A Timely Object Lesson on the Dangers of Tight Coupling in Financial Markets, and Hence the Lunacy of Fetishizing Algorithms
FTAlphaville had a fascinating piece this week in which it described a discussion at a CFTC roundtable debating the FTX proposal that is generating so much tumult in DerivativesWorld. In a nutshell, Chris Edmonds of ICE revealed that due to a “technical issue” during the market chaos of March 2020 (which I wrote about in a Journal of Applied Corporate Finance piece) a large market participant was arguably in default to the ICE clearinghouse, but ICE (after consulting with the CEO, i.e., Jeff Sprecher) did not pull the trigger and call a default. Instead, it gave some time for the incipient defaulter to resolve the issue.
This raises an issue that I have written about for going on 15 years–the “tight coupling” of the clearing mechanism, and the acute destabilizing potential thereof. Tightly coupled systems are subject to”normal accidents” (also known as systemic collapses–shitshows): in a tightly coupled system, everything must operate in a tight sequence, and the failure of one piece of the system can cause the collapse of the entire system.
If ICE had acted in a mechanical fashion, and declared a default, the default of a large member could have caused the failure of ICE clearing, which would have had serious consequences for the entire financial system, especially in its COVID-induced febrile state. But ICE had people in the loop, which loosened the coupling and prevented a “normal accident” (i.e., the failure of ICE clearing and perhaps the financial system).
I have a sneaking suspicion that the exact same thing happened with LME during the nickel cluster almost exactly two years after the ICE situation. It is evident that LME uncoupled the entire system–by shutting down trading altogether, apparently suspending some margin calls, and even tearing up trades.
Put differently, it’s a good thing that important elements of the financial system have ways of loosening the coupling when by-the-book (or by-the algorithm) operation would lead to its destruction.
The ICE event was apparently a “technical issue.” Well that’s exactly the point–failures of technology can lead to the collapse of tightly coupled systems. And these failures are ubiquitous: remember the failures of FedWire on 19 October, 1987, which caused huge problems. (Well, you’re probably not old enough to remember. That’s why you need me.)
This issue came up during the FTX roundtable precisely because FTX (and its fanboyz) tout its algorithmic, no-man-in-the-loop operation as its innovation, and its virtue. But that gets it exactly backwards: it is its greatest vulnerability, and its greatest threat to the financial markets more generally. We should be thankful ICE had adults, not algos, in charge.
As I pointed out in my post on FTX in March:
The mechanical means of addressing margin shortfalls on a real time frequency increases the tight coupling on the exchange, and is tailor made to create destabilizing positive feedback loops: prices move a lot leading to margin shortfalls in real time that trigger real time trades that accentuate the price movement. It is like seeding the market with huge numbers of stop orders, which are inherently destabilizing. Further, they can create incentives to manipulate. Anyone who can get some idea of where the stops are can “gun the stops” and trigger big price moves.
It’s particularly remarkable that FTX still is the subject of widespread adulation in light of Terra’s spiraling into the terra firma. As I said in my Luna post, it is lunatic to algorithmize positive feedback (i.e., doom) loops. (You might guess I don’t have a Luna tattoo. Not getting an FTX tattoo either!*)
FTX’s Sam Bankman-Fried is backtracking somewhat:
In the face of the agricultural industry complaints, Bankman-Fried gave ground. While maintaining his position that automated liquidations could prevent bad situations from growing worse, he said the FTX approach was better suited to “digitally settled” contracts — such as those for crypto — than to trades where physical collateral such as wheat or corn is used
Sorry, Sam, but digital settlement vs. physical settlement matters fuck all. (And “physical collateral”? Wut?) And you are deluded if you believe that “automated liquidations” generally prevent bad situations from growing worse. If you think that, you don’t get it, and are a positive threat to the financial markets.
