New names for old ideas
On Tuesday June 8, the content delivery network Fastly went down for about an hour, causing major websites, including the Financial Times, to throw 503 "Service Unavailable" errors across the globe. Other web infrastructure providers, including Amazon Web Services and Cloudflare, have experienced comparable outages in recent months.
Such outages are annoying and disconcerting for users, and costly for businesses that depend on the services directly. But they also indicate the possibility of a more systemic financial risk. It's not just that downtime could prevent trading: services could actually become unavailable because of financial activity. That means that the ability to trade could disappear when it matters most.
Blockchain "network congestion"
In fact, I think that we have already begun to see this happening. On May 19, during a two-week period in which major crypto asset prices dropped by a third, Coinbase and Binance both experienced significant disruptions. Both companies cited "network congestion".
Prices for many crypto assets dropped sharply on that day. On this graph of Bitcoin, Ethereum and Litecoin prices, May 19 is marked by a vertical line:
This is not the same thing as Fastly's outage this week. But neither can it be dismissed as just unexpectedly high traffic on the day of some major event, such as a news website might experience.
Crypto exchanges' downtime is a financial phenomenon, arising from their market-making activity in a fundamental and inescapable way. A few weeks ago I argued that the difference between on- and off-chain crypto transactions could cause problems for the exchange during times of high transaction volume. May 19 is─I think─an example of just that.
Exchanges that hold crypto assets in custody, as Coinbase does, can transfer funds off-chain among their customers using only their own accounts. Such transfers can be completed quickly on the exchange's own servers. But a market-wide move from crypto assets to money, or the reverse, is more likely to require the exchange to trade with others in on-chain transactions, which are by design slow and costly.
A dealer model describes the situation fairly well. Coinbase is a dealer because it holds an inventory of crypto assets that it uses to facilitate trading among its users. Transactions that do not result in large net flow can be accommodated without much price movement. Even if volume is high, the assets can just be passed back and forth at low cost among the exchange's users. In the dealer model, the company sets its inside bid-ask spread within the bounds set by the wider market's outside spread:
But Coinbase's position has a limit. At some point it cannot continue to buy from its customers, and the company must sell inventory on-chain. During the big price drop of May 19, Coinbase customers were selling crypto assets. The company moved toward its upper position limit, and would have had to sell on-chain or be stuck with a huge outright positions, the value of which was dropping quickly at the time. But on-chain transactions are slow, and so Coinbase made its customers wait. Coinbase stopped making markets rather than have to trade at a loss.
The exchange vanished precisely at the moment when trading opportunities were most highly valued. This is not incidental, it's exactly what should be expected from a market-maker at its limit. It's one thing to make a profit by offering trading opportunities when markets are quiescent. Market-making through a big price adjustment is expensive, as was amply demonstrated in 2008.
Note, this is interpretation on my part—I don't have any privileged information on Coinbase's or Binance's decision-making. Nor does that particularly interest me—I am looking for the big picture.
One of the hallmarks of crypto and DeFi projects is that they use the language and trappings of software development: they share their code on GitHub, they offer documentation rather than prospectuses, they typeset their white papers using LaTeX.
I would not discount the potential for these ideas to disrupt established practices in finance. Some of the changes will endure, as work on central bank digital currencies is showing.
But when an old phenomenon shows up in a new guise, we should call it by its name. And the name for "network congestion" is illiquidity.