and what the negative effects on both other companies and also the social fabric of society will be. Each of these is a significant source of uncertainty. For example, if the timing or the extent to which it can be monetized is not up to what fuels the huge investment boom, that implies a big correction in the market. If some sectors don't benefit, or if the AI boom displaces some companies, that also has major risks. But even more importantly, given how disruptive AI is, has already been even though the productivity effects are nowhere to be seen yet, it's been very disruptive in education. It can be very disruptive in production, especially for many types of workers who might lose their jobs or youngsters who may not be able to get jobs in the sectors that they have been trained for or in any sector will see. And some of the other disruptive effects, it is possible that there will be a lot of regulation and the implications of those regulations, both for profitability and the direction of AI and and its consequences are generally uncertain. I think nobody can say with great confidence how the policy space will look like as far as AI is concerned, and even 2 or 3 years time. So that's just an incredible amount of uncertainty. Add to that the whole issue of where AI leadership will lie, where profit opportunities will be. So it may well be that some of the expectations About advances in foundation models such as large language models, the GPT, etc. are realized that does not guarantee huge profits for companies such as OpenAI, Google or Anthropic, because there may be close enough substitutes that the foundation layer of the AI stack may be commodified. And as a result, that's not where the profits are going to be. So there's so much uncertainty here that is really mind boggling.

Speaker 2: When you think about that uncertainty, right? And I want to maybe take a step back and talk about this, this concept of, you know, these inclusive institutions because like that's.

Speaker 3: Which includes institutions, they're not well, I'm kidding.

Speaker 2: Well, that's what made me think about it, right? Because you're talking about AI. Um, the potential to put some controls around it. And where I go is, do we have the institutions that are ready to put controls around, around AI? So maybe for the benefit of people who haven't read your books yet, um, talk a little bit about just that concept, inclusive versus extractive institutions, institutions. And then, and then maybe we can talk about, you know, is there a way to, you know, put some scaffolding in on the AI world?

Speaker 3: Yeah. So, uh. James Robinson and I, in our work, especially in why nations fail, we define inclusive institutions as the kinds of things that we sometimes take for granted. The scaffolding for well-functioning markets, secure property rights, reliable legal system. Uh, no restrictions on people choosing their occupations and where they're going to be employed. Fair rules for competition. So that, for example, when there is a procurement, it doesn't go to the company that's connected to some powerful politician, but it's the lowest bidder or the one with the best technology that gets it. An environment that creates incentives for innovation and investment. None of these are either automatic or universal. Far from it. Throughout human history, we've seen a rich array of institutional arrangements, together with supporting norms, and very few of them approximate what I just described as inclusive institutions. And I just, you know, focused on the Economic aspect. But then, you know, these inclusive economic institutions need to be bolstered by political arrangements. And the ones that we argue in why nations fail are synergistic to inclusive economic institutions are inclusive political institutions where the sort of the logic is similar that instead of concentrating power, you create a more level playing field in the political domain and include the right sort of checks in the same way that the legal system and the right regulatory environment does in the economic domains, both inclusive economic and political institutions have been the exception rather than the norm over human history. We've seen plenty of examples of very top down systems designed so that a particular group is advantaged or power is concentrated in the hands of a family or a person or a social group, or there is complete breakdown of laws and enforcement so that property rights are not secure. Corruption could be endemic.

Speaker 3: People have very rarely experienced freedom of movement or freedom of occupation. So those are the realities that we face. And we often, in sitting in the United States, underestimate the degree to which those kinds of institutions have shaped the economic trajectories of other nations, and often not for the good. But the ironic thing is that when we were writing this book in the late 2000, the US was, despite all its idiosyncrasies, a great example of inclusive institutions. Uh, a couple of months ago, I got an email from a journalist that said this paragraph in your book, were you anticipating what's going to what was going to happen in the United States? And I said, oh, I wish no, that was a description of failing African nations. But, you know, these issues of institutions is not one that's far from home. We are living through a period of major institutional changes. And if you believe half of what we say and what I have spent decades researching, that is going to have economic consequences, not sometimes. Sometimes that may be a little delayed, not this year, but some of the implications about less competition, more favoritism, less, uh, fairness in the legal system. All of these things could have major impacts on innovation, which has been the engine driving the US economy. And the thing is. Once those effects start appearing, they are harder to reverse. Because another aspect of these institutional perspective that we emphasize is that there is a lot of persistence. There is a vicious circle. Once institutions get weaker, they empower those who are benefiting from those institutions. They also destabilize the situation. They destroy the norms that were bolstering the institutions. So it could be a rather dangerous dynamic that we are very much in the middle of. Yeah.

Speaker 2: Um, so institutions I was at dinner on Sunday where our. Former fed chair, Jay Powell, was awarded a profile in courage right here at the JFK Memorial Library. And he gave a very pointed acceptance speech. Um, what what are your thoughts about that independence of the fed? You know, what will you be looking for in Kevin Warsh and like the evolution of those fed governors and the the organization, like what will be what will what will be red flags and what's going to give you confidence?

Speaker 3: Well, look, the fact that you we came to award a courage award to the fed chairman for doing his job already tells you where we are. And this is a.

Speaker 2: I want to say it, but.

Speaker 3: I can say it.

Speaker 2: Thank you.

Speaker 3: Thanks for now. You know It's not guaranteed. That's one of the things that we take for granted, too. And that's part of liberal democracy. But. The fed situation is important both because it matters so much for the economy. But it's also emblematic of the broader changes that have been happening for quite a while. It's not just Trump, although Trump's administration, uh, first administration and especially the second administration accelerated it, but there has been an increase in the concentration of power in presidential hands, partly because of other dynamics related to, uh, breakdown of bipartisanship. But if you look at the number of executive orders, they have been increasing steadily since, uh, the Bush terms and, uh, but. The implications of those are really much more problematic when that kind of executive power starts undercutting the independence of institutions, especially those that have been built in order to ensure the smooth functioning of the economy. Independence of central banks hasn't come out as a theory out of thin air. It was a response to a situation that we witnessed repeatedly across the world, in the United States and other industrialized nations, and also in the developing countries, where monetary policy became not a tool for ensuring price stability and, and, and sort of, uh, well calibrated expectations on the side of the investors and the business sector in terms of what's going on with credit and, prices, but it was a tool for incumbent politicians, presidents or prime ministers to play with the economy in order to increase their popularity.

Speaker 3: And that, you know, led to disastrous effects with the stagflation experience of the 1970s, which gave rise to demands for independence of central banks. And then that was also incorporated into the Fed's mandate and functioning. And it hasn't been challenged until what we are experiencing now. And the way its challenge is particularly ugly. It's not simply that a president is putting pressure privately on the fed to adopt monetary policy. That would be in his short term interest. But it's taking the form of very questionable tactics, both in terms of rhetoric and in terms of. Court cases and so on. So that's really an alarming change in. And it's. The tip of the iceberg, because when you see the most well protected agency in terms of its independence, be subject to this, you know, imagine what's going on for those that are more behind closed doors and don't have the same sort of protections.

Speaker 2: Mhm. Um, maybe along those lines, um, I think it was today or yesterday. Um, we, uh, there was an announcement that the fed, so the tariffs were struck down, tariffs were struck down earlier in the year, obviously, but now the administration has found is approaching a way to introduce certain tariffs through sections 301 and 302. I think that was the announcement today or yesterday. Um, how do you think about that and how do you think about, um, again, sort of institutions, right. And the legitimacy of institutions when you, when we are working to work around court rulings and judicial rulings and find other ways to enable in this case tariffs, is that does that do anything to, you know, compromise the soft power of governments and governmental institutions? What are your thoughts there?

Speaker 3: Well, say anything you like about this administration, but they've been very creative in using strange statutes for imposing tariffs. But let me separate your question into two parts because they're both very, very important. And thanks for raising them. The first one is about tariffs. There is nothing in economic theory that says you should never, ever have any tariffs. Moderate tariffs could be justified under certain circumstances, but there is no justification whatsoever for having arbitrary tariffs that you can change that are country specific, sector specific, company specific. That is not just a particularly pernicious way in which you can threaten not just other countries, but domestic companies, because you can say, I can put tariffs on your supply chain, and it creates a huge amount of uncertainty tomorrow. You don't know what the tariff will be. Therefore, we don't know how much your inputs are going to cost and what's going to be what's going to happen to trading patterns. And when you look at it, it looks very much not like an economic rationale, but a political rationale. It's just part of this agenda of concentrating power in the hands of the president or the or the administration. If the president can set tariffs today and change them completely tomorrow on a specific country, that is a huge amount of power. But the bigger issue is exactly what you have hinted at with soft power.

Speaker 3: We live in a globalized economy, globalized world, whether you like it or not. And many of the challenges facing today facing us today, including AI, including climate change, nuclear proliferation, pandemics, but also broader issues of inequality and taxation, have a very major global element. What that means is that we need global governance and global rules, some sort of coordination, some sort of expectations, the way that previous American administrations have approached that, like other leading countries, is by trying to. Sustain a system of rules and expectations, at least with allies. For example, during the Cold War, bolstered by both hard power and soft power, but mostly soft power. What that means is that it isn't just the threat of sending bombers and. Ships to punish a country that deviates from the trading patterns that or trading approaches that the US favours. But trying to create the kind of moral suasion that comes from the expectation that the US is going to support the system. It is. It has an agenda that's not just selfish but will benefit other countries. And there is support for things that are more sort of inspiring for everybody, like democracy, freedom, economic flourishing and so on. Has the United States done that consistently? Far, far from it. You know, we've supported CIA coups around the world.

Speaker 3: We've sometimes, uh, thrown our support behind, uh, dictators such as Mobutu in, uh, in the Congo or Pinochet in Chile. But the general approach for sort of trying to win, win hearts and minds was a very important part of US policy, and it worked to some degree. People became supportive of democratic institutions in their own country all around the world, partly because of the US experience, partly because of the US image and leadership, and partly because of US support this administration, together with a philosophical shift that goes back to several ideas in the first half of the 20th century has completely eschewed soft power. This is like the elimination of the USAID, for example, was part of that and has adopted an attitude that all that matters is hard power. And I think that's ultimately self-defeating. It's ultimately self-defeating because hard power doesn't have that much capability to bolster a system. And unless you have the normative justification that what you are doing has some benefit and has some objectively good features, it's going to be very difficult to sort of sustain it as a system. So that means that we have that that that was what I meant by the geopolitical risk. The big geopolitical risks aren't just rooted in US-China competition. They are rooted in the United States undermining a system that worked, albeit imperfectly.

Speaker 2: Mhm. So since you mentioned wars and we couldn't leave here without talking about current Middle East conflict and the markets, and you're in front of a group of markets professionals here, but the markets have largely been shrugging it off. I mean, obviously, you know we trade with the opening and any news on Hormuz? But for the most part, its markets have been. It's been a dampened response to to news in the Middle East. Um, thinking about what you're saying about soft power. Are the markets right. Or is there something more structural or is there a bigger shift that we should be watching?

Speaker 3: Well, look, this is your area. You should give me the answers, but let me pose the question, or at least explain my puzzlement and ignorance so that you can give me better answers. I'm truly baffled by how resilient the markets have been interpreting the US economy at least, but also to some extent, the global economy. Because the US economy is so integrated with the global economy as completely robust. It's like a Teflon to which nothing sticks. And I'm puzzled by that. I'm puzzled by the fact that the risk of all of this massive AI investments are not going to pay off is nowhere priced in the market. I'm puzzled that the energy crisis that we are in the midst of, which is then very, very likely to get more severe because the reserves that we have available are not going to be able to be enough to prevent the oil price, so far as I can see, from increasing for perhaps another 30, 40, even 50%. How that's going to then affect food prices and all sorts of inputs into other industries. So those risks, as well as the risks from US institutions and the geopolitical balances being topsy turvy, I don't see them anywhere in the stock market. Investments in bond spreads or anything like that. So you tell me why there shouldn't be there.

Speaker 2: We are going to have time for questions at the end and comments and answers.

Speaker 4: In a Q&A.

Speaker 2: Session. Um, I don't know. Maybe that's a good opening. If there are, does anybody have any answers or questions to that?

Speaker 3: Yeah.

Speaker 4: Let's start with I don't know.

Speaker 3: Okay. Somebody has the answer. Great. I think they're going to bring you a mic. Thank you.

Speaker 5: So if I can answer your question with a question. What's.

Speaker 3: That's not what I was hoping for.

Speaker 4: But I'm fine.

Speaker 5: What's the alternative? So where else should the investments go?

Speaker 3: Good question. But I think despite all the risks that. The US institutional fabric is facing. U.s. Treasurys, municipal bonds and corporate bonds from the Non-ai sector, for example, are much less risky than investing in highly AI or highly energy exposed parts of the stock market. So ultimately, of course, people wouldn't want to put all of their investments and nor would fund managers in the safest part of the market. But the way I would have expected is that some of that pressure would lead to a greater risk premium on the more exposed parts of the market, including the tech sector, including the sort of the at least six of the Magnificent Seven that are very, very heavily AI exposed and now make up 60% of Nasdaq. So that is the picture that I find puzzling.

Speaker 2: What about regionally? Right. So you talk about the confidence in institutions. And then I've got to think about Europe. Right. So I think in Europe it's fair to say people have confidence in their in their institutions. They might not always like them. They might not think they move quickly enough or responsive enough, but there's confidence in them. Despite that, European prosperity is not what US prosperity is. Growth is not what US growth is. Is that it's more socially balanced, obviously in Europe. But is that a question? Is that is that an alternative for the markets? Is that is it an area where we might see a pickup in long term growth or prosperity?

Speaker 3: Yeah, I think those are the right questions. Let me give my perspective on that, which is that there is no alternative to putting a significant fraction of an international investor's portfolio in the United States. But how much? So we're not talking about investors suddenly stopping all investments in U.S. treasuries, corporate bonds, and the stock market or the trading system, turning its back on the US dollar. But a portfolio adjustment. The part of the American secret sauce has been that U.S. assets appear and are safe, liquid, and resilient in the eyes of the international investors. Think of the global financial crisis. It was clearly caused by American banks and American policy making. And then what did international investors do? They doubled down and invested more in the United States because at a time of turbulence, US assets were the more reliable ones. Why is that? Well, because the US market has a lot of liquidity. Us companies have a lot of potential and and possibilities for further growth. But very importantly also all of that is bolstered by US institutions. So if you are a foreign investors in China investor in China, you're not just taking on the risk of Huawei or or BYD, you are taking on a significant risk from a complete change in government policy.

Speaker 3: And you're taking also a risk from Chinese legal system if you ever have any problems. All of these things were the secure part of the American offering and also great liquidity that because it's a very liquid market, if you want to get out, you will likely find somebody on the other side. All of these things would be questionable if the American economy. Starts tumbling and in an associated way, American institutions become weaker. So that's the that's the risk we're talking about at the global level and in terms of portfolio adjustment. Yeah. I mean, I think you wouldn't want to increase your exposure to China hugely, although there are many interesting things going on in China, it's still a very authoritarian country and it has its own risks. And Europe has been anemic in some ways, but it's not a disaster. Many sectors in the in Europe are continuing to grow. And if you compare their productivity to the United States, they're not doing that badly. Europe is doing badly in two very important respects. One is AI and tech more generally. The US economy over the last several years has been fueled by AI investments. European economy hasn't. If you take that AI premium, American growth would have been slower than European ones.

Speaker 3: Most probably. And second, the European Union over time has become more inefficient and more gridlocked because of many small countries playing the game of holding out against directives or against regulations and so on. So that's created a policy quagmire in the European Union. Now you can have a very pessimistic view about Europe that this institutional problem will not be solved and Europe will never catch up in AI and tech, and those are the future. Or you can be more sort of optimistic that the problem is now very alive. And many people are talking about it, and they're trying to come up with ways of reforming both the actual letter and the practice of EU governance. And European democracies are more appear not perfect by any stretch. They have a lot of problems. There are some risks, but they appear a little more robust than the American ones at the moment. And they are also quite aware that innovation in tech has fallen so far behind. So there is a possibility of a turnaround in Europe as well. I wouldn't hold my breath that that's going to happen right away. There are some big challenges, but I wouldn't write off Europe as a contender in the global economy either.

Speaker 2: Thank you.

Speaker 3: Yes, please. I think they're going to bring the mic. Thank you.

Speaker 6: Thanks very much for your remarks. I found it interesting that you draw a parallel between Asian countries and European countries and and the United States. There's 2 or 3 key differences also. Right. So for the economic powerhouse of the United States, we have only 350 million people. And depending on how you count European countries, there are almost twice the population. And in the last 25 years, they've gone from a similar GDP to the US to maybe two thirds the GDP of the US. And that trend is continuing. I mean, that spread is growing and growing, and there is not at any time the threat of one American state seceding from the Union or attacking another neighboring state. Of Arkansas is not going to attack Texas, and California is not going to attack Nevada. But European countries and Asian countries attack each other and they kill each other. And they do that every decade or two. Right? And I'm not making light of it. And it's a serious problem, but I'm just saying that the source of stability that you were referring to about the strength of the economy and the lack of a viable alternative for investment assets, is because we don't have those problems. We don't have internal strife the way Europe and Asia. And I think maybe Africa and South America do. And we don't have the, what you call anemic sort of economic growth. I mean, we don't beat the rest of the world by a little bit. We beat the rest of the world like by a lot. And I don't think anybody will catch up to us anytime soon. I mean, I think that is just the reality of our economy. So, um, you know, I very much would love a global pluralistic society and I think we'll get there, but I think the United States will be at the head of it for a long time.

Speaker 3: I do not disagree that the US has had many advantages. The European economy is still very far from an integrated market. That's actually part of the problem in tech, actually. Uh, and certainly China has much more complicated institutional weaknesses, and some of them are very severe. That being said, I think the issue is not whether China, say, will reach the US GDP per capita or whether France will reach the US GDP per capita in the next five years. It's again a question of where the global risks are. So if the AI sector, for example, tumbles, that's going to have a huge effect on the American economy and not so much on Europe. So that's a risk that we bear when we invest in the United States, and not as much when we invest in Europe. Why has the American economy outperformed Europe? I've mentioned AI, but there's another very important factor to bear in mind. You see that divergence between Europe and the US really most clearly not in the 1990s, not even in the 2000. You see that after the financial crisis and after Covid, both of these were blips for US GDP and had much bigger and long lasting effects on Europe. Why is that? Again, I would say it's about the institutional strengths of the US. Us borrowed a lot and it could because corporates could borrow in the United States for the reasons that I tried to articulate.

Speaker 3: And the government could borrow very for very cheap and they increased spending. And that really helped us alleviate some of the burdens. Europe didn't and couldn't do it. Some of it was because they had much higher risk premia, and some of it was because of institutional problems, both within the European Union and in Germany, for example. China has tremendous institutional problems, and I wouldn't bet on China. But right now, manufacturing in China is doing much better than manufacturing in the United States. If you look at how they are integrating technology, China is now far ahead. It wasn't three years ago, but, you know, investment in robots, investment in AI, in the production process and integration of consumer. Products together with the production process is advanced in. In China, far beyond what we do in the United States. In fact, you know, when we talk of us having the lead in AI, we're talking about LLMs. Us does not have the lead in AI. When you look at consumer products like how consumers use AI for, for, for searching or for getting products and getting recommendations for payment systems. And certainly us doesn't have the lead when it comes to integrating AI into the production process. So again, none of that implies that in five years China will have even come close to US GDP per capita. But I wouldn't take it for granted that US will be able to coast smoothly.

Speaker 6: I fully agree with you. The challenge may be for an investor is how do I harness these blips of of brilliance and opportunity around the world while mitigating the risks.

Speaker 3: I have no idea on that. And that's why I was getting hoping to get some recommendations here.

Speaker 6: Thanks for your thoughts.

Speaker 7: Thank you. Um, if, uh, liberal democracies are at risk, to which extent do you think electoral institutions would be responsible for that in this context of social media and new technologies? And if that's the case, isn't that like growing pains and.

Speaker 3: Growing.

Speaker 7: Pains?

Speaker 3: Pains? Okay.

Speaker 7: There is not so much we can do about it. And the process to get there would be like up and down and like full of, uh, just steps ahead, one step backwards. What's your take on that? Electoral institutions?

Speaker 3: Well, I can talk about that all night. And you don't want me to. Because, as Lee said, some of you are interested in gin and tonic or other things, but self-advertisement. In two months, my new book, What Happened to Liberal Democracy is coming out. I would recommend that as well. Very, very objectively. I think it's a great book, and some of the answers may be there. But briefly, I do not think that we can blame liberal democracies crisis on social media. But social media is making things worse. And we also cannot blame. Liberal democracies, problems in the United States, in our particular voting rules, because they've been around for 250 years. But again, they're not helping. So, for example, if the United States had proportional representation, I think some of the current problems would be alleviated, not solved. Are we likely to go to proportional representation in the United States? Not any time soon. But things like fusion, voting and other things. Multiple lists are now being practiced in a number of local elections and state elections. Social media. Absolutely. I think part of it may be transitional, although I wouldn't bank on that because with artificial intelligence, the nature of social media and some of the fault lines that social media was exploiting could actually get more severe.

Speaker 3: But also, we're not helpless there. So I think and this is broader than just political effects, but many of the pernicious consequences of social media from the mental health effects on Teenagers, the impact on reduced physical on reduced real world social interactions to echo chambers and all of that have gotten much, much worse after algorithmic feeds. So if, for example, we ban algorithmic feeds so that we go back to the early days of social media, where you see posts from your selected news sources and your friends in, in a, in a chronological order, at one fell swoop, some not all of the negative effects of social media would be would be reversed. Now, are we likely to get there any time soon? No, but I don't think we are completely helpless in any of these things. And I think that's one of the perspectives that I've always tried to emphasize, which is that if our problems have an institutional nature, that also means that there are things that we can do because institutions are not easy to change, but at the end of the day, they are our collective choices and we need to take responsibility for what institutions do, however difficult it may be to influence their current trajectory. Please.

Speaker 8: Oh, sorry. Sorry. Somebody else is here. So, uh, great remarks. And I think one thing really stick to me is it seems the weakening of institutions is, uh, tends to persist, as you said. And in that case, almost the failure is inevitable, right? So bad behavior reinforce itself and continues. And, uh, nations just fail. I certainly hope that's not the case here, but I want to hear your thoughts on. Okay. Um, is there any way kind of, kind of get away from that trend and any historical instances that the nations will be able to reset and stop that, you know, decline or what will be the typical trigger to make that happen.

Speaker 3: Thank you. Thank you very much. That's a very important question. And exactly like you said, my work emphasizes some self-reinforcing dynamics. You can get into vicious circles where things are bad, and then that creates a tendency for things to get even worse, and likewise with virtuous circles. But those are tendencies, not laws. So there is no law that says once things start sliding, you can never stop them. The fastest growing country in the world for much of the last 40 years was Botswana, which started its independence with two college graduates and a few miles of railway track and paved roads. So it was a huge institutional overhaul that laid the foundation of Botswana's subsequent economic growth. There was no guarantee that that would be taken in the United States, in the US history. We see the same thing. Things look very dire at the beginning of the 20th century, at the end of the Gilded Age, when not only inequality reached very high levels, a number of corporations became very, very powerful without any antitrust laws. And the government had not many of the tools that we take for granted today, not antitrust, not federal income tax rate, none of the regulations or not much of the safety net that we have today.

Speaker 3: And worse, the politics was generally accepted to be hugely corrupt, with senators not being directly elected and many of them in the pockets of the powerful trusts. But it was in that context that the progressive movement gained steam and completely transformed the face of US politics. Us legal system, as far as monopolies and cartels were were concerned, and the original, the and the, uh, and the roots of the regulatory state, which, you know, has had some positives and some negatives were laid during that period as well. Now, does that also guarantee that today we're going to have our own progressive moment? No, it doesn't guarantee it. But I would definitely not despair and say we are out of options and the only way is down. But I would definitely take the possibility that the way down is a possibility. Seriously. And you know, what can we do? We have to take action as citizens, as civil society, as corporations and as politicians, and hopefully as some ethical politicians who will stand up for bolstering US institutions, not just for their own power and, uh, and, and influence. Oh. Yes.

Speaker 1: So, um, I believe you know, Kevin Walsh.

Speaker 3: Yes, I do.

Speaker 1: Um, what's your confidence that he will push back when required and do the right thing?

Speaker 3: Are we on record or off record here?

Speaker 2: You got eight seconds to go in the off the record.

Speaker 3: Okay. Four. Three. Two. One. Well, my guess is that Kevin will do whatever is good for Kevin or whatever he thinks is good for Kevin, and the chances are that he will not want to go down in history as Trump's puppet. But I wouldn't necessarily say that that's a done deal either. So there is some uncertainty there. But he is very he's quite capable and he's very aware of all sides of the bargain here, both what Trump is expecting of him and what he should really be doing. I think he's quite aware of that. I am 100% sure of that. The question is what? He will judge his room for manoeuvre and what he will judge what is in his best interest, and hopefully he'll make the right choices.

Speaker 2: All right. I think that's a good way place to end it with a big question mark. So thank you very much.

Speaker 4: My pleasure.

Speaker 7: Okay. Welcome back. Um, so I do have some bad news, I'm afraid. Um, and that is that you're going to have to put up with me as MC for the remainder of the day. Um, I decided that Kayla was too good, and she was making me look bad, so I kicked her off the stage. I mean, I'm quite capable of making myself look bad all on my own. I don't need Kayla's help for that. Um, but there is good news. Um, we are now in the final stretch. Cocktails are not far away. Um, I'm going to need one by then. And quite honestly, I need one right now. Um, but even better news. We have some great sessions to finish off the day. We're going to talk about AI and human capabilities. We're going to pitch you three trade ideas in our Shark Tank. And then we're going to end with our keynote speaker, Professor Daron Acemoglu, who's going to take part in extended fireside chat with Sara Saleh, who's head of investor services in North America for State Street. There's going to be lots of times for audience questions in that as well. So get your thinking caps on. And then cocktails. Um, but anyway, priorities. So let's get on with it.

Speaker 7: Now before I introduce our next speaker, I have to, I realize with questions after my speech this morning, I actually didn't square the circle. I said to you, I would explain why it made sense for investors to be overweight equities. And I never actually did that. So my point was that we're in this sweet spot for equities, insomuch as we have a higher inflation world, but a fed that's not really inclined to do anything about it and certainly aren't going to hike rates aggressively to slow growth. But at the same time, fed has built up credibility so that long end rates and inflation expectations have not become de anchored. So in that world, you don't want cash because inflation is high. So real rates are much lower. You don't want duration because the risk of higher inflation will undermine duration. And so what are you left with. You're left with equities. So the point is from a macro environment we are now in this sweet spot for equities of higher inflation benign fed fed but one that has extended credibility. So you have a macro pro equity backdrop. And then you have a narrative in equities. And that's AI. Which brings me very neatly to our first speaker.

