2025 State Street Research Retreat

Alberto Cavallo: Thank you. Thank you. It's a pleasure to be here. See so many familiar faces. I should clarify from my bio. I'm no longer a member of the advisory committee of the Bureau of Labor Statistics because President Trump shut that down. So we need to update that bio. Value. But really great to be here. I was, uh, I was looking at Michael today. Michael Metcalf I remember in London last year, um, you know, I presented some results that showed that inflation was falling. And I think you said, Alberto, if inflation comes down, what are you going to talk about next year? So yeah we have tariffs now and I'm getting asked a lot are tariffs going to bring inflation back. So I've been talking a lot about this um updating the the slides as you can imagine very quickly because the policies are changing and the results are also changing. I'm going to show you some of the latest results. And I'd love to make this as interactive as possible. Okay. So I'll show you some results and then ask you to give me ideas of what could be driving some of these patterns. And we'll we'll react to that. Okay. We'll make it a bit more like a, like it would be in an HBS class. But let me start first showing you what is happening with the price stats indicators for the US. And then I'll go into the research to explain some some of these.

Alberto Cavallo: This chart that you're looking at is the US monthly inflation rate. The latest numbers. This screenshot I took it a few days ago. You can see both the oranges are line. The CPI is the blue one that ticked up. They both ticked up last last month. But actually this month we're seeing a bit of a slowdown. The current rate is 0.2%. So there's no if you look at the aggregate index itself, there's no evidence yet of any significant acceleration in our index and also in the CPI for now. Okay, I like and if you've seen me present the last few years, I like when there's a turning point moment happening. I like to actually look at the price index rather than the monthly and the annual rates. And this chart that you have on the screen can be used for that. We can actually calculate the current trajectory with the price index is in, which has started in December of this year. And we can compare it to what we had remember at the beginning of Q1 of 2024. The slope is actually lower, so the trajectory is not as inflationary at that scare that we had at the beginning of 2024. But I will mention it's longer lasting. So think in terms of seasonal adjustment. This trend seems to be persisting for longer. And you'll see some of the reasons for that. I you know I helped create the index price that produces them. The macro strategy team here with Lee and Michael produces these great documents.

So I borrowed a couple of their charts, which illustrates something very nicely. So I showed you before the Price Stats Index compared to last year, the graph on the right actually shows you that if you build a range of values between January and June of each year, the current index is actually well within the range. It's actually even below the median levels we have seen in the last ten years, of course, affected this average and median by some of the bigger numbers we saw a couple of years ago. Now the chart on the right and this is from Michaels. Metcalfe's 8th of May results for April. He's actually comparing each of the subindices the price that generates versus the historical average. And you'll notice a couple of things if you look closely, the current levels, particularly in April but also in May, if we were to update these numbers, we're relatively high for two sectors in particular. Household goods and electronics. Obviously, two sectors where we might more likely be more likely to see some of the impact of the tariffs. And in fact, if you go and look at some of the other sectors, you may notice transportation for a while actually had some upward pressures. But it is important to understand that that sector and fuel in particular, is compensating some of the effects we see of tariffs in the other sectors.

So just to make this clear, this is our fuel monthly inflation rate. Actually the transportation index which is greatly affected by fuel. You can see that we've had some spikes in the last few months happening there, but the level that you see there compared to what we experienced a year ago is much, much lower. So this is actually this is a sector that is likely putting downward pressure on the aggregate index right now. And if we end up going into a recession, it's going to push down aggregate inflation for several months. So that's important to remember. So anyway just to summarize where is the evidence of more of an impact of tariffs happening that's happening in sectors like these. This is household and furniture. And I'm showing you the price index here. You can clearly see around March when some of these tariffs started to become binding, we start seeing a new trajectory happening there. So the question becomes is this connected to the the to the tariffs themselves or not. So I'm going to actually show you some research I've done on this topic. I'm going to use an old and a new paper okay. The old paper is written in 2021. It was about the first trade war. I'm going to try to extract some lessons that we can use to understand what is happening today. Okay. This is a paper like I put there that was apparently misused by the US Trade representative to justify the reciprocal tariffs announced on Liberation Day.

So I won't walk you through the details, but you may remember Trump with all this list of big tariffs. He actually had a formula for it apparently. Or at least the trade representative tried to justify them. That formula required some parametrization of the sensitivity of prices, for example, to the tariff rates. And they seem to have gone to our paper to get that number, but they got the wrong number. If they had actually gotten the right number, the tariff rates would have been a fourth of what they actually were. Okay, so I can give you the details if you're interested later on. But perhaps more importantly, I want you to take away something valuable of that paper to understand what is happening today. So I'm going to focus on the other bullet points, trying to understand what did we learn from the first tariff trade wars. So why was this a useful paper? We were able to look at prices at the border and at the retail level. If you can do those two things, you can answer a fundamental question, which is who is paying for the tariffs? Any tariff could be paid by three parties essentially. It could be foreign exporters. And certainly Trump when he introduced the tariffs. And now he's hoping the foreign exporters will bear the burden they will have to lower their export costs. But it could also be paid by Americans.

And you have to split there between firms that end up bringing the goods, and then consumers who may end up getting the cost if the firms decide to pass it on. Okay. Now it's unclear a priori who is actually going to pay for this. We looked at the first trade war, and we found a year and a half, a year and a half after the tariffs were introduced, most of the burden was lying on US firms. So the Chinese exporters, who had not dropped their prices and the firms had decided not to fully pass it on to the consumers. So both ends of the political spectrum were wrong. Trump was accused, was hoping that the Chinese would pay. The Democrats were hoping that consumers would pay, or at least claiming that would be the case. We found that it was US firms, and I want to walk you through some of those results of why they happen, so we can understand the situation today. First, as I mentioned, there was when we look at the border full import pass through because the Chinese did not lower their prices despite the RMB depreciation. I'm going to make this interactive. Why would this be the case? Why would the Chinese decide not to lower their export prices? Who wants to answer that question? If you're a Chinese exporters, why would you not lower your prices if suddenly getting into this market is more expensive? Yes. Can you speak up?

Speaker 5: It's already.

Alberto Cavallo: [00:13:56] The margin. The profit is. You mean the margin? The margin is too low. They may have no ability to actually do that. That's possibly true in some sectors. So they had no option but to keep their prices as they are. Any other ideas? Yes.

Speaker 6: Demand is so strong for their products. They don't have to lower right?

Alberto Cavallo: Americans would still pay for it or they think they would, so they decide not to do it. Why would the demand be high? No alternatives? Yes. Very good point. There's no alternatives. We were importing from China goods that at the time we could not bring from other places. This is actually very important to understand. Who pays for the tariffs. If you have options, you're actually going to put the pressure on the outside exporters. Okay. So all these reasons are very valid. In fact, this graph shows you what end up happening with the prices that importers in the US paid when the tariffs were introduced. And if you look at that chart, it jumps up by nearly the full amount of the tariff. So just to give you some magnitude for a 20% average tariff rate, importers ended up paying 18.4% more. Okay. And this is something that lasted a year and a half after the tariffs were introduced. You may be wondering wait a second. At that time the renminbi depreciated, which actually gives incentives to the Chinese exporters to lower their prices in dollars. But that depreciation at the time, and particularly for the US, does not have much of an impact because US goods or actually global goods are transacted in dollars. They are contracted in dollars. So usually exchange rate effects take a very long time in a country like the US to have an impact on prices. That's sort of explains why we saw 18% instead of 20.

Now here's the evidence that how differentiated the goods are and how many options you have is important. We flipped it around and we looked at exporters from the from the US being affected by the retaliation tariffs. So China decided to retaliate. And specifically they focused on agricultural products. These are products that they could bring from other places. They could bring it from Brazil, from Argentina, soybeans for example. That meant that US exporters did have to lower their prices to become competitive in such a, you know, obviously competitive commodity markets. They ended up lowering their prices by 7% for an average tariff rate of 15%. So how differentiated the composition of exports and exports is, is tremendously important. Has this changed now? Probably not that much. Although you can argue that a lot of firms have had a lot of experience, or at least have been thinking a lot about doing trade diversion, about finding alternative suppliers, and this could actually end up putting more pressure on foreign exporters this time around. Now, let me focus on the retail for a second. Here we found something I think is quite important today as well. That size of the tariffs matters and that the visibility how much consumers understand if a sector is being affected for tariffs matters. So I'm going to make that clear with this chart. There's a lot of lines here. But this is essentially the additional inflation rate that we saw a certain number of days after the tariffs were introduced for several goods.

The blue line I know it's probably very small in the screen. The blue line is washing machines. If you remember, washing machines was the first tariffs that Trump introduced widely announced. And everybody knew they were getting hit. It was a 20% tariff rate. We saw an increase in the CPI. Academics started showing it. We showed it as well. Very quick pass through okay. Now why the tariffs affect washing machines so much and not the other goods that you see listed there. I think it was two reasons. First, the size of the tariff. So washing machines go to 20%. Many of the goods go to initially only 10%. If you get a tariff. Firms may decide to wait it out. If you get a big one, the cost is very high. You may have some worries and concerns about your customers and things like that. But obviously the size of the shock matters. If you're interested in this, I actually have a paper on Covid called Large Shocks Travel Fast that that shows this in particular. Also during other types of cost shocks. That is important. Today we're talking about the magnitude of tariff rates that is extremely high in some cases, although we don't know what it will be. Second, visibility matters. And this is where I think it was new for washing machines. Everybody knew they were getting hit. It's easier for a firm that is trying to pass this on to consumers to justify it, and avoid antagonizing and creating anger on the parts of of the customers.

Smart. Anyone here from Canada? No, I guess we we don't travel back and forth. This is actually another. Another. But I was in Canada presenting some of this stuff and talking to some people, retailers over there advertising clearly at the the websites which goods are getting tariffed. It's actually a very smart move because if you as a customer know that good is getting hit, you're more willing to accept a higher pass through a higher. You sort of don't blame the retailer, you blame the tariff and the policy. But as you can see there, as Trump expanded into other goods, no one was really clear what was getting affected or not. That's one of the effects of uncertainty in terms of of the effect in retail prices. Okay. Now we also find a lot of evidence for adjustments on the firm side. We tend to assume in economics. And when we write academic papers that firms have this very simple price setting decisions, they get a cost, they have a certain margin and they pass it on very quickly. But the pricing decision depends on lots of factors, including things like what I'm listing here, how much do you actually have of a stock? How how long do you think this trade war is going to last? So what you're looking at here is actually a chart that shows you the tons of material or goods that are imported by two very large retailers in the US.

We actually went to the customs records, and we could see how many times they were bringing products in certain categories from China and from other countries. And what you can see there in the graph, the blue line shows you the tons that were imported from China. Those actually spike before and during the time of the tariffs being implemented. That's the front loading that is certainly happening today as well. This allows firms to reduce their pass through quite significantly at the beginning at least wait for for a while. The second one that you see there is the red line that's straight diversion. Those are the goods that these retailers started to bring from other countries. We saw that increase significantly after the tariffs were implemented in particular. But here's something interesting. The share of goods imported from China was 90% before the tariffs. It went quickly down to about 70%. And then we saw it stay right there. So there was evidence that there's some trade diversion that is easy to do at the beginning. And then things becomes more complicated. You have to make, you know, more expensive decisions of relocations, of having connections with suppliers that are actually more expensive for many of these firms. But certainly that's playing a role as well. Today, we did find a lot of evidence of reduction in margins.

Many of the companies we spoke to told us that they expected the tariffs to be temporary. So if you think they're going to be temporary, you may be willing to take a hit to your margins for a while and delay it. And overall uncertainty just makes pricing decisions lower. So given all this, what can we expect today? I'm going to give you arguments for why I actually think the pass through can be stronger today. I've listed them here. I've mentioned them already. One is the size. Okay. This is one of the slides I have to update all the time. I don't know if I should put 150 5030. But these are very high numbers compared to the last time. We may not think of it now because we got 150. Now there are only 30. But these are these are huge. If they had just announced 30, we would have considered these quite big. And I should mention this is higher than what we saw retailers reacting to in the first trade war. There are more sectors and countries affected. Overall, it's broader effect. I think there's an expectation on many of these firms that something will remain. So there's uncertainty about the level. But we all sort of expect that some trade tension will remain. Some tariff level will remain. Even, by the way, if the Democrats come in because we saw Biden coming in and not removing the tariffs that Trump had introduced in the first trade war.

So that increases the willingness to pass it on to the consumers. And we do come out. And the fed has sort of mentioned this several times. When you hear them speak, they're worried that we just came out of a high inflation environment where firms have more flexible pricing. I think you're right. Firms are looking and attuned to their cost a lot more. And that could mechanically lead to higher pass through. But I'm going to give you the reasons why the impact on inflation will not be as large as many economists like to announce. Okay. And I'm an economist. I have to give you these two sides. This one is actually quite important. I think most people tend to over assume how much pass through there is. So let me go through them. First, a tariff shock is not as systemic as an energy shock like we saw in 2022. Energy feeds into the production of everything important inputs they do, but they don't matter as much. So it's not something that should affect us as much as I showed you at the beginning, fuel is actually which is a very important sector in the CPI basket, may actually end up lowering the pressure on inflation significantly. Second, trade diversion is still possible. If they had applied a tariff rate that is equal to everybody, you cannot do trade diversion. You cannot escape the tariffs.

But since there appears to be the case that we're going to get the differentiated one, I think this will will matter. And in fact, many firms, like I said before, have been thinking about doing this for quite some time. And perhaps the most important one, I think there's this huge uncertainty that we're always talking about in terms of affecting investment and consumption decisions. They also affect pricing. Firms will not make very important pricing decisions until they're sure what exactly the policy is and how that could affect their cost. So it's adding delays. It's going to make this pass through quite a slow. So that's the predictions based on the first paper, I'm actually going to show you what is happening now. And this is where I also ask you to help me interpret what we're seeing in the data. This is a new paper. We put it out on April 20th, and I've been updating it every week, adding a little more data. It's it's using price stats data. But we use essentially four large retailers where we know the country of origin, either because we can see it online or because we have this methodology to essentially ask ChatGPT or a generative AI model to find out what the country of origin is, which actually works really well if you think about it, a generative AI has the ability to search online, so if you give it a good product description, it will search online in somewhere.

It will find information about the manufacturer. Tell us where the good comes from. So we use that to fill in the blanks in our data with country of origin. And you're seeing here we have about 331,000 Thousand goods. They come from 108 countries. But given the nature of these retailers, most of those goods actually come from China when they're imported, about 36% of them. So I cannot tell you the names of these companies. But if you think of your mind right now, a very large retailer, chances are you got the right, the right one of them at least. Okay, so these are and think about the goods that these companies produce. They sell a lot of household goods a lot of electronics goods for the home. So those are significantly represented here, which is precisely where we would hope we would see the the effects of the tariff. So furnishing household goods, recreation and culture, which is essentially electronics is highly represented. Okay. So what I'm going to do is I'm going to build a price index with these subsets of goods and try to understand how they are reacting to the tariffs. So let me show you the first chart. This chart normalizes the price level in domestic and imported goods at one on October 1st of 2024. And then you're essentially seeing what the price level did in each one of these sectors. First, focus on the first part of the graph.

The graph on the left, you see this big decline in imported goods that then come up. Anyone wants to venture an explanation for that? Why did imported goods decline and then went back up? Sorry. Black Friday? Yes. Or seasonal, though that happens around December. In fact, it happens at the end of Thanksgiving. So it's a way of validating the data. You would expect to see this, particularly on electronics and on things that are imported. That explains that decline. Second thing you should note is these are downward sloping indices. That's actually pretty normal for these type of goods. Many of these goods get launched at high prices and they're discounted through the life of the goods. So in normal times you build an index with these goods they would have these downward trajectory that you see on both of them. Now look at those vertical lines. Those are the moments when tariffs were implemented or some announcement important announcement was made. The first one that you see there was March 4th. That's the day when the US actually implements the 25% tariff rate on Canada, on Mexico, and an additional 10% on China that had already applied at 10% in in February. Immediately after we see the retailers increasing prices. Now this is actually quite interesting. You see the red line jumps up. The blue line also jumps up. So domestic goods go up. Why would that be the case. Why would domestic goods prices increase if they're not affected by tariffs.

Speaker 6: Okay.

Alberto Cavallo: Raise your hand. It's like a class. Yes. Over there. Increase demand. What do you mean?

Speaker 5: Say more people are purposely avoiding important events because they're expecting prices higher.

Alberto Cavallo: Yes that's right. And in fact, prices are rising. So you switch your demand to domestic goods. That gives the domestic goods firm an opportunity to raise their price. It's more pricing power. We tend to to say okay, so that's one reason. Any other reasons? No not really. You've seen me present this many times. So you know the answer. Like put yourself in the shoes of a domestic manufacturer. Why would you raise your price even if you still haven't been affected by a tariff? You'll never be affected by a tariff. Probably because. Yes. Over there.

Speaker 7: Because you can.

Alberto Cavallo: Because you can. That's the argument. Yeah you can because the other prices are rising. Yes. You guys don't manufacture anything. Well, wait, let me put it this way. Imagine you're selling services. Probably financial services. Imagine your competitors are forced to raise their price. One is that you say, oh, there's going to be more demand for me. Actually, it's a bad example. Services doesn't work? No. What would be another reason? Yes. You have foreign components. Exactly. Well, you may have some foreign components, like the people you bring in to do your work and stuff like that. Like me. I'm providing an input to production and may be kicked out in a few months. But, um. Yeah. So if you fear that you're going to have some imported inputs, even, you know, you're saying this will affect me, that could explain some of what we see there. And in fact, there's a jump and there's they both start increasing. So if you look at the trajectories, it's more pronounced for imported goods but also for, for domestic goods. We start seeing those increases. The second line is actually Liberation Day. Um, it's interesting if you look at the imported ones, there was a jump and then it stabilized. And after Liberation Day, they just just start going up significantly on a new trajectory. So there's a sense here that the trade tensions are going to remain.

The third line is actually the China Post. So for a while we saw, you know, when when Trump announced that it was actually signing a pause until they can negotiate. We did see some of that decline. It has bounced back up. So bottom line we do see some rapid responses. This is more this is faster. I wasn't expecting to see this. This is faster than we had seen in the first trade war. These reactions happening within just a few days. Now. Domestic goods are also affected for the reasons you mentioned more pricing power, imported inputs, and simply just the uncertainty about whether these goods will be affected. But I want to be clear, the magnitude is still relatively small if you look closely at the y axis. That's an increase since the early March for imported goods of about 2.5%. And if you add the fact that maybe the downward sloping line would have kept on going, perhaps another extra percent, but 3.5% compared to a size of tariffs, that could be up to 125 or potentially 30, even 30% is not that big. And this is the uncertainty I think, filtering through a lot of firms are some firms are adjusting, but a lot of them are simply not doing anything yet. Okay. So that's important to remember. We do see spillover effects into unaffected categories.

In fact this line what it does is you have the red line is imported goods. You have the blue lines are domestic goods in categories that are directly affected competing with imported goods. And in categories that are not affected at all. And what's interesting about this chart is the not affected. Domestic goods are actually also increasing more gradually, but they are increasing. I think this is a strong sign of some of this uncertainty about whether there will be indirect cost pressures affecting them, and it could be imported inputs, but it could also be just a fear that inflation is going to rise. That's going to put pressure on many costs of distribution, many costs or potentially wages, and retailers are gradually adjusting to that. Now I can actually look more closely into imported goods and tell you where the pressures are actually happening. This chart splits the imported goods between China, Mexico and Canada. China is the red one. Mexico is the green one, which coincides with their team soccer team's shirt, by the way. That was a coincidence. And then the yellow one is Canada. So if you look closely there, their patterns are very similar until the tariffs appear. Then on March we see most of them jump up. So there's a reaction happening there. But then things really start to diverge. On Liberation Day the Chinese prices are the ones that start increasing the most followed by Canada.

But certainly the Chinese goods are the ones experiencing. And I think this is simply reflecting the fact that these retailers and potentially the manufacturers know that the tensions with China will continue. And, you know, maybe they're going to sign a deal with Canada. Maybe they're going to sign a good deal with Mexico. In fact, this suggests that they don't believe Mexico will get hit by any tariffs at all because the prices have kept on coming. But these expectations are playing a role in creating this divergence for the prices of many, many of these goods. I should mention also Canadian and Mexican prices, not just a matter of expectations that a deal will be signed. They actually got a lot of exceptions because Trump realized that, you know, we have a trade agreement already that he signed. So he gave a lot of exemptions to them as well. So this could matter a lot. I will also mention there are significant differences across categories. We are seeing these effects, but mostly on household goods on recreation and electronics. There's nothing happening in food, in sorry in food. Let me go back on that one. In fact, domestic goods prices have increased more during this period than imported goods. Much of this is because imported goods from Mexico, which we bring a lot of food from.

There are not really being affected, as you saw in the previous graph. And then finally, if I go even deeper, some of you, I was talking to some of you before ask me about goods I. Do we see anything on individual goods? I actually get that question a lot from journalists. So they're all obsessed about avocado prices. Um, so if you read the news in the last few months, everybody's trying to figure out about avocados. And I don't know if it's a millennial thing or what, but you know what is happening to avocado. So I did it. I said, I'm going to measure the price index of avocados, drawing from all the data we have at price. So this is what you're looking at. It actually starts in 2023. So you can see these cycles. Those are the seasonalities. And there's certainly been a spike and unseasonably high spike at the beginning of 2025, which leads to an inflation rate of about 10%. Okay, so but my point is be careful because this is a single good. Okay. What happens if I look for other goods that we bring from Mexico and Canada. So for example, what else do we bring from Mexico? Raise your hand. What comes to mind? Yes. Tequila. I thought exactly the same thing. I don't know what it says about us, but if I do that for tequila, there's just nothing there.

Okay. What do we bring from Canada? Who said it? Someone. Yes. Over there. Canadian. I didn't thought about that. Do we really bring whiskey? Okay. It's not alcoholic. Maple syrup. Very good. That's the first one I thought. Maple syrup. There's nothing there. There was actually a little bit of a spike, but then it came down. So I gave this to the journalists. They didn't want to publish any of it because it was not confirming the story. But this makes a point. We need these large data sets if we just focus on a few selected goals, you wouldn't get meaningful information. So a very specific one. All right. Good. So main takeaways that I want you to to keep in the aggregate index. We do not see signs of an acceleration yet. I do think there might be if we get some clarity on what the actual policy is and people are you know, the level of uncertainty declines. But it is important to remember this upward pressure will be contained initially in sectors that are directly affected by this and to some extent very gradually in other sectors as people anticipate pressures. I do think it's important to remember this is not just a potential shock about tariffs. This has created uncertainty potentially can affect demand, may make firms decide to scale back their investment and make consumers more afraid of the future.

If this becomes a negative demand shock as well, pretty much like we had at the beginning of the pandemic, we won't see much happening with inflation, because we'll have these two forces compensating each other and particularly fuel. If it comes down, oil comes down. We will see that putting a lot of downward pressure on the aggregate index. So those sectors will compensate. We obviously have to wait to see what happens, whether we enter in a recession or not. But remember, it's a supply but also a demand potentially shock. The microdata results that I showed you are actually very consistent with what we found the first trade war. The effect is going to be gradual. Even if it's large, it's going to be gradual. We won't see it happening right away. There's going to be some quick responses, particularly in sectors that are clearly being affected. We also get domestic goods that are impacted even in affected categories, but the magnitude is still small relative to the size of these tariffs. And we should expect that to happen very gradually over time. I actually think this is very important for political reasons. You know, when when people say, oh, there's going to be a lot of inflation and then inflation doesn't show up, it looks like the policy did not hurt American consumers even though it is hurting them.

It does so very gradually that it doesn't become in the minds of the consumers, connected necessarily to the tariffs. So it makes the tariffs quite popular in the short run. So that's actually very important. I get asked a lot about other effects that are not just prices. So for example, even if prices don't increase because firms don't find a way to to justify them and, and pass it on to consumers, they may decide not to bring that many goods from, from other countries. And that could shrink product variety. We have been looking at the data. There's still nothing there. The retailers have spoken. Even we saw Walmart and others go to the white House and they said, if you don't reverse the policy, we will see empty shelves. That's still not in the data. I just want to be clear, but we're going to keep an eye on that. And overall, I want you to remember there's just so much uncertainty that I think we should not expect much happening until some of it goes away. There's going to be delayed pricing decisions and and pass through. Okay. So I'm going to end it there. Thank you so much I'm happy to take any questions. Yes. Thank you. Yes. You can clap.

Speaker 8: Thank you Alberto. My question is about inventories. I know you also track them. Do you see anything in inventories? Are we running out of products?

Alberto Cavallo: No. That's good. So we we can actually not see how many inventories firms have. What we can see is when inventories run out and then we get stock outs. Um, that's what I mentioned at the end that we don't see that happening yet. If you remember, we used to have this indicator for the pandemic. And there it was very quick. We saw out of stocks happening right away. And then many goods started disappearing from the stores. That is not yet happening here, in part because I think some of these tariffs have been scaled back. And we saw an increase again in some inputs. And and I think the inventories are probably large enough to get us through the next few months in many of these categories. But it is an important dimension, I think, um, that may become noticeable to consumers in America. Perhaps there's so much variety that people won't mind having a few, you know, a few less options. But if this remains for a long time, you may end up seeing a decline and it's not yet in the data. Yes.

Speaker 6: Oh, sorry.

Speaker 4: I was just going to remind folks, you can also submit questions through the Q and A. And so there's some on here as well.

Speaker 9: But so um, I saw your chart about the price rally when the Trump administration announced that there's a 25% tariff. Right. But then I wonder why. Because for importer exporter, they probably know Trump will do something. So why don't they just react before the announcement?

Alberto Cavallo: Oh, before the announcement? Well, I think it's part of this uncertainty that he makes an announcement and you're not sure if it's actually going to implement it. And he does implement it. So as soon as he people saw that that was in the code and implemented, that's when we saw the reaction. So I think it was more a matter of not believing that, you know, because Trump all the time says this is a negotiation strategy. And the more he does it, the less likely it is you're going to believe it until you actually see it happening somehow. Or he makes an even more outrageous type of policy announcement like on April 2nd. So I think that's part of it. The other thing that is connected to that, by the way, is that if you think about it, none of those firms have actually paid the tariffs two days after the tariff is implemented. So why are they raising their prices? That's very normal. You try to smooth out whatever increase you think you're going to have. It doesn't make sense to wait until you get the new inventories that came with a higher cost because that would create a big jump in the index. So you do it sort of gradually. One dimension that is, I think for them hard right now is you don't know where you need to end up. It's going to be at 30. Is it going to be at 50. That is actually going to be part of the the explanations for these bumps that you see along the road. Yes. Over there.

Speaker 10: So it looked like a lot of your data focused on the retail sector. I'm very curious about the auto sector, given that I'd say Mexico plays a significant role in the American auto industry. Very curious about your thoughts on inflation there with regards to policy.

Alberto Cavallo: Yeah, that's a great question. So we don't have prices of cars ourselves here, but we do have a lot of auto parts. And I believe we haven't yet seen much happening there. But but the tariffs have been implemented and Then that might be part of the some exemptions that are being given. But it's a great question. Um, I'm in the next version of the paper. I'm going to try to have a subsection on auto parts. So so you can have some more information.

Speaker 4: Why don't we take one from that's come through on the tablet. So you've talked a little bit about um, price changes in in aggregate from different sectors. But do you see a change in or does it matter if a firm is a larger company or a smaller company. And that that how does that impact tariff pass through?

Speaker 11: That's a great question.

Alberto Cavallo: So larger firms have a lot more options to absorb the tariffs. 1st May be that they have higher margins and the other is they can actually stockpile or do trade diversions in ways most companies cannot. So the graphs I showed you about stockpiling and trade diversion, Those came from two large retailers that they can import goods themselves. So that's evidence that large firms have ways to adjust. Smaller firms may not. But those smaller firms compete with these large companies and they have to react to their prices. So they're actually going to delay some of that pass through. And many of them will. You know, there are claims that many of them will actually suffer greatly if this is not scaled back. But I don't expect them to necessarily be a big driver of, of, of inflation that we may not be capturing here because we focus only on large firms. Traditionally, large firms are the ones that drive most of that volatility. You end up seeing in the aggregate CPI.

Speaker 12: Is it surprising to you that given the exchange rate moves we've seen, right, the dollar has depreciated quite substantially since Liberation Day that you haven't seen that reflected in prices. Is that consistent with what you'd expect?

Speaker 11: Actually, that's a great question.

Alberto Cavallo: Normally when we saw tariffs before we would actually see a currency appreciate. And that's what happened in the first trade war. The renminbi depreciated. And when the renminbi depreciated you think of the Chinese exporters. They actually have an incentive to lower the price in dollars, because those dollars that are bringing in make, you know, allow them to pay more their workers and stuff like that. This time it's actually flipped around. So we saw the tariffs and we see a depreciation in the dollar that in principle, in theory could help alleviate some of the price pressures of the consumers. When a currency appreciates then on the import side the tradable goods can actually fall more. But just like it happened the first trade war, the pass through from changes into prices in the US, I think it's going to be very, very low. And that's specific to the US. And it's connected to the role of the dollar in the trading system. The fact that, you know, importers and exporters all over the world sign contracts in dollars means that you get an exchange rate movement that doesn't necessarily affect the price in dollars that the US faces. So I think both it doesn't matter the direction it should help in terms of inflation, but it won't matter much in magnitude because of this nature of the dollar being the trade currency of the world. It will matter in other countries. By the way, if you're thinking of inflation in Europe, that will matter more. If you think of inflation in developing countries. There are all sorts of effects here in other countries that we haven't discussed today, but I'm actually out of time. So I'm happy to talk to you about some of the other research we're doing on on those dimensions as well.

Speaker 4: Alberto, thank you so much.

Alberto Cavallo: Thank you.

Speaker 1: We have Dan Drezner. Dan is professor of international politics at the Fletcher School of Law and Diplomacy at Tufts University. He is also a non-resident fellow at the Chicago Council on Global Affairs and co-director of Fletcher's Russia and Eurasia Program. He has previously held positions with the Civic Education Project, the Rand Corporation, and the US Department of the Treasury. He has received fellowships from the German Marshall Fund of the US, the Council on Foreign Relations, and Harvard University. He has also found the time to publish seven books, including All Politics is Global and Theories of International Politics and Zombies. I think there's another book in there coming out this afternoon that you can work on. Um, today, though, he is going to face his most daunting task to date and take us into the mind of Donald Trump. So looking at the first 100 days of the administration and what that tells us about what we can look forward to from here. Um, now, I heard this speech in New York recently, and it got really dark at one point. But today he's going to be more upbeat. He's going to be more twitterfall. So Dan, please join me on stage. And I promise as you're speaking, I'll keep you up to date with what's going on. Okay.

Speaker 2: Thank you very much. And thank you very much to Lee as well. If Lee and I ever get into a Twitter war, it is going to be because he's had to read that stupid bio of mine again, again and again and again. And it's just too long. And I do apologize for that. Um, so I'm going to talk a little bit about why Trump 2.0 looks so different from Trump 1.0, although we're starting to see some resemblance now. Um, the sort of known knowns we have about Trump's first 100 days, uh, what the effects have been to date, what Trump's actual constraints are. Because one of the themes about Trump 2.0 is that he has fewer of them than he did during his first term. And then what this implies about the future. Also, I'll ask all of you. You know, by the time I'm done with this, you're going to know more about the state of the Trump Musk flame war. Just the first question. If someone can just update me, I would appreciate that. Um, the thing to realize about Trump's first term is that when he came in, he wasn't expecting to win. You know, remember, Trump wins. And the very first thing he does is he fires his transition director, Chris Christie, which meant he was behind the eight ball from the get go. He had no idea how to staff his administration. And he wound up staffing them, staffing it mostly with sort of conventional Republicans, which Donald Trump is not.

Speaker 2: And so, as a result, a lot of Trump's first term was an administration at war with itself. All right. There were Trump's policy instincts. He had to deal with Congress, which was dominated by a GOP that wasn't entirely Trumpy. He had to deal with courts. He had to deal with the civil service that was not used to dealing with what Trump wanted. And he had to deal with the so-called adults in the room, one of which we'll be hearing from after me, that try to sort of guide Trump's impulses into somewhat more normal policy this time around. And part of the reason we've seen so much activity, let's put it in the first 100 plus days, is that Trump just faces fewer constraints than he did in 2017. And this is for three reasons. The first is the nature of Trump's victory. The second is what Trump has done to the GOP. And the third is that even outside of the Republican Party, there are fewer guardrails on the president. And this is a trend that long predates Trump. So in terms of the Trump victory in 2024, it wasn't an overwhelming victory. I mean, he won by a couple percentage points. I think it was something like the second narrowest victory of this century. But what was striking was the degree to which almost every county, not all of them, but something like 80% of the counties in the United States, the vote shifted towards Trump.

Speaker 2: So even in the states where Kamala Harris won, she won by a smaller margin than Joe Biden did in 2020. So this was clearly a case where you could argue Trump didn't just win. He won in a way that people were not expecting prior to the election. And, you know, again, elections have consequences. And the result of his victory is that, um, you know, it was seen as more legitimate than what happened in 2016, when you could have dismissed it as sort of an electoral College fluke. Another difference is that in contrast to 2017, the Republican Party in 2025 is fully magnified, essentially. In other words, you have anyone who's in a Republican in Congress or anyone who is serving in his administration, is someone who legitimately and genuinely believes in the kind of populist nationalism that Trump has espoused, very different from, let's say, the George W Bush brand of republicanism, um, that you would have heard, uh, even ten years ago. And indeed, yesterday, if you heard Kevin Warsh speaking, you know, the Kevin Warsh of today in terms of what he was articulating, sounded a little bit different from the Kevin Warsh of a decade ago. And that's because Warsh is, you know, fully bought into MAGA. Similarly, in contrast to 2017, Trump this time around had four years out of power to essentially build a policy apparatus of folks who actually believe what he is selling. So, you know, you're all probably familiar with project 2025, but also, Trump set up a Russell Vought, who's now the OMB director, set up the America First Policy Institute.

Speaker 2: You saw the creation of think tanks devoted to implementing what Trump's policy instincts were. And the Heritage Foundation basically went full MAGA in some cases, causing former Heritage Foundation fellows to resign because the think tank was no longer espousing what they were they believed in. So as a result, in contrast to 2017, it's not just that Trump won. He won with a staff that actually wants to implement his policy agenda. And you can argue the sort of pivotal transition moment, the moment that demonstrated that Trump was going to have more political capital this time around than last time was when he nominated Pete Hegseth to be secretary of defense. And, you know, ran into some trouble from Joni Ernst, the senator from Iowa. Now, Ernst is a sort of rock ribbed Republican, but one of Ernst's sort of core signature issues has been dealing with sexual abuse in the military. This has been a signature issue for her. She's worked across the aisle with Kirsten Gillibrand of New York and Hegseth because of his prior track record. Was sort of a poster boy for someone you don't want to put in the Defense Department if you care about sexual harassment. Ernst went on television after the first meeting with Hegseth, sort of saying, well, I don't know about this guy at this point. I'm not a yes.

Speaker 2: At which point Donald Trump and at the time, Elon Musk, sort of unleashed hell within Iowa's congressional districts, causing Ernst's phone lines to ring. Ring constantly, basically saying, you better confirm him, you better confirm him. Ernst flipped within 48 hours, saying that she would be prepared to endorse Hegseth. And the moment Ernst did that, the remaining Republican opposition and there was some legitimate queasiness to this, um, mostly melted away because Ernst wasn't going to vote for against Hegseth. You know, she wasn't providing political cover for other Republicans. And so that was the closest vote. Um, and he still got through on a tiebreaker. All of other Trump's other appointees, with the exception of Matt Gaetz. Um, to put it this way, the ones who all went to a vote all got through. The other thing to realize is that over the last 75 years, honestly, there has been a steady shift in terms of the, you know, the balance of power, in terms of separation of powers from the legislative and judicial branches to the presidency. Um, you know, the The Constitution empowers Congress with the right to either both set tariffs and declare war. And essentially, over the last 100 years, Congress has said, we don't want this responsibility anymore. We will give it to the president. And as Congress has become more and more politically polarized, Congress has responded to this by essentially delegating more and more powers to the presidency in the thought that the president was the last adult in the room.

Speaker 2: And so, as a result, presidents now in the 21st century have considerably more leeway to do things without congressional checks than was the case 50 or 100 years ago. This is particularly true if your party controls both the presidency and Congress. There's just going to be no investigation or check on presidential power. This is also true because you're seeing Trump embrace a version of what is called unitary executive theory, which basically says that as President, Trump can do whatever he wants within the executive branch. And that is an argument that is somewhat rooted in the Constitution, but not entirely. There's supposed to be things like civil service reforms or protections and what have you, and he's sort of running roughshod over that. But also, Trump has made it very clear that he is on vengeance 2 or 2025 and a fear of government retribution is also cowed. Members of Congress and, to a lesser extent, the courts in terms of dealing with Trump. Lisa murkowski, who's a Republican senator from Alaska, admitted, you know, in a town meeting that she was afraid to speak up in criticism of Trump. If, you know, if Senator Murkowski is afraid, you can imagine what the ordinary citizen feels like if they want to criticize him. And the result is, is that the GOP is, by and large, gone along with whatever Donald Trump has wanted. And the Democratic Party has been so dispirited and demoralized from what happened in 2024 that they are still, you know, engaged in shooting each other in the foot, you know, sort of revelations about Joe Biden more recently, an example of, you know, that kind of debate, which is almost separate from Trump.

Speaker 2: So when we talk about Trump 2.0, when we talk about what he wants to do, you have to realize that he has far fewer checks on his preferences actually getting implemented into policy than was the case back in 2017. And although this is not a sentence, I often say Donald Trump does have a few core policy convictions. He really does. He believes in three things very clearly. First, he is convinced that trade deficits are a sign of geopolitical and geo economic weakness. He thinks that if a country runs a trade deficit somehow, it means we're giving money away to other countries and we are being impoverished. Now, anyone who has had econ 101 knows that's not quite how it works. But nonetheless, this is what he has believed since the late 1980s. And because the US has run a persistent large trade deficit for the entire 21st century, this is something he wants to correct. The second core conviction that Trump has is that US allies have long been free riders off of the US economic and security umbrella. He thinks that US allies in Europe and the Pacific Rim don't pay enough in terms of defence. Don't kick in enough in terms of global public goods.

Speaker 2: He thinks that the liberal international order has screwed over the United States, and he wants to rewrite the rules such that you have a greater amount of burden sharing. And the third core conviction that Donald Trump has is that he is antithetical to multilateralism. He is extremely suspicious of any sort of multilateral institution, because he feels that any multilateral institution is a constraint on the United States, and an opportunity for other countries to gang up on the United States. So part of the reason that you can argue the Trump administration did this sort of Liberation Day tariffs and sort of says, okay, we'll set up 90 days to negotiate with trade partners, is he wants the United States to negotiate one on one with other countries rather than within, let's say, the say, the World Trade Organization, because in a WTO setting, these other countries can talk to each other, whereas at least in his mind, if you do it one on one with the United States, the United States is usually going to have the bargaining advantage in terms of Trump's management style. There's no denying that Trump is better prepared in 2025 than he was in 2017. Donald Trump has had zero experience in government prior to being elected in 2016. This time around, he's been president for four years. He's a little more familiar with the levers of power. He's learned some not a ton of lessons, but learned some of them.

