In Financial Terms, Bank Reserves Are Arpanet
In March 1982, the Assistant Secretary of Defense for Communications, Command, Control, and Intelligence Richard DeLauer circulated an official memorandum essentially declaring the TCP/IP protocol as DoD requirement. It was the culmination of several efforts that stretched back into the 1970's and the internet's earlier version, Arpanet developed by DARPA. The official history begins in December 1978 where Principal Deputy Under Secretary of Defense for Research and Engineering (you have to love bureaucratic titles if not the bureaucracy itself) Gerald Dinneen had issued a prior memo mandating a switch to TCP/IP for all "packet-oriented data networks where there is potential for host-to-host connectivity."
This mandate was making both the next technological leap while addressing increasing shortfalls in capabilities. A DoD report written in July 1980 hailed the success of the first iteration of Arpanet but also noted its limitations. It counted 66 network nodes to that point, connecting as many as 5,000 users, which was quite a feat given those limitations. Going further meant overcoming a fundamental chokepoint, which was described bluntly as, "the basic hardware and software are becoming obsolete."
The primitive network was run using "minicomputers" developed in the 1960's called Interface Message Processors. While they performed very well functionally, they were no longer sufficient resources to handle the growing technical requirements of the DoD and Arpanet. Part of the push for TCP/IP to replace Host Protocols Network Control Program, the existing network protocol standards, was the proliferation of not just various new hardware but the setups and activities of often disparate working groups and all their idiosyncrasies. Further, as anything with the DoD, in military applications there is always a desire for adaptability and redundancy. As the March 1982 memo declared, "Military requirements for interoperability, security, reliability and survivability are sufficiently pressing to have justified the development and adoption of TCP and IP in the absence of satisfactory nongovernment protocol standards."
It was no easy task, of course, as the bureaucracy mandated a specific date and then charged the technical teams for just making it happen. One of those people was Mike Muuss, a research computer scientist at the US Army Ballistics Research Laboratory. He had started TCP/IP Digest as a way to connect with all the various computer researchers working around the world to find more elegant solutions. Unlike other Defense Department initiatives, Arpanet and its internal workings were a public-private collaboration. The first two-network test of the TCP/IP protocols had been carried out in 1975 between one network located at Stanford University in California and another at University College London.
Mike Muuss was at one of the DARPA sites that had been charged with making the protocol transition. He described it as,
"This type of edict is not uncommon when working for the DoD; some manager will stipulate that something will happen ‘absolutely' by a certain date. All the technical people start worrying, and screaming, and carrying on, claiming that ‘it can't be done on time.'...
"I happen to feel that TCP and IP are ‘good' protocols, and certainly much better than what we are using now. It seems something of a miracle that they have been adopted as a standard; usually standards are things like COBOL that people go out of their way to avoid...There exists AT LEAST one choice of software for UNIX systems (all machines), T(w)enexes, Multics, and IBM's, so the majority of ‘ordinary' systems will at least be able to talk, even if not conveniently."
That was the brilliance of the standards only revealed over time, as it was meant for itself to be as simple as possible while keeping all the complications and distinguishing features to the outside among the individual networks. It was meant as a means to develop simple communication elements that could bring together networks that might have almost nothing in common otherwise - it was the primary component of what would become the internet, as the history of TCP/IP protocols is in many ways the history of the internet itself.
There is more to the standard than the acronym implies. While the Transmission Control Protocol (TCP) and Internet Protocol (IP) are specified, there is also the Dynamic Host Configuration Protocol, the User Datagram Protocol, and Internet Control Message Protocol. These work together so that information sent in packets has common, immediately recognizable formatting, addressing, routing, and receiving. These functions correspond to four "layers" of standards.
The top layer, closest to the user and more recognizable to the layperson, is the application layer. The next layer is the transport layer responsible for end-to-end message transfer, segmentation, congestion control, error control, and flow control. The third is the internet layer that takes on host addressing and identification, as well as packet routing. The final layer is the link layer that moves data packets between the internet layer among different hosts. Robert Kahn and Vinton Cerf, primary developers of TCP/IP and thus afforded the honorific of "Fathers of the Internet" sketched a simple diagram of this layer topology here.
While the standards for communication of packets were absolutely crucial, there remains significant need of continuous exertion to still make it all work. Part of that is just the natural evolution of technology, which actually speaks very well of the protocols' central premise, that they can be in use generations of machines later and to undertake functions that might never have been imagined when they were developed. But as a 1993 technical paper by Leland, Taqqu, et al. details, there were individual problems among packet transmission functions that seriously threatened network functionality just as the internet was taking off as the world wide web. In other words, hardware running packet protocols had to fit functionality as well.
