[air whooshing] - You're watching Access Returns, a channel that makes complex investing ideas simple enough to actually use or better questions. We need to better decisions. Most people talk about markets and think, it's all war and peace. But I, my dear Access Returns watchers, know, when you talk macro, it's really all war and pies. How you doing, Warren? What's good? - Good, welcome to the show. - I appreciate that intro. Thank you, Matt. - The Leo Tolstoy of Twitter threads and LinkedIn posts with some of this stuff here. - You're killing me with the content. It's great. - I'll take it. - What is it? Every beer market is an allegory or whatever. I don't know. - Yeah, you got it. You got it right up there with the family quotes. - You're well first. I know you're well read. - Speaking of well read though, I'm starting with Fernando. Fernando is not here with us, but isn't spirit always. He said that being invested in the market right now means constantly fighting against the day loose of bearish arguments about AI. - I mean, it feels like everything. Because when you sort through it, which of the bearish AI arguments do you think actually have any teeth? - Yeah, I mean, I think that's when you zoom way out, that's kind of markets in general, right? It's just you're always assessing risks and that's how we're wired. And anything that's new or novel is gonna be especially fertile ground for concerns to crop up and for bear market stories and risks to crop up. And so our job is whatever it is, AI, whatever it is to as researchers is to test the theories as they come and try to be objective and level headed about it. And so that's what we do often with AI, I'm fortunate to work with Fernando. Like you said, he has a background in machine learning and so this is not something that's like he had to pick up on the fly. It's in the area of the world he's been involved in for over a decade. And so it's his massively fortuitous for me 'cause I'm just like a dumb macro guy with a background in energy. And so I'm just trying to pick this up on the fly. And I can understand why there's so many generalists who are skeptical and worried about this. This build out and it's turned them off and has made them just nervous and probably caused a lot of frustration because in order to, I mean, tech is now what? 50% of the market in order to outperform, if tech's outperforming and you already feel like tech's a big overweight, you have to be even more overweight to outperform. And so it's been a really tough market. I've seen from a behavioral standpoint over the last couple of years. So yeah, that's a totally the right description. And I think that came from Fernando. I do think that there's always worries. I'm wired to be worried. I'm always peppering Fernando as our dynamic. He's more of an optimist, a techno optimist that I'm more of like I said, a traditional macro warrior. And I'm always peppering him with questions and saying, like, what about this? What about this? And he sometimes things land and sometimes they don't. I think that right now we're in this middle in the midst of really thinking through what I would say is the open source panic. We really wanted to get to the bottom of that. Some of the charts we've seen floating around within more traditional macro. My, our, our read on that and we can talk about it is that that's not really something to worry about. That's not an existential threat at this moment in time. The data and analytics that people are pulling from really when you dig into them aren't too worrisome. And so that's not a worry. I think if you ask me the biggest concern is regulatory. What happens? I do think that the messaging around data centers, you could call it the horse who affects you, call whatever you want, or you know, you're seeing both the left and the right start to say, why are we building data centers? And I mean, I'm somewhat sympathetic to that. I mean, my son actually asked me that this weekend, he's 15 and he was like, Dad, do you think we should be building data centers? So I mean, this is becoming an issue. It's becoming something people are really talking about. And so I think the messaging and the guys who run the labs are horrible, horrible, spoke men for this world. They're not the marketers. For not at all. And so, you know, if our fate, if the AI fate is left in the hands of Dario and Sam, we're all screwed in my opinion, just because they give off a vibe of their messaging is objectively scary. Like Dario just says, you know, we're just going to lose a lot of jobs and be ready and hand-rearing and be nervous. And Sam is, I think most people find them to be kind of shady and not trustworthy. And so to me, both of those guys are kind of a nice figureheads for this problem out here, the regulatory problem. I don't think it's a today issue. I think it looms out in the future. Like that 2028 deadline is kind of important. What kind of progress do we make between then now? What kind of messaging changes between now and then? Those are all important things. So that's a big deal. And then just, I think market sentiment, but this is usually something that's-- Market sentiment sounds so general, but we do. There is a large degree of financing when it comes to the labs and their training budgets and things like that. And so if market sentiment was to sour for some reason, then you could see progress stall out. And I think model progress, as we've said, is really the lifeblood of the bull market today. Is you need to see models improve. And if models improve, you can envision deeper adoption, enterprise adoption that flows into compute demand. All of that stuff comes together to then ratify the concerns around hyperscaler CapEx. That's a big part of the market too. And all these things start to flow together synergistically. But the tip of the spear, and that's why I think that mythos model announcement back in early April, was really like when we started this generational semi-run to us, that was a big deal. And so that's where we're at right now. I don't see a problem when it comes to market sentiment, but a regulatory and then