Speaker 2: And on the issues that he cares about, he does know something about the world in a way that he didn't back in 2017. Now, that said, Donald Trump is still Donald Trump. All right. He is an impatient person. He has extremely poor impulse control. And he's the oppositional behavior of a four year old. All right. Now, that said, in his interactions with Elon Musk, Trump might have found the one person that is less mature than he is. Years. And so this might be an instance where he actually has a bargaining advantage. I'm just going to point that out. Trump is also a firm believer in the madman theory. The madman theory, which was coined by Richard Nixon, was the idea that when you engage in international bargaining, if you can act like the crazy person, you might actually get other countries to make concessions that they would not otherwise make because they think that you are crazy enough to do the really dumb thing that would hurt both of you. And so he's been very clear on this point. He thinks that if he can act crazy, he can cow China, he can cow the European Union, he can cow other allies. The result is a sort of split presidency in terms of Trump 2.0. All right. On the issues that Trump cares about. There is much less likely to be freelancing, which was the case in Trump 1.0, where you very often had sort of policy subordinates like, let's say Gary Cohn going off the reservation, for example, or hiding paper from Trump, so he couldn't act on things this time around.

Speaker 2: On the issues that Trump cares about, you are much less likely to see people sort of going off the reservation. All right. On trade, on immigration, on alliances, and on dealing with other leaders. This is Trump in his element. This is the stuff he actually cares about. On the other hand, there are a whole array of government policies that Donald Trump has zero interest in. And in those areas, you are still going to see a replay of Trump 1.0. Those are the areas where policy subordinates are probably going to be able to do whatever they want. And unless the issue gets to the top of Fox News, he's not going to care. Okay, so on issues like antitrust or cryptocurrency or regulation or health care, Donald Trump, by and large, is not intrinsically interested in these things. He will get interested in these things if they become a news peg. Otherwise, he's not going to care. In terms of the effects to date? Well, you know, doge is doge, and we've sort of seen what Elon Musk did to the federal government and that he pretty much did to the federal government what he did to Twitter when he took it over, which is he tried to engage in mass layoffs and buyouts, much like Twitter. He tried to sell the leases that government buildings had or sell government property to describe what Musk did in terms of using AI as a crude management tool, that actually is an insult to AI.

Speaker 2: What Musk actually seemed to do, or what the Doge team seemed to do, was sort of control F searches for things in terms of looking to, to cut. We've obviously had service disruptions in terms of things like air traffic control or weather forecasting. And unsurprisingly, this hasn't necessarily worked well because and as a political scientist, I feel like I can actually say this the federal government is not a private corporation, and trying to run it like a private corporation is probably not the way to go, because you expect different things from government than you do from a private sector corporation. Operation. Nonetheless, there has been a considerable amount of carnage that has been wreaked. And while the courts have stepped in as a constraint, and I'll talk about this in a little bit, there is sort of a Humpty Dumpty problem here, which is there are many instances in which Trump and Musk, when he was running Doge, have succeeded in destroying government institutions like, let's say, USAID, the US agency for International Development. And even if a court eventually rules that Trump can't do that, you have the Humpty Dumpty problem, which is once you've broken the egg, you can't get the egg back in the shell again. And so even if courts eventually potentially check on Trump in terms of what he's doing to the executive branch, it's not clear how much of that can be resurrected.

Speaker 2: We're seeing this now, for example, with the US Institute for peace, which is an independent think tank that Trump said that he could take over. A court ruled no. But that was about a month or two after Trump did what he did. I have no idea what's going to happen to us now. So we're more than 100 days in. How is Donald Trump doing politically? Not so great. Um, you know, he's underwater, uh, in sort of a significant decline. The only presidency that did worse than this in this century is Donald Trump's first term, where he was somewhat more unpopular at this point, but nonetheless, he's underwater. We still have a massive amount of policy uncertainty going forward. Um, there's another slide that will sort of show this. But again, what is what is shocking is the degree to which this dwarfs what we saw with either Covid or for that matter, the 2008 financial crisis. And we also have a really odd jerry rigged government at this point, which is at this point, Marco Rubio has four jobs. He's the secretary of state. He's the acting national security advisor. He's the acting USAID administrator, although admittedly, that's not that big of a deal at this point. And he's also the acting national archivist for some reason. Okay. He's not the only one, by the way, with multiple job titles. So this gentleman named Daniel Driscoll, for example, who is simultaneously the Secretary of the Army and the head of the Bureau of Alcohol, Tobacco and Firearms.

Speaker 2: And then there's Jameson Greer, who is the US Trade representative, which you would think would be a full time job at this point, who is also directing the acting director of the Office of Special Counsel in the white House and the Office of Government Ethics. And he's doing a bang up job. Let me just say. So I'm sure you are all familiar. If you read the Financial Times with the idea of the taco trade, the idea that logic that Trump always chickens out. Now, I think this is actually unfair to Trump. Trump is is, you know, when he feels like he's in a situation where he can win, he's not going to chicken out. He actually likes to bully smaller countries and or weaker actors. He will go to the max on that. But in any instance where he actually recognizes that he might feel some pain, there is a suggestion that he will back down. What I am curious about, particularly with respect to the tariffs, is whether or not we're going to be shifting from Taco to what I call talk, which is Trump always loses in court. In other words, we're now increasingly seeing, particularly on the tariffs, Trump running into a sort of a head wall in terms of the judicial branch in no small part. It's not that Trump can't raise tariffs if he wants to.

Speaker 2: It's that the way he did it, namely through what is called Iipa, the International Emergency Economic Powers Act was such a sort of rush job because Trump is impatient that he likely violated the rules of doing it. And so he will probably wind up losing in that area. Which doesn't mean that Trump can't raise tariffs. It's just that he's going to have to find a slower means of doing it. More interestingly, you sort of have to wonder about the state of state capacity. So this is a picture of the abbey Library in Saint Gallen, Switzerland. I took this picture. Actually, I'm very happy because I got to teach a short course on economic statecraft there in April. It is a gorgeous library. If you ever go to Zurich. Saint Gallen is only an hour away by train. I highly recommend it. This library is extraordinary because it was basically set up around 800 A.D., and since then it existed as a sort of repository of knowledge. And any time, any sort of, you know, actor came close to ransacking the Abbey library, the monks would immediately take all the manuscripts and try to hide them to make sure they didn't lose that sort of institutional knowledge. And to be blunt, my concern about what Trump and up until last week Musk was doing to the federal government is that we are beginning to see the erosion of institutional knowledge and institutional memory within the federal government.

Speaker 2: You're seeing the sort of senior people depart. And you're seeing increasing difficulty in terms of data collection and data dissemination. I'm sure you all read the Wall Street Journal story yesterday about the difficulties in terms of collecting consumer price information. And we're now increasingly seeing reports that are being delayed in terms of going out because they are politically inconvenient. For example, the US Department of Agriculture delayed a report by a couple of days, revealing that in fact, the US farm trade was actually a bigger trade deficit now than it was last year because that was politically inconvenient for Donald Trump. So what are Trump's hard constraints? Right. I think there aren't that many of them. As I said, politically, he's not facing huge amounts, but there are three. The first are the bond markets, the second is public opinion, and the third is the international system. So on the bond market, I'm going to defer to you all. You all know a lot more about the bond market than I do. All I will say is that what is striking to me about Trump's first 125 days has been the degree to which the stock market has seemingly reacted to sort of news stories in an almost hyperbolic fashion, either overly optimistic or overly pessimistic. Whereas the bond market, by and large, has been far more, you know, slightly gloomier. And what was most interesting was during Liberation Day. And in the reaction to that, it wasn't just the stock market fell.

Speaker 2: This was normally a moment where in sort of high periods of geopolitical uncertainty, you would have the so-called flight to quality. This happened in 2008. This happened in 20 2020, in which when it seemed like everything was crashing down, people would rush into the dollar because the dollar was seen as the safest asset and because treasuries were seen as the safest, deepest, most liquid capital markets in the world. And that didn't happen in April. And that's interesting. It's not to say that I think the dollar is finished. I really don't, because it boils back to Tina. There is no alternative still, but nonetheless, the fact that you even had this sort of perturbation means that for the first time, you've got people thinking, well, what could happen? Post if Trump does something else? Could it be the case that, in fact, the dollar will no longer be the safe haven. And if you start people seeing people hedge about that, that's going to cause a few priors to have to be recalculated. Second constraint is public opinion, which Trump is still sort of captive to. It's worth remembering he has a razor thin margin in the house. You know, the one big beautiful Bill act, which is actually the name of the legislation passed by a single vote. It's going to have to go back to the House after the Senate does something to it. And there's a question about whether or not it's going to get through.

Speaker 2: This is part of the reason why Trump wound up withdrawing Elise Stefanik to be his choice to be the US ambassador to the United Nations. He couldn't afford to leave that House seat empty. And similarly, if senators start sensing that Trump is not all that popular and they actually start standing up like Tom Tillis did when it came to whether Ed Martin was going to be the US attorney for the District of Columbia the moment he said, no, Trump didn't go ballistic. He didn't try to browbeat Tillis. He basically just sort of accepted it as a fait accompli. And so there are instances where Trump, recognizing the handwriting on the wall, is not going to be able to bully others, particularly if he's less politically popular. And then finally, there's the international system. And here, the important thing to remember, there are sort of two things that are worth remembering. The first is that for a lot of key leaders on the global stage, this is not their first go around with Donald Trump, right? Xi Jinping, Vladimir Putin, Emmanuel Macron they've all dealt with Trump for more than a decade now, right? They're used to his shtick. And so they're not going to be thrown when he tries to pull his madman theory because they've seen how he reacts. The other issue you have to deal with with respect to Trump is that with, you know, in terms of dealing with other democracies, Trump is actually sort of had a negative effect.

Speaker 2: Poland actually is somewhat of an exception because recently the Polish presidency went towards a sort of MAGA favorite. But by and large, the elections that have been held in 2025, in Canada, in Australia, in Romania and elsewhere have gone against the candidate that Donald Trump and JD Vance have boosted. In other words, negative polarization is no longer just a domestic political phenomenon. It's an international political phenomenon. And so as a result, Trump, by acting the way that he did, managed to guarantee that a conservative would not win in Canada and would not win in Australia either. So in terms of outcomes, I think relative to what the market perhaps expected in January of 2020, I think the expectations on deregulation and on taxes and on crypto and maybe on reducing the size of government were pretty much accurate. These have sort of, by and large, gone, as you would have expected, coming from what Trump said during the campaign. Where I think, however, the market was not, you know, clearly clued in was that Trump was serious about tariffs. He was very serious about immigration restrictions. I think they actually underestimate Trump's tendency to use force, not, um, you know, not so much against China, but certainly in the Middle East and also in Latin America, and also just the persistent degree of uncertainty that Trump has done, especially with respect to tariffs. Now, part of the issue here is that, as I said, Trump loves tariffs.

Speaker 2: I mean, just loves them. But the problem is, is that he loves them for mutually contradictory reasons. Right. Donald Trump thinks of tariffs the same way. The father in My Big Fat Greek Wedding thinks of Windex. He thinks it cures everything. He thinks it can do everything right. Donald Trump thinks that tariffs can be useful as a tool of coercive bargaining, as a means of industrial protection, and as an alternative means of raising revenue as opposed to normal taxes. The problem is, at best, you get one of these three things. If you think the tariffs are a tool of coercive bargaining, you can't keep them up as a tool of industrial protection, and you can't keep them up as a tool for revenue collection. Because if the other side acquiesces, you have to lower the tariffs if you want to use them as a tool for industrial protection, to create more industry in the United States. Well, they can't be a tool of coercive bargaining because you're not going to bargain. And they also eventually can't be a tool of revenue collection, because the whole idea is to create a domestic substitute industry that eventually Americans consume. And if you think of tariffs as a means of collecting revenue, well, then you can't create a domestic industry that's competitive with imports, and you certainly can't use it as a tool of coercive bargaining. And the thing that I think we are still not entirely sure about is which of these three things does Trump want to use tariffs for? And then of course, there's just sort of the mass uncertainty.

Speaker 2: This is actually I think from I think this is up to date, but this sort of shows all the sort of fluctuations in terms of Tufts. Tufts. Tufts. Trump's tariffs. Announcements. Sorry. Um, you know, over his second term, his office. And the truth is, with this kind of fluctuation is extremely difficult, you know, for businesses, particularly small businesses, to make any kind of calculation or any kind of long term plans in terms of investment. And we are now, I believe, you know, more than halfway through the sort of 90 day extension on the Liberation Day tariffs. Um, the US Trade Representative, I think, actually gave all trading partners until yesterday to sort of send in their best offer. And this leads to what I like to call the curious incident of the trade negotiator in the nighttime, which means we have not heard yet from the Trump White House whether there have actually been any deals announced or any offers put in. And that silence is deafening in the sense of, I don't know whether or not other countries are actually taking this seriously, in no small part because the courts have suggested that Trump can't do what he does. So what is the future going to look like? Well, we're probably going to see some crises. Um, air safety is obviously an issue.

Speaker 2: I wish Godspeed to everyone flying back to Newark. Um, in terms of epidemics and pandemics, what we are seeing coming out of HHS and the CDC is disturbing. And whether or not we're actually going to have any data in terms of any kind of new pandemic is also problematic internationally. Shockingly, Ukraine has not been solved in 24 hours. Um, and it now seems increasingly that Trump's attitude is something along the lines of, well, I'll just let them play, fight each other and then we'll sort it out. Um, Gaza is going to continue to be a problem. No one's really talked about Kim Jong UN all that much. Um, for the last 125 days, which is always a good sign that Kim Jong UN is going to do something so that people start talking about him. Um, he likes getting attention. And I do think that we are, you know, not quite realizing the degree to which Trump wants to take action in Mexico, particularly with respect to drug cartels, as a way, and also sort of coyote smugglers in terms of halting illegal immigration. I am very concerned about the long term effects on US economic dynamism. This has less to do with tariffs and much more to do with immigration restrictions. And what we've seen in terms of Trump's jihad against Harvard and other institutes of higher education. Now, I admit I'm a professor, so I've got some skin in the game here.

Speaker 2: But I guarantee you that one of the reasons, one of the sort of, you know, comparative advantages of the United States has always been that we tend to attract the best of the best in terms of people who want to get PhDs and people who want to do stimulating research. If we make it harder for those people to come into the country, they're eventually going to have to find alternatives. And then we sort of lose the sort of the idea of the clustering effect, we lose the fact that people come here because everyone else is coming here as well. And it really doesn't take that much to disrupt this. All it takes is you've got to disrupt the visa system and cancel research grants and basically make it harder for universities to be the repositories for cutting edge research. And I am also a little concerned about the effects on national security. The analogy I would make is that you should think of the sort of US security architecture as like a Jenga tower. And with changes being made in terms of the elimination of USAID cuts to the State Department, changes in the Defense Department, it's like taking out 1 or 2 of these little pillars in a Jenga tower. The tower can still last, can still stand for a while, and then eventually it all comes crashing down. And that is honestly my concern. My concern is whether or not the United States will actually be able to handle a crisis.

Speaker 2: It's extraordinary to me that a week ago, Pete Hegseth, that Secretary of Defense claimed that China's actions was China was going to have to imminently take military action in Taiwan. And no one paid attention because no one took Pete Hegseth seriously. In terms of wild cards. The problem we have is because Trump has fewer checks on him. There is simply more variance in terms of his behavior than there was in the first term. Again, he really does believe in the madman theory. I do think we're going to see a more militarized foreign policy in this hemisphere. And I also think you're probably going to see some nuclear proliferation among US allies less so than US adversaries. As you're seeing countries in NATO and Japan and South Korea realize they can no longer completely count on the US security umbrella. The cheapest way for them to create deterrence is, in fact, to develop nuclear weapons. And then finally, whether or not there's going to be a third term, I really don't think there is. But Trump does talk about this in these sort of half joking, half serious way. And again, I think there's a 98% chance that this is not going to happen. But the very fact that I think there's a 2% chance that he might try to sort of force something through is not good. And generally a rather disturbing sign. And so, in conclusion, I am happy to take some questions. Thank you very much.

Speaker 1: So do we have any questions in the room? Please raise your hand. Oh one there. Yeah.

Speaker 3: I was just sorry. I was just wondering if you had any thoughts on whether or not the Republican Party heading into the midterm elections. Do you think they were going to try to separate themselves from Trump a little bit, or sort of hedge against maybe more unpopularity or like more popularity depending on it because it seems like it moves day to day, so I don't have to kind of take a bet against whether he's going to be popular at the time of the midterms if they're going to, you know, handle that in any particular way.

Speaker 2: I mean, I would say a couple of things in this. The first is bear in mind that almost nothing that is happening right now will matter for the midterms. Part of this is because, you know, the American voter doesn't necessarily the American any. Most American voters are not paying attention right now. Um, and nothing that happens right now is going to affect necessarily what they're going to do in terms of voting in the midterms. Um, that said, what they do care about is whether Donald Trump is going to make their life difficult by primarying them. And so the really interesting question is, once the Oba goes back to the house, to what extent does Trump, even more than Mike Johnson, try to use his leverage to get members of Congress that are reluctant to cast this vote to eventually do so? Um, in terms of the actual midterms, one of the sort of underrated dynamics that have happened over the last decade politically, is that there's been an inversion of what used to be the case. Um, when I was your age, frankly. Which is it? It used to be the case that during Non-presidential years, Republicans would always outperform Democrats. And the reason is, is that when the president's not running, when there's not a presidential election, the people who are much more likely to vote are people who have more education and senior citizens.

Speaker 2: And those two demographics always trended Republican. The one change over the last ten years has been that there's been an inversion of that. Those two groups are now more likely to vote Democratic than than Republican. It goes back to this dynamic that happened during 2024, which is if you take a look at voters who said they were paying attention to the election, Kamala Harris actually won those voters by something like 55 to 45. But if you for the 25% of voters that said they actually weren't paying that much attention, but nonetheless cast their vote. Trump won those that group 75 to 25. That group is less likely to come out in the midterms, no matter what Trump wants to do, because his name is not on the ballot. So I think regardless, um, Democrats should win back seats. I suspect they will win back the House. I'm much more skeptical than winning back the Senate, because the states that are sort of up for contention in the Senate are a much more difficult path for for Democrats to win.

Speaker 1: Anyone else in the audience at the moment with the question. Okay, let me take someone from the iPad. Um. What are the chances the big beautiful bill passing and when?

Speaker 2: Um, this is actually where the Musk Trump blow up matters a little bit. Um, because Elon Musk railing against this bill. I mean, Elon Musk is one of the, you know, someone who's actually less popular than Donald Trump. So you would think it would not matter that much. But him railing against the bill does create a news peg, and it actually does, in theory, embolden some Republicans who were not thrilled about this bill to potentially resist it. That said, in a cage match between Musk and Trump, I would actually put my money on Trump in no small part because he holds more levers of power. It's worth remembering the degree to which Elon Musk's wealth is tied up in federal government dollars, and the lack of federal government prosecutions. Um, and so it'll be very interesting to see what happens if this, uh, dispute escalates further. Um, so I suspect that he will be able to get it through the Senate. The question is whether the Senate version will then get back through the House, because what the Senate is likely to do to it is strip the Medicare cuts that got the fiscal hawks in the, in the the House Freedom Caucus to buy onto it. That said, you know gun to my head. I think it'll pass if for no other reason that this is the one bill that Trump wants from Congress. And you know, this is where Republicans decide are they going to hang together or are they going to hang separately? And I think on this, they're going to hang together. I think what's been underrated, frankly, has been the degree to which Mike Johnson has actually gotten things through a house that is even more narrowly in his favor than, let's say, Nancy Pelosi dealt with four years ago.

Speaker 1: So here's another one. Considering Trump won the popular vote while campaigning on a populist platform, how do you think the left will respond in the coming years, both strategically and rhetorically?

Speaker 2: The left will respond in a mature, disciplined, you know, forward leaning. I'm sorry. I couldn't resist that. You know, the the left will respond frankly to some extent much the way the right is responding. It's the end of Reservoir Dogs. They're going to shoot each other, you know. And to be fair to both the left and the right on this, very often when we try to figure out why someone wins the argument, a lot of times what we want to presume is that, oh, someone figured out the, the message to to cause voters to switch parties from Republican to Democrat. And this is where I agree with what Kevin Warsh said yesterday. The reason Donald Trump won in 2024 was had nothing to do with what Donald Trump did. It had everything to do with 9% inflation for a couple of years. And Americans really hate inflation. So if the economy goes into recession or we go into stagflation, it doesn't matter what the left does and it doesn't matter what the right does. The Democrats are going to have a good series of elections.

Speaker 1: What do you think of foreign governments response to Donald Trump so far?

Speaker 2: Um, I mean. Mostly, I think the most interesting response, and this is going to be the one that that really does sort of tell us what the contours of the global economy are going to look like, has been China and how China responded to the Liberation Day tariffs. And I think, you know, that is an instance in which Trump, I think, was surprised by the way that China responded. I think he thought they were going to roll over or at least be a little more deferential. And he then discovered that, in fact, China has some economic policy tools at its disposal. And that led to, among other things, the phone call this week, which my understanding is that Trump initiated, um, you know, Trump has an instinct to find human weakness. And so leaders who want to be deferential, he will take advantage of. But leaders who stand up to him and also have the means to not necessarily be scared by him, he is probably going to take a step back at some point and realize, okay, I need to rethink this.

Speaker 1: Okay. Final one. Will the next president be more or less extreme than Trump?

Speaker 2: Uh. Oof! I guess I will say that I'm assuming the next president will be a Democrat, and I'm assuming that the next Democrat is not going to be Bernie Sanders. So probably less extreme, but you should, you know, give that guess all due weight.

Speaker 1: Which Democrat will it be?

Speaker 2: So I like to call a yacht question, by which I mean, if I had the answer to that question, I would not be here in front of you lovely people. I would be on a yacht somewhere with many, many people taking care of me. Um, I'm honestly not sure. I mean, I honestly don't know at this point if I, you know, gun to my head, I would say, uh, JD Pritzker from Illinois. But that is a wild ass guess.

Speaker 1: Then thank you very much.

Speaker 4: Thank you. Cheers.

Speaker 1: Welcome Roberto Rigobon.

Roberto Rigobon: Thank you so much. It's great to be back. Indeed, the first time I spoke here was right after the financial crisis in 2009. And I remember I made a bunch of outrageous statements, and I have no idea why Mark thought that that was worth it. And then they said they invite me again. So, um. Eh. So let me let me tell you more or less what what I am going to do. Uh, this is a this talk is, is part of the research that I'm doing with Isabella Loayza. She's a post-doctoral student at MIT, and she's a computational social scientist. So we have been working for a while on trying to understand, you know, the impact of a general purpose technology like AI is by by looking at this problem differently. And you will see in a second the way we do this. Okay. So first what is a general purpose technology. These are things that have happened before. We have had many in in our lifetime. These are technologies that when they are invented they are not that good. But they have this feature that they improve relatively fast after implementation in that process of getting better and better. They have implications and have consequences in sectors and in many aspects. They are unexpected, so the internet is one of them. At the beginning, the internet was kind of lousy and as it picked up, then it starts having an impact on e-commerce and totally unexpected.

And and what happened with these technologies is that they are very sneaky as they improve and then they are as they are producing innovation in other sectors. They tend to become very disruptive. And in fact, this word disruptive was the definition of general purpose technologies at the end of the 19th century. I was surprised that I found it there because, you know, every time I talk to someone in Silicon Valley, they they say that they invented the word disruptive, but apparently it was not. They are disruptive because they change significantly the way we behave humans. Some of them are disruptive in terms of life expectancy, like vaccines and pasteurization. Some others are very impactful in the in the impact of the labor market. And that's the one I want to talk today about how they change the labor market. So they are they are more than these three. But I have grouped these three major, major changes in in our labor market. So the first and second are very important. This is 150 years of history. So it's not. It's not like it happened just recently. Okay. Almost, almost actually 200 years of history. So, um, we are the first and second industrial revolutions that, that. Mostly affected our our physical ability. It was mostly substituting what we do with our hands. And to be honest, we are like a very weak species, given the size and our weight and and our mass.

I mean, do you understand that insects look at us with pity. Is that okay? Like. Like you cannot even you cannot even, you know, lift your own weight. What a loser. Is that okay? So? So when when when we have machinery and procedures that substitute our incredible weakness. I mean, it produces a transformation. But truly, I want to ask the question. I mean, 3000 years ago, every one of us was carrying rocks in Egypt to manufacture the the pyramids. Who wants the job back? Raise your hand if you have to go back and carry rocks. You know the donkeys are the only ones that want their job back. Because now, instead of carrying rocks, the donkeys are carrying American tourists in Santorini. And and, you know, they weigh the same. And the Americans are very entitled. Is that okay? So, so truly, they want their rocks back. But truly, you know, nobody wants to go and work in in agriculture in the same way that we used to work. We substituted that with a truck. So it took a transformation. Just to give you an idea, in 1850, about 90% of the labor force in the world was in manufacturing. Today is less than five. So it's a tremendous transformation. We are feeding more people than ever before. We have less hunger than ever before, and we are using a tiny proportion of our population. So truly in improve our standards of living. The most recent ones and I put this computer on the internet just it's actually the transistor.

But, you know, I didn't want to sound just a computer and the internet, these are the products that ended up with these innovations. What they're doing is substituting our part of our cognitive activity. So at the beginning they were scary because we thought that we are the only thing that can add and subtract. So in fact actually I remember my dad, he he was really good at adding, but that's what he was an accountant but truly was mostly accountant. Is that okay? So he forgot to put the A because you give him a 1010 numbers of four digits and he looks at that and gives you the actual answer. And I remember I was in the university, I am old enough that I started the university with the engineers rule. You remember that? Who remembers the engineer rule? Raise your hand. Oh my God, you guys are old. So this will be probably your 21st research conference. Is that okay? And so that's how I started in my engineering career. And the day I got the Texas instrument, I told my dad, I said, you are unemployed. He kicked me out of the home, so he had some minor consequences. But anyway. But but there's a saying in, in computer science that if you can explain it, you can codify it, because what we, what we have been able to do is that if I tell you how to use the differential equations to calculate this building, and this building would not be possible to be calculated by a human, given the shape of this building, this needs a computer.

The differential equations for the forces in all the nodes of this building are extraordinarily hard to solve for a normal human being. So that's why all the buildings that were designed by civil engineers look exactly like a square again, that we can compute everything else. We can't. But all the buildings today have these curves because a computer, once I explain the physics, they can estimate unsolved differential equations that are incredibly nonlinear. And that's fine for a computer. So we taught them how to do it and they do it. So if I can explain how to play chess they will play chess. And by the way what they do is they replicate our procedures okay. So when we when they add they actually replicate our additional. So the way you add is exactly the way they will add. It's just binary as opposed to us decimal. But truly it's actually the same procedure. So uh, by the way, this this revolution is having a tremendous impact in humankind. It still is. Half this is still you know, we have been doing this for 50 years, and still many companies have not done the transformation, the digital transformation or the data transformation. So it's not like we are already there.

Is that clear? Everybody knows Excel but that's it okay. So we need to we need to continue investing. And I will talk a little bit about this. Ai feels different because AI is replicating things that we cannot explain. They are repeated, but we cannot explain them. So the best example is to talk about my family. If you have ever been in these conferences, you know that I talk a lot about my families. I have an incredible, unique ability that not a single human being has, that I can recognize my wife. This has been very important to be able to be happily married for 32 years, because I say like, is that you? And she said, yeah, that's me, you bastard. Is that clear? So she makes very clear who she is. So I know and I recognize her. Can you imagine the likelihood of being married if I were unable to recognize my wife? And so this is my daughter. Actually, this Richard has a lot to do with psychology. She has been an instrumental in the research that I'm going to show you. Her name is Veronica. If you ask me, do you recognize Veronica? Yes, I do. How? I have no clue. If you said. Well, what is the distance between the two eyes? No idea. What is the ending of the of the ear relative to the start of the eye? No idea. Height. About this size. Weight.

Okay. I can lift her. Is that okay? So less than me. I mean, like, I have no clue, but I am incredibly good at recognizing it. In fact, I actually think that this is what I recognize from her. This is. I asked a computer to give me the most minimalistic view. I think it's just her hair, but the computer doesn't do it the same way I do it, because I cannot explain it. If I don't understand it, I cannot explain it. So what the computer does is they do it differently. That's why when they when you are taking a picture for your passport, they tell you to be slightly diagonal. Why? Because it happens to be that the shape of your ear and where it ends and where it starts. The eye is a very important piece of information to recognize your face. Who recognizes your love partner by the ear? Raise your hand, please. One. You must be like a really funny guy. Yeah, but truly, when you think about it, you say like, no, that's why. And for example, for the ladies, they always ask you to remove the hair from the ear. They have to see the ear to recognize it. They look at the distance nose one. But that more or less most people have one only. So so truly, it's not about the smile. It's about certain features that that the individuals have. So we do it differently. Now what is interesting is that the underlying statistical process that our brain uses and the underlying statistical process that artificial intelligence use, we use the exact same function.

It's called a universal approximation function. These are things that humans have been actually doing for ages. In fact, from Aristotle, we started doing approximations. We started to use the language of math to try to simplify reality. So I'm going to talk a lot about universal approximation functions. So be ready. No I'm joking. No, no. So so what has happened is that through time we have improved. In fact, this kind of our greatest minds have actually contributed to this. So, you know, a big step was Isaac Newton and a big step was tailored. And so Gauss, these are the people that provided the tools that we use today to approximate life. These patterns that we look at life, our approximations of what we see. And these patterns are coming from all these statistics that we have created, all the approximation of functions, all the orthogonal functions that we have created, Chebyshev Taylor expansions, Fourier transforms. All of that has been with the purpose of being able to take reality and simplify it in their components that are simpler for us. The beauty. Our brain doesn't work that way. Doesn't. Our brain has a very flexible set of functions. And therefore what we do is that these functions are super simple. They all look like a step function. Okay, but there are billions of them.

And then we approximate reality and the patterns of reality through that, the computer does exactly the same thing. Those universal approximation functions have limitations. So what our research has been try to understand what we humans do. Duel where the universal functions will fail. So these are the tasks, the actions and the activities that we humans do that are not going to be replicated now, nor ever by a computer. Is that clear? Because the underlying piece that leads to artificial intelligence has limitations like our brain has limitations. The difference is that we operate in a collective action society. We correct our stupidity by others. Is that okay? And right now you can see that there's a lot of stupidity being corrected okay, in the world. But truly, what we do is that we act in a collective, in that imposes context, that imposes restrictions that a single mind might not have. So let me tell you what are these limitations? There are five limitations. One is about having very little data. So the brain works in an amazing way. How many of you have kids? Raise your hand. Okay. How many times did you can you tell me, do your kids, will your kids recognize what a chair is? Yes. Okay. How many times did you tell your kid that that was a chair? Probably zero times. What you said is sit in the chair. You're eating like a pig in the chair.

You just kick the chair and we say, how many times? 4 or 5 times. And then the brain recognizes what the chair is. That's remarkable. A computer needs 10,000 pictures of a chair to recognize a chair. My son, confused by the way, he was an intern in Stay straight. So actually he he knows many of you and he is a. Well, I'm not going to say anything anyway. So to protect his identity. So but the guy that is British that talks about macro okay. There are many. There are many. So you never know. So but uh, but uh, I do remember that he at the beginning was confusing a chair with a sofa. How many times did I have to explain that a sofa was a sofa? Probably 4 or 5 times. And then he recognized it. Our brain has an incredible capacity to collect data in a very different manner than a computer. In a computer, we actually have to provide the pictures. But what that means is that we operate very badly when we don't have too much data. Okay, by the way, we make judgments about people just by the way they walk, and we make mistakes and we humans make mistakes. Second one is extrapolation. So if this is the universe of the data, all the data that I observe, machine learning is extremely good at finding patterns inside patterns that you and I probably will miss. Is that clear? So inside the data, they are extraordinarily good.

They are extraordinarily good at predicting. Probably they are very good at the boundaries of this data. They are terrible here. Terrible because these functions behave in a very chaotic way. By the way, that also happens to humans in isolation, humans in isolation. We produce terrible ideas, but our innovation and our creativity does not happen without context and purpose and social restrictions and culture and norms. You see, almost everything that restricts our creativity comes from the fact that we interact with other humans and that we have a purpose to serve other humans. And that's the part that a single computer is missing, has no purpose. That's why we're so afraid. What will the computer creatively create? Well, it's a chaotic process. So you might create something that looks actually like a Dali painting. Is that okay? But extrapolation is a very important problem that computers have. Another one that's very important is about multiplicity. There are many outcomes in life that actually two possible recommendations are equally valid. We call that a moral dilemma. And and when you think about it, we don't solve a moral dilemma by flipping a coin. We go through a process to create trust in one of the outcomes. And, and and a solution to the problem is not necessarily that is correct. It's irrelevant. The answer Irrelevant. I will give an example in a second. But. But the multiplicity is that, you know, how do you deal when there are two valid and justifiable answers that are different? We again, the way we do this is we solve this by cooperation.

Is that clear? We get together and we develop trust in an answer and then we choose. Is that clear? No, no. One single mind. Fourth one happens to me all the time. I forgot to tell you that I am. I am an engineer and an economist. So I only need lawyer to have all the worst professions in humankind. Is that clear? But you see, when I talk to my wife and she says I have a problem, I go immediately to solution mode. Is that okay? Like, yes, baby. Like, you know, where is Tony? Yeah. You know, Mission Impossible. This is mission super possible. Is that okay? Is it, like, super trivial. And in fact, for some reason, I tend to answer all the reactions of my wife by using the word just. Is that okay? Oh, you just have to do this. And at that time, she gets very upset when I give them a solution that I know is the correct one. Has that ever happened to you? And what does the wife typically tells you? Like I don't want a solution. Is that a typical one? Have you ever said that to your partners? Have you ever said that to your partner? Yes. Yes. Yes. Be honest. It's okay. We cannot help ourselves.

Okay. But see, it's because they want the outcome has to be a relation. It's a relational outcome. It's understanding. It's what we do with empathy. Empathy is not about finding a solution. Sympathy is what we do when we find a solution. Sympathy is the ability to see that the other person is suffering, or is scared or has anxiety, and we provide something to reduce that sentiment. So I give you a chocolate. I give you a hug. So do we provide something to compensate? Is that clear? That is called sympathy in psychology. Computers are really good not at hugging. They are horrible at horror, but they are really good at detecting a state of mind. Really, really good. And then offer a recommendation. My wife will hate that because she doesn't want a recommendation. She wants someone to listen. Someone that develops empathy. Someone that develops compassion. And that's actually about a connection. And that connection is really difficult when you are expecting the connection really difficult to obtain, to provide it requires authenticity. Let me put it this way. If if a woman that is pregnant has been fired from the job, I can offer compassion. Is that clear? Because. Of course, I have not gone through the process. Is that clear? I have not been fired because I was pregnant. I mean, trivially. I mean, I guess so, but I can understand the pain that she's suffering at that moment because I have seen unfairness.

But when a Venezuelan comes to me and says, I am afraid that I'm going to be kicked out of this country. Oh man, I can tell. I know exactly what you feel, because I was going through my green card exactly in September 11th, and we had to repeat everything, and we were treated like criminals. Like you have no idea. Thank God that I was working on MIT, but it was a no fun moment for my life. So when the students come to me and says, I feel unsafe, I said, oh, I know, and I don't have to say anything else. There's a connection because we both know that we went through the same experience. That authenticity matters a lot. It's not about the statement. I will come back to this in a second with an example. So lastly, beliefs. We the best decisions that humankind has made were not based on data. I know that we tell everyone that we base our decisions on data, and to be honest, in 99.99% of the circumstances, that's the best procedure. Okay, base your information and your decisions based on data. But the best decisions of humankind were not founded on data. So let me just give you an example. Since the moment we became a species, what we tend to do is we gather all together. Okay. We go to the town next to ours. We beat the hell out of those bastards and we enslave them.

That's what we have been doing for thousands and thousands and thousands of years. So let's look at the big data. We have thousands and thousands of years that we are better off by taking what others have and by enslaving those people. And we did that for thousands and thousands and thousands of years. Which data point was used by the first person that decided to abolish slavery? Which data point? What did they do? What was in the brain? In a world that there was no fairness. In a world that there was no equality. What about vote for women? We improve our life expectancy by double without women voting. So if you give that to the to the computer and you ask, what should I do tomorrow? It will say just replicate today. But somehow we added something to that question. We added a notion of equity. And when you add that notion, which is not in the data, then we suddenly decided one vote for each. Our decisions actually in humankind, abolishing, for example, you know, public executions, providing rights to people to defend themselves, things like that. Do you process the institutions that we have created in all the developed nations? They were not necessarily supported by data, at least the first one, the first country that gave the citizen ship right at birth in the country. This started in the Napoleonic laws. It was not in the Constitution. Then the second country that was 1802, the first country that put that in the Constitution was Haiti.

But it was conditional. It was not unconditional. So you had to be either black or not from Portugal, Spain and and France. The first country that put that in the constitution in 1811 was Venezuela. By the way, in that Constitution also was the first time that we said that the Social Security is a right of the citizens. Social security is a responsibility of the state and a right for the for the citizens. That's exactly how it's written in my Constitution. Of course, we violate the Constitution two weeks later, but that's a minor point. But, um, the United States did that in 1860s, 1865. It had to go through a civil war to put that and and the conditions and the reasons of that birthright citizenship were different. In our case, it was about creating a nation. So we wanted to make sure that everyone that was there had a sense of belonging before that, before countries were created, the only thing that you needed to do was to pledge your allegiance to a king. This was not a kingdom, and because it was not a kingdom, there was nothing to pledge to and therefore we needed to actually change. It was a sense of equality and a sense of ownership. Almost all of that was created by negating the past. So what we have done is to look at. So I just want to highlight, I am always a very positive person, and I hope that you can see that.

I see a lot of beauty in the way we interact with each other. Is that clear? We correct each other. That doesn't mean that we don't make mistakes. We make huge mistakes of inference when we actually don't use proper data to make a claim. We make huge mistake of extrapolation because we kind of invent something like a nuclear bomb that we probably should have not done it. We have millions of dilemmas in moral aspects that we don't know exactly how to choose. We make millions of mistakes in marriages because we don't produce relational outcomes, and we have actually made millions of mistakes because we were not courageous enough to reject our data. But at the end of the day, our interactions as humans has allowed us to build what we have, and that is what makes us different. Our weaknesses we have solved by a cooperation, by solidarity that's difficult to achieve in a computer. So what are these characteristics that humans do that are related to this? And this is where the word epoch comes. What we have done is to went through all the psychologies. And this is where my daughter has been absolutely instrumental. And what we have done is to look at what are the things that we do that cannot be replicated. So this is empathy and emotional intelligence. Compassion is there.

This is a human contact. I don't know who here likes sports, but there's no way you derive the same satisfaction of sports by watching alone that we're watching with people that you can hug and high five. There's something about ethics, judgment and openness. These are all the aspects that are related to to those decisions of multiple outcomes. And openness is our ability to change our opinion. Is that clear? So it's a combination of using small data and then evidence to change. There's creativity and imagination. And this is the creativity and imagination that I described that is really far from where the data is. And the last one is about hope and vision. Hope is thinking that something will happen when the actual probability is zero. That's what an entrepreneur does. Hope is also misused is when someone buys a lottery ticket, thinking that there is actually a net present value positive from that transaction. So these are not universally good aspects of humans. Is that clear? But this is what actually makes us human. So we actually went through this and match each of them to one of the data weaknesses of, uh, of artificial intelligence. This is kind of a little bit less of a is more on the technical side. But again, creativity is defined as inventing something that did not exist before. Uh, is uh, is a problem of inference in a small data, and it's a problem of extrapolation. So the web pages that actually write a song for you, that's we don't call I am a musician.