Speaker 7: So, Professor Robert Roberta, Roberto Rigobon. I'd never met you before. Is the Society of Sloan Fellows Professor of Applied Economics at the MIT Sloan School of Management. He is also a research associate associate with the NBER and a visiting professor at Iiasa in Venezuela. Roberto joined Sloan in 1997 and has been named teacher of the year five times. He has also received three excellent Excellence in Teaching awards from the school. He was, of course, one of the two founders of the Billion Prices Project with Alberto, who spoke earlier. Um, would you better know now as price State Street price starts. Roberto's career has been about measurement, about economic, social and ESG measurement. He's also co-founder, founder and director of the Aggregate Confusion Project, which studies how to improve ESG measures. Recently, his work has focused on how human capabilities can complement AI, and that's the subject he's going to speak about today. Now, before he starts, I will ask you to be kind to him because Roberto is a huge football fan, but he's also Venezuelan, and Venezuela is the only South American country never to qualify for the World Cup. This year. They've expanded it to 48 teams and they still didn't qualify. Roberto. Over to you.

Speaker 3: Yes.

Speaker 8: Yeah. Actually, every every country makes it to the World Cup except Venezuela. Yes. Very proud of that. Yes. Uh, so, um, it's great to be back in State Street. Um, let me tell you a little bit about, uh, about this research. So in 2017, ai was did not have the the amount of panic that we see today, but still the questions about how computers were going to substitute humans was a very valid question at the time, was about business analytics and, and robotics. Um, and I remember at the time I was thinking about, um, that, that is a terrible question, like asking what the machine is going to do that, that, that we do. So what aspect of our life is the machine going to replace is the incorrect question. Um, I think the correct question is well given what the machine is going to do, what are we going to be able to do. So how we are going to complement or what do we have to do? How do we prepare how to complement? I mean, most of the people here, I will assume that if you're here, you know how to add and subtract. Is that a good assumption? Yeah. Now, if I give you 20 numbers of four digits and I ask you to add, uh, you probably will use a phone, isn't it? Yeah. So aren't you afraid that you are going to lose your job because the phone is substituting your capabilities of addition and subtraction? What about long division? I remember one student asked me once, what is your most important academic accomplishment? I said, long division man, that's a tough one.

Speaker 8: So and now the computer has actually substituted my most important attribute. So I just think it's just the incorrect question truly. So the question is what we're going to do. So therefore we need to ask what this technology cannot perform. Not now not ever. So we can prepare for that. And so what I mean not now, not ever means that it doesn't matter how many more complicated you make the neural networks, how much bigger, faster, you put more energy. There are things that they cannot do. They cannot perform that we perform differently. Now, when you actually look at the brain And and the neural networks, which is kind of the base of, of all artificial intelligence from the architectural point of view were identical, which is kind of interesting. So, um, we have a bunch of sensors, our eyes, our ears, our, you know, our smells, you know, that bunch of sensors come into a single node and those nodes are activated and they are signals one zero. They come to a node. That node has a certain, you know, function that it has been estimated, it becomes zero and one. And that goes to another neuron and another another. And so that's how it is. And and when you think about a neural network, it's actually organized exactly in the same way.

Speaker 8: So, you know, when you think about it that way, the two have the same machinery. So it should be the case if we have the same machinery. It's not surprising that people think that they can do exactly the same of what we do. But what is interesting is that when I started asking the question to people that study the brain neuroscientists, psychologists, sociologists, anthropologists. It was very interesting because they had a completely different answer. It was not about the architecture. It was not about the machinery. It's about how we use that machine. And I mean, I'm not going to summarize nine years of being studying neuroscience, moral philosophy and all of that in an hour. But but let me tell you a little bit of the things that I learned. Okay. So first and foremost, I think that the most important difference that almost every single discipline that studied the brains will agree is that the brain does not optimize the brain controls. However, the machine tends to optimize. That's a very important difference. So we we kind of actually, when we're doing what is called robust control. So we set a set of boundaries of acceptable outcomes. And within those boundaries, we actually make decisions like someone, for example, that works in, I don't know, compliance in a bank. They have a set of acceptable outcomes. And then within those outcomes, where day one is that the performance moves within those ranges.

Speaker 8: And we kind of think about it that way. So we have boundaries. These boundaries are very important because somehow we develop relationships with entities and with individuals based on those boundaries. And that implies that every transaction or every information that goes through that channel has a context. So for example, I have three kids, and even though I have tried many times to look cool in front of my kids, they, they, they don't find me cool at all. It's okay. And I said, you know, but I play for ten years in a rock band, which is actually true. I have evidence of that. And they said, yeah, that you used to be cool, but you are not. Now you, you are such a good rocker that you teach economics anyway. So? So there's no way I can look cool because I am dead. I am dead for them. And by the way, they are my kids. And in my relationship there are certain boundaries. And therefore when I say something, they know it comes from that. When they talk to their friends, they know that it comes from a friend. When you when they talk to their husbands or wives, it comes from a husband and a wife. And this matters a lot because those boundaries imply that there's a certain content on the information that is attached to the context. So so that leads to the third characteristic is that we do not absolutely, under no circumstance we ever let the data speak for itself.

Speaker 8: Never under absolutely no circumstances, not a single brain, not a single father or mother for that matter, will ever say, I will let the data speak for itself. You will never tell your kids, you know what? Go to the internet. Read whatever you want, and you just make your own choices. Is that okay? You can use that as a truthful reflection of humankind. And no one will ever say that. And why? Because we actually, when we look at the data, we use our values. And again, those boundaries. And we know that, for example, when our kids are talking maybe to their mother or their father or their aunt or, you know, any family member, you know that the information that goes through that channel is different from the information that will come from the internet. And therefore, the advice will be different in these two instances. Well, it depends on the uncle. Almost all my uncles were drunk, so, you know, I also had I also had warnings. They all they all came with warnings. But yeah. So it is important to understand that we act differently. Okay. It's interesting that we act differently. And these features about how our brain works, it just makes us different. So notice that I'm not saying better. Is that clear? I'm just saying I'm using the word different. So if we're going to complement the machine we're going to complement them on those characteristics.

Speaker 8: That makes us different. And there might be some problems in which our capabilities are well suited for the problem. And there are others in which we are not. And therefore that's kind of what is going to decide how our jobs are going to look like. So we have organized all of these in, in five relatively big, uh, set of activities that we do in a second and we'll explain a little bit better. But I just want to tell you where we end. There are activities that, that we do, we organize them this in epoch. So it reads epoch. There's no particular reason for the choices. There's a there's a statistical decision that goes through this, but it's kind of minor. So the E stands for empathy, emotional intelligence, Compassion. Things of belonging. Uh, the P stands for presence. Networking. Trust the fact that you develop human contact and therefore human relationships. They all stands for openness. Opinions, judgments, ethics, all the moral philosophy aspects. This instance for creativity, imagination, curiosity, intuition and the H stands for leadership, hope and vision and so on. Now all of these guys actually are solving different statistical problems. So this is how a statistician will look at the limitations of, uh, of the fundamental, um, estimation procedure that is inherent in AI. They are really bad in inferencing, small data, for example. Uh, they are very chaotic on extrapolation. Uh, they tend to be really bad when they have what is called a moral dilemma.

Speaker 8: So, um, a moral dilemma implies that there are two answers that are both valid to the exact same problem. Otherwise, it's not a dilemma. Is that clear? So otherwise there's an optimization here. There is no optimization. That's the whole point. There's a lot of trade and transactions that we do in our life that is about relational outcomes. For example, I actually thought that I had a relationship with Lee because I have known him for nine years. But now I realize that he doesn't know my name. So maybe I should change the boundaries of that relationship. Is that a no? Like, you know, so and then after not knowing my name, he makes fun of my country. I mean, I mean, it's I mean, clearly it's not enough that they kidnapped my president. He said, okay, like, you know, so, you know, like, let's, let's actually crush it's okay. You will have a chance. Yeah, yeah. Well he's British. What are you gonna expect? Yeah. Anyway, so no, by the way, when I told you that some humans have empathy, we know that not all of them. Is that okay? So. And then the subjective leaves is, is something interesting that that the way the neuroscientists explain this to me was absolutely fascinating because once I tell you, you will realize that it's trivial, but I never thought about it. I actually never thought about it. They said, you know, the brain, which is using the data that has been collected, is the one that decides what data to collect, because the brain decides if my eyes are open, if I am listening, if I'm actually texting or not, if I'm actually paying attention, if I'm sleeping or not.

Speaker 8: So if you are in a in a meeting and you're bored, your brain will disconnect. You will stop collecting data. So we and in fact, if you are talking to someone that has terrible values that you do not respect, for example, you disconnect. So it's interesting because it's not that we are waiting the data differently. Is that actually where some data we just disregard completely? Is that clear? So we just don't even bother to collect. So. The epoch actually, by the way, we thought that we were going to be able to find one activity for each of the statistical problems, but that's not the case. So these words that are here actually what happened is that it's not like we take actions that are particularly designed to solve one of the statistical issues. That's not the case. We solve many. Okay. So every one of these actions have different weights. Um, my, my younger daughter is a psychologist and she has helped us a lot to do this. So I just want to give you an idea about how we think about it when we think about the word of hope. I actually remember asking Veronica, I said, like, you know, what would be? I mean, she's a very good statistician.

Speaker 8: She, she was studying psychology and she was a TA in statistics. So I said, um, how a statistician and a psychologist will define hope. And she said, well, hope is for someone to believe that something might occur when actually the probability is zero. And I said, that's actually defining an entrepreneur, isn't it? It's beautiful. I said, wow, what a beautiful definition. And just to show you these things, it is not the case that we use them universally for goodness. Okay. We make mistakes because hope is what a leadership will tell you that we can do. It is what an entrepreneur thinks they can do. But also hope is the reason why we actually play the lottery. So we use it for goodness and for mistakes. Hope is again, there are 48 teams in the World Cup. You understand that the computer will never do that. A computer will say, that's stupid. What a waste of resources. It's like they are going to have what you are going to have all these people coming to the United States, do we have to issue visas for all of them? They're going to invade my country. I said okay. Like we should stop them from coming. By the way, they have no chance of winning. Why play the game? Why do the emotional investment to lose? Okay, then you pay for the tickets that are super expensive, and then you go to the stadium and then you see them lose and you see them losing and everybody else making fun of you.

Speaker 8: Like, why do you want to go through that process? That's why Venezuela doesn't ever qualify. We are very rational. Our AI has decided, let's not make it. Let's not suffer all together. So but surprisingly, we don't care about efficiency. We don't care about optimization. We care about the experience. And the reason is because we have a relationship with our team, with our nation. It's kind of inconsequential. If I'm going to play Brazil. I know I'm going to lose, but I still want to try. Maybe there's a miracle happening and I don't want to miss the miracle. So truly. This is what actually kind of makes humans very human. Almost every one of these. I have a very long list in the paper that talks about this. So I just highlight some of the words. What we did was to go through each of these kind of groupings to try to see what are the type of activities that we do that are solving this statistical problem. So, um, we operate really well in a small data and that comes from that relationship. So there's a book on anthropology that I find absolutely fascinating, where they talk about the superpower of humans and the superpower of humans is that we transfer knowledge without requiring experimentation. That's kind of the statistical view. So how is that happening? Well, you know, 20,000 years ago, when the end of the day comes, we see it around the fire and the elderly tell us stories.

Speaker 8: At that time, the life expectancy was 20 years old. So if you were 40, which that's elderly, by the way, if you were 40, you have something to teach me. The fact that you are alive means that you might know something. Therefore, I might want to listen. By the way, why on earth the elderly will tell me is because I'm going to show respect, treat them well with dignity. And almost everyone here, when you were young, you were told that you need to respect your elders, your elders. Isn't it? That's a story that goes back 20,000 years. And the important thing is trust. When someone that knows more than me and I know that they know more than me, I do not need all the data. I need the sentence. So if you think that the sentence is a data point, just one sentence, one advice, one question at the right moment, I will actually make a decision. I need one data point and only one data point to make a decision. Because with that, all that context that we have built in 2000 years with very simple rules. By the way, one of the things that I, that I experienced when I was reading that that book is I felt kind of very sad because nowadays one thing that has happened is because of our life expectancy has become much, much bigger.

Speaker 8: The distance between the elderly and the young has become bigger, much, much bigger. And they communicate in different mediums. So that conversation, that advice from the elderly to the young is not taking place in the homes of most of most of the young people today in the world. So we're kind of disconnected. They don't live in the same city. So it's an exception. So I was thinking like, how can we actually maintain that superpower? Um, by being able to enhance that connection. So we are losing some of that. By the way, we're also losing the capacity of tolerance, which is to learn from a sentence of someone that disagrees with me. So we just cannot listen. And we have lost the ability to listen to others. So, um, so these are things that I find quite interesting because when you think about from the informational point of view, I should be paying attention to everybody, isn't it? I can disregard, but why not listen? So anyway, so anyway, aside from my depressions, my team of psychologists will take care of that. Let me just tell you more or less what we do now once, once we have all these words. And by the way, I in the paper, I can show you it's about 60 words. Is that okay? 60 things that we do. Okay, so it's more than. Than what I'm just showing you, but, um, um, what we are doing is kind of evaluating task by task in the United States labor data.

Speaker 8: Uh, what is the amount of epoch that this has? In other words, how much human skills are embedded in the task? So for example, this is a person that is a physical therapist and a physical therapist. This is, you know, one of the tasks that is not organized in the order of importance or anything like that. Okay. Just to show you some tasks, one of them says, monitor and record client's progress to ensure that goals and objectives are met. And so when we have that sentence, what we do is we ask, well, how close is that to a definition of empathy? So we don't ask empathy directly. We ask. We have about 20 definitions of empathy where kind of different sociologists, anthropologists will disagree. They are not exactly the same. So we ask, well, how close is this? Or how much empathy, given those definitions that sentence implies. And we do that with every one of the words that we have in our list and so on. So that gives me a score for the E, the first oh, sorry, the first. The first box will be the score on empathy. The second box is the about trust and and presence and networking and relationships and so on. That's how that's how we go along. The tasks can have a lot of presence. For example, a dentist might actually need you to be there.

Speaker 8: Is that okay? So that but but that doesn't that doesn't have any empathy. Nothing else. Is that okay? Just the presence. So what is interesting is that what we do, task by task is defined well, how substitutable this task is, how how can, how much can this task be performed by either a robot or an AI? This is everything that is for robotics or AI and. Um, and we make an assessment about task by task, then we aggregate that occupation by occupation. This is the data that we have 1000 occupations in the United States. This is 2014. I'm going to show you why in a second why I put 2014. Uh, but, you know, they have some professions that have a lot of epoch like management has a lot of finance. You have some individuals that are working in the finance industry that have relatively low epoch. You can see that it's like a little mode on the on the bottom quintile and so on. So law is very interesting. They have two very big, uh, modes, the paralegals and the other ones. Okay. So they're very, very big modes. This is the one interesting thing is that epoch is not perfectly correlated with education. In fact, the day you get a PhD, you clearly lose empathy. Is that clear? So and so. So it's not they're not necessarily associated with that. Uh, look, for example, nannies have a very high epoch.

Speaker 8: They are close to environmental economists, and my co-author, Isabella Loayza, has the decency not to show where macroeconomics is. Ah, but macroeconomics are between tax preparers. These are TurboTax people. Okay. They are between TurboTax people and the guys that read meters. Okay. So macroeconomics is here. Not only that is if you organize this by hope. Just. Just hope. Okay. One of the 60. If you organize this by hope, macroeconomics have the lowest score on hope of every single occupation in the United States. Is that okay? A mortician has more hope than I do. Is that okay? So we are the lowest? No. In fact, we have about 20 tasks that Macroeconomists do, and not a single one has hope. And the reason why 0.2 is because 0.02 is because the number cannot be zero. Is that clear? So but it's just the error in the machine. So why this matters. We did that in 2014 to try to see if we can predict the changes in the labor force that are taking place for that. And in fact, between 2015 and 2020 for the United States created 8.5 million jobs. 7.5 of those million jobs were created in occupations that are on the top 20 percentile of the epoch. Okay. So they actually were created on activities that have a lot of human capital. And again, these are not related to education. There are many activities in which this is not related necessarily to education.

Speaker 8: And also there are different ways of forming education. I don't know why people tell me always that the nurses is a is an exception to education. They spend ages educating themselves. I don't understand that. But people think that nurses have lower education. Nurses is one profession that where the education is highly correlated with the epoch. Is that okay? It's actually they need to have, if you think about it, empathy, leadership, to tell a person that is suffering that the pain is, you know, somehow worth it. I mean, can you imagine all the things, creativity. I mean, they actually have a lot, a lot, a lot of epoch and also a lot of education. But for some reason people think that they are not educated. I understand that they might be less educated than a surgeon. I understand that, but but not totally clear. Okay. So anyway, so this is not perfectly correlated with the level of education. In fact, after you control the level of education, we still are predicting, uh, the jobs that were created. Now, we did not stop there. We actually look at the hirings that happened in the first quarter of this year, and we look at the job postings. This will be a combination of what? Um, so this is actually, for example, linking the CVS, for example, and, and also job postings from the firms. And, and we look at what are these job postings and firms highlighting.

Speaker 8: What are they searching for and what is the difference between that same occupation ten years ago? In other words, we look at when you are trying to hire a programmer, are you asking for the same things? Let me tell you, what is so interesting is that not a single one of the programmers postings that we saw. As for Python. Not a single one. Not a single one. Everyone was asking ability to to talk to the clients, to develop relationships, to work in teams, etc. they were actually highlighting 100% epoch, 100%. There were technical activities. Okay, so it's not like this is not technical, but when you look at the difference, what changed from one place to the other is they asked for less. So we actually explain a humongous amount of, of the hiring statements that you see in the job postings and on the things that people were highlighting on their LinkedIn profiles. And then we are looking at the projections of the BLS. And here the beauty is that we were expecting that we were not going to be able to predict the projections of the of the BLS, the Bureau of Labor Statistics. Every year they do a projection of which occupations are going to grow and which occupations are going to go down, and they are pretty good, I'm telling you, extremely, extremely good. So but we said, well, it has to be the case that the that our predictor cannot predict what the BLS does.

Speaker 8: Is that clear? And why? Because they don't have the data. Is that okay? They they they they have education. And in fact, once you control for education, we predict absolutely zero of the projections of the BLS. Now we're going to give them the data. We will see what happens afterwards. But. But if you go back in time, the mistakes that the BLS made in 2014 projection are all explained by. Well, not all 80% of them are explained by epoch. So, um, let me just, uh, let me just finish here. So I give at least ten minutes for Q&A. Let me just give you what are kind of the lessons we are going to be fine. That's the first we're going to be totally fine. Uh, there are many things that we do that makes us different from the computers. What we need to do is to make sure that our jobs, uh, are highlighting those aspects. And second, and very importantly, that we teach people and make sure that people learn those aspects like moral compass. And some people tell me, oh, these are soft skills. And I say, I mean, there's nothing soft about it. Like, how do I teach honesty? Anyone here knows? How do you teach? Uh, creativity. Curiosity is not in our education system. In fact, actually, I remember my son. My son was asking why every single instant he couldn't stop asking why until he went to school, and a month later he stopped asking why.

Speaker 8: You know why? Because I'm pretty sure the teacher told her. They said, you know what? One plus one is two because I say so. I remember my son asking me, dad is one plus one equal to two. And I said, not always, by the way, my wife wanted to kill me, but that's a minor point. I mean, he was like six years old. Five years old. Okay, so I don't remember exactly his age, but he said why one plus one is one plus one equal to two. I said, no, not necessarily. And he said, for example, I said, well, one pile of dirty clothes plus one pile of dirty clothes is one pile of dirty clothes. And he said, but one kilogram of dirty clothes plus one kilogram of dirty clothes is two kilograms of dirty clothes. So the measurement matters and I am obsessed with measurements, as Lee said. I said, so this depends on how you define your units. And my wife told me to shut up. Well, my son is teaching now in Princeton statistics, so he didn't do that badly with my influence. But but that curiosity is crushed by our by our education system. So if I give you 70 million workers that need to learn humanity, I have no idea how we're going to do that. So but these are the good news if we implement correctly. And with this I finish. So we took every one of these jobs and characterized two two dimensions.

Speaker 8: One is risk and the other one is augmentation. What is risk? Risk. Imply that I have a job that is a collection of tasks. And this collection of tasks can be dismantled. What does that imply is that imagine I have 15 tasks, ten can be substituted by AI and the other 15 can be reallocated among other workers in the society. So it can be it can be dismantled. These are the jobs that are here. These jobs gained very little with the technology and they can be dismantled. Now, the size of the of the circle is how many people are in that occupation. There are 1000 circles here. And and when you add them, this is something between 4 and 7 million. And this depends on the cutoff. Is that okay. Which is arbitrary. But there are 4 to 7 million jobs that are at risk for seven, four and seven, not 35. Okay. 4 to 7. The ones that are here, the augmentation is that they are substituting a task that is not particularly important for my activity. For example, as a faculty at MIT, I have to do three things. One is to prepare the content. The other one is prepared the jokes. And the third one is to prepare the PowerPoint. So actually in practice, um, I for a class of an hour and a half, it takes me about ten minutes to prepare the content.

Speaker 8: Uh, about 3 or 4 hours for the jokes. Okay. And about 16 hours for the presentation. I am completely incompetent in PowerPoint. When I send a presentation to here to State Street, they said, how can you be so incompetent in PowerPoint? I mean, like, this is the stuff that was trying to be nice. Is that okay? Like Lee, I was trying to say, be nice to me. So so they say like, you're terrible at PowerPoint. I said, yeah, I know, that's why I'm sending you this well in advance. Can you just fix it? Because I am completely and absolutely a PowerPoint idiot. Okay. Um, if you take the PowerPoint out of my daily job. My jokes will improve. So this is the percentage of occupations that are going to be replaced that are not the core activity. This is the reporting of the surgeon. This is the reporting of the nurse. So actually when you count all these little circles that are above 30%, that's my cutoff. But again, it doesn't matter. That's about 100 to 120 million jobs. So there's a fantastic opportunity here. So let's make sure that we don't waste it. By the way, if you're asking what is this guy here? Is that okay. That has no risk and no augmentation. This is the people that is universally hated in all organizations. That's the CFO. Okay, good. Let me stop there and see if there's any questions. Thank you so much. Yes. Question here.

Speaker 9: Hi. Great presentation. Thank you. Um, I don't know if this question fits your your framework here, but what about we notice I think empirically we notice that there is a lot of people who are very successful at high levels of government and companies who would actually score very negative on many of these epoch measures. Yes. They are not personable. They are jerks. They are aggressive. Yeah. They put people down. Um.

Speaker 8: They or their orange. Small hands. Small hands. Yes yes yes.

Speaker 9: I had nobody specific.

Speaker 8: Yeah. No no no. Nobody. Me. Me neither. Me neither. Yeah, yeah. Um.

Speaker 9: But not just government, really. You see that they rise. They do really well. So I don't know if most of your presentation seems to suggest people because you talked about values and everything, people who are positive on these measures and epoch, they tend to do well and they're successful. So I don't know whether that fits in. How would you explain some of this? No. This is outside the purview of what you presented.

Speaker 8: So no no no no no. So so one thing that I cannot do. So so a couple of things, if you notice what I said is that the value that you have for the labor and the, and for the society as a whole will depend on these characteristics. But I can only measure these characteristics on the jobs, not on the individuals. Some of the people that you are describing are incredibly creative, have incredible visions. Okay. So it really doesn't matter what your opinion is about Elon Musk. It's undeniable that the person has a tremendous drive that makes people work toward that vision with an incredible ability. And when the guy says that he's going to put data centers on SpaceX, I want to know who here really doubts that that will be possible. That's it. See that leadership is on the H. The guy. How much empathy does he has? I mean, I don't have to measure it. Is that okay? Like how much compassion he has? Like zero zero. Absolutely. But I don't measure this at the individual level. But think about all the individuals that you have there.
Speaker 8: They are extremely good communicators. For example, some of them, they have very good visions. They are incredibly creative, incredibly innovative. So and they are incredibly crazy, incredibly crazy. So if you ask them if I if Elon Musk tells you ten years ago that he wants to put data centers on space, you will say hopeless, but the guy what he has is hope. So, so truly so again, but I don't measure this at the individual level. I would love to be able to learn. But you imagine this is a small violation of privacy. But but I do think that the people, when I actually look at these individuals, what they have is a lot of 1 or 2. But that's the key. The key is that you need some of these characteristics of human. A person that is in the middle of everything will be a good person, okay? And it will be a great teacher. Yes, yes. So yes, yes. Okay. Well, this is the role of the questions. This is a splash zone. Is that okay? Good.

Speaker 10: So one question I had and I'm all for computers doing everything computers can do. I don't want to do anything that a computer can do. But there's the computers taking the jobs. And then there's also like other people in other locations. And this sort of comparative advantage, like moving work around the globe. Did you do any like is most of the data you showed us LinkedIn jobs and have you looked globally, and how do you think that sort of plays into some of these trends?

Speaker 8: It's a fantastic question. And, um, the reason why we look at the US is that the US has the absolute best statistical office. So we have 1000 occupations. There's about 20,000 tasks, and that data is updated every two months. And we have the data for 20 years. Okay. So so the data is just extraordinarily good. So it was easy to implement and it's very well, um, they have mistakes. Okay. For example, the dentist, the joke that I made about the dentist is that they forgot to say that the dentist need to be present. Okay. And we told them, don't you think that you need to add a task that the person, in order for you to destroy the person and punish and torture the person, it has to be present. It's really hard to torture a person when it's not present. Is that okay? So, um, so so it's not perfect. The data is that okay? But but it's the best in Europe. Just to give you an idea, the number of they have a different taxonomy of jobs and it reduces by almost one order of magnitude. So now the jobs are to aggregated in Europe. So it's hard. So we are using actually we're trying to implement this in different areas. For example, one is uh In a LinkedIn postings because that one we have access to, I don't know, Brazil alone will be 30 million and Europe will be hundreds of millions. And in the job postings, we might be able. We have to see how much we can learn. Okay. But that's the one that we're doing.