sentiment to help finance this, to continue AI experiment. I want to get back into that in a little bit at the model improvement part. I think that is super, super important. I think tying it back to the news of the last couple of months is really, really relevant with the mythos part. But first, especially because you got a 15-year-old bringing this up, and we have midterms this year, and whatever else is going to come. You call this the single best chart to capture the current macro. I want to get this technology versus housing investment chart up. This is fascinating on another way. This is coming home to roost on Packwell. We're seeing it. Yeah, so this is the part of the side effects when we talk about the regulatory, why are people are people happy with the AI build out? And I don't even think this is something that the average person can feel yet, but maybe they'll know it intellectually. Maybe they feel it is that AI is part of the crowding out of some of these traditional consumer sectors. And so we, as a macro shop, one of the areas, which I think we'll talk about, our source of truth is the residential, at least it usually is the residential construction payroll gauge. So how many people are employed in residential and housing construction? That's a huge-- it's an important gauge. I mean, usually you see that it's the best leading slice of the job market before recessions. It slows with a nice lead time, usually an 8% drawdown in that slice of the job's market happens before each modern recession. So it's a good-- it gives you a good lead and a good signal. In this cycle, we've seen week housing, week residence-- in starting to see week residential construction payroll data. And you're actually seeing residential fixed investment plateau and then fall since 2022. And you would expect this to be a week overall economy, a week consumer. But instead, what we've had is, as we went from 2022 to 2023, and the AI mega theme took off, we've seen IT spin information processing, and equipment spending combined has overtaken residential fixed investment. So we're at 1.5 trillion versus 1.1 trillion in favor of tech spending fixed investment. So it's a crowding out. You have housings having to make room for this. It's taking up real resources in the economy to build out data centers and do all this work. The whole anecdotal, you can't find a plumber, you can't find a electrician. Trades are being utilized and maybe even pulled away from the housing sector. And so that drives up cost. It also potentially raises the cost of money, the neutral rate. So this goes to the Fed dilemma that we're going to deal with in the second half of the year. Will the Fed slow inflation, address inflation, and do raise rates enough to slow that big tech boom that we're experiencing? One of our working theories is that it probably not. All else equal because in order to-- the most responsive area of the economy to monetary policy is still the housing market and residential fixed investment. And so they would have to do a lot of leaning on that sector of the economy before they even touch this data center build. I like the hyperscaler cap ex going up above a trillion
next year and then beyond a 25, 50, 75 basis points of Hikes is not going to change that trajectory. They're going to have to do something much more that believed into the capital markets and hits the wealth effect and all these other things. So that's the dilemma. That's the K-shape economy, the bifurcated economy. That one chart in that dichotomy between residential fixed investment, plateauing and going down in IT spending and tech fixed investment, just moonshawting higher, that captures all these, all these issues. So it's me. It's the number one chart to start framing a macro discussion. Drilling on this just for a second because this does make an incredibly important point for the Fed right now and where they are because hyperscaler CapEx, we're looking at like 2.5% of GDP and a huge chunk of this rise in core PCE is coming from information processing equipment spend. That's going to flow through the way we think about inflation and growth. Yeah, you're starting to see the Fed talk in, they did inflation X, tear, inflation X oil. Now you're hearing some inflation X, inflation X tech spending. I mean, PCE just from like you said, those two PCE categories, which are small categories, but just explode. We never talk about it like this. Yeah, I mean, I think we were adding, we did the math. I think it was like, we were adding seven basis points on a month over a month basis to core PC. So I mean, that's the difference. Like if you're at 0.2 versus 0.27, that's a big difference if you annualize that rate out. And so yeah, it is in the inflation data. And so the Fed has to take that in and then they have to decide, do they have the tools to address that? And I've had a lot of debates around this and like they do have a dual mandate too, which makes this complicated. So the Fed has to, a lot of people pretend the Fed is like one of the other global central banks where they just have price stability as a mandate. They have price stability and full employment. And so if they were to just address inflation, they could end up killing Lairmer because it doesn't click. There's any real tightness in the labor market or any inflation coming from the labor market. It's coming from a generational capital spending boom out of the tech sector and out of the fiscal, the fiscal deficits that we're seeing, which are kind of taking a back seat, but they're constantly running in the background. So zoom in on jobs a little bit. And partly because I so appreciate the way you always frame residential construction and stuff like that here, but also because the K shaped part of the labor market, that is where we're looking at Eric Pakman data for the people's been highlighting this a lot for like a year and a half. The jobs we're adding are not the $200,000 white collar job. We're adding nurses, home health aids, things like that in the 30, 40,000 dollar a year range. That's not helping consumption. It probably distorts the way the Fed