To receive a free full-text email of The Zeitgeist whenever we publish to the website, please sign up here. You’ll get two or three of these emails every week, and your email will not be shared with anyone. Ever.
It took me four meetings to realize what was happening.
Sometimes I’m a little slow on the uptake. But I get there eventually.
You see, that fourth meeting was the very moment that I converted to the church of Epsilon Theory. It took place a good four years before the first words Ben ever published under that label and put to words what I had felt for quite a while. But on a dime, it changed my questions, my due diligence process and my concept of the set of behaviors which could even conceivably produce true idiosyncratic alpha.
My conversion on the Road to Reykjavík, if you will, took place during my time covering external equity and macro managers for a large public pension. I was in New York, as I often was, making the rounds with existing, prospective and emerging investment managers, both long-only and long-short. Five meetings a day for five days. At least a few dinners. When I referred to the fourth meeting, what I meant was the fourth meeting out of ten or so in which precisely the same observation was presented to me in almost precisely the same language:
“The right way to think about aluminum is as a mechanism for storing and profiting from access to low-cost energy.”
The logic here obviously isn’t earth-shattering. Aluminum production is notoriously energy-intensive. You haul in bauxite from Australia, crush it, and throw it in the industrial equivalent of a pressure cooker with lye at around 350 degrees. You filter it and seed it with aluminum hydroxide crystals so that larger crystals can form as the disgusting aluminum oxide slurry cools. The real problem comes when you have to turn that aluminum oxide into aluminum. The former’s melting point is prohibitively high – think like 3,700 degrees, about twice as hot as the actual flame in your average charcoal grill – but there are some fancy workarounds that permit electrolytic extraction at a much more reasonable 1,700 degrees. Still, when the process is considered as a whole, aluminum remains very energy intensive to produce. That’s one of the reasons aluminum production has so often been attached to hydroelectric and geothermal energy sources.
It is an interesting factoid, and it is fun to learn how much of Iceland’s power production, for example, has historically been devoted to refining aluminum. Look it up. It’s an insane amount. But this didn’t stick in my head because these four people had the same perfunctory observation to make about the components of margins for a metal refiner. It stuck in my head because they used the exact same, odd linguistic construction for characterizing and describing it at roughly the same time. All of this was brought to my mind, as it happens, by one of the stories that rose to the top of today’s Zeitgeist.
Now, when I got back home, I searched through recent sell-side research for this language. Nothing. Maybe there’s a relationship between these individuals? Maybe it’s just the usual idea circuit? But I couldn’t find any connections between the PM’s backgrounds. What’s more, three of these meetings were with equity managers. One was with a discretionary global macro fund. The context of the observation related to different securities in each case. I’d characterize three as treating the observation as a novel research-based driver of a long thesis, and one as a novel research-based driver of a short thesis. This wasn’t your classic case of the emergence of a crowded trade.
Instead, what turned up was a series of three related articles from major financial publications in the month prior, each of which conjured some variant of the above language.
I came away with three strong, if loosely held, beliefs. Each forms a part of our current views on the proper use of natural language processing in investment applications, and a big part of what we think most shops are getting wrong as they explore these questions:
- Narrative is not (just) sentiment. Nearly all present applications of NLP to investment management treat sentiment detection as a primary – if not exclusive – aim. Narrative has explanatory structure independent of the affect of language used in it.
- Narrative is not (just) crowded ideas. Decision-making happens at the margin, and common knowledge drives second- and third-order decisions. Conflating narrative with an expectation of lockstep first-degree thinking from those who hear its associated missionary statements is wrong.
- Narrative is not (just) idea propagation. Most scraping, data-driven, NLP and sentiment-based models in the investment world have become heavily tilted toward a belief that social media’s reach has long since eclipsed that of traditional media. We agree. The demons agree (and tremble). Everyone agrees. But here’s the problem: reach isn’t the same as common knowledge. Except perhaps for the tweeter-in-chief, there is still no social media account in the world which everyone can assume that everyone else has seen. In politics and finance, we think many of you are discounting the power of missionaries far too much.
Of course, Ben had made all these leaps in the political world years before. It formed the core of his dissertation and the book that followed it. We all have our personal Road to Reykjavík. I’m sure there are more than a few members of the pack with a similar story, too.