Speaker 8: And the other one is on on syllabus syllabus from universities. We have about 17 million syllabus. Five of them are from the United States. And we have this through time for different careers. And we're asking our mathematicians more epoch today than in the past. The answer is no. Okay. But our sociology is more epoch today than in the past. And the answer is yes. And we have. So I'm trying to do that. Not with MBAs. Okay. But I have to do the research in such a way that Harvard looks worse than MIT. Otherwise, I am not even writing the paper. That will be the. So if you don't see that paper, you know who won the race. This is not a race, but I want to win it. Okay, so but but there are many areas and the best area is also your human resources. See you and your organization. You have very detailed descriptions of what the people in your organization do. If I combine that with, for example, what they actually do, they are ways to collect this information from what the actual activities that the person is doing on the communications inside the company. We could actually compute the EPoC inside organizations, and then help organizations redesign the jobs to make it more consistent with what, with what the technology is going to do. Yeah. I have a fantastically large zero here. I don't know if I have time for more questions. Oh, one more question. Okay. It's perfect. Well, thank you very much. Thank you for being here.

Speaker 1: Thank you Alberto. So now that we know about inflation, we're going to switch to more of a prescriptive presentation on what to do about all of that inflation, or maybe not as much inflation. So last year when I introduced our macro outlook, we talked about the three T's and a U. So Trump tariffs and turbulence and the U was uncertainty. And fast forward a year and all of those things are still in the market ecosystem. They haven't really disappeared. Instead the market has just gotten more has more topics to navigate. So we have international conflict that seems to just suck all the oxygen out of the room. And now we're thinking about the implications of that conflict. So intensifying inflation risks, interest rate risks. We also have private credit concerns, crowded trades. What does AI mean for the economy. And so the next question to ask is how should we position for this? And so fortunately with us, we have Lee Farage who will tackle that very challenging problem. But he's going to help us really look through the static of all these topics and show us how we can use some of our proprietary indicators to create actionable trade ideas to navigate the medium term market. So it is my absolute pleasure to introduce Lee Farage, head of Multi-Asset Strategy for North America at State Street Markets, and also my boss.

Speaker 2: Thank you for that, Kayla. You can stay. So yeah. Thank you. I'm not sure I can recover all of that. I'll do my best, but not all of it. But I'm going to give you yeah macro outlook view on the world. Um it's called how will the party end. Um, I don't mean this party we're having here. This has just begun. And I know this end, it's going to end at 530 with drinks on the terrace outside, so that's not the question. The bit I'm talking about is the returns we're seeing. So this shows you um annual returns in the S&P and the Nasdaq. The point here is six of the last seven years the Nasdaq has been up 20%. And the S&P has been up by 15%. We've never seen that before. We've never seen returns on that scale clustered in that way. The closest we came was actually the late 1990s, where we had five years of very strong returns, five years of 20% Nasdaq and 15% S&P. And we all know how that ended with.com This year. Now we are six out of seven, and so far this year we have the Nasdaq up by 16% and the S&P up by 11% even in the middle of a war. You look here, you look at the returns so far since the war started. So since we had the biggest disruption to global oil markets we've ever seen, the Nasdaq is up by close to 20% in three months.

Speaker 2: The S&P is up around 10% in three months. Whilst there is a war going on and we're seeing this huge increase in oil prices, which, you know, we're sort of stealing a little bit from what Alberto talks about here. I'm comparing Brent futures, the active contract with what happened in Ukraine. And you can see a much bigger impact. In actual fact this morning it will have gone up again even further. This was as of Monday. We're up over 40% now since the war started. Obviously hopes of a deal. All this talk of a deal. Is there going to be a. I don't know, no one knows. Let's be honest here. Um, a lot of hope over the weekend seems to be dissipating. And don't forget, the deal we're talking about now is just an extension of the cease fire. It's not actually a final deal. It's a two month extension of the cease fire, the two week cease fire we signed about a month ago, where we still seem to be firing and it's not ceased. But anyway, don't worry about that. The fact is, this thing is still going on. And what we're learning is it takes two to taco. And this is the problem. We know Trump wants this to end, but the Iranians don't seem as willing to negotiate. What does that mean? Look, these are possible scenarios for the oil price.

Speaker 2: And I did this on the 12th of April. So you can look at it where inventories take most of the strain, which is the bottom 1 in 6 months time. If the street is still closed, we get to about 120. For Brent. We're about 98 this morning. If you get a partial offset with with reserves, we get to about 150. If there's very little offset, we get to nearly $300 a barrel in six months time. And Trump tweeted overnight that he thinks the strait will be open by Labor Day. Well, that's getting us close to some of these scenarios here. So look, that's the backdrop to this 20% rise in the Nasdaq over the last three months and this 10% rise in the S&P. And then when we look at our own sentiment measures, we get a similar positive picture from real money investors. So this slightly messy looking chart here this is our behavioral risk scorecard. This is our broadest measure of real money sentiment not from surveys or anything else. This is actually from what they're buying and what they're selling. Risky assets versus safe havens. Where are we? We've got positive sentiment. It's come down a bit over the last week or two, but generally, apart from the very first week of the conflict, sentiment has remained positive and remains so today. And then if we look at the other elements of this.

Speaker 2: So the behavioral risk scorecard measures sentiment. And then we also within that we look at positioning in risky assets. Are they overweight or underweight risky assets. And that's the light blue line here. And I've just done a three month average to clean it up a bit. What you see is positioning in risky assets right now is actually the most overweight we've seen from real money investors since 2018. Three months into this war the oil price up 40%. You know, the Straits still closed and looking like they're going to be closed for a while. Real money investors are positioned overweight risk the most they've been in eight years. And a lot of that comes from their positioning in equities. So this chart on the left here this is something called our asset class weight series. So this is the 50 trillion we custody. How much of that is held in stocks? How much in bonds and how much in cash? The horizontal lines are just the long run averages. Real money investors have been overweight equities for over four years. We saw a slight reduction in that at the start of the conflict. Where are we now? Where are we in the latest data today? We're a couple of days ago. We've got the biggest equity overweight we've seen since 2007. That's where we are now. If you look at the chart on the right, it shows you the equity positioning versus that in fixed income there, we're over 30% equities over fixed income.

Speaker 2: The average there, the long run average is 20%. You think about a 6040 portfolio. We're over 30% equities over fixed income. And it's increased since the war started. So here I've got the change in each of those asset classes since the start of the year and then since February 27th. So the allocation to equities has gone up by around two percentage points since the start of the year, from 56 to 58 and close to one and a half percentage points since the conflict started again on the surface. Remarkable makes no sense. We've come out of cash and bonds have been reduced a little bit. Except it does make sense. It makes sense because of the risks that are out there and the reaction function we're seeing. And I'm going to try and explain to you why this actually does make sense and is not insane. So look where the risks lie. And this first bit here will um, sort of talk about what Alberto just mentioned, but I've got prettier charts than him. Um, so chart on the left. This shows you price stats daily. Um, all all I've done is reset the daily index back to 100 at the start of the year, and then just plotted it each year, each day through the year. So the light blue line, that's the median of the last ten years.

Speaker 2: The dark blue line is last year. And what you see, inflation was fairly well behaved. And then the tariffs came. And as Alberto talked about, they pushed inflation above the green line. The green line is this year. That's the impacts of fuel. So price stats is up close to 4% about 3.8% so far this year. Chart on the right shows you the impact of tariffs and fuel together. So what I've done on sectors. So what I've done is take the year on year rate for the sectors and the overall number on the first of Jan last year and then where we are now. So that captures the tariff impact. And it captures. In addition, the war impacts. So what you see, nearly every sector is higher apart from healthcare, which is purely domestic or nearly, almost, almost domestic. Every other sector is up there running around 5%. And then we've got the fuel and energy, the transportation and energy, which is up 14%. And as Alberto said, it's not just in the US here, which luckily it backs that nice chart that you had that sort of goes through time here. I've just got the change in the year on year rate since Feb 27th. The biggest increase has actually been in the US in terms of the year on year rate, followed by New Zealand, which were the two countries that Alberto highlighted. Australia, as you mentioned, is the only G7 country or G10, where the year on year price tax rate is actually lower today than it was at the start of the conflict.

Speaker 2: And that's because they have fuel subsidies. Japan, it's barely moved, but pretty much across the board we're seeing this increase. But this is all headline. Central banks care about core inflation. So this chart here looks at if oil goes from $70 to $110 and stays there for between 3 and 6 months, which is looking increasingly likely as we look at the tweets overnight. What does that mean for headline inflation. And then what does it mean for core. And this is using work from the fed, Killian and Xu, IMF and others and basically bringing them together using AI because I'm calling trendy. So using AI to bring them together. And you get these projections for headline and core. A couple of things to note. Biggest pass through to core comes in the UK and the eurozone. And so you had a question about the ECB. You do see more of a pass through into core more rigid labor markets more activity tied to headline inflation means it gets passed through into core US and Canada. Japan. Big increase in headline. Very little increase in decor. The weight of of um of fuel getting passed through into core is one of the things Alberto mentioned also the impact on growth in Japan. Japanese growth gets hurt the most.

Speaker 2: That's why it doesn't get through into core UK, US and Canada somewhere in the middle. Um very big headline reasonably well behaved core in the case of the US. But here in the same way as Alberto talked about the ECB and inflation was rising before we had the oil price shock. When it comes to core US inflation, the starting point matters. And the starting point is above target inflation. The starting point right now is PCE corps at 3.3%. The starting point of this war was the fact that PCE core inflation has been above the 2% target for 62 months. It's over five years we've been above target. So the fed adopted their 2% PCE core in July 96th. Implicitly. It was an explicit until 2012. But implicitly we know from fed meeting transcripts. They adopted it in July 96th. We've never seen a run of above target inflation like this since they adopted the target. The closest we got was 30 months back in the 2000, just before the financial crisis, 62 months. And as this chart shows you, which is core PCE -2%, where are we going? We're moving away. We're not just moving away. Since the war started, we were moving away gradually from 2% before the war started. So look, before the conflict, we were pricing more than two cuts by the fed this year. Why? Well, the argument was that policy is restrictive.

Speaker 2: That's getting really hard to make that argument. Now if you use a natural rate model. So the Holston low back Williams is the most famous argues that natural rates should be about 1%. Well, that would imply that policy now is accommodative because real rates now 3.753.30.4 in terms of the core rate. If you use the Richmond Fed natural rate model rates should be even higher. Even if you go back to a Taylor rule. The Taylor rule right now is arguing fed funds should be close to 5%, not 3.75. So the idea or the argument that that policy is restrictive at this level, where is the evidence for that. Where is the evidence in growth. Where whereas the evidence in activity. Where is the evidence in the labour market, the labour market people will say, well, look at payrolls. And a lot of the fed governors have argued, you know, when they were cutting last year, it was about the labor market, it was about payrolls, except they really got the labor market wrong for all the researchers and everyone else. They've got the fed, they've got the labor market wrong. Because if you look here, look, this is payrolls 12 month moving average. We're around 20,000. Um historically, whenever the 12 month moving average of payrolls has been around this level, we've been in recession. We were revising up growth forecasts for this year ahead of the conflict.

Speaker 2: And we'll start revising them up again now given the recent data. But we were up to sort of 2.7, 2.8%, um, not in recession going the other way at the same time. Look at the unemployment rate. The unemployment rate is 4.3%. We can go back over 50 years, and there's only been a couple of periods where it's been lower than this. Just before the.com and just before the pandemic. Normally, if payrolls drops, then the unemployment rate will rise. It's not rocket science. That's a fairly obvious thing, except when the bigger issue in the labor market is supply, not demand. Labor supply is something the fed have been really late to realize was a factor. Um, this here is data from the United Nations US workforce growth. You look at the 80s, 90s early 2001 to 1.5% a year, sort of fairly average workforce growth. Um, projections made in 2023 was workforce growth would be about a quarter percent a year over the next ten years. Purely demographics. Baby boomers are retiring the replacement rate. The birth rate has been dropping for years. The labor force will shrink. The Green Line shows you what the UN projected. If you have zero migration, they projected the workforce will start to shrink in 2026. And that's exactly what we're seeing. The workforce now is shrinking. You look at the chart on the right, shows the labor force splits into those who are foreign born and those who are domestic born.

Speaker 2: And what you see is any growth. We saw, and I started this in 22, because if I start during the pandemic, all the lines get messed up. So I could go back ten years and show you the same thing. The only real growth you've seen in the workforce over the last ten years or so is through immigration. If you look at domestic born workers, it actually started to shrink in 23 and 24. Where are we now? Both are shrinking. Both domestic born has dropped year on year, and the foreign born workforce has dropped year on year. Because obviously immigration policy, which means the labor force overall is now shrinking. And if you go back again, 50 odd years. This has only happened before really during the financial crisis and then during the pandemic. And yet here we are with two, two and a half, 2.7% growth, and the labor force is shrinking. That's why payrolls is low. That's why the unemployment rate has not gone up. And actually, if you look at the jolts to unemployed ratio. So using yesterday's Jolts data, which I've just added in the Jolts unemployed ratio is back above one. Now it's 1.03 after yesterday's number. Why does this matter? Because look at the relationship between the jolts to unemployed ratio and average hourly earnings. If we continue to grow in expectations that are increasing about growth this year, there will be positive demand for labor.

Speaker 2: If payrolls replacement rate right now, generously is zero, any demand for payrolls, if you've got a shrinking workforce, will tighten the labor market. You can make an argument. Actually, the replacement rate is negative, but that gets a bit complicated, so just call it zero. Any positive payrolls print tightens the workforce when the labour force is shrinking. So that jolts to unemployed ratio will move higher. If that jolts to unemployed ratio moves higher, then wages will be sticky at best and at worst will start to rise again. Why does that matter? Because when you look at PCE and here I'm splitting it into services and goods prices year on year, we can see the impacts of the war on goods prices. You can see how the year on year there has accelerated. But what I'm concentrating on is the green bit here. Services year on year. Because the point here is if you look over the last two years, this has been sticky. Services have been sticky around 4%. Why? Because wages have been sticky around 4%. Wages drive services inflation prior to the GFC or prior to the pandemic. Sorry. Wage inflation averaged around 2%. Services inflation was around 2%. The fed was hitting its target actually a little bit below it. Now we're at four and we're not getting down to the target. So I'm going to ask a quick question.

Speaker 2: We'll do a show of hands. What's going to be the next fed move. All those who think oh no we've got an actual poll. Apparently you can vote now with the thing or I'm going to do a quick show of hands. Anyway, all those who think the next move will be a hike, raise your hand. About a third. All those things are going to be a cut. Yeah, about two thirds. Everything I've told you. No one in this room should be expecting a cut. But I do as well. I do as well. I totally agree with you. I think there will be a cut next. And that's because of the Warsh effect. Yeah. Thank you. Prove my point. So if we go back to the slides, um, the concern is over. Wash. Um, look, we know what the president wants when it comes to interest rates. We know whoever got the job will have sort of hinted that, you know, he might be able to deliver for him. Um, and when Kevin Walsh did his hearing at the Senate, um, he started to lay out the intellectual arguments for lower rates. One of his big ones is AI. Kayla mentioned AI and the impact on productivity. His view is that AI will raise productivity. That lowers the neutral rate. That gives us room to cut rates. Absolutely right. If this was 2032. But it's not any impact from AI in terms of meaningful impact in terms of productivity is not going to come yet.

Speaker 2: And this is a combination of studies from OECD, IMF, Penn, Wharton, etc.. And I used AI because I'm calling trendy. So that's where you get in terms of estimates of productivity boosts across the US and the G7. It's minimal this year, next year, and arguably in 28 as well. In actual fact, you can easily make an argument that the level of investment going into AI will raise inflation in the next year or two, before AI eventually causes productivity to rise and inflation pressure to fall. Walsh is talking about rates now. He's also mentioned the labor market about how you're going to have all these job losses because of AI. It's probably overstated anyway. It's certainly overstated for the next couple of years. Ai is complementing most jobs. It's not replacing them. And this is estimates of net jobs replaced by AI. It's a lot lower than the gross number because you get a lot of people re-employed. You get a lot of labor market churn. But again, next couple of years not relevant. The other argument that Walsh made, which which Roberto alluded to, was this thing about a billion prices. He does know you never actually collected a billion prices, right, Alberto, because he talks about he wants to collect a billion, but we don't collect a billion every day, do we, by 10 million?

Speaker 3: No. Right. So every 40 days.

Speaker 2: Every 40 days, we get a billion. All right. He wants to collect a billion price. I think he means a billion different ones. Um, and his argument being, and this is what he said in his hearing, what I'm really most interested in, Senator, is what's the change of that 500,000,001 price, because that's inflation. Median inflation is what he means because that's inflation. No it's not. That's absolute nonsense. That's rubbish. Because if you get a billion prices and you rank them from the one moving up the the least, the one moving up the most, and then you pick the one in the middle, that's not inflation. Because what if the 500 million at the bottom, no one's spending any money on all the money is spent on the 500 million at the top? That's why when we look at inflation measures, we weight the factors based on consumption. You can't just rank them and say the median. That's that's inflation. That's absolute nonsense. Absolute nonsense. Um, but what it does do if you use median look. So this is called PCE 3.3%. Now the highest since November 2023 trending up. Can't deny that. But median inflation that's drifting down until last month when we had the impacts of the war. But median inflation has been drifting down for the last few years. So if you want to make an argument that inflation is trending towards 2% and we have room to get ahead of that, then you can make that argument. And that's where Warsh is coming from.

Speaker 2: And look, the pushback here is well it's a committee. It's 12 people who vote. He's only one vote. I think the chair has more influence than that. Powell himself has talked about he would decide what he wants to do two weeks before a vote and then go and build consensus. It's going to be harder for Warsh to start with. And it may he may struggle to get it. He certainly will struggle to get a cut in his first few meetings. But if the war did come to an end, oil prices come down. I still think. I still think the next move is a cut. And I still think, actually, if the war ends over the summer, he will get it in by Q4 and maybe even a second one if he's lucky. But I think he could still get a cut in by the end of the year. But I certainly think Warsh is next. His first six months or one year in charge is not going to raise rates. And that in itself is wrong. And you can make the argument as well. That even though Powell and Trump have not had the best relationship, and Trump is not the biggest fan of Powell, His pressure on the central bank has had an influence over the last couple of years or so. Here I'm looking at core PCE, the annual average, the deviation from 2%, and then what the fed did each calendar year when inflation is above 2%, they're hiking.

Speaker 2: When it's below we are at zero interest rate here. But when it's below here they're cutting. When it's above they're hiking. Look at 24 and 25. The two red bars PCE core inflation was above target. The fed cut both years. But they'll say it's about inflation expectations. That's what's important. You got to look at inflation expectations. Chart on the right is exactly the same thing with inflation expectations. When they're above 2% you hike. When they're below you cut 24 and 25 inflation expectations above 2%. Fed cut both years. So yeah, Powell might not have been as aggressive in cutting rates as Trump wanted. But the pressure on the fed, I would argue, has influenced monetary policy and it has meant the fed is more dovish than it maybe should have been. Now look in terms of inflation expectations now they're not running away. You look at five year breaks. They're fairly well behaved. You look at um a broader measure though. The Cleveland Fed five year inflation expectations still pretty well behaved 2.6 2.7% but the highest since 2007 now. And inflation expectations are one of those things. It's fine. And they're well behaved until they're not. And I think fed credibility is being reduced. Inflation expectations are ticking higher. And even though we're seeing that real money investors are reacting. So the chart on the top left this shows you flows to US treasuries. So duration weighted aggregate flows to US treasuries.

Speaker 2: What I've done I've taken the last two years. So May 24th to May 26th and compared it with the two years prior to that May 22nd to May 24th. When you look at flows to treasuries, there are around 20% of their level the last two years than they were the prior two years. If I look at tips flows, they're stronger over the last two years. And don't forget that prior two years, the May 22nd to May 24th includes the Covid inflation era. We're still seeing stronger buying of tips over the last two years, with this more dovish than it should have been fed, and it's not structural across all bond markets. I look at European government bonds. There are about 80%. The level. I look at Canadian government bonds we're about 90%. The level treasuries has been a structural break. That's real money. Investors saying the fed is not doing the right thing on monetary policy. It's not their inflation expectations that but it's their inflows. That's what we're starting to see. So look what does this mean for FX and rates. We are seeing tightening expectations everywhere. Um you know before the out the war started the aggregate one year rate change over the next ten years. Next year was -50 basis points. We're now at about 450 basis points of hikes. Um, if I look at it by country, the change in the expectations for December of this year. You can see we have nearly three hikes just over two hikes for the ECB.

Speaker 2: Um couple for the Bank of England. Everyone else is priced to hike at least once the RBA has come back actually. But not the fed not the fed. Over five years of above target inflation the strongest growth you know the smallest impact on growth because it's an energy exporter. Why haven't we got a full hike priced in for the fed. When we have for the ECB the BOE, Bank of Canada etc.. Because people know there's an influence here. And actually you can see the impact on growth here. Using the same scenario 70 to $100, 3 to 6 months. Japan, as I mentioned before, gets hit hardest. Uk eurozone get hit. Us Canada actually benefits because oil is such a big part of the economy. Us growth barely gets hit and yet we still don't have that full hike priced in. Um so I mentioned flows from real money investors. What you're also starting to see, I think is this be reflected in relative ten year yields. So here I take ten year yields just US and Canada. And the reason why I pick Canada is because historically Canadian and US ten year yields are so tight together. But look at the last two years the US is starting to pay a premium over Canada. Why? Fiscal policy is obviously a part of it. The level of the deficits we're running is a part of it. But I think in there as well is again, this uncertainty over the fed.

Speaker 2: And I'm not saying fed credibility is gone. It hasn't Because things would be a lot worse if it had. But it's being eroded. And this is where, if I'm right and water is going to come in and try and cut rates as feasibly as he can, that credibility continues to get eroded. And that means a steeper curve. That means a higher term premium. Yes, the curve is steepened. Um, but look, compared with where we used to be at 200 basis points, we're currently at around two stands about 70 odd basis points. There's a lot further to go. And what about the dollar. So here we see the chart on the left is dollar positioning what we call holdings or excess holdings in the dollar. And you can see the start of the conflict. We had almost the biggest dollar underweight since 2021. So when the war started as you would expect we saw dollar buying. Actually this chart shows it better. It's a bit messy, but these are five and 20 day flows. The dollar. And then positioning is the green line holdings. So you see at the start of the conflict, we saw straight away buying of the dollar. And then in the five day and then the 20 day followed obviously with the lag. Why? Well, yeah, a couple of reasons. One safe haven, but also because the dollar was the biggest underway we had in FX. And normally in you know, what you do at a time of uncertainty and conflict is you reduce position size, whether it's an underweight or overweight, you sort of move back towards benchmark dollar was the biggest FX underway.

Speaker 2: It is also a safe haven. So we did see that buying. But as soon as the ceasefire was signed, the two week one a week ago, we were still fighting two weeks, a month ago, we were still firing that ceasefire. Um, they started selling the dollar again as soon as that ceasefire was fined and was still signed. And we're still seeing dollar selling. Now in terms of positioning, I've got that same holdings chart. But now what I've done I've extended the full history to 2005. And what you see is, yes, there's still a dollar underway, actually quite small. When you look at it from the longer term perspective, it was the biggest underweight since 2021. But compared with history, there's a lot further for real money investors to start actually selling dollars again. Um, crucial to this is hedge ratios. Um, here are the hedge ratios for us investors and foreigners to the US. So what you see is the big actual buyers of the dollar when the conflict started were US domestic investors. That's the dark blue line which you see shot higher. So they increase their foreign hedge ratio from 10% up to 25%. They hedge their foreign assets, increasingly hedge them in that. That was the dollar buying we saw at the start of the conflict.

Speaker 2: Foreign investors actually on the right hand scale, inverted foreign investors have started to increase their hedge ratio in the US. And that's dollar selling. And what you see is they didn't do it last year. People talked about last year about foreigners selling Dollar Hedge America, Sell America, etc. after the trade tariffs. That's not what happened. And this is why. So this shows you the hedge ratio again that foreign hedge ratio on the right hand scale inverted. So if the the light blue line goes down that's selling of dollars because foreigners are increasing their hedge in the US. And you see last year didn't move. It was sort of between 55 and 58% nearly all year. Why? Because the cost of the hedge the cost is crucial here. So if you look back to 2022, before the fed started hiking in Q1 2022, the hedging cost was about 352535 basis points minimal. The foreign hedge ratio into the dollar into the US assets was nearly 80%. And then the fed started hiking. So the cost of the hedge. This is a three month rate spread, by the way. I should have explained us minus DXY weighted basket. Dark blue line starts moving up the cost. The hedge goes from 25 basis points to 2%. Foreign hedge ratio gets reduced because no one likes to lock in a loss and that's what they would do. So you see that foreign hedge ratio start to decline. Light blue line going up.

Speaker 2: And then we've seen it start to increase now as relative rates have narrowed. Now if I'm right and the next move by the fed is a cut not a hike. But we do get an ECB hike. We do get the Rbnz hike and we do get the Bank of Canada hiking. Then that relative rate is going to narrow further. And then the next move by the fed to cut narrows further. So the foreign hedge ratio will go higher because the dark blue line comes down and it's cheaper to hedge. And that's where dollar selling comes from. And this is why I think we will end the year with euro dollar close to 120, if not higher, particularly if the ECB follow through on these hikes and others do as well. And generally I think the DXY is going to fall by at least 5% over the second half of the year. And that will accelerate into 27 when Warsh gets the ability to cut rates further. Quick word on dollar yen. Um why can't the yen strengthen. Because it comes back to the hedging cost again. So here what I have this is from the perspective of a Japanese investor buying ten year treasuries and hedging in three month rates. If these dark blue thing is positive, that means the Japanese investor can buy treasuries hedging three month and he outperforms Jgbs. If it's negative, he underperforms. Therefore he can't hedge. So what you see is when they can hedge then the yen strengthens up to its on.

Speaker 2: The yen is on that side inverted. So here you get to ¥100 and then we get to ¥70 because here they can hedge. And so the current account surplus does not get recycled through the bond market in that world. And therefore you have a natural overweight in yen. The yen strengthens. The only time it didn't work was this period here. And that was Abenomics where you had the three arrows and they wanted to he companies were instructed to buy foreign assets Unhedged at that time. Where are we now? About a -100 basis points. So Japanese investors when they buy treasuries, have to be unhedged. This is why the yen can strengthen. And so we're stuck between this 160 that the BOJ 160 the 161 BOJ line in the sand. And the fact that the natural flows are still not yen positive despite what everyone wants them to be. It's about a -100 basis points right now. Boj could hike this week or next week they could hike again. Treasury yields rise. That 100 basis points starts to get eroded, but it has to go positive. This relationship before the yen can really start to strengthen consistently. So for now play the range until this point and then we will see dollar yen fall aggressively. And I am pretty much out of time. But I do probably have a couple of minutes for questions if anyone has any. Somebody's got to have.