looks at the labor markets. What do you think about that? Yeah, I mean, again, it goes back to really monetary policy. It's a surgical instrument. It's a it's a sledgehammer in a lot of ways. You can't just surgically stop one area of the economy and hold another one constantly. If they want to raise rates to really impact inflation, they're going to end up killing a lot of activity. It's in these areas that are already weak that will suffer the most. Like you said, if you go back one year, we're ex-healthcare education, government jobs, the non-farm payrolls is flat. That's not a sign of a really strong labor market. It's actually if you go back in time, that looks like a kind of recessionary levels. We've had two months in a row and a two per two plus percent drawdown down the residential construction payroll sliver that we look at, which again, 8 percent is like a pre-recession level, but a two percent drawdown and two months in a row, that's a yellow light flashing in the economy traditionally. If you look at wage data, if the labor market really takes a lot of people say, hey, you're looking at the establishment survey. Because of immigration policy, the labor force is shrinking. We don't need to be growing jobs in this type of a backdrop in order to stabilize the labor market. If that was true, I think you'd be seeing wages stabilize at the very least, but wage growth has stagnated. It keeps making new cycle lows if you were like the Atlanta Fed wage tracker. If you dig into that wage tracker, college educated workers, that's the weakest wage growth sliver within that. You're also seeing the percentage of workers who are getting no raise go back above pre-pandemic levels. It's something like almost 14 percent of workers getting zero raise in that data set. You start putting all this stuff together. I don't think it's a building data center's out. It takes up a lot of certain resources and certain trades, but it is far from this like economy-wide labor market boom taking place. It is. It's crowding out a lot of the more labor-intense areas of the market that we rely on usually. I don't see, when I put all that together, even though we've had three straight, decent, non-farm payroll reports, and that's kind of the market's trying to digest that, I don't see a picture of a labor market that's really reheating yet. Then we had the Jolt data come out, and that's also bad response rates and things like that. Opening's, openings are up, but if you look at the quits rate, it's down. I don't think I'm going to over-index that data set. When I weigh all the evidence, and I'm open-minded, maybe we're getting to reheat. I think we did have a fiscal expansion in the first half of the year. We have a big cap-ex boom and wealth boom and things that could power this economy, but I am. It's been a model-through labor market for a while now, and I think that's the reality that we're facing, the feds facing. Let's go back to GPU demand, because you keep telling me this, and I believe this, this has been really useful in framing this argument for other people. We'll get the chart of this up here, too. Why is availability basically a cleaner read than CapEx announcements or revenue and this fast index? I want to talk about what that's telling you. This is something going back again, being very fortunate to be working with Fernando. This is something that back in 2023, when the AI story was really taking off, and we were doing the back of the envelope calculations on what this meant for the economy and the labor market and white color displacement. We were free big skeptics to be honest back then. Fernando started compiling data where he'd go out to the various neoclouds and on an hour by hour basis every day of all these years going back and starting in 2023, we would check how available each GPU was. We've done this starting back with the A100s and going through with the hoppers and then now with black wells and we're testing it always and tracking how frequently we're able to obtain a GPU from a neocloud. We compile that together and we average it and basically have an index and it tells us it gives us a good read on supply and demand, which GPU supply and demand. This is the bleeding edge of the market. On demand GPUs from the neoclouds. That's what we look at and again, we initially started collecting the data saying from a place a posture of skepticism and while we watched this whole thing unfold, it was actually allowed converted us into believers where we could see compute demand just continually growing and the GPU availability would be trending down and we could get these cycles and you would see GPU and then you started seeing GPU availability collapse for like the H100 and then you would see the rental prices going up about two months after that. We saw that again with black well, we saw black well B200 availability collapsed this year, the beginning this year. I mean, we went through the spring of this year and saw the availability for all GPUs go to 0%. Black well went to 0% and then rental rates exploded off the back of that. So availability is a real time, hour by hour, day by day read on supply and demand dynamics in the GPU rental market and I think it's, you know, to me, that's the kind of analytics that this whole thing is this ecosystem lacks. And so we really pay close attention to it. We've built out a lot of different views. We now have foreign availability that we have by different vintage and what's and we've created an index that's weighted newer vintage over older vintage. And again, it's, it's, to me, like you go through this period of time where I don't know if the open source story is necessarily an anti compute demand story, but there's always latent fear out there. And what we've seen is that just in recent weeks, black well availability, it creeped up maybe like two or three percent back on the floor on available. We can't get it. There's a scramble for compute that's been going on in this market since the big really sense the agentic age in the early part of the year and it has not let up and our data has, it does not show a sign of that.