Uh, we don't call that creativity. We call that rip off. Is that clear? So can I create a song like Taylor Swift? Of course I can, because I go to the web and I said, I want the song to sound like Taylor Swift. Is that okay? I can do the dancing. Is that okay? But the question is, is that really original or this is a rip off? It's a rip off. This is almost like saying that you will pay $10 million for a student drawing something like Van Gogh. I give them any image of the Boston skyline, and I ask a student of art, which is actually here four blocks from here. And I say to that person, can you draw me that as if you were Van Gogh? They will be able to do it, and it will look identical. The computer can also do it. I want to know who will pay $10 million for that. It's authenticity that matters. The other one is a rip off. So anyway, so what we did, and this is what I want to end today and then give a chance for questions. We took this a. This can go back. Yes. We took each one of these definitions of empathy, definitions of compassion and many definitions. And what we did was to take these definitions. By the way, you are going to share the the the slides.

No. Yeah. The you can get the slides. Just so you don't have to take pictures. I just want to let you know. Now if you want a picture, just let me know and I'll do something like, uh. You know, and that's a picture worth having. But other than that, you know, so what we did was to look at the jobs in the United States. The census has a database that has about 900 private sector occupations. Each of the occupations is described by a set of tasks. And the tasks are, uh, I don't know, a 27, 24. I don't remember exactly, but it's on average a job has 2427. The job that has more tasks is actually a plumber. Is that okay? Uh, CEOs have very few tasks. Send an email and the tasks are like that. Send an email, create a vision. You know, they actually when you read the sentences, you can see that there are many aspects in the sentence that you said, oh, this seems to be that you need hope and this you need presence. A nurse, for example, in the description of the task of the nurse, I mean, there are different nurses, but let's assume it's like the one that I remember is about physical therapy. See, the patient is on massive pain. So that nurse that that person that is assisting the individual that is doing the treatment needs to convince you to go through pain at home because they have an hour that you are with that person is not going to do it.

You need to do the exercises at home. So it needs to motivate you, needs to tell you why the pain is actually useful, you will fix your shoulder or whatever. So they need to create a lot of vision. They need to present a hopeful recovery. And in fact, when you look at the task, it's impressive that they actually say on the task that you have to create a plan. Is that okay, a plan for recovery. So um, so what we did was to take each one of them task by task and evaluated all these five dimensions, the epoch. And we gave them a score between 0 and 1. And then we aggregated at the level of the professions. And we produce this. By the way, the data is, is, is available. Um, right now we are doing this for all the Latin American countries. So we are getting the data in all the Latin, I mean all the Latin American countries that have this data. So we are trying to actually get all the countries, and we are kind of figuring it out how we can do this in Spanish and Portuguese, because we can do this in English and French. So it just happens to be that we chose the two wrong languages. Okay. So so what is interesting is that we when you look at the different sectors.

So let me just talk about financials. Oh sorry. If you look at oops I think I screwed if you look at, at uh at the top Even this is a laser node doesn't have a laser. Okay. But if you look at business and finance, you can see that it has a very wide spread. What that means is that there are some some tasks, some occupations in the financial system that have a relatively low Epoc. Okay. That means that many of the activities that they are doing can be replicated by AI. That doesn't mean that there will be replicated now because that's an economic decision, by the way. It's not even clear that the technology exists. What we're doing is evaluating what can be substituted now or at some point in time. I had this discussion with the government of Spain, where they were asking me if a, if a there was one profession that in the government of Spain that, that, that we showed there was substituted, that were hunters, people that hunt. No. And they said that how can you substitute that? And, you know, to be honest, I didn't pay attention. I mean, this is 1000 occupations, so I really don't care. Okay. So I said, oh, you know, I didn't think about that. And they said, this has to be all wrong because how can you substitute fishermen? And they were substituting fishermen and and and and hunters that you must be all wrong.

And they said, you know what? You know, I am all for communication. But but if you criticize my research, I become really nasty. Okay. So I told them, you know, after a while when they were criticizing my research that occurs, my answer, I said, okay, let me substitute the hunter. It's going to be a little tractor that will go through the forest with a camera and a gun on the shoulder. It will go about ten centimeters from the deer because it will not smell like a human, and it will shoot the deer at the right age. And exactly here. So no pain. Say, substituted your hunter. He said, by the way, the military already has that device, but not for deer. And and then they were shut up. And then they look at me and said, what about the fishermen? I said, well, for the fishermen, we can use a technology that is very new, which is called a net. No, you get a net and you get the tuna and the dolphins, and then we use a facial recognition and we send the tuna on one door and the dolphin on the other. And he said, I said this was, by the way, was the vice president. Anyway, I love to shut up politicians anyway. So so this is like. But will that be economically viable? I said, that's not what I'm answering.

I don't care. I said, can it be substituted. It can. Is then an economic decision whether or not. By the way, I remove economists from this list because economists are between taxpayers Taxpayer. Tax preparers. So TurboTax guys is that okay? And meter readers okay. So economists came between those two. And I said you will remove the economists from this list because I have a reputation to protect. He said I want it to be below this. Now what is important is also about what is the collection of of of tasks. What occurs is that you might have some of the tasks to be substituted, but that will enhance your job because these are jobs in which the tasks are all intertwined. So this is almost like an accountant that needs to make a judgment about where to do the accounting entry and adding and subtracting. If I give an accountant a calculator, I cannot separate these two tasks. There are some properties in the data that you can detect how intertwined the tasks are. For example, my hobbies are golf, carpentry and cooking, so they are completely independent. My wife can take one of them and it will not necessarily affect the other. I will be unhappy, but that's a minor point. I will be able to continue performing the others because the tasks are not intertwined. So here you can. The risk measure that is on the bottom is a proportion of the tasks that will be substituted in a job where you can dismantle.

Is that okay? Augmentation is that there's a proportion of tasks that will be substituted in a job that cannot be dismantled. This is a surgeon that needs to make a judgment about where to cut, and I can improve the equipment to get there. Is that clear? That will make the job of the surgeon better. But I cannot separate the cutting the judgment from the actual, you know, activity. So what is very interesting about this graph is that in the United States, the size of the bubble is the number of people in each of the occupations. We have about 7 million jobs that are at risk. Seven. Not 2 million. 7 million. It's not a small number. Is that clear? But it's not 50% of the labor force. Is that clear? It's 5%. 5% of the labor force. And by the way, anything that is above 30 is a very big number for augmentation. The way we compute it in the paper, we explain what it means. It just means that 30% of the tasks will be improved by AI. So this means that 30% of what you do will be improved. This is almost 70 million, seven 0 million jobs. So, um, properly deployed AI I will improve dramatically our life. Let me just tell you now how to do it. And with this I end. There's a difference between wealth and welfare. And so.

As I said, I'm a golfer, I love golf. I think I have kind of a relatively good handicap. It fluctuates between 6 and 8. I am 62, so it's not a bad handicap. But if you ask me, will you be able to beat, uh, Scottie Scheffler? No. Probably not. Is that okay? And I said, well, I will help you with AI. Let's create a robot that hits the ball for you. And an AI that will measure the wind, the temperature and everything. So do you think that with that technology, I will be able to beat Scottie? I'm absolutely certain the answer is yes. The robot will be way more effective than that. Him actually will have better information. I actually will be able to measure the distance perfectly that he can't. Okay. And so I will beat the bastard. Would that be fun? Would that be fun? You see, when we design, it has to be with a purpose. Right now we are using and designing AI to substitute some of our tasks to create a competitive advantage, as opposed to design AI to make our life better, to make our jobs better. These two things are different because if I need to design someone to substitute their hair, okay, I need the AI to do everything that Dale does, but better. Is that clear? The bar. Well, in the case of Dale is kind of low, but anyway, for almost everybody, the bar is really, really, really high because I need to do everything that Dale does, but better.

If my answer, my question is, can I do something that will allow Dale to make better decisions? Then the bar is really low. It's it's not. I just need to provide him with a pattern or a recommendation that he had not thought. It's a humongous difference between these two. Humongous difference. And we keep keep designing to substitute as opposed to design for uh, for complement. So our challenge is to create institutions to make sure that we design for complements. There's no doubt in my mind that artificial intelligence will create a tremendous amount of wealth in the world. And in fact, I think that computers and the internet have done that. We have failed to produce welfare. For that, we need to provide training. We need to provide training that is complementary. When we invented the car, the car was worse than the Horses. Horses could go everywhere. Cars couldn't. Everybody knew how to ride. Nobody knew how to drive. What did we do as a society? We could have taken choice. One which is to create better horses, faster horses and better riders. We decided not to do that. We decided to learn how to drive. So we invested in complementary skills as opposed to investing in skills that are competing with the technology. We invested on skills that were complementary, and that's what the epoch is trying to to answer.

Second, we invested in public goods. We provided roads, paved roads. And by the way, a lot of those roads are with free access. We pay a taxation to the center and then everybody can use them. So we provided the public goods. You understand that this has not happened on the internet. We don't provide free access. The access is actually bundled in a product. There is a very expensive product. So we have not done that. We have invested in the in the complimentary skills. Third, we have to provide norms and regulations. How are we going to work? You see, when everybody has a horse, nobody, nobody rides on the right. This is irrelevant. It's irrelevant. When you are on a horse, you just go. You don't stop. When you see an octagon red, you just shoot it. Is that okay? That's what you do when you're in a horse here. When you see an octagon that is red, you kind of stop. And it's universal. And they. And the light is a universal sign. We invested not only on education, not only on the complementary capital investment when we invested in the norms to all these, to all of these, uh. Technologies that are disrupted on the labor side, we also invested in labor standards. We used to work 30 200 hours 2000 years ago. Today we're working between 16 and 18. Well, if you're in France, it's like 12. But anyway, that's a minor point.

So, so but, but but we actually are working half of the time of what we used to work. We have better labor conditions. The auto industries that responsible for the 40 hour, five days a week that regulation came from the auto industry and the auto industry actually doubled the minimum wage at the beginning. It was coming from the fact that you want this technology, the combustion engine, to be a source of welfare that we all can enjoy and make it better. So this is our challenge. Our challenge is not whether or not we're going to be substituted by AI. The challenge is are we going to design it so we can complement it? And in fact, the best business that I think of is one in which we are able to teach empathy, compassion, presence, a hug, creativity, curiosity, hope, vision. I don't know where you learn hope, but I learn hope from my mother. I was born in a relatively humble household. My both my parents were escaping very bad conditions in their country. My dad from Argentina, my mother from Spain. So I was very humble. But. But you know what my mother told me about how to develop a friendship, how to have hope, how to have a vision, how to have a responsibility about my future. So by some measures, I could have been a low class, but by what? My mother told me, I was absolutely rich. Our challenge is to find a way to teach humans a little bit of humanity. Thank you so much.

Speaker 3: Okay.

Speaker 4: You know, I'm going to take back what I said in the introduction. There is no way I is replicating that. What a wonderful start. So we've got two different ways to ask questions. We can take hands. You can also scan the QR code. And oh by the way you can also use the app. Have I mentioned that I've got one question already that someone submitted. So I'll let me let me take this first one and then we'll see if there's any hands. Um are there any sectors or parts of the economy where AI has not been applied currently, but you think should be?

Roberto Rigobon: I actually, I think that no one is using AI properly now yet they are using AI mostly to substitute what we do. It's like Uber. Uber, you can see it everywhere except in the statistics. So if you think about it, Uber should be. I have a useless piece of capital, which is, you know, in the garage and I have a useless individual eating potato chips watching the TV. Is that okay? So I take a useless human being and a useless capital, put them together and offer a service that in math, should be infinite productivity. But we don't see it. Why? Because Uber has been designed to substitute taxis as opposed to change car ownership. If I'm going to substitute a taxi, well, then I will go from A to B. A person that doesn't know me, I will just make a payment. And what they did was to a little bit improve the payment system, which by the way, has cut up everywhere and a little bit about the cleanliness of the car. And that's it. And by the way, they are doing this this by circumventing labor laws. So it's not clear to me if this is welfare improving as it is. If you have a capital that is 95% of the time idle, which is probably your car. Yeah, we could do better. Uber should be designed to change the way we own cars. This technology is not to move you from A to B. This technology should be used so you don't have. In fact, the future is that everybody has a Toyota Corolla white. Is that okay? These are indestructible cars. They are impossible. It's impossible to kill them. In fact, even if you burn them, they they they, you know, resuscitate. Is that okay? So. So I need white Toyota Corollas for everybody. Everybody's self-driving, and you only own a little piece of that asset. That is when you have the the impact. Right now, AI is being used to substitute to threat. There's more fear than actually hope.

Speaker 3: But that's human collaboration, though, isn't it? We need to share the cars. Cause.

Roberto Rigobon: Exactly. But, but but we are not using it with that purpose. So we are creating wealth, but we're not creating welfare. And in fact, actually, if you think about it, Uber if I, if I try, I mean, I have not done the research, but I cannot imagine that the welfare improvement of of of, of Uber are that big. Because the only thing that we point is about the better quality of the car and the cheaper price, and both are coming from the fact that they are violating or circumventing labor laws. So not clear to me that this is welfare improving. So next question.

Speaker 3: So okay. Well so this this actually goes back to the poll I think. So so specifically within finance. Um, could you comment on the epoch score for things like investing?

Roberto Rigobon: Oh, I love your question. Can you have an AI managing a portfolio? Yes, I love that question because immediately I answer like an economist. Is that clear? I said it depends.

Speaker 6: Yeah. I mean, of course.

Roberto Rigobon: It depends. No, but. So what is the most important aspect that State Street has? It's trust. Yeah, it's actually trust. So let me give you an example. Imagine this actually happened to a cousin of mine. Not what I'm gonna describe, but this disease that I'm going to highlight is a true disease. So imagine that I go to Google and I write on Google and I say, you know what? I have muscle mass that I'm losing on my left arm. This bone is breaking repeatedly. They try to heal it but then heals incorrectly. So you have to break it again. I know I don't have osteoporosis. What should I do? And Google answers. You have a rare form of cancer that has no cure. You need to amputate your arm immediately. Please raise your hand if you will amputate your arm with that search. Alberto. All? Yeah.

Speaker 6: Nobody.

Roberto Rigobon: Okay. Okay. Why would you do you. What would you. You look like young and that you love technology. What would you do? You will go.

Speaker 6: To a doctor. Yes.

Roberto Rigobon: And after you go to a doctor, imagine that. The doctor says you have a rare disease. Do the test. You have a rare form of cancer. And actually, you need to amputate your arm. What would you do next?

Speaker 6: You go to another doctor.

Roberto Rigobon: So you want a second opinion? Will you put the second opinion on ChatGPT? No.

Speaker 6: Why not? No.

Roberto Rigobon: Okay. So you go to another doctor. Does 400 tests and after the 400 test says. In fact, you have a rare form of cancer that is eating your bones. If it gets to the clavicle, you will die. I need to amputate your arm. What do you do then? You amputate your arm. Is that correct? It's unfortunate. In fact, that's exactly what happened to my cousin. Have you realized that Google was right. Trust is not about the answer. Trust is about the process. And when you interact with State Street, it's not because they have the right answer. The right answer contributes to trust. It's the process. It's the process. So thinking that I can substitute a portfolio with a computer, it needs your trust. Michael, when you tell to someone State Street has this procedure we have tested we have Dave Tarkenton. We have, you know, a will that holds. Megan, I don't know who is actually doing.

Speaker 6: The test, but anyone anyone from SSA have.

Roberto Rigobon: Done the testing and they tell you this portfolio is better. That statement doesn't come because it's a computer. It comes from your judgment. So, in fact, how long will it take for a computer to substitute my portfolio? Never. Unless the portfolio is a stupid portfolio. Which, by the way, are all the ones that I design. The reason why I don't talk about investment is because in the only conference that I talk about investment, we are a financial crisis.

Speaker 3: Okay, I think we have time for one one.

Roberto Rigobon: Oh, there you go. The boss says one more question.

Speaker 3: So I have one on here, but let's take let's take it. Oh, let's take a voice. One.

Speaker 7: Stanislav Seltzer from Sergey. Um, about 15 years ago, Garry Kasparov lost a match. Chess match to to Big Blue designed chess computer. And in a press conference, he looked shocked. And they asked him, what's the future of computers in chess? He says, yes, I will lose the supercomputer, but give me a laptop and I will crash supercomputer. What do you think of that?

Roberto Rigobon: I think that, uh, I think that playing, watching two computers play, play chess is a really boring thing in life. Yes. So, like I said about golf, I'm also a very bad tennis player. Is that okay? But in this case, I'm really bad. I think that it will be no fun to create something that makes me better enough by substituting everything that I do in the tennis to win one game against Rafael Nadal. Truly, truly for me will be. I will just play with Rafael Nadal. I don't mind losing. And the fun will be to experience losing, not the winning. So when you design the AI to substitute my tennis swing, it will be useless now if you use the AI. So I improve my tennis playing, then. Great. You're exactly using that information to make me a better golfer, a better tennis player, and a better carpenter. Is that okay? So that that truly is is how this should be used. I think that watching chess, I used to like chess, but watching every chess game now and on ties is really, really boring. It's almost like the premier soccer. Just like boring. I know, I'm joking, Lou. I just saw Lou here. So. But you know, we want we want outcomes that are unexpected and we want outcomes that have a chance. It's the whole point of being a fan is that you hope to beat the odds. Being a fan of a team is that you hope to beat the odds. So it cannot be Just a computer playing against each other. Is that clear? So I don't know about these statements, but watching people play chess and always end on a tie sounds like a really, really boring game.

Roberto Rigobon: Okay. And one human creativity, talent and mistakes. To be part of the sport or be part of our decisions, be part of our life. So what makes what makes a life actually beautiful is the fact that we overcome our own mistakes and we make many, many mistakes. And it's in that process of a mistake and the correction of the mistake that we truly feel fulfillment. And in fact, if you do that for somebody else, it will be even better. Make the mistakes with others and then actually make sure that you compensate. And then at the end of the day, that sense of fulfillment, you will never get it from the computer. Never, ever. It's your accomplishment. And I would like my kids to actually feel exactly the same way. That's the that's the society we need to strive to build. That we make better decisions. That we make mistakes, control our mistakes. But more importantly, every time we make one that we find a solution. So I love I love stupid politicians, I love them, I absolutely love them first because increases the demand for my classes. So I have more students that come to my class. But the second and the most important one is it gives gives us the chance to rethink what is the society we want to build. So it's costly. But I am absolutely certain that at the end we will overcome what we are and we will be much better off. Our life expectancy will go up and the quality of our life will go up. So that's what I'm looking forward.

Speaker 4: Brilliant. Thank.

So today, Lee is going to show us how to use proprietary indicators to create actionable trade ideas to cut through all of this noise and navigate markets over the medium term. So it is my absolute pleasure to welcome to the stage Lee Ferridge, head of multi-asset strategy for North America at State Street Markets.

Speaker 2: Thank you. You're welcome.

Speaker 3: Thank you. Kayla. Um, so I'm sorry to to let you down, but you said I'm going to solve the problem of uncertainty. But as my title suggests, I'm really not. Um. What next? Who knows? I'm being honest. Um, you're not wrong about the level of uncertainty, though, that's for sure. So, um, where we are. This is where we are. So this is the Baker, Blum and Davies economic policy uncertainty index. And what you see is it's at a record high. So this chart goes back to 1985. So the only time that's even come close to this really was the pandemic. So we're thinking GFC here. We're thinking dotcom crash. You're thinking 87 crash. None of it comes close in terms of the level of policy uncertainty, which makes some sense actually when you think about it. Because during those crashes, financial crisis, you knew what policy was going to do. It was going to go all in and try and correct the problem. Um, as Kevin Warsh talks about yesterday in zero eight, um, this time we just don't know. There isn't a problem or there wasn't. And now we have this level of uncertainty. So Kayla mentioned the S&P 500 companies. Now unfortunately I don't have the option that they do to reduce to to remove their guidance for the rest of the year. I'd love to do that. But Michael told me I'm not allowed to, that I've actually got to give a view for the rest of the year.

Speaker 3: So that's what I'm going to try and do and try and give you a view for the macro outlook FX rates. Et cetera. Et cetera. Obviously leaning into our proprietary data that you've heard a lot about so far. Um, but look, this is my starting point. I'm getting my excuses in early. Look at the level of uncertainty. So if I'm wrong, you can't blame me. I'll just come back and say it was somebody else. Um, so the policy uncertainty is high and that's reflected in market pricing as well. So this chart on the left here, which I will admit I stole from Kayla, this chart on the left here shows you rates expectations for the rest of the year in the US. But what I've done I've taken specific times. Um, so you can see where we were on the 1st of April. So the day before Liberation Day, that's where the blue bars sort of signify. And then we get to the low or the maximum number of cuts priced post Liberation Day, which actually happened sort of on the Friday after the Wednesday was Liberation Day or the Monday, the 7th or the 8th of April. And you can see we got to a point where we were getting close to five cuts price for this year. Where are we now? Have you got less cuts priced in now than we did the day before Liberation Day, even though the level of tariffs that we're likely to get is still probably higher than the expectation on the day before Liberation Day.

Speaker 3: The market is sort of shrugged it off. It's sort of, you know, moved to a point where that's not too bad. So Alberto touched upon this. This is anchoring. Maybe it's the art of the deal. I'm not even sure it's that explicit, but this is anchoring. So you go in and you say it's going to be 150, and then you bring it down to 50 or 30 or whatever. Um, and the market sort of goes, you know, 150. That's horrific. You come down to 50, the market's like, okay, we can live with that. Yeah, but that's higher than you thought it was going to be. But it doesn't matter. The market moves on and you will see that in prices. Um some charts that are on in terms of the equity market. So the interesting thing here as well is the chart on the right though, which shows you this expectation now. So we have 2 or 3 cuts priced in for this year. That's not really what the market expects. So I'm using the options skew or the options market in the chart on the right. So 1 to 3 cuts is given about a 50% probability. But then what you have is no change in rates is given around a 30% probability.

Speaker 3: And more than six cuts is about a 20% 10 to 20% probability. So really what you're looking at when the market is pricing 2 to 3 cuts this year, it's almost a weighted average because it's hard to see the scenario where you get 2 or 3 cuts. It's going to be zero or it's going to be a lot. Those are the two options I think. And what you see in terms of market pricing is that weighted probability in the middle. So the market's uncertain policy is uncertain and so is the fed. Um you know Kevin Warsh obviously has some misgivings about current fed policy. Should we put it that way. Um, but what we're actually seeing from the extensive chat we get from the fed, which obviously, if we get a warsh fed in the future, there'll be a lot less of that. But what we get right now is a lot of fed disagreements. So this is work that Ronnie talked about yesterday looking at, you know, the tone of central bank speech. So what we're doing here is looking at disagreement amongst fed speakers. That's the dark blue line here. And you can see it's been rising consistently. And that's reflected or was reflected in one year one year pricing. The standard deviation the volatility around one year one year pricing started to come off a little bit more recently. But what's interesting is this level of disagreement we're getting in the fed.

Speaker 3: Normally that's a build up to a change in rates. The green lines here represent or change in cycle. The green lines here represent the start of the hiking cycle, the end of it, then the start of the cutting cycle. Rates are on hold. They have been since December. And yet the messaging from the fed, the disagreement continues to rise. What about our own indicators here? So turbulence, our measure of market unusualness. Um, actually coming back down to more normal levels. You can see what happened after Liberation Day. So turbulence global equity turbulence. Turbulence actually at 100%. Which means it was the highest reading we've seen in at least five years. So that's a measure of market unusualness not just price volatility but a breakdown in correlations. And you think in fact that's probably the movement the dollar compared with others, which was highly unusual. Now though we're coming back. We've come back to around the medium. So there's some normalization going on certainly in prices. And also in sentiment. You know, given this level of uncertainty, given that previous level of turbulence, you'd You thought real money investors would be extremely risk averse, almost running for the hills, if you like. That's not what we see. So this is our behavioral risk scorecard, our broadest measure of risk appetite from our investor behavior, from our real money flows. What are they buying? What are they selling across asset classes.

Speaker 3: We have 22 different versions of it all the risk on risk off decision. We add them up. You get a score for sentiment. Where are we now. We're at plus eight. The most positive sentiment reading we've seen since February. But more interesting in some ways perhaps, is even at the height of the uncertainty, even at the height of that turbulence, this sentiment measure never really got that low. It was negative for sure. We got down to about minus six, but look where it can go. -16. That's a sort of panic mode. We never got there. We were cautious, but we never got down to any sort of panic level, and you can sort of see that better. Maybe in this chart, which looks at the three month moving average. So just smooths it all out a bit more. And what you see is yeah, we're negative now. We were positive through most of 2024. Returns were extremely high last year. Why wouldn't sentiment be positive now though. It's negative. But again nervous not panicked. And I think that is a key takeaway from investor sentiment. They never got panicked, even with that level of uncertainty and something else just to complete the picture of where we are right now. So Ronnie spoke yesterday, um, from MQTT and this is the narrative map that he introduced. So this is the one for FX. So basically we have the narratives coming from our MQTT media stats data.

Speaker 3: Um, the y axis is the media intensity scale from minus one to plus one, so anything above zero is above average. Focus on that narrative for the financial media. Anything below zero is below average. And then what we do is we overlay it with the DXY. So changes in the DXY. What's the r squared of the correlation between changes in the DXY and changes in the number of stories on that narrative. That scale from minus one to plus one as well. So a negative reading here does not mean a negative correlation. It's just below average R squared above zero. Above average r squared. What are the key takeaways. Well you can see the main topics that top right is something that is a focus for the financial media and is driving prices. What have we got up there. International trade. Trade war. And writing that he'd love that Donald Trump the most important single factor there. Um, we've also got fiscal up there. Federal reserve just about in. But a more striking thing on this for me is actually the number of topics that are over on this right hand side. So whether it's a focus for financial media or not, markets are reacting to a whole range of different topics here. So you've got very few on the left hand side of the chart. You've got a lot on the right hand side. So the market really doesn't know what is the thing I'm meant to be focused on.

Speaker 3: Every story, every narrative seems to be producing a reaction. That's what we're seeing from this. Um, and I'm going to dig into a couple of these more, a little bit more now. Um, so taking some tentative macro thoughts, tentative. Remember that next year when I was wrong, they were tentative. Um, let me talk about trade tariffs. Everyone's talked about it so far. Um. What's the motivation? Salvation. That's the key question here, I think. And I think Kevin Warsh hinted at this to some degree yesterday. What's the motivation for the trade tariffs? I think one of the main ones is revenue. We see the struggle of the bill going through Congress at the moment. I know revenue from tariffs is not part of the bill, but revenue from tariffs is being put to members of the Senate. Members of the House say, look, we're going to get this money here. This is going to help offset what we're doing. Do not worry about the deficit too much. Um, and Trump has made no secret of the desire for revenue from this. This was a chart, a version of a chart he tweeted out on January the 3rd, showing sources of US government revenue going back to the 1800s. Customs duties up until really the the early 1900s were over 50% of government revenue. Um, and then we had income tax and we're here.

Speaker 3: Um, So again, Kevin Walsh hinted at this yesterday about how imports are 14% of revenue in GDP and how he would much rather hit that sector than the 80% that's hit by tax. Revenue is a key driver. I think Trump wants to get customs duties probably up to around $500 billion a year. Peter Navarro, just before Liberation Day, said 600,000,000,500 billion gets you to about 10% of government revenue. Probably means an average tariff rate of 15 to 20%, but it won't be evenly distributed. As Alberto talks about, you're going to have different tariff rates across different countries. There are four areas that have to get hit the most. If you look on the right hand chart here. There are four areas that have to get hit the most eurozone Mexico, China, Canada. Which of the four areas for countries that he talks about the most? Mexico and Canada. We've still got a 25 cent tariff with a Usmca carve out. Yes, which brings the effective to around 15%. But they were the start, as Alberto said, they were the first ones, 25% tariffs slapped on them. Fentanyl at the border was the excuse. China is China always easy to hit China? We got 150. We're down to 50. Eurozone was trickier. Managed to pick a wall with the eurozone. Now though you know 50% tariff. They're not negotiating properly. There's not going to be a trade deal with the eurozone.

Speaker 3: It's 27 countries. If you get a trade deal with the EU it has to be ratified by each individual country, some of which involves a parliamentary vote. The Netherlands is one place where you need a parliamentary vote. The government's just collapsed. There is no parliamentary vote going to happen there. You're not going to get a trade deal in the next 90 days in the eurozone. So do you want an effective tariff rate of 15 to 20%? These four have to be hit the most. You can do 10% with the others. Trade deal with the UK. Easy. The reason why the UK got the first trade deal. By the way, it's one of the very few countries where the US actually runs a trade surplus. So yeah, you can do a trade deal with the UK. Dead easy. These four are going to get singled out. We've got 50% on steel and aluminium now, 25% on cars, autos maybe something on pharma. Get you to that. 15 to 20% get you to that 500 billion a year. Where are we now? Early days. We haven't really seen much of an impact yet because there hasn't been one. If we look at tariff revenue. So the chart on the left shows you customs duties year to date. We're at about 70 billion so far this year. Now this is up around 80% from this time last year, but it's $70 billion in a $30 trillion economy.

Speaker 3: It's a rounding error right now 500 billion is more significant. But we're not there. And we have this 90 day pause. We've pushed it out. And because of that, we haven't seen much impact on inflation. Alberto stole my chart earlier, so I'm going to steal it back. My colors are better than yours as well. But as you see, if we look so far this year actually as Michael's chart, I can see it. I saw you react. It was Michael's chart to start with. All right. Um, if you look at where we are, you start at 100 with price stats and just run the index through, compare it with the light blue line on this chart is the median over the last ten years. Obviously take the median because the average will be skewed by what happened in 21 and 22. Prices so far this year are below the median, so you're not seeing a big impact on the consumer on spending power now. There's a big offset here on fuel prices, I think, that are playing into this as Alberto's already talked about. But bottom line is in terms of consumer spending power, nothing to see here and there won't be because we've got this 90 day delay and then you get front loading. And as Alberto said this will come through slowly. It's not going to come through in one big hit. And that's why when you look at the data so far this year, it's actually okay.

Speaker 3: Here we've got Bloomberg Economic Surprise Index. I've split it into the hard data and the soft data. We know the collapse in confidence. We saw that across the consumer to a lesser extent amongst manufacturers. Yes, we had the big collapse in confidence. And the view was in February March whenever the hard data would follow, didn't. In actual fact, what you're seeing is confidence is starting to come back. The consumer confidence numbers have picked back up. Ism is a little bit as well. The hard data is fine. Slower than last year? Yep, sure. But still pretty good. And that is an important takeaway right now is the tariff impact is coming. It's not here yet. It's going to be much delayed and it's going to be more gradual. And for now the economy is fine and so is the consumer. Now the other area I wanted to talk about was one on that narrative map that actually wasn't in that top right, but should be and this is immigration policy, um, purely economic point of view. I'm not going to get into any of the social questions or anything else. That's not my job. From a purely economic point of view, the change in immigration policy could not have come at a worse time, both if we think about the medium to long term and also where we are now medium to long term.

Speaker 3: This chart explains it. You see here. This is data from United Nations changing the US workforce. You see, through the 80s 90s 2000 US workforce grew by 1 to 1.5% a year on average. Projection for the next ten years is a quarter of a percent a year. With no migration. The US workforce starts to shrink in 2026, and that is a problem when we think ahead, because demographics are not on our side. And so if we're not going to have immigration, where's the growth in the workforce going to come from? Why specifically now is a bad time? Because the labor market is actually still pretty tight. We have a 4.2% unemployment rate. Yes, it's gone up from where it was. But if you look back over the last 30 years, there's only two periods where unemployment has been lower than it is now. And that was just before dotcom. And just before the pandemic. 30 years. And we're starting here at 4.2%. And now we've just curbed immigration. You know, this demographic problem has gone on for a while. We know it has. How have we corrected for it? Well, there's been two elements. One is increased participation. You see that on the chart on the left here. If you look at the 25 to 54 year old participation rate, which is the one you should look at, you can't look at the adult one because now the population is so skewed to the over 60s you have to look at sort of peak working years 25 to 54 recently hit its highest level in 25 years, since the early months of 2000.

Speaker 3: It's down a little bit from there, .1.2. But the fact is we're pretty much at the limit in terms of participation. We're not going to get above this level. We've taken up the participation rate. That's not going to correct for a lack of labor supply going forward. The other thing we did was immigration. The chart on the right sort of highlights this. So here I'm starting in Jan 2020. And it's the cumulative change in workers. But based on whether they were born in the US or they were foreign born. So native born workers. Us born over since Jan 2020 has risen by 2 million people. The workforce sounds a lot, but payrolls averaged over the last 12 months, 175,000 a month, 2 million people is about 12 months supply. This is five years and four months we're looking at here. So that native born workforce could supply one year's worth of that labour demand over five years and four months. What do you do for the other four years and four months? Foreign born workers, up by 5 million. That's going to slow from here. Actually dropped by half a million last month. And this is why I say this policy could not have come at a worse time. Because the fact is, when we look at the jolts to unemployed ratio jolts data yesterday a bit better than expected, we're at 1.1, which seems like it's in balance.

Speaker 3: But you look at this chart here, you look from 2000 to 2018, jolts to unemployed ratio was always below one. There were always more people out of work than there were job openings, i.e. excess supply of labor actually change prior to the pandemic. That's when baby boomers started to retire en masse. They were born between 46 and 64. They started to retire en masse in 2018, 2017, 2018. And that's when you see that move back above one. That period to 2018, when we were below one, was when you had pretty low wage growth and the fed could easily hit the 2% inflation target. What happened in 2018? Look at the chart on the right, which shows you the jolts to unemployed against average hourly earnings. What happened was, as we went above one, you got excess demand for labor. Wage growth started to pick up. The fed was hiking in 2018 until the equity market collapsed in Q4. And then they started cutting again. But they were worried about wage inflation because you had this excess demand for labor. Well, today we're starting at a JOLTS to unemployed of 1.1. We've just curbed immigration. We've said the economy is actually holding up pretty well. We said the impacts of the tariffs is going to take a while to come through.

Speaker 3: So demand for labor is going to hold up. But where's the supply. So the risk from here is actually that jolts to unemployed ratio starts to rise again. And that leads to what increased wage growth at the very least it's stable and wages stay sticky here. And in terms of a 2% inflation target, that makes life hard for the fed. And that's why when it comes to immigration people say there's a growth negative. It is over the long term because it lowers the level of trend growth in the US for this year. Actually, because of our starting point, it could be a growth positive because what it will mean is higher real wages for incumbent workers. So the chart on the left here shows you real wages. So average hourly earnings minus PCE plotted against real consumption expenditure. Real wages are rising. That boosts consumption. You get into that virtuous circle. If real wages continue to rise real consumption rises, demand for labor increases, wages increase, etc., etc. you get that wage price spiral, which is the sort of thing that is the only inflation that Kevin Warsh was talking about yesterday. He doesn't believe in the wage price spiral. He thinks it's the other way around. I disagree with him on that and a number of other things. But anyway, that's an aside. The fact is that that's the risk here because of that change in immigration policy.

Speaker 3: And that's why I think when we look at growth expectations for the US for this year, I think we're at the low. This is where I do agree with Kevin Walsh. I think we're at the low point. You already see consensus for 25 and 26 seems to be bottoming out. This is the low point. I think you'll see growth expectations actually start to drift back up not to where we were. It's 2% possible this year. Yeah I think so. Which would be close to the 5% nominal that Kevin thinks as well. So what does this all mean for policy and markets. Start with the fed. That 2 to 3 cuts we've got priced in as I said really you're looking at zero or you're looking at lots. Where am I? Zero. I don't think the fed will cut this year. I think if wage inflation holds up, the economy holds up. You get some pass through on the inflation side, the inflation side. It's the labor market that worries me more than tariffs quite honestly. But it's enough to keep the fed on hold. It's enough uncertainty. It's no real sign of a slowdown actually maybe some acceleration. Don't forget we're going to get this fiscal package. We're going to get some sort of stimulus there. Deregulation is still to come through. You know Trump had a sequencing problem in terms of the policy moves this year. Two of his big policy moves are growth negative.

Speaker 3: That was Doge and that was tariffs. They came first. Immigration is mixed as we've talked about. But then you've got growth positive policy changes. The fiscal side the deregulation which is starting to come through. And then energy policy. Energy policy hasn't really been needed because oil prices are low. Although the fact that a couple of weeks after a visit to the Middle East, you see OPEC announcing its third consecutive increase in production, even though oil prices are low, maybe there isn't energy policy there. Who knows? But the fact of the matter is, you're probably going to see more growth, positive news coming through in the coming months. And we're already starting to see that. And that is enough to leave the fed on hold. And this is where I see it for the rest of the year. Equity markets. What's the big deal. What's the big deal on equity markets. We're flat for the year. We're back to where we were in November. And again, going back to that power of anchoring. We're looking at this level of tariffs that if you just said in January, no way are we going to get tariffs that high. Equity market will be on the floor if that's the case. It's not. We're looking at this much higher level of tariffs than expected. And yet because of that anchoring process. We've hit the lows, we've come back and we're flat for the year.

Speaker 3: Where do we go from here? Probably continue to drift higher. If we look at real money investors. So this is our asset class weight series 47 trillion with custody bulk of that real money investors. How much do they hold in stocks. How much in bonds. How much in cash. Simple as that. The vertical lines the horizontal lines are the long run averages. Real money have been overweight equities for around two and a half years or so. We saw a reduction quite dramatic one in the early months of this year. Where are we now? You look at the last month. So the chart on the right over the last month, real money investors have moved back into equities consistent with that behavioral risk scorecard going back to positive territory plus eight. As I said, the highest reading since February. What are they moving out of bonds. Look at the one month change in bonds. A month or two ago we were seeing a shift towards the bond market. You can see the three month change for bonds is still positive, but the last one month that's really turned around. What about from here? Well look, this shows you deviation from the long run average. If you look at where we are now in terms of equities, we're meaningfully lower than we were at the one year high. And also where we were at the start of the year.

Speaker 3: Now the one year high, we're just over 4% above the long run average for equities. That was the highest reading in over 16 years. Actually, it was the highest reading since June 2008, which is never a date. You want cropping up in this stuff. You never want to go back to mid 2008 because bad things happened after that. But not to worry. It's come down. The overweight in equities is nowhere near as big as it was. The underweight in bonds is still very significant But there's room to rebuy if they want to. And the other element to consider here and sort of touches on some of what Kevin was talking about yesterday with the balance sheet as well. Look at excess reserves. There's so much liquidity still in the system. So here I've got that asset class, that equity holdings in the asset class weights plotted against excess reserves. And what you see is when there's plenty of liquidity investors tend to move towards equities. Earlier this year there was the big break during tariff policy the height of that uncertainty turbulence at 100%. Now what do we see. We see them starting to close back up. What are they not buying? They are not buying duration. So you saw the underweight in the asset class weighed series. The chart on the right here. This is where I look at the treasury curve. But you split it into sectors. So it's flows and holdings.

Speaker 3: Y axis is holdings positioning. X axis is flow's 20 day flows. Where's the biggest underweight in the curve? The ten year plus sector. They're still selling. Where's the second biggest? Underweight. The 7 to 10 year. They're still selling real money. Want nothing to do with duration right now. The only part of the curve they're buying is the 1 to 3 year. Anything to do with duration? They are not interested. Who can blame them? Right. You look at where ten year yields are 4.5% high but not excessively so. But if we look at this chart here a lot of people talk about term premium being at ten year highs. Yes it is. But the world didn't start in 2010 or 2008. There is this you know, Kevin touched upon this yesterday that, you know, the financial repression, the balance sheet, the QE one, the QE2, the QE three, all of that lowered term premium to artificial levels. And what is term premium term premium is the extra return that investors require to take on duration risk, so it reflects the uncertainty of time. We just started off with uncertainties at record levels. Inflation uncertainty because of tariffs is at very high levels. Fiscal uncertainty is extremely high. You have 6.5% deficits projected. As we stand. We have this huge, big beautiful bill going through Parliament or Elon calls it something else. But the big beautiful bill going through Parliament through Congress.