Speaker 3: I don't want to take this place, but if you can go back to you had a chart on domestic versus foreign born workers. And that's the puzzle in my head, because in last year you had an increase in the domestic. Yeah. And how do we explain that big shift from the positive move to the big negative that's happened very recently? That really puzzles me. And I'm wondering if you have any thoughts on that.

Speaker 2: I do.

Speaker 3: I had no doubt.

Speaker 2: What happened, actually. And I confirmed this with Treasury as well. What happened last year was, you know, these are surveys, right? So the BLS finds people up and says, were you born in the US or were you born overseas and overseas? People who are born overseas actually said I was born in the US. So they lied. So they've adjusted for that now. And that's why you had that big rise in domestic workforce last year. And now it's come back down because actually they realized that people were lying because the government rings you up last year and asks you if you're born in the US or born overseas, kind of say you're born in the US. Even I did. They didn't believe me, but I did. But that's what happened. I think it was the accent. Yeah, apparently. Oh, but that.

Speaker 3: Brings a lot of other problems, right?

Speaker 2: Well, they've adjusted for it now, and that's why, you see, you know, the domestic workforce we know from demographics has to shrink. I mean, that's pretty straightforward. It's the foreign born workforce has been that supplies or will supply growth in terms of the labor force going forward. And if we have zero net migration, the workforce has to shrink because we know domestically has to. The domestic workforce has to decline. This is why people had this debate the other day as well. People talk about the participation rate. So it's really low participation rates really low. Therefore you can see that come up and that's going to raise unemployment. Please do not look at the participation rate anymore for the overall adult population, because you have this increased amount of retirees who aren't coming back to the workforce, so that overall participation rate will go down. You have to look at the 25 to 50 5 or 20 5 to 65. And those participation rates are very high. And so people look at the overall participation rate and say, well, we've got all these people on the sidelines. We don't. They've retired. That's what's happened. Every time for one more. Yeah. Colin.

Speaker 4: Thanks. So I'm trying to square, um, what you said about the labor market, you know, should be relatively strong given supply issues. Um, also Alberto's presentation saying that in at least headline inflation is likely heading up above four. You know, you mentioned there's some pass through to core. It may be somewhat limited, but I'm trying to I think Warsh would have to be the most persuasive man on the planet to get the broader committee to cut in that environment with inflation rising and employment solid. So how does a cut happen with that backdrop?

Speaker 2: I think to get the cut you have to have the war at an end, oil price starting to revert back down to where you know it should have been. And Warsh will make the argument that and he made the argument in the committee hearing as well. You have to look at inflation a year's time or 18 months time, and we're mismeasuring inflation and productivity from AI, etc.. Look, I agree with you. It's a stretch. It's an absolute stretch. But there are those on the committee who will agree with him. You know, probably probably Bowman and whoever Powell. I don't think will disagree with him either, which I think is interesting. He said he's going to be a passive member. He's staying on to make a point about the inquiry, but he doesn't want to be vocal or anything else. I suspect Powell might just go with the chair out of respect for the position. So there's only 2 or 3 more you need to get in there. And I think, you know, look, if it's a stretch for now for sure. But if we're starting to come, you know, three months from now, war is over. But real wages have been reduced by the rise in headline inflation.

Speaker 2: So perhaps there's some cracks in the consumer at that point. Then I think Warsh can be very persuasive. Look, you know when he sat on this the stage last year, a lot of people thought he made a lot of sense until you actually took a step back and listened to what he said, and then it didn't. I mean, I got to spend some time one on one with him last year, and it's time in my life I will never get back. So, look, he can be persuasive on the surface, but, you know, when you dig deeper, um, yeah, it doesn't it doesn't tally up. So, look, he'll have his job cut out, but I already showed over the last two years, I think the fed has been more dovish than it should have been because the pressure from the president. And now you've got a chair backing that pressure. And I think that will still have influence. So look I agree. Look every bone in my body tells me they should be hiking. They shouldn't have cut last year. But unfortunately my my belief in the fed is getting diminished. And I am probably out of time. Thank you Kayla.

Speaker 4: Talk about raising expectations. I can I'm not sure I can tell you everything about inflation, but I'll tell you about what price does is showing us because clearly inflation is coming back through a series of supply shocks. So I call this presentation inflation in transition. My idea was I'm going to talk about tariffs. Then the shock, you know, with the war in Iran. And now I'm thinking we're going to have to talk about tariffs again pretty soon. But we have the succession of supply shocks. And honestly, I think this is the moment where price actually shines because it's able to show us policy makers, economists and people in the financial sector what is really happening in real time, not only in real time. It also allows us to look deeper and measure things we couldn't measure otherwise. Like I'm going to show you today with tariffs and also with the Iran shock. And I can say that without sounding self-promoting because as Tony has said, it's actually State Street Company now, but it is truly a unique data set. So let's jump into it and I'll explain what is happening, mostly by looking first at the aggregate inflation rate around the world. As you probably know, if you're a client of State Street and you've been using price that series in the past, that price stats collects data in real time in 27 countries. What you're looking at here is the World Inflation Index, were. Those 27 countries are aggregated with weights depending on the size of the economy.

Speaker 4: But essentially I'm showing you the annual inflation rate since 2020. You can see a deep there during Covid. Then we see the increase in 2021 spike in 2022, right after the war in Ukraine, things start to improve significantly in the end of 2022, although we reached a level of about 4% and we seem to be stuck at that level, not much going on there. As you can see in 2025 with the tariffs, although I'm going to go deeper into that. And then finally we have the jump. This is updated until just a few days ago. You see the impact already happening globally in terms of the war in Iran. So I'm going to walk you through these last two events. The events on the tariffs itself is still lingering and it may pick back up, but obviously the attention is now focused a lot on the war in Iran, which is that that jump that you see there. It's actually a quite dramatic increase in all of price that series. So what I'm doing here, as I'm showing you all the countries where price starts, collects data, you have these bars that show you the range of annual inflation recorded in the last ten years. The blue dot is where price starts. Annual inflation was just a couple of months ago at the beginning of the war in Iran. You see those red dots? That's where we are today. So dramatic increases in inflation rates all over the place.

Speaker 4: I'm only living outside Argentina and Turkey because the inflation rates are so high that they actually distort the x axis. But every other country that you see there has suffered an increase in inflation. In fact, if you look closely, you'll notice the developed nations are the most affected ones at this point in time. They're getting this big jump, mostly because they tend to allow fuel to rise more freely. Developing countries are still far from their top ranges because we haven't seen much inflation in food. So I'm going to give you some details of that as well. But overall, perhaps nowhere is it more obvious if you use price data or the CPI in fact is if you look at the US and this is the monthly inflation rate of the US, big spike in price starts reaching 1.5% and has come down actually right now. This is remember a price level change for the moment. So inflation spikes and then it starts to come down. But as you can see there, it seems to be stabilizing at a level of 0.6%, which is in fact quite high on a historical basis. So we're going to start seeing some of these effects of that, obviously, depending on what happens with the war. But I promised you I was going to start with tariffs. So let me start there. Tell you how much pressure there is left in tariffs and what we might see moving forward. Before we go into details of the full shock, a couple of things.

Speaker 4: If you saw me present the last maybe two years actually, you know last year this happened only last year It feels longer. When the president announced the tariffs on on Liberation Day, we started doing some research looking at the microdata that price has. We can. We complemented it with a country of origin information for each individual good. In about five of the largest retailers in the US. And we made charts like this. If you look closely here, you have essentially two price indices one for domestic goods and one for imported goods. Imported are the the orange ones you see, for example, declines around the holidays because there's a lot of electronics. They're coming from China. And what becomes obvious when you look at this graph, there has been an impact of tariffs on inflation, which is quite, quite obvious. The numbers on the on the right at the end shows you the deviation in percentage terms from the trend that those price levels had before the tariffs were put in place. So roughly about six 6% for imported goods, about 5% higher for domestic goods. Domestic goods get affected, obviously because they compete with imported goods. They also have some imported inputs, so both types are affected. Now there's a punch line in terms of how much we have actually seen for the 2,023% increase in tariff rates in applied tariff rates. After accounting for all exemptions, we essentially found a 5% increase at the at the retail level.

Speaker 4: Okay. That's like a partial percentage pass through, we call it. And that's the number that ends up affecting inflation. You're going to notice several things in this chart. Quick reaction to news. But the big change is the change in the trend happening over time. In terms of trend. You might also notice that there's a slowdown that happens right around October November last year. It feels like a long time ago, but it's actually very recently. If you remember what happened there, we had a series of trade deals announced by the administration. We also had, some tariffs that were rolled back, although not advertised that much. And finally, we had already in November some information coming out that suggested that the Supreme Court was going to strike out the the tariffs, which they eventually did in February. So all this seems to have contributed to a relatively slow down in that pressure that we were seeing on inflation. Perhaps nowhere is it more obvious that if you look at the papers results by country, you would notice that in the red line that you see here, which is the price index for China, China's prices are obviously very important for this bundle of imported goods in the US. They reacted very quickly to this news. And right now they're actually on a trajectory and a trend that looks very similar to what they had pre-pandemic. So all this led me to believe that some of that pressure was already passing on by the beginning of, of this this year in terms of how much it actually affected inflation dynamics.

Speaker 4: You have this chart over here which shows you the cumulative effect over time. So just to describe this, the way we do this, we calculate at the sector level the most disaggregated sectors we can the deviations from the trend. And then we aggregate it up using official weights. So this basically tells you how much the headline index deviated from the previous trends. And we're kind of assuming every other sectors like housing all that is not being affected by the tariffs. So that's the numbers we get. We essentially have to account for imports and domestic because domestic also get affected. We get about 0.76 percentage points increase in the headline CPI or headline index. So it sounds like a small number, particularly if you come from a country like mine, 0.76% inflation sounds really small, but if you put it in context, that basically means that the CPI in the US, which in October was 3%, it would actually have been closer to 2.2% much, much closer to the Fed's target. And I think the discussion would have been quite, quite different. In any case, one question I often get when I show this 5% increase for the 20% tariff is how much pressure is there left? I mean, there's obviously some incomplete pass through in percentages. And here it depends on which side of the aisle you are. If you're a Democrat, you would probably face it like, oh, that means that there's a lot of pressure left.

Speaker 4: You know, only 25% had been passed on to consumers. So maybe more is coming. You know, that's that's one way to phrase it. The other way to phrase it is the one if you're a Republican, you would actually say something like, wow, consumers only pay 25% of the tariff. So it means that the tariffs have worked. And, you know, the ones who are paying for the tariffs are the foreigners or the firms. Okay. And that's actually what the president did in his op ed in January of this year, he wrote the Wall Street Journal. He said, my tariffs have brought America back. And he, in fact, cited our work. He said, you know, you look at the results from this paper. There's this study at Harvard Business School that shows that 80% of this cost is being paid by others. Okay. That's basically the opposite of the 20%, 25% that we detected at the consumer level, which is great. I love it when the president cites me. I actually think it's probably a very rare occasion where the, you know, the president's writings, there's actually an academic paper being cited. So I'm very honored by this. No, I really am. I truly am. And, you know, we I write papers so that policymakers know this. So this is in fact, very, very nice for me. But there's there's there's something I want to clarify. There's a misinterpretation here. Pass through in percentages is not telling you much about the incident.

Speaker 4: Well, not tells you something but not, not directly reflecting the incidence of the tariffs or who is paying. It's also not really telling you about how much pressure is left by itself, because maybe the 25% pass through is where we'll get a stock. In fact, to know this answer, we need to know the pass through in dollars. Okay. The pass through in terms of percentages, how much for each dollar that has been paid at the border is actually being paid by consumers. Okay. And that's something just to illustrate how different this is. Let me use a very simple hypothetical example. Think of a good that is imported for $10 at the border. This is the top number that you see there. Now I apply a 20% tariff to that. Good. That's a $2 increase. The price goes up to 12 at the border. Again this is the the price that importers pays. That's the 20% change that we basically saw. Now at the retail level this good would normally be sold for probably twice as much. Okay. Why is that? Well we have distribution costs, we have local costs. We have margins that are applied by the wholesaler, by the retailer. So in normal times pre tariffs these sells for 20. Now imagine those $2. The full amount paid by the tariffs are passed on to the consumer. The price rises to $22 at the retail level. And that's only a 10% increase. Okay. So if you did the calculation only 50% pass through in percentages.

Speaker 4: Okay. That's that's the thing. Now even though there's been this full pass through and the consumers have paid all, and there's essentially no more tariff pressure left to be passed on to the consumers. This is an important distinction that often gets not reflected much in the discussions. Um, part of the reason is we don't know much about these differences. It varies by goods. So it's hard to to measure it. But roughly half in normal papers have found that roughly half of the price that we see at the retail level is driven by distribution, lower costs and margins, and are not directly affected by the tariffs. To make these numbers a little bit more realistic. By the end of last year, what happened is that basically for every $2 of tariffs paid at the border, $1 was already being paid by the consumers. That actually gives you the percentage numbers I showed you before, 5% increase at the retail level for a 20% percent change at the border. In fact, again, because we have this great price tax data, we actually went further. We said we're going to estimate for each category of goods. The share of the distribution. We're going to run a regression to try to measure how much consumers have paid over time. These are the results we have in an update of the paper. You can see in October 2025, 43% of the dollars had been paid. Tariff dollars had been paid by the consumers.

Speaker 4: Very similar to the the kind of an average number we had before. And by March of this year, that was that number was 73%. So three quarters of the tariffs were by the beginning of the year already being paid by consumers. And it also suggests that much of the pressure had already dissipated at that point in time. And we were not going to see a big increase in inflation still lingering unless tariffs are brought back up, which explains why those monthly rates in price stats were actually falling. There are some seasonality there, but overall this was looking very benign. By the beginning of the year, things look like, you know, the inflation was going to start to fall down for 3% back to to target a lot of optimism on that side. I was wondering, what am I going to talk about in the retreat? But then fortunately for me, not for the world, we get a new shock. Okay. And this one is actually much more complicated. Well, in some ways it's complicated because this is energy. We've seen energy shocks before. They feed into everything. Eventually they're global in nature. So it can rise, can have a big increase, but they can also be temporary and they are affected by expectations. So the results could be very short lived. So just to give you a sense of what is actually happening to this day, I'm going to talk about fuel passthrough and first show you in the US what happened to gas prices over time.

Speaker 4: So I took the price stats, fuel index, car fuel index normalized it to 100 on the day the war started. So the y axis shows you the cumulative effect. You can basically see that prices have risen 50% at the pump in the US at this point in time. Quite a dramatic increase. You know, a lot happening at the beginning. Then things seem to stabilize. We saw another jump and things seems to stabilize again. The CPI has gradually caught up to this trend. So quite a dramatic increase happening which explains why we saw that big spike in the price index. This is the same monthly inflation graph I showed you before. Only I'm taking a screenshot from the insights platform. You can see it peaking at 1.6% actually, and then coming down and stabilizing at 0.6. So energy is car fuel is about 3% of the CPI basket. That's a direct effect. Energy. Energy for the home, things like that. If you add them up that's about 7.5%. So this is roughly the effect you would expect. But it was not the only country affected. In fact, every single country in the stats database seems to have had some impact. These are the same fuel indices over time. You can see Australia was greatly hit at the beginning. They decided to lower taxes on fuel and that seemed to keep things under control for them. New Zealand became the country most affected and gradually we saw the US taking over in that chart.

Speaker 4: By the end of this graph, which is just a few days ago, we essentially have kind of three sets of countries. We have the US, South Africa, New Zealand and Chile as being the most affected right now. These are countries that have basically led markets to whatever they needed to do supply and demand. Prices rose. In fact, Chile is very interesting here. You're going to see Chile kind of flat and then jumping back up by a lot. So Chile has a policy to kind of regulate this through subsidies. But they decided that this shock was too large to be paid with subsidies. They were going to have too much of a fiscal cost. They actually let that price come up dramatically. So we have this bunch of countries that are greatly affected in the short run. We have several countries like Australia. I mentioned like Spain, you're going to see there that implemented tax reductions or sometimes imposed provide subsidies to temporarily bring this down. And then we have some extreme countries where they basically regulate fuel. And we have seen not much of an effect, for example, happening in Russia here. So lots of heterogeneity that is going to impact differently. These these countries inflation. Now I've looked at something very similar, very similar shock through my research. This is work I did leveraging again price stats for the war in Ukraine. I wrote this paper called Large Shocks travel fast, which is the idea that the energy shocks are quickly passed on into other things, particularly when it's a very large kind of shock.

Speaker 4: The patterns we saw back then are actually very similar to some of the patterns we're seeing right now. In terms of fuel prices. You have that chart on the right kind of showing you the same thing. But at the time of Ukraine, big jumps in the US kind of in between. We have several European countries. And then you have Japan, which is basically flat because they subsidize and control the price of fuel. They keep it low as they can, like they're doing right now. So kind of similar patterns happening there. Um, now what was interesting in the case of Ukraine is that it did not remain only an energy shock by itself. The context at the time was one where we had had this supply shock pushing prices up, creating a lot of pressure. When the energy shock comes, we start seeing that pressure being liberated across many sectors. And for example, this is the effects that it had on the food frequency of price changes on the left in many European countries immediately after the war, the prices of a lot of food items started being adjusted quite frequently and leading to more inflation. We saw the same thing in the US and in several other countries. It seemed to trigger a massive increase of all this pressure, um, or release of all this pressure in other sectors as well, of course, contributed by the fact that Ukraine was a shock on a country that also produced food.

Speaker 4: We at the same time had the effect on fertilizer prices that had been rising. Fertilizers use a lot of fossil fuels in their production, so they usually come up with energy. And we're starting to see, by the way, some of that pressure, although milder right now. And it led to this effect that it had on food inflation. This is the chart for the US during 2020 223, the worst time of the inflation crisis in the last few years. Now, are we going to see the same thing? That's what I wanted to show you. Well, if I first compare the fuel impact of the Iran war with the Ukraine war, which I'm doing in this chart, it's actually worse. You can see there both indices normalized to the start of the of the war. Prices started rising sooner in the case of the Iran war. And then they ended up rising pretty much like 50% more. So that's a significant higher fuel pass through in this case. Part of this has to do with the behavior and expectations about oil. I also think many countries decided not to intervene too much this time as compared to the war in Ukraine, probably because we were in a high inflation environment back then. They were more concerned about this than perhaps they are now thinking that this is a temporary effect. But regardless of that, it has a bigger effect on fuel, which immediately affects transportation prices.

Speaker 4: You can see here I'm putting all the countries together and separating these bands essentially show you the heterogeneity across countries. And you can see a comparison between the impact on transportation prices of the Iran war and the Ukraine war. Again, more impact on transport. Now, what's really interesting is this crisis is behaving differently in terms of the impact it is having on other sectors. And you can see that also clearly on the right of this chart. That's food and non-alcoholic beverages Even though Ukraine seemed to have this impact and upward trend, we do not see much passthrough yet happening. In the case of the war in Iran, it's far more contained on that dimension. Just to make this more obvious. Let me show you some heat maps where I. The colors actually represent the value of a passthrough coefficient from fuel into the price indices. That price that monitors in each one of these sectors. So you have transportation on the top, you have food and beverages and all the other main sectors. We do this in every country. So and on the x axis you have the number of days since the shock hit. And one thing you'll notice here is there's a lot of red in transportation. What I showed you before, in fact, it's been worse. In the case of Iran, it's darker towards day 60, but there's not much pass through into the other sectors as compared to what happened in Ukraine. Yes.

Speaker 5: Can I ask a question here? So on the point of pass through is that pass through or directly, because Ukraine is a big agricultural suppliers. Yes, I sense that that would be different, right?

Speaker 4: Yes I agree. The context is very important. Not only the country that is getting hit. There were a lot of commodity food staples that increased at that point in time. I also think what I was describing at the beginning, the context of relatively high pressure, uh, succession of Covid shocks had led to a moment of high inflation. And then when you get an energy shock, all that passes on quickly. There could be various factors here, some direct effects and some indirect effects as well. We're actually working actively to try to identify at the good level what the percentage of energy and that each good can can have. So we can measure the pass through coefficient more cleanly that you can see here. But, but regardless of whether it was the Ukraine itself or not, there's, there's not much going on in terms of pass through yet to other sectors. Now we do expect it to happen if the oil price remains high for a significant amount of time. But if you look at the pricing behavior of firms at this moment in time, there doesn't seem to be a belief that this is going to be a long lasting type of crisis. We do not see much of that anticipation in prices, which we saw, for example, in the case of of the of the tariffs. So for now, the takeaway I want you to, to get from this discussion is that Iran is mostly a fuel inflation story at the moment. We are roughly three months from the start. You can see it's you know, the response across countries is greatly aligned to the response of fuel.

Speaker 4: So there's there's still not that additional component, even though, let me point out that leads to essentially the same kind of the same aggregate effect because we have more impact on fuel, as I showed you at the beginning, less impact and pass through on the other sectors. So those two things compensate each other. And roughly on an aggregate basis we are seeing similar effect. In fact, I've been updating this in the last few retreats that started like a month ago. You can see it flattening out in the case of Iran, again suggesting this effect is still contained to fuel. And naturally, fuel has been inflation has been coming down and flattening again this curve. So this should be relatively good news. If you're a central banker, for example, you know, this looks like a canonical type of temporary cost shock. But it obviously depends on what happens to the world moving forward. I will say, though, it has dramatically affected annual rates, like I was telling you before, and these rates will be here at least for 12 months due to the base effects. This is the case of price stats for the US. You can see reaching 5% roughly. It should stay at that level. What I did here is build a counterfactual where I removed the direct effect of fuel from this, and the rate would have been around 3.4. So still a little bit higher than some of the impact on the other sectors probably filtering through. But most of the story right now is about about fuel affecting this.

Speaker 4: There are some additional contributions if we want to look at the details. I am running out of time. But just to give you a sense, obviously I've shown you the fuel change. That's the percentage change in car fuels. We also have to take into account how much fuel has in the basket. That's the second column that you see there, explains why China is one of the countries that has received kind of the biggest impact here, because it has a 20% increase, but an 8% share of fuel. The ranking doesn't really change much when you add the weights to it, except that, for example, if you look at Turkey, they're on the second line. There's been a significant impact. But Turkey's inflation is already so high that the fuel contribution of this shock, which is what you have on the second to last column, that's the the fuel share in the annual rate is actually not a big deal in the case of Turkey, quite a big deal for China. And many of the countries listed at the top. Japan is also interested. So Japan has kept the fuel prices around 9%. Not necessarily a big actually not a big at all weight in the CPI basket. But still, since annual inflation is so low in Japan, this has led to a meaningful contribution of about 13% to the annual rate. In fact, Japan's inflation has been falling quite significantly. The CPI is catching up to that trend. Lots of heterogeneity I'd love to tell you about and happy to discuss in in the Q&A.

Speaker 4: But let me just summarize some of the key takeaways so far. So inflation pressures are obviously rising again. The sauce is shifting though from tariffs into fuel tariff passthrough. If you compare it to the first trade war was clearly faster this in 2025. But it was still gradual and incomplete in percentage terms by March of 2026. Most of that tariff pressure, according to our numbers, had already dissipated away. So we were looking into a more benign inflation scenario. But now we get this Iran shock that is quite dramatic. It is a fuel shock. Good side to it. We're not seeing much of an impact yet on the other sectors. So so far it remains mostly a fuel inflation story with limited spillovers. Of course, as I put on the title of that paper, large shocks do travel fast. The key risk here is that higher fuel costs will gradually creep into the prices of other goods. Again, though, you would have to remember what share is fuel of the total production cost. Kind of a similar discussion of the percentage pass through in dollar pass through that I showed you at the beginning, but it could still lead to higher expectations. Of course, we are not seeing it in the data, but the beauty of price stats is if we were to see it, we would pick it up quite quickly and we would let you as clients know, okay, so I'm going to leave you to that and then I'm happy to take any questions. Thank you.

Speaker 5: Yes. Ask a question. I don't know if you want to comment on based on this. It would not seem that the ECB should rush into hiking. Would you? Well, if what you are saying here that this is mostly a fuel story.

Speaker 4: Yes.

Speaker 5: It would be a mistake.

Speaker 4: Well, actually in the UK we already had detected higher inflation before this. So that complicates in the in the in Europe. Now you talked about the ECB. Your question was about ECB. So I would be very cautious in the UK and I would lean towards hiking rates. In that case the US is a little bit different in the US. I think the story is more cleanly what I have just described. And in that case I would certainly not rush to lower rates right now. But if you're Kevin Warsh thinking about what to do, you can wait it out and hope that this will, you know, start to, to, to fade away. Now I, it all depends on what happens with the price of oil, right? Because if the price of oil remains high, because the war is persistent, it is true that the energy fuel shock will dissipate away. But we should gradually see some upward pressure happening in the other sectors. So, um, it's not clear that this is in the absence of a resolution to the war, this can still be a problem that creeps up and pushes inflation up, um, moving forward. So um, do you want me to be an American central banker or a peon? If I'm an American, I would just wait. I think that's probably what many of them are thinking. We're not going to lower rates. Let's let's see what happens in the next few months. And if we start seeing some pass through, then maybe it might be time to hike. If I am a European, the overall context was actually according to the prices data, picking up stronger so I would be more inclined to hike. Uh, and potentially the, the, you know, since we have two wars there and we have also the war in Ukraine, it might be more persistent in this case. Yeah. Yes, sir.

Speaker 6: Another way of asking that question is before the war in Iran, were you sick? Was your data showing a trend toward the 2% type of number?

Speaker 4: A not quite yet. It was, um, you know, before the tariffs were put in place, we did detect that decline. Then the tariffs were put in place in the US. We saw a sort of a 3%. And by the beginning of the year, it seemed from the monthly rates that we were gradually creeping back down. I saw it from the tariff trends that some of those trends were returning back to pre levels. So my expectation was that in the US we would see those come further down. Okay. So but that is all changed now with the war in Iran. Yes.

Speaker 7: Sorry. Just just a clarifying question on your 25% pass through calculation with the dollar slide that you showed. So in that example, would it be the $1 retail versus the $2 or $0.50 versus the $2. I'm just trying to understand your 25% pass through.

Speaker 4: Yes, I think it would take me. I cannot go back. But if you think of that example, essentially what happened is we had a 20% tariff at the border. So think of that good ten. That was a $2. And at the retail level we only saw half of that. So $1 being passed on to the consumer, which gives you this roughly 5% that we get at the retail level for the increase of the goods relative to the 20% tariff rate.

