Shouldn’t stock market record highs be commonplace?

Factual Questions
21

By way of elaboration, the S&P 500 components currently make up 83% of the Wilshire 5000, the preferred index covering large, medium, and small capitalization stocks (but excluding penny stocks and very small companies). I’m not sure I would call members of the Wilshire 4500 major, but it does include companies such as Door Dash, Cloudfare, Coinbase, and a lot of companies I have never heard of such as Middleby Corp. or UFP Industries Inc.

Since the 4500 is positively correlated with the 500, the S&P 500 is a decent proxy for the overall stock market, even if it covers exclusively large capitalization stocks. The Dow is a terrible measure because it covers only 30 stocks chosen arbitrarily and weighted by price rather than market capitalization. (Admittedly the Dow was convenient during the age of slide rulers.) The 500 and 5000 are poor measures of the economic health of the country (for that, look at the unemployment rate and productivity growth) but they do measure the overall stock market well.

22

Oh, I agree. I’m just saying that in the context of the recent stock price performance of other companies in the space based on the perceived long term potential of AI, AAPL going nowhere has pretty much been the equivalent of dropping sharply in any other context. It’s already kind of discounting them not being at the bleeding edge. If they don’t show something credible in AI this year, I don’t think the market will be much more patient.

23

To be fair to Apple, they have had AI accelerator processors in their phone’s since 2019. Apple have designed their own processors for some time now. Whilst Samsung might have a dedicated chip, Apple have been ahead of the game and integrated the capability right into the SoC.

IMHO, a lot of people are going to take a bath over the AI hype train. It exemplifies the problem with the market that has been with us for quite some time. There is too much money chasing too few next big things. Maybe there is a killer use case that the general public are going to be prepared to pay real money for. But its very unlikely to be the next iPhone level product. This massive inflation of the value of next big thing companies can only be viewed as a market distortion. And a worrying one.

24

The problem I find at the moment is that there are huge potential shocks everywhere.

We have a potential Trump reelection in the US. The immediate problem with that is more social, but people are dreaming if they think dismantling democracy and the rule of law will be just hunky dory for business.

In Europe we could have the Putin war expanding.

In Asia, Xi deciding his legacy is to reunify China and fuck the consequences.

But you can’t just cash out, partly on general principles that trying to time the market is a mug’s game, none of this may happen, and because if you’re out of stocks forever inflation may kill you.

I’m just trying to stay ultra-diversified, and I really don’t feel Google at a forward P/E of 21 or Amazon where it’s possible just AWS may be worth most of the current market cap are the most risky things out there that I need to worry about.

25

If this is a bubble, I’ve seen a lot dumber ones. The magnificent seven are profitable and in an industry growing faster than the remainder of the economy. Interest rates will fall over the next year or so. AI has game-changing potential, although whether LLM models will plateau is unclear. It’s also uncertain who will vacuum up the surplus – suppliers, downstream companies, or consumers – but NVIDIA will surely take a nice fraction of investment spending, as will cloud providers.

NVIDIA has a PE ratio of 66 - hefty, relative to the SP500’s average of 23. But consider that their earnings per share grew almost 7 fold last year, and demand is expected to pick up moving forwards. This month Sam Altman of OpenAI made the rounds hoping to raise 7 trillion dollars of funding. Even half of that is an insane amount of money for an investment project: a cutting edge chip fabrication plant might have total cost of operations over 10 years of 35-43 billion. $7 trillion would buy you 162 of those plants, assuming the $43 billion figure.

This isn’t investment advice: I just poked around the internet for 10 minutes. Everything could go sour. I’m just saying this isn’t as stupid as pets.com, packaged liar loans, or any other antecedent of random white-swan financial crises occurring once or twice per decade.

Agree with Riemann that the geopolitics are ugly and that our CEOs are hothouse flowers wholly unprepared for an environment of strongman cronyism, like that imposed on the people of Russia.

26

I think I’m more neutral-to-positive than that. AI doesn’t need to realize superintelligence to transform the economy drastically. It seems to me that the potential outcome is so asymmetric that the price can fairly reflect expected value taken across all outcomes even if it’s 75% certain than the stocks are overvalued. Optionality, essentially.

The only thing I’ve tweaked is that I’m closer to equal-weighted across the 6 (excluding Tesla) than the market cap weighting that purely passive indexing would give me.

27

I’m having a hard time figuring out the AI space. So far, most of the uses of it I’ve seen out in the wild have been poor - online magazines trying to replace writers with AI, Air Canada trying to use an AI for flight bookings, etc. None have been real successes.

And yet, when I think about my own job before I retired, I can see massive possible productivity improvements. My wife says the same thing. So a lot of the value is hidden in back offices and vertical markets. It might be that AI companies per se don’t become massively valuable, but all companies increase in value due to efficiency gains.

Some analysts think the big foundation models (ChatGPT, Gemini, Claude, Grok, etc) will have a value of zero. Open source is too strong and the models may just converge in ability and become commodities and sell for little more than their compute cost.

I generally approach bubbles on the gold rush theory - the people who got rich in the gold rush weren’t the miners, but the people who sold goods and services to the miners. So I’d be looking at chip manufacturers as a better investment. Maybe cloud providers, but I’m not sure AI will remain mostly in the cloud. And I did invest in AMD a few years ago, for full disclosure.

But here’s the thing: LLMs are coming out which need much less compute for the same quality of output, and new chip manufacturers like groq are getting into the game. So I don’t know how much longer Nvidia can stay on top. Chips might be pcome commodities as well.

It might be ,that even if the big cloud based LLMs are ultimately smarter and better, the difference may not be worth it for 90% of AI tasks and most LLMs will be embedded in devices.