starts, if the market starts loosening, you know, people are worried that they met as coming in, coming to market, and they're going to put like more start running out there cloud capacity. And hey, we'll see that. That will flow into the neoclows. That will flow into our hour by hour, day by day data. And we'll know. We'll know. Is this actually impairing the story? But again, we're data driven. And that's going to be the thing that leads us. The other thing I love that you're tracking inside of this because you're raising this point, which is we need other ways to track this. It's kind of like I'm thinking about World Cup tickets and people thinking they can just go to Stubhub or Seed Kikur whatever and get them. And they go, oh, this isn't the way this works for my local baseball team or football team. I have to think about this in a whole new light. This is where we are with this. I want to talk about the token maxing thing. You were flying online basically how it's fading token costs were collapsing as the bill came to you got an open router and you actually measure it. What's the data show? Yeah. Again, this is our, I think the way from the beginning of the conversation you take in the bearish argument, assess it objectively and try to weigh its validity. And so the most recent thing that's been crapping up is two things. And I love the guys over at Silicon Data, you know, and I'm interested in the work they do. And I think they're doing good work. And I talk to the folks over there. Pretty frequently. So I'm not integrating their stuff whatsoever. But they had this token cost index that came out is really interesting series. It, you know, it purports to basically show the price of token. And it had a strong rise through the year and then it collapsed. And I started seeing things like dual wisecale charts of this, this token cost index versus mag seven. And that's like become this the macro, the macro answer, everything is dual wisecale charts of two series. They're both going down at the same time. And you just say, aha, that's that's something. But anyway, so it became a very big deal out in the macro world. So that caught our eye. And it's like basically the implication is that the cost of the price of tokens are falling is on a global whatever basis. However, they're doing it. And that implies that the just token scramble the token maxing thing is falling off. The other thing we saw was the open router data. So open router platform for routing token consumption and model usage. And it's like what you saw was the chart of of open source Chinese models, open weight model of taking massive share on open router from the frontier models. And so this was another, you know, became another bearish argument. So again, Fernando, who is deep in the weeds on the stuff. And if you ever really want to get deep in the weeds, he would be a great guest. But Fernando went through and we've been collecting open router data for a while actually internally. And so went through and looked at the pricing of these of these tokens for frontier tokens on open router and in open source token pricing. And it's true. There has been a met an explosion in open source tokens process. And it's, but it give the pathetic context. The total inference month over month tokens processed on open routers up 40% month over month. Um, open source has taken gone from like 45% to 35% or 43 to 33 something like that over this month. And so if you, you can back into what's the growth rate here is probably something, I don't know, like 60, 65% month over month growth in open source tokens on open router. But still double digit month over a month growth in frontier tokens and moreover to go back to the pricing point. We're seeing pricing for frontier tokens rising on open router. And we're seeing open source token pricing falling on open router. So when my, my, the signal that I take from that, just as a macro generalist is that this worry that overall token demand is falling off. It doesn't wash. This is a small sample of the overall market. And it's very, it would not be represented if necessarily it would understate the strength in token demand from the frontier models of anything. So you're not going to open router to access frontier models typically to going straight to the, to the lab. So to me, the price rising and growth of tokens process for the frontier tokens rising tells me that these concerns are way overblown. But in fact, the, the token scramble continues. This whole usage rate continues. Now I have no doubt there's going to be ebbs and flows in that. The token maxing thing is kind of stupid. Like it's a stupid metric to just say, your, the more tokens you use, the better job you're doing. I mean, there's obviously going to be, it's a stupid metric that comes with stupid stories. Like you say that then you're going to get the thing about some Yahoo at work doing something, bozo thing. That makes headlines. Now everybody makes fun of it. But that, that's not the story whenever they do that to you. Yeah. And so to me, that's, I get it. But it, again, in it, you know, I'm totally, maybe that's happening. I don't have a strong token. The tokens are beanie babies. Maybe, maybe, maybe, maybe that's happening. And if it is, we'll, we should see it in the data. Like in my view, if that was, if we had a token maxing craze and something came over for for American, Uber blew through their budget and all these companies blew through their, their, their token budget. And now they're pulling back aggressively. And the open router data is showing they're all switching to open source models. And that's their, their open source models are good enough now to do all the tasks that we really wanted done on the frontier. We would see price for frontier tokens in our data falling. We'd see demand for frontier tokens on open router falling. You would have just see this, the mix shifting, but the old, you would see the, the act that part of the pie going down. Instead, the pie is just growing rapidly. The total pie is growing rapidly, frontier growing, but open source exploding. And that makes sense. I mean, it makes sense to be more efficient with your token usage. And the open source models are getting better. We run our own AI application that we, we, it's a research assistant that Fernando created and he does his own custom benchmarking. And you can see it in our benchmarking that the open source models have gotten very good, but they're not good enough. They're, they're not good enough to displace the frontier right now for our purposes. And if not for us, then probably not for most. I'm putting this chart up here. This is the chart of basically those open source models getting closer and closer to the frontier. And basically the, the threat of the models to where they stand. You want to just explain real quick, because I, I think this is interesting. You have code correctness instruction following some of your accuracy, chart quality along the bottom and how you're doing that custom benchmarking. Well, I don't do it. It's Fernando does it, but it's, we, we have a, an app that's a, that's a, it's an AI assistant. It takes, it's taken all of our code and it's like it in our data and everything and you can come to it. And I guess this could be a marketing segment too. Like if you're interested, you can put your name in and try it out. And this has also been the thing. It was an internal tool to begin with. We call it Caliban. It was an internal tool. We used it to create a large amount of our studies, but it wasn't good enough to do everything. And then about December of January, when every, when all the models leveled up, we were able to start doing almost all of our stuff. And now I would say we just live in that, in that environment to create charts and studies. And you should, you mean, you see our work as I, I don't think anybody's doing more complex stuff than we are. And it's all with living within that app. So that's what it's doing. We're selling it to, we're giving it to clients. We're selling it to outsiders. And they're doing analysis, their own custom analysis and creating charts, creating bundles, doing all that stuff. So it's an analysis and charting tool. And Fernando is in there as the creator of this and benchmarking it across these different dynamics that you just read off and seen how good is it? Is it good? Because we would love to, one of the few things we've learned in this is kind of helps us with the tokenomics understanding is whenever a new client comes to us, we have to, we have to kind of guard the amount of tokens they use because people can blow through a budget and the whole thing becomes un-economic for us. So we would love it if we could switch everyone to cheap open source Chinese models, you know, and, and, but our testing says there's still a significant gap, you know, even though it looks like it is getting closer, they are getting better. I don't want to underplay that. But for our purposes across those four dimensions, it's still not good enough. And I think that that experience is, it's extremely instructive and helpful for us in understanding this the overall landscape that we're dealing with here as investors. So let's get into markets a little bit more. It's called semi is the meta-group underneath the AI story. When that leadership started to crack with semis, drop in double digits, handful of sessions, a lot of pain, a lot of red on the tape, you argued that it looked more like a viable dip rather than a top. I want you to explain what you saw.
