Streetwise Professor

November 27, 2019

Back to the Future: Exchanges That Price Other Things Don’t Price Their Own Things Right

Filed under: Derivatives,Economics,Exchanges — cpirrong @ 7:54 pm

The first post I wrote almost 14 years ago was about an electronic trading system (in Japan) being overwhelmed by a avalanche of orders. The main conclusions I drew were: (a) since communications capacity is costly, it is uneconomic to build so much capacity as to make such outages impossible, and (b) pricing of capacity is the best way to ensure that it is utilized efficiently.

What is old is new again. A few weeks ago the CME was strained by a deluge of message traffic in the Eurodollar futures market. The surge was in part the result of something only dimly–if that–grasped 14 years ago: algorithmic trading. More specifically, algorithmic trading combined with the order matching and confirmation protocols of the CME and Eurodollar markets. Eurodollar futures utilize a pro rata secondary priority rule in the absence of a “TOP” order, i.e., an order that improves the best bid or offer. Moreover, the biggest order at the inside market gets informed of executions slightly before smaller orders. This advantage (on the order of 10-20 millionths of a second) provides a valuable edge.

This institutional setup provided incentives for two algorithmic traders to attempt to get a slight edge in quote size. Thus, each would send a message to increase its size when the other had the edge, which induced the prior leader to send a message to increase its size to regain the lead, which induced the other to send a message to send a message to regain the lead . . . which continued until the maximum quote size was reached. This race for priority in the book led to a surge in message traffic which put stress on the CME system.

Matt Levine wondered why the two algos didn’t just keep their orders at the maximum size. My surmise is risk: the larger size you show, the greater the risk. Furthermore, if someone else improved the price and became the TOP order, size didn’t confer an edge, so it makes sense to cut size.

The CME has responded by announcing fines of $10,000, with the threat of cutting off a violator’s trading connection altogether, for violations of a message traffic threshold. This is a rough form of pricing, so why call it a fine? Why not just call it a non-linear message pricing schedule?

I guarantee that a variant on this story will recur. Because capacity is costly, exchanges don’t price it, and rules adopted for other reasons (e.g., priority rules intended to promote quoting in size) provide an incentive to use this unpriced resource.

As I argued years ago, it’s odd that institutions–exchanges–that specialize in providing the platform to allow the pricing of scarce resources don’t find a better way to price their own scarce platform resources.

Print Friendly, PDF & Email


  1. A friend of mine achieved a Ph.D. in topological mathematics from Stanford University. He joined forces with two of his math advisors to develop a protocol to search for correlations within very (very!) large data sets, based upon the mathematics he had derived. He’s pretty much genius level in math.

    So, they opened a start-up to exploit the method. Typical Silicon Valley stuff. One of his first demonstration projects was to look at the behavior of the futures market. He discovered that over time the movements of certain futures became correlated. When that happened, the market would invariably crash.

    He said the cause was algorithmic trading. He said the algorithms were simple, stupid, similar, and made the same decisions. It was apparently easy to extract their behavior from the reams of market data.

    They got seed money from a VC. The company was successful (lots of hard work), they got pretty rich, they apparently assisted in some medical breakthroughs, and recently sold the company to someone in India. Friend got bored with running a company, got tired of all the travel, and wanted to explore new intellectual territory. He won’t need an external income for awhile.

    But it seems to me, with unintelligent algorithms making family-related trades, some crackerjack math whiz should be able to develop a way to parasitize them. Spot the trends their correlated trades produce and anticipate the market movement they will inevitably cause.

    Evidently Adam Smith’s invisible hand is materializing. He could never have anticipated the emergent visible hand.

    Comment by Pat Frank — November 28, 2019 @ 1:55 am

  2. allocation algorithm was supposed to keep smaller traders in the game but because of the messaging feature you highlighted above, it squeezed smaller traders out.

    Comment by jeff — November 30, 2019 @ 12:36 pm

  3. forgot to mention that the book in Eurodollars also shows best bid/ask given spreads. Bid/Ask can look larger than it really is because of the spreads. Where US Treasury contracts have spreads they are more of a scalper contract. Eurodollars are a spreaders paradise

    Comment by jeff — November 30, 2019 @ 12:38 pm

  4. 20 *micro*seconds? Not 20 *milli*seconds? 20 usec is the relativistic delay on a 2-mile round-trip, so these trades must all be running on a hosted service at the exchange. Even there, 20 usec is only about 50,000 clock pulses, so even a locally running process (assuming, generously, a dedicated CPU core because this time scale is small compared to a multitasking context switch) can implement only a fairly simple decision tree.

    The Prof is right, of course. Exchanges should either price their timeliness (like package couriers) or batch all the trades into granules whose time intervals are large compared to network delays. This, BTW, is why blockchains are, y’know, chains of blocks, rather than continuous streams. BTC’s 10-minute interval is quite conservative even for the crypto world, but I never understood what the problem is with, say, 2-second batching on a conventional centralized exchange. Human traders would never notice.

    Comment by M. Rad. — November 30, 2019 @ 8:23 pm

RSS feed for comments on this post. TrackBack URI

Leave a comment

Powered by WordPress