|
*WARNING: those allergic to middle school mathematics may want to jump down to the possum drawings..
Is the S&P 500 ripe for a rally? A meltdown? Let’s ask our friendly neighborhood scattergram for some guidance. (For those of you with better things to occupy your thoughts, a scattergram is a graph that simply shows the relationship between two sets of numbers.) In this case we are using a scattergram that plots annual forward P/E ratios on the horizontal axis - or X axis as the math nerds call it - and the subsequent 10-year annualized returns of the S&P 500 on the vertical (Y) axis. In other words, given the forward P/E ratio of the S&P 500 today, what should we expect for an annual S&P 500 return, on average, for each of the next 10 years?
| | |
If any of you are having an algebra flashback looking at y = -0.0146x + 0.3203, I feel your pain. That’s the slope-intercept equation that you’ve been trying to forget since 7th grade, although back then we knew it as y=mx+b. In this case it represents what we call the best fit line, or the line that best represents the relationship between our x and y variables.
So, with that best fit line providing all the fortune telling ammunition we need, we can confidently plop an orange dot down for the known x variable – forward P/E ratio as of December, 2025 – and know the future.
| | |
The news isn’t great for the S&P 500: a forward P/E of about 22.5 to start the year indicates a compound annual growth rate of about -1% per year for the next decade. For those of us who aren’t excited about the possibility of taking a decade to turn your $1.00 into $0.90, this is bad news.
A short detour with more math follows. Feel free to ignore this paragraph. Let’s say you have two sets of data, each with four numbers representing four outcomes. The first set consists of 100, 0, 100, and 0. The second set consists of 48,49, 51, and 52. Both sets have the same average: 50. But one set of data probably gives the impression it would be better at predicting what might happen next than the other set. The 1st set, while averaging 50, has none of its numbers anywhere near that average. In the second set, all four data points are clustered around 50. It turns out there’s a measure, R2 that tells us how predictive our y=mx+b. equation is. From our original graph, we can see R2 = 0.7293, meaning the Forward P/E of the S&P 500 is 72.93% accurate at predicting the following 10-year average return. We now return you to your regularly scheduled programming.
There is, however, a silver lining as, over the last couple of years, the S&P 500 has effectively split: the Magnificent 7 (Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla.) and the Other 493. Let’s look at the two in a little more detail.
7 vs 493
As of December 31st, 2025, the S&P 500, which encompasses about 75% of the country’s total market value, was worth $58.44 Trillion. The Magnificent 7 is around $21.7 Trillion of that total. Graphically, it’s something like this:
| | That would be fine and dandy if the Magnificent 7 contributed a proportionate amount of the total earnings – a.k.a. profit – of the S&P 500. They don’t. | | |
Note: the above chart measures the percentage of Market Capitalization and Earnings contributed by the 10 largest components of the 10 largest companies of the S&P 500, not just the Magnificent 7. (The additional three companies are Broadcom, JPMorgan Chase, and Berkshire Hathaway.) However, the point still stands: a small handful of companies make up a disproportionately large percentage of the value of the market but don’t come anywhere close to earning enough money to justify their valuations.
Elevated Valuations Justified?
Of course, to many, elevated valuations in the AI space is more than justified by the expected earnings growth in the future. For those earnings to materialize, AI will need to prove itself a worthwhile investment for businesses. But Torsten Slok of Apollo Capital recently noted, “According to a Duke CFO survey, most CFOs are seeing no impact from AI on labor productivity, decision-making speed, customer satisfaction or time spent on value-added tasks.” Chart below.
| | |
And, just to be clear, valuations are more than elevated. Take Nvidia, the manufacturer of AI’s picks and shovels, the chips that make the AI engine run. As mentioned in a prior note, the market capitalization of Nvidia ($4.5 trillion) is smaller than the GDP of only the US ($30.5), China (19.2) and Germany ($4.7). It is bigger than every other country on the planet.