Speaker 3: Um, still English. Um, but the fact is, term premiums should be much higher. The level of uncertainty is, as I've shown, has rarely been higher than it is now. Why isn't term premium higher? That's got to be the risk from here, right? The risk has got to be that term premium reverts to more normal levels, which generally 150 to 200 basis points, we're currently 50 to 100. That easily sees long end yields push higher. I think we'll see 5% on ten years this year. I think there's a risk next year that it goes up further. Um, I'll go back to Kevin Warsh yesterday, the one bit I didn't understand his idea that the fed shrinks the balance sheet, which I don't disagree with. But if the fed shrinks the balance sheet from 7 to 3 trillion, we have 6.5% deficits at least going out from here. We have 10 trillion of debt to roll over in the next 18 months. How the long end yields come down in that world. They can't. Who's buying this? So the pressure from here, even without the shrinkage in the balance sheet. But if you get a warsh fed that's something else to consider. But the pressure here has got to be for a steeper curve and for higher long end rates. Now let me finish off with with pH. Let me finish off with the dollar. What are we seeing? Chart on the left us excess holdings.

Speaker 3: This is real money positioning in the dollar. We have the first underweight in the dollar from real money investors in three years. So since March 2022, where they ran an overweight when the fed started hiking rates, they are now underweight. And as you can see, the line continues to move lower. Where are they? Overweight. They're overweight. Aussie yen. Swiss Kiwi. Um. Cad. Third biggest underweight is the dollar. Only the scanned. Is there more underweight sell America trade right. You know the tariffs came on sell USA. All these foreigners selling the dollar. Um that explains it right. That's the narrative. Except it doesn't it's not what's happened. So we can see that through our hedge ratio data. So there's two forms of the hedge ratio. So there's the foreign investor hedge ratio, which is foreign investors into the US or whichever country we're looking at. How hedged are their portfolios in terms of their FX risk? Then for some countries we have domestic investor hedge ratio. So for the US this would be US investors in their overseas portfolios. How hedged is their FX. So both lines are on this chart here for the US the foreign investor is on the right hand scale inverted. And I've inverted it because when these lines move down either of them it means you're seeing selling of the dollar. Look at the foreign investor hedge ratio. It has not moved. It's pretty much been around 60% all year.

Speaker 3: So who's selling the dollar US investors. You see that US domestic investor hedge ratio. It's gone from around 25% in February. March to about 13%. Now US investors were selling the dollar, not foreign investors selling the dollar. So the hedge ratio this is a stock, not a float. And this can move for 1 or 2 reasons. Either they remove existing hedges or what you do is you get a flow of money out of the US into foreign assets. Unhedged. And that brings down the average. That's what's been going on here because we saw buying of European equities, US investors, this hedge ratio came down, US investors have moved into European equities and it's unhedged. Why isn't the foreign investor hedge ratio change something called prospect theory. This is the Kahneman Tversky the fathers of behavioral finance. Basically we do not treat losses and profits the same way we are. We are loss averse irrationally so. So us investors, by not hedging their overseas assets, they give up a potential profit because obviously you get a positive return if you're a US investor heading overseas, particularly in Europe. Foreign investors to the US. If they want to increase that hedge ratio, they would have to lock in a loss and they don't want to do that. That's what we keep seeing and they haven't done it so far. The dollar has stabilized over the last month. Us equities have started to outperform Europe.

Speaker 3: So are we going to continue to see those moves into European equities unhedged. Or are we going to see European investors now decide to lock in their dollar losses at this level. Have we moved 10%. I'd argue no foreign investor hedge ratio. The US domestic investor hedge ratio by the way six year lows. Can that go lower. It can. It's unlikely. Look at the foreign investor hedge ratio sterling in the eurozone. That's where they went. Look how that those hedge ratios declined. That was US investors buying the assets. Look at towards the end of these lines though. Look, they're stabilizing. Eurozone one is turning up a little bit a sign that these big moves may be over. How big were the moves? Well, if we look at the Z score, the three month change, the foreign hedge ratio for the UK and eurozone 3 to 4 z score move z score. I am English z score move. Look at the change in the domestic hedge ratio for the US. Nearly A6Z score move. That's how unusual and big those moves were. What are the latest signs? They're starting to stop. And if we sit here today and we think about us, growth actually probably picks up from here. Us equities started to outperform. We've seen the eurozone equity positioning is back to neutral. Heavily underweight back to neutral. Dollar positioning now underway. This is where I come back to a positive dollar view.

Speaker 3: I actually think the dollar will outperform over the next six months. I guess it's a form of Kevin Walsh's theory about better growth attracts capital. I think that's where we are. I think we've seen the move the other way and from here, not back to where we were. The DXY is down 10%, but we could probably recapture 5% of it. What against the euro is the obvious one. We saw some very strange moves when you look at the ten year yield spread Germany against us against euro dollar. Around Liberation Day you saw some very strange moves reminiscent of Liz Truss days where yields went up and the dollar went down. Sterling was the case obviously with Liz Truss. But that spread, that ten year spread is around 190 basis points. That's a big anchor for euro dollar to try and drag along if it wants to go higher because you will see flows into eurozone. And this is why I think we'll see Euro dollar back below 110 this year. And then no presentation of mine would be complete without a negative sterling view. Here it is. Us UK relative one year change in rates. The UK is priced to cut less than the fed this year. That makes no sense to me. Cable's already overshot. I have to finish with a negative sterling view and that is it. I am very pleased to take questions. So Kayla.

Speaker 1: Great. Thank you.

Speaker 4: Lee. Thank you. Cheery as always.

Speaker 1: Okay I'm going to take one from that's come through on the iPad. Um, if anyone else has questions in the audience, we can address them with the microphone too. So this one, uh, this question that came through, uh, has to do a little bit with what Europe can bring to the table in terms of trade negotiations. And one specific question that they're asking is why not solve the trade balance, um, with by building up their defensive, their defense sector, by buying more US defensive goods? Why is that not being discussed? And is that an option?

Speaker 3: It is an option. I mean, I guess, I mean, you know, the German debt break and the removal of that was obviously a big driver of those flows I've just been talking about there. Um, the message is, though, they're going to buy domestic defence. That's why we've seen the Dax outperform so much. Um, look, I think the optimism over eurozone fiscal policy over others will follow Germany. I think it's been heavily overplayed. We're already seeing a little bit of sort of pulling back in Germany. You know, Mertz doesn't maybe have the power he thought he had. He lost the first round of the Chancellor vote, won the second. Um, but maybe there isn't suddenly this free. And let's go and spend loads of money in Germany. Attitude. That seemed to be the view in the first days after the election. And then also there was an assumption that others would follow Germany. Others don't have the room fiscally. Think back to a year ago with France when we were about left wing government or right wing government, and you saw the OMC bond spread blow out significantly. Um, Italy doesn't have the room, Spain doesn't have the room. The only country that really has the room is Germany. Do they really have the inclination to see this through, to the extent that the markets seem to expect? I think we're starting to doubt that now. And if you look at deficit projections for the eurozone for the next three years, around 3% of GDP, the US is 6.5%. So you've still got that differentiation. And this is where I think a lot of that optimism towards eurozone equities towards the eurozone is probably overplayed. If the fiscal side is the card. But to answer the question, could we solve the deficit if they bought us manufactured military? Yes. But that's not the message we're getting so far.

Speaker 1: Yeah, yeah. And we also see outflows in terms of real money out of European equities too. So real money is certainly participating in less optimism around.

Speaker 4: Europe.

Speaker 3: Having been strong buyers up until the last few weeks or so.

Speaker 4: Yeah, definitely.

Speaker 1: Okay. I know there was another hand. Oh. Right there. Sorry.

Speaker 5: Hi, Lee. Hi. Long time listener. First time caller. You did a pretty good job of saying the uncertainties are great, so you're probably wrong. And in a year's time, maybe you'll have to apologize. But where are you going to be wrong in a year's time?

Speaker 3: Um, it's a very good question. Where am I going to be wrong if the US economy is much weaker than I expect? If, you know, the tariffs come through much more quickly, that hurts the consumer demand for labor slackens as a result. And so the economy does nosedive pretty quickly. You know, the confidence numbers are right. Everyone pulls back because the confidence numbers the fed will cut more than three times or 2 or 3 times in that world, so that it really is predicated on the US economy doing better than people think. Now, my view here is predicated on that. Where I'll be wrong is if the US economy really does that, that confidence, that uncertainty really pulls back on consumer spending. It forces us to pull back on CapEx, etc. and the economy slows sharply and the fed has to react to that. If that happens, then I'm wrong. The dollar goes down from here. Listen, beyond this year, I'm sort of negative on the dollar from next year anyway. Um, I think the economy, everything we've talked about here catches up in 2026. Um, I do worry about the new fed chair, whoever it is and what that means for steeper curve and how that slows the economy, etc.. So I'm really looking at the last hurrah of the dollar rally this year. But if I'm wrong on growth this year and it doesn't outperform from here, then the dollar decline that so many people seem to expect over the next 2 or 3 years starts in 25, rather than my view, it starts in 26.

Speaker 1: I think we have one more.

Speaker 6: Hello. So since the beginning of the year, uh, Stephen Mirren's article is quite popular. So what's your view about about this article.

Speaker 3: Um, my view on the Steve Moran article. For those who don't know, this is the Mar a Lago accord. I think we have to take it seriously. I don't think we can dismiss it. I honestly think it is a desired policy. Um. Project 2025 for international finance, if you like. So for those of you who don't know, basically, you know, the mirror and paper Mar a Lago accord is basically supposed to be akin to the Plaza Accord in the 1980s. It's an attempt to devalue the dollar by about 25% with everyone. Foreign cooperation. Everyone sells the dollar. Um, and we get a 25% devaluation. Why do they sell the dollar? Why do they agree to this? Well, one the paper lays out and I'd recommend you all read it. It's published in November when it was at Harbor Bay capital. Um, the first reason is trade tariffs that, you know, you get relief from trade tariffs if you agree to this policy. The second one is the security blanket going back to Europe and military spending. Basically, if you don't agree to it, then you lose the US security blanket. Those are the two arguments behind it. I first read it months ago, thought, this is rubbish. You're never going to do this. And then as time has gone on, you think it's ticking each box when you read the paper and then you see the policy, it ticks each box down to details like including VAT and the tariff calculation. One that red flag yesterday. Listened to Kevin Walsh by the way. He talked about the fed has a triple mandate. He mentioned that yesterday afternoon. Not just, you know, stable prices, maximum employment.

Speaker 3: Um, the third one being. Reasonable and steady long term interest rates. That's what he said yesterday. That's the third mandate of the fed. I've only ever seen that mentioned in one other place. And that was Stephen Miron's paper. I've never heard of that third mandate for the fed. I still don't know if it's true or not, but it was in Miron's paper. And then Kevin Warsh said it yesterday, and he obviously is connected to the circles within the administration. The fact he brought that up was another sort of bell in my head that this policy is real. I think it's something they would like to do. Whether they'll actually get there, I don't know. I think it's the second half of next year story. Not anything before then because you need a compliant fed in order to do it. Powell doesn't go until May, so you need a new fed governor in place before you can attempt to do it. Also, you want to start from a low inflation point. Well, if the tariffs go on at some point in July it's a one off adjustment relative price. Next July they will drop out of the equation. So therefore you will probably start from a lower sort of inflation point. And you have a fed chair who's more likely to be compliant, so everyone should read it. Put your skepticism aside and just think about all the bits in there that are actually being ticked off as we go through. But next year's retreat, depending if we do around June, it might be a really hot topic next June at the retreat. So come back.

Speaker 1: Lee, thank you so much.

Speaker 3: Thank you Kayla. Thank you everyone.

Speaker 1: I'm not sure if the Mar a Lago accord is really a beach read material, but I'll add it to my list. Okay, so up next we're going to take a quick break about 15 minutes. Um, you know, stretch your legs. Let all of that good information sink in. Um, and we'll meet back here and start our shark tank. So thanks, everyone.

Speaker 3: All right. Well, thanks for that introduction. And yes, we'll be talking a lot about geopolitical fragmentation both today and with the two sessions tomorrow. Um, so what I want to do in the next 40 or so minutes is give you a quick introduction to Geo Quant and kind of how we try to provide a data driven approach to measuring political risks under Trump 2.0 and then really focus in on three different areas that we're tracking in terms of political volatility, geopolitical tensions and finally regime stability. And to start off a little bit with talking about what Geo Quant is. If you're not familiar with Geo Quant, we were formed almost about ten years ago, and the genesis for the company was really based off of the tacit recognition that politics matters. But in terms of being able to measure the impact of politics on market or commercial outcomes, that's traditionally been really difficult to do because we lack the same set of systematic, quantitative and standardized metrics that we have in macroeconomics. And so geo quant solution to this then, was to become a benchmark political risk data provider. And we are able to do that really only because of three innovations that have come about in the last ten years. First is the quantitative term in political science. So this allows us to kind of operationalize traditional political concepts like institutional quality, social instability or bilateral tensions. Second is the explosion of big data, particularly with news feeds, which allows us to track and collect information in real time on different political developments around the world.

Speaker 3: And then finally, are advances in AI, natural language processing, and time series forecasting. And so essentially what it does is we take the news and we turn it into numbers. And this is really useful because by providing a set of political risk metrics, we're able to not only provide two year forecasts of where we think political risks are headed, but we can also create early warning sign monitors for, say, regime collapse or interstate conflict, which I'll show you a quick preview of. Or we can apply it in Cross-asset and macroeconomic modeling, which is some of the cool research we've been able to do and is available on the insights platform. If you download the app and, you know, while there aren't a lot of comparable data providers in this space, we think that we're picking up some kind of signal of of substance because we have found that we're able to kind of consistently predict election outcomes, including last year's Trump victory, and that we're also able to kind of predict extreme asset movements by incorporating different politics. So in order to understand how Geo Quant is kind of tracking the different political risks, I think it's important to understand what are the political risks today kind of a reaction to and in many senses, Trump 2.0 is a reaction to the post 1945 international order, which has traditionally been characterized by kind of three pillars open markets, free trade, globalization, limited regulation, open borders, migration, capital mobility, territorial sovereignty and integrity norms, and then finally open ideas.

Speaker 3: Liberal internationalism, democracy shared norms about the use of force. But now what we see is kind of three different political challenges to these pillars. First, protectionism not only with the Trump tariffs, but overall trade uncertainty and policy volatility. Open borders under duress by kind of a US nationalism and America first foreign policy that encourages transactional dialogues and kind of self-help systems. And then finally open ideas challenged by the concepts of populism, anti-establishment sentiment and institutional distrust, not only in the US, but I think globally, as we saw with a lot of elections last year. So in order to kind of understand what what is happening, Geo Quant measures these three different political risks on a daily basis and then provides forecasts and analysis about what are the implications of this. So in the remainder of my talk, I'll give you some insight into kind of how we're kind of measuring and operationalizing protectionism, nationalism and populism, as well as leave you with some answers to what we think are some interesting questions about their consequences for markets in terms of the the kind of future of US safe haven status. The winners and losers of Liberation Day, and kind of the consequences of Trump's foreign policy for international order and stability more broadly. So let's start off with what's, I think, on top of everyone's minds, protectionism and really focusing in on policy volatility.

Speaker 3: And to understand a little bit more about how we measure protectionism. Let me tell you how it works. So what we do is we scrape about 4000 news articles every day from 323 media sources in nine different languages, and we apply different topic modeling and sentiment analysis algorithms to track about 40 different political risks, many of which you see in this taxonomy on the left hand side. And what's nice about this taxonomy is it allows us to track different types of, of of risks. So protectionism can affect, say, policy risks related to the predictability, quality or stability of monetary, fiscal, regulatory policy. We can track nationalism and populism based off of different social grievances or different, um, protest events. And then finally we can think about geopolitical and foreign policy risks not only related to overt military behavior that we might be seeing, but also diplomatic and military engagements. And by applying kind of this sentiment analysis, the different news articles we're tracking, we're able to get a really high frequency data series, which you see on the right hand side, where sentiment scores above zero correspond to higher risk, more negative news stories. And what we can then do is start to track some really interesting evolutionary trends in, say, policy risk, which is elevated during the first Trump administration above the white line, as we might expect that it goes down during the Biden administration.

Speaker 3: And as you can see, a very, very sharp rise, uh, since November, uh, and particularly with the inauguration and kind of the first 100 days in office, which Professor Drezner will talk more about tomorrow. And one of the nice things about having this high frequency data series is we can apply various kind of transformations of the data to create measures of political volatility. That is, we can just look at the historical variance of that political risk data I'm showing you and measure with kind of, uh, interesting, uh, clarity. How is political volatility behaving. And hopefully the geo Quant data kind of confirms that pit in your stomach you've been feeling for the last few months, which is that whether or not we're looking at a seven day rolling average in in lighter red, or a 30 day rolling average in darker red. That political volatility is up year over year, that it really began to rise last June following the first presidential election debate. And then it kind of has kind of consistently remained elevated since then and and remained high. And what's nice about this political volatility measure is because Giocante is able to provide forward looking estimates. We can see that political volatility is forecast to remain high, to actually rise over the summer, which I think would be interesting to kind of contrast with, uh, where this taco trade is going. Our data would say, actually, you should be very cautious about about Trump actually chickening out.

Speaker 3: Now, what's cool about kind of having this political volatility measure, if I can just geek out for a second, is that it seems to be capturing something distinctly different than how we traditionally think about volatility that if you look at, say, the VIX or the move or minval indices, we find that the way market volatility tends to move is not always aligned with how political volatility moves. And that's important for estimating its effect on on markets. Now if politics didn't affect market outcomes this wouldn't matter. But if politics is providing a distinctly orthogonal influence, then it provides interesting opportunities to examine that that risk premia. And kind of two ways to illustrate this is by thinking about how political volatility has begun to kind of erode the safe haven status of US assets. And I'll tell you a story in two ways. First, on on treasuries and then on gold. So starting last year at Geocon, we began to notice that this political volatility was really beginning to weigh in on, on the bond market, particularly around kind of the Trump trade narrative that we were seeing and that we thought it would be interesting to kind of create two different asset models, one where we create kind of a fundamental bond model that has kind of our general correlates of ten year yields, and then have a separate one that looks at what if we only looked at politics as kind of the input behind these yields.

Speaker 3: And of course, you know, macro fundamentals tend to drive most of the predicted behavior in yields consistent with what we would expect. But what we have now found is really since about last September, this interesting divergence between what the fundamentals would predict the bond market is doing and what our politics only model is now seeing. And our kind of internal explanation for this, and this is some of the research we published on on State Street Insights in January, is that this heightened policy risk and domestic instability is really outweighing kind of the more conventional geopolitical stressors that that would cause a pile in on the bond market, and that these political influences are expected to keep yields higher for longer. We conducted the same type of exercise looking at gold, where again, we had our macro fundamental model, and we compared that to a model that had a politics overlay on top of that. And once again, we found that really in the last few months, politics seems to be having a very distinct and interesting upward rise on on gold prices that US trade uncertainty, as well as larger fiscal sustainability concerns around, say, Trump's big beautiful bill are really causing this elevation in gold pricing. And that's what our kind of political model forecasts will keep gold elevated above 3000 really through through the end of the summer. Now, up until this point, I've talked about kind of policy risk and political volatility within the US.

Speaker 3: But of course, the US is not an isolated actor that US policy and political volatility has spillover effects on the rest of the world. And one of the cool things I think, that we've been able to do with our data is we can take that country risk data, which you see featured on the left hand side for, say, investment trade policy or international relations risk. And we can start to tease out bilateral engagements. That is, we can look at investment trade policy between the US and other countries around the world, using what we call our kind of contingent or bilateral risk measures. And so what you see in the map, then on the right hand side is a snapshot of bilateral investment, trade policy risk or bilateral economic tensions between the US and the rest of the world since January. And I think, unsurprisingly, you can see that risk has risen for countries like Mexico, Canada and China. What makes this kind of bilateral relations data really interesting, though, is the fact that it allows us to then explore in greater detail what are the effects of the Trump tariffs. And what we find kind of most surprising in our analysis is that there's been very disparate or differential reactions to the Trump tariff since Liberation Day. So, unsurprisingly, I think if you look at, say, economic tensions with China, Mexico or Canada, we see large spikes in that news flow, sentiment risk, that there have been very negative consequences that has led to this downturn in in economic relations.

Speaker 3: Um, but what strikes us is more interesting is how kind of more muted the reactions have been for countries like South Korea, India, Indonesia. And our hypothesis on this is, first that those countries have just less trade exposure to, to to US policy. And so they're a little bit more shielded. But in other cases, India or South Korea have already begun to kind of benefit from some of the regionalization or de-risking that was coming out of China even a few years ago, um, that South Korea was one of the first to enter into dialogue and negotiations to try to get ahead and institute kind of that 90 day pause. And so, looking ahead, what the data kind of forecasts is that industries that have really flown, um, to, to other countries will really tamper the long term volatility or risk associated with these tariffs. Um, and that the ongoing court cases, while they remain kind of uncertain, we think that they will probably have an influence on on negotiations and the bargaining positions, which creates interesting opportunities, particularly, I think, in APAC region for those countries to to kind of, um, grab better deals. Now, the kind of so what from this is, yes, geo quant data can kind of monitor the effect of these tariffs in real time. But what do we do from this from an investor standpoint.

Speaker 3: And one of the things that we've been thinking about at Geo Quant is how can investors really leverage these measures of geopolitical volatility that I've shown you that we've created. And what we have found in in a white paper that we just produced or published on on Monday is that there are some really interesting smart beta applications here where looking a month ahead or a quarter ahead, using that forecast data we have, we find that investors can really reasonably hedge against rising geopolitical volatility for different markets is a function of their trade exposure to the US as a function of kind of their their policy stability relative to the US, and that by kind of using this, this geopolitical tilt in a multifactor index, you can produce risk adjusted returns, which you see in kind of the color lines here for Canada, China, France and Taiwan that are on average, you know, 50 to 70 bips above kind of the benchmark index. And so that creates really nice and interesting kind of, um, opportunities to to kind of integrate this into more systematic strategies. So beyond protectionism, I also want to talk about two other risks nationalism and populism. And in thinking about US nationalism, I really want to focus in on foreign policy and kind of the geopolitical fragmentation and the way that we measure kind of these geopolitical risks is using that same relational database of contingent, um, data, this time focusing more on measures of military or diplomatic tensions between country pairs.

Speaker 3: So here you see, for example, our year ahead forecasts for US bilateral relations, um, with with the rest of the world between now and next June. And I think the main themes that kind of jump out in these forecasts is, first, that we expect kind of a downturn in diplomatic and military relations, especially in kind of Asia and Africa. And that a lot of this is kind of the ongoing consequences of Trump's new foreign policy alignment, that there's a lot of polarization of existing allies, not only in Europe and kind of trying to get them to increase their defense spending commitments, but also, um, in the APAC region, especially after this visit to the Shangri-La dialogue last week that we see the US kind of transactional foreign policy dialogues encouraging more arms sales, like the 142 billion sale with with Saudi Arabia earlier in May, that these arms sales and kind of transactional dynamics create more of a self-help system or what the international relations theory literature would call defensive realism, that everyone is kind of on their own, that they can't rely on existing allies to come in and defend them. And so they need to take measures to protect themselves. And this is kind of a dangerous situation to to kind of start being unfolding, because we also think that there's some kind of degradation or challenge to those post 45 norms about the use of force, that respect for territorial integrity, um, is is no longer as protected, that opportunities for mediation or dialogue, either through trusted inter-governmental organizations or diplomacy, is no longer the primary route or first kind of pair, And that can then have kind of these chilling effects on on bilateral relations around the world.

Speaker 3: Now, this kind of chill in US diplomatic and military relations is not unique to Trump. And in fact, if we kind of zoom back and look at how these bilateral relations have been changing since the start of the Biden administration, what we find is that actually there's been kind of a downturn in relations everywhere. And a lot of this comes out of the Biden foreign policy that, you know, some of the the, the bills that were passed with the chips or the IRA act and these other protectionist measures, isolated allies and economic partners that miscommunication from the Biden administration around, say, the Aukus submarine deal, one-China policy or other kind of democracy versus autocracy rhetoric all had this chilling effect on on on relations. And this kind of downturn in geopolitical cooperation with the rest of the world is significant because while the rest of the world was starting to drift away from the US, we find in our data that they were starting to move more towards China. That is, China was really exploiting a lot of opportunities during the Biden administration and are continuing to do so now by pursuing trade talks, by pursuing military base agreements or arms agreements with with other countries.

Speaker 3: And this has consequences for thinking about geopolitical fragmentation and realignments, because our data says that it's not a win. Will the the world start to fragment? Our data says that this is already happening And what our projection is, is that a Trump foreign policy really just accelerates this realignment towards towards China, which is what you see in this data here. And when I look at this, I think one of the things that kind of I found most surprising is how much of Europe and Latin America has really moved towards China and away from, from the US over this period. So what are the implications of this? Well, if we think about a more, uh, America first foreign policy or transactional dynamics, um, the consequence is an international order that is more unstable. And the implication is that the likelihood of international conflict rises under these systems. Why? Well, we know that the main tenets of international conflict tend to come from isolated allies. From increases in arms sales. Defensive posturing or changing rules based orders that limit opportunities for dialogue and mediation. And collectively, these risks can increase the chance of inter-state conflict, not by any kind of, um, belligerent, intentional invasion, like like the Russian invasion of Ukraine, but rather we see an increased risk of instability and conflict because of miscalculations, because of misperception and spiral logics that can arise from uncertainty about what are states arming for, what are they trying to kind of achieve that defensive posturing can often come across as aggressive, and that these can also contribute to increased incentives for preemptive strikes to to kind of lock in gains, to maintain my own security before you get too strong.

Speaker 3: And so what this means then is, in the context of that diplomatic and military tensions data that we're monitoring, we can look at select pairs of countries that we're most concerned about us, China, Russia, Ukraine, India, Pakistan. And look at what are the forecast changes over the next six months under kind of this this umbrella. And what we find kind of most concerning is this increased instability in South Asia that in just a few weeks, the India-Pakistan situation went from being relatively stable to all of a sudden escalating very quickly and then cooling off just as quickly. And that type of volatility is very unnerving. It creates a lot of kind of apprehension that it could happen again, that there could be some other type of kind of miscalculation. Um, we also see kind of Iran standing out here as a source of of instability either because these nuclear negotiation talks do not end well, which I think is consistent with kind of the terms that have been released by Wyckoff's team or Israel decides to kind of engage in this preemptive strike that has long been kind of a shadow threat against them.

Speaker 3: So that then has kind of, I think, strong market implications for not only Asia, but also global supply chains. The last thing I'll talk about is kind of populism and the broader wave of anti-establishment trends and kind of regime stability dynamics. Now, last year when I spoke at the research retreat, I spent a lot of my time talking about the global election super cycle. That last year was a very unusual, um, timing period in, in, in the world, and that we saw an inordinate amount of elections. And one of the most kind of interesting consequences of those elections Actions was how they turned out that there was a lot more incumbent turnover than than usual. The data that you see here shows on average, in the ten years before last year, we saw incumbents losing power about 34% of the time. Last year, we saw a dramatic spike that moved it up to 46%. As a consequence of kind of high government and regime instability, frustration over kind of post-pandemic restrictions, inflation, and that the high turnover rate then lent itself to large amounts of kind of policy change, volatility and ideological swings. Now in 2025 so far, we seem to maybe be turning back from that. Um, thus far we've only seen about 29% of elections result in turnover. And that part of that is due to the unusual effect foreign policy, namely the Trump trade effect, has had on elections in, say, Australia or Canada, where we would have ordinarily expected the incumbent parties to lose if, uh, the Trump tariff announcements had not galvanized kind of this rallying effect.

Speaker 3: That said, uh, while these anti-establishment tones seem to have softened, I think they're still really interesting or notable gains for the far right or even center right parties. Uh, AfD in Germany still made very strong gains in in February that the Poland and Portugal elections also saw strong movement towards, towards the right. Um, and that this is going to likely lend itself to greater amounts of institutional gridlock. And we think that this is consequential because, again, speaking to kind of the effects work that we've done for for State Street, we know that institutional gridlock can sometimes be positive for currencies that in cases like Poland, for example, the zloty might actually have some upside because it constrains the ability for tusk to kind of really pursue his agenda, and that that is oftentimes treated positively because it means stability in a country's kind of trajectory. But in other cases, like Korea, you can make the opposite argument that Lee's proposed kind of economic reforms for more decentralization, for increased checks on executive power, for softening relations with North Korea could also be really positive for for the yuan there. And we'll be interesting to watch. Now, outside of countries that hold elections, there are a lot of other markets at play. And we try to monitor those other elections through our regime instability indicators, which very much like our conflict warning monitors, try to look for early warning signs of regime collapse.

Speaker 3: Now, the main kind of warning signs we're looking for in in our data indicators are things like institutional redundancies or personnel stacking, which can often be used to kind of limit the power of elites or other institutional actors. Security force hollowing, which can check against coup risk or rebellion attempts, as well as kind of political and criminal violence growing or socioeconomic grievances broadening. And so what you see in the figure here is a measure of our regime instability indicator with levels on the x axis. Higher levels corresponding to higher regime instability. And then on the y axis the two year change in these indicators. And what we find is on the the markets that we have listed here, the vast majority have seen kind of a substantial rise in the last two years that this is generally driven by kind of the last two. Warning signs that I highlight this growing kind of political violence and instability, as well as social tensions, which suggests that populism sentiment remains kind of a potent force not only in countries holding election, but also in these other markets. And that's then consequential, because if we're looking for kind of the most precarious countries to keep an eye on, we want to look at those in the top right corner where we see already elevated levels of regime instability risk, as well as substantial rises over the next over the last two years.

Speaker 3: So Russia, for example, might seem like it's relatively stable, but our indicators would suggest that another Wagner insurrection like attempt is not at all out of the question, and that other countries like Ethiopia, which now has a de facto kind of conflict in Amhara brewing again and trying to match those IMF reforms, or Venezuela, which is constantly trying to, uh, limit, limit the limit instability, um, might actually be more unstable. And that could then have really interesting spillover effects for neighbors like Colombia, Brazil, Kenya, Egypt. So, in summary, won't be the the last person to say this, but a volatile world is here. And the implication is that politics is now starting to fray. Uh, on on markets for a longer time and with greater influence that this has implications for not only kind of safe haven assets, um, but that it also creates differential opportunities for looking at other countries, uh, trade policy, um, as well as their ability to kind of maintain policy stability in the wake of different elections for international order and conflict. What we find is that geopolitical fragmentation started a while ago, that the Trump foreign policy likely accelerates these these realignments. And that is particularly concerning because it creates a very destabilizing set of forces, not only in South Asia, but in select regimes in emerging markets as well. So thank you for that and happy to take any questions.

Speaker 1: So I think we have time for for one or maybe two. I think we actually have one here already. So here we go. Well done.

Speaker 4: So just a question. So is there any data that speaks to the topological structure of the world order. So are we heading towards a more US-China bipolar world or is more towards a multi-regional fragmentation?

Speaker 3: Yeah. So we've looked at this a little bit. I've played around with some Markov models, and I think it really depends on what are your base assumptions about how many different multi-polar worlds we could could see? Um, in most of the simulations I've run, it seems that the movement towards the US or China seems to outweigh kind of other regional blocs. Um, so that would suggest more of a bipolar or even centralization.

Speaker 1: So, so I have a good one here on the app, actually. Um, so, um, does geo political volatility measure control for the cadence of the electoral cycle? And if I may, I'm just going to do a little rider on it as well. Yeah. What would change the inputs into your forecast to predict political volatility to come down. What are the kind of factors.

Speaker 3: So the political volatility measure doesn't explicitly take into account, um, election cycles. Although I think just from the US measure we saw it kind of moves around elections in general. Um, in order for it to come down, I think we would have to see some of the underlying inputs be relatively stable. That is, we would want to see a lot more stability in, say, our social indicators or our policy indicators. Um, and those are hard to, to kind of, uh, stabilize because they tend to be very fast acting or reactionary, uh, to, to kind of news events, to, um, different policy announcements. Um, but that's, that's what I would look for.

Speaker 1: Got it. Thank you. Um, any any other questions? Okay. Well, Iris, thank you very much. And brilliant that you plugged the app. I mean, special, special prize for that. Thank you. Um, so.

Speaker 3: Thanks, Lee. Thank you. Lee. You'll notice he said that I was the past head of our PhD program in business economics. And that's been fortuitous for me because we certainly are having challenges bringing students, unfortunately, to Harvard these days. Um, so this is a back to basics talk. So what I'm going to be speaking to you about today is really trying to understand, uh, in the world of flows, you know, State Street puts out thousands of different flow series. And you've heard from Lee earlier and also in previous presentations about some of those series and how they relate to, to current trends and with with Alex Kim of Fox and with Haoran Jia, who's also here. We've really been trying to dig into this market and try to understand on the spot side and on the forward side what is driving, um, these flows. So what I would start with is just a very simple decomposition for you, which is start by just thinking about the US dollar and think about the players in this market and why they trade. Um, so if you we're going to start with the foreign buyers and sellers, and we'll get to the domestic. So us players in a in a moment. But if you're just thinking about foreign buyers and sellers, how do they trade in FX? Well, a couple of different ways. Well first in the spot market. So for example, a Euro based investor who buys, say, US equities or fixed income would go and have to buy in the spot market, have to buy dollars in order to do that, right.

Speaker 3: They may also play in the forward market. So really how could they play in the forward market really three different ways two in hedging and one in speculation. So the first is to the extent that they are hedging their purchase of their US say they're buying US equities just for sake of argument, they would be selling dollars forward according to whatever target hedge ratio they might have. You know, it's common parlance in the FX world that if you're a fixed income investor, you're most likely doing a full hedge of the position. Certainly many investors do that. If you're an equity equity player, You may or may not be hedging, and you may or may not be using a hedge ratio of of one. That's kind of 111 piece of that. The second piece is that unrelated to your current flows, depending on what you own already in that asset, as that asset changes value, you have to change your hedge position. So just for example, suppose that you are a European investor and you own a $100 of US equities. And the US equity market is on a tear and it's now at 150. Well, your $100 short US dollar hedge is no longer perfectly matched with your position. So you need to go and adjust that position. So you would have to respond to appreciation or depreciation of US dollar assets by selling additional dollars to get back to target hedge ratio.

Speaker 3: And then the third piece is of course, you might be doing some kind of speculating on the exchange rate. And by the way, that's observationally equivalent to just changing your hedge ratio. So for example, if I have a target hedge ratio of one and I don't keep that hedge ratio, it's kind of like I'm layering on a speculative position on top of that. Okay. Now on the right side here we have the domestic buyers and sellers. And it's really just parallel to what the the foreign investors are doing with one change, which is of course, just buying and selling within US dollar assets doesn't affect their FX position, but anything of the US dollar asset vis a vis the rest of the world, of course, is going to end up having a spot effect. And then in the forward market, it's exactly it's basically a parallel but flipping the sign relative to, um, relative to the foreign buyers and sellers. So you can see here if you just did this for people buying US equities, you can see we already have seven different, uh, excuse me, eight different categories. Right. We've got the spot and then two categories of hedging and one speculation on the one side. And then we have the same thing on the domestic side. And that's just for US equities. You could do that again for fixed income.

Speaker 3: And then you can do that again for the pure speculating category. So you can see that there's just what we're going to try to do in this. This paper is do that decomposition and see if there are things that we can learn from that. And really I'm most interested for today's purposes in making some quantitative statements like how important are these different pieces. So if you want to understand the FX market, should you be focused for US dollars, for example, on the hedging activity with respect to buys and sells today, or should you be focused on these other things, should be focused on speculation. What should you track? And so that's really our main focus today is to kind of give you some quantitative statements. And then I'll show you some results where we're looking at what is the relationship with current and then future returns in terms of predictability okay. So getting into the specifics a little bit more, just a bit of notation here. So suppose that you were we're going to focus on the left hand board for a second. Suppose you're buying US equities. And we're going to focus on equities today say of of value x. What does that mean in terms of these other categories. Well if you have a hedge ratio target let's call it HR. It means you're going to be short. So you're going to have a negative position of air times that x okay. That's going to be to hedge that position.

Speaker 3: And then similarly um if you have an existing position you will be adjusting your hedge ratio according to Air Times, whatever that return is that's experienced over that window times the position. Okay. And then we're going to call everything else that's left that you do in the forward market. We're going to call that residual speculation okay. Now I'll get to this a little bit later. But what one thing you'll see when you start to play with this data is that you have to be careful in what you call the residual or what you call speculation. Some of what we might call speculation is just people getting around to their hedging a little bit late. So you have to be careful in terms of what horizons you analyze the data. Okay. So what I'm going to focus on is really today mostly on the equity and foreign investor driven flows. We'll do a little bit on the domestic. And our hope is this is the first part of this research project. In the future, we're going to be adding the fixed income and domestic investors and really trying to paint you a very complete picture of this market and tell you what exactly you need to track and how you need to think about it. Okay. So one thing you might ask yourself is why do this back to basics stuff? Why don't we just go straight into developing a bunch of indicators, thinking about how predictive they are and so on? And I would say I'm going to give you four reasons that I think it's it's worth thinking about this.

Speaker 3: The first is that all of these components are linked to each other, right. So for example, you buy in the spot market, it automatically implies activity in the forward market. If you are a hedger um second is that they vary in their persistence. So for example if you're going out and buying US equities today, say you're a Euro based investor that actually predicts that you're going to be buying euros tomorrow. Euro based, excuse me, US dollar based equities tomorrow and the next day and actually out for a couple of months. So that means that it also because it's linked to your forward transactions in dollars. There's also generates cross-asset predictability. So in other words buying dollar equities today predicts that you will be selling in the forward market two months from now US dollars. Okay. There's also variation in their predictability that goes beyond the persistence because of cross-asset correlations. And then lastly they may vary. Although we didn't find a lot of evidence on this, they may vary in whether at that moment they're liquidity seeking or liquidity providing. In other words, when you're trading in one market, are you pushing the price against you or are you pushing or are you taking advantage of of, um, opportunities? Okay. So this is a one slide that I think is, is to motivate how, um, why we think this is interesting.

Speaker 3: And this is, I think the probably the first prediction that you might have as you start to think about what we've done here, which is to say that hedging modulates the relationship between returns and asset flows. That sounds fancy, but let me explain what I mean by that. What I mean is that if asset flows are liquidity seeking in FX markets, all that just means is when people are out buying stuff, it's moving the price up. Okay. Um, then what do we think? Well, if you're buying us equities, you're driving up the price of US equities. You're also driving up, say potentially the price of the dollar vis a vis foreign currencies. What we think is that to the extent that you're hedging that position and so you're say you have a perfect hedge, you're undoing that pressure on the dollar. That's a basic prediction that you would have. Okay. So here this is a very simple investigation of that idea. All we're doing is we're looking at each currency. The correlation between the returns and the dollar flows into equities. Okay. And you can see that being plotted on the vertical axis and on the horizontal axis. We've just got the average effective hedge ratio over that time. And what do you see. You see that in fact the higher is the hedge ratio. The lower is that correlation between returns and flows. Pretty intuitive. If the flows are completely hedged then it makes no difference if people are out buying equities or buying treasuries or whatever.