Well, I mean, I think that the number one the number one thing is going back to the availability the idea is that This is a mega trend. I do think The other the other factor here is that if you look back to the 90s and yeah, I think My hypothesis we could be wrong and we'll see how how everything goes my hypothesis is that this will end up being a bigger Cycle for semiconductor than we saw in the late 90s and at 95 to 2000 period Now let's talk about like kind of the negatives. So the negatives is we run a bubble scream or what we call a blow-off top blow-off terminal top screen We run it for all these markets and we run it for all the industries within the S&P 500 In simmies in construction engineering are the two industries right now that screen Positive as a potential blow-off terminal top This this condition though first started a back-in Gosh, I think it was 30 months ago at this point and so we've seen it for a while back in the 90s we first had it in 1995 with the first time you saw Simmies register as a terminal blow-off top in that condition lasted for 63 months on and off But basically we saw the the bull market run for 63 months and I think over the course of that time at 11x as a group and And we're at like four x right now and tracking exactly where we were in the 90s the other thing though is that back then we saw It's once you get into these periods and you're in this mega trend this this cap x explosion We did see four big Pullbacks or bear markets like 20% plus drawdowns in the semi group. We've had two of them so far. So everything lines up Well, and it's like one a year you're gonna get one pullback like this a year 120% drop a year in the group Be ready for it the question you have to ask yourself is is this the top is this the end of the cycle and Dub just I don't see that in the fundamental data in the GPU availability data You can there's enough visibility for spend and for catbacks and for every in the other things that we've talked about and so to me You have to let technicals guide you in those moments and then have conviction to step in if you're if your fundamentals are in place and I think if you know something Again, remember in 2000. There was a couple factors I mean, there is a pretty hard stop to that bull market has a Y2K so everything got pulled forward and we're gonna have anything like that this time around in Number two, you had the fed hiking race which we talked about so if the fed started hiking rates And then you saw semi's pullback. I would have to do a re-evaluation of Is this really the end is the Fed gonna break the back of this this thing earlier than I expect But that's not my forecast and I think for 30 months in 31 months in at this point maybe We had a 63 month run back in the 90s and I do think If anything, this is gonna be a Longer a longer and stronger bull market doesn't mean we're gonna think there were the 38% drop at some point in that late 90s period We could definitely do that at some point We go up a hundred percent and three months and then you know, you can easily pull back 25% But the question is have we seen the absolute top I would say no I have to ask this very Pointedly about this are semis still cyclical as an industry and we're just in this phase where temporarily they don't act like cyclicals because they grow grow grow grow grow grow grow grow And then they become cyclical again on the other end or They suddenly wide-mode businesses Man, that's like the big question and I I'm not smart enough to answer but I am smart enough to acknowledge that there are some things happening under the surface of the market that that suggests the way these things are being priced Suggest that the investors are treating them as At least less cyclical than they've been historically and this is a big deal for me in the way The overall market gets valued so we came into the year Bullish with bullish last we've been bullish and our One of the big bear arguments to talk about things that create anxiety has been valuations and we've talked about this said I don't know if I talked on this broadcast. I think I might have But one of the reasons we don't we've not seen over valuation in this market is that We have massive margin expansion out of non-cyclical pockets of the market and when you get Mark margin expansion from Non-cyclical park pockets of the market you should expect to see multiples Expand and when we adjust for everything the market has been Rational and how it's done that the semi-cases may be the first Bit of exuberance that we're seeing though. So semis traditionally the most cyclical tech spot us industry the Therefore when margins expand if you look at a historic chart of margins for the semi-group in price to sales multiple When margins hit a cycle peaks by price to sales starts going down to a cycle trough You don't want to pay up for peak cycle margins in this cyclical type of group That tendency has totally broken down. We've had margins explode obviously out of the semi-group You know we can see you can see micron you can see in a video you can see all these companies where Bottle necks or whatever have exploded margins and instead of getting like price to sales for the group going way down You know to one or two topics twice the sales as exploded so For us this is one of those things where how do we factor this into the overall market value? Which because semis are like 20% of the market now and It's not a small it's not a small factor where you're trying to do your your overall top-down analysis so They are priced right now the answer the question is they are priced right now like they won't be cyclical so I think there's a knee-jerk Kind of healing to because of history and everything I laid out to fade say this is a bubble probably is the beginning of a bubble in some spots for sure In this very specific part of the market. I don't think the overall market's a bubble But there's always air pockets of exuberance, and I think this is one But again going back to the 90s. I don't think we're it's this cycle is going to be long It's going to be the longest cycle we've ever seen and so there is it's not They're not I don't I'm not buying into the sci fi world