Speaker 5: But it was a higher base, right? Instead of the ten, it was a 20.

Speaker 4: Yes, exactly. So this $1, which is half of the tariff that the importers paid, was paid by the consumer is from a base that is. 20 so that leads you to the.

Speaker 7: 5% over the 20% is your quarter.

Speaker 4: Yes, exactly. There's a 40% pass through in percentages. Yes. But if you calculate that in labels, it's a 50% pass through. That's correct. That's the right calculation. And this difference is what. And by the way, this happens with any any shock, we tend to think, oh, you know, tariffs increased 20%, retail prices increased 20%. That's not the case. Because of this. We tend to think we get an energy shock, the 40%. That means goods are going to rise 40%. No, it depends on the share of that. That cost, which would be fuel on the overall retail price that we have today for those goods. Usually this information is very hard to come by. We can have estimates in the case of fuel of how much energy intensity there is and the cost structure of these goods to estimate this pass through well, and we can do it in percentages and also in levels to get us the levels. By the way, each one of these measures is useful for something different. The percentage measure is useful to understand essentially this 5% increase how it's going to affect inflation because that's really what feeds into inflation. The other the one the pass through in levels is more important. If we're talking about who is paying for the tariffs and also about how much pressure we think there might be left. If we assume that every single dollar that is paid at the border will be passed on to consumers, then you can simply use this as a as an idea of how much is still going to happen moving forward. Yes, sir.

Speaker 8: I believe last last year, when Chair Walsh took the stage here, uh, he expressed his preference for real time data instead of lagging data. So I'm just curious, with all the movements in inflation, are you getting more calls from the fed or your colleagues from states like what's the interaction over the past few months?

Speaker 4: So I am not these days. I don't hear, you know, if they are selling more, the stuff I don't hear because it's their company. So I cannot tell you if they're selling more to the fed. But I can tell you on the research side, the research department definitely is, uh, is reaching out, but they were already last year with all the tariffs. Price has this ability to you know, since we have this not just the high frequency but also the details, we can do stuff like I showed you, like I identify the country of origin and things like that in ways that official statistics never provide. So on the research side, there's always been some some of that. You're right. That was said in a response that he plans to use a billion prices. That may be a reference to the Billion Prices project, which is the academic research project that we had with Roberto and MIT, and sort of was the beginning of all the price cuts. I don't know if he was making a direct reference, but I know, obviously, since he was here last year that he knows about all this. So I imagine we at least inspired him a lot. And if I were State Street, I would be willing to give him the data because policymakers need better information to make better decisions, for sure. Yeah. Very good. All right. Thank you so much.

Thanks, Michael. Nice to be with all of you in London. As Michael mentioned, this is a presentation that is trying to go really back to basics in trying to understand foreign exchange demand really at a core level of what's driving FX. It says four dimensions; it's actually going to be more like sixteen or maybe thirty-two, depending on how… So it gets complicated very, very fast and so what I'm hoping is that I'm going to take you through step by step how we think about this and how we're ultimately using some of the State Street data to try to shed light on the underlying drivers of FX demand. Now, I will also say that I'm mindful that I'm speaking in the UK. I'm going to be mostly focussed on the dollar versus the euro today. That's mostly as a matter of convenience because just looking at a bilateral, two currencies makes it a little bit easier. Already here, you see we have 16 or 32 depending on how you look at it dimensions. So once you start adding more, it gets even more complex. So this is work that I've been doing with Alex Cheema Fox - you've probably seen him on this stage here many times before - also [?Haran Shia 0:01:26.2]. It also comes out of work I would call also fundamental-type work that Alex and I were doing and that I presented at these retreats, which was trying to understand how real money investors - so equity investors, fixed income investors - do their FX hedging and how much that FX hedging contributes ultimately to FX demand. When we were doing that work a few years ago, everybody asked us, 'Okay, fine, so you've shown us that fixed income folks, they tend to hedge more, tend to hedge more with ratios closer to one. Equity investors, more like a half; depends on which domicile currency and so on. Okay, can you do the next step and back out: well, given that and all the reshuffling you see around the world, what are the implications for forward demand? What are the implications for spot demand and so on?' So that's what we're going to be doing in this work, is really taking what we know about how real money investors do their hedging and also their speculation, and trying to take the next step and say, well, how does it all add up to something? How does that get us to rethink what's going on in the FX markets? Now, the reason that you want to do this is because these are really pretty economically distinct components. For example, the hedging component of FX demand is quite different in its nature than the speculative piece of FX demand. It might have different implications for correlations with underlying returns, with future returns and so on. So it feels a little abstract right now, but it'll become more clear when I get into it. Here's the simplest schematic. This is a four-dimensional piece or a four-component piece. I'm at HBS; we always like to do everything in a two-by-two matrix, partly because you can fit that on the board whenever you're breaking things down. It's a little incomplete here so I had to stuff a lot in each one of the boxes, but you'll see where this is going. Why don't we start with: imagine a UK investor or a European investor who is buying US assets, so I'm going to call that foreign. Right now I'm using foreign from the perspective of the US, so they're buying US assets. That generates spot demand for dollars. That's the only way that you can buy, say, US… Let's say you want to buy shares of Nvidia. Does it generate anything in the forward market? Not mechanically. However, if you hedge your position, so you're say a European investor buying Nvidia shares and you want to hedge your foreign exchange risk, well, let's say you have a hedge ratio of 0.5 and you're buying $100 worth of Nvidia, you'd be shorting $50 worth in the forward market. So you'd be minus that hedge ratio 0.5 times whatever that asset purchase was. Then in addition, you might be speculating on FX, so you might have a view on the FX markets. You might take it an addition, something going on there. Then this one turns out to be relatively important but people forget about it, is there's a huge rebalancing piece as well, which is to say: a month passes, two months pass and your Nvidia stock doubles. Well, your hedge is now out of line and you had a target hedge of 0.5 and so you're now short $50. But your Nvidia is $200 so you're no longer at 0.5, so you need to re-up your hedge. You need to go short an additional $50 in order to get to that target hedge ratio. That's this rebalancing piece. This is after the US dollar asset appreciation. This is just for the foreign investors. Domestic investors, same thing. This would be, say, a US dollar investor. They're buying foreign assets; they have to sell US dollars to do that. They're hedging by buying US dollars, forward rebalancing after the foreign asset appreciates. Say they buying Siemens stock, let's say, and then there's also speculation on FX which we would call the residual. Now, you can see right now I'm just talking about Nvidia and, say, Siemens. So it's just two equity investors in two markets, but it gets pretty complicated because we don't just have equity investors. We don't have just two currencies. We have fixed income investors. We have multiple currencies and so lots of different components driving, ultimately, both the spot demand and the forward demand. State Street, as you know - most of you I think are on this insights platform - can track the underlying flows in the FX markets for many of these real money investors. What we're going to try to do is say: let's try to decompose according to these different motivations and try to say, can we figure out how much, say, the rebalancing or the hedging piece or the speculation piece is ultimately driving what's going on in the FX markets? That's what I hope to show you today. The first set of messages is really how to think about which pieces of this are important. I think one of the central things to come out of it is that this rebalancing piece is hugely important relative to what we thought. Most people don't even think about that when they think about FX demand, but it's absolutely central to thinking about flows in these markets. Then the second thing is that many of these things happen with lags, so people don't rebalance their portfolios right away. Actually that generates a fair amount of predictability in the flows in the FX markets. Just give you a simple one, which is: imagine that Nvidia position I was talking about. European investor goes and buys Nvidia. Well, maybe they don't do their hedge right away; maybe they do that after a month. Nvidia doubles. Maybe they wait to rebalance to get to their target hedge ratio again. Maybe they wait, even do it at the end of the quarter. That generates this predictability in the flow relationships and I'm going to show you all of that you can see in the underlying data. Actually it gets us to think a little bit differently about demand for the dollar, demand for the euro. As I said, this is how we came up with four dimensions. Really it's spot versus forward, so there are two places where you can trade in the markets. One of the things I'm not going to talk about today, but we're very interested in doing next, is looking at the currency basis and the relationship between these different components. There's the currency perspective, so do you view the flow from the perspective of the investor's base currency or the foreign currency? Are you dollar based, euro based and so on or pound based? Asset class, so equity investors and then ultimately the domicile. We're going to be looking in this space here, so European investors trading locally and trading foreign, and American investors trading foreign and domestic. We're looking at both spot, forward, equity and bond funds so like I said, lots of different pieces. I'm going to drill down on one or two of these where it's easy to see what's going on, but just keep in mind that again, there are lots of these different components that ultimately you'd want to keep track of. Last thing I'll say is - I mentioned this already - but these pieces of currency demand, they're linked to each other, some of them mechanically. They vary in their persistence and their predictability. By mechanical what I mean is, of course, if you buy $100 of assets and you have a target hedge ratio, there's just a mechanical corresponding demand in the FX markets. That's a mechanical piece and then there's a persistence piece as well that's quite interesting, which is to say if European investors are buying Nvidia today, it's actually quite likely that they're buying Nvidia tomorrow. That generates a persistence also in the FX demand to the extent that they are hedging in a consistent manner. Last bit of motivation here is in terms of thinking about: why this potentially matters is that hedging might modulate the relationship between returns and asset flows. I'll just - again, this is maybe a little bit complicated - but try to give you some intuition. So let's say it's a European investor buying US equities. Normally, that's going to create demand for the US dollar. Think about that as they're buying dollars in order to buy those Nvidia shares. However, if the investor is fully hedged, then they're going to be selling the dollars in the forward market. So there's essentially a corresponding offset and that's going to mute the FX pressure of asset flows on the currency. I hope that's somewhat intuitive. Now, there's a little bit of evidence on… This is a very simple schematic. I want to be careful in taking a causal interpretation here. But if you just do something simple and say, gosh, let's look at the average hedge ratio of different investors and look at the correlation between their returns, their asset flows and the returns, you can see that the folks who have the highest hedge ratios - the lower hedge ratios, are associated with stronger positive correlation between flows and returns, which is consistent with this modulation story. Again, in other words, in places where you're hedging more, the relationship between buying the underlying asset and the currency return is much weaker. Pretty intuitive. Mostly, this is related to some previous work. I would say there's been an explosion of academic and practitioner work in this area over the past three or four years. It's very, very exciting and it's really been driven by the data availability. I think relative to even some of this very cool work that's out there, the State Street data is ten, maybe a hundred times better because it covers so many more different actors and it's so much more precise - and really is the only data that allows us to do this decomposition in terms of the motivation of why these different folks are trading. How do we do it? What do we observe? Well, we observe here the spot asset flows, so we observe stage three. We'll see in the custody data whether, say, a European investor is buying a share of Nvidia. Then we see we can back out these other pieces so we can say, gosh, well, if you tended to hedge with a hedge ratio of, say, 0.5 then the corresponding hedging of that flow would be this FX forward hedging piece. We can calculate that. If Nvidia experiences some returns, we can then compute: well, how much would you have to adjust your hedge for those returns? Then at the same time we also see how much you're actually doing every day in terms of what are those forward positions that you're executing. We know it all has to add up, so anything that is left, we just call speculation. In other words, if your hedging motivations suggest that you should be short $50 today and instead we see that you're short $75 today, we say that $25 piece, we're going to call that speculation; that delta between what you're doing from a hedging purpose versus what we see you actually doing. Now, there is a question which is: well, how do you measure what your target hedge ratio is? Here we're relying on previous work that Alex and I did, where we showed that there is a remarkable stability at the investor level in terms of how people actually do their FX hedging. So people do try to typically stick with target hedge ratios. They tend to rebalance relatively quickly; somewhere between a month and three months on average. Especially in the fixed income world, you see a huge amount of stability in terms of hedging programmes of the investor. So we're relying on our ability to estimate that; otherwise this decomposition wouldn't be particularly interesting or useful. One more preamble set of data before I really get into the meat of what we're able to do here. This is a picture from our other paper. I'm showing here, this is European hedging. This is US dollar funds. We also do this for the counterpart. The top picture here is for equity funds - the fraction of those funds that do any FX hedging at all - and you see that in the blue, which has trended up over time since the beginning of the sample. Fixed income, the vast majority of funds tend to hedge out their exposure. Then when you go to the very bottom picture, the blue is equities, the red is the fixed income, is the average hedge ratio for those entities that choose to hedge. What you see is not very surprising here. For the fixed income investors they tend to have hedge ratios close to one. The equity investors here, the numbers tend to be, for the folks that hedge it all tend to have numbers that are closer to about 0.5 on average. Let me show you a decomposition and here I want to go quite slowly because otherwise, the rest of it is not going to make sense. Let's just start and look over here. The leftmost picture, this is European equity funds buying dollar assets, so what do you see here over time? The black line - the dark blue line, I should say, the darkest line here; it's a navy - is showing the asset flow. So this is starting in about late 2002, meaning European funds are buying US assets, buying US equities into 2008. Then they're shedding them for about a decade and then there's into COVID buying assets again and then shedding them That's on the equity side piece. You've probably seen, I think, presentations from Michael and others about these kind of numbers of over time. Now, what's interesting here is they own a lot of assets overall. What's happened to US dollar assets over the equity assets over the past 20 years? Anybody? Yes, they've been killing it. If you've owned any equities in dollars, unless you have the worst stock selection in history, you've been making money on your US equities relative to your portfolio and the other assets. What does that mean? Well, if you're hedging, it means actually you're consecutively having to go more and more short the dollar as those US equity assets appreciate. That is this piece here, this return hedging; this green one right here so this is driven entirely by the US stock market doing incredibly well and having to shed assets. You can see here this yellow line is the total forward position. This is what I'm saying; this is a hugely underappreciated thing that's come out of the data is that actually, the net forward position that you see investors having largely tracks that return piece. In other words, being able to track the relative performance in the different asset classes is central to understanding what's going on in the FX, the foreign exchange flows. The last thing I'll show you here. I started with the asset flow here. In terms of the flow hedging, you see that that has a flipped sign relative to the asset flow. That's natural because if you're long the asset, you're short in the hedge. You can see that in the equity world, the flow piece is not super-important. What's been centrally important, at least thinking about dollar assets, is this return hedging piece. So that's in the equities world. Now, when you shift over to the fixed income world, so here where European fixed income funds, so these are European funds they're buying primarily US treasuries, agencies and some corporate bonds and so on. So you can see here the picture in the flows is quite different. Over time, they've just been steadily accumulating US dollar fixed income assets over time. Here, the central messages on the returns haven't been so important here in driving what's going on. However, just the hedging behaviour has been very, very important. In other words, people like to talk about Treasury demand driving demand for the dollar. Well, actually a lot of the US Treasury demand is hedged. So you see that it's almost completely offset here because European investors tend to hedge with hedge ratios close to one. You need to think about, when you think about what's going on with US treasuries, they can be actually quite distinct from what's going on with the dollar because of that hedging behaviour. Here, in other words, you learn two quite different things in terms of the underlying drivers of the FX markets from the equity side and the fixed income side. We can now flip the script a little bit here. Let's look at US dollar funds buying European assets. Dollar funds, the asset, here the picture is more over 25 years essentially shedding on net European assets in terms of on the equity side. However, the hedge ratios tend to be so low for European investors that it doesn't generate a lot of mechanical demand or net demand for the euro, which would normally be the offsetting piece here. Actually here, a huge driver of the demand is just the residual. In other words, when you're thinking about US equity funds and they're trading vis-à-vis the euro, the hedging behaviour is less important and less a thing that you should focus on. Similarly, the return piece here has been, I would say, a little bit important but not quite as… Certainly not as important for the European investors, and that's partly because the hedge ratios overall are just lower for the US domiciled investors. Then when you turn to US fixed income funds buying European assets, here the hedge ratios are a little bit larger. So it actually looks very similar to that previous picture I showed you on the right side of the screen in the sense that the asset flows tend to be essentially somewhat undone by the hedging behaviour when you're thinking about the consequences for the foreign exchange markets. I'm going to skip this. You can keep going in terms of these dimensions. I've just focussed here: when I was talking about US investors, I was talking about them buying European assets. But also, you can think about them buying domestic assets - which also shifts the relative demand. I'm going to skip that here; it's going to be too much for everybody to contemplate in a short presentation. So I showed you the picture, but you can actually be a bit more formal about this by doing what people would call a variance decomposition. What's a variance decomposition? It's saying, okay, well, we know how much total demand for foreign exchange there is, say, on a given day or a given month and so on. I just showed you, you can decompose it into all these different pieces. They all have to add up to that total. Well, how important are the relative pieces is the first question you should probably ask. You can do that using a variance decomposition. What you do in a variance decomposition is, I'll say let's just take the variance of all those individual components. That's not enough because there's also covariance between those different components. So then we have to add all of those things up. That gives us: how important are these different things in driving the variation? I'm not going to go through - this is a completely overwhelming table of numbers. I will say, what are the main conclusions that we get from this? There really are two. The first is that this return hedging piece is very important. So thinking about the relative performance of the, say, equities - especially in equities, to some extent in fixed income - is important for thinking about underlying drivers of foreign exchange demand. You can see that here in terms of the share here of the variance just driven by return hedging is enormous. The second thing I would show you is that there's this negative thing here at the bottom where I'm saying there's a negative covariance between the return hedge and the residual. What does this mean? Remember, I'm going to be doing everything at the monthly level and so that means if I think you should be short $50, let's say, because you just took on a $100 Nvidia position and you wanted to be short with a hedge ratio of 0.5. But let's say it takes you some time to do that, so maybe it takes you two months to get that hedge in place or three months to get that hedge in place. Well, it's going to look like today you have a zero and tomorrow you have minus 0.5. So I'm going to be giving you a big negative residual here and you're going to end up finding this negative correlation between the return hedge and the residual, driven by the fact that all of this behaviour plays out over time. You see this very consistently in the bottom, which is to say that not just with the return hedge but also with the flow hedge, it just takes time for people to implement this. In other words, this generates a little bit of predictability in the underlying foreign exchange demand. Those are the two main things to learn here. The other thing that is very natural to look at is monthly autocorrelations. So how persistent are these different pieces? Now, if you're a trader why do you care about autocorrelations? Autocorrelations also generate potentially predictability in returns, so that's one of the main reasons that traders like to think about these things. I would say that the main thing to emerge here is that the flow hedge piece, the flow-driven piece tends to have some persistence. Why is that? It's really because the underlying flows actually have persistence to them. In other words, when European investors are buying dollar assets today, they tend to buy them tomorrow and next month and actually over a period of time. That doesn't happen instantly. As a result, to the extent that they're also doing hedging of all of those positions, it generates this predictability and persistence in the underlying foreign exchange demand. You see that it's stronger in equities than it is in bonds, but there's a little bit of it in the bonds as well. Another interesting question is - and this is when the Europeans are buying, are the Americans selling? Where is the liquidity coming from? The answer to that is: on average, yes, when Americans are buying, Europeans are selling and vice versa. That's this negative correlation here between European flows and dollar denominated flows in total. It's modest; it's minus 20 per cent so it means that, roughly speaking, 20 per cent of the time when an American is selling, it's a European buying and so on. Interestingly, a huge driver of that is this return hedging piece. In other words, a huge piece is actually relatively mechanical - and that's pretty intuitive. It's because when dollar assets are performing well, say, relative to European equities. What does that mean? It means that drives the Europeans to sell dollars and it drives American investors to buy dollars back. So you get this natural offset that is a liquidity provision in the FX markets and it's really driven mechanically by this hedging behaviour. That's actually much stronger than any of these other pieces in driving what's going on. If you look at just the discretionary things like asset flows, it goes in the opposite direction. So it doesn't tend to be true that Europeans are providing liquidity to dollar investors and vice versa. Again, another set of a huge number of numbers. I've showed you these contemporaneous correlations. You can also look at these correlations with leads and lags. In other words, when A buys today, does that predict B buying tomorrow and so on in these different components. We've done that here. You can see that because you have a huge number of components, that can quickly get complicated. I'll just emphasise one thing that you do learn from this, which is that the hedging, the residuals are negatively correlated with each other, but positively with lagged return hedges. What does that mean? That just means that when you look at return hedges and flow hedges, they all take place over time. So it creates this predictability from returns ultimately to the FX flows happening over the course of a month, two months in the future and so on. I'm running out of time here. Let me just show you two more sets of quick results. You can also look at the relationships with returns. This is looking at the European - the return of the euro relative to the USD just in spot terms. Here, we're looking at on the equity side you get the largest correlations here, so in the range of 0.1, 0.2 with these different components. This is doing all of that in a regression from, so trying to understand ultimately what is most correlated with contemporaneous and future returns? You can see that these decompositions essentially allow you to explain a fair amount more of what's going on in terms of in an R squared sense. Understanding the ultimate drivers of returns. We've also done a little bit of work trying to understand whether these generate predictability and trading strategies. I'm going to spend not much time on that today. Alex has some other work on this topic. Essentially, what you find is, you don't find very much in the immediate T to T+1 but as you start to look out… I'm just showing you the T+2 returns here, but as you go out T+2, T+3 and so on, some of this predictability and underlying FX demand driven by these return and flow hedged pieces actually generates some fair amount of the momentum effect in currencies that many of you know about. I think given I'm running out of time, I'm going to stop at this slide, which is: I think there's still work to do here. Ultimately, what I've showed you is that understanding the hedging behaviour of real money investors can really give you a bigger insight into what's driving demand in the foreign exchange markets. I think the next step is really a deeper dive into, ultimately, which of these… I showed you some of this with the returns but ultimately, as you add this up - not just with Europeans and Americans but across all domiciles - what are the ultimate drivers of liquidity? Who is seeking liquidity, providing liquidity and so on? These are two pictures from previously-presented work from Alex in some of these research retreats. One of these around G10 FX rotations and another one on DXY timing that's using the direct indicator versions of the State Street data. We're not using the indicators in our paper here that I showed you today. We're using the underlying raw data and then building it up. But some of the portfolio strategies that have been talked about in the past use indicator versions of these different flow measures. Really, I've pushed the time here a little bit, but I think I've got time for one or two - or maybe three questions. Yes?

M: Thank you for a great presentation. I wanted to ask you about: all of the banks now have month-end models for dollar demand based on this idea that if US equities do well, then people will be selling dollars. Everyone tried to play that at month end. Actually, in fact you have the corporate month end and the month end at 4:00 pm - which is quite a lottery these days. So you mentioned the timing of these things can take time, but the industry seems to think it's literally done within 15 minutes and not so much money is behind that. None of it, by the way, is making any money when you look at these factors. So I'm wondering if you guys looked at specifically timing around month end, the 4:00 pm fix and applying this in trading strategies.

We have not. What I can report is just that when you look at the underlying flows, it takes time. So I think maybe the right way to think about this is that maybe 80 per cent of it happens very predictably right at month end. Then there's a lot of arbitrage activity around that. I believe that that is true, but then there's another 20 per cent that just takes its time, has maybe a looser hedge ratio and ends up happening. That is certainly what we see in the data. Now, how that plays out exactly in the returns, that I can't tell you; it's not something we've studied very carefully. Again, everything here we've looked at is in just trying to understand them at a monthly level, the returns. But I believe both of those things could exist in the data.

M: Many hedge funds nowadays are looking at cross-asset correlation with dollar index. Sometimes correlation comes and goes: 20 days, 5 days, 252 days. When you've got the effect of a positive euro dollar correlation with the US dollar equity assets, by selling a dollar you are enhancing this correlation; hence this positivity of, in this case, long US equities. Have you done any studies on this at all?

Well, I would say that first picture that I showed you… I've got a different slide here. Remember that first picture that I showed you between the correlations and then the average hedge ratios? I would say that's really the only thing that we've done on this topic but if you - which is to say that it's very suggestive of hedging modulating this correlation between the asset flows and the FX. There are other intuitive things, like if you have asset flow but it's happening disproportionately in equities versus fixed income, we know equities are less hedged. That's going to generate higher FX correlations than if those flows are coming predominantly in fixed income, which tend to be hedged. That type of thing, so I think we have those intuitions. There is some suggestive evidence, but we've done these things; they're based on very long-time correlations. What you're referring to is temporally, like this month versus next month, that type of thing. I'm very sympathetic to the idea that that's true in the data. We haven't looked at that in the data ourselves. Michael, you've got to…

M: I can't let you escape without asking you about bubbles.

Yes.

M: What are you thinking right now?

We're getting close to the final stages. If you think about your bubble checklist…

M: I know. That's why I'm asking.

If you think about the bubble checklist, there was this one… I often give a talk on bubbles. You think about the different components. I'm going to focus on US equities for now - but it's not just US equities. We've had high valuations for a long time. We've had a lot of speculative investment. At the end stages of almost every famous bubble - like 1929, 1999 and so on - we've had this massive wave of issuance to the public. We are getting all of these announcements now. The thing I would tell you about that, that I think is interesting: when you think back in '99, you had lots of issuance but really the problem started when the lock-up stared expiring. So there's been all this press. This is where I think the press is getting it wrong and it's very focussed on who's going to get to market first. Who cares? What matters ultimately, these guys are only listing - OpenAI and all these folks, Anthropic - tiny fractions of their float. When the full float is released, that's going to be when we're going to start seeing some of the problems. Actually, I guess in pure, real bubble terms this is not investment advice, but I'm still actually pretty… What is a bubble, by the way? How would you define a bubble? I have this debate with lots of people. I think a bubble is when the near-term expectations are probably still quite positive, but the long-term expected returns on the asset are negative. It's one natural definition of a bubble. I feel like that's a place where we are right now. You're also seeing some of the other type indicators; name changes, folks adding .AI to their names. There are famous examples of this back in dotcom. There were 75 companies that added .com to their name - and the average return to adding .com to your name in 1999 was 75 per cent, measured over a 2-month horizon. Now, you're not seeing quite that today, but the other thing: you're seeing lots of volatility. You've got a lot of the different markers here.

M: Well, hopefully we can get you back to talk about it next year and the bubble [?would've burst 0:38:31.5].

Well, it's not exciting to talk about it after it's happened, but yes.

M: Exactly. Thank you, Robin. Thank you.

