Five Easy Pieces:
- Magnificent 7 stocks keep rolling, driving market concentration to its highest level since 1972
- The improving performance of free open source large language models and implications for closed model revenue moats
- The No Labels movement risks triggering a 12th Amendment Contingent Election if it wins electors in a few states
- The Armageddonists: an update on one of my guilty pleasures
- Bottom-fishing in Chinese equities and parallels to the 2008 TARP bill in the US
Watch the Podcast
Good morning, everybody. Welcome to the February 2024 Eye on the Market podcast. This one's called "Five Easy Pieces." It was a movie from 1970 with Jack Nicholson where he plays an oil rig worker who was also a former classical pianist. So in Hollywood, anything is possible.
There's five topics I wanted to talk briefly about that we wrote about in the Eye on the Market this week. One of them has to do with the unending dominance of the Magnificent 7 stocks. I would like to talk a little bit about-- on a related topic on open-source large language models given that NVIDIA is such a huge part of the Mag 7 right now; a quick follow-up on the No Labels movement, where I got into some debates with some clients at some conference recently; an Armageddonist update, which I'll explain; and then some comments on bottom fishing and Chinese equities.
So let's start with a discussion of the Mag 7 stocks which, of course, you all know at this point are completely dominating equity markets. Last year they returned 76%. The rest of the market returned 14%. But I think it's important to keep in mind a couple of things.
First, unlike 2000, 2001, these stocks are making a lot of money. Their sales growth is seven times higher than the rest of the market. Their margins were expanding at the end of last year instead of contracting. Their margins are three times-- almost three times higher than the rest of the market.
So it's not a profitless boom. The bigger concern is that it's a profit-oriented boom and that these companies are increasingly dominating not just market capitalization, but income as well. So that's why we spent as much time as we did in the outlook thinking about antitrust issues. Because that's really the only thing on the horizon that I think can seriously dent the overall Mag 7 story.
Tesla's run into a buzz saw that's very specific to it recently, mostly related to pricing competition in Europe, the US, and in China. And also, note that their fourth quarter earnings were flattered by a one-time non-cash tax allowance adjustment that I explained in the Eye on the Market.
But, in any case, the Mag 7 story just keeps rolling. And we have a chart in here from some of my colleagues in the investment bank who do excellent research on how the market concentration has now reached the highest level since 1972. They don't think that's a great thing. They note that historically, surges in market concentration have either preceded or coincided prior recessions. So they're not agnostic about this.
One thing I know for sure is that market concentration makes life a lot more difficult for active equity managers. Last year was one of the worst alpha years on record in terms of excess returns. Only 23% of managers managing against the Russell 1000 type index outperformed. And that compares to almost triple that level-- 66%-- in 2022. So this kind of stock market concentration stuff is very tough generally on the active management industry.
Again, this is quite different than the bubble that took place in 2000, 2001. And the big risks here are whether it's the Biden administration or a future Trump administration or whoever, a political reaction to this kind of earnings concentration in addition to market concentration.
Now, with the AI revolution being at the core of Mag 7 outperformance, I wanted to just talk a little bit about this report that came out quietly from Microsoft, of all people, last year. So Microsoft, as everybody knows, is heavily invested in the success of OpenAI. They're also doing their own work with open-source language models. And this raises a lot of questions about the monetization of large language models, the big closed ones, whether it's OpenAI or Google's version or anything else.
So what Microsoft did-- and this is not a typo. They took an open-source model from Meta that Meta had released called LLaMA. They adapted it in ways we describe in the piece. And then they went to see how it would do on questions related to biomedicine, finance, and law, compared to some of the big highly trained private closed models.
So one example in finance, they compared it to Bloomberg GPT, which took like a billion computing hours to create, $1 million to create. I think Microsoft created-- Microsoft spent less than $100 reportedly on their version of an open-source model. And, as you can see here in the chart and was discussed the piece, the performance was roughly the same in biomedicine, finance, and law as the big, expensive, complicated closed-source models.
The open-source models require work to integrate. You need a lot of programmers to kind of figure out what to do with them. But once you have that in source talent, you have greater transparency. You have more version control. You can use whatever servers and cloud providers you want instead of the ones required by the closed models.