Speaker 3: It's not impacting the currency. So this is sort of illustrating this this basic idea. So Alex and I have worked in this area a little bit. So we have another paper that I presented here a couple of years back, and we're working off of some of the facts in that, in that previous work. Um, you know, particularly just to remind you of some of the things that we found. Uh, first of all, institutional fixed income investors tend to hedge their currency risk much more intensively and regularly and programmatically than equity investors. Uh, non US dollar investors are much more aggressive about their hedging than dollar based asset holders. Um, and then there's much more hedging over time. So hedging activity has risen hugely since 1988. But especially after the financial crisis, there's just been basically a step change in the levels of activity. And one of the things that we investigated in that paper was what we call dynamic hedging. Again, that's just a fancy way of saying that people tend to stick to their hedge ratios. So if you look in the data and you see that, uh, Lee looks like he has a hedge ratio of 0.5, that means that if there's a change in the underlying asset positions that he holds, we can predict pretty well that he is going to adjust his hedge ratios as well.

Speaker 3: So we see that in the data that people are pretty regular in their, um, their approaches and also lots of other, uh, lots of other work in this area. I think one of the advantages that we have here at State Street is it's just incredible data and insight into what people are actually doing. I mean, there's I can't tell you the number of academics who have ventured into this area. And one of the big challenges has been matching the data with the underlying portfolio data. And that's one of the things that you all in the audience to have access to, because you have both of these different flow series that, uh, that you can look at. Okay. So again, I'm going to come back to these demand equations that I mentioned to you before. Just as a reminder, we're going to call total demand the sum of spot asset flows. And then the forward demand which is this forward hedging, the forward hedging of returns and then the speculation component. And I already talked about how we would calculate this. And this slide really has just two kind of niche points about that calculation. The first is that these are pieces that we call hedging. You actually have to estimate at any moment you don't know what they are. You just know what how people have hedged in the past. And then you know how they're trading today. And it's really a question of can you label that hedging or not.

Speaker 3: And so what we do is we say let's look at what your hedge ratio ratio was in the past, and let's apply that to whatever you did in the spot market. And we're going to say that piece of whatever you're doing today that's your hedging piece of FX demand. Okay. That's the first thing to keep in mind. The second thing is you have to be very careful about horizon. So we're going to do this all of this analysis at the monthly level. You could do it at any horizon. You could do it daily. For example. One of the problems with doing it at too short a horizon is that there's some leads and lags, and it takes a while for people to actually execute these trades. So just for example, while you might execute if you're hedging a purchase of bonds, if you're buying some bonds today and you're hedging them, right, you might hedge those right away. On the other hand, if you own a bunch of dollar assets and they've appreciated it might take you a couple of weeks or a month or maybe even two months to adjust that underlying FX hedge position. So you've got to be careful. We might end up then calling something speculation, when it's really just that you've just taken a while to get around to your to your hedge, and you're actually going to see that a little bit in the data. Okay.

Speaker 3: And again, just as a caveat, we're going to focus on equity investors. Um, we think uh, and we're going to focus on foreign the foreigners for now. But I will show you some results with dollar domestics as well. Okay. How am I doing for time? Okay. Okay. All right. So as I said, we're going to estimate FX forward hedging. It's going to be this hedge ratio which is just a past number. And we've got that at the currency level times the spot FX asset flow okay. Again that's just think of it in this dollar example. This is Europeans and Japanese and others buying dollar assets times whatever their hedge ratio was in the past. That's an estimate of their hedging demand at that at that moment. As I said we have to estimate this hedge ratio. And it combines two elements. The first is intensive margin and then extensive margin. So what does this mean. The intensive margin is how much do investors who are inclined to hedge. So they're got some hedging program. How much do they do. So okay. So conditional on hedging. In other words how much do you hedge? And then the second is how likely are investors to hedge. This is the extensive margin. Okay. And you see quite a bit of variation across currencies in all of these things. We've detailed that in the other paper. But here is just a snapshot just to give you a sense. This is at the bottom.

Speaker 3: You'll see. This is um for US dollar investors. And this is just for a handful of currencies that I'm showing you. You'll see for the Aussie dollar for for US dollar, US dollar investors buying Australian assets for example, only 21% of them choose to hedge at all in the equity for equities. And when they do, the average hedge ratio of those folks is very low. It's only 4%. Okay. So basically a non-issue for American investors buying Aussie equities. Okay. When you look at it for American investors buying a European assets you can see the percentage you hedge. This is the average Going back to the late 90s is a little higher 29%. And when they hedge, it tends to be at a higher ratio. This is an average ratio of 38%. So really think of kind of back of the envelope. It's really the product of those two things. The 29% times the 38% that's giving you in aggregate the overall hedge ratio of that population of folks. Okay. So when you do that for the universe that we're looking at today, you'll see something that looks like in the equities world, something like that picture above. So that's the average foreign hedge ratio over this long period of time for the universe of people that we're looking at. The first thing I would point out is for equities. These numbers are kind of low right. So the largest one is for us. So investors do tend to be hedging their US equities more than pretty much anything else.

Speaker 3: But here even here it's 20. It's only 20%. Again think of that as reflecting both the fact that some people just don't hedge at all. And some people hedge with ratios that are less than one. So you get an average, which is about 20% here. And then when you look at some of these others like Australia, which would be consistent with some of those numbers I showed you earlier, the numbers are really are pretty small. They're sort of down below five 5%. Okay. Um, yeah. So anyway, all of this again this is aggregated. The key thing that makes the two the two figures are not actually directly comparable because the bottom one is just for US dollar investors. And the top one is actually aggregating over for foreign investors of all types for that particular currency. Okay. All right. So let me show you some numbers here. The first one this was a little bit hard to see, but hopefully you can see a little bit of it here. This is doing it for the dollar. Okay. Going back to the late 1990s. So the blue line I think is probably easier if I just point here. Um, see, the the blue line is the asset flow piece that's in the spot market. Okay. And then the two orange lines are the flow hedge and the return hedge. And then that purple line is the residual.

Speaker 3: It's that speculative piece of it. Okay. And it looks like a pretty noisy figure I would say the main thing that I take from this is that, um, components three and four at any meaningful horizon are kind of the important ones, which is to say the asset driven demand isn't it's important in some, you know, it happens, but it's not nearly as important as this return driven piece of it on people's existing portfolios, driving adjustments in their hedge position. And then this speculative piece, this residual. Right. So again this is for the dollar. But this this main lesson that it's these these items three and four. The return hedging and the speculative demand kind of carries through almost all of the currencies that we have that we have looked at. Okay. Here is the same picture for the euro and the yen okay. Again this is focusing just on the foreign investors for now. Just to give you a sense, um, so you see um, for the euro, the return driven piece is actually, if anything, the most important, um, you will see here sometimes what looks like a negative correlation between the speculative and the return hedge piece. Of course, that's kind of intuitive because the speculative piece is, uh, is this, uh, residual. So some of that is driven by the fact that it just takes a bit of time for people to actually adjust their, um, their hedge positions to what's happened with returns. Um, when you look at Japan, it's actually a little bit different.

Speaker 3: The asset flows tend to matter a little bit more here in terms of thinking about about what's happening. And then the returns based peace and flow based pieces are less important. Okay. So I keep using these expressions like less important, more important. If I want to be a little bit more rigorous about how I might say that. Here's a simple way we would do it. We'd say, okay, let's look at just the ratios to total volatility. So we'll compute each one of these series for each currency. And then we'll just compute the volatility of each one of those compared to the total flow that we're seeing here. And these are those ratios. And you can see in general it's not true everywhere. So um so Australia for example would be an exception to this. But in general you're seeing kind of just more volatility in the return driven piece of the FX behavior. And the speculative piece of this um, affects FX trading. Okay. I mentioned earlier these autocorrelations. So how persistent are these different pieces? So here we're looking at asset flows. Remember asset flows the flow hedge return hedge. And then the speculative piece. You can see the asset flows in the flow hedge are quite persistent even out to four 4 or 5 six months. The other piece is much less so. I think that's pretty intuitive. Remember numbers two and three here are just imputed numbers based on our estimated hedge ratios.

Speaker 3: Asset flow is inherently quite persistent because when people buy US dollar assets, they tend to buy them tomorrow and they tend to buy them five, five weeks from now and so on. That means that the flow hedge piece is also quite persistent, right? Because if you're hedging, then we sort of know that not only are you going to be buying more dollar assets tomorrow, you're also going to be playing in the FX market accordingly. Again, less persistence in those in those other three. Okay, this one's a little trickier, but I want to go through it with you, which is that the leads and lags here of these currency flow pieces are also interesting and somewhat intuitive. So for example, you've seen I'm not going to uh, we've blanked out here, of course, the diagonal. But you could see here, for example, that asset flows have some persistence from month to month. That number is 45% okay. You can also see here, um, that the asset flow and the flow hedge for example, are correlated in a cross-serial way. So here you see this -38%. More interesting thing here is if you focus on this last column the speculative. And you'll see things here like the speculative piece is correlated with the return hedge that we imputed last month. And it's actually to the of 18%. So that seems like a pretty high correlation. Why is that happening. That's happening because the there's a lot of hedging that's happening with a lag in response to past returns.

Speaker 3: Right. So if you have a portfolio it appreciates in value to some extent. You'll be doing that hedge today. But you'll be actually executing that trading over a period of time okay. So again important lead lag relationships driven by the dynamics of hedging. One of the challenges is thinking about exactly what horizon we should be doing that at. We've done this here at the monthly horizon okay. So one question you could ask is you could say let's try to understand. Which of these pieces are liquidity seeking in the sense that they're correlated with current returns or not. Right. And I'm just going to show you a little bit of evidence along those lines here. So this is doing this in a panel regression. And we're just looking here at the current spot FX return. So this is just the percentage change in the exchange rate all expressed in US dollar vis a vis the US dollar okay. And then we're looking at that for each of these pieces. There's the asset flow again the flow hedge return hedge and the speculative piece okay. And then one thing that I've done here is you can combine one and two right. So you can say basically let's look at the asset flow net of the hedging. And we could call that the hedged flow. Right. So that would be just another way of of looking at this data.

Speaker 3: And what you see is if you look in the right hand column here overall all of these are to seem to be positively correlated with with current returns. I would say this was actually somewhat surprising to me, which is to say that all of these are liquidity seeking, including the speculative, speculative piece of this. Again, this is foreign investor. So foreign investors are not, uh, you know, they're basically looking like momentum investors here, uh, vis a vis. And each one of these components vis a vis the currency markets. Okay. So, so far, everything I showed you was with just the foreign investors. You could, of course, add in the domestic players as well. Remember, the domestic buyers and sellers, this is us investors in the spot market. They're buying and selling dollars. If they're trading in foreign assets, um, in the forward market, they would be doing that by hedging purchases of their foreign assets, uh, in which case they'd be buying US dollars forward, responding to appreciation of foreign assets by buying additional US dollars forward or speculating on the exchange rate. Here is a place where we have less granularity at the moment. And so the only way we can look at this right now is by aggregating all of the components in the forward markets, and we're going to call that domestic FX demand. So we'll have one more one more bucket to look at. Um, now one of the reasons to do that, frankly, is we just have very good data and much better data at State Street because of our data.

Speaker 3: Some of our coverage. So we have lots of dollar based, for example, domestic players. And this is just showing you just a sense of the fraction of volume that we're able to track in foreign versus domestic players. So it's a it's an important set of players to look at right now. We don't have that full level of granularity that but we're working on that that we would love to to have. Okay. So I think some of the things that you might look at, I'm really just going to focus on this first one right here. It's really a mixed picture that emerges when you look at what the role of the domestic players is vis a vis the foreign investors. The first is if you look here with correlations with foreign demand, it looks like the domestic players are sort of providing liquidity in response to foreigners hedging returns and hedging returns with a lag. So they're stepping in to essentially provide that liquidity in that market. You can see that's the correlation between the foreigner position and the forward market. And then those different components. Okay. Now the challenge is that the one you look at returns, the domestics look more like momentum players. So it's kind of it's a little bit of a that's why I said it was a mixed picture. They look like they're momentum players in the sense that they tend to be buying dollars when US dollar assets are doing well.

Speaker 3: On the other hand, they tend to be providing liquidity with on average with respect to the FX transactions of of the foreign investors. Okay. You can then of course look at these folks here as well. This is doing that same regression but now adding in the domestic flow piece. And I would say there's some negative negative loading here. Um aligning with some, some some reversal. Okay. Pardon me. Oh I believe um, actually this is why I said, um, this is looking at the future returns. And so now we can look at to the extent that you have persistence or reversion, you can use these different components to try to understand, uh, relationship with with future future returns here. So overall, um, you know, it hasn't been our main focus really is trying to generate, uh, forecasting results. I think one of the challenges with forecasting results is you want to think a little bit more carefully about the scaling. So right now we're thinking about everything in dollar terms as adding up. But when you're thinking about it from a forecasting regression, you probably want to have more natural units for each individual currency. All of that said, what we find is we find some forecasting power when we combine all of these in a, in a, in a, in a regression, you get actually much more action when you start doing currency by currency type analyses.

Speaker 3: So this is looking at for example the Japanese yen. So contemporaneous returns and future returns. And um, you know, you get uh, basically what looks like, uh, versions of momentum with respect to flows driving future, future returns in the data if you do that. Um, yeah. We've looked at some different currencies like euros, euro as, as, as well. So the currency by currency level It's more interesting if your main goal is to try to do forecasting. Okay. There's some related applications. So Alex and Jaron can talk to you in the break if you're interested. They've put out some other work on this. There's some notes on dollar neutral pH strategies. This one is based on essentially a form of a currency momentum. But it's drawing on the flows and hedge ratios to do so. And building on some of the intuitions that that I've talked about here. There's also a dollar timing, some work that that Alex has done, um, that is related to some of these ideas here. I won't go into the full details of how these strategies were connected, but if those of you who are interested, um, can can find it. So in terms of our next steps, I think what we're interested in is further dissecting that domestic component of flow. As I said, we have that in a very aggregated way, um, at the moment. And I think once you start to think about the world in this very disaggregated way, you get a bunch of interesting insights emerge.

Speaker 3: I'll give you an example, which is, um, you start to kind of see, well, for example, when a, uh, assets in a particular currency appreciate, you can predict the rebalancing by some investors back to target levels. Well, once you can do that in the underlying assets that generates, if you know how they hedge their positions, also generates predictable FX movements. And to the extent that they're playing in these markets. But of course, they're doing all of these things with leads and lags. And so there's interesting action in terms of thinking about what happens in one market today and how it plays out in another market in the future. Now, one of the challenges is, of course, separating out that speculative piece, that residual. But even on just the mechanical piece, there's um, there are interesting things to do. And then the last thing I would just point out is this has been very focused on the equities world, where the hedging is actually there's not that much hedging going on. But in the fixed income world, we know hedging is a much more important phenomenon. And so the hedging modulates some of this cross-currency action much more significantly there. So that's our next step. And next time I see you I'll be bringing you some of those results. And we're going to try to look at all of those in an integrated manner. So let me just stop there. Happy to take questions discussion feedback.

Speaker 4: Thank you Robin. Super great presentation as always. Um, have you considered introducing a normative component to this analysis? You know, you're showing what people do. Yeah. You could at least conceptually, um, estimate what the risk minimizing hedge ratios are for all of these portfolios. And then look at how what people do differs from what they should do. Obviously, you'd have to make some assumptions, but you're an economist, you're happy to do that. But is that is that something that you think would be. And it also gives you another angle at what's speculative and what's not.

Speaker 3: It's a very I actually I love that idea. So Alex and I have this previous paper, which is probably probably easier to do it in that context than right here, because right here we've already aggregated. Right. But actually because of the amazing data, you can look at the individual players and see what they do, and you can compare it to whatever benchmark you'd like to construct and then ask the question. I think that's a very interesting exercise. Again, I think you would want to do that on an individual basis. By the way, it gives you a related question, which is to the extent that people are or investors are deviating from whatever rules we might give them. Is it serving them well or is it serving them poorly? It's something that we've talked about extensively. Can we give people advice like, geez, you know, historically over the past 20 years, you've had this terrible strategy. Why don't you try this? So I think we could certainly the calculations aren't that hard to do. The data is a bit more challenging because it's that you're looking at the portfolio level, and you really have to be sure that you're getting both the piece and the underlying assets right, for the individuals. We have much less of that concern when we're doing this aggregate work, because we're already kind of adding it all up and looking at the currency level. But at the individual level, it's a very interesting exercise.

Speaker 5: I have a question. Thanks for the presentation. Um, I'm pretty sure you have this data on a customer client level, but also there is a lot of heterogeneity Entity within a client. So same client, same investor may have multiple. Different portfolios with different restrictions. And I think there may. Be a lot of information there. There's just one example. You're investing in equities in two accounts let's say. Same client or portfolio manager. One allows for derivatives and the other does not. Do you have that data in that granularity. And if you do, do you think it would be useful to separate the speculative versus the hedging ratios?

Speaker 3: It's a great question. I have to be I don't want I don't want to speak out of turn here. The you know, one of the things that I've learned, one of the great pleasures of working with folks at State Street is with how much respect they treat, the confidentiality of the customer data. So, in fact, I don't believe that one would be able to do what you're describing conceptually. It would be a great thing to do, but we are mostly looking at the underlying behavior and trying to guess that kind of thing rather than, um, uh, really, uh, use that. And again, that's due to confidentiality constraints. I couldn't I'm not that well versed in what those are, but we, they, they do tend to be very careful, of course, to protect client information.

Speaker 2: So Robin, let me ask you one from the iPad. Um, if you were building a predictive strategy for FX, what series would you combine to do that?

Speaker 3: Um. We have what we have seen. I mean, I would rather defer this question, I think, to Alex, if he can, because he's written on this. Alex, can you can we hand Alex the mic? He's he's written on this. And I think it speaks to the work that I've done, but I don't think I think he can give you a better answer.

Speaker 6: Yeah. I mean the short answer is all of them. They all sort of sorry, they all sort of have some value to the the pretty squiggly lines that were going up. They're basically they're combining all these components. They're combining the hedging by the equity people, the hedging by the bond people, the overall FX flow and the cross-border demand for equities, which is, as Robin has shown you, it's generally unhedged in the majority of its its volume. So roughly speaking, if you do a simple model where you lump them all together, add a bunch of Z scores, you actually do pretty well. And with that, that picture was showing you is a sort of a waterfall of complexity. So the line at the bottom is what happens when you use two components. And the more complexity you add, the more value that you get. So we basically found that there are incremental there are incremental pieces of value to each one of these components of FX flow. And the work shown today has shown that there's yet more value. And slicing them up into the at least approximate motivations of the hedging flow. So we anticipate that will result in stronger potential applications in the future.

Speaker 2: Thank you. Alex. Um, any other questions from the room at the moment? I've got a couple more on here. Okay, let me go with the iPad. So, um, one here about any plans to use this in emerging markets? Um, we obviously have the data on emerging markets, much lower hedge ratios. But have you thought about trying to see if you can do this work with emerging market currencies as well?

Speaker 3: Yeah, I think that would be very interesting. One of the challenges with emerging markets is when people do proxy hedge, I call it proxy hedging. I don't know what the right term would be, but.

Speaker 2: That's a good term.

Speaker 3: But basically using a close cousin to do the hedge relative to the actual thing. So that complicates then how you would think about that analysis.

Speaker 2: So using the Mex to the Mexican peso to hedge some other risk because it's it's more liquid.

Speaker 3: Exactly. And everything we've done that I showed you is on very much this matching concept. Right. We've got your your buying dollar assets and you're hedging in that same currency. So we would have to think very carefully once we kind of go outside the G10 of how to deal with that and deal with some of those correlations. But I think in principle it's a great exercise to do.

Speaker 2: I think Dave and Mark have done work on that previously, but the Aussie is also used as a proxy hedge for emerging markets, so that might impact the Aussie results as well.

Speaker 3: That would be very interesting to look at.

Speaker 2: Yeah. Um, anyone else here? Okay. Um, early on you showed an average US dollar hedge ratio of 20%. If you conditioned just on those who hedge, what would the average be?

Speaker 3: Do you remember the number in our paper?

Speaker 2: So it's about 50%.

Speaker 6: About 50% is about 50% for, uh, we look at Euro CAD and sterling. It's in the paper by the way. But it was about half.

Speaker 3: Yeah. So meaning about half of them hedge and conditional on hedging. Yeah. They hedge about at a ratio of one half that's in the equities. But in there once you go to fixed income the numbers are higher across the.

Speaker 2: Board, like 50% for equities is high anyway. Yeah, but.

Speaker 3: That's for dollar assets.

Speaker 2: Yes.

Speaker 3: For dollar assets the hedge ratios are lower once you go to the other currencies.

Speaker 2: Yeah I'm time for one more quick one. Um is there any application of this work for predicting month end flows? Because a lot of the hedging you're talking about happens on the last day of the month. The month end flows are very important in FX, and for FX returns any application of this to predict what currencies are going to be bought and sold at month end.

Speaker 3: I think that question kind of answers itself right, in a way that, um, the many of the results that I showed you, you have these cross-serial pieces and the Cross-serial pieces are very intuitive there. You see, the returns go up and then you see the FX trading a bit later. Right. So I think it would give you the prediction would be to I think you would see that concentrated at month end.

Speaker 2: Yeah. I will tell our spot traders they will be very interested.

Speaker 3: Yeah. Very good. Thank you.

Speaker3: Okay. Thank you. Lee. Thank all of you for joining us this afternoon. So one thing I will disagree with, I think, uh, Dave or Will or Meg could probably do a better job than me, but that's okay. Um, so yes, I'm going to talk about, uh, nowcasting and talk about how we can use relevance based prediction to do nowcasting. This is joint work with, um, Will Kinlaw and Dave Turkington. And it's based on earlier work with Meg Czasonis as well. So I guess most of you know what a nowcast is. I actually didn't until Dave and Will asked me to get involved with this project. So it's a prediction of an outcome that's already in process and whose final value has not yet been determined. But there's information that's being released along the way to this final value. So GDP is a really good example of a variable that is suitable to nowcasting, because its components are revealed throughout the period to which it pertains. So, you know, during a quarter goods and services are being produced. Right. But we don't know what that's all going to aggregate to. But we're getting information along the way. We get an initial estimate, you know, a first release of GDP. And then typically we get a revision. So what we're going we actually have some empirical results, and we're going to nowcast the final value of GDP. That's the project here. And as I mentioned, we're going to use relevance based prediction to do so.

Speaker3: I've spoken on this topic a few times to some of you. So a little bit will be repeated. But there's a lot of new stuff that we've done since the last time I've talked about relevance. So anyway, relevance based prediction is a model free routine that forms a prediction as a weighted average of outcomes in which the weights are based on a very precisely defined statistic called relevance. And it has several appealing features. One is that it's prediction specific, right. It tailors the choice of the observations and the variables to each individual prediction task, where you know that you can contrast that with a model, which is, you know, it's a single calibration that's designed to handle all prediction tasks. Right. We're going to do this prediction by prediction. It's fully transparent. So it reveals precisely how each observation informs the prediction. And it shows you exactly how each predictive variable contributes to the reliability of the prediction. And unlike any other prediction system that I know of and I know about a lot of them, it gives you advanced guidance about the reliability of each individual prediction. So before you even make the prediction, you know whether or not it's going to be a good prediction. That's very useful. It's theoretically justified by information theory, the central limit theorem, the Mahalanobis distance, and some really cool mathematical convergences which I'll highlight along the way.

Speaker 3: And what's really pertinent to Nowcasting is that it's it's, um, it's very resilient to missing information because that's the challenge of an outcast. There's information that's missing. We have some information, but not all of the information. And this will preserve more information than model based approaches to predictions. And the other thing is that it enables the formation of a nowcast for variables that have no observable components. Okay. So let me show you more specifically how relevance based prediction works. It has three key features relevance, fit, and grid prediction. You should be thankful. I used to take about six slides to go through this information. So it's all it's very compact now. So relevance measures the importance of an observation to prediction. So you see the first line of this formula shows that it's composed of similarity and the average of the informativeness of an observation and the informativeness of the current circumstances. And the second line just expands, you know, it shows you what we mean specifically by similarity. So just set aside for a moment minus one half and look at the rest of that second line. This is a mahalanobis distance. This is the distance between an observation and current circumstances. So x I is a vector of the values of a set of predicted variables for a particular observation. And XT is a vector of the values of those predicted variables for current circumstances. So that's just showing you independently how the variables of observation differ from the variables of the variables for the current circumstances.

Speaker 3: Then we multiply by the inverse covariance matrix. This accomplishes two things. One is it captures the interaction of the variables, and the other is that it standardizes everything by dividing by variance. Right. And then we post multiply by the transpose of the difference in those two vectors. And that collapses this into a single number. So the y the minus sign because the Mahalanobis distance is a measure of differences. Right. And we're looking for similarity. So we need to put a minus sign in front of it and then y1 half. This is a little trickier. It turns out that the average squared distance of observations from each other is twice as large as the average squared distances from the average, right. So in order to put similarity and informativeness on the same scale, we have to divide by two for similarity. The next line down is how we measure the informativeness of an observation. And again, it's a mahalanobis distance. The only difference here is that instead of XT, which are the values of the variables for current circumstances, the prediction circumstances, it is the average, right. So here we're measuring not the distance between an observation and current circumstances, but the distance between the observation and the average of all the variables. Right. How unusual is this observation. And this comes from information theory. The fact that we include this comes from information theory which tells us that um, information is inversely related to probability.

Speaker 3: So unusual circumstances or unusual observations convey more information than common observations. And then the next line is again it's a measure of the distance from average. But here we're just measuring the difference of the current circumstances from average. And the reason we do this is because by doing so we center relevance on zero. So I don't know if you know, that may seem complicated to you or not, but the you know, the bottom line here is what you see in that blue box. Observations that are like current circumstances but different from average are more relevant than those that are not. And we're going to form our prediction, as I mentioned earlier, as a weighted average of past outcomes where the weights are this are based on this measure of relevance. The next key feature is what we call fit. So fit tells us it quantifies the prevalence of useful patterns in a data set. To the extent there are more patterns than we're going to be able to come up with a more reliable prediction. So the way to think about this, in fact, a lot of this research on relevance, a lot of the insights came from viewing data not as distributions around means as we've all been taught, but rather as collections of pairs. So consider a pair of observations that go into a prediction task, right.

Speaker 3: Each observation has a relevance weight and it has an outcome. So if they both have high relevance and high outcomes, they're aligned. If one has high relevance and a low outcome and the other has high relevance and a low outcome, they're also aligned. But if one has high relevance in a low outcome and the other has high relevance in a high outcome, they're not aligned. So what fit is doing is measuring the average alignment of all of the pairs of observations that go into a prediction task and using standardized terms. Perhaps a more intuitive way to think about fit is that it's the squared correlation between the relevance weights and the outcomes. And here's one of the interesting convergences I want to talk to you about average fit. So if you take fit and you take a weighted average of it across all of the prediction tasks, it converges precisely to r squared, right? So r squared tells us how much how confident we should be in a model. Right. But it's it's an average. It's based on some good predictions some bad predictions some so-so predictions. Right. What fit gives us is the r squared analog for an individual prediction. It tells us how how confident we should be in that specific prediction. And then the, um, the final feature of relevance based prediction is what we call grid prediction. And this allows us to consider a vast number of combinations of predictive variables and observations to come up with a composite prediction.

Speaker 3: So the way I like to think about it is to contrast it with, say, a neural network. A neural network proliferates a massive amount of parameters to handle all of the complexities in a data set. Right. What we're doing is considering a massive amount of combinations of predictive variables and observations to handle all of these complexities in a data set. And what we're doing, I think, is much better than, say, a neural network because it's more efficient. Right. We're doing this task by task. When you build a neural network, you have to be, you know, that what you end up with, that model has to handle the complexities for all of the prediction tasks. But we're doing this task by task, right. So we only have to handle the complexities and the data for a specific prediction. And then the other thing is you'll see is this has this is fully transparent. You see everything about how the prediction is formed. And and you know how reliable it is. So let me just, um, go through a little toy example to illustrate how this works. So as I mentioned, Rbpp relevance based prediction computes a prediction for a unique combination of observations and predictive variables, which is then entered into this grid. Right. So here we have on the left we have observations t for the current circumstances and then observations.

Speaker 3: Then you know, for all the observations out there we have the values for the variable we're predicting y. And then we have all of the predictive variable values. Right. All of the x's right. So based on those formulas that I just showed you from the x's we compute similarity the informativeness of an observation. The informativeness of the current circumstances. And from that we get relevance. Right. And then we rescale relevance so that the weights sum to one, and we form a prediction as a weighted average of that final. So we take the final column and multiply it by the y column. And we get a prediction which in this case is 19.4. And it has a fit associated with it. Right which is 2.32. So that is just one of the cells in this grid. So this is the prediction grid. It computes a composite prediction is a reliability weighted average of all of these cells. So you so each column is some combination of predictive variables. And each row is a subsample of observations based on a relevance threshold. Right. So each cell has within it a prediction and an associated fit for that prediction. So what we could do and what we used to do until Meg told us not to. We used to just look through this grid and find the cell with the highest fit. Right. But that could be problematic because there could be a data error there, or there could just be some reason why we don't want to rely exclusively on that one cell.

Speaker 3: Or you could have like you do in this case, you have two cells that have the highest fit, right? And they have different predictions. So what we instead do, rather than just taking a single cell with the highest fit, we take a weighted average of all of the cells where the weights are. The relative fits the relative reliability of each prediction. So we're sort of diversifying across many different calibrations. And you can see here if you go down the first column we're using all the predictive variables and the 20% most relevant observations. And that's what we got from that slide just before. So that's just one entry into this grid. Now if we if if you look at the upper left cell this is another remarkable equivalence right. We're using all of the predictive variables and all of the observations. The answer we get from that cell is exactly identical to what you would get if you ran a regression analysis. So regression analysis is a special case of relevance based prediction. And any of the predictions that come from the first row assume that the relationship is linear. Once we go beneath this first row right we're looking at different subsamples. And those different subsamples are capturing conditionalities in the data. They're capturing situations where the relationship between the predictive variables and the outcomes changes from what the overarching relationship is.

Speaker 3: Right? A neural network would have a parameter for that kind of a situation. We just have a different subsample of observations. Okay. So the grid actually has another great feature, which is that it naturally gives a comprehensive measure of variable importance, which is better than any other measure of variable importance. Um, it's basically what we do is we take the average adjusted fit of the grid cells that include a given variable, minus the average adjusted fit of the grid cells that omit that variable. Right. And we have to scale it properly. And that's how we measure variable importance. Now this overcomes the limitations of t statistic or equivalently a p value which only measure marginal importance. Right. So if you're looking at a t statistic and you have two collinear variables, they both may be really important independently, but when you put them together, they cancel each other out and they don't show up as being important. Well, relevance based importance measures total importance rather than just marginal importance. It also because we're looking at those those rows below the top row, it accounts for shifts in the relationship between the predictor variables and the outcomes, which t stats cannot do. And it's very close to a Shapley value. For those of you who do machine learning and use Shapley values, this is very close to Shapley value, but it's better than Shapley because it accounts for an individual prediction's reliability, whereas Shapley doesn't, because Shapley is just dependent.

Speaker 3: You know, it's it's computed from the predictions. It doesn't consider the outcomes. We're considering both predictions and outcomes. So another useful feature of this approach, which I mentioned at the outset, is the fact that it's resilient to missing information. And the way we handle missing information is we just assign a zero to the relevance weight of observations with missing data, which seems rather simple, but you see, it's actually quite elegant. So here's an example that demonstrates why it preserves more information than model based approaches. So a toy example only two predictive variables right. And we have some of those observations have missing information. So we have a choice. We can get rid of the observations with the missing information. Or we can get rid of the variables with the missing information. Right. So if we get rid of the observations with the missing information you see that's the first panel. We're going to give up that information that's there right. If you go down for variables two, three, and four. We're giving up the values associated with x2, and for variables seven and eight we're giving up the values associated with x1. So rather than getting rid of observations with missing information, we can get rid of variables with missing information. But again we lose a lot of information that way right. So if we get rid of x2 then we're missing all of that information plus three observations for x1.

Speaker 3: And if we get rid of x1 similar situation. So if we get rid of the observations with missing data we lose ten pieces of information. Or there are ten pieces that are missing. If we get rid of if we omit variable two, we're missing 13 pieces. Variable 112 pieces. But because the grid is considering all of these calibrations, right, it's only missing five pieces of information. Now this is we're not talking about missing information for observations, but what about missing information for. Well, this this is, uh. Excuse me, this slide is missing information for an observation. So, as I mentioned, it assigns zero to the observations with missing information. Now, why does the other feature here is in addition to preserving more information than you would from a model based approach, this takes into account the relative importance of the information that's missing. Right. So, um, if you're missing information for an observation that's relatively unimportant, right? That observation is going to have a relevance weight close to zero anyway. So by assigning zero there's not going to be much difference. It's just not going to matter if you're missing information. For an observation that really is important to the prediction, right? Then that observation is going to get a zero weight, right? So that particular element in the grid is going to have a much lower weight in the composite prediction.

Speaker 3: Right. And the composite prediction. So it won't you know it'll change the composite prediction in a meaningful way. And the composite prediction itself will be less reliable. Um, but if we're in the case of Nowcasting, the information we're missing is about the XT, right? The prediction circumstances. If this is the case, if the values for the prediction circumstances are missing, we can't compute similarity, which means we can't compute relevance. So therefore we have to give a zero weight to the entire cell, not just the observations with missing information. Um, so the predictions formed only from variables whose prediction circumstances are known. But again, by doing so, by assigning signing zero, we're taking into account the relative importance of the missing information. Okay. So I mentioned that one of the reasons that I like this approach is that it has a really solid theoretical foundation. It's um, it's not just based on heuristics or some empirical validation, but it's grounded in some really important theory. So it's grounded in information theory, which shows us that the information contained in an observation is the negative logarithm of its likelihood. It's grounded in the central limit theorem that tells us that the relative likelihood of an observation from a multivariate normal is proportional to the exponential of a negative Mahalanobis distance. That's a good line to remember when you're at cocktail parties. Um, but so you can think of the Mahalanobis distance, it's like a squared Z score.

Speaker 3: You know, it's a multivariate version of that. And then it has these really cool convergences. Like as I mentioned, if you use if you apply RVP across the full sample of observations and predictive variables, it gives you the same prediction that you'd get from a regression equation and regression. You know, there's some really solid theoretical justification that Carl Gauss gave us for, uh, for regressions. Then we have this convergence of fit to R-squared across all the prediction tasks. And then we have this convergence of RBI relevant relevant based importance to a T stat right. Rbi averaged across all prediction tasks converges to a t stat in the absence of collinearity and conditionality. And then we have the fact that RBI also converges to a Shapley value across all of the prediction tasks. And the final plug I'll make for this approach is that the cornerstone of ChatGPT in all large language models is a concept called attention, which is really relevance. So ChatGPT uses relevance to predict the contextualized meanings of words from their unconditional meanings. Exact same formula. So let me now get into how we can apply this to Nowcasting. So we're going to nowcast quarterly GDP growth final revised value seasonally adjusted. But what we'll show you our annualized values and um Nominal not real. And these are the predictive variables that we're using. Four of them are prior circumstances and the rest are variables where we get information released about them throughout the period.

Speaker 3: The ones that are in light blue are actual components of GDP, and the ones that are in dark blue are not components. So these are our predictive variables. Well, did you want me to say more about those. Okay. And here are some results. So what we're showing here is the average GDP outcomes for high versus low nowcasts. So full sample is all of the predictions that we or all the nowcasts that we made. And what you see is for those that we predicted, those periods that we predicted would be in the top half for the full sample, the annualized GDP growth nominal was 1.27. For those that we predicted would be in the bottom half, it was 0.89. So this is pretty decent separation in the ability to predict high versus low outcomes. As I mentioned, though, we know ahead of time which predictions are more reliable, which ones we can trust more. So the middle column there, those are the outcomes for the predictions that were the 50% most reliable predictions known in advance. And here you see even greater separation. And then if we go all the way over to the right, those are the ones that were the 20% Percent most reliable and greater separation still. Okay, so there are two takeaways here. One is that this successfully predicted high GDP growth from low GDP growth. And it successfully distinguished predictions that you should have more confidence than from those that you should have less confidence in.

Speaker 3: So this is a so what I want to do now is just get into the transparency and how this transparency can be very useful in terms of how you interpret and trust the results. So on this chart here, we're just taking a small segment of the entire history of predictions that we made. Because as you'll see in a minute, it's hard to see much when you put them all there in one slide. So this is just a period from 2020, the end of 2021 through the most recent quarter. So you see those blue dots are the nowcasts from relevance based prediction. The line is the actual final revised value and the gray bars are the fit. Right. So you see that there are lots of blue dots because we're making lots of predictions along the way. And then what you see is that they tend to slope up, which means that, you know, as you're as you're accumulating more and more information, your prediction is becoming more reliable. Um, and again, you know, you have this information about the extent to which you should trust the, the nowcast that you're making. So this is the full sample of predictions. Um, so we in the results that I showed you, we excluded the, um, two Covid years. And you can see there. You can see where they are.

Speaker3: You can see how it you know, the predictions and the outcomes were super extreme. Um, but because this approach is so transparent and because it tells you ahead of time whether or not you can trust your prediction. So what happened here is we were able to predict the Covid collapse pretty reliably, right? You can see the nowcast go right down to that line. But we did not predict the Covid rebound very reliably, reliably. And I'll show you why. Toward the end. But one thing that you can see right here is that the average fit was 1.24 for the nowcast of the collapse, but only 0.63 for the nowcast of the rebound. So we would have known not to trust are now cast as much for the rebound, which you would not be able to do with. With a model, you wouldn't have that information. As I mentioned, we can also see which variables were more important, contributed more to the reliability of the predictions. And what you'll notice here is that on average, the releases, the components of GDP on average were not that important. And the reason for that is because there's a lot of missing information in those variables. Right. You only had it, you know, like the third release, you only got one of those toward the end. So on average they don't show up as being important. But you can see the range for some for some of the nowcast they are super important.

Speaker 3: So I want to show you just two particular periods. One is the financial crisis and the other is, um, Covid. So this is the nowcast of GDP growth for the global financial crisis. So what you see there in the top left panel is the actual nowcast, along with the fit of each of those nowcast. And you can see that prior to the collapse there was not much, you know, we didn't have much confidence in the nowcast. But once it collapsed, um, we had very, very high confidence. Below what you see is the importance of housing permits, right? This was a housing generated crisis. And what you can see is these are the three releases of housing permits. And you can see how those releases became more and more important to forecasting the collapse of GDP during the global financial crisis. And over to the right, what you see are the, um, observation weights. This is the importance of those prior periods on predicting what happened during the global financial crisis. And one thing you'll notice is before, if you go back into the 90s, um, those periods were not they had no weight in the nowcast. And that's because there were a lot of the information for some of those variables did not exist in the 90s. So it's just this is just a way of showing you how much transparency there is in this approach. This is the same information, but for the, um, the current.

Speaker 3: I said I was going to show you the, um, Covid, but instead what I meant to say is I'm going to show you the current quarter or the most recent quarter that we have in Nowcast for. And this is actually pretty insightful as well. So our. Nowcast is for quarter one I think is 0.99, right. Annualized nominal, which is a little under 4% annualized. And then if you take out inflation, which is also about 4%, it's about zero. So it's about flat, which is fairly close to the first release of GDP for the first quarter. We don't know yet what the final release will say. But I think what's really interesting here is, you know what's driving this trade, right? This is showing you the importance of trade as information is being released. And you see, it becomes really, really important toward the end when, you know, all of this craziness has been going on. But there are a couple of things to look at. Look at the fit. Right. So here's our now we have no confidence in this nowcast because I know I'd probably say things I shouldn't say, but you know why? I mean, there's a lot of chaos going on right now around trade, right? So, you know, this shows this is telling you. Yeah. You know, you can get a forecast and you can see that it's being driven or nowcast.