Good morning, everyone. It's a pleasure being back here. I was here a couple of years ago. I want to talk to you about something that I think is very important. Very important. I've been in the market for almost 25 years now. We used to think that the market is driven by factors, size, momentum, value, idiosyncratic volatility, quality. At this point there's something called like the factor zoo, anyone familiar with that term? I think it's more of a jungle at this point. I don't think that's how the market works. The market works by narratives. Think about the last couple of years, it was all narrative driven. Does that make sense to everyone? Do you guys agree with me? It's not about size, value, it's about do we have military escalation? Did we have trade tensions? Are people fearing about inflation? We had a pandemic. All of these are narratives, they're stories, okay? That's what really drives the market. Traders wake up in the morning, they think about what's the next story, they don't really think about the factors. The problem is that it's difficult to measure this. How do you measure these intangible stories? So the whole point of my discussion today is really about quantifying narratives. I'm going to show you a specific application to central bank tone. What's really overarching this whole discussion is really narratives. What I want to do is I want to bring narratives to the investment community. Think about these borrow factor models, I want to do the same thing with narratives, and I think that's how the market works and I have a lot of research to show that that's, indeed, the case. The question, again, is how do we measure it? Our solution is really technological. The way we think about narratives is we look at the media and we see, what are people talking about? That's what we're trying to do. Now, to do this effectively you really need to look at a large set of data. So what we do is every day we created a machine that every day goes to the media, it goes to online and scrapes digital media. We started this about 15 years ago. Point in time, every day we go and collect almost a copy of the Internet. We're talking about almost one million articles per day. Then we see what people talk about. The way we quantify a narrative is very easy. Say you're looking at trade tensions, you have a million articles per day, just compute what's the per cent of articles that talk about trade tensions. That could be ten per cent one day, can be five per cent another day, six per cent, maybe jumps up to 20. Then you have the Rose Garden, what was it, April 2nd 2025, suddenly it's up to 50 per cent discussion. So by quantifying the narratives this way, we can have a good quantitative measure of telling us what people care about. Does that make sense to people? We're trying to understand where is the attention? We're able to quantify that. The implications of that, there are a lot of implications. There are implications for risk and there's implications for alpha. Why alpha? Because people have limited attention, right? How many things can you really think about? I wake up in the morning, I get the Wall Street Journal, I maybe have five minutes within the whole hassle of the kids in the morning. I have five minutes to look what's going on. I can maybe pay attention to three or four topics, that's it. Our system looks at more than 1000 different narratives, it looks at it systematically. So at any given point of time I can see what's bubbling. I see the discussion that's bubbling and I can see in 2021 people are talking about inflation. You have limited attention. Inflation only realised 2022. People were talking about AI in 2023, it took a long time until it bubbles and now everyone talks about it now. I'll show you some slides about it today, okay? So what I'm trying to do is to measure economic narratives, and I'm going to show you that they're important. So I'm going to show you three things. Two is too little, four is too much, so I'm going to show you three things. One, I'm going to talk to you about narratives in general, okay? So I'm going to show you how we quantify economic narratives and I'm going to show you how we create it with State Street. On State Street Insights we have a portal that shows you, what are the narratives that are floating around every day? Second, I'm going to take this idea of narratives and I'm going to focus on central bank communication. I'm going to look at central bank coverage in the media and I'm going to try to help understand what's the tone of the different central banks around the world, and I'm going to show you that it can predict yields. Third, I'm going to take a specific narrative that has been bubbling over the last year, which is Fed independence and Fed credibility. I'm going to show you that by quantifying this narrative we can even get a better measure of predictability for yields using central bank tone. Okay? So that's the plan. Let's start with the data. I mentioned we create a copy of the Internet every day. More specifically, this is point-in-time data, we've been collecting it point in time since May 2011. We have more than 150,000 sources, at this point it's closer to 200,000 sources every day, about five to seven million articles a week. Which means one million articles per day. When we look at this media, there's a lot of sources. So first thing just to help people understand a little bit better, we classify different sources into different media reservoirs, we call it. So there's general press, and there's Patient Investor Magazine, Absolute Return Magazine, these are information comes from the industry. There's press releases that come from companies. There's social media. There's trade magazines. FX traders talk a lot about what's happening in the world all the time. So we classify these - every media source into different reservoirs of information. Another reason we do that is because the media inherently has some bias. Does that make sense to people? You guys agree that the media is biased, is that's something completely surprising to you? It makes sense, right? The issue is how do we correct for these biases? So we do a lot of work in the background. I'm not going to have time to get into all the details, but I'll just give you a few examples of things that we correct for. General press is typically negative, does that make sense? You open up the Wall Street Journal every morning, it's always negative. Always negative. There's a war happening, there's fraud, whatever, it's negative. Why? Because negative news sells, okay? If you've ever lived in Manhattan, and I've done it a couple of times in my life, look, the media, they get how people think. What you don't want to see is your neighbour from the penthouse with a key to get in, or now it's the card to get into the suite, in the penthouse suite. You don't want to see they're doing well, you want to see they're crashing and burning. You're not going to say it, but the media gets it. Negative news sells. On the other hand, information coming from the company press releases are typically positive, right? Information comes from us, we're in the industry, we're typically portraying an industry in a positive manner. So what we do is we correct for the overall tendency of each source to be either positive or negative. We're going to look at the sentiment and I'll talk to you a little bit about sentiment. We can correct this with a three-month window, so that at any given point of time we'll look at whether, oh, is the Wall Street Journal more negative than average for this article? Okay, so that's how we correct for that. There's a lot of other biases. There's geographical biases. When Microsoft is mentioned in Seattle. So I've lived three years in Seattle, no one in Seattle is allowed to write anything negative about Microsoft. I'm just telling you. No one in Seattle PI writes anything negative. If you want to see a little bit and understand a little bit something more impartial about Microsoft, go to New York Times. So there's a geographical bias. Another bias, by the way, the one bias that I love the most, because I didn't expect this, is article length. Short articles are more informative than long articles, because short articles, their journalist is forced to come up with a view immediately. Long articles, on the other hand, the journalist has enough time, enough place to write on the one hand, this, on the other hand, that. If you graph the distribution of sentiment of long articles you can see it converges to neutrality. So we actually need to multiply by the squared number of words to get a distribution that you can compare short articles to long articles. It relates to central theorem, if you remember that from statistics class. So all of that is in the background. You have to correct for these biases to get something more meaningful when you aggregate all the things together. Then we use this data to create different indicators that you guys can use. We create indicators for single meme equity, we create indicators for macro effects, we create indicators for central banks, which is what I'm going to focus on today. We'll also have different indicators for thematics. Again, I'm going to show you part of this today. So that's the data. So let's start talking about the three things that I'm going to cover today. First, let's talk about narrative economics. What I did here is I plotted four series, if you look at the graph on the top, there's four series that are graphed here. Just showing you how we quantify narratives. The vertical axis is the per cent of articles on any given day that talk about the subject. You can see COVID-19 is a narrative. The discussion in COVID-19 back in March 2020 was 75 per cent. 75 per cent of all articles talked about COVID, does that make sense to you? That was the topic of conversation. You can see how that went down over time. The graph on the bottom, by the way, opens up the discussion of COVID for the different reservoirs. I think it's pretty interesting that you see the FX reservoir is the one that you see a lot of - there's an increase in discussion, in intensity of discussion, from FX-related sources. You see that? That's because traders already talked about COVID or some pandemic happening mid-January 2020. They were first to the game. They talked about, oh, there's something pandemic happening somewhere in the world. FX traders were already talking about it and our indicators already found that. After that there was corporate, corporate start talking about it, and then general press, and then like a lot of things we see, political venues start talking about it the last. They're the last to the party, but then you can see at some point they start talking about it a lot. Especially in the US, it was a very divisive are you going to take the shot, not take the shot? So that became like a political thing. Going back to the graph on the top, let's look at other narratives. So there's look at inflation. I really like the inflation narrative, because you can see that there's discussion on inflation all through 2021. The only material is later, right? In 2022. January 2022, that's when we start collecting information, we identified the inflation is happening, but that already came in discussion before. Armed conflict, you see the spike February 24th 2022. Russian invasion of Ukraine, you can see the spike, and you can see how it attenuated. By the way, there's spikes after that. You see the Middle East conflicts? They don't receive the same worldwide attention as the conflict here in Europe. Even the most recently you can see the jump over there, but it's not the same. Again, the nice thing about this is you can take this quantitative measure. We're all quants here in the room. What we can do is we can take changes in this measure and start calculating betas. You can start calculating exposures to narratives. You no longer need to think, oh, what's going to happen if there's a war? What's going to happen if there's trade tension? Well, here's what happens. You take the series that we have here and we do that. We calculate exposure, stake stock returns, stake commodity returns, stake FX returns, regress. Is it okay to say regressions here? You regress. I mean Europe, typically that's okay, everyone has a master's in quantimetrics. You regress on changes in the narratives and you get exposure, and that immediately gives you a view. The last one here is trade tension. You can see how I mentioned the Rose Garden speech, April 2025, you can see how there's a rise in discussion of trade tension, that also happened in the first Trump administration, 2018 to 2020. You see the bumps there too. Okay? So here, what I'm showing you is really how we can measure these narratives. We can know it's no longer these elusive intangibles, we're able to bring it to a place where it's tangible, it's quantifiable, measurable, you can use it for understanding risk and also alpha. I wanted to show you AI, we're talking about AI. So here's the AI narrative. Here's the discussion of AI over the years, you can see how it's pretty clear that after 2022, with ChatGPT, there's an increase in the intensity of discussion on AI. There's more discussion in corporate than just the general press. If you look at the bottom graph, not all AI is the same. When you think about AI, you can go more granular and you can ask, well, is the discussion about chips? Is the discussion about energy usage? Is the discussion about data centres? Turns out, most of the discussion is about data centres. Again, this is a way for you to understand, this is a way to quantify the attention to the different facets of AI. So what we developed with State Street is we developed this dashboard that you can actually observe narratives as they happen and understand their exposure to different financial assets. What you see here is I took a screenshot of State Street Insights and it's a two-by-two. Let me explain what this is. Let me explain how to read this. Vertical axis is the intensity of discussion, this is what I mentioned, the per cent of articles talking about every topic. Horizontal axis is the regression R-squared. In this case I took the US equity, I took weekly returns of SPY and ran a regression on changes in weekly intensity of each narrative. So each narrative point here is a dot of the amount of discussion, how much are people talking about it versus how much does it affect the market. That's the two-by-two. Think about the different regions here. When you look at the bottom left, this is a region where the narratives that are there, these dots, the narratives that are there are narratives that no one's talking about it and it doesn't really affect the market. That's typically where narratives start. Then they typically move to the right. Moving to the right, bottom right, means there's not a lot of discussion, but it's starting to affect the market. Typically this is when managers, this is when us, as an industry, we're starting to price the narrative. Then it goes up, okay? Then we're in the top right. The top right is there's a lot of discussion and it's being priced. So this is when the media picks up on it, you start seeing it everywhere and it affects the market, high R-squared, in a lot of discussion. Finally, in the lifecycle of a narrative, typically it goes back left, the media are still talking about it. We already priced it, but the media is still talking about it, and we moved on to price something else. That's a typical lifecycle of a narrative. So look at AI. Right now AI, and this is from a week ago, or a couple of weeks ago. This is AI, a lot of discussion and it's pricing the market. When I click on it, you can see the bottom graph shows you the time series of that. The dark line, the solid line here, this is the amount of discussion on AI in the last year. Does this make sense to you? To me it actually makes sense, it's what I observed. When you look at last year, we started to talk about AI, here it's almost ten per cent, and then over Q4 a lot of discussion. We had an AI meltdown here, but still people, a lot of people are talking about it. The interesting thing here is the dotted line here. Can you guys see the dotted line from the back? The dotted line is R-squared. The point is that even though there's a lot of discussion on AI, that's not what's driving the market in any given point. Of course, in Q4 we know it drove the market and that's where we saw this in our AI meltdown. People are still talking about it, we moved to other narratives, we had the war in the middle, right? Now people are talking about it again. I mean now AI is pricing the market again. You see that? So the nice thing is that you can see in any given point what's really driving security prices. Does this make sense? So you're going from the top right, top left, you're moving between these two. That's what this graph shows. Are we good so far? Any questions on this? The nice thing about this, it's updated every day, refreshed, there is a bunch of narratives here, there's a drop-down menu. You can look at US equities, you can look at the dollar, you can look at other factors, you could look at treasuries. Then what you have here is popular narrative, but there's macro narratives, there's micro narratives, there's other sets of narratives. Every time you use the drop-down menu you can see a different graph. You click a narrative, you see the time series of it. So that was narrative economics. I can tell you the way people use it, again there's two ways, but I think the way that most people use it is just understanding risk of the different assets they have. You can apply it to commodities, you can apply it to FX, factors, sectors, whatever. In particular I want to take a deep dive to central bank. So I'm going to create narratives around central banks and I want to understand the monetary tone. What I want to understand is the central bank becoming more hawkish or more dovish. That's what I'm going to do. So why do I think the media is a good avenue to talk about central banks? There's two things that have been happening over the last several years. One is central bank communication has become very opaque. Does that make sense to people? There's so much scrutiny on the words that are being used by central bankers that they programme this so much, they think about it so much, everything is planned, that the communication becomes worthless. Because you can't say anything. So then we turn to the media to interpret it. The other problem, of course, certainly in the US, they meet eight times a year. So what happens between the six weeks? Between the two meetings, rights? When you look at the media you have a score every single day. So what I'm going to do is I'm going to look at every day, I'm going to look at the articles and talk about the Fed, and then I'm going to extend it to 15 central banks in the world. I'm going to specifically first look at the Fed, look at all articles talking about the Fed, and then check are the articles talking about interest rate increases or interest rate decreases? Is it talking about tightening? What is it talking about? So I'm going to create this measure that if it's talking about interest rate increases, I'm going to call it a plus one, increase rate decreases are going to be a negative one. So then I can create a measure that you can take, say, all the communication that suggests that the bank is going to be more hawkish, minus all the communication that they're going to be more dovish, divide by the sum, you get a normalised measure between one and minus one every day. The graph that you see here, I took this from Bloomberg in 2017. I'm going to show you an updated version of this, but they looked at the central bank governors and they looked at the correlation between their height and the average rate. They tried to predict, in 2017, what happens when Jerome Powell is going to become central bank, the chairman. So they're predicting, well, he's taller, maybe rates are going to be higher. Now, I think we're going to come up with a better measure and I'm going to update this in a minute, you're going to see how it looks like. Everything I'm talking about here, we've published papers, I'm going to show you some updated data. We've published three papers, we're working on the fourth now, that shows you how to use these measures in a predictable way. So first thing is I wanted to show you how the different central banks look like right now, or over time. So you can see in the last six-seven years how do the three central banks have - the Fed, and the ECB and the BOE. How do they look like? Okay, so again, the number is positive, it means you're more hawkish, negative means more dovish. All of them have the same pattern, I don't think that's surprising. At least it's interesting that the way we measure it conforms to what we think should happen. What I'm going to show you, the interesting thing is what happens when you start measuring different people and you start looking at differences between the chairman of the Fed and the other voting members, okay? So this is how it looks like right now. Let's start looking at some interesting differences. So we look, for example, here on the left, look at this is how Jerome Powell looks like, the chair, versus the other voting members. So you can see sometimes the chair, according to the media, is more dovish than the other members. Sometimes they're more hawkish. See that? Interestingly, this suggests that there's some disagreement on, therefore, the committee. We're going to use that. When there's more disagreement, there's more uncertainty. That could predict higher yields to compensate for that uncertainty. That's one of the results we find. You can see how it looks like the ECB, and you can look also at the Bank of England. I want to show you some results on how we can use this to predict yields. So what we've done is we've collected all this media information, and I'm going to show you after we compute these measure of hawkish-dovishness, we're going to regress it on - we're going to take yields and regress it on these measures. I'm going to show you that it predicts yields. This is a graph, this is what happens when we added Kevin Warsh to this system. You see that? So this is how I did this. I put the graph, the initial one, into ChatGPT, I added the correct height of Kevin Warsh, and I asked ChatGPT to reproduce the figure with Kevin Warsh, okay? So you can see how that looks like. He's actually pretty tall, I didn't realise that. Here's what happens when you compare Kevin Warsh to Jerome Powell. So look at the middle section here. Instead of looking top down, looking at articles that talk about the Fed and whether the Fed is mentioned in a positive or hawkish way, or dovish way. I can look at the particular person. The way we do that specifically, we look at the mentioning of the person, the individual, and then we look at the words around the individual, so that we don't capture the entire article. Just around the individual, to try to target whether this individual is portrayed more hawkish or more dovish. What I think is really interesting about this graph in the middle is look what happens, this is over the last maybe year-and-a-half. So look at the fall of 2025 when there's discussion about Kevin Warsh, and he's trying to get the nomination. Look how he's becoming more dovish. You see that? He's becoming more dovish. Jerome Powell actually lags him, he become more dovish after. After Kevin Warsh officially was chosen, look how now he's becoming a little bit more hawkish. I think that's pretty interesting. I'm sure you're going to write about this at some point. Jerome Powell is becoming more hawkish too, but actually Kevin Warsh is the only one who actually has a positive score. Which goes against what you would think, but when you look at the data, it actually suggests that he actually is more hawkish than everyone else. So when you compare Warsh to other voting members and regional and Powell and other FOMC members, he's actually the one that is the more hawkish. At least the data provides a little bit of a different view and this is looking, again, at Powell, and Kevin Warsh, and then in the middle graph, the middle line is all the average of the other voting members on the FOMC. Again, this is point in time, because there's a rotation, but this is at any given point in time you just look at the voting members and this is what it looks like, okay? So they're somewhere in the middle. So to show you that this is actually - that was just the data, but to show you that this is useful I ran some regressions. So I took this from the paper. Let me explain to you what I've done here. We looked at a time series of changes in yields, okay? So we looked at you have here the one-month yield, and the three months, and six months, and you have T plus one until T plus six, this is measured in weeks. Typically we do like Friday to Friday. So what do we do? We look at a given week, Friday to Friday, look at the average tone of the Fed. So this is top down, you look at the Fed articles, whether it's hawkish or dovish. You calculate it every day and then you average over the last seven days. Then you look at changes in the yield from the end of that week to the next week, and then the following week, and the following week. What is interesting is you can see how the tone has a positive and statistically-significant coefficient. That is to say in weeks when you observe the Fed has become more hawkish, typically yields go up in the next several weeks. When you interact this measure of hawkish-dovish with the FOMC meeting weeks, you can see that let's say in the three months or the six months, when you add interaction term, part of this effect goes away. Because it comes, you see, with a negative sign. So what's interesting about this, when the FOMC meeting happens, it draws everyone's attention. Everyone is looking at it and so there's no predictability, everything is priced the same week. When people don't pay attention to it, that's when you start seeing the predictability. Similar results are found when you look at other central banks. So again, this is a time series per central bank. We did some experiments with - we did our own bag-of-words approach, we also used LLMs. Overall, the result here suggests that it doesn't really matter how you - which LLM you use and which approach. All of them are going to give you a positive and statistically-significant coefficient. So we took an example, let's take the two-year yield. When you run regressions of the hawkish-dovish on yields, on the two-year yield, it doesn't really matter whether you use generation one, Roberta, ChatGPT, at this point we actually looked at ChatGPT-4, or Llama-2, all of them provide statistically-significant coefficient. We produce all of these measures on a daily basis, it goes to State Street servers and you can download from there. We actually recently updated, you can get the Gen1, you can get the ChatGPT-4, all of them are available for you. The message here really, and I've been doing this for almost 15 years, it doesn't really matter which approach you use to calculate the sentiment. It doesn't really matter. When you look at the predictability there on the bottom here, you can see all of them have a positive predictability on future yields. When you look at overall predictability of rates using either the overall Fed versus just the voice of the chairman, versus sigma, which is the variation, disagreement across the other voting members. All of these have effects on yields. Just looking at the top down explains about 50 per cent of the explainability comes from there, and then about 25 per cent for each the individual voice of the chairman versus the disagreement across all the other voices on the committee. So that was just the Fed, but I can also look at cross-sectional variation, cross-sectional strategies across 15 central banks around the world. So what I've done here in the top graph, every week I calculated the hawkish-dovish indicator per central bank, and then what I did, I created this long-short strategy, just to show there's predictability. You buy the top, the hawkish yields of the sovereign bank related to those central banks, and you short, you minus those that have a dovish tone. That spread, if you look at the cumulative numbers, that spread is significant over time. The top one has used bag of words, the bottom one has used ChatGPT. So it doesn't matter which way you use to calculate hawkish-dovish, you see this predictability. So the point here, really, is not only time series it works, it also works cross-sectionally. If you want to understand, if you're interested in central banks around the world, the hawkish-dovish indicator could be very useful for you, because it could predict yields up to a month or six weeks after the measurement. So, so far I covered narrative economics in general, specifically about central banks, and now I want to add something more topical, which is the Fed independence and credibility. So let me dive into that particular narrative. Here's what we've done. We looked at the articles, the talk about the Fed, and then we also looked at them and said, how many of these articles are talking about Fed independence? What I think is really interesting is look what happened over the last year. Suddenly you see a huge jump in the amount of discussion, in the intensity of discussion on Fed independence. This is now top of mind. People really care about this. They care about it more in the FX reservoir. So these are traders, they're talking about rates around the world, they're really worried about the Fed independence. What I'm going to try to do now is create a measure, not just independence, but overall credibility. So here's what we did. We actually - we're not the first to do this, but we used our data and we followed the measure of central bank credibility by Aikman, and we did the following. We took an article that talks about the Fed and then we asked ChatGPT, we asked it, 'Does this article, is it expressing the Fed's integrity, competence and credibility as either critical, neutral or supportive?' In other words, here's a few examples. Here's the top one is Larry Summers saying, 'Stephen Miran, I heard his speech, it was really bad.' That's going to be a negative one on credibility. Then you have all this thing about the mortgage, potentially mortgage fraud of Lisa Cook. Again, that's going to be a negative one in terms of credibility, okay? So we're going to calculate, again, just the same way we calculate hawkish-dovish indicator. We're going to calculate here all the articles that are positive, or not positive, but suggesting that the Fed is credible, minus all those that suggest the Fed is not credible. This minus that, divided by the sum, is going to give you a number between minus one and one. Here's how it looks like over the period, we did both for the Fed and for the ECB. When you look at the last year I think it's pretty cool. I think it's consistent with my prior that the Fed seems less credible than the ECB. Okay, if we just look at this last year, that's how it looks like. There is a time variation, credibility over time, so this is kind of interesting, but I'm going to use this for the regressions. If we just look at the last several years, I think it makes sense. So I'm going to take the same regressions I showed you before and then I'm going to interact them with Fed credibility. The interesting result is what you see here at the end, beta-seven. So this is a regression with seven variations, but I think I just want to highlight here this last one. What is this beta-seven? I'm taking yields, and I took this yield a month out from T to T-plus-four. I regressed on the tone is what I showed before, that's hawkish-dovish, and then I add the interaction with low credibility. Low credibility are all the bottom 25 per cent of weeks in terms of credibility score. Also interacted with FOMC meeting weeks. That interaction, for the one-month treasury is positive, but for the ten-year treasury is negative. What does this mean in terms of economics? Here's why I think this is super interesting. Let's think about it this way. When the Fed is deemed less credible, the way we think about it, at least the measure it seems like today. When the Fed is saying, 'We're going to cut rates,' so we're going to be more dovish, the one-year response has a positive coefficient. So when the Fed is more dovish, the one-year rate is going to drop, the one-month rate is going to drop. Investors understand that they're doing it for the wrong reason and maybe that's going to bring inflation in the future. So the ten-year treasury actually has a negative coefficient. So when the Fed is saying, 'Okay, we're going to reduce the rates,' it has a negative effect. So long-run rate's going to go up. So this is consistent with the market understanding. Then when the Fed is less credible, when they're saying we're going to drop the rates, it signals to people inflation's going to come because they're lowering it too much. So that economic interpretation that I think strategists talk about, you have this in the data, it's right there. I think that's the most interesting result from this whole research that I've showed you, that coefficient there. It's hugely significant, and this is measuring - it's a really, really large number. We did something very similar with funding. Look at so far, minus the two-year rate. I'm not going to talk about this a lot, but you see similar results. There's statistical significance here. So what's going on here is when the Fed is becoming either more hawkish or more dovish, it affects the entire yield curve. That difference between the funding rate and the short-run funding rate and the two-year yield, that gap itself increases with respect to the tone of the Fed and its credibility. When the Fed is less credible, that gap becomes even larger. All right, so let me summarise. What I tried to show you here is a very different approach to what at least I've seen for the last couple decades. What we're trying to do is understand the narrative. These are stories, these are thoughts that people talk about, right? I don't know, at least my circle of friends, we don't go out Friday night and talk about, 'Oh, did you see this value factor today?' Maybe last week we did talk about momentum factor, the momentum crash, but normally we talk about the narratives. This is what's happening in the world, we're going to have more taxes, there are going to be tariffs, there's a war, that's what we talk about. How do you translate that into investment decisions? First of all, you need to quantify the narrative, that's what I've showed you how we did it. The way we did it here is by looking at the media worldwide. Then I showed you particular application for central bank communication or for central bank tone. Then I also showed you what happens when you interact it with credibility. Questions?

So Kevin Warsh, one of his plans is less communication from the Fed, how do you think that's going to impact the research you're doing here?

Oh, I think it's going to be much more impactful.

Yes?

Because this is the whole point. When the Fed is not communicating, people want to understand what's happening. So you're going to look at media interpretation and I think that's what's going to affect people's minds. So I think these indicators are more valuable when there's less communication, absolutely. You're going to have a lot more to talk about over these indicators. Yes, please. I'm sorry? Go ahead.

So does the market make sense that it's pricing too much rate increase? What do you think?

Well, I don't have it here, but I think you could look at - we have USD, in State Street, in the portal insights, we show, and I looked at it a week ago, how certainly AI, what was it, trade tension was really priced into yields and into USD. So you could look at that. I don't have it in front of me, but certainly that's the point of this tool. That you can see whether the R-squared is meaningful.

Understood, thank you, but what's your personal view? Now the US rates, at this level, the ten years was very close to five at one point, several days ago. Do you think it has been - gone too far? Do you think that the markets pricing in one rate increase by the end of March 2027, is it too exaggerated or is it completely making sense? So what's your personal view?

I really don't know. I think, look, a lot of things, when you look at what happened in the last three months with the war, everything a little bit went upside down, right? We thought treasuries are going to be safe, gold, it wasn't like that. So I don't know what to tell you. Things behaved very differently this last episode than previous military escalations, yes.

You showed us a chart where you compared COVID, tariff and the war, basically the war impact. I was wondering whether the spikes in those different charts or time series there are actually meaningfully comparable.

To each other?

Yes.