You have less exposure to the business issues, like the shenanigans that took place at OpenAI this year. You don't have to share your private data with people who own that model who might censor what you do with it. And you could even run some of these open-source models on a single GPU or even a MacBook instead of a huge GPU cluster.
So this was a quiet paper. Microsoft didn't make a lot of fanfare about it. But to me, it does raise questions about the profitability of some of these large language model efforts.
And also, there's a guy that I talk to a lot about this kind of stuff who will remain nameless. But he does tell me the golden age of large language models in some ways is already over. And I said, What do you mean by that?
And he said, for the last few years, language models have been able to surreptitiously just scrape all the data without the people who own it knowing, repackaging it, and selling it, and calling it research. And he said, that golden age is over.
So there are a lot of interesting use cases for large language models. But I'm very curious to see the follow-through in terms of monetization and then what impact that would have on some of the premiums priced into some of these stocks.
OK. So another thing, interestingly, that happened was I was speaking at a conference in Utah about this weird triple witching hour scenario, where the No Labels party in the United States runs a slate, which they say they may do. And, let's say, they win four or five states, and they win enough delegates to prevent both Trump and Biden, presuming they win their respective nominations, from reaching 270.
So I started to talk about what would happen if no candidate passed 270, and the rules around contingent elections that go into the House of Representatives to pick the president. And a hand was raised in the front of the audience from a CEO who is actually so active in the No Labels movement that their email handle is-- it's nolabels.com
And this person objected to my line of thinking by saying, Well, we would never let it get to a contingent election. We would do some horse trading to form a unity government before a contingent election took place. In other words, we would throw our electoral support to either Trump or Biden in some kind of negotiated, coherent way. Maybe.
This is real complicated. The US system is not set up for that kind of horse trading. You can do it at a convention, but you can't do it after the general election. At least, I don't think so. And there's four major reasons why.
First, No Labels can't force their electors to switch votes. They can decide whatever it is they want to do. Now, they can they can suggest they do it. They can try to compel them to do it, but they cannot control them. And those electors, should they win any, would be able to do whatever they want.
Number two-- around 2/3 of states have actual elector binding laws in place that expressly prohibit electors from switching their votes. And even in the states that don't have them, I could imagine an avalanche of constitutional challenges from voters saying they were disenfranchised. So unless somehow the states that No Labels win happen to be states that have absolutely no elector binding laws and that all constitutional challenges fail, I see that as a huge hurdle.
Two other things-- even if they get past that point, on January 6, there's an Electoral Count Reform Act that requires electors to be faithfully given. And then this is a term that's long existed in the constitutional law community for a couple of 100 years.
But the bottom line is a faithless elector is somebody that votes for a candidate that's different than the one that they were allocated to based on the general election results. And you could imagine that Congress, on January 6, would basically reject some of those No Labels electors that switch parties.
And then the last thing is the No Labels people have been very kind of vocal about, Well, we would throw our support, depending upon a unity government where we get certain cabinet posts and this and that. There are laws, federal know and criminal laws against-- they'd have to avoid violating by horse trading their support for their electoral votes.
So I think this is a huge gauntlet here. And in the weird, triple witching scenario, triple witching hour scenario, where No Labels wins enough electors to prevent Trump and Biden from reaching 270, I think the higher probability is that you would end up with a 12th Amendment contingent election in the House. So you can read more about that if you're interested. I enjoy that kind of stuff.
One of the other things I enjoy, and one of my guilty pleasures-- everybody has guilty pleasures. Mine is rather benign, in addition to fishing, is I like to look at the consequences of Armageddonism, which is the media tends to flock to people who have the most horrifyingly, terrifyingly calamitous things to say.
Now, there's been a lot of books written on behavioral human instinct towards bad news rather than good news. And a bunch of newspapers and magazines have historically conducted experiments where all they do is put good news on the cover, and their newsstand sales go down by 2/3. So the media loves quoting a bunch of these people, and we show the names and the chart. And you can-- you'll probably recognize a bunch of these doomsayers.