Speaker 3: You see that it's being driven by trade. But you also are warned. Don't trust this. Okay. This shows you the importance of the predictive variables again for quarter one 2025, um, relative to their historical importance. Look at the release one and two of trade, right? They are outliers compared to history. So again, this is just evidence of the degree of transparency that you have with this, um, prediction technology. So let me let me just summarize. So in Nowcast, as I mentioned, it's a prediction of an outcome that's in process whose final value has not yet been determined, but for which useful information is being released along the way. We applied RPP to forecast of GDP growth from variables that are not all components of GDP, and without any insight into the composition of GDP P. And aside from the Covid rebound, this approach did a really good job of nowcasting high and low GDP and even more reliably for those high conviction nowcasts. So why did it fail for the Covid rebound? Well, the most important predictive variable for that rebound was prior quarter GDP growth. So you can see why that wouldn't work right. Because it was super negative. And then we had this rebound. But here's the thing. Well that's what Biden used to say right. Here's the thing. Forget sorry. That's a scary sign. Oh dear. That is scary. Now, what was I going to say? No.

Speaker 4: It's not getting better.

Speaker3: So you would have seen that this this nowcast was being driven by the prior quarters GDP growth, and you would have known to overrule that. So you wouldn't be able to see that in a model, and you wouldn't know whether whether you should trust that nowcast or not, but because you have this insight into which variables are driving the nowcast, you have the opportunity to impose your judgment. So there's that. Um. Then, as I showed you earlier, we had this advance warning that the nowcast was not reliable. But the explanation that I like the most is that the Bureau of Economic Analysis collected hardly any data at all to come up with that estimate of GDP growth. Li told me not to trust them. Right. So think about it. You have a bunch of government bureaucrats pretending to work from home. Right. Are you going to trust them, or are you going to trust Dave and Will? Who would you trust more? Right. So it could be the case that our nowcast is actually a better estimate of what really happened that period than what was reported to have happened. So, um, anyway, so RVP identified predictive variables that contributed most to the reliability of each nowcast it identified housing starts as the most important variable for the global financial crisis, and trade is the most important variable for Q1 25. So I will stop there and I'm happy to answer any questions. Yes, sir.

Speaker 5: Thank you Mark. Very thoughtful as usual. Um, as you know, I'm a big fan of this work, so I have a lot of questions. I'll ask two, and then maybe if you have time, I'll ask one later. Have you thought about using the prediction error as an x variable? Like thinking of in the regression space. Your residual can be an AR one process, right?

Speaker 3: Yeah.

Speaker 5: And the second about the grid search that you're doing, you used to choose the best one. Now you're taking the weighted average. Is there a middle point. Like for example if I was doing it manually for a for an estimate for a forecast. And one variable would always fit every time in any time I would just drop that variable. So is there a in between using the whole sample and just one variable?

Speaker 3: Yeah, both are very good questions. So your first question reminds me of Garch right where you're reintroducing the errors to improve the prediction. We we have not considered doing that I don't think. But you know it's it's a it's a very good suggestion. So thank you for that. Regarding how we search the grid. So. If we had. So I just showed you six predictive variables, and I didn't show you an exhaustive list of the combinations. Right. If you had 20 predictive variables, there's more than a million combinations of those 20 variables. And then if you're considering, um, ten subsamples of observations, now you're up to 10 million, right. And then we also filter not only on relevance, but we filter on similarity by itself as well. So now we're at 20 million. So here's a secret. We don't do an exhaustive search with 20 million cells. Right, Dave. And one of our colleagues at Wyndham shall have come up with some super clever engineering to sample the grid intelligently. Knowing certain cells, you always want to include certain cells. That doesn't matter. So they have an algorithm for doing just what you said. The middle ground come up with a, um, you know, a sample of the of the full grid to form the prediction. So we do that. But we that's one thing that's, uh, that's a house secret. How we do that.

Speaker 2: Mark, I have a question on the iPad. I have a number, actually. Um, could this be used to predict inflation by combining price stats with other variables? So, for example, Alberto said they don't include car prices in price stats, but could you use this method to use price stats and other variables to predict inflation?

Speaker 3: Yes, and I predict that our prediction would be better.

Speaker 2: So would you please do that?

Speaker 3: We will. I'll talk to Alberto. Said he was going to come here, I don't see. Oh, there he is. Okay. We'll chat.

Speaker 2: You chat. We need that. We'd like that. Um, anyone else in the audience for the question? Okay, I've got a couple more here I can go for. Um, how does this relate or complement your recession likelihood index? Because you're predicting GDP. Is there any way you can use this in your recession likelihood?

Speaker 3: There are close cousins there. Close cousins. They both rely on the Mahalanobis distance. But there there are differences too.

Speaker 2: Um, how does the Nowcast compare with analysts consensus? Have you looked at that?

Speaker 3: I don't know who. I don't know what their consensus is. Do you?

Speaker 2: You can get it on Bloomberg. Yeah. So we can compare that if you haven't done it.

Speaker 3: Yeah. We so we did want to compare this to GDP now which is produced by the Atlanta Fed. Yeah. Um these results are not directly comparable because we're now casting nominal GDP growth. And the Atlanta Fed is now casting real GDP growth, but we are in process of collecting the data that we need to do something that would be directly comparable to what the Atlanta Fed does. So that's that's coming soon.

Speaker 2: Okay. Um, anyone in the audience? Okay. I've got one last one here. We can use them before Aaron shouts at me. Um, can you use this idea of regimes and similar times to predict the missing variables?

Speaker 3: So that's a that's a very good question. So I said that I implied that you either have to get rid of the observation with missing information, or you have to get rid of the predictive variables with missing information. But you could use techniques like maximum likelihood estimation, right, to impute that missing information. We discourage that. Um, because, if you have a really strong linear relationship, very simple, right? Then ML probably would do an okay job. But if you're dealing with a very complex data set where the relationships are, um, you know, they're changing across the sample, it becomes harder and harder to impute values for the missing information so that those values become more and more tenuous. And what's worse, or maybe not worse, but what compounds that problem is the model doesn't know the difference between data that you actually have versus data that you're making up. And it places as much importance on the data that you're making up as the data you have. So we think assigning a relevance weight of zero is far superior.

Speaker 2: We have loads of questions, but sadly we are out of time. Mark, thank you very much.

Speaker 3: Thanks for that introduction, Michael. It was great. Invite you every time to give. This should put it on YouTube I love it. I'll send it to my kids. Maybe I'll get some some likes. How's everyone doing? Good. You're good. So far? It's been a nice conference. Everyone hyped up here. Come on, give me some. Like some love. It's not easy, you know. It's real. Every time they put me in this 330, 4:00. You know, you've seen a few speakers. You're kind of tired. It's not easy. It's not easy. But thanks for sticking around. I think earlier, Michael, you said that what we do is insight and measurement. There was a third thing. I forget what it was. Application. I'm not sure where we stand at media stats with that, but I think we do part of everything. But today our focus on measurement and then insight. The big I guess the big picture question of what we do. Is the following. We in general we think narratives drive the market. We don't think so much that it's a momentum factor or a book to market factor or a size factor. We think narratives are what's important. Okay. If you've been paying attention since the election, I think that's not very difficult to convince people that that's what's happening. Okay. That narratives are very dominating. And I think that's what drives the market. I don't think necessarily that's new because Bob Shiller came and talked in this conference.

Speaker 3: Maybe I think it was four and a half years ago and Covid September 20th. There was we had a conference. It was online. Bob Shaw talked about this narrative economics. The innovation here is how do we actually bring this idea of narrative economics into practice? Our solution to quantifying narratives is by using the media. So what we've done is we engaged in this big data project We've been doing it for more than ten years with State Street Associates. What we've done is we've created a machine that goes online and collects media sources from digital media every day. Right now, it's about 150,000 sources. It's 5 to 7 million articles every week. That's even without social. I'm going to add social media component later today. Okay. And so we're able to understand what people are talking about, what they're reading, what they're thinking, what they're talking about. The solution to understanding what the narratives are is by coding and looking at every article. What is it talking about? Think about a narrative trade war. We can look at the words, or we can use the LMS to understand if an article is talking about trade war. Then we can look at all the articles we have in a given day. We can code. We can quantify how many articles talk about trade war. And then we can look at the ratio. And that ratio is going to tell us how many articles we're talking about trade war.

Speaker 3: And we can quantify it every day. And look at that ratio how it moves over time. The last five years, you know, of course, the the topic that was the most discussed in any given point of time was the pandemic, Covid 19. Up to 75% of all articles were talking about Covid. The next hot topic since then was the trade war. If you look at recently with our data, a little bit over 50% of articles were talking about trade war. There's no other topic that was so highly discussed as these two topics that I just mentioned. Okay. And we can look every week. We can look what's being mentioned, what's hot the last week, I can tell you the two the top two credit rating is number one. And number two was nuclear energy. Yeah, these are narratives again. That's the cool thing about it. It's not like priced assets. How do you bring this to pricing? I'm going to show you a little bit about that. So that's generally what we what we're about today I'm going to focus on central banks. I'm going to try to understand what's the narrative. What's the tone of central banks using this media information okay. So I talked to you a little bit about the data. A few things that are important in this respect. One is that it's a point in time data. So we don't go and backfill. This is every day we're able to understand what the market is discussing okay.

Speaker 3: This is not anything backfilled. Turns out this is huge hugely important. I'm just telling you if you try to do go and kind of look at a firm that you're interested in, look at all the data that you can go online and Google it and figure out the data. You can get news about that next week. Try to do the same thing. On today's date, 50% of the information is going to be gone, you're not going to be able to find it. Therefore, it's very important to get all that information. Point in time. There's other issues. When you talk about media, there's always issues with bias. Okay. Is that surprising to anyone? The media could be biased. Media is not biased. No, no. It's not that general press is always positive, right? That's not the case. Typically they're negative, right. So there's some kind of bias there. If you look at it from like Microsoft I've lived. You didn't mention but I've lived three years in Seattle. No one says it's not allowed. I don't think there's a written law, but you can't say anything negative about Microsoft when you live in Seattle. That's just the reality. If you want to know something more impartial, go to the New York Times. So there's geographical bias. There's actually a size of article bias. Long articles they convert to neutral sentiment. Because you have to say, on the one hand, this.

Speaker 3: On the other hand that we look at sentiment analysis. There's nothing there you can't really understand. It's neutral. If you want to know something, you look at short articles, short articles to the point this was a good talk or bad talk. That's it. You can understand very quickly. So we try to control for these biases by organizing information into reservoirs. So you have several reservoirs of information. So you have all the articles talk about general issues, all the articles about corporates, all the articles about effects, all the cross talk about politics. In this particular application. Today I'm going to use the reservoir. We find that the reservoir for macro stuff is very important. It turns out to be leading other types of reservoirs. Why? Because we think that. I think it makes sense, because I think that traders and FX traders around the world are always on top of things, and they were the first one to report that there was a pandemic back in mid-January 2020. So when it comes to central bank information, we find that it's very important. So I'm going to focus on the FX reservoir for the applications that I'm going to show you. So the point of the day is going to be central bank communication. Now why what's the issue with central bank communication. I think there's at least two issues. One opacity. What do I mean by that? Central bank I mean bank governors know that they're being looked at all the time to try and understand.

Speaker 3: The market is very sensitive to what they say. So when you look at official transcripts, they end up being meaningless. Why? Because they're so careful about choosing every word that they're going to use. They kind of go over it again and again and again to try to see, oh, did we use this word before in the previous time? Well, you know, it's I don't want to use this word. It's going to sound too strong. Whatever. So at the end of the day it becomes almost irrelevant. It. That's one issue. Second is they meet only several times a year. Right. So if they're going to meet every six weeks, in normal times, what happens in the middle? The market is always thirsty for information. So what happens? You're going to wait six weeks. You're not going to hear anything. So the our solution again is looking at the media. The media is there every day. They talk about these issues all the time. So if we're clever enough we can come up with a measure, the daily measure, that at any given point is going to understand the pulse, figure out the pulse of what? How the media is interpreting, whether the fed is going to be more hawkish or more dovish. We published a few papers and I mentioned it here. A few folks, colleagues and friends from State Street Associates. And I'm going to present you some of the results from those papers.

Speaker 3: So to give you a little bit idea of the numbers here we're talking about. At this point, it's almost about a million. If you look at, up until today, a million data points of articles that have talked about the fed over the last ten years, and we're going to devise the following measure. We're going to create a measure which we called hawkish dovish. How do we do this. We go every day get the data in media articles. We look at media articles. We're trying to figure out there's different ways of doing it. Let's just start with top bottom. We'll look at articles to talk about the fed. Among the articles talk about the fed we look at the words around it plus or -5 to 7 words around fed. And we try and understand are they're talking about interest rate increases or interest rate decreases. So we take all these articles that are talked to talk about the fed. We take the number of articles every day. Number of articles talk about But interest rate increases the number of articles. Talk about increased interest rate decreases. This minus that divided by the sum is our measure of daily hawkish dovish indicator. What I'm going to show you here is I'm going to use a seven day moving average. So basically I'm going to do like a Friday to Friday every every week I'm going to look at the average of the hawkish dovish over the last week.

Speaker 3: And then I'm going to try to show whether it can predict bond deals or change in bond yields the next week and the week after that, the week after that. We can also measure it bottom up. What does that mean? I'm going to actually look instead of looking at the fed I'm going to actually look at the members. I'm going to look at Jerome Powell. I'm going to look at the members of the board of governors. I'm going to look at the regional central banks. Because data could be sparse for individuals. I'm going to group them. So I can look at overall hawkish dovishness of, let's say, the voting members because it's a rotation thing, right. So you can look at the voting members, you can look at the the chairman of the fed and what you see here. If you guys can see in your monitors I plotted the time series, I think the last three months or last six months of the hawkish dovish indicator. And what you see here on, on the um, on the left is, um, the chairman of the fed versus the voting members. And in the center you have the ECB and then you have the Bank of England. So if you focus on this one here, I think it's pretty interesting that the peak here, the top is, was uh, Liberation Day. So this is April 2nd.

Speaker 3: Okay. And after that, actually they're becoming much more dovish. This was updated May 4th or last month. Since then they have turned more more hawkish. So both of them since then have turned more hawkish. I want to mention a couple of, um, State Street articles that I found really useful. I kind of copied and pasted from PDF. But I think it's really cool graphs, because one thing you can ask me, it's kind of a side note, but is the market really paying attention to fed communication? How can I measure that? I can look at S&P five or I can look at here it's DXY and I look at I can look at changes in that interest in that um changes in these indices. And I can try to correlate with the amount of intensity of discussion that the media has about the fed. And what I like a lot. And this left side graph is what you can see here. The orange line is just the amount of intensity, how much news there is about the fed. The blue line here is the R square. I'm going to use the word regression. It's okay. Use regression. Regression is okay. Still fair game. Regression of the index. Here's DXY on the changes in the amount of intensity of discussion about the fed. And what you can see is sometimes the market or the DXY is not affected by fed discussion. Sometimes it is.

Speaker 3: So sometimes the market is narrative driven driven versus not driven. Narrative driven driven. And recently it has been recently DXY has been moving because of um because or in conjunction with discussion on the fed. So I thought that's kind of interesting. The other one, you look at the right hand side, It's measuring voter. It's measuring among the voting members. The disagreement in their in their hawkish dovish indicator. So it can measure hawkish dovish of each member. And I can look at the standard deviation of that. And what you can see is that standard deviation is going up over time. So it seems there's more disagreement. Okay. The media when you look at the media it seems that it's projecting that there's more disagreement among fed members. I think that's pretty cool. Okay. Here I'm just zoning in on on on the fed. And as I mentioned, we measure um, hawkish per per person. And we kind of aggregated it. This is very similar to what I showed you before. Let me show you some empirical results. So I'm going to do two types of tests. I'm going to show you time series test. I'm going to show you cross-sectional test. For the time series test I'm going to focus here on the fed. And I'm going to look at Friday to Friday hawkish dovish indicator. So it's weekly Us Weekly frequency. And we're going to look at the change in hawkish dovish indicator. And I'm going to run a regression of the changes in the yield on the hawkish dovish indicator.

Speaker 3: Starting for from one one week after two weeks three weeks four weeks etc.. The results you see here, we did it across the yield curve. And you can see this panel and this panel you can see um the following. When you look at the T plus one the next week you can see the coefficient. You guys can can you see the in brackets the 3.4 t statistic. Once the can you guys see 3.4. What is the 3.4 signify. It means that when the indicator is more hawkish than the next week, interest rates are going to increase by 3.46, and the week after that is going to be by 3.6, the week after that, 3.48. So it means that in a week where the media is saying the fed is more hawkish and the next the following weeks interest rates go up or yields bond yields go up. When you look at this bottom one here I really like this result. The reason is you see if we add a dummy for the FOMC meeting weeks. So let's think about this. Every six weeks they meet we add a dummy variable to indicate whether that week the FOMC met. And what you can see is that that coefficient is almost a similar magnitude but negative. What does that mean? It means that there's predictability, but that happens outside of the meetings.

Speaker 3: Fomc meetings. So during the so the weeks that there's meetings, there's a lot of attention to what the fed is saying. And I think the the movement is more contemporaneous. The predictability comes when there's under-reaction. So in between the meetings there's still information in the media. It seems it seems that people underreact to hence the predictability. We've done similar results for ECB and and BOE. So this is a time series result. We've done a few things with tweaking the measures. So I explain to you a little bit the measure that we call Gen one, which is looking at the words of interest rate increases versus decreases. But now we have you know, we have LMS where I have large language models that can be helpful. And so we in this study here, what we did is we reran our results by using Roberta ChatGPT llama two. And overall you can see that the using LMS improve the results. So we actually produce now a new set of indicators on top of the original one. We produce another one based on ChatGPT the two tests we did, by the way, to understand whether LMS can help are twofold. One is we did the relevancy test. Two is we did a yield prediction. Relevancy test is about contemporaneously when the media is more hawkish, our yields going up. That's contemporaneously. We expect that if the market is efficient then we expect some some discussion to be immediately included into important into asset prices.

Speaker 3: Surprises. The prediction part is when you look at t plus one, t plus two, T plus three, etc.. So the graph on the top is the contemporaneous and the graph on the bottom is the prediction. So you can see it doesn't matter if you use the LMS you get a little bit better. All of them seem to be pointing in the right direction in the same direction. So that's just to show you that we've also used MLPs. We also use LMS. And both of these type of indicators are provided to you through State Street. If you're interested in understanding a little bit more about the overall volatility of yield, what we find is that the top bottom discussion about the fed explains about 50% of the R squared, and then specific discussion of ronpaul versus the variation among the different members account for the the other 25% of each. Okay, so all of them kind of matter for discussion in general. Jerome Powell specifically, and the dispersion or disagreement that I mentioned before among the members, all of them contribute to understanding changes in yields. Okay. So that was time series. What I'm going to show you here is cross-sectional. So what we've done here is we looked at. Central bank tone this hawkish dovish indicator across 15 different central banks. And what we're doing here is we're comparing the changes, the weekly changes in yields of the bonds respective to each of the central banks.

Speaker 3: We're trying to see whether when the media is more hawkish about a central bank, does that predict more yields for that country in the next week and the week after that and the week after that? So these are results of cross-sectional regressions Fama. Macbeth. Regressions of T plus one. Changes in yield on time t. Hawkish dovish central bank. We do a cross-sectional regression with 15 observations. And we kind of look at we do it every week and we sum the coefficients. And that gives you an idea of how a return would look like with the top one is with the gen one. The bottom one is with the with ChatGPT. So this shows you that even in the cross section of of different countries, hawkish dovish indicator seems to be working okay. Finally about the central bank indicators. Um, a client request, client request, client request from Asia asked us a little bit about central bank intervention. Now intervention when central banks intervene in the currency they don't go and say, hey, we're intervening. You have to understand that a little bit from you. Try to understand it for the media. You're trying to understand it from variations in the in the rates. So what we've done is we we looked at the monthly level, we coded this and we tried to see when does the media think that there is some intervention. And so we we created the measure of central bank intervention.

Speaker 3: We calculated this for 17 central banks. And what you can see on this chart is the contemporaneous relation between changes in volatility of the currency and the indicator of whether there was a significant intervention from the central bank. So you see that that's a positive relation. So when there's an intervention during the month of intervention there seems to be more volatility. What happens after that. What I think is very interesting is that we look at the top graph. You see there is contemporaneous correlation in terms of volatility at time t, but after that volatility seems to drop. So that seems to suggest that the intervention helped. We see that there's more volatility happening before they intervene and then volatility drops after it. That's that top graph. The bottom graph does everything in event time and kind of looks at percent changes. So that's just showing you that when you look at more like emerging markets it seems that that's much more of a drop than developed markets. I think that makes sense. Okay. So that was a little bit about central bank central banks. It was really using traditional media, digital media, but traditional media in the remaining of my time. I want to talk to you about social media. Who uses social media here? Amazing. Am I the only one not using social media? Geez, I'm so outdated. Who uses Reddit? You're so cool. You're too cool for me.

Speaker 3: Amazing. Well, I didn't really know about Reddit until, I don't know, 3 or 4 years ago. If you want to use Reddit to do sentiment analysis, it's not easy. Why? Because no one writes anything. There are no words. You guys know what I'm talking about. It's all emojis. Thumbs up, thumbs down. Rockets, moon yolos, all that stuff flying around. There's no words. So what do you need to do? You need to count the rockets. Yeah. Yeah. So we actually created these flat files. When you look at stocks, individual firms, we actually have a product. You can get it through State Street, where every day you have the number of rockets, the number of yolos, the number of moons, and every single company. Russell three. And we're using it, by the way, as a short squeeze indicator. So when you have a firm that has high short interest, if it's also if there's also a lot of discussion on social media, then then it might be subject to squeezes. But how we're going to use Reddit for the purposes of macro. What we're going to do is we're going to look at instances where people talk about, oh, I'm going to buy this, I'm going to buy that. So we're going to try to understand social. We're going to try to understand household spending. How does the fed deal with this. They have surveys right. A few hundred households and ask them questions.

Speaker 3: We're going to use Reddit, and we're going to ask 21 million people how they feel about the economy. Look at this. Don't you love these bullet points here? That's great. Social bullet points. So what's the problem with the current surveys? The problem with the surveys. Limited number of households. There's all these questionnaire biases. I've never I've never done one of these questionnaires. So there's a bias there. And it's only released every four every four months. So similar to the fed communication. What happens in between. And is it enough cross-sectional. Is there enough. Are these houses representative enough to understand what's happening in the economy? If we look at social media, we can create these measures every day based on millions of users. The results, I think, are absolutely fantastic. I'm super excited about them. And you guys have the benefit because I've presented it already a few times before. Some State Street representatives here have heard me and I try to be a little bit more. But, you know, I get it that they've heard me so many times don't hear me anymore. So therefore I added more results because after I was in Toronto a couple of weeks ago, I got some questions. I'm going to show you even more results. And again, they're all, I think, super exciting. What do we do here? We're going to look at personal finance subreddit. Right now it's about 21 million users 24 million posts. And we're going to look at the measure we're going to look at is the number of times did you see increase in spending versus decrease in spending.

Speaker 3: Very similar to what we did with the the fed hawkish Dolphins. Look at the top the top graph here. You see it's much more volatile because it's every day right. We have a measure and that measures the propensity of of Redditors to consume to spend. When it goes up. It means there's more discussion on I just bought this about about that. And when it goes down is I can't buy this. I didn't buy that. And I plotted it against the New York Fed. Um, which one I think is here, the New York Fed. There's two there's experienced and expected spending. The one above there is uh, is experienced. So you can see what I say again, I'm a little bit too close to this. I think there's some predictability here. That when you look at the social discussion, it comes before The reported discussion by, um, by the fed, uh, the New York Fed, uh, surveys, of course, the New York Fed surveys. It's a it's a step function here because we're kind of keeping it flat for four months every time. Right. And you see, in between the social measures, I'm going to show you in the next slide some regression results to verify that there's some predictability. This graph here um, it's a super graph. But but it's a little bit disturbing, at least to me.

Speaker 3: I have three young kids and I'm looking I'm looking at this graph and basically what this is saying this is measuring financial situation. So people are just commenting on what's their financial situation, how they feel about the economy, how they feel about their finances. That graph over the last decade has just been going down. It's just been going down. When you think about who's reading and who's contributing, I think what this is signaling that, you know, a lot of our younger generation doesn't feel so good about the economy. So I think it's spectacular that you can get that information. But on the other hand, I'm not sure that I really like what it suggests. Okay. Some regression results. The top panel contemporaneous regression. So this is the MKT media static sentiment over 120 days. Because we need to work in this frequency of the New York Fed, um, surveys. So when you look at over a four month period and again here, the number of observations is only like 30, right? So it's like a decade contemporaneous correlation. The in parentheses. It's the t statistic. You can see that it's highly correlated contemporaneously with both both New York Fed experiences and expectations. We look at the bottom graph. This is about predictability. So this is time T plus one. This is a four month interval T plus one versus t. You can see that mkt the social measure has quite significant predictability certainly for experiences not so much for expectations but for experiences.

Speaker 3: Okay so one question I received last time I presented this was well, maybe you can tell us a little bit about the demographics because look social media, maybe it's just a particular part of the of the population, but it doesn't include everyone. So what can you tell us about where is it more important? So fortunately when you look at these surveys they bifurcate by different groups. So you can have an age different age groups and you have different regions and education and income numeracy. I think it's super interesting. What do you see here. The again in parentheses is the t statistic. So when you look at this predictability of the social measure trying to predict the survey of household spending, it seems to be more predictive for the younger age group. Sensible to me. Younger age group. High earners. That to me stands out. High earners. Young high earners. That's what this is capturing. It's not that it's not capturing other things but that really and also educated. So I think that's pretty interesting. It really opens up and explains to you what this is trying. You know what this is capturing which segment of the population high numeracy, high income, educated young. In this one. It shows it from the Midwest. Not sure how to think about it. So I'm going to flip to Michigan's survey. There when you look at the region, it's actually it's actually correlated with all of them.

Speaker 3: So a little bit I guess more comfortable with that. But it points to the same idea. The highest the highest correlation is with the low age group and high income. So it's virtually when you look at both Michigan Survey and New York Fed Survey, and it's the same social media measure that I mentioned, both of them predicting these two surveys, highly correlated top earners, um, young top earners. Okay. Finally, um, just to give you a taste of what we're doing, we're trying to do this all over the world. So. Because I'm flying to Europe next week, I also included here the ECB. And again different places have different benchmarks that we can use. Ecb actually publishes something um publishes uh what is this consumer expectation survey once a month. And I then run correlations but it seems to be related seems to be related. Michael, you asked me what it looks like. It seems that the spending you see at the end there seems that going up goes up a little bit. Okay. So more on that next time I see you. Okay. So we're doing this with Asia. We're doing it for a bunch of countries okay. So this is what we're this is what we're doing. We think this is a big question I guess measurement. But it could be very insightful okay. So we're going to continue to push on that.

Speaker 3: If you're interested. Give us the feedback. Because we want to do things that we think that you also think it's important. So if you think this is important, we certainly think it is. We have the capability to measure it and we're going to keep doing it. So let me summarize. I hope I hope I convinced you that the media could be very helpful. Okay. It's true. There's a lot of biases. But we're here in the business of aggregation and statistics. We have a lot of information. And we think on average by averaging we're getting a lot of the noise out. So I showed you how the media, in combination with some kind of word count or LMS, could help understand a little bit more about central tone as central bank tone. I showed you why it's important both in the time series and the cross section. And then finally in the second half of the talk or third of the talk, I explained a little bit about social media, how that could be used to better understand macro. The macro economy, the environment that we're in. I showed you like we call this a POC for us Europe and Canada. We also did Canada. I didn't show it. But uh, Michael Guidi I think last couple of weeks ago showed those results. And the other countries we're working, we are working on are listed right there. Okay. Thank you very much for your attention. I'm happy to take any questions.

Speaker 4: Okay. There you go.

Speaker 5: Thanks very much for that. Um, the internet can be a bit of a sarcastic place. I'm curious how you think about sarcasm, especially on Reddit, and how that feeds into your your process.

Speaker 3: Look. You can also tell me that the internet could be a place where there's a lot of fake news. What I'm trying to do. Okay, I'm trying to measure what people are saying. Okay. The. The way I think about, um, you know, is it useful? What I'm trying to do is correlate that contemporaneously or lead lag, and try to see whether this has something to say about the economy. So it could have been that, yes, maybe sarcasm or fake news or any other biases that would work against me finding anything. But the result actually is not the case. So it does seem to be informative. Now just a little bit about I mentioned fake news. It's related to your comment. It could be that what's said on social media is wrong. It could be the what said in the public press is is wrong. But if people react to it, okay, I want you to be able to know that. So some narratives drive the market even if they're not right. But if enough people talk about it. So that's again, if there's one or more or two people that are, you know, either sarcasm as you mentioned, but, uh, bias or fake news in the aggregate, I hope it's going to wash out. But if there's a narrative that everyone is talking about, it might be a false narrative, but everyone is talking about it and it affects the market, I want to be able to measure it. So that's the point. You can make an assumption or whether or not you think that that narrative will have a temporary effect, and then you can take the other side of the trade. But my job, I think, is just to try to help you understand what in aggregate, people are talking about. Thank you for the question.

Speaker 1: So, Ronnie, I have one on here. Um, have you looked at the persistence of mean reversion in your hawkish, dovish indicators?

Speaker 3: Have you looked at what.

Speaker 1: The persistence or mean reversion or persistence versus mean reversion?

Speaker 3: Um.

Speaker 1: And you have a minute.

Speaker 3: Um, we haven't looked at it specifically. I mean, we looked at, uh, whether these variables are persistent, there's some persistence, but we try to take out the persistence by looking at changes. So to the extent there's some persistence, because I worry about doing time series predictability regressions with persistent variables. So we try to take care of it that way.

Speaker 1:  You did it. You did that in under a minute. The time for one more question. There was the one at the back, please.

Speaker 6: Hi. Were you able to figure out if, um, you know, people are using ETFs to confirm the opinions that you're extracting from this analysis.

Speaker 3: Whether people using ETFs.

Speaker 6: Or, you know, if they're able to actually.

Speaker 3: Here's here's my answer to you. Um, I think let me just. Rephrase the question. Our asset prices reacting to the narrative. The answer is definitely okay. Definitely. When you look at we did this following analysis. We have hundreds of narratives coded in our system. And we looked at take a three year period. We did it on a rolling basis three year period. And look at the univariate regressions as a regression. Again, a regression of spy on each of the narratives and then pick the top five, top five R-squared univariate, then run a regression of the spy on the top five univariate. Narratives. You will get an R, sometimes R square of about 50%. Okay, so that has variation. So sometimes I think the market is more is narrative driven and sometimes it's less narrative driven. Okay. But definitely when you look on average we're talking about significant sizable R squares. So I definitely think and the results can can prove it. And we've written an academic paper on that that the asset prices even in the macro seems to be they seem to be driven by discussion and narratives. Yeah. Thanks for.

Speaker 1: The question. And I would just add, actually, if you haven't downloaded the app, you can see all of that on the narrative map, which proves exactly what you're saying. Thank you. Now unfortunately we are now at a time, but thanks very much for your questions. So thank you very much.

Tenreyro: Thank you very much, Michael. That was a very generous introduction. Thank you for having me. It's great to be here in this conference. There is a lot going on at the moment in economics and in politics. In my presentation today, I will try to focus on the big picture, where we are coming from, where we are now, and where we might be heading. In doing so, I will be drawing comparisons between the US and Europe, including the UK. So starting from where we are coming from, we've had two colossal once-in-a-generation shocks, the pandemic and the Russian invasion of Ukraine, which led to an extraordinary increase in the prices of energy and other commodities, all in the space of less than three years. Adding to Brexit, this is the most remarkable concentration of tail events, or rare events, as we call them in the literature. To give a sense of magnitude, the energy shock alone was comparable, if not larger, than the oil shocks in the 1970s. This is illustrated in this chart, which shows the growth rate of the price of crude oil and European natural gas since the 1970s. This makes clear that the 2020 shock was indeed larger than the oil shock of the 1970s. This picture is about wholesale energy prices. Let me show you what happened to retail energy prices, what consumers and companies actually pay. This is a component that actually enters in the CPI basket. So we can see here that despite this being a global shock, it had different effects across different jurisdictions. So if we take the Euro area in orange here, we can see that the cumulative increase in retail energy prices reached a peak of 80 per cent, compared to a peak of just below 40 per cent in the US. We can also see a gap of 20 percentage points opening up off-peak. If you think that this is bad enough for the Euro area, for the UK, it was much worse. The cumulative increase in retail energy prices reached a peak of 115 per cent. That was much more persistent than in either the US or the Euro area. So 115 per cent, that's 80 percentage points higher than in the US, 35 percentage points higher than in the Euro area. These are huge differences, and this represented an enormous shock for the UK economy. Now, crucially for the Euro area and for the UK, this increase in energy prices will set negative terms of trade shock because both the UK and the Euro area are net importers of energy and other commodities. This is very different from the US economy, which is a net exporter of energy and commodities. For the US, this represented a positive terms of trade shock, pushing up on activity and on prices. Now, negative terms of trade shocks are a problem for central banks because they pose a big trade-off. They push up on prices but down on activity, and that's the challenge. I guess we will come back to that in the next session because tariffs are a little bit like that. Instead, just to be clear, positive terms of trade shocks are just like a standard demand shock. They boost aggregate demand, the external aggregate demand of the country. In the case of the US, it had yet another boost given that it's a main exporter of defence and military equipment. So the conflict also added another boost to its external demand. Of course, this is not just a story of energy prices and commodities. Coming from the pandemic that led to a big increase in the demand for goods relative to services, which, along with supply restrictions, particularly in Asia, led to a big increase in the prices of goods. We saw that playing a big role during the pandemic and its aftermath. So one thing that was different, however, across the Atlantic was the response of fiscal policy to the pandemic. In the US the response was much more generous than in Europe. We can see that reflected in household incomes, for example. In the US, household incomes increased very early in the pandemic, going above their pre-COVID trends. This is very unusual for a recession to have household incomes going above trend and even increasing. Whereas in Europe and in the UK, we saw a decline initially in household incomes relative to trend. Only recently, we've been seeing a catch-up with their pre-COVID trends. So bottom line, these rare events had very different impacts across the Atlantic. In the US, the energy shock was a positive terms of trade shock, which, along with the large fiscal stimulus, led to a much bigger impact on aggregate demand. Whereas in the Euro area and in the UK, the shock was a negative hit to the terms of trade, making it much more difficult for central banks as they had to handle this trade-off. Now, as you know, monetary policy is an aggregate demand tool. Raising interest rates can lower inflation by slowing down current and expected aggregate demand. So when inflation is driven by aggregate demand, we face what we call the divine coincidence. By increasing interest rates, we can both restore inflation back to its target and output back to its potential. However, when inflation is driven by a supply shock or a cost-push shock as an increase in energy prices or commodity prices for a net importer of those commodities, then monetary policy faces a trade-off. Increasing interest rates to lower inflation exacerbates the weakness in activity generated by the shock. So that's the trade-off that central banks in Europe and in the UK were facing, a very different situation from the US. In the occasion, interest rates, policy interest rates increased by roughly the same magnitudes in all these jurisdictions. The dynamics for inflation were very similar across all of them. What was different, and we can see it in this chart showing patterns of consumption for the US, Europe and Euro area, and the United Kingdom, is what happened with the aggregate demand, aggregate activity. This shows a picture for consumption. You will see the same similar picture for investment. In the US, consumption picked up very quickly early in 2021, going above its pre-COVID trend. In the Euro area and the UK, in contrast, consumption is barely back now to its pre-COVID level, well below its pre-COVID trend. So this is where we see the big differences in these shocks. The dynamics for inflation expectations for longer-term inflation expectations they were fairly similar across the three jurisdictions. There was much more movement in the short end, understandably, because those were reflecting the dynamics of the actual shock, as we can see in green here, the 12 month ahead measure of inflation expectations. This is households. Whereas in blue for the three countries you can see that there was barely any movement. This shows a five-year ahead. There was much more movement when we look at financial market measures of inflation expectations. There was a big increase in inflation expectations, particularly in the Euro area, mostly a re-anchoring after many years of being well below the two per cent mark. There was a big increase also in the US and the UK. We are now seeing a convergence back and, in fact, in the case of the UK, these measures are back to their historical averages pre-financial crisis. Now, this plot here shows the dynamics of inflation during this period. We can see that for the UK, the US and Europe, inflation more or less behave in a similar way, a big surge and then a quite rapid decline. I want to stress here the differences from the 1970s. Despite the shock being larger than back then the dynamics of inflation was different in the sense that it went up and down much more quickly, and the increase was much more contained, certainly in the case of the UK. So this was a big test for the monetary policy framework, particularly for Europe and the UK. We had a 1970s shock, plus the pandemic, with a post-1990s monetary policy framework. I think one conclusion we can draw, particularly for the case of Europe and the UK, is that the framework held up reasonably well, returning inflation to target within a reasonable period, reasonable given the lags in the transmission mechanism. So take away from this, the flexible inflation targeting framework fared fairly well in this big test. Now just as a footnote here, the combination of supply shocks or cost-push shocks and lags in the transmission of monetary policy is a reason why most modern central banks have flexible rather than strict inflation targeting frameworks. The role of monetary policy in the face of this large trade-off inducing shock is to return inflation to target within a reasonable period of time, trying to minimise the losses for future potential output. So where are we now? We haven't seen the unwind of the energy shock and all its indirect effects, as well as a renormalisation of supply chains. At the same time, we've seen the effects of a restrictive monetary policy which is still in place. This has been shaping the dynamics of inflation and activity for some time. In particular, let me focus here on Europe and the UK, where the supply component was much bigger than in the US. So the unwind of the supply shocks is effectively a positive supply shock. It pushes up on activity and down on price pressures. On the other hand, restrictive monetary policy leads to lower inflationary pressures and weaker activity. So the net impact of these two forces, demand and supply, have been pushing down on inflation in these jurisdictions. On net, the two forces pushing up in different directions in this tug of war activity has ended up pretty flat or close to stagnation. We can see that summarised in this GDP chart comparing the three jurisdictions. Again, the UK is always in green, Europe in orange and the US in blue. We can see how much bigger the shock was for the UK, with Europe being in the middle. So, where are we heading? What's next? In the short term, the biggest threat to the global economic order are tariffs and trade wars and the uncertainty surrounding these policy decisions. Let me start with tariffs here. You're probably familiar with this chart, which shows the US average tariff rate going back to the 1920s. We can see in the most recent period what happened with the first Trump administration, tariffs went from an average of roughly 1.5 per cent to 3.5 per cent, and then went down during Biden to roughly 3, so a small movement down with Biden. Here I plotted three scenarios. The first one, scenario A, is one in which we have tariffs of 10 per cent on all trading partners. That would be called Scott Bessent’s dream. Second one, scenario B, is a 25 per cent tariff on the European Union, Canada, Mexico and China. The C scenario is similar to B, but the tariffs on China reach 60 per cent, as in the electoral campaign of President Trump. So we can see that the middle scenario here takes us back to levels of tariffs that we saw back in the mid-1930s, whereas the top scenario, the C scenario, gets us way beyond the peaks that we saw in 1930 with the Smoot-Hawley act. So these are much more radical changes in tariffs than what we saw before in the Trump administration. Alberto will talk about those in his presentation after me. I wanted to also emphasise that the US economy is now more open than it was back then. So the effect of these tariffs, both on the US and its trading partners, will be bigger, commensurately bigger. Now, of course, there is a lot of uncertainty surrounding these measures. We don't know where we will end up. This is picked up in indices of trade policy uncertainty. This is the one put together by Caldara and co-authors at the Fed. This is dominating the dynamics and making it much more difficult for companies to take decisions. One thing is to know that the tariffs will be higher, but not knowing exactly what they will be is paralysing decisions, particularly when it comes to investment. So we don't know where they will land, but at least we can speculate using theory about what their effects might be. I think there's consensus that they will have an adverse effect on global growth. The effect on prices, again, we'll hear more later. In principle, we should expect relatively limited effects on price inflation in the US, but I'll discuss some back-of-the-envelope calculations in a minute. The effects on other countries are, in general, ambiguous. So let me walk you through these two last points. In the US, US imports are roughly 10 per cent of its GDP; 60 per cent of the imports come from the European Union, Mexico, Canada and China. So, if we focus on my middle scenario, the B scenario, a 25 percentage point increase in tariffs on all four would lead to an increase in inflation of roughly 1.5 percentage points. That's the direct effect of the tariffs. If we were to add ten per cent on all other trading partners, that would get us to roughly two per cent on inflation. That's the direct one-off price impact assuming full pass-through. Now, in the past, we saw that pass pass-through was reduced, was much less than one for one. In part, that was because of the appreciation of the dollar, which this time is not happening. If anything, the dollar has depreciated. The other factor was exceptions. There were many exceptions to different product categories. That might happen this time, so we will have to watch. There were two other effects on margins, exporters cutting prices and importers cutting their margins, which again reduced the pass-through from tariffs to CPI inflation. There was significant trade diversion, imports from non-tariff countries increased relative to the tariff countries. Now there's reason to think that this low pass-through that we saw in the first administration will not repeat itself this time. Why is that? These tariffs are much larger and much more visible. So it's very hard for consumers and companies to ignore these 25 per cent increases. Moreover, one could argue that with the new immigration measures, the labour market might tighten more and be tighter than it was back then. So the risk of second round effects this time is much larger. So that will create in the US again, some of the dynamics that we saw in Europe and in the UK in terms of cost-push shock, that will push activity down and prices up. I would expect those effects to be bigger this time around. Now let's now pivot to countries on which these tariffs are imposed. In principle, we should see lower export prices, given this lower demand. So that should be a disinflationary force. On the other hand, with retaliation, we would see higher imported goods prices. So the effect in principle is ambiguous. It depends on what retaliatory measures these countries take. For third countries that are not directly affected by the tariffs we should see lower inflationary pressures because the combination of lower global growth and lower demand will tend to push price pressures down there on top of the excess supply from other countries that will need to place those products elsewhere. Let me zoom in on the UK. Policymakers were very concerned about many factors that could impinge on inflation. On red here you have the inflationary factors, and on green, the disinflationary factors. The first one that is always in the mind of policymakers is the dollar appreciation. Again, that is not materialising this time. The dollar is moving in a different direction. The second is tariff retaliation, but the government has done very little. One advantage of seeing the UK is that in principle, those negotiations have already happened. To the extent that those stick, the scope for retaliation has been minimal. There is a third concern. People were concerned that tariffs will create supply disruptions in the same way that we saw disruptions during the pandemic. This situation is very different because this is about bilateral tariffs between the US and different trading partners. So the only products that will be affected are those that do the round trip to the US and back to the global market. That should be a relatively small set. There are three factors here in green that should be disinflationary. The adverse impact on global growth, both from the impact of the tariffs themselves and the uncertainties surrounding them. Second, lower demand for UK exports from the direct impact of US tariffs on the UK. Third, is the excess supply of those foreign products. So all the products, say, from China, we need to find a new destination elsewhere. Those will be exerting downward pressure in the UK and other countries with limited retaliation. The final, which should be an inflationary factor, is something to consider. There were some anecdotal articles in the news pointing to some multinational companies increasing markups everywhere in order to make up for their losses in the US. That might play out. We should see that in markets in which there is very high sellers' power. We would not see them in competitive markets, but it's something that might play out and something to watch in the future. So the next question is how should central banks react to these tariffs? Again, for third countries that are not directly affected by the tariffs, there should be no trade-off. The tariffs will lead to lower activity and lower price pressures. So if anything, monetary policy would have to be a bit more accommodative relative to whatever plan there was. For the US or countries imposing retaliatory tariffs, we are in trade-off territory. These tariffs will lower activity and increase price pressure. So it's a very similar situation in which the UK and Europe have found themselves with the energy shocks. So if these price effects are small, the orthodox recommendation would be look through, ignore it and focus on the medium-term inflation prospects. However, if these price effects are more substantial and again, the administration ends up going for the high tariffs, and there is a risk of second round effects along with tight labour markets, we will need to see some tightening relative to the non-tariff scenario. So this is going to present the Fed and other central banks with retaliatory measures in their countries, with a very tough trade-off to handle. Effectively, in the case of the US, it might lead to delay loosening or perhaps not loosening, depending on how those tariffs play out. For all countries, we will see increased uncertainty. This will be counterbalancing these pressures, and because this uncertainty will not create any trade-offs, they will lead to lower activity and lower inflationary pressure. So this will be the other element that central banks will have to weigh in in their decision. Now, let me, before I move to the medium- term, let me show you a picture that is more for you to think about. This shows ten-year yields on the right for the three jurisdictions we're looking at: Euro area, UK and the US. I also plotted the policy rate. So just before the pandemic, the UK policy rate was roughly between the US, the Fed Funds Rate and the deposit facility rate in the Euro area. After the pandemic, the UK moved very closely to the US. We see that movement also in the ten-year yields. Now this might change given that these economies, again, are coming from divergent paths and are moving into different situations. Again, the impact of the tariffs in the UK will be very different than the US. So I would suggest that this pattern might not stay for too long, given how these economies are diverging. So let me talk about the medium-term factors. So in the medium-term horizon, there are four factors that will be shaping the dynamics for activity and inflation. The first one is increased defence and infrastructure spending in Europe. This should give a demand boost to the European continent with spillovers to the UK. It might have positive effects on supply capacity further down the line. There are many papers now showing that these expansions, especially military spending, can have positive impacts on the supply capacity of the private sector in the medium-term. The second factor, which I think will play a role, is lower weight of the US dollar on international portfolios. This uncertainty created by the administration will diminish the dominance of the dollar in international trade transactions and the presence or the relevance of treasuries as safe assets in international portfolios. Again, I don't think the dollar will lose its dominance, but it will be diminished to some extent. Third factor is further fragmentation. I talked about one-off tariffs coming from the US, but we might also see a movement towards more fresh shoring or home shoring or reshoring, where companies move from the most efficient supplier towards the friendly or more reliable supplier. That will create effects that are akin to a tariff increase. Same with climate change effects, they might lead to effects on productivity. Here is very important to think about the demand and supply effects. So unambiguously, these factors will lead to a weaker economy, but they will have impacts on both supply and demand, meaning that the effects on prices will be ambiguous. Here is important to think about the pace of that fragmentation. So in work I have with Ludovica Ambrosino and Jenny Chan, we show that if the pace of fragmentation is very gradual and this is anticipated by companies and households, then this fragmentation might be disinflationary because demand and investment in particular are just in advance. That might be very different from a fragmentation that happens very abruptly and leads to immediate supply effects, with demand adjusting with a lag. In that scenario, we will see much stronger inflationary pressures. The final factor that will be shaping our economies is AI. We heard quite a bit this morning. There's a lot of uncertainty about what these effects will be, the timeline. We can say with some confidence that the effects on productivity will be large, but this will again have effects on supply and demand. There is a question of how much consumption and investment will react in advance to these future productivity gains. More importantly, there will be a question on how the labour market reacts to this, what are the complementarities and substitutability capabilities in the labour market. So there will be some big challenges for governments to deal with: AI, its impact on the labour market and potential effects on inequality. This is all coming. We need a lot more research on that. I will pause here and be happy to hear your questions or comments. Thank you.