But…
Is it possible that AI doesn’t – and won’t - work in the way so many are hoping? Is Artificial General Intelligence (AGI) still just a pipedream? I for one think AI is useful for mundane data gathering tasks but is almost entirely useless for evaluating even the most basic of questions. To illustrate the point, let’s turn to the humble possum.
In a series of articles, Gary Smith, the recently retired Chair of the School of Economics at Pomona College, gave various AI platforms the prompt, “Draw me a picture of a possum with five body parts labeled.”
Open AI’s Chat GPT-5 produced this:
| | |
The possum itself looks rather possum like, so there’s that. I suppose getting two of the five labels correct is better than getting them all wrong.
Google’s Gemini answered the prompt this way:
| | |
This is definitely better than the Chat GTP-5 response.
However, a day later Gemini gave a very different answer to the exact same prompt:
| | So, a very different response, albeit with six labels instead of the requested five. So, to throw a curveball, the prompt was then changed to ask for six labels instead of five. That was, apparently a bridge too far as Gemini lost its – supposed – mind: | | I kind of like that of the two “grasping hind feet” labels, one points to the nose and one to a front foot. There are many more examples, but I think the point is made. I’ll leave the possum drawing discussion with my favorite unhinged possum-labeling-body-parts prompt: | | |
If you’d like to read more on the inability of AI to respond accurately to basic prompts, this article may be of interest. It explores Chat GPT’s assessment of how a game of Tic-Tac-Toe is altered when, before the game starts, the 3x3 Tic-Tac-Toe grid is rotated 90° to the right.
And, if I may, a brief sidebar regarding internet traffic. Since the launch of Chat GPT, the first widely available AI platform, in late 2022, AI generated traffic on the internet has ballooned from a tiny fraction of total internet traffic to half of all internet traffic. You can read the report here. With that much of the internet’s traffic being questionable at best, it’s not surprising that Merriam Webster’s word of the year for 2025 was AI Slop.
| | |
So, why all the picking on AI stocks and the flaws of AI in general? Because, as I mentioned at the top, when something is inevitable, i.e. an Ai stock meltdown, that doesn’t mean its imminent. This mania may well stretch out through the year… or it might start this week. Time will tell. In the meantime, we’ll continue with our strategy of owning modestly valued companies with good management, well managed levels of debt and , in many cases, healthy dividends. In addition, we’ll continue to hold larger than usual levels of cash and T-Bills.
Just before the dot.com bubble burst in March of 2000, Cisco Systems was the world’s largest company. They made – and still make – the switches and routers that made the internet work. The picks and shovels if you will. That didn’t save the stock as it fell from a high of $51.80/share in March 2000 down to $7.25/share in September of 2001. A while after it’s huge sell off, we started buying CSCO as the company still had a very profitable business model and was selling at a bargain price. We have owned the stock for many years now and it has been a solid, if unspectacular, performer. Who knows, maybe one day we’ll pick up some Nvidia as well.
Office Update
Juliet passed her Uniform Investment Adviser Law exam, otherwise known as the Series 65, in December. Next up are CFA exams that will stretch out over a few years. She’s been working with me to get our Black Diamond platform, a replacement for Morningstar, up and running. It has been quite a chore as 2026 marks our 25th year in business. It turns out that 25 years of financial data is quite a bit. We should be unveiling our new client portal this coming week.
In other office news our resident bull, Fernando, suffered a catastrophic fall from the windowsill and sustained a gruesome injury. Look away if you’re squeamish.
| | |
We have Dr. Krazy Glue on the case and expect a full recovery.
Juliet blames the office bear for pushing Fernando and has gone to AI to manufacture evidence.
The accused:
| | |
I’m not sure it would hold up in court, but it’s sure compelling.
As always, thank you for entrusting us with managing your hard-earned assets. We’re looking forward to a challenging year and look forward to your questions, jokes, stories and life updates.
Happy New Year,
Bob McClure
| | | | |
