I’ve said this before, and I’m quite sure I’ll say it again after today: The dominance of mega-cap growth stocks in cap-weighted indexes makes US equities a kind of leveraged long duration bet. And because some of the same stocks which dominate US indexes are likewise dominant in global benchmarks, the same can be true for those gauges.
That’s self-evident to “professionals” (the scare quotes are there to denote that the definition of “professional” is always malleable when it comes to imprecise occupations like predicting the near-term trajectory of asset prices driven in part by human emotions), but it’s not well understood by everyday investors, which is problematic in an environment where persistent inflation and hawkish central banks make it difficult to say where long-end yields are ultimately headed and how policy will evolve.
If inflation stays elevated, for example, and fiscal excess continues, bond yields could stay “high” (relative to the last 15 years). If inflation refuses to abate such that policymakers are satisfied, markets may have to adjust to a new reality wherein the cost of money is nowhere near free. Both of those outcomes are challenging when equity performance depends so heavily on growth shares.
“Equities have been doing much better since the US 10-year bond yield peaked in October last year,” SocGen’s Andrew Lapthorne noted on Monday, before exclaiming, “When equities are expensive, the discount rate seems more important than, say, a recession or a banking crisis!”
The figure above from SocGen shows that nearly half the weight in the MSCI World Index likes falling bond yields. As Lapthorne noted, that nearly matches the TMT bubble. The chart uses rolling correlations to benchmark Treasury yields.
Given the remarkable outperformance of mega-cap tech shares in 2023, it’s worth noting that the better those stocks do relative to other index constituents, the more duration-sensitive big-cap benchmarks generally are — or, more simply, the more dependent the market becomes on falling bond yields.
“The bond/equity correlation is then once again a major issue for equity investors, leaving equity markets vulnerable to a pick-up in bond yields,” Lapthorne went on to say. “Or to put it another way, to better economic news.”
I’ve been very bullish on mega-cap tech the last 6 months, but it’s getting harder to maintain that stance when I can get an easy 5% on my money instead of worrying about when the next shoe might drop. I still maintain that mega-cap has favorable cost dynamics going forward with all the layoffs and generative AI, but generative AI might also undercut some mega-cap business as well, especially Google.
If you haven’t read it already, I’d encourage everyone to read this about Google. Their search business may be most at risk out of the mega-cap titans. https://www.semianalysis.com/p/google-we-have-no-moat-and-neither.
Do you have any thoughts about the big Wall St banks and AI? My thinking is, there’s any number of firms utilizing algo’s that they spent a great deal of time and money developing, employing experts in varied fields to develop these tools……and so it’s an easy jump to the thought that building an AI program that could find a new edge in trading is already well underway. How might this affect equities in general, and which companies are the worth watching for new developments?
This is way outside my area of expertise so I have no idea how much this changes trading or how the algos operate, but here is interesting thread about what this does for investment banking. BloombergGPT will certainly be interesting to watch.
https://twitter.com/palmtreeshinobi/status/1655934201835945984
That was an interesting article, even if other AI folks are pushing back. The possibilities I took from it are
– It may not be possible for the megas to dominate AI as they do search, social media, cloud computing, productivity
– It may be that they can use AI to be better at what they do, but the new things that AI births may be hard to corral and make proprietary
– A lot of the megas domination comes from network effects, but does that apply to AI – if I use one AI tool to do task X, why does that mean you need to use the same AI tool when you do task X
– It may also be that the dream of AI consuming orders of magnitude more silicon every year for endless semi growth will not play out, if smaller models and more efficient chips can do the job with fewer resources
– Useful AI might mean specialized AI, trained on the small fraction of content actually relevant to reading CT scans, analyzing stock patterns, adjusting insurance claims, etc. Specialized could mean smaller models, a fragmented field, no dominant gatekeeper.
Yeah, these are great observations. Initially, I thought this would be another area dominated by the existing titans, but now I’m not so sure. Proprietary data may become very valuable and the tech titans certainly have that in spades, but if quality of data is what really matters and you can’t corral what you’ve got, you’re right that it may break some monopolies. As I said yesterday, a 5% guaranteed return isn’t looking too bad right now while this plays out 🙂
Hey JL –
I’ve seen the idea in your last paragraph talked up in tech circles. It is just not cost-effective to run the mass wide-scale learning processes for all uses. Perhaps not even technologically feasible at the moment. Specialized AI will be a good route to start.
Yeah, if you’re training an inventory management AI on a company’s internal data and processes, an investing AI on stock market data and company financials, a military AI on Red and Blue force weapons and tactics, you probably don’t need a 100 billion or 100 trillion parameter training dataset, and you probably want to run inference on compute resources lighter-weight than AWS cloud – it might have to run locally in the nose of an NGAD fighter aircraft, or in milliseconds in some HF traders’ datacenter.