Lost in translation: LLMs, LMMs and the future of ETFs

Originally published in David’s LinkedIn newsletter, “One Mann’s ETF Opinion.” Subscribe for the latest updates.

I was casting about for possible blog topics recently while attending the annual ICI exchange-traded funds (ETF) conference in Nashville—usually fertile ground for ideas. Tokenization, complex ETFs and market making were candidate topics I was kicking around when I suddenly felt a bit of an existential crisis related to this newsletter. I’ve been noticing recently that many articles begin with three or four artificial intelligence (AI)-generated summary bullets. This distilling of content reminded me that my wife also shared recently that she’s started to drop some podcasts into AI engines that then turn an hour of content into a five-minute summary, available in both written and audio form. Personally, I like to read articles and listen to podcasts in their entirety; however, if most people just want the highlights, then I am potentially wasting my time writing full-length articles when a few bullet points will suffice. 

Naturally, I immediately consulted our marketing team since sending them a couple of sentences would certainly save everyone more time and effort. Unfortunately, they informed me that I had it completely backwards. To gain popularity with AI engines, my content needed to be longer as well as clear and more detailed. This would better grab the attention of the LLMs (large language models). Perfect! I felt humbled that our community of LMMs (lead market makers)—the firms responsible for providing liquidity in ETFs—would be such fans of my content. Furthermore, with more than 5,000 ETFs currently in the US market and over 100 new ETFs launched each month, the strain on LMMs is a real concern within the ETF ecosystem. 

Upon opining aloud about this interpretation, I was caught off guard when they started laughing at me.  How could this be? ETF liquidity is no laughing matter! As it turned out, the reason for writing longer content was not for the benefit of our LMMs but for LLMs. This was disappointing on a personal level (but potentially encouraging on a professional one). The machines may not be quoting me at cocktail parties, but perhaps they are using content like this to better understand/explain ETFs, ETF liquidity and the plumbing that allows the ecosystem to function. I suppose that the optimistic view of such technology is that these machines are using my word to improve the ETF ecosystem. That’s right…yours truly might very well be teaching AI! The student has become the teacher.

Now, my bridge. At first glance, ETFs and AI technology may not seem connected. But the more I thought about it, the more the connection made sense.

AI summaries are essentially a new wrapper around written content. The original article is still the underlying asset. The summary is the representation layer. It is faster, more portable and easier to digest. But it is only useful if it remains connected to something accurate, complete and current underneath.

That sounds a lot like ETFs.

An ETF can turn a complex basket of securities into one ticker. Investors may experience that ETF as a simple trade, but behind the scenes there are holdings, baskets, market makers, authorized participants, creation and redemption mechanics, spreads, premiums and discounts and a lot of other roles involved in ensuring the ETF trades the way investors expect.

The wrapper is simple. The plumbing is not.

Returning to the ICI conference, one of the main recurring themes was how these new technologies are helping improve the ETF ecosystem. Fifteen years ago, “complex” ETFs were funds that held high-yield bonds or bank loans. Now, investors do not even think twice about getting exposure to those securities within the ETF vehicle and instead are asking questions about private credit, digital assets, structured products, swaps and even equity-linked notes. ETF tokenization could completely reimagine the ETF ecosystem as blockchain technologies improve the settlement and trading experience for all market participants. 

Meanwhile, AI and LLMs (the robot kind) are serving as important tools to assist with the next chapter of the ETF story. I would also not be surprised if newer LLMs were helping older LMMs with their important role within the ETF liquidity ecosystem. Maybe that’s a topic for another post.