where semis are just no longer cyclical But I'm also not buying the idea that this is close to a cycle top and that we should be like Expect the expect the margin growth to slow down Anytime soon and so to me there is no finite answer to that right now. It's a little bit of both It's just a matter of being aware and baking that into your overall market forecast Invaluations are a horrible timing tool. So I'm not going to time anything off that just gonna observe Don't time anything off it and if I learned anything from my own childhood in the 80s into the 90s It's that I don't want to grow up and I want to be a toyser s kid and The spoilers I grew up in toyser s1 bankrupt after multiple turns in private equity. It's this all-in's badly one way or another You've got this line about bull markets. They don't die of old age. They get murdered by the central bank I know we mentioned a little bit before But I think with the backdrop of what you just explained about semis and their contributing to margins and where we are in the cycle here It feels like Fed mistake is now top of the list for the next couple of years Well, yeah, I mean we're off the Iran war obviously Are we though? I mean we haven't no one's talked about we haven't talked about for 40 minutes So that means we're off that means we're off I think but so the what we're left with this kind of the after effects of that I do think that the war we spike the oil prices. We've had Look, I don't think that the inflation shock we saw coming out of COVID was really a supply shock But the Fed claimed it as such so they had they sit with a said brief transitory inflation back then It was a supply shock will look through it Then we had terror and that didn't really work out for them. Then we had tariffs last year again I think that's more of a valid supply shock. They look at they said we need to look through this there's Issues with the labor market and yada yada yada and now of course we had a supply driven spike in the price oil straight Or moves closed and this is your quit essential Supply shock, but it's all these things together have really put the pressure on the Fed so now you have the Fed Back in the corner where people are starting they're worried people are standing lose faith in them and So they could end up being switching there to a more real hawkish reaction function and obviously you have war shen So there is a change of leadership And so these are this is a real risk. I think it's a time where the Fed is on a true hold We've been in this default cutting position And I think today is a true hold where the next move is the next move is actually more likely to be a height than a cut But the next move could be either one they're on a true hold And so that's that's that's a different spot for the equity market and I think whenever you look at valuation multiples or when major tops are formed Fed tightening shows up around these
these periods. If you think about multiple expansion, you don't get multiple expansions during tightening cycles during hiking cycles, you get multiple expansion when the Fed is cutting. And that was the assumption for us coming into the year. You have to take that out. So earnings up 25% this year, that's great. But we should, we should expect multiple contraction against that. Earnings up 25% this year. But as we talked about, it's a more cyclical stream of earnings. It's semis and energy driving the marginal gain in earnings. So you should expect multiple contraction. So these are kind of things that cut against the bull story. And you have to just weigh that out. It has to be part of your analysis. And so if the Fed were to actually take the move from a true hold to a hike, I, you know, I think you cut risk. That would be my, that will see what happens. Then I don't want to pre-judge it. I'm sorry, Kevin Worsh. I don't want to pre-judge it. But, you know, I, I think that would be your very simple game plan. Our simple rule of thumb, we've told people clients for years of that, Hey, if, if you don't see a recession in your indicators and the Fed is easing, you can't be underweight equities. And that's changing if the Fed gets into a true hold and then potentially a hike cycle. So that's the, that's a big deal to me. And as an, as a macro guy, I could see Waller Williams, Worshan Warren, you guys could all throw up there with pal and three in the mix, the, the, W, W, P. Yeah. No, I do think that's the big, that's the big voting block. That's why I don't think there's going to be a hike this year. You have Williams, Worsh, Waller, Powell. That's the core for. They are all, if you read the T leaves, I think they all believe rates are restrictive and believe that they're just fine sitting here and holding at three and three quarters or whatever. And then you have cash car. You came out and basically agreed with that view. And so that's five out of the 12, you really need a clean sweep of the rest to start a hike cycle here. And I don't think we're going to get that. It would really take a lot out of the job data to, to move in that direction. In my opinion, I don't think we're going to get that. So if for now, I'm holding tight that, you know, we're going to be okay in the equity sign. Taking back the oil just for a second, because I want to talk about, is there anything more there? Do we have the lingering effects of an oil shock? Is that a myth? Is there more nuance to this that we're going to find out in three months when some piece of supply doesn't get delivered to the right place? And we all freak out again. I think that we dodged the big bullet. We're still releasing SPR. So the, the odd feeling is that right now the market's transitioning from a severe shortage to a bit of a blood because of all the SPR releases because China hasn't returned to the market. And so that's kind of, it's kind of crazy to say, but and if China were to just run back in, because China killed about 4% of global oil demand, just like overnight by stopping their imports. As they do, they just, they just do that. We have swing producers. We now have swing consumers. And, and so China released stress tested their ability to do that. And as long as they don't