Yes, because all of them are looking at per cent of total articles. So you have the same base, it's all the articles that we collect. So the denominator is the same for all of them. Yes, so what you see here, well, we don't have it here, but it's true that COVID was the highest. Second highest over the last 14-15 years was the tariffs, that was the second highest in terms of it reached more than 50 per cent of the discussion. If you remember that Rose Garden event, everyone was talking about it and the next day everyone in Europe was talking about it. So this lingered for a while and it really captured a lot of people's attention. Yes, so in that sense maybe it is priced correctly. I think if you wanted an alpha strategy you need to look at the other narratives that people underreact to when one narrative is capturing everyone's attention.

How do you think that the diffusion and adaptation - adoption, sorry, of AI in society will impact the significance of narrative on the market?

Look, I think we have seen the last couple of years, and I think we're going to continue to see that the market is more narrative driven. I think the cycles though that narratives are moving around might be quicker. They might be quicker. The cycles that I've seen have been people have been underreacting to narratives over a period of 6 to 12 months. That has been shortened. People are more reactive, and maybe it's because we're better able to get data, etc. It's true that it used to be the market is talking about something for a month, two months, three months, up to six months. Now it seems like every month there's a different narrative. Again, the point of this is that it's measurable. You can look at it, at any given point of time, and you can see what's driving it, yes.

Just one thing, you said that you are screening, you found suitable one million articles per day, but did you ever try to assess which ones are the more impactful, the more predictable? Because in the end maybe only ten of them make sense to look at, you know?

Yes, it's a great question. So our approach has been more democratic, in the sense that we let every vote, every count, every media count. I can tell you a few things about this. First, we saw - I've shown you that FX magazines, FX reservoir seems to be leading. That's the one we used for the hawkish-dovish indicator. Having said that, the power of this is really understanding aggregate discussion. When a narrative starts being discussed it's only going to affect the market when everyone is talking about it. The kind of thing is which media is right and which one is first, that's a different game. That's a game of either HFT, like high-frequency trading, which we're not doing here and I don't think any of our institutional clients really worry about that too much. So we're not really interested about who's first. We're also not interested in who's right. What we're trying to measure here is not what's correct, we're trying to figure out what is. I've done this for a while, I can tell you it's more important to understand what people talk about than how they talk about it. Whether or not they talk about it in a positive or negative way, that's a second-order effect. The question is, what are they talking about? Do you see what I mean? That's much more important than whether it's a negative or positive thing. It's almost like saying there's nothing like negative news, or bad publicity, it's just what are people discussing?

We've got to finish.

Should I take the last one?

No, we've got to go.

We're good? I'll be around, I'll stick around, if anyone wants to ask questions, I'm happy to. Thanks very much, everyone. Thank you.

Before I'm going to tell you what AI can do, I want to show you something that AI does. For those of you who manage assets, or risk managers, I think that should actually be quite worrying, I would say. Here are two sentences. Operating profit rose ten per cent to €566 million from €550 million. The second sentence: operating profit increased to €566 million, up from €515 million. It looks the same, right? Maybe the first sentence is something I would've written. The second sentence might have been a Newswire journalist, but let's ask what a large language model that earns its living by analysing financial news data would attach as a sentiment to these two sentences. It turns out that FinBERT, which is a modern language model that is trained on a domain-specific task - which is sentiment analysis - actually would classify the first sentence as positive and the second sentence as neutral at least. This is strange because we humans would attach exactly the same sentiment to this sentence. This sentence comes from the same company, from the same quarter. These are the same numbers and so on and so forth. So this example is actually the main idea that I want to showcase to you in the next 40 or 30 minutes. It shows that there are two sides of one coin of AI, which are fragility but also power. We have to be very, very careful in balancing these two sides of the same coin. So just as the agenda, first I want to make you a believer in AI so I'm going to start with the bright side on how AI can actually scale human capabilities. Then I want to pivot a little bit and I want to emphasise the illusion of understanding that AI usually has. Then I want to really go into the dark side and I want to showcase three or four fragilities that current AI systems actually have. Eventually I'm going to tell you what this actually means. Lee was saying this morning that you should have some takeaways from today's seminar. I think that the final takeaways - I put them in one slide for you - are actually your take-home exams or take-home tasks that you should really look at when you want to use AI for your daily work. In the beginning was the word - and then we tokenised it. So that's not exactly what the Holy Bible is saying, but this was something that really triggered me when I was looking into natural language processing because in finance, most of the information is actually in words. Maybe 80 per cent of all business-relevant information is in words and not in numbers. I'm going to show you one particular case that we looked at, at our lab at the University of Zurich. I'm going to just show you and try to convince you that using AI the right way can actually be something very meaningful. For those who know me, I do a lot in climate-relevant research so we wanted to look at physical risk in financial markets; how physical climate risk would actually impact companies. We were looking at almost two million SCC filings. That's a lot of data - and nobody can really read this data so you need to have a machine in the loop to analyse this data. In the end, these are almost 3.5 billion paragraphs that you need to go through. Now, what we did is we wanted to analyse how companies are affected by physical climate risks. The saying goes that physical climate risks will increase in the future and they will have actually a substantial impact on the economy and on financial markets. So how can you really measure that? The bad thing about extreme weather events and physical risks is, from a statistical viewpoint, they do not happen that often. So to make statistical inference is really hard because you have really a shortage of data. What we were doing is, we were taking the NOAA database that shows which natural catastrophes, extreme weather events had a substantial damage. We were using this information to map it to what companies are reporting in their 10-K filings and 8-K filings as well, so basically all the filings. We were training then a palette of different large language models because you don't always need the biggest model to do certain things. So we used a simple model to identify whether a company was actually talking about an incident related to the NOAA event database. Then we were using a model to analyse the textual content of the paragraph, whether it's actually impacted directly or indirectly through the supply chain. Eventually we were using a larger model to really classify the impact on the business of the related company. So it's basically state-of-the-art computer science that we did there. We were then looking at what happens with these firms. It turns out that if you look at the data over this period from, I think, 2010 until 2024… I'm going to quickly tell you why 2024; 2024 is the time when the NOAA database stopped. The reason is that, according to the White House, climate change is not an issue; it's just a hoax so therefore there are no data available anymore from the NOAA database. It's going to be a little bit tricky to extract this extreme weather event data from some other sources. Just to tell you a little bit what is happening. So we're analysing the data. We were looking at something which is also currently very en vogue in - for instance MSCI does a lot in this direction using satellite data, geolocation of the companies, and look what happens at the geolocation of the company. But we find actually that textual data is much more important for addressing or analysing the losses of the companies. If you just look at the geolocation of the data, you miss all these dependencies that are written down actually in 10-K or 8-K filings. Dependencies related to supply chain and other information. So you don't really see that from the sky. We showed that you actually are able to extract a lot of information from this huge amount of textual data. So this is really nice - and it has also economically-significant effect. So the cumulative abnormal return when an event happens is actually minus 2.36 per cent and highly statistically significant. For finance terms, that's actually a big number. You can also differentiate between different industries. For instance the mining industry was damaged the most - which is easy to understand - but also the construction industry was actually profiting from natural catastrophes, which is also clear if you think about it. But you can really analyse this data along these dimensions, so this is the bright side of AI. You can analyse an ocean of data in half-an-hour, so that's really good. Roberto this morning was really talking about the human capabilities that complement AI, or vice versa. Ronnie was talking about the signal that you can extract from the talk of central banks in order to extract the tone and then use that as a signal for making an investment decision. I was talking in the last five minutes about how you can use it to really analyse physical damage due to climate change from textual data. So these are all about signals, but what happens if AI can also have an impact on the quality of the signal, if it can actually fake the signal? Think about these two sentences that I showed you. One was positively classified and one neutral. We are academics; we like to break up things from time to time so I thought, that's an interesting research idea. Let's make an adversarial attacker that attacks the sentiment classifiers that are used in the finance industry, and think about whether it actually works or not. This is the pivot - the start of the pivot - that should make you a little bit aware of the deficiencies of large language models. I use Mary Shelley's 'Frankenstein' as a quote. 'We worked for years to give it speech, and only after it spoke did we ask what we had made.' This is really true if you think about… I work on AI as much as I do on finance - and probably a little bit more on AI - and we develop things. Once we develop things, we are so hyped because it looks so great that we forgot about thinking about how to evaluate what we have actually produced. This is happening also in the LLM space. What can go wrong with large language models? Well, of course it can read at superhuman speed so it can read three billion paragraphs in seconds. It can process unstructured text. So in the early days when we did natural language processing, we had to put a lot of emphasis on structuring the data. Now everything goes automatically as long as your PDF reader is actually a good one. It can match patterns across a wide range of data, but what LLMs cannot do - and even if you think they do it - they cannot do causal reasoning. They might sound like reasonable machines, but that's not causal reasoning. It's not causal identification; it's a causal assistant. It generates text that would appear to have some causal reasoning behind it, but that's not causal reasoning as we humans actually do. It's more like a politician, maybe, but not a true scientist. It lacks robust semantic understanding - and this is actually you who worsened the situation because in finance, you always come up with new terminologies. Think about the GameStop saga. During that time on Reddit and all these platforms, all kinds of new synonyms or acronyms and terms came up. An LLM cannot really understand these terms, so they're trained on past data and not on actual data. Of course, you can link it to a search engine like the Google search engine. It's really hard to do self-verification. You can ask an LLM, 'Are you really sure' and the process starts again, but in the end it's just a stochastic machine and it gives you back an answer. But it's not taking your request really seriously that it should double-check whether the previous statement was grounded. So these are all problems that you need to be aware of. If you use LLMs, do not treat it as an infallible oracle, so always be aware of the deficiencies. There's also literature; I think there is a very nice paper that used Boston Consulting Group consultants to test the power of large language models. They had to solve a task where the large language model was actually fit, was capable to do it, because it was part of the knowledge base of the large language models. Actually the AI users did 12.2 per cent more of the tasks that they were supposed to do compared to the control group that was not using AI. It was also 25 per cent faster, so more and faster, at higher quality - which looks great. But if you have to make a task which is outside of the model's capability, the model will not admit that, hey, I don't know. I'm not sure, because they're not trained that way. They're trained to please you, to make you happy. That's what we do at Google, so if you do the same experiment with a task which is beyond the capability of the AI, 90 per cent of the users were not correct anymore. Why did this happen? They also showed that the users, the most experienced users, were making these mistakes because they thought, well, the AI, it's producing a reasonable answer; I'm going to trust it. This is called the jagged frontier. So you don't know where the LLM is actually outperforming because you don't know what is going on inside the large language model. Now I'm going to talk about the dark side, so therefore abandon all priors you who enter here - according to Dante's 'Inferno'. Of course, I replaced 'hope' with 'priors' because not all hope is lost. I'm not an AI doomist at all because I make part of my living out of using and developing AI. So what is the dark side? First, going back again to these two sentences, this was such a fun project and also a nice title for the paper. It was called 'The Battle of Transformers'. What did we do? We were looking at transformer models for sentiment analysis, and FinGPT or FinBERT are two of the usual transformers that you make - that you use for large-scale sentiment analysis, whether it's on stocks, whether it's on monetary authority talk, whatever. We were creating an attacker transformer; that's why it's called 'The Battle of Transformers', based on GPT-4o, who should paraphrase the sentences in a way so that the sentiment analyser is somehow going in the other direction. From positive sentiment to negative sentiment, but as a human you would still call it a positive sentiment - but the machine would clearly attach a negative sentiment to it. We were using that on the standard datasets that you have in computer science and sentiment analysis. We found that actually 20-to-54 per cent of the predictions flip the sign. So it's not just positive to neutral, but positive; 20-to-54 per cent flip the sign. We are just a bunch of academics from Zurich, not even MIT. We created a monster, an adversarial attacker, out of thin air that would change almost 50 per cent of your sentiment analyser that you often use in the finance industry. It was kind of fun to do actually and so that's the first vulnerability, the first fragility that you should be really aware of. I was telling you the signal is important. Your business is in the signal business; you transform or digest signals into strategy. What happens if this signal is actually a fraud? This was the first fragility. The second fragility is: what happens if everybody is using the same kind of technology, the same kind of LLM in building in investment strategy? Everybody today, you use Claude Code or Gemini or GPT; these are the most powerful ones. Maybe you use Grok as well, so if everybody is using these tools, you can hypothesise - at least theoretically - that there will be some kind of homogenisation in the investment principles that you apply. The second thing that can happen is that, hey, even my students in their 'wohngemeinschaft' - in their apartment - can do something that Goldman Sachs was able to do ten years ago exclusively. You can analyse, as a student, all three billions paragraphs and do some investing, do some investing advice, so everybody can do that. So this is a democratisation process for investing. These are different things that actually can happen in financial markets. I did some testing on that. So far the evidence is sparse, but definitely something that is happening that is; analysts are much quicker in reading corporate reports, corporate announcements, 8-K filings for instance because they come at a surprise. The digesting of 8-K filings has been increased significantly since the arrival of transformers. So it was not conditioning it on the arrival of ChatGPT, but on the arrival of transformer technologies back in 2017. One thing about this signal is, I put there a formula; signal is equal to the true value - this is what we all want to extract - unique data because everybody of you is using maybe some unique database. But also there might be some kind of homogenisation in terms of the databases that you're using because there are companies that are selling this data. Noise plus a common AI error. So far, times have been quiet, I would say - although in the US there's this President in the White House - but times, at least in the financial markets, have been quiet. But what happens if everybody is using these large language models and if something unprecedented will happen? The LLMs will all go in the same direction because they don't have this training data to differentiate. So trades might be coming crowded and this is definitely a danger depending on how much AI is already determining the investment process in the industry. There is a paper actually in computer science that showed that the more skilful a large language model gets, the more correlated the errors will become among the large language models. Do you remember maybe the time when ChatGPT came out? We were using it for writing birthday cards; it was doing a great job so eventually, this pre-Christmas time was very efficiently handled because you can write Christmas cards, birthday cards and all these cards very easily. So this is a low-stake application, but with the increased skill of large language models, we now use these large language models maybe to manage an atomic reactor. So the stakes get higher and higher and we rely on these large language models. It turned out in this research that the more skill these large language models have, the more correlated their errors will become - which is scary, I would say. At the same time as the skills of the large language models increase, it's becoming harder and harder to detect the fallacies of the large language models because we are not experts in everything. But the large language model pretends to be an expert in everything, so we rely on that information without critically asking. We might fall into what other researchers call epistemia, so not epistemic foundation of our scientific knowledge but something that looks kind of scientific. That's a lot of danger and brings me to the third fragility, which is the knowledge problem. It's not your fault; it's probably my fault or the fault of my colleagues at university. How do we generate science? We generate science by writing papers. These papers will be picked up actually by you to do a factor strategy, if you're still in the factor soux era or do something else. So you pick up these papers, these scientific papers. But the more capable AI becomes, we have to ask ourselves: who wrote actually these scientific papers? Was it just AI that was used to generate - well, first of all a hypothesis and then writing the paper, doing some statistical analysis, writing the code for the statistical analysis? At university we need to produce papers; these papers are our currency. So what a great thing that is arriving here with AI. Generative AI, now I can write 100 papers in 1 year and I can flood the review processes at the big journals. This is very problematic because some of the research might actually not be good research. I tested this hypothesis in the following way. I was using from a computer science conference - one of the top three computer science conferences last year, last December - actually some data. So at computer science conferences you usually have 20,000 submissions; 10,000 submissions from China, 5000 from the US and the rest from the rest of the world. You have 75,000 reviews, referee reports for these papers. Last year ICLR was attaching a score to how much of the content was AI generated, both from the papers as well as from the reviews. So if you think that there is a sweet spot between using AI 100 per cent and not using AI at all, like a Markowitz Efficient Frontier of AI usage, I was able to test this with this data. Actually it turns out that even a slight increase in AI usage will lead to a degeneration of the quality of the paper. The human-AI collaboration frontier as of last year, or early this year, is still constructed in a way that degenerates the quality of the research. This is not research by philosophers or some other domain that probably won't use AI, but AI scientists that used AI to write papers. It turns out that the quality of this research is degrading right from the start. Also what turns out is that if people are using AI to review - because I get swamped by review requests and I just say no. I could say yes and just use AI to do a review. The review process, if you use AI, is such that it's more lenient. This is how AI is trained. LLMs are trained to be polite, so it would accept papers much easier than a human would actually accept papers. If the paper gets accepted then it enters the scientific knowledge. Then it enters the AI labs that train large language models. Then it enters your AI process that is used to make investment decisions. This has a big impact and you have to be therefore very careful in how you design workflows around human and AI collaboration. If you go down the naïve road, then eventually our knowledge base, our scientific knowledge - and I'm not just talking about AI knowledge, but in general - will actually run into the danger of degenerating. That would be really a big problem. Now, again to really keep the frame, the signal is so important. Signal can be reframed by an adversarial attack. The signal can be contaminated by AI that is doing this epistemia thing. The signal can become crowded if everybody relies on the same AI stack, but what happens if the signal is actually cut off? Everybody is using AI these days, so we use foundation models. Which foundation models? Apertus. That's a Swiss joke; Apertus is a Swiss large language model developed by ETH and we hope to make it great. But which foundation models are we actually using? We are using OpenAI, using Claude, using Google DeepMind so there are very few providers so you create a big dependency, a supply chain dependency for your knowledge processing. You rely on cloud platforms like AWS, Azure, Google Cloud; there are only three hyperscalers. You have AI accelerators like Nvidia, Google TPUs, of course, and you have some semiconductors that create - are at the start of this whole process. You have this ultra-extreme ultraviolet lithography in the Netherlands with ASML. So you create a lot of dependencies. The question is: what really happens in the financial markets if you have these dependencies? If, for instance, Joe Biden at the top of his cognitive capabilities in January 16th 2024, was actually signing the AI Diffusion Act. The AI Diffusion Act was determining which countries are problematic to send AI to, AI generated by or created by US companies. Guess what? Switzerland was in the same list as Iran and Russia. You have to be aware of these dependencies. So what we did in our research is we created an ADI - we always need to have the three-letter acronyms in finance. ADI is: algorithmic dependency index, so we're crawling the 10-K filings with some language models. We're trying to extract information about the dependency of the companies on the whole AI stack, going from the large language models to semiconductors and so on and so forth. We then were looking at whether the market is actually reacting to these dependencies or not. Then we were looking at around 13 different events, political events at the White House, what new regulation came out in terms of restricting AI, restricting Nvidia from selling chips to China and all these other things. By using the ADI index to categorise the companies into the different buckets, we found that whenever such an event is actually happening, when a new rule comes out there is a big impact in terms of cumulative abnormal returns in the company. So markets are reacting to this algorithmic dependency and therefore, it's probably an interesting measure to look at in the future to design hedging strategies when a certain president would say that, well, no cloud services outside the US. I think it mentioned there the Davos World Economic Forum sometimes in between the lines. Or for instance in May 2025, the AI diffusion rule was taken back and replaced with something that said, well, we're going to decide case by case. So we don't have the countries, but we're going to decide case by case - which is similar as if you would take out the speed limits from the streets and the police would say, 'We're going to decide case by case whether we fine you or not for being too fast on the street.' So what does this mean? It increases the uncertainty in the market. Also with this new AI diffusion rule, which was basically less restrictive but increased uncertainty, the market reacted with a negative effect for the companies that had a high algorithmic dependency. Moving forward, there's of course hope in the whole process. Quoting T.S. Eliot from his poem 'The Hollow Man', I think it's even in 1925, 'This is the way the world ends. Not with a bang, but with a prompt,' which means it's not going to… We're not going to go into a crisis and then everything stops, but it's this slow process which is dangerous - and which has already started - that we have to be aware of. If we are prompting a machine, we first have to know that when we prompt it, it's giving back what we most like as an answer and not whether it's true or false. We did an experiment. I should've said earlier that my wife told me that I should wear a suit today and I told her that, 'Hey, I'm at Google DeepMind; I don't have to wear suits anymore. I swear to God that back in 2018, when I joined Google Research at the time, no more suits.' But given that today I'm wearing a suit, it means that whatever I say has nothing to do with Google DeepMind, so that's my disclaimer here. So whenever you see me with a suit, I'm not Google DeepMind. But we did an experiment at Google DeepMind. I'm going to talk about it in the last five minutes. We thought that it would be interesting to think about a problem where you really have high stakes. Research synthesis, I know this is not a common word in finance - unfortunately not - but research synthesis is very important in health, climate, medicine because it synthesises the state-of-the-art research - not just summarisation but a synthesis - informing policymakers about in which direction it should actually go. Unfortunately, we do not have that in finance so it's the consensus of the most important researchers about a certain topic. So we wanted to understand: what happens with large language models if we delegate research synthesis to a large language model? We're taking one particular example called the AMOC - the Atlantic Meridional Overturning Circulation - which is a big debate in climate science. They ask whether this is a tipping point or not, whether it will stop. If it stops, then we can go skiing in London all over the year, so it will become much cooler in Europe, so we did this experiment. We created last year, end of last year, a state-of-the-art language model that would extract information from thousands of articles, condense this information. Then we were giving this information, this research synthesis, to the globally-leading climate scientists to change it. You cannot do that with PhD students because the large language model is already that good that the PhD student would say, 'Okay, rubber stamped.' But what did the top climate scientists - globally top climate scientists - say? Well, they changed around 60 per cent of the text that was generated. Well, eventually 60 per cent of the whole document was rewritten by the climate scientists - which is a large fraction - but you see the problem. The large language model generates something which is looking so great. I would not have been able to change anything from that document because I'm not the super-expert on that. But if you ask super-experts, they will change 60 per cent. What did they actually change? Well, they changed some of the conclusions of the large language model, because the large language model is trained to be nice. It's always nice and it's always consensus. It's like the Pope. So it always tries to do consensus. It did not acknowledge the differences - the subtle differences - in the science community about the probability of an AMOC decrease. It also did not pick up some of the more extreme views, so it always condenses and we call it… In computer science, you call that mode collapse. You always convert to the mean also in terms of the knowledge that you eventually provide to the user. So this is something very important and it also brings me back to the problem: when you use large language models in your finance workflow in the sense that it's always around consensus, a general consensus. It creates this consensus. It decompresses the variance of the output and therefore, be very careful if you apply that. So the conclusion that we got from this DeepMind research project is that it's super-important to create a workflow where human and AI collaborate. That's super-tricky and it's also part of my current research. How can we make sure that we put enough friction in the human AI collaboration so that we don't fall prey to these four fallacies that I was telling you? They have the same origin - and the origin of these four fallacies was actually the two sentences that I mentioned at the beginning of my talk. So five things to do for you tomorrow, because Lee was telling you that we should have a to-do list. Audit your model supply chain. If you're not happy with it, at ETH they tried to build the Swiss LM. The first version was not so much of a success, but the next version - 1.5 - will be out this summer and we hope to build up a downstream finance pipeline with them. Use some adversarial attacks, so 'Battle of Transformers'. Remember that. Always monitor the provenance of your training data and map your infrastructure dependencies so that you have a migration plan. Most importantly, keep the humans in the loop. It's human judgement that is important - and human judgement, there are actually two aspects here. The first one is current human judgement and the second one is future human judgement. We have to be sure that we train the younger generation to ask the critical questions as well. If we now rely just on AI, we lose this tacit knowledge that we, our generation, has built up in failing along our career paths many, many times. If you just rely on AI, you use this tacit knowledge that is so important to make the decision. So just in one sentence, my conclusion: the fragilities - these four fragilities that I was mentioning - they are real. But that doesn't mean that all hope is gone, because they are your choices. Thank you. Questions?

F: About research papers being generated by AI versus those who are not: which are the most obvious characteristics that define, differentiate the ones done by people and the ones not? How to spot a paper generated by ChatGPT.

It's almost impossible. There are companies that sell these kind of tools. You see the obvious cases and you probably could tell the difference between an AI-generated… If you write birthday cards it's obvious, but there you can just write a prompt or, let's say, a skill in Claude Code that it should actually write me an article like Ronnie would write an article or like Roberto would write an article. Then it takes over the style of the person, so it's very hard to detect AI these days with some tools. At ICLR they were using Pangram, which is currently the state-of-the-art AI tool to detect AI-generated text. But also there is an error rate involved and AI gets better and better and can trick you as well. So it's very hard to detect. For us, think about what it means for science. Back in 1665, there was the first paper published in a journal by the Philosophical Transactions of the Royal Society in London. Since then we are writing papers and this writing process has changed dramatically, so we are all using AI. How can you evaluate whether a paper is really scientific, solid or not? We have to rethink a little bit also the scientific process. The scientific process is very similar to what you do in industry because you also want to have a model, whether it's a published model or your inner model, to extract signals to eventually invest in the market. More and more stuff is just generated by AI and it's really hard to follow. I know that also from this company I'm also working with that the code base that was generated over the last couple of months is just exploding. It's very hard even for that company to keep track.

M: Thank you. On your third fragility, when you talked about how more AI means worse papers, what surprised me the most is that you said, basically, from the first per cent you see that decline. What is your explanation for that? Are they using it wrong? It just puzzles me.

Well, of course it's a statistical average that you see, so there might be papers that are actually better than others, but this was a mean analysis. I tried many different things to really robust-ify this event, so I used paper fixed effects and all these kind of things. It could be that people are too naïvely using the AI, at least back then. It might be that it improves, but it's just a warning sign that we should be very careful on how we design these whole processes where we collaborate with AI. Humans collaborate and not just one single human, but also a group of humans together with artificial intelligence. I think in the future you will see a lot of developments in this direction. That's all I can say about that because I'm wearing a suit.

M: Maybe final question. As AI evolves, we all have… There are still separate pools of knowledge hermetically sealed off from, basically, the common knowledge base. AI will eventually be integrated into… Basically, some models will be able to access hermetically-sealed knowledge and data in their work to generate new insights. How do you think about that world where, basically, if you think about State Street and all the other big companies, we have some proprietary information. So our AI knows things that others do not. Basically, you still get that diversity of thought, so to speak, in that age. I just wondered whether you could speak to that a little bit.

Well, clearly LLMs have a tendency to compress to the mean, so diversity of thought is something that is usually killed by large language models. There's also a paper in 'Nature' that just came out which is more in the medical research domain. They showed that diversity of research question is compressed by large language models. One thing that you have to be really aware of is that for instance when we write a paper, at least in the old days, we first generated a lot of negative results. We know what is not working. However, we cannot publish that. The large language model is just learning from the tip of the iceberg that generated this scientific paper. But it doesn't know all this tacit knowledge that was necessary in order to eventually come up with what works. So my idea was actually to think about hypothesis-generating machines that would work on a continuous basis and just generate a pool of negative results that could then inform modern large language models exactly about this tacit knowledge that is actually not working. So that could be an interesting idea as well - but be very careful; LLMs stay compressed to the mean. This is a statistical phenomenon that we are all aware of, at least in computer science, and therefore you should really be careful. Thank you.