The real value may lie in the training data. That’s the only thing that’s really proprietary in the whole AI stack. So I’d invest in the compamies that have a huge database of proprietary data. Alphabet, Meta, Twitter, Tiktok, etc.

Anyway, back to the OP. A picture is worth a thousand words. Have a look at this graph of the Dow. The trend is clearly up, but look how often you would be down from the last new high if you had invested in it. Sometimes there isn’t another new high for decades. But even in normal times, new highs in total value are rare.

https://www.macrotrends.net/1319/dow-jones-100-year-historical-chart

As an aside, look how important timing is. If you invested in the market in 1928, you would not get back to breakeven for 30 years! If you invested in 1932 you would have quadrupled your money in four years.

28

New Scientist magazine has an extensive section of articles on new developments in science. AI is now being mentioned with great frequency, almost always in the context that it has found things that were unknown or unfeasible for manual searches. AI is already transforming all the sciences and that means that it will transform all of our futures.

However, the term AI is also already problematic. It’s impossible to know whether the AI used to find new protein foldings has the same set of meanings as the AI used to generate the most handsome men in each country. I’m old enough to remember when the term “computer” was responsible for aiding scientists in breakthroughs. Terms get fudged by journalists who need to convey procedures to a lay audience.

Which companies use which types of AI for which advances will guide future value. What won’t change will be the need to dig beneath surface articles to fully understand what companies are doing, a need that has existed since corporations were invented.

29

We should move this discussion to an AI theead. It’s a good one.

30

[Moderating]
Remember what forum you’re in.

31

It occurs to me that the rate at which new record highs occur should be a function both of how rapidly the market is growing, and of how much random fluctuation there is in the market. A market that’s growing only very slowly, but with zero noise, would always be at a new record high. On the other hand, a high noise could mask the effect of underlying growth for a long time. Has anyone studied how noisy the stock market is over time, or how noisy it is right now compared to historically-typical values?

32

Take a closer look at your chart. New highs occur all the time - but they are supplanted by newer new highs. As an example consider the SP500 at FRED:

From Aug 2020 to December 2021, new highs were a regular occurrence. Same for April 2019 to Feb 2020. At the Dow chart, Aug 1995 to Dec 1999 had new high after new high. Lots of new highs between Dec 2013 and October 2021.

There’s a ton of study on that, because there are financial instruments that trade on market volatility. One interesting result is that while the stock price is roughly a random walk (daily returns are not serially correlated), volatility itself is serially correlated. So a big jump up or down today makes a big jump tomorrow more likely - it’s just that the direction isn’t clear. Financial professionals trade stock options on the basis of these models.

FWIW near term stock volatility, as reflected in options pricing, is below average:

Discussion:

Also daily stock returns aren’t normally distributed: the distributions have longer tails (leptokurtotic if you want the statistical term - leap like a kangaroo). Over-reliance on the normal distribution led to a lot of red ink during past financial crises, along with reporting that claimed that you should only see such events every x thousand years. Erm, no: that applies only if you assume that particular distribution against all historical experience.

33

If you invested $1,000 at the start of the chart, how often would your investment be at a new absolute high? There are lots of local peaks, but the absolute high moving up is more rare.

For example, at the very beginning your investment goes up in the first quarter, then down a bit, then up, and up again. But then your investment does not get larger again for almost 12 years - 46 or 47 quarters. After that, I get 10 investment peaks in about 20 quarters or so during a big bull market. But then… your investment craters in 1929 and doesn’t return to its peak until 1959, or roughly 120 quarters. That’s 13 times your bankroll hit a new high out of about 170 trials. All the rest of the time, your investment was below its peak.

If we look at the second half of the century to now, your bankroll would have gone up and down until 1964, with a few more absolute peaks. But then more disaster, and your investment doesn’t recover to its 1964 peak until August 1995…

I lived this when I was a gambler. Even when you are steadily making money, you always feel like you are losing because of the way the variance works.

34

This is not true. See this list of the S&P 500 stocks by performance year-to-date. Actually, there’s 503 of them. 302 are up, 201 are down. More than 100 are up by more than 10 %, whereas only 44 are down by more than 10 %.

Even if your claim were true as a matter of fact, it wouldn’t go to prove much. If you eliminate the best-performing members of a mixed set, then of course the remaining set will perform worse. That’s not because the remaining set is what “really matters”, it’s simply because you’re ignoring the top performers.

35

Wouldn’t you expect that, whenever you hit a new high, the very next day you’d be 50% likely to hit an even higher new high?

36

And in this case the set members are not independent. If the top performers didn’t exist, much of the investment dollars would move to other stocks.

37

I got bored and made the calculation. For the S&P 500 between 1/3/1950 and 2/16/2022, a new nominal high was reached on 7.64% of all trading days. I assume they are clustered together, but that would be .0764*252 trading days = 19.2 days per year on average. Calculated with Stata.

38

And on the other hand, if the top performers didn’t exist, some of the other stocks might not be able to exist, either (or exist at a much lower level). Both of the major cell phone OSes (and the third-place distant runner up) are made by those top 7, for instance: What would an independent company that makes phone apps look like without them?

39

Those concerned about magnificent 7 overvaluation could consider the SP500 Equally Weighted Index. You can get an ETF based on it, though annual expenses will total to 0.20% rather than about 0.04% - five times higher, but much less than any actively managed fund.

It has underperformed for the past 2 years (duh), the past 5 years, but has outperformed over the past 20 years.

This is not investment advice and I don’t own any of the SP500 Equally Weighted Index. I do have a pronounced exposure to the SP500, my favorite, as well as some in the Wilshire 5000.

40

Yeah. And that’s close enough to the 5% number mentioned before, accounting for variance. I know, it’s surprisingly less than you’d think.