*FTX bought the naming rights for a stadium in Miami. I say only slightly in jest that this is another indication of the dangers posed by FTX and its messianic founder. FFS, you’d think after the 2000 tech meltdown people would recognize that buying naming rights is often a great short selling signal, and a harbinger of future collapse. To say that those who forget the past are condemned to repeat it is too strong, but those who follow in the footsteps of failures that took place before their time betray an an arrogance (or an ignorance) that greatly raises the odds of repeating failure.
Sam – aka SBF – is a man whose Hubris is in urgent search of its Nemesis
Everybody should listen to the notorious OddLots episode featuring SBF being questioned by Matt Levine. It features the hilarious introduction of a magic “Box” the contents of which and their ‘value’ are determined by … something like popular acclaim. It’s the ultimate distillation of Keynes’s depiction of the stock market as beauty parade.
You can’t even call it a numbers game.
As for Nemesis, if you haven’t seen this …https://www.reddit.com/r/wallstreetbets/comments/univjj/saving_capital_markets/
(The Prof is a Boomer who not only lived through Dotcom Bust but the October 19 meltdown. Shellshock came for the traders that day 🙂 )
Comment by Simple Simon — May 29, 2022 @ 10:31 am
@simple Simon. “Boomer who not only lived through Dotcom Bust but the October 19 meltdown. Shellshock came for the traders that day.” Sounds like me. And I’m pretty sure I was closer to the action on Black Monday than he was ;-). Couldn’t be any closer than 30 South Wacker. That was the epicenter. And it’s where I learned first hand about margin doom loops.
Comment by cpirrong — May 29, 2022 @ 2:02 pm
So why not practice one of those risk transformations we’ve all read about? Algorithmically build in negative feedback (circuit breakers, liquidity turnstiles, etc.), leading to more-frequent but less-consequential freeze-ups as the algorithmic bounds get approached and traders react. No (mythical) guarantee of continuous trading, but no risk of various runaways, either.
Comment by SRP — May 29, 2022 @ 2:28 pm
It is something of an adage in computer programming that you can’t automate something until you really, really understand it. The decade-long hype-despair cycles in the AI world, for example, came from the fact that we don’t know what human intelligence actually is. Early attempts among top-notch cypherpunk coders to implement Wei Dai’s B-Money proposal similarly failed until Nakamoto, who likely came from a financial background and was not a top-notch coder, merely a competent one, put the pieces together. Today there are thoughtful analysts in the crypto world, who, in the aftermath of the 2017 ICO scamfest, have set themselves to the yeoman work of understanding how traditional contract arrangements for tech-startups actually work so that, if the same obligations and incentives can be expressed in code, the world of startup investing can be democratized, with startups (and smaller investors) bypassing the monopoly rents of Sand Hill VC firms.
There is always pressure, of course, to run with the code you have, and I suppose we have to experience failure from time to time to see the flaws in our coded implementations. As sad as the TerraUSD debacle was, it is at least heartening that even insiders saw the need to start collateralizing with Bitcoin *before* the crash…not soon enough, of course, and it should have been apparent from the start, but hey, five years ago the debate over a crash like that would still be stuck-on-stupid “tulip bulbs” vs “to the moon”, so this is progress.
Comment by M. Rad. — May 30, 2022 @ 8:00 am
Breaking news (which you may have seen): Elliott Associates will sue the LME for $456 million, for its decision to protect one of its large Chinese clients by shutting down trading and cancelling trades after the recent short squeeze.
Comment by Ex-Global Super-Regulator on Lunch Break — June 6, 2022 @ 7:49 pm
And now we have the Financial Times report about ICE making a (human) judgment call not to declare a default by Citi in March 2020 due to a missed margin payment. The ICE official said declaring default “would’ve been cataclysmic”.
I’d love to see your comments in a post.
Comment by Dave Mason — June 10, 2022 @ 1:18 pm
@Dave–That’s exactly what the post is about. I lead with that.
Comment by cpirrong — June 16, 2022 @ 6:43 pm