Welcome. Thank you. Why don't we just kick off very briefly with your current roles, your background, just a minute or so on, on who you are. Chris, I'm going to start with you and go down the line.

Thanks. Hi, everybody. My name is Christoph Gort. I'm with Cambridge Associates for almost two years. I joined via the acquisition of Cambridge Associates of SIGLO Capital Advisors that I founded almost 20 years ago, and we focused on alternatives. Cambridge Associates is definitely more familiar in the US than it is in Switzerland, but we're basically an investment firm working with all sorts of institutional clients, family offices. Overall, more than 1000, but we don't have products. What we do is we select third party managers and strategies that are complementary to our specific client's needs. In Switzerland, we do the same, like our colleagues everywhere in the world. We try to understand what the clients are looking for and then find strategies and managers who implement accordingly.

Alexandra.

Alexandra Janssen. I'm an economist by training, and I'm the CEO of a company, Ecofin. We advise private clients and institutional clients how they should, could invest their funds, their money. Similar to what Chris just said, we are independent. So no own products. We're somebody who is selecting products, trying to understand the markets, trying to understand our clients and looking forward. We often have a macro lens. So looking forward to the panel discussion.

I'm Elliot Hentov. I'm a little bit Lee Ferridge's alter ego because I sit on the asset management side of State Street. My clients are much more concerned about a little bit more medium-term strategic asset allocation questions. Particularly in the macro side, we look at investment themes that play out over months and quarters, kind of 3-to-18-month time horizon.

Okay, let's start off by setting the global backdrop. What do you see as the major issues we should be considering right now when it comes to global investing? There are some obvious ones, but maybe some we haven't thought of as well. Elliot, I'm going to start with you. What should we be focused on?

I'll keep it short, but the obvious ones are the right ones. Obviously, the war you want to hedge, the risk outcomes from a negative supply shock, and then most importantly, you want to be positioned for the AI CapEx cycle. To what degree does that demand shock for AI infrastructure, the potential huge revenue streams that could be emerging in that new sphere, how is that weighed off versus the threat of higher rates, the circularity of investing in that space, the increase in leverage, the increase in insider selling in the space as well? I mean, it's the obvious ones, but if you get that wrong, the rest of your asset allocation will not matter.

Alexandra, same question to you.

Which issues matter are actually investor specific, right? There are risks that might matter to a family but might not matter to an investment pension fund and vice versa. I think important is to understand which issues actually matter. In my view, there is a difference between the issues at hand. We have things like AI valuation and concentration risk and indices, and these things, they're risks. They are manageable. I can think about them and then optimise my portfolio accordingly, but then we have effects of unsustainable government debt and high geopolitical risks, and there we're, in my view, more on the side of uncertainty. Uncertainty, where you sometimes don't even know in which probability space you're actually working. Depending on which risks are actually relevant for you as an investor, you might be able to do normal portfolio optimisation, or you might need to think very differently about the uncertainties at hand.

Chris, any thoughts?

Yes. Last, make it short. What is interesting is that ex-ante, we often talk a lot about risks with clients, potential risks, etc., and with hindsight, we typically forget about all the risks we discussed and it's only about performance. The challenging part is that there are more than enough reasons to be worried. Actually, always. Of course, also today. But then at the end of the day, if the client only cares about the performance, the biggest mistake that's been done since '09 was to be not enough in equities. To me, this risk is interesting because I got my scars in the tech bubble thinking that I'm smart, making a lot of money in '98 and '99, losing everything the year after, and then realising, probably I didn't understand as much as I thought. That scar is on my back forever. Buying the dip is a strategy that worked extremely well the last 18 years and I'm not really sure… I mean, the people who made a lot of money by just buying the dip all the time were right, but I wouldn't feel comfortable. I think this is one of the biggest risks to see this development, to see the valuations. I don't want to be the old, frustrated dude who thinks we're too risky, but at the same time, there is a limit to where it can go and this is what fundamentally worries me a little bit.

Can I just add to that? I think that's always interesting on macro panels because you're looking with the macro lens, you're focused a lot on risks. As soon as you talk to entrepreneurs or people looking at single companies, then you're a lot more optimistic. I think for people with a macro lens, it's always important to change sides as well to not get too pessimistic.

Well, I think it's what Roberto said this morning, macro strategists. There is no hope in that role. I think that's what he talked about. To that end I'm going to talk about the conflict in Iran now. The conflict, it continues to rumble on despite much talk of a peace deal being two days away and it's going to be the greatest peace deal we've ever seen and everything's wonderful. There's no real sign of a meaningful deal here. How are you thinking about the risks around that? How are you thinking about what if this thing does carry on? What if the oil market cannot continue to shrug it off when the Straits close in six months' time, and we're $160-$170 a barrel? Which remains a real risk, that has not gone away. Alexandra, I'm going to start with you on that one.

If you talk to entrepreneurs, what I think is really interesting is that they're saying we're still in the supply chain shocks from COVID, from everything that happened from the war in Ukraine, and now the next shock is already potentially coming or depending on the areas, it's already there. I think the system is already stressed and one thing that I think is important as an economist, there are economic principles that apply as well to geopolitical conflicts, and here we have a lot of externalities, right? The US is an important party here, but countries and areas that are affected are very different. Europe is much more affected, Asia is more affected, and it's difficult to protect against these. I think as soon as we're getting into late summer and early autumn, and we see no resolving then, especially for Europe, the situation is going to be very, very difficult. I think then adjustments in the financial markets will be coming as well.

Why is there no deal? It's been two months since the ceasefire. There was no deal a month ago, not a week ago. Why should there be one next week or in three weeks? The answer is you're going to need a trigger for a deal. So now let's just think about what is that potential trigger going to be? The one that I think is the most obvious one is you get the economic pain and you get the market pressure and one of the sides softens its position, most likely the US. That's one scenario. In that case, you need the event, you need the downside event at least to be building up to some degree in order to get the geopolitical outcome. The other ones are, I think, less likely. There's an idea that maybe the Iranians eventually feel squeezed, that their timeline is shorter. That could very well be. I think none of us really know. I think the more compelling one is you probably need more economic pain. You do need oil well above $100 again, ten-year bonds approaching five and so forth in order to actually get the diplomatic process moving.

I think I mentioned earlier, we know Trump wants out. He wanted out from day two, but it takes two to Taco and this is the problem he's got. Iran is the one that's really stopping the deal going forward. Trump wants an off-ramp anywhere he can find one. I take your point about the economic pain, but I think Iran has to start to feel economic pain as well to get that process moving.

Well, that's bad news for all of us investors because their timeline is several months and their timeline is beyond the world economies' pain threshold.

Chris, any thoughts?

No, not many thoughts. I think what we try to do with clients is to not make bets on single events and single risks but try to construct portfolios that make us feel comfortable to move ahead, regardless of particular scenarios. Obviously, that sounds very easy. It's difficult in practice. We don't try to come up with strategies that are dedicated to particular macro events. We like to think of portfolio construction as an exercise that prepares us for all sorts of future events. There are many other risks, including what Marcus said before, that we should take into account and probably should tweak the portfolio a little, and this is why we're not focusing that much on single scenarios.

Final question on the global macro backdrop. We're a week away now from the first meeting of the Warsh Fed. Do you think the change in Fed leadership will be significant to the investment landscape? What are the risks or concerns around a Kevin Warsh led Fed? A lot of people seem to think it's just the Fed. It's 1 vote out of 12, etc. I'm not convinced. Alexandra, let me get your take on that.

I read quite some things that Kevin Warsh wrote or said in podcasts and appearances. I think the vision he has is very different from the Fed today, right? He wants to revision the Fed, reduce the balance sheet. He has quite a monetarist view of inflation, which I personally think is high time that it re-enters central banks. The problem is that he enters a situation where most of the people in the Fed think differently, and he will enter as this one person. Of course, he can influence change, but this will be, I think, a very hard fight that they will have in the Fed. He doesn't have a year to just convince everybody in there and then start the work, but from day one, the policy needs to stand. I think this creates some friction and as well risks that at the moment is, in my view, somewhat dangerous as well for the financial markets, that it's not clear where the Fed as a system stands.

So you think Warsh would be a good addition to the Fed; it's just that the institution is not going to be carried along by him?

I don't know, right? I take him at exactly what he says. I think there are very good ideas in that, that would, in my view, influence the Fed in a positive way. The problem is, if these ideas now hit the Fed, that is very established in the way it works, then it's going to create frictions and uncertainty in a situation like now where you have oil prices rising, price indices rising, but monetary inflation, what Warsh more looks at, not really rising. That creates a lot of discussion and frictions, which is actually good that this discussion happened, but they happen too late given the situation we're in. I fear that could create some uncertainty and friction for financial markets.

What are the good things? Do you think it's the balance sheet shrinkage or the shift in inflation focus? What specifically do you think he brings new that's positive?

For example, I agree very much on is his view of inflation that after the war in Ukraine, energy prices increased and the Fed and the ECB, the first inflation, they all blamed energy prices on Putin, and of course, these effects were there. Looking at the monetary aggregates, it was very clear that this is true monetary inflation and they reacted much too late, in my view, to all that. I think there, he has an important point about the view on inflation that these central banks should have, and I think that's very positive but the effect that it now has on policy, I'm not so sure about that.

Elliot, thoughts on Warsh?

A lot of thoughts on Warsh. I'll start off with let's simplify central banking. What's central banking simplified? You obviously affect short term rates through the rate channel. You can maybe manipulate long term rates through balance sheet. Then the third function really is market communication. So let's look at what a Warsh Fed will look like. The first one is market communication will be very different. I think we're, at least in my professional life, in an unprecedented era. We had 25 years of every single meeting consensus vote of the FOMC governors. That's no longer the case. That's obviously ended. We'll have even more friction. I know that's usual for other central banks, that's common, but for the US and the US bond market, that's new. We have to learn the muscle of what does it mean if we have six, eight different views popping out of each meeting? What does that suggest about future policy? The other one, this is where Alexandra and part ways, she just said, 'I take Warsh at his word.' I do not. I think he's full of BS and I'm really talking about the balance sheet. That's what I meant with BS. I think it's absolute hogwash. I do not think he's going to shrink the balance sheet. I don't think it's a possibility. I do not understand, in my macro lens, how it adds up given the fiscal expansion and where things are, the parameters, how the Fed will actually engineer a balance sheet shrinkage. I think quite the contrary. We're going to have stress in the bond market and that stress, suddenly the financial stability mandate will come in, and I see the Fed actually re-engaging in balance sheet expansion. Completely contrary to what Warsh was telling us and to who he has been all his life. I'm a big sceptic here. It's the one contrarian take you'll hear from me today. Otherwise, I'm very much [unclear word 0:15:33.7] in the consensus, but on Warsh, I really part ways.

Any Warsh thoughts?

No. The juicy statements have been made; nothing to add here for me.

If I just can add, what I think is interesting is in the communication that he says, you can potentially shrink the balance sheet and in return you can decrease interest rate, but he would start by decreasing interest rates, right? This already makes one cautious whether the shrinkage of the balance sheet is really there and will come. I agree it depends a lot on circumstances, but I just think that his different view that he brings into the Fed is a more monetarist view. This is actually really healthy for the Fed, but in my view, it comes a bit at a at a bad time.

I'll add a bit of context here. Last week in Boston, we had as our keynote speaker, Dr Daron Acemoglu, who for those who don't know, he wrote Why Nations Fail. He won the Nobel Prize in economics in 2024. He actually knows Kevin Warsh quite well. He said that he's a very polished, very intelligent man and he will always do 100 per cent the right thing for himself. That was a direct quote from Daron Acemoglu last week. I would take it at its word. Let's bring this back to Switzerland. I promised we would talk about it from a Swiss perspective. We talk globally about rising rates. We talk about inflation concerns. We talk about currency weakness. Swiss investors, you face a very different set of challenges. Chris and Alexandra specifically, would you care to expand on that? What makes the environment for Swiss investors so unique and what do you do about it?

Well, I probably would take too long but let me start with what makes it a little different. Inflation in Switzerland has been very low for decades. Rates have been very low. We're usually always involved in the carry trade on the one side for the cheap borrowing. This makes the whole management and also return expectations a little different. The Swiss ten year is around 40 bips to 30 years, whatever, 60 bips. That shows you that it's not that easy to create or to generate high performance in Switzerland. I have to remind my colleagues in our headquarter in Boston that target returns of 10 to 12 just sounds a little bit unrealistic to Swiss investors, especially if it's net of fees and hedging in Swiss francs.

Those Boston people, they're the worst!

Ah, yes! I mean, you know what you're talking about being English, so how could you be there for 17 years? No, just kidding. The thing is it creates a different set of expectations. The currency hedging is obviously a very big topic. In real returns, the performance of the US and the Swiss investors hedged real are fairly close. This is not that surprising, but a lot of the discussion we see goes is four per cent hedging costs too high? Should we do it; should we not do it? Well first of all, it's not a cost. Second, it's more of a risk management decision than anything else. This makes the Swiss set up more challenging because generating high performance is difficult in that in Swiss francs very strong currency hedging is an omnipresent topic. The third one I would list is a strong home bias among the institutional investors in Switzerland. As you all know, three stocks make up half of our market cap index. Not saying anything about these stocks. They are great and there are reasons why they're so big. At the end of the day, clients who invest on a market cap basis plus with a Swiss home bias have a lot of money in seven magnificent stocks and three large stocks in Switzerland. You cannot deny that this looks risky and with respect to concentration might be a little bit of a worry and that's from a Swiss lens, different, than for most other countries.

In one respect, Swiss investors aren't that different, right? They're invested globally, most of them, some with the home buyers, mostly I think for good reason, sometimes bad reasons. There is not much difference except that the reference currency is the Swiss franc and there the issue just is that very often, the Swiss franc appreciates in times of crisis. Exactly when you have the crisis hitting globally, the Swiss franc appreciates or other currencies depreciate relatively and then you're hit twice. That's, I think, one of the main issues why the hedging plays such an important role. As soon as we're talking, not about the equity side, but about the bond side where you have nominal payoffs, then it's just a lot more difficult to reach net of fees, significant returns. This is especially important first for pension funds who have to fund their liabilities in a safe and secure way and as well, for example, private investors with high liquidity. If you have already, again, at some points in time implemented negative interest rate at Swiss banks. It's not an easy task to do and it's difficult to manage the risks accordingly. In the end, I think the issues that Swiss investors face are very similar to all the others investors globally.

You mentioned about the safe haven. Japan could have been - or it used to be in the same situation. What's different now? Why is Japan struggling with a weak yen? They're intervening to stop it. We've got the same global backdrop. You've got two countries with huge positive international investment positions. Why is the yen so weak and the Swiss franc so strong? What have they done differently and could Switzerland learn from that or not? Any clues?

I don't know whether Switzerland should learn from that!

No, not necessarily a good thing!

I hope not! I think it's difficult to say. I don't have the answer. I think the Swiss franc is strong because the economy… I mean, you always have two things, right? You have an economy that needs to work well and you have a monetary environment that needs to fit this economy. In Switzerland, both areas are strong. The economy is relatively strong and the monetary cloth is relatively tight fitting to the economy. That's why the Swiss franc is strong. Japan, I think is a very different case. I mean, we have the issue with the high government debt and the economy that I think works very differently. In the end, the reason, I don't know, maybe you have more insights.

I'm not an expert on Japan either. I mean, obviously the debt to GDP ratio is totally different. The widowmaker trade among hedge funds is well known, saying that there will be an explosion in rates in Japan. I think no other trade in the last 30 years created so many losses among hedge funds like that one. It's always been said, 'Next year, Japan's going to break, the debt to GDP is too high.' Well, it didn't happen for 25 years. I'm not saying it doesn't happen for the next 25 but it seems that there is more resilience around that situation than what we here in Europe might understand. I'm not in a position to comment in more detail.

Elliot, thoughts on Japan and the difference, why we've seen this divergence?

I think the lesson from Japan for Switzerland is that… Switzerland's experienced it with, it used to be S&P sovereign ratings, we all remember 2015, is that when you do get stretched conditions, policymakers are forced to act. You actually don't have the policy freedom and to the degree we see that, that's the lesson from Japan. Yes, the yen is weak but look what it's doing. It's forcing them to act constantly. It's tugging them. It's almost the market is pulling them like a marionette, like puppets a little bit, to respond. I think that's the lesson for Switzerland. We will get moments again where policymakers are going to just simply be forced to respond to the effects dynamics, because there's limits to how much you can let the natural market forces unfold.

Let me pick up on that potential Swiss policy responses. If the Fed under Warsh is as dovish as I expect, and I think they will be, I think he's been put there to do one thing and that's cut rates, and this leads to renewed dollar weakness which I mentioned before. I think we could see a ten per cent decline in the dollar over the next 12 months. You're going to see Swiss franc strength. How do you think the Swiss authorities will react to this? Are we going to go back to a specific FX target like we had prior that ended spectacularly? Do we think currency intervention? Go back to negative rates? How will the Swiss authorities respond if you do get another ten per cent appreciation against the dollar? Chris, I'm going to start with you on that.

Well, it's a good question. The feelings about the negative rates that we had for a couple of years are mixed, definitely, depending on the industry you've been working in. It was also surprising that, for example, pension funds was never even a discussion whether they are exempted. From a financial perspective, delivering a strong performance with negative rates is going to be terribly difficult and for every saver and pensioner, this is not good news. For every active member, it's not good news either. Given that the guarantees in Switzerland are fixed, it just means higher contributions. Now, we're a low tax country compared to our friends in Europe, but still higher contributions are not attractive either. I truly hope that we do not enter the negative interest rate territory again, but am I totally sure in a scenario like you describe? Well, no, not at all. So it's definitely an option. I don't hope it materialises, but it would be naïve to completely rule it out.

First, I think or I hope that the central bank wouldn't intervene on the FX market or even with a lower or upper bound, because I think the risk associated with these actions for Switzerland are quite large. I think if we only have a depreciation of the US dollar and not the general appreciation of the Swiss franc vis a vis many currencies, then it would be even more difficult to actually argue for more interventions. Now, if that would be combined with a global stress scenario, which is probably not unlikely, then the situation might be different. If we really have a very fast depreciation and over appreciation, however that is defined, then the Swiss National Bank might be keen to intervene more strongly again. I hope not, because I already think from an economics perspective, interest rates are very low in Switzerland, a bit too low, in my view, and lowering them further or intervening just brings a lot of risk for the real economy in the end.

Plus, some wealth transfer where the Swiss National Bank just doesn't have the mandate for. Let's not forget the Swiss franc is good for some sectors and industries and bad for others. I am not aware that it's the central bank's job to organise a wealth transfer. From that perspective, it would be probably good to make sure we understand the policy and the goals and probably redefine if we think it's necessary but that wealth distribution, in my view, can be a little bit of a problem. I hope that the SNB is hesitant taking that issue into account.

Let me jump to something else. The last few days, the last week or so, back in Boston, there started to be a focus on the referendum you've got coming here, the vote about the 10 million population cap. The way I look at it, this is another extension of populism that Brexit was an example of populism. Donald Trump is an example of populism. This is in the same vein as those two things. What are your thoughts on it? Will it pass? Will it not? Is it a good thing? Is it a bad thing? I'd love to hear your perspectives. Alexandra, go first.

I don't know whether it's going to pass or not. I think it's very tight. I disagree a bit that the feeling in the population that immigration is too high, population growth is too high. That's not, I think, a populist thing. That's really a fear and the problems that we see in real life every day. I think this is very merited. The problem that I see is that the solutions that we are presented, there is just one solution that's this initiative and if it's not that, then it's going to be a similar one in five years. If now the people say no, then it's going to go forth as it has gone up until now and then people in Switzerland are not going to be happy at all. I don't trust in other parties to really have a good solution. The question is, from an economic perspective, I think there would be good ways how to structure immigration into a country and the initiative that we're looking at right now is not that. Nevertheless, I think if there is a no now, then it goes on as it did before, and that's not a good solution either. I don't know what is going to happen, but I am not sure if it's as populist as we see in other countries.

I agree on your perspective, and I'm not here to tell anybody what to vote, but the thing is, it's an expression that the Swiss population is apparently not happy. About a decade ago, we had another voting. It was called 'masseneinwanderungsinitiative' that actually was accepted by the public, but sort of vaguely. Some would say not at all implemented. What we see today is just an extended level of disappointment with the situation. I think this per se is a little worrisome. Now, the solution of the initiative might be a little bit myopic and short sighted but what worries me more is that we have the discussion for over a decade and apparently just never took it really seriously, bringing us into the situation where we have an initiative that is very harsh, and probably from an economic perspective, not a good solution. At the end of the day, it also shows - and this is why the polls are so tight - that a significant part of the population is disappointed. It's a little frustrating that we even got into this situation because the problem is not new. Switzerland benefits from smart people coming here, and I think nobody in our country has a problem with that, but we just stretched it a little too far and what we view with this initiative is the increased level of frustration. I think you should have respect for that feeling, even if a day in Hong Kong or Singapore and probably New York will show you what density really means. I'm not even talking about the subway in Tokyo, but this is the way the population feels here, and I think they have all the rights to express that feeling.

I just want to bring a global view here. I really do not know anything about local politics, but it could be a solution to the other problems. Meaning you're worried about currency strength. Well then, follow the British example. Come up with silly policy actions and your currency will suffer!

Let me remind you, you actually live in London!

It will depreciate! It could be a solution. I mean, the British case is really - when the referendum happened for Brexit, the UK, many people forget this, after Greece, had the worst real wage income growth in the previous decade. From 2007 to 2016, the UK was the number two worst performer in the developed world. It had a cardiac arrest after the financial crisis, and then it decided to give itself a stab wound on top. Now, when I think about Switzerland, one of the reasons the currency is strong, the economy is strong, the monetary environment is because the policy environment is solid. Obviously, it's a threat to the global investors' perceptions of that. Notwithstanding that the feelings are genuine. Brexit, the feelings were very sincere. The economy had underperformed. There was an underclass growing, there was neglect of public concerns and all those issues, emotions were real. The solution, unfortunately, was very counterproductive.

I was going to say, how did that work out in tackling those problems? How did it work out for that? Is it the economy is booming now and everyone's doing great?

Can I just add one point what I think from an investor as well, global investor perspective is interesting is real estate prices in Switzerland. They have been pushed very high in parts due to the monetary policy, in parts due to the well going economy, but in parts as well due to immigration. If you really have lower immigration in the future, it's going to be interesting to see what happens with real estate prices. I'm, at the moment, looking at real estate prices and wondering what would happen if there was a yes because I think there should be a reaction in the market, but let's see.

I want to leave some time for questions from the audience. So please, anyone in the audience got a question? We've got three people here you can ask. Come on, somebody's got to have something.

Thank you very much for the discussion. It was very insightful. It's maybe more a comment than a question. Just this morning, I came across a very interesting presentation by the University of Lucerne that was addressing this question with the initiative. It was quite interesting that he presented, actually, whether yes or no, there will be growth of the Swiss economy, a little bit more if we vote for more people, so GDP will grow more slightly, but then GDP per capita will roughly stay the same. I found that very interesting that we need to basically make a decision on how we want to live in Switzerland, how much density we would like to accept and maybe find a way to deal with it.

From my perspective, I think GDP per capita is a much-underutilised measure. We look at real GDP growth, Japan, we look at real GDP growth. If you look at GDP per capita, Japan is actually doing much better than people think because the population is declining. I think GDP per capita is a much-underutilised way of thinking about thing as economists generally. I think that's an interesting point.

I think that's the point driving the discussion, right? It's not growing into the width of the country but growing per capita. The question is, which is the best way to reach that? I think that's the hard question at hand. I think implicitly, that's the question that people who are voting are trying to answer. They're only interested in GDP per capita, if at all, and definitely not the overall GDP of the country, right?

Any other questions? All right. We've got a couple of minutes left. Let me finish with one. Elliot, I'm going to start with you. Put yourself in the shoes of a Swiss investor facing low domestic rates, persistent currency strength. How would you, as an investor, tackle this? What would you do?

First of all, I'm very grateful that I'm not because it's a very tough position. I go to Australia and I made the joke, the world's upside down there, but it's really upside down here. There are very few markets that we deal with where the clients face this, where their base currency has the challenges that it does. I mean, obviously at the core, FX management sits at the centre of portfolio management in a way that it doesn't anywhere else. What that looks like in practice, I think is very specific. Alexandra talked about pension funds and others that have large liabilities, but it's central in a way that it isn't. I think that's all pretty obvious. I think, Christoph, the two of you are better placed to explain what those strategies would look like. The one thing I would draw from the low-rate environment, the lesson from the 2010s though, which I think applies to Switzerland and elsewhere, is there's no point in hunting for yield, stretching that search for just that extra few basis points in your base currency, in your own fixed income market. The lesson from the 2010s is that that doesn't play well. Much better, you focus on your global diversified portfolio. You use the carry trade, use the Swiss funding rates to amps returns, and all of those strategies make much more sense than trying to eke out that extra basis point in the home market. I'll leave it to the locals.

Alexandra and Chris, what is he missing, or you can answer the question, what is the biggest risk or the biggest mistake a Swiss investor can make?

Hard to say. I agree with what you said. I think in the end, if the reference currency is really the Swiss franc, then there is no escaping the Swiss yield curve, right? One just needs to accept that. I think it's as well important to understand that in real terms, it's not such a bad landscape to be in. I think there is a lot as well, not for professional institutional investors, but on the private side, we need to do a lot of educating, why there isn't four per cent here, and for good reasons, and why that isn't so bad, right? In real terms, we're in a good situation. So it's a bit of education on the private side and as an institutional investor, optimising, given the yield structure that there is.

I mean, maybe we created the wrong impression that the low-interest rate environment in Switzerland, in Swiss francs is a reason to not get up happy in the morning. Well, it's clearly not. I mean, there's so many interesting things you can do to think about how to improve a portfolio, make it more robust, more resilient, better performing. We have the private markets. I'm not generalising that everything is great there, but it opens up fantastic opportunities. We have amazing universities in Zurich producing a lot of interesting spin-offs. Unfortunately, the money is then usually made by US private equity managers and not by us, but it's a very interesting thought how to improve that. You have the alternative world with hedge funds that is fascinating, even if proportionally more black sheep than in other industries, but still a lot of very interesting stuff. What I think should be appreciated is the fact that the real returns, as you say, are pretty okay, and the nominals are low. This is motivation to do a good job in the morning when you get up. I think this is inspiring, not frustrating. I'm not sure if I would want to change with anybody else just because our nominal expected rates are a little lower. It's actually a great motivation to think about how to improve your portfolio and this part in itself is super interesting.

On that positive note, I'm going to close. That was brilliant. Thank you, all three.