And so we looked-- at the end of 2019, we looked at their forecasts and we said, Well, what if on the day of their prediction of disaster we switched $1 from stocks into bonds, by taking the dollar from the S&P 500 and moving it to the Barclays aggregate. And by the end of 2019, you would have lost somewhere between 30% and 60% by listening to these statements which were made between, let's say, 2010 and 2016.
So COVID hits. This is amazing. COVID hits. The market collapses. And now, the Armageddonists think, OK, I've been bailed out by a global pandemic which I didn't predict. And so then after the markets had already declined by 20%, 30%, they doubled down with some of these hilarious quotes about we're heading-- this is going to be the worst bear market in my lifetime.
This is a deep depression. I expect the S&P to lose 2/3 of its value over the coming years. And then-- so we have a chart in here where we plot these-- where we plot the timing of these statements against what happened in the market which, of course, has almost doubled since a lot of these statements were made.
You know, this is all fun and games, but it is-- it's a reminder that the timing of investing is important. There can be deep corrections in markets for different reasons. But usually, the worst time to double down on them is after the market's already gone down.
Now, one place where Armageddon is happening is if you're an investor in Chinese equities, which has over the last couple of years been quite the train wreck. And there's lots of different equity indices in China. Most of them are doing roughly the same kind of thing.
There was a spike in trading volumes recently. And a lot of times a spike in trading volumes tends to coincide with the bottom in a market because you get kind of seller capitulation. So we're looking at that. But there's some bigger picture issues that I just want to talk about for a minute.
So we have a chart showing that the P/E multiple on China has gone down to around 10 times, which is pretty low, particularly compared to valuations in the developed world. But I just want to close with two things.
First, China at a 10 P/E looks cheap. But there's a lot of things that trade at below a 10 P/E. And we have this giant bar chart in here that looks at them. So if you're value hunting, European Energy, US Energy, S&P 500 banks, Asia-Pacific Energy, S&P 500 Telecom, a small cap telecom, Asia-Pacific Utilities, European Financials, Brazil, Italy, Poland, Austria, Hungary, Turkey-- all of these things trade below a 10 P/E. So you know if you're interested in bottom fishing, China's not the only place to look.
And then the other thing I want to close with is just some comments on China bottom fishing, and then what to watch for and the TARP bill in 2008. So in 2008, I wrote a piece on October 14 talking about how we had become very bullish on US equities at that point in time. Why?
And up until then, the government was dealing with a crisis of solvency perceptions on the banks. And their first plan was to buy all the bad loans from the banks, put them in a bad bank, work them out over time, and the losses would basically accrue to taxpayers who would be funding the acquisition.
The markets didn't respond to that. And I talked to some people at the IMF who had done this study showing that in prior decades, if you have a banking crisis of some kind, or a solvency crisis of some kind, buying bad loans doesn't work well in terms of boosting GDP and the equity market.
But when the government steps in and buys the liabilities and equity positions of the banks, you get a kind of a durable recovery and confidence growth in the stock market. So the TARP bill, which failed the first time it went up for a vote, but then passed. When TARP passed, we became more confident that we had hit a bottom. And we had another bottom test in March, 2009. But buying in October of 2008 would have made you a lot of money over the next two, three, five years.
So my conclusion on China is they've had this monstrous real estate bubble. The value of the housing stock to personal spending is one example. It's three times higher than the US at its peak. Home price to income ratio is four times higher. This was a crazy bubble in China.
And so if we got to the point where China did what the US government did in 2008, which is to say, OK, we're going to attack the heart of this problem, or we're going to recapitalize not just the regular banks, but all those shadow banks as well-- a lot of the larger ones have been failing recently-- that would give me more confidence that China was an investment rather than a trade.
Yet to try to bounce 10% or 15%, go up and down based on some monetary policy announcements-- maybe. But to really get bullish on China for the longer term, I would need to see an aggressive recapitalization of the banking system, both the regular one and the shadow one. And certainly not strong-arming domestic mutual funds into buying shares and going after short sellers.
So that's my take on Chinese bottom fishing and a bunch of other things. Thank you for listening. Our next Eye on the Market paper will be in early March and will be our annual energy piece. And this year it's called "Electrovision," for reasons you might be able to imagine. So see you next time.