Metcalfe: Okay, you should know the drill by now. So you've got the QR code up here. We'll take hands and we'll take questions from here. I'm going to start, while you think, would you be able to reflect on a way to… At our Boston event last week, we had Kevin Walsh talking about the future of the Fed, and he had some suggested changes to their framework, but you're actually involved in their framework review. So I'm curious to hear where you think their framework review will go this year.

Okay. So I have to say I'm involved only with the tools. He invited me to discuss the tools and my paper was discussing in particular QE and forward guidance. So within that, our recommendation or conclusion was that we need to keep these tools in the toolkit of monetary policy. We don't know what shocks might hit the economy, and they might be needed, so there's no reason to eliminate them. We made some recommendations. In particular, if you think of forward guidance and how it might have constrained the Fed from taking fast decisions or faster decisions during the pandemic and its aftermath. So our recommendation is to stress test this forward guidance and think about scenarios rather than committing to a particular path for asset purchases. Try to stress test it and think about under which conditions they would like to keep that recommendation, how likely they are. So do a scenario analysis, as we do with interest rates, to think about a more robust way of doing monetary policy. So, my role was on the side of the tools and how to improve the use of the tools. So these tools are effective, can be very effective in the future. We should separate that aspect from actually the monetary policy stance that the central bank wishes to take, so we don't get involved with that. In terms of what will happen with the framework, my sense is that they might de-emphasise the part on employment that took centre stage and so level the two, inflation and employment, and put inflation first. The second aspect is the average inflation targeting. My guess, and it's just a guess, is that there will be a comeback to the standard rule of inflation target. Inflexible inflation targeting, we will be facing many supply shocks with climate change and geopolitical tensions. So you need a flexible, inflation targeting regime, but not the average targeting. That is very problematic. If you think about the average inflation targeting taken at the extreme, it means that the US right now should be causing a deep recession in order to generate this inflation. That's obviously that will be very costly. So I think there will be a return to the standard flexible inflation targeting and putting inflation first, yes, and cutting the asymmetry that they introduced with employment.

Metcalfe: So any hands on the floor?

Speaker 1: Given everything you said, are you surprised at how in sync the ECB Bank of England Fed policies have been over recent years, and given what you've been saying about tariffs and the coming shocks, would you be surprised to see quite a lot more divergence in the next couple of years, say?

Tenreyro: Well, I think the Euro area has already, the ECB has already decoupled. I think what's surprising is that I would expect to see the UK much closer to Europe than the US, and that's not happening. The US is still following one for one the movements in the Fed Funds Rate and ten-year yields are much closer to the US. That's a bit of a puzzle given where we are going. So that's the part that puzzles me. The ECB is taking already a different turn and has already taken a different turn.

Speaker 2: Actually, just on that point, how much do you think that's driven by the fiscal dominance, especially in the UK over the ECB? Do you think that might also be a risk in the next five to ten years and might make the monetary policy a bit less effective than it was over the last couple of business cycles?

Tenreyro: Yes. So I think that fiscal dominance is a bit overweight. If I think about the fundamentals of these different jurisdictions, again, the UK is much closer to the Euro area, both from where it's coming and where it's heading to, but yes, clearly there's a lot more weight attached to some fiscal. Again, I think the US is in another level on the fiscal space right now than the UK. So that closeness to me is hard to reconcile with what we are seeing and what governments are saying.

Speaker 3: Do movements in US monetary policy, and changing global financial conditions make the job of setting interest rates for the Bank of England even harder.

Tenreyro: So, US monetary policy is an important factor, of course, in defining global financial conditions, and it's something that the Monetary Policy Committee or the ECB will take into account when making its decisions. If the tightening in financial conditions globally is putting at stake the achievement of the two per cent target in the medium-term for the UK economy, obviously, the reaction will be to loosen. So it can generate the opposite reaction, again depending on what's the forecast for inflation and in the economy and how this impacts. So, in general, this will tend to tighten conditions. It's something that the MPC will take into account. Now I think what's interesting this time is that, unlike other episodes of tight financial conditions in the US, the dollar is going the wrong way. So that's a relief from the perspective of the Monetary Policy Committee, because it reduces the trade of having an appreciated dollar or a depreciated sterling. So in that sense, this time around, the conditions are more favourable from the perspective of the MPC when thinking about inflation.

Speaker 4: Staying on the same theme about the UK policy following us rather than the Eurozone, I think the big reason is obviously the stickiness of UK inflation. Have you any sense why it's proving so sticky? Given how weak growth is, why is inflation proving so difficult to get down in the UK as opposed to the Eurozone?

Tenreyro: Yes. So what's interesting is that the first to touch the line, the two per cent, was the UK back in 2024, right? So in the spring, last spring, UK inflation was at two per cent. Then it went down to 1.7% in September. So it's interesting how it went down. What we see now is a second leg that is associated to other measures that the Chancellor has taken. The government has taken minimum wage, then pricing of services, so this is an add-on. Without those, I think we will be telling a very similar story to Europe and the rest of the continent. So the hope is that these, going back to hopes, these are just one-off, and given the situation with the economy and the state of the economy, there will not be large second-round effects from this round. So that's the one to be tested in the next year. I don't think the dynamics was that different, but we just saw different measures taking place in the UK this year that have pushed inflation on another leg. Again, in principle, they should be one-off, but given what we've seen, we should be alert to second-round effects and follow that carefully.

Metcalfe: Okay. I've got one from the app. Let's do you first, and I've got one from the app.

Speaker 5: I'm sorry. Sticking to the UK theme, we obviously had payrolls data today, as confusing as ever. We'd love to hear what your assessment is of the UK labour market right now.

Tenreyro: The labour market has loosened quite a bit. Obviously, the first thing I should say is we have a problem with information right now, so that's a big limitation. I think most complementary sources of labour market measures are pointing to a big loosening relative to what we had. I think the combination of the pandemic, Brexit was really lethal to the labour market. We are over the worst of that, and we've seen vacancies declining substantially. We're now back to pre-COVID levels and even below, which is quite surprising given that it's now much cheaper to post a vacancy. So I think the labour market looks a lot looser, but we have, and this is related to the previous questions, we've had some big shocks on the price side with the minimum wage, that obviously creates extra inflation, extra relative to where the tightness of the labour market lies. So that I think is what we're seeing in the figures on the pay side. Overall, especially if you look at the pay data and vacancies etc., we've seen a big loosening. We should be back, that's my take, to what we had back in 2018, 2019. So to me, this is looking normalised right now. Just we have to go through this last round of combination of NI, National Insurance Contributions and the last leg of minimum wage.

Metcalfe: I don't know whether you can do this in 20 seconds. The question on the thing was you talked about terms of trade boost to the US, also incomes, explaining the difference. Do you get the sense in the US that they are now beginning to reverse?

Tenreyro: All that enormous strength and boost coming from all these external factors are certainly at the very least plateauing. I would expect a reversal on something much more similar to what we saw here because the tariffs are now a cost-push shock that we generate that trade off in the US, something that they have not seen, at least not in the same magnitudes as Europe or the UK during the energy price shock. That was a big boost for the US, no question. So, yes, to your question, I would expect a reversal of that. It's still being helped by, again, it's a big producer of military and defence equipment. To the extent that these wars and conflicts globally continue, this will be given the extra boost. We're not seeing new hikes in commodities or energy prices, so that should really level for the US, and it will have that extra boost.

Metcalfe: Perfect. Well, thank you for taking all the questions and a great talk. Thank you very much.

Turkington: Thank you, Michael. Every year I tell the conference organisers I have two simple requests. I need Roberto as my opening act, and I want Michael to praise me for at least 30 seconds. I think this is the first time that they both happened, so thank you both so much. It's wonderful to be here. I want to talk about interpretable AI for investment decisions. The word interpretable is really key here. Whether you like it or not, I suspect that most of you are spending more time interpreting AI-generated content these days and maybe a little less time interpreting people-generated content. Most of us are pretty good at interpreting people, but we're not necessarily that great yet at interpreting AI. That's a problem for the reason that I'll describe now. I want to give you an example first of interpretation gone wrong. In 1997, chess master Garry Kasparov famously played against IBM's supercomputer Deep Blue. In this encounter, Deep Blue went on to become the first computer to beat a grandmaster at chess. It was hailed as one of the early victories of AI. But what I want to focus on is something that occurred in the first game of the second match, when Deep Blue took its 44th move. It was stunning. It moved its rook all the way across the board for a reason that nobody could understand. Kasparov himself thought that the computer understood something he didn't. He said it was a sign of superior intelligence. Kasparov was visibly shaken by this and he went on to lose the next match. Now, 15 years later, an IBM engineer revealed what actually happened. What happened was, the computer crashed and it was programmed to default to its last resort of making a move completely at random. So, the rook move was completely random, there was zero intelligence. It is very unsettling when we can't tell the difference between genius and idiocy - but that's where we're headed with AI. As we try to get smarter about making investment decisions, we face this challenge. Is a model or an AI solution genius or idiotic? It relates largely to this issue of overfitting. I want to talk about, what really is overfitting in terms of predictive modelling? I want to argue that transparency is essential. This goes back to that theme of trust that Roberta was just talking about. We need trust in our model, so we need better techniques. I want to talk about two techniques that we've developed at State Street Associates to address this issue. AI is all about finding patterns - hopefully useful patterns. People, of course, are quite good at this. Roberto had wonderful examples of that. This is what the IQ test was designed to measure, is finding patterns. But machines have some superhuman abilities. One example that I love is, there was a team of science researchers who asked AI to find patterns in retina scans within the eye. It could reliably distinguish between male and female retinas. No doctors knew how to tell the difference, so there are superhuman abilities that can be very powerful. What can we do with them? Well, I think there are fundamentally two ways that these types of capabilities can help us. We can go faster, so we can perform our work faster, we can augment ourselves. But what's the problem with faster? It creates the risk of mistakes. If you start driving too fast, every time we go too fast, we pay less attention and there's a risk of accidents. Now, we can control this, we can mitigate it by being more careful. We can do things like not allow the AI to use certain tools; only the ones we want. We can use it only on routine, low-risk tasks. We can verify and check the answer when that's possible. These are controls to make sure that we don't crash the car when we're going faster. But what I'm much more interested in for this talk is getting smarter. Getting smarter is solving problems where we don't already know the answer. No matter how long we work on it, we might not get it, so it's fundamentally about prediction, predicting some unknown. Like Kasparov with the rook move, if we just look at a prediction we can't tell the difference between if it's really good or potentially really bad unless we understand it more deeply and earn that trust. Keep in mind that problems of investing and finance are not routine and they're not low stakes. They're fundamentally hard problems for three reasons. First, the fact that markets are competitive is a feature, not a bug. It leads to price discovery, efficiency but it also means that there's a lot of noise that prices should be hard to predict. So, there's fundamentally a lot of noise versus signal in markets. Second, markets continually adapt. If someone finds the greatest ever alpha strategy and they start trading on it and maybe someone else starts trading on it, it disappears through competition. Old data are not necessarily always relevant. We face this perpetual small data problem in finance. Third, mistakes can be disasters. Misclassifying a picture of a cat is not that bad, but a bad investment decision could wipe out years of gains in a single week. So, this means we have to use data very carefully in our field. How can the use of data go wrong and what can we do about it? I took the liberty here of updating the Goldilocks fairytale to apply to the next ten years. If you have young children, I think this is much more important than porridge and beds. You can overfit a model. You can underfit a model - but it needs to be fit just right. Of course, underfitting is a problem where you fail to identify the most important patterns in the data. Overfitting is where you see patterns that aren't really there. You mistake noise for signal. Now, people of course overfed all the time. It's not just the machines. Going back 50 years, Kahneman and Tversky, the famous psychologists, they documented that random sequences of heads and tails for coin flips don't appear random to people. We see streaks like three heads in a row or four tails in a row as evidence that there's some kind of pattern happening, but it's not really there. Maybe you've seen some of these experiments where researchers generate random stock return lines - pure random - and traders confidently point out the head and shoulders patterns and the reversals in the chart, but it's not there. Now, of course, real data will have real patterns but my point is, it's hard to distinguish the noise from the signal. It's genuinely a hard problem. So, overfitting can be subtle but it's like a model going crazy. It's giving you this complex story where the truth is really simple. Let's look at some examples of overfitting from ChatGPT, from image generation. First, I want to ask for a picture of two golden retrievers, wearing watches, shaking hands. Now, golden retrievers are very common images online and they're well-represented in the training data. The AI trains really well, so here's what we get. I think it's great. I love this picture. They look happy. Technically they're shaking paws, not hands, but it's realistic and they're wearing watches like I asked. But now I want to switch the animal. Has anyone heard of a saola before? This is one of the rarest animals on Earth. It's estimated to be critically endangered, only a few hundred left. It lives in the mountain ranges between Vietnam and Laos and it was first discovered in 1993. Very few people have seen it. Scientists have never seen it; they learned about it through the locals. There are a few pictures of a saola, but not very many - and they're not very good. This is what we get. Now, I've never seen one but there's something suspicious about the hands, okay? It might look like this, but it's a nice picture, so there's something going on with the saola's hands. There's something going on with the watch hands, too. I don't know if you can notice in those pictures. Create a photo of a watch. AI-generated watch images almost always show the hands and exactly showing the time of 10:09, with about 35 seconds because that's what's always shown in an ad for watches. This is the most visually pleasing time. I'm sure you want to buy that made-up watch. Now, we can ask for the watch upside-down, okay, so they're pretty impressive. It doesn't really move the hands very much but it flips all the numbers and it does it well. Now it's showing 4:40, but here's a task. Create a photo of a watch upside-down showing the time 10:10. Okay, ouch. I'm sorry that I asked. It's going crazy here, so it won't move the hands but now it's making up numbers. It has a 21 and some of them are upside-down and there's repeating eight o'clock twice. This is going crazy. One more example. Write the alphabet forwards and backwards. Forwards is very good - and actually AI didn't used to be able to do this. But the backwards one repeats the X, starts making up letters. Now, if the AIs are listening to me here, take note because I want to defend you a little bit. People can't recite the alphabet backwards either, okay? We're not that great. If anyone can do it, let me know. I'm happy to see. But the machines make a mistake writing the alphabet backwards - and so do people. But in a similar way to what Roberto said, they make mistakes in a bit of a different way. This is not exactly the way that a person would make that mistake. Okay, so let's now come back to data; structured data like we use to predict in finance and investing. Here are ten hypothetical datapoints. First, I'm going to run a regression of the Y values on the X values and this is what we get. This is clearly underfitting. Some of the predictions are very bad. Linear regression. Well, what if we regress on not only the values of X but the values of X squared? Now it looks really good - and it should because I actually made these data with X squared with a little bit of noise. So, this is pretty close to the truth. But what if we add X cubed, X to the fourth all the way up to X to the seventh? Now it's overfitting. The extrapolations could be very, very bad but it's making less error on the actual datapoints themselves. So, again this overfitting occurs but how can we detect when to stop? When is the right amount of complexity when we don't know the truth ahead of time? This is an issue with neural networks. Neural networks are incredibly powerful. Maybe some of you work on building neural network models or you've experimented with them. Maybe you manage teams who do it. Maybe other people that you work with have pitched you the idea of using neural networks to solve complicated problems and make predictions. Well, they're state-of-the-art. They're great at modelling high complexity and they learn iteratively by updating their parameters based on data fed in, in repeated sequences called epochs. So, there's a question: how many times do you train? How many epochs do you want? With too many, they learn too much nuance and they overfit - like that yellow line. But with too few, they underfit and they miss the pattern. So, determining in advance, before you actually do your important application, whether your model is properly calibrated is really the central challenge of machine learning. The usual approach to this is something called cross-validation. I'm not going to dwell on what that is. But basically, you go back to the data, start from scratch and you take subsets of the data, train your model and then test it on the leftovers. You do that with different subsets and folds of the data. It works okay. but it has problems. You cannot evaluate the final prediction function that you want to use for your future, difficult, important problem that's trained on all the available data. You only know how it works when it's sliced up. The splits introduce noise into the process. It takes a lot of loops and extra calculation and time. It requires full knowledge of how the prediction model works. If you and I, if we all wanted to test the integrity of a ChatGPT answer, we can't retrain it on subsets of the data. We need a different way. Recently, some colleagues and I, we proposed this method called interrogation. Picture a brilliant trial lawyer on TV, confidently grilling the witness and seeing if their answers are consistent. That's basically what we want to try to do to a model. We have a model here that's literally a black box. We don't know how it works, but that's okay. We want to ask it lots of questions and we want to get answers to predictions. We want to ask questions in a very specific way where we give it these prediction tasks - those are the questions - and we see what it predicts. By doing it in the right way, we can map out all of the aspects of its logic. There are four fundamental components: linear reasoning, non-linear, pairwise interactions and high-order interactions. Let me just give an example if we were studying currency prices, for instance. An example of a linear effect would be: higher interest rate currencies tend to appreciate. The carry trade. Example of a non-linear effect would be: this carry trade depends mostly on the sign of the differential in interest rates and not so much on the size. An example of a pairwise effect would be that in turbulent markets, the carry trade actually reverses and crashes. It's an interaction. A high-order interaction would be something like these reversals and crashes during turbulent markets are more severe if the currency is overvalued. So, there are higher and higher levels of conditional logic that allow us to tell a story about the data. This is an intuitive way for people to explain their own logic, and for us to explain what a mystery box is doing. It could be the mystery box in your economist's head, by the way. If they're willing to be subjected to this questioning, you could do this to a person. But what we do, importantly, is we consolidate these components as a decomposition of the prediction function. We regress these components, or rather we regress the actual outcomes in the past on these four components to test if the beta is equal one. Essentially what we're doing here is a statistical test for: are these logical parts of reasoning in your model well-calibrated and consistent? Let me give you a real example of how this works. In our research we simulated some data with the regime process and non-linear aspects, and we fit a neural network on it. We start with zero epochs. Then as it's learning, this chart fills from left to right. That gets us to about 160 epochs. The first thing we can do here is, while it's learning, we can see how much emphasis the prediction model puts on different logical components. It's relying a lot on linear relationships, simple relationships. Keep training it and it keeps relying on linear relationships. But it starts to get more sophisticated and it starts to get nuanced and introduces some of these higher-order effects. Where on this chart should we stop the training? Where has it learned enough, but not too much? Well, here's the test that we do. We have a measure of instability for each of these components - and this is a statistical test. It's how far the betas are away from one. Are they statistically different than one? What we want to see here is as little instability as possible, these values close to zero. Technically, you could do an F-test. You could do a joint test for: are all the betas less - are all the betas equal to one, and consolidate this into a line like this aggregate instability score. This tells us that we should stop training at about a little before 200 cycles. Now, here's the actual training error on unseen testing data, the actual error that happened. Our stopping point was pretty good, pretty close to the optimal. Now, this is pretty profound because without seeing the testing data, just by interrogating this box we can understand if it's going to make a good decision. Let me show you another example so you don't think I just picked the one that worked. Here's another one. Here, you see a very different picture. Here, there's a lot of pairwise stuff going on so it's learning something much more complicated. Where to stop? This picture looks different. That picture looks different. We should stop after 300. Again, that was quite good. Let me stress that this is not about picking the simplest possible model. The world may not be simple, your problem may not be simple. But we want something where the logical components are in harmony. It works on any model - even the black box - and it doesn't require the same data used for training. We actually have applied this approach to LLM answers, looking at LLMs to predict economic linkages and seeing: how credible are they based on, does their logic hold up? You can almost imagine a person. Maybe some of us have friends where you talk to them and everything's really simple. They've got these rules of thumb for all questions. But then if you catch them in a contradiction it's, 'Oh, there's all this crazy, complicated conditionality. In this case, you have to consider all these other things.' It doesn't necessarily mean that they're wrong, but it's a sign of potential instability. Let me move to the second topic here. What if the prediction model doesn't have to be a black box in the first place? What if it's not a mystery to start with? We can reframe prediction as taking an average of things that actually happened in the past. What is the best thing about data from the past? We know it's possible. These things actually happened. They're in the data, so that's a great starting point for prediction. How does overfitting look in terms of which observations we use from the past to extrapolate? Let's go back to this example from earlier. It's the ten-datapoint example. I just want to see about the prediction for the leftmost datapoint. So, predict what happens when X is negative 0.875. Remember, the first model in grey is a linear regression on just X. In the middle it has the squared term. On the right it has all these high-order terms. We can decompose regressions into exactly how much weight they put on every datapoint to predict the next datapoint. This gives you a really interesting insight. You can see what the underfit model is doing. The just-right model does something smart. It puts a lot of weight on the datapoint that's near where we're trying to make the prediction. It also puts a lot on the distant one on the right because that's also extreme. It has to put some negative weight on the ones in the middle, so this is great. Now, the overfit one is a really interesting one. What it's doing here is focussing exclusively on one datapoint. It's memorising the data. It's not generalising, so what's the problem with memorising data, why is that bad? Memorising things is kind of okay, but the problem is that all of the noise, any noise in that one datapoint, gets totally embedded into the prediction, whereas in the middle, the noise is diversified. So, it's good to diversify the noise away. Now, what I want to do here is a little experiment. We're going to shift that line for our prediction input just a tiny bit to the left, and I mean really, really tiny. So, you have to look closely because some of these differences are really hard to see. Here we go. The underfit model barely changes its weights. The middle one, actually it's smart. It scales them all up and it boosts the prediction a little bit. The overfit one now is going crazy. This is like the watch face. It doesn't really understand what matters here. It's putting huge positive and negative weights on some of these observations. Now, before I said that it was bad because it didn't diversify the noise of a single point, but now it's even worse. Now it is amplifying the noise in all these other datapoints, so this is not good. This is an interesting perspective on overfitting. In fact, some research that Mark and I and our colleague Megan have done, found that high-complexity models, some popular approaches, basically have this problem. They memorise the data and then they go crazy for new ideas. The problem is, when the model is complicated, it's not obvious that it's doing this. But if you simply look at: how is it extrapolating from the past, you can see it. We can reframe prediction as a selection of weights. Over the past five years or so, Mark and I and Megan have come up with a new approach to doing this carefully and specifically for each prediction task. It's called relevance-based prediction. We're going to determine weights task by task. Mark will describe this later today in more detail. But an important feature is that with this level of transparency, you know in advance how much confidence you should have in each prediction because you can see if there's a good pattern or not. There are lots of advantages here. You can see this confidence in advance. You have transparency to tell a story. Which historical events am I extrapolating from, and does it make sense? This approach can adapt instantly to regime shifts. You don't have to wait for a rolling window to move far enough. It even handles missing data well. I'm just going to show you an example of how it works. We're predicting the volatility of the US stock market within the next three months. We're using a bunch of market and financial and economic conditions as our X variables. So, the X variables are going to determine the relevance of all prior time periods in our data to the prediction task at hand. What ended up happening is that the relevance-based prediction worked about as well as the best neural network model trained on the same data. So, we're finding the same degree of complexity and non-linear patterns as a neural network, but in a transparent way. Now, what's even more important, I said that we can see in advance how reliable each prediction is, based on if there's a strong pattern. The predictions that we have confidence in advance in performed much, much better than those where we had low confidence. Here's an example of the transparency. Two different predictions; one from 2008, one from 2014. You can see among hundreds, we're only showing the top three, bottom three here. What weight did each prior experience receive to get the total prediction? The prediction is just a weighted average of things that actually happened. So, in case A it puts a lot of weight on these bad market crashes, October '87, etc. In case B, it was a quiet period and so it puts weight on quiescent times. Something like October '87, it was very important for A and it was among the least important for B. This tells a story, using the past to extrapolate to the future. We can also see which variables mattered. Well, implied volatility always matters, but in case B we should pay attention to the credit spread and payrolls. But they didn't matter that much in case A. Extreme interpretability that you can get from this kind of approach. So, to summarise relevance: we can see how every experience and every variable contributes to every prediction. Very, very transparent and very explicit. We can see the unique reliability of each prediction in advance. It adapts instantly and it can handle missing data - for reasons that Mark will get into later today. To finish off, let's make this practical. How can you use this? We've created API tools for our clients to apply these methods to any data that you have. The relevance one is on version two, actually. The interrogation one will be coming soon. This can be used to get much more trust and intuition for the data. These approaches are sort of counterculture. They're contradictory to what we learn in classical statistics, but why should they be? Why should they be controversial? After all, when we think about trusting a person, we ask lots of questions. We don't take on blind faith their track record. Past performance does not necessarily predict future results. We want to understand, so we can do that. We shouldn't just give up. We could understand the models. People, after all, have 86 billion neurones in their brain and we're somehow able to understand them, so it's possible. With relevance, the natural way that people think about extrapolating from the past to the future is by finding the times that are similar to now, but unusual and memorable. This is what relevance-based prediction does. It allows us to debate these parallels and the appropriateness of learning from prior episodes or not. In summary: it makes sense to get this trust. If we can create more of a dialogue with these models, with these approaches, then we don't have to turn over the keys to the kingdom and trust on faith what the data say. We can augment and add the data to what we think. I'll stop there and I'm happy to take questions.

Metcalfe: Great talk, Dave. As we had before, you can either submit through the app, you can do this… There are so many options. Or you could just raise your hand. While you're thinking actually, I'm going to ask the first one, if I may. Have we actually interrogated any of our existing models? I'm thinking mind and machine.

Turkington: We're in the process. They're being interrogated now. I'm kidding. We're in the process of doing that, so that will come soon. This is actually quite new work. We have a paper that's available on it, so there's full detail on that. But it's something that we'll use more and I think it will help us better calibrate and train these models. There are a lot of interesting extensions to it that I'm excited about. For example, I think you could interrogate using different types of tasks. So, same model, but let's say: what if we use this model to predict in a volatile environment? How reliable? Now, let's interrogate it for a different use case for a calm market. Maybe we see that it's more reliable in one case than the other. So, the model doesn't have to be good or bad at everything. Again, I love the analogies to people. You can have an expert in history that knows nothing about math, but it's not a good or bad person - just like it's not a good or a bad model, necessarily. What is it good at doing? I think we could tease that out, so a lot of extensions.

Metcalfe: Yes, as someone that runs a team of strategists, the whole idea of interrogating actual people, that's a totally different thing. We can take that offline, maybe. I don't think we actually seem to have anything on the…

Speaker 1: Hello there. Are there any asset classes or timeframes over which AI is probably most applicable, the easiest first to do?

Turkington: Yes, so we've focussed mostly on tactical monthly horizon predictions, but we've started expanding that into longer horizons. I think it's an excellent question because, as I said before, markets inherently should be difficult to predict anything that's dynamic and competitive, but for different reasons. I think in the high frequency, you have more noise factors that are not diversifying each other. But in the long run, you have more potential for things to happen. So, there's a sweet spot, probably, in the middle where in principle maybe we should have more hope for prediction. I think monthly, two, three, six months, that's actually where we found a lot of success predicting both… Economic outcomes are more amenable to prediction because they don't have that inherent competitive dynamic, as well as predicting volatility - like the example I showed. Those are still very useful, but predicting straight-out returns is really hard. But we have found great examples in predicting currency markets and other applications, so monthly roughly and a little bit longer so far.

Speaker 2: You talk about there not necessarily being a good model or a bad model. But I'm just wondering, is there a way that you can combine the models so that you can utilise, I guess, the best features of each of the models to be able to predict more robustly?

Turkington: Yes, that is a fantastic point, I love that. Yes. We've seen that work in the traditional machine learning context in the mind and machine project that Michael's been talking about, where we have three different models: a neural network, random forest, boosted machine. They all reason very differently so when we apply these transparency techniques in the interrogation, they work differently but they harmonise in a good way. When they all agree with each other, those predictions have worked the best. So, there's getting a consensus and also it's useful to know when there's disagreement. When different models that usually work all of a sudden disagree - like if you went to your three doctors and they all disagreed - you would have less confidence. You would seek more. So, it's good to know when the answer is more obvious, less obvious. Also, relevant space prediction inherently does this. It inherently takes… I didn't get into this detail, but it takes an ensemble of the answers you get from different combinations of predictive variables and different subsets of the experience, and blends them according to how much confidence you have. So, in the traditional literature overall for machine learning, it's been powerful to ensemble and combine models together. We absolutely see that. I think that's a fascinating extension to it. What I would say is, you need to know when each one is working and making sense, and when it's not. Just taking everything and combining them and always going with the average might not be enough. Thank you. Oh, is there… Well, we might be out of time. Quick, okay.

Speaker 3: What's your current forecast there for the three months' S&P? Maria says it's going up.

Turkington: I've got to go back and interrogate someone.

Metcalfe: If I can run here and just take that for you. What I will say to the point about the three machines right now for the month of June, all three for once agree and they sell Swiss Franc. So, it's a very pro-carry environment. Dave would be bullish on the S&P via his machines. There we go.

Thank you.

Veitmane: Thank you, team, and thank you everyone for sticking with us. The question for this talk is, is there any alternative to US stocks? Anybody who knows me, obviously, knows that the answer is emphatically no. There is no alternative to US stocks. For me, the reason is really very simple. So that's earnings growth, there is US, and there is everything else. Maybe at this stage I probably should just mike-drop and walk off. Realistically, that's exactly what has been happening over nearly the last two decades. The dark blue line is the performance of US stocks relative to the rest of the world, and brighter blue line, relative earnings, US, relative to the rest of the world. I don't know what else we need to tell. I think the question for this talk is really - so US stocks are exceptional. We can accept that. The question is, has US administration and the new policy uncertainty that they created, has it broken that exceptionalism? We start with the left-hand side chart, so Mark Kritzman, thank you very much, turbulence, systemic risk. Yes, markets are fragile. Markets are uncertain. Loads of things happening. Even more worrying, our right-hand side chart, we see that growth expectations have come down this year. Inflation expectations have come up a lot. Yes, and that's a really unpleasant scenario. What does that mean for US stocks? I was very lucky to take a lot from being the last speaker of the day. Yes, we heard that maybe inflation will be not as scary this year at least. It might take time or might not happen. Yes, Lee, I actually very much agree with your views that you can never count US consumer out. Yes, a 25 per cent interest rate on credit cards has not stopped them spending. Yes, people still have jobs. So for now, we're at least in good hands. More importantly for me, as an equity strategist, is to think about earnings expectation. Lee has shown you the chart of soft versus hard economic data. That's my alternative. Looking at earnings expectation, I think that the biggest message from this earnings season, Q1 earnings season, was that companies really don't know what's going to happen next. The majority of them pull guidance out, they don't know, the word uncertainty is the most mentioned word. So whichever way you want to measure it, the sentiment is not great. Look at harder data. The left-hand side - is it left? The right-hand side chart with those bubbles is that earning surprises. The size of the bubble is number of companies. The horizontal axis is median earnings surprises. Vertical axis is proportion of companies surprising an upside. Again, it's US, and everything else. Yes, US companies are much more, they have higher earnings expectation and they exceeded those earnings expectations. I think the chart on the right really gives you even more insight in what's happening. Yes, the majority of companies have done really well. Which are those companies? Where have those earnings expectations been so strong and why? It's really the tech sector. Yes, everybody expected last year that tech was growing 30 per cent or so. We expect slowdown. Analysts put numbers, almost like halved them. No, we're still strong, we're still doing well. Even the rest of the market still produced reasonably decent numbers. Really important for tech, those were the only companies that were able to exceed expectations strongly, but they were able to guide. They are the ones who did not withdraw guidance, and they continued to talk about strong capex, strong order books, strong demand for their products, at least for the next couple of years. I don't think there is any company in the world outside tech that can give you guidance for the next two years. So that's where the strength is. If we look at broader, regional earnings expectation, again, there is the US, there is everything else. That was true at the beginning of the year. That is still true now. For me, as an equity analyst, what's really important is what's happening to the tech sector. That's a chart of profit margins. I think Mark that touches a little bit on your question about mark-up after COVID. So lighter blue line, outside tech. Margins didn't really come up. Companies here, they were able to charge higher prices, but their costs came up, so margins were really flat. Where we saw an increase in margins is the tech sector. I think the tech sector is absolutely paramount. Yes, analysts are being analysts, they still expect broadening in earnings. We just need to wait a little bit more and the rest of the market will catch up. For me, it's really the tech story. As I was saying, tech is able to guide higher. Why tech is able to guide higher. We started this today with a talk about AI and how it will change our lives. I think that is exactly what is in those numbers. What is in those margins and profits, is that we are having a secular transfer of capital from the non-tech sector into the tech sector. As Roberto was saying, 70 per cent of jobs will be augmented. That means that pretty much every sector, almost every company needs to invest into AI, needs to find those technologies to make us more productive and more inventive. Maybe get rid of boring jobs or substitute jobs or substitute parts of jobs. The adoption of AI is there. It's somewhat ironic that innovation in cheapening of running LLMs came from China. Notwithstanding that, we know that LLMs are, running all those AIs is getting cheaper. That means broader, faster adoption. That only means one thing, more confident profits in the tech sector. That's exactly what we're saying. To me, the fundamental case for tech, for US equities is absolutely there. What happened with investor behaviour? We can help you a little bit more on that. This chart shows you, the darker blue line is investors' allocation of benchmark allocation to US stocks, and brighter blue line is allocation to US tech. Both of them were very high at the start of the year. Then what happens? You create uncertainty. You create potential for stagflation and worries. What is a natural reaction from investors? You reduce positions. You go back to benchmark. You don't sell the 'worst asset' or the assets that you expect to underperform the most, would be hit the most. You sell what you have, and that was tech. Investors started selling US tech. Prices came down. Portfolio losses accumulated, then you need to sell more, and then prices come down more. Then that creates momentum. That creates a snowball effect. That continued pretty much until earning season. It continued pretty much until tech CEOs stood up and said, 'Actually, we are really profitable companies. We are almost independent of the business cycle because we have this AI secular trend.' So trade policy, yes, it matters to an extent, but again, who were the first companies to get the exemption from tariffs? Also thinking about how hard it is to tax, to tariff and tax the service industry, which a lot of where tech resides. So that pretty much was the trough in investor selling. If you look at the left-hand side, that are daily measures of investor flows, pretty much early April investors started to go back to US stocks. They started buying tech. Right now, tech flow is the strongest of all sectors we're seeing. US is seeing accelerating in flows. Left-hand side chart shows hedge fund shorting activity. Line going up means that hedge funds are shorting more, so in addition to being custody bank is big security lender providers. So when hedge funds want to short there, their prime brokers come to us to borrow stocks, so that's our proxy for shorting again. All hedge fund managers got a lot more concerned in March time. Short position increased everywhere. Then the US was the first place where those positions were taken off as the market's still somewhat elevated. I think the summary is very simple, is that the US has superior earnings growth. That has not changed yet. We see very strong earnings, margins, return on equity. Whichever way you want to measure it, the US really stands out relative to everyone else, but institutional investors have reduced their positions in the first few months of the year, and now they're beginning to come back. That's a very simple case for US stocks. I feel like I had a very easy job. I've been rehearsing that story for the last decade! Cayla, on the other hand, needs to give you an insight, what else can be done? Where are the other opportunities? Good luck, Cayla.