come back, I think the market's pretty over, over supplied now, the one thing that's worked, you know, because I think every, all every oil forecaster who went through this, expected us to have a harder time dealing with the straight-worm who's being closed. I did. I'm not going to run from that. You got, you got energy analyst phones and you've skirt and we did. I mean, it was a scary time. But the thing that's worked, if you were just really just, if you could have stripped away all the scary stories as technicals and positioning data. And right now the positioning data is back to very stretched. There is about a 40% managed money short position, which in my framework is extreme pessimism and short position. And the best signal for the last 10 years, the single best signal is positioning in the oil market. So we're over, we're stretched on the short side. If you get a short covering, there'll be a bid into the price of oil. Do you want to buy that and hold that long term in the face of now OPEC starting to overproduce and SPR oil floating around it, you got to get your bullcases. Now, what we're going to buy back SPR and refill it. That's not a good bullcase. And so I think you're going to get a spike here, like back to 8590 brand. As we whenever we get that short cover, you got to wait for it though. And that's, that's possible in, in my view. But the long term clearing price of oil is not going to be it's not going to be significantly higher than where we are right now, in my opinion, like 7580, we have raised it. It's higher than it was before this glut, before the war, but it doesn't feel like a macro risk to me. It doesn't feel like anything we can't deal with. It won't be a problem as long as we don't, if we go back into the war and cutting down this closing the straight warm moves. Yeah, I mean, anything's on the table then. But I just did, I get the sense that the administration has no appetite for that. Any, and it's okay if you don't have anything on this, it's just curious because when I think about other cyclical industries, like I think about the refiners, and I think about where they are right now with where crack spreads have moved to and the whole just shape of this mess and oil as it gets worked out, as the kinks get worked out, basically, assuming this stops, assuming we revert towards normal with some extra embedded costs, assuming we get some short covering and, and they're allowed to normalize out there too. Do we think about some of those other cyclical industries? Are there any possible flare ups or flare downs that come out of this for equity markets writ large? Not really. I think, I mean, refine product is when people say we need to watch inventories. In my view, we really say any to watch refine product inventories. You need to watch gasoline inventories and discipline inventories specifically in the United States. We are at multi decade lows for both. You know, we did drain inventories aggressively and we now have refiners that are running full out like higher than we've seen them run at this point of the year or really any time of the year in the last, you know, five post-COVID era. And so those refiners are minting money right now. They're buying cheap oil and selling decently elevated refine product and capturing that spread. I don't see it creating any kind of systemic problem. It's just going to create more tweets from Trump or Trump is like, why is the price of gasoline not coming down as fast as the price of oil? And this is just how the markets work and stuff like that. So, but yeah, I mean, they're they're decent. They're a decent enough play in the energy space for that reason, but don't see it as anything systemic. It's interesting too, because it's still such a tiny part of the broader markets. And actually, let me use this to steer us here. Let's go back to earnings. Let's go back to how strong those estimates have been when you said about multiples before. Let's get the S&P 500 yearly trailing 12 month earnings growth chart up. You said that it's basically front loaded to match the AI build out. Where do we think we're going in a year? Right. I mean, we're probably going to be in mid-double digits for a bit. And my reason for posting that chart is you see this. So everyone's also looking for this idea that we're an earnings bubble. That's like one of the big things that's going on right now. It's fun to say. It's a earnings bubble. And there is a long-term analyst estimate chart that's split around that, you know, I think it was Yardini's who made the chart. And there is a field for long-term analyst estimates. I think it's an iBus or something. And it's partially populated. Like if you're doing like annual estimates, you've like 90 something analysts who are filling in estimates. And for this series, there might be like three. I mean, it's not representative. No one takes a serious and it shows 25% annual growth in earnings for five years. And I just can't you can't make sense of that number. I mean, I think it's capweighted versus earnings weighted. It's done a lot of different ways. But the true numbers, the consensus numbers are what we show in our chart. And it's much less scary now. A five-year growth rate, including the 23% we're growing this year, is more like 15 or 16%. That's what we should see. We did one where we stripped out this year and said, because it is front loaded like you said, and said, what's the growth rate in earnings? I think it's about 14 or 15% out through the end of the decade. Strong, but not in this historically crazy world where we can't hit it without nominal GDP being 10%. And that's what I hear a lot of the macro bears saying, oh, nominal GDP is there's a big dig. There's a big divergence between where this GDP is and the speed limit of the economy and what the analyst have baked in. And they're just taking faulty numbers. These aren't real numbers. 25% five-year growth rate, annualized growth rate over the next five years. That's not real. That's not happening. No one's that's not you're not if you're using that to base a whole macro theory off of you're not you're not interacting with like the reality of what analysts are saying. Part of what's fascinating about this to me when I look at your five-year projection or the way you're looking at the projection here is the composition of markets is going to change our that period of time. It goes back to that 30 month versus 60 month run up in the 90s and what we see where that Peter's out and then where the potential effects of this land in the rest of the real economy. So when I think if I'm going to think out two or three years, how do I think through that? Yeah, I mean, I think to be totally honest, I think looking out two to three years is it's that of a fool's errand. There's too much stuff that's going to happen between that. I can't think about two or three years from now I can make a prediction and who knows where we'll be in three years to follow you and be able to you'll be able to find
to come back and hold my feet to the fire. I think it's too far out to really have any confidence. The thing you can say, and I think the thing that people are dealing is, "Do the, does the three-year-out estimate where we are today, does it align with the cone of probabilities that we're going to go through?" And the argument when you use something like a 25% annualized number is that we're analysts are seriously removed from the economic reality. But that those aren't real numbers. The real baked-in numbers for like 2028, 2029 is, you know, 13% Arnie's growth, 12% Arnie's growth. Do we hit that? I don't know. Is it possible? Do we need some kind of economic regime change to hit that? No, we don't. Speaking of economic regime change, last time we talked about the debatement regime, we talked about why you would not be underweight equities, and that's proven very correct. I'll just put it at that. As we think in our media term though, do we stay positive on it? That that base yield just laid out doesn't sound like a bad path. Doesn't mean we won't go up and down. We won't have drawdowns and some pullbacks, but still feels like you're saying stay long, the equity markets. Yes, we came in the year and we said we thought H1, first half of the year, peak Goldilocks. So we saw disinflation continuing in the form of oil and the form of shelter in the form of labor. And we saw fiscal expansion from the one big beautiful bill and AI powering. We wrote about this and Arnie's explosion. So we thought there'd be an Ernie's explosion, fiscal expansion, but still disinflation. So it would be this perfect Goldilocks environment in the bulk of the returns for the year would come in the first half. Anti midterm election year cycle. I kind of like look at the cycle charts and then fading them because they're so people over, they overweight them. So I think we got that except the Iran war really disturbed the Goldilocks balance. So now it's about like we said, 15 through what we're left with coming into age two. And I think that the Iran war has taken away the Goldilocks environment. The AI build out that we talked about in the spending of tech and the crowding out of things and raising inflation a bit in the economy. That's also overheating. So we are transitioning into the risk of overheating and bringing the Fed in. And so that's the big risk. The positive is that earnings, even though we thought there'd be an Ernie's explosion, I don't think we expected 25% earnings growth this year. Ernie's is is really taking the baton. So when I put all this together, we came into the year with a very aggressive, I'd say pretty aggressive targets on the S&P 500 of 7850. We came out and just basically said, we're holding a steady. We thought we think that their earnings will be higher than we expected. The Fed will be tighter than we expected. Multiple will contract more than we expected. But that because we have earnings so high, we can still land at that 7850 level. So we came in at 7850. We held our target study. It's not the most rip-roaring bullish thing to say. But I think we'll be just fine. That's a pretty good year. That's a 15% up year on back to 20% up years. And I don't think you have any reason to believe that the top, the secular top is in until the Fed comes in and starts tightening. Or the economy starts looking like it's going to weaken. It's very difficult to get a recession. Like we said, when hyper-scalers are spending 3% of GDP and the fiscal deficits at 5, 6%. So generally a good backdrop for equities. For more and pies, mouth to Leo to Tolstoy's years. That's where I'm taking this home. It's just fine, Leo. It's just fine. Or if people want to check out the research, they want to see this new tool, which is amazing. It's on the website. You want to check this tool out. It's really, really cool. Where should we send them to buggy on the internet? 314research.com, the number three spell out 14research.com. You can find me on Twitter. It does name of the AI research assistant is Calabans. If you reach research, if you Google or search for 314research, Calaban, you'll find that landing page. And we are an institutional shop. So in general, we service institutional investors. And so put your name in, fill out the information. And remember the team will send you some sample research. And we'll discuss if it's a good fit. If you're out there, you're professionally running money. You're doing this for yourself. You want some off the beaten path charts, data, and tools. I can't suggest enough 314research. Super, super, super cool. Warren, thanks so much for joining me today. Thank you for having me. Appreciate it. You're watching Access Returns. Check out the substackle of transcripts, posts, all sorts of stuff about this episode going live there in the days ahead. Like, comment, subscribe, all the things below. And we are out. Thank you for tuning into this episode. If you found this discussion interesting and valuable, please subscribe on your favorite audio platform or on YouTube. You can also follow all the podcasts in the Access Returns network at accessreturnspod.com. If you have any feedback or questions, you can contact us at
[email protected]. No information on this podcast should be construed as investment advice. Securities discussed in the podcast may be holdings of the firms of the hosts or their clients.
PodcastTranscript AI