Seder: Thanks. Obviously, we're not going to move away from the US. I think Marija made a really strong case for why it's important to stick with US stocks. However, what we've learnt throughout this year is there is a case to diversify. What you see over on the left-hand side is that - sorry, right-hand side, you can see this too - is that the black line, those are real money positions. The black line is US. That spaghetti noise, mess of all those other lines, that's everywhere else. What that tells us is that real money investors, they're pretty much neutral across a lot of different regions. This is specific to DM. There's no EM holdings on here. Fifteen minutes is just not enough time to talk about the entire world. I'm sorry. Within the DM space we do see, there are some signs of opportunity there. This is really with the premise of, okay, you're keeping your US overweight, but as Elliot had mentioned in the conversation earlier, maybe the magnitude of what's going into the US is smaller, and that gives you a little room to be flexible elsewhere. What have real money investors done this year? This is showing change in positioning year to date. You can see a lot of, a sizeable increase into Europe ex UK. Other places that have seen some increased interest is places like Pac ex Japan, a little bit into the UK. The US is starting to move a little higher. Avoiding Japan and avoiding EM. Let's talk a little bit about Europe. This is showing real money positioning for year into mainly European equities, going all the way back to 1998. Anything above that zero number is overweight, anything below is underweight. You can see it's very rare for institutional investors to go overweight European equities, but they did recently. That's what we saw at the beginning of this year, where briefly there was a short-term overweight, and then we started to see positioning move back below. We're right about neutral. I threw that blue bar in there though just to give you a sense of what's typical. That is the average positioning in European equities over the last ten years. You can see that positioning, although underweight, is still a little above where we would usually expect it to be. Now where did all of this interest in European equities go? Was it broad-based or was it just a couple of sectors? Really, that's what this chart on the left is trying to show, is when we look at different sectors, we can see that a lot of this allocation was really primarily only going into banks. It was mostly going into banks and capital goods. That's why on the left - sorry, on the right-hand side chart, what you can see is that those two lines are bank positioning and capital goods positioning. Right now, that exposure is near all-time highs for cap goods, and is very high for banks. You have to ask, okay, where are we going to start, what's going to drive investment into those two industry groups further? Are we going to start to see a broadening out? At least what we're seeing on the flow side of things is that real money investors are starting to net sell banks. There is still some interest in cap goods, but because of that very substantial overweight relative to history, we're questioning, will that continue for long? If we take a step back and ask, okay, what are flows doing for the region as a whole? That's what the right-hand chart is showing, is that we've really seen pretty sizeable net selling since April. This ties into certainly the trade war and tariff conversation, where I think Q1, we knew tariffs were coming, and there is this idea that, okay, tariffs will hurt the US, but this isn't, I think it would be a mistake to assume this is a zero-sum game and say they're going to hurt the US and no one else. This hurts everyone. I think this is what the flows are saying, is that this above-average exposure in European equities could be at risk. What are we seeing in terms of hedge fund data? Marija had talked a little bit around this borrowing data. You can see that increase right around Liberation Day of increased shorting of European equities. That was something we saw really across the board. We saw a general increase in shorting activity throughout all regions. From our perspective, we want to see, how does this shorting of European equities compare to the rest of the DM space? That's what the chart on the right is trying to show, is that we saw a greater increase in European equities than we did the rest of the world or the rest of the DM space. As we started to get more positive news, that shorting activity has started to ease a little bit. It's eased more in other parts of the DM world, not as much in Europe. Now what about profitability? What we see is that earnings expectations in Europe continue to be revised lower more than the rest of the world. Also on the macro picture, I think there's a couple of really interesting things that are going on here. The top right chart shows you improving economic surprise data out of Europe. Some signs of improving business conditions. If it was actually going to lead to better growth, we would expect to see GDP expectations be revised higher. That hasn't been the case. So that's what the bottom right chart is showing you, is GDP expectations for this year and for 2026. They've continued to be stable, but generally trend lower, especially for 2025. What that tells me is maybe there's too much pessimism in these expectations because we're starting to see economic surprise move higher. That's a sign to me that the benchmark was too low, but the outlook is still not great because GDP forecasts continue to be revised lower. Is there anywhere in the world where we actually are seeing GDP expectations hold up? That's one thing that we're thinking about in terms of this increasing allocation on the margin conversation. One place that looks better is actually Australia. What this chart is showing you is the year-to-date change in GDP expectations. That's the Y axis. The X axis is just the level or the actual forecast. Australia really stands out because not only is GDP expected to be higher, those expectations are relatively more resilient than the rest of the world. Actually, we've seen Australian equities outperform over the past couple of weeks. That's what the left-hand side chart is showing, that yellow line is Australia. When we think about what does the Australian equity market actually look at or what makes up that market? I think when you think of Australia, we certainly think about commodities. We think about exposure to China. The bulk of the sector allocation is actually in financials. That's what that dark, navy blue section is. Really what's key is what's happening to the credit cycle in Australia. We've actually started to see some positive signs. When we look at year-over-year lending in Australia, it's growing at a pace that's above pre-COVID averages. We are starting to see some signs of improvement there. Additionally, what's happening on the macro outlook? Will the RBA be able to cut more? Is inflation going to be an issue? We've already established that, okay, GDP is in a better place relative to the rest of the world, but what about inflation? If inflation doesn't continue to come down, will the RBA be stuck? Really, we're not getting that sign from price stats. Price stats are a daily measure of inflation. Bottom left chart, we see the disinflation story continuing. In many ways, it seems like the RBA has been able to pull off somewhat of a soft landing. I know the GDP numbers from last week were weaker than expected, but still, relative to the rest of the world, this is looking to be okay. When we look at our media sentiment for the RBA, that is what's on the right-hand side chart, and that light blue line is our measure of hawkish or dovishness. Anything below zero is considered dovish. What you can see is recently we've started to see this oscillation of a little more dovish, then less dovish, now we're flipping back to more dovish. I think the point is that inflation is playing a game here. Inflation is playing ball and continues to show signs of disinflation. Therefore, the RBA can cut if needed, and protect some of that growth within the market. What about institutional investors? What is real money doing? We continue to see that real money investors are underweight. They're also more underweight than they usually are. That's what the left-hand side chart is showing. In terms of our flow data, that's what the top right-hand side chart is. That's where we continue to see net inflows, really since the end of March. That's been pretty consistent. On the hedge fund data side of things, we don't see extreme signs of appetite to shore Australian equities either. All of that collates into a pretty positive view for us. Another thing we've started to see recently is investors looking for value. There's a couple of places that stand out on this front. It's not the US. What the left-hand side chart is showing you is anything in the top right quadrant is positive net buying, underweight positioning, and lower multiples. Two places that stand out on that front are the UK, as well as Japan. When we think about investor appetite, we are seeing some positive signs there from real money. When we think about what hedge funds are doing, hedge funds have this split view where they're slightly more pessimistic on the UK, less pessimistic on Japan. I bring up this chart on the right-hand side to show you in the past what has happened when investors have gotten nervous about US equities. It's very rare to go underweight US stocks, which US exposure is the black line or the navy, the dark line on the right-hand side chart. Any time there's been an underweight position in the US, it's really been UK equities that have benefitted. Now, those two instances of the US underweight are pre-Brexit. So times have changed. Some of this does have to do with sector allocation as well, which I'll talk about in a second. Then when it comes to Japan, position is about benchmark, so there's really room to add. I think the reason why it would make sense to maybe increase our allocation to the UK though is really their sector profile. If you are trying to diversify away from the US, you have a lot of exposure to cyclicals, a lot of exposure to tech. That's something that the UK doesn't have. Whereas if you're looking at Japan, if you're investing in Japan, you're going to be investing in industrials. You're going to be investing in consumer discretionary. So you already, you're really doubling-up on your cyclical exposure there. Whereas the UK, you have more exposure to lower beta parts in the market, like staples. If we think about what's happening on the FX component, Lee talked about his bearish view for sterling, that would be a pro because all of these markets are very globally exposed. That makes them more attractive in the global marketplace. Then for Japan, if the yen does appreciate, that would be more of a challenge for those markets. Then finally, earnings. When we think about our analysts increasing their expectations or decreasing, and what does that look like relative to the rest of the marketplace? It's not entirely encouraging. I think that's what makes this really challenging, is that it's hard to move away from the US. You saw that profitability chart that Marija was showing. That's what makes this part of the presentation so challenging. I think if you really are trying to diversify from a sector perspective, that puts the edge for the UK over Japan if you're looking for value. I think, yes, that's pretty much it. I'll open it up to questions. There's a lot of topics we covered.

Moderator: There are. Let me quick unlock the computer that has questions. While I'm doing that actually, I had one that I was going to ask Ramu and Elliot, but I'll ask you. Actually, Cayla, your chart on Europe ex UK holdings brought it to mind in terms of the volatility it's seen. It was really consistently underweight for years. Then in the last three or four years you've seen big overweight, big underweight, big overweight, now heading back to neutral. Do you have a view, either of you, on herding, and whether crowding will be as possible - and this is a capstone of the day, I think - in this area of fragmentation that we've been talking about throughout the day, and disruption, and disruption technology, disruption in policy? Do you have a view on whether positions will be as consistently held in the future as they have been in the last 20 years? Is the TINA trade dead, in other words?

Seder: I don't think it is. Especially when we think about what's happening with the Fed. If we don't see rates come down, that means that investors are going to be forced to stick to parts of the market that have very strong balance sheets and actually are able to invest in AI. I think it's going to be hard for some of these especially smaller cap companies to benefit from AI if they can't finance that investment. So that leaves fewer choices out there. I think it's going to drive concentration further. You'll have less variability or less volatility in terms of positioning because you have fewer good choices.

Moderator: I have a question on the app, but is there anybody in the audience who'd like to raise their hand first? If not, this was actually another question I was thinking of asking, and actually we talked about it on the podcast - another shameless plug there - last week. What does it take for US tech to underperform?

Veitmane: That's a very difficult question for me! I love tech for many, many years. I'm a very simple-minded equity strategist. It's all about earnings. Right now, tech earnings really dwarf everything else. So we need to imagine a scenario where that's not true. I think the times when you buy low and sell high, that's probably gone. There are not many opportunities. It's really buy high and sell higher, that's the mantra for equity investors, has been for the last, more than a decade. That's why that earning chart really works. You really need to find this scenario. I think what we discussed on the podcast was that we created a world where there is a set of assets, like tech, which have fantastic earnings from balance sheets, from cashflows, fantastic margins, and huge moats around it, but they're very, very expensive. Then you have another set of assets, which are really cheap, but they're not really good quality. There's very little in between. I feel for a lot of investors who need to make that choice. I felt very uncomfortable for the last couple of years coming to clients and saying, 'You have to stick with the expensive, crowded tech trade.' The amount of grief Michael gives me, I recall! It is true. That goes against everything that we learnt in terms of financial theory and analysis and valuation and Fama-French factors. That has not worked. So investors need to make this choice. I think it's this earnings divergence. The chart I started with, there is US or tech, however you want to call it, and everything else. So unless we see the convergence of this chart, either tech earnings disappearing or, I don't know, we're learning that it's very difficult to monetise AI or it's failing as promised, or you alternatively need to see the profits in other parts of the market come up very quickly. Again, that was the emphasis of this make Europe great again, we're going to spend, fiscal impulse and infrastructure spending in Europe will generate earnings. That seems to be failing more and more in terms of earnings expectation, but that needs to happen. You need to see the convergence in this US tech, exceptional earnings, and everything else. That's really what is the trigger. For now, we're not there.

Seder: I think I have three things, if I can add. The first is that could put back, put on some pressure is multiples. I think that's what made the DeepSeek news so challenging is that you had to ask, is this realistic? Is all the good news priced in? You saw how quickly the market reacted to that. Then we've also seen that's not going to be a long-term trade because what we're seeing now is investors are coming back to tech because they're reminded of their profitability. So I do think multiples have a role to play, but I do question, I think that will just catalyse medium to short-term volatility, as opposed to ending this long-term tech trade. The second thing that we've seen also is right now investors are deciding between, okay, if I'm nervous about the economy and global growth, do I go for a lower beta? Do I stick with high or high quality? That's a really tough thing because there's very low, there's not many options if you want low beta and high quality right now. I think that supports your point about tech, is that they're choosing the higher quality. They're not choosing to diversify as much in lower beta. Then my last point, there's been a couple of interesting things we've seen in terms of positioning data. Crowding can be a concern. In the equity rebound that we've seen in terms of tech and AI exposure, when we look at the industry group level, we've seen positions rebound in semi-conductors and software, but we haven't really seen tech hardware participate as much. If we get good news in terms of China/US negotiations, that's one place that could really rally on that news because it doesn't look to be priced in or hasn't been being priced in over the past couple of weeks.

Veitmane: What about regulation?

It's always a threat there.

That's always a threat, but I don't know, I'm being a very cynical equity analyst. Who was the front row of Donald Trump inauguration?

This time round…

Yes, I think the risk of regulation is lower now than it used to be a few years ago. I think with the importance of, geopolitical importance of the tech sector, I think that's lower. It's not non-existent and it's always there, and Europe is trying its hand, but I think the risk is a lot lower than it used to be.

Moderator: What about the pressure to reshore? Apple is the classic case here, where they're so inextricably linked to China. There's a new book out about this that covers exactly that. What challenges do you think they have to surmount for that not to be a problem for the multiple question especially?

Veitmane: Yes, I agree in general that tech hardware is in a much more difficult position than, for example, software is. Again, we know Tim Cook, he's probably the most shrewd operator. I have no doubt in my mind - and please don't quote me on that - in, I don't know, a month's time we'll have an announcement from Apple, announcing a new plant somewhere in the US. Probably something that started a few years ago, or signing some paper. That's the direction of travel. I think, in general, those companies became so big and so important, and moats around their businesses, Warren Buffett's argument that it's really next to impossible to compete with them. Their cashflows are just unparalleled. So I think Cayla was talking a little bit - that's probably not directly answering your question - but Cayla was talking a little bit, you need to go into safety. So what's safe? It's not low beta. Healthcare is not safe, the staples are not safe. What's safe is huge balance sheets, and that's your tech behemoths.

Seder: To your point too about…

Veitmane: We can do it all day!

Seder: …China and companies' exposure to China, this isn't just a tech problem. For example, LVMH tried to open factories in Texas, and it's been a disaster. It's more expensive and lower quality. I think that leads you back to the more profitable balance sheet companies' stronger fundamentals because that's not just going to be a problem for Apple, it's going to be a problem for any goods producer in the US because very few goods are produced here.

Moderator: Cayla, Marija, thank you so much.

Graf: I mentioned fragmentation as the key theme here. One of the questions we want to talk about in managing money and asset allocation is the breakdown of long-standing relationships that we've come to take for granted in some senses over the last 30 years. Elliot, maybe to start with you, as an opening question, how much do some of these relationships, like the relationship of real rates to gold, or rates in the dollar was a big one we had to think about a lot in FX markets in the first few months of the year, how much do you think these breakdowns are now permanent features? How much do you suspect they'll just maybe be aberrations that return to the norm eventually?

Hentov: Yes, I was also pleased to see fragmentation come up as the top choice. It means my job prospects, employment security remain robust, so first of all, so thank you for that. I think the news headlines speak for themselves. Fragmentation is a process that's ongoing. Can you fragment forever? No, you cannot fragment forever. Eventually, you find an equilibrium. The problem is we are in a transition period. Transition periods are short. they're not very long, but we're talking years, maybe one or two decades, until you resettle. During this process, the correlations do not work. We basically have to reinvent our guidebook or our roadmap as to what happens. I think we should be clear, what does fragmentation mean from a capital markets perspective? We're talking about two major processes. One is supply disruption, supply shocks. I think that's the conventional sense that we know. So it's the supply side that gets impaired, but in a non-linear fashion, which is why these correlations don't work in the same way because it's breaks and trends. Then the second one is distortions. You have active market distortions that take place, where basically you have an exogenous force that's messing around with how markets would actually otherwise be pricing. Do you think distortions are anywhere close to ending tomorrow? Probably not. Therefore, it will take, we're still in a period of many, many years of where basically we do not find an equilibrium, where we have basically a fixed set of correlations that we can go back to portfolio construction.

Graf: In taking that multi-year perspective, Ramu, and the folks that you talk to, client base, what adaptations do you see both asset owners and asset managers making in response? Do they differ?

Thiagarajan: Well, I think the set of tectonic shocks that the world economy has seen has resulted in a lot of rethinking of how to think about asset allocation in markets. As you said, correlations have broken down. We have some history of how these things evolved in the '80s, but this time is slightly different. What happened in the '80s, as Elliot said, there were supply shocks that took place that resulted in higher inflation. Monetary policy authorities felt that the only way to bring that inflation under control was to engineer a recession. Therefore, at that time, rates sold off, and because growth prospects were poor, so did equities. Therefore there was this, the positive correlation that we saw during high-inflation regimes. Now the question that we have is that where we are likely to end up? The answer in my mind partly depends on what is the market expectation with respect to how the central bank will engineer this landing? If the expectation, as it is today, is that it is relatively a soft landing, then I do think correlations are likely to come back, albeit over a longer period of time. That also depends on the credibility of the central bank itself, which has also come under discussion. What are our clients doing? For the large part what they are doing is trying to think about counterfactuals that they've not thought about and coming up with risks and ideas. Standard risk models have these coherence metrics built on correlations that have worked in the last 20 years. The type of correlations that we're seeing today, some of which you highlight in your question, are not something that we have seen. What needs to happen, what certainly I'm hearing from clients is that we are now trying to stress-test portfolios for scenarios, scenarios that we have not experienced in the past, and what is likely to be the loss in those situations? Should we look elsewhere for diversifying assets? Those are some of the things.

Graf: You bring up the stock/bond correlation. 2022, that was the experience in spades. We did have almost a Volcker-like shock to monetary policy that then, as a consequence of a high-inflation environment, drove that positive correlation of returns in a negative sense. Both of them going down in stocks and bonds, that is. Of course, since then we've seen equities in the current episode very briefly be positively correlated with fixed income returns, but now we've almost not reverted to the pre-COVID correlation structure, but we have seen equities far more resilient, especially in the last couple of months. Despite the fact that uncertainty, as we've seen on the screen a lot today, uncertainty remains very, very high. I think monetary policy uncertainty, you can map to that as being very high. The stabilities of equities though is where I, in a very long-winded way, want to get to here. I am wondering if we need to now reassess equity risk premium. Elliot, I'll start with you. I'm curious if you have any thoughts as to whether that is a reality, and if it's a permanent reality.

Hentov: Being a macro guy, for me equity risk premium is just a derivative of where yields sit. Yields obviously have to be higher. You imagine the equity risk premium would be derivative, and therefore, equity returns should be lower. You're asking about stability. So we have to think about what's the nature of the shocks to the system that we're getting? Where are they landing? We've had them, obviously, through the rate channels, through inflation and rates. That's obviously very, bonds are hypersensitive to that. One could say equities are just a derivative, so a little bit more insulated. We've had them to global trade, so that's particularly goods. I'd say that equities that are depending on complex supply chains, complex goods trade, clearly should be vulnerable to a reset in the trading system. That means a large chunk of the equity market is probably a little bit more insulated. Particularly, obviously, US tech comes to mind. What we try to do sometimes is identify geopolitical risk, specifically in portfolios. The paradox is that large, US service companies have less of that. Once you rotate out of, into other sectors in the US market or non-US, you're taking on geopolitical risk. Yes, in April that looked like a great choice, and in Q1. There's an argument from a return perspective, but you should be very clear that you're taking on risk. To your stability point, some of that in recent weeks is illusory. It's very fragile. Whereas I think there's a part of the equity market that does indeed have a protective layer around it in terms of its earning structure, its cash balances, dependency and exposure to supply chain and so forth.

Graf: That US-centricity, Ramu, I wanted to ask you about. I will say, our last session of the day with Cayla and Marija, we are going to talk a lot about US exceptionalism in stocks. I think it's important from an asset allocation, very long-term thinking perspective, insofar as that was the theme of Q1, even really at the start of, sorry, at the end of last year you started to see these rotations as a consequence of European fiscal policy, amongst other factors. In what you're talking to clients at the firm about, do you see this as a lasting theme, or as Elliot noted, is it more illusory and temporary? One of those reallocation trades away from the US really coming to a halt.

Thiagarajan: I think there is a lot more thought now with respect to thinking about alternatives to the US. If you think about in terms of realised returns, the US has done exceptionally well in the last several years. The standard finance arguments of either a small stock premium or the emerging market premium just not existed in the marketplace, has defied expectations, has not worked at all. I've talked to many asset owners who have made this allocation to emerging markets and have got burnt. What has happened in terms of realised returns have done, US has done extraordinarily well. That speaks to, part of it is exceptionalism, and if you look at the MAG7- and Marija knows more, some more about this and will talk about that - the growth has been around 30 per cent, which has been extraordinary. It defies fundamental economics. As the form scales of the returns, the market should come down. No, it's actually gone up, for a number of reasons due to inorganic acquisitions and so forth. So I do think the stability comes from the fact there's a tonne of cash on the sidelines. Bonds are seen to be a higher-risk asset, for sure, because term premium has gone up, and it has seen there is, of course, there's talk about what is likely to happen as a result of a lot of issuance that's coming to the market, whether it is premium going up. So bonds are seen to be the riskier asset here. Therefore, by default, there is a little bit of a rotation into the US in the equities. Your question is, is that likely to continue? I do think there's some fragility risk here, as Elliot noted. We are seeing earnings growth come down somewhat from what we're seeing in the top seven stocks. There's other concerns coming from geopolitical risk and policy uncertainty. I think there is discussion now, so should we go to LATAM, for example? Equity risk premium is very, very high in LATAM. People are thinking about other markets that potentially give you diversification away from the risks that the US is subject to.

Graf: Fixed income is where I wanted to take this. Actually, you've given me the perfect segue in that in talking about term premium, but also talking about issuance. Of course, it's all well and good to think about rotating more from bonds to equities, but when the bond market is growing to the extent that it is, that's quite a big hurdle to get over. Elliot, can you start us off maybe thinking about how far along we are in that process? If we've made any progress in that process. Where you think that long-term asset allocation decision needs to go?

Hentov: I'm not going to start you off, you'll have to cut me off because we're on my sweet spot…

Graf: Well, the clock is there.

Hentov: Yes, look, I co-wrote a paper with Ramu last year, a year ago, titled, 'Who will buy all these treasuries?' We did a simple model, basically some basic assumptions, assuming policy was neutral. Again, that was a big assumption. We came up with a discovery that, and the model suggested the risk premium had to be 95 basis points higher. That's not a hard flag in the sand, but a rough barometer. Since then, I'd say the market has come about halfway, maybe two thirds from where we were last year. The reality is we have not yet adjusted to the world we're in, where supply issuance carries on. Actually, the slope steepens a little bit. Most importantly, global supply issuance, not only the US, there's global supply of safe assets is rising at a faster pace than it was in recent years. At the same time, how you want to measure it, global demand for safe assets is actually slowing down. It's somewhat flattish, after a decade where we had a big, steep risk from all sorts of new buyers, central banks, QE and so forth. That obviously is no longer with us. What you get now is really a very different dynamic in bond markets, which suggests that our model was very conservative. Starting with the fact that we thought the policy would be neutral. I did pull out a few numbers because I do want to stress this. My colleagues are sick of hearing this. They know I'm a bond bear for long-term bonds, for good reason. When something looks unsustainable, it probably is. The reality is the US is on an unsustainable fiscal path. That does not mean an acute crisis is imminent, but it means the price clearing has to be at a higher level. Absolutely. So yes, my bond colleagues got excited, they said, 'Hey, some premiums are back to 20-year highs.' I'm like, 'Yes, for very good reasons.' We have not seen the end of it. In fact, I know you want to cut me off, but I did pull up some numbers.

Graf: I have a follow-up, that's all.

Hentov: I'm just getting warmed up to be very honest. The term premium, 20-year highs, it's roughly where it was between the late '90s and 2008. What was happening in the US back then? The debt stock was half of what it was today. All the sovereign metrics - I used to be at S&P's sovereign group - and the US got downgraded in 2011. On all of our classic sovereign metrics, roughly half or today it's double as bad as it was before, with those interest payments as a show of revenue. Whatever you want to take, any flow metric, it's twice as bad as it was before. What was the term premium doing during that period or the primary deficit? What was the primary deficit? Today's at three per cent. There we didn't have one. There was a primary surplus.

Graf: We had negative term premium as well.

Hentov: Well, the term was roughly, it was small, but it's comparable. Today we're at a term, a primary deficit at levels the highest in US history outside of a war and recession. The highest for any industrialised country ever outside of war and recession. These things, I tend to think, believe in these heuristics that they tell me, 'Well, the price-clearing level has to be structurally higher, even higher where it is today.' I think we still have some way to climb. I will just finish off on one thing, just because I did pull out this data before I came down to the panel.

Graf: Well, we've got to use it then.

Hentov: In sovereign analysis, you also look at how the debt is structured, maturities' profiles and so forth. The US has made some very, very risky endeavours in recent years. It has to rollover close to a third of its debt stock every calendar year. That compares to about 13 per cent, 14 per cent for Italy. That compares to about 8 per cent for the UK. The short-term share of stock is over a fifth of the US stock because they issued a lot of bills into the market. That makes the sovereign hyper-vulnerable to changes in market conditions. I do think we don't get a crisis, but we do get a reset coming this year, where I think there will be a widespread acknowledgement that, well, the ten years probably should be closer to five and to four, for a prolonged period. Possibly higher.

Graf: Ramu, can you follow that up and talk about what you think that will look like in market terms? I guess it gets to this question, and Elliot's kind of answered it, but do issuers who issue in their own currency, do they have the potential for debt crises? A very simple question, but a very loaded question. What does that look like when things go wrong?

Thiagarajan: By the way, Elliot and I don't agree on too many things, but on this one, we do agree. We did write papers on this point. I do think that, the question is are we at the tipping point on some of this? I think the answer it not quite, but as he said, the risk premium has to be higher. In the OECD countries, $17 trillion is going to be issued in 2025. That's up from $14 trillion. The world is awash in sovereign debt, and it's likely going higher. The title of our papers was 'Who's going to buy these and at what price?' As we said, it is true that it's going to, you're going to have to command a higher risk premium to offload these things. Also, what part of the yield curve are we going to issue it on? That's another aspect to it too. Are we at a tipping point? I don't think we are at a tipping point to the extent that it could result in a crisis, in a major crisis. I do think there is going to be significant repricing. One thing I will say that there's a demand side - two things. One is that the price-insensitive buyers have walked away from the marketplace. For the large part, the Fed is not buying, the central banks are not buying. Japan and China have started, are not buying anymore, and their holdings have stalled. Japan slightly higher than China. Given that the market is dominated by the price-sensitive buyers, who get to have a higher risk premium, I don't think we are at a tipping point. The demand side, one thing I would say is that the demographics is something that will kick in as an effective, you roll your time out to next five, ten years - and Lee and I have been talking about this. This is something that will result in a demand for safe paper, particularly something that can get you a decent yield. So are we at a tipping point? No, but there is enough policy uncertainty in the place that people are concerned about tail risk situations of that sort.

Graf: This is for either of you, whoever wants to answer it. It's a similar question to the question I asked about equities. During the most recent period of volatility, you had corporate and high-yield spreads relatively well behaved. Do we need to think about spread product differently as part of this broader question of asset allocation as well? Are the fundamentals of corporates more trustworthy, in other words, relative to governments?

Thiagarajan: The fundamentals have been very strong. The corporate fundamentals have been very strong, whether it be cash on the balance sheet or levels of debts servicing capability, that's been quite strong. In terms of the spread, so the spread side has been pretty well behaved. We are now in the high yield around 7.6 per cent with OAS, of around 300 basis points or treasuries, and that's very reasonable, given where we are. So corporate fundamentals are pretty good. There is this issue on the sovereign side, and that risk can translate on the corporate side, through channels that he is talking about. At this point, the corporate bond market is reasonably well behaved. I think the expectation is that it will continue, given that the fundamentals, particularly in the US, are not that bad.

Hentov: I think, on that point I would just add, if you have an orderly resetting of the sovereign risk curve, then spread products will perform fine. The risk is that the tail risk of something going wrong has fattened quite a bit.

Graf: That's what I'm getting to, it's like why…?

Hentov: Why is that not captured in the conventional spread metrics? That's maybe a question for the credit guys to ask. It certainly has, there's no question about that. The risks of something going wrong have fattened, and that should be reflected somewhere.

Thiagarajan: Yes, I do think that the tail risk, the probability of a tail risk has certainly gone up. What is happening in many asset owner communities are thinking about tail risk scenarios for a complete blow-out on some of these markets and what portfolios will behave like, but with not a very clear idea of the timeline as to when that might happen. Those conversations are becoming more frequent, that's to the point of the tail having fattened.

Graf: I'm going to finish before I go to audience questions, and I'll grab the laptop here in a second, but Elliot, if you can talk a little bit about the dollar. What role that is playing in currency hedging in particular. We're seeing in our indicators, and I think Lee will talk a little bit about this in a moment, a higher degree of dollar hedging by foreigners of US assets, but particularly of US owners of foreign assets hedging back to dollars less. Do you have a sense of how much further this can extend? Not necessarily thinking about de-dollarisation. I don't think that's the conversation to have right now. Just how greater is the propensity of the clients you talk to to hedge away dollar risk?

Hentov: Yes, I think I'll answer that anecdotally. I don't know, I spent years talking to American clients coming in, 'We have EM, we all these beautiful things outside of the US, any interest?' and they were just like, 'No, thank you.' That literally, that conversation has flipped so quickly in Q1 this year. Even before April, that conversation with US clients is a lot about, 'Well, actually, it makes sense, from the FX angle too, just to have a much higher non-US-dollar exposure somewhere, across all asset classes.' The flipside is with clients, what I call Eurasia, broadly, is about, well, we've had this constant, simple recipe. Well, any new influence that we get as a pension fund, boom, send a lot of it into the US stocks, and then figure out what to do with the residual. Now the conversation is much more tempered about, well, I think we'll still continue to allocate, but we'll allocate less. What do we do with what we keep? It seems to be a bit of a rise in home-buyers. What you see here is, anecdotally, and I think the data is now bearing it out, first with the hedging, but I think we'll see it in the flow data in the coming months is you'll see higher US capital flows outward, and you'll see lower inflows. Again, I don't think the direction changes. I think it's the magnitude that shifts. That is, obviously, dollar-negative on both fronts, beyond a doubt. I think that is why the short dollar is the most crowded trade space.

Graf: Can you talk about what that means for us as a bank? Do you have any thoughts on that?

Thiagarajan: I think, for us as a bank, given the macro flows that he just outlined, which is very much something that I'm hearing from clients as well, having products and services to facilitate that effect is really important. We have that. We have on the asset management side, we have products that can do that. In terms of servicing assets outside the US, whatever form they are, we have that. We have that capability. I do think to facilitate that is something that we offer as a bank. The other thing that I do want to say is that we have talked a lot about higher-rate environments ease the flow into private markets. I do think that given the expectation that rates are likely to remain higher longer, certainly in the US, and Lee will talk about this also, is that there is a hunt for assets that in expectation give you higher yield and higher return in private, so private markets becomes the logical choice. In realised terms that's not happened, we all know that. We just know that in the last ten years and so forth public markets have done a lot better on aggregate compared to private markets. Let alone the smoothing arguments and so forth. What is true is that the flows, we have never seen flows or discussions about private markets as much as we are seeing now. That's also the one by the rate, expected rate environment. I think, again, that's an area that we can certainly service in a big way, to facilitate that with our clients.

Graf: We have about five minutes. I'm going to quick go grab the laptop. In the meantime if there's any questions from you in the audience, please do raise your hand. We've got mikes - my screen has just died, which is never a good sign.

Down on the right here?

See who's first.

Speaker 1: Hi, thanks. I'm not talking my own book or anything, but I just wanted to give a bit of pushback on the bond blow-out thing. I would just say that we've seen concepts of things being too big to fail, unorthodox policies, no surprise since financial crisis. I just think that you would start to see governments could easily manipulate the markets through incentivising things by taxation or forcing reserve quality or what have you. They've got plenty of tools to prevent that from happening. I just thought, what are your thoughts on that?

Hentov: I think it's a great question. I think that's, my standard presentation has this bullet point, number two, financial repression because that's the logical consequence. If there's not enough demand growth, you create it. To a degree, I think that's part of this reset, which is why I'm still in the camp that I think it should be orderly. We'll just have slightly higher yields than we've had in the past. The pathway to get there, to avoid the crisis, is through regulatory means you create more demand. You allow perhaps above-average inflation. You may have the central bank involved in bond-buying. All these are part of the mix to make sure the path is orderly and you don't get a bond blow because it's not necessary. You are a sovereign issuing a reserve currency. You do have tools to avoid that. Those unconventional measures are certainly part of the package.

Graf: Any other questions from the room? Cayla? No.

Speaker 2: We talked a little bit about rising term premium, especially in the US. A conversation that we continue to have with clients is the story of US exceptionalism. Is that at risk? We'll probably see rising term premiums, especially at the long end. What about term premiums outside of the US, should they also be rising? If so, does that just mean the US is still maybe the safest bet, it's just the overall risk is rising everywhere?

Thiagarajan: I think that the notion of bond issuance, the world being awash in sovereign debt is true not just in the US, but in other markets as well. I do think that the need to finance, because a big part of the COVID and post-COVID recovery in many of these economies came from fiscal policies, which required financing. I think the answer to your question is yes. I do think you should expect term premium to be higher in these markets as well, depending on the economy. It can vary from market to market. Depending on that, you should expect term premium to go up for longer duration bonds in most markets. The US, of course, is the deepest, most liquid market. The fact that the US term premium goes up also has an effect on other markets. I think the broad answer is yes. I think you should expect term premium to go higher across most markets.

Hentov: I would add a nuance to that, that if you look at US exceptionalism, the past decade, and disaggregate the source of it, yes, there's an innovation premium. Yes, there's higher business investment. The majority of the growth differential over the past ten years has been because of fiscal excess. That's going to have to come down. Therefore, there is going to be a shrinkage of that exceptionalism story. The question is, how much does it shrink? I think it will shrink quite considerably, even though you'll still get a little bit, particularly that innovation premium that comes out of this.

Thiagarajan: I'll add one quick point. There was a fascinating paper by Cram Hall in the Jackson Hole conference, I think last year, where they did a very interesting insight to see what really happens to the bond market when there is fiscal policy announcements. What you find is that inter-temporarily fiscal announcements by any of the governments, starting with the US government, has resulted in significant increase in those premia around those times. So what that tells you is the market has become very attuned to what is happening on the fiscal side, and I think that is going to result absolutely in higher volatility, and potentially higher risk premium, return premium. You had a question? Somebody had a question.

Graf: Any other questions from the floor? I do have about…

Speaker 3: Yes, a couple on FX. Is the dollar smile dead in your view, for the most part? On rates, you don't like the long end, obviously, but is there anywhere on the curve you do like? What's the consequence for your view on inflation, given the fiscal dynamics?

Hentov: I think the short run still is subject to classic macro dynamics. Therefore, disinflationary periods followed by lower rates. Therefore, I think that's, the macro data still works in that part of the curve. Inflation generally structurally, I don't know what your priors are and what you want to compare it to, but obviously, post-2020, structurally we're in a higher inflationary regime going forward. Your first question was on FX, with the dollar smirk. I think I'll leave that to, we are hosted by State Street. Right?

Graf: We've got a whole session on that, so I'm going to defer it again to Lee.

Hentov: You'll hear from the real experts afterwards.

Graf: We've got, yes, probably, yes, we've got time for one more.

Speaker 4: How long can the US keep running six per cent fiscal deficits annually? What's going to change it?

Hentov: The headline doesn't concern me, it can run six per cent indefinitely, with some side effects. I'm more concerned about the primary deficit, which is close to three per cent, that's the deficit before interest payments. That, for an economy, if you're running three per cent of primary deficit, you'd better have a very high, fast-powered economy, which the US does not have to that degree. The maths is very simple. I can explain it to my teenagers at home even how it will eventually run out of steam. It's like the coyote off the cliff. The question is what does that adjustment process look like? How do we get to a smaller deficit? Do we get it through responsible politics and consensus policymaking? I think that's not the credible path. Do you get it through higher inflation? That's one angle that can happen, although that has its own problems and doesn't solve the issue. Do you get it through market pressure that eventually compels the deficit smaller?

Thiagarajan: Or do you get it rewriting social contracts? That is starting to surface now, which is a very big issue, in terms of the benefits that you get, whether it be healthcare benefits or pension benefits. I'm starting to hear a lot more of that. Just a third rail, which people never wanted to talk about that, now there is discussion about variations of how do we do COLA adjustments, cost-of-living adjustments on these things? That sucks up billions of dollars. That's another way to… All of these are fair game at this point.

Graf: That might be next year's research conference, I think, to talk about that. Sadly though, this session at this year's conference, we're out of time. Thank you very much to my panellists, to Mark as well for this session, very much appreciate your time and your thoughts.