Any ethernet station will apply (I have no idea if this is still the process, I only relate was described of it in 1993, though I imagine that the basic ideas survive in the current frame of technology) a "carrier sense technique" that is meant to sense whether or not another packet from another station might already be in transit. If so, the sensing station will defer so as not to cause packet collision and therefore error via data loss. If two stations simultaneously detect an idle channel and both send packets at the same time, there is collision which both stations detect (transport layer) as "the signal on the channel [that] does not match its own output." Each station then calculates retransmission at random intervals.
Network traffic design had to account for congestion and collision avoidance, but early networks were having problems of reliability especially at higher use times. Congestion avoidance was "basically a prediction problem and involves detecting when congestion is imminent, and taking actions designed to prevent it." The problem of predictability was not solved by traditional probability distributions, as network engineers kept running into fractal behavior and self-similarity.
"Leland and Wilson (1991) present a preliminary statistical analysis of this unique high-quality data and comment in detail on the presence of ‘burstiness' across an extremely wide range of time scales: traffic "spikes" ride on longer-term "ripples", that in turn ride on still longer term "swells", etc. This self-similar or fractal-like behavior of aggregate Ethernet LAN traffic is very different both from conventional telephone traffic and from currently considered formal models for packet traffic (e.g., pure Poisson or Poisson-related models such as Poisson-batch or Markov-Modulated Poisson processes (Heffes and Lucantoni (1986)), packet-train models (Jain and Routhier (1986)), fluid flow models (Anick et al. (1982), etc.)." [emphasis in original]
Until engineers had updated their prediction processes out of normal probability architecture to start controlling for self-similarity and using a multi-fractal approach, high-use networks were prone to these patterns of "burstiness" that would actually degrade and, if serious enough, collapse them.
I am quite (very) often asked to describe or define a eurodollar. That's actually a good thing overall because despite it being the epicenter/cause for the events of the Great Recession and panic, hardly anyone even knows it exists. Last week I paraphrased Charles Dickens' description of insurance to call it a system where one bank that holds no dollars gets another bank that holds no dollars to guarantee everyone has dollars. While I was reaching for levity, there is a lot of truth in that formulation. The word itself, eurodollar, is a misnomer through and through. There are no dollars in it, and it isn't really tied to Europe (and often gets confused as if some amalgamation with the euro, which it most certainly is not).
I struggle to answer because a eurodollar is not a thing like a dollar is a thing; the eurodollar system is therefore not a redistributable pile of eurodollars. If we are trying to be as specific and comprehensive as possible, the eurodollar is really just a system of standards that allows common functions to be carried out among various and often quite disparate systems - it resembles in so many ways the TCP/IP protocols that delivered the internet. That isn't so much an accident, either, as the eurodollar system grew up and evolved alongside the same technology and functionality as the internet. As computer power was shifting into the 1980's, so, too, was modern banking and finance: interest rate swaps were born then, as were eurodollar futures, the early forms of math-as-money (Basel) and bridging all of that together were computer networks that allowed more instantaneous settling and communication.
The eurodollar can thus be described as a multi-dimensional series of often parallel but intersecting financial functions all designed to encompass a single, global whole. The nature of those functions just happens to be monetary, so that the end result of eurodollar functionality is monetary or financial. In most simple terms, it solves the basic equation of how to get numbers all over the world to balance and stay balanced. That used to be quite cumbersome and often dramatic, as under the prior currency standard, gold or dollar/sterling, meant taking some accountant(s) down to the vault and performing a physical audit. By removing the dollars from the eurodollar (in reality, there weren't many in it to begin with) or money from the monetary, the system could perform in unimaginable ways since it would be free to reinvent itself over and over. The standards would remain but the methods and products did not.
That is a very promising and potentially awesome power; but it is also frightening at the same time. Among the earliest warnings were of this self-isolation, which means that though like a computer network its monetary duties meant vulnerability would have very grave consequences. However, because that did not happen throughout most of its history, the eurodollar was simply ignored both officially and publicly. It carried on in all its new dimensions with only one true but minor disruption (LTCM, Asian flu) until August 2007.
There are any number of examples of the eurodollar system acting among these new dimensions. Perhaps the easiest to understand were the repo market problems of the panic era. Recalling these events a few months ago, I wrote:
"According to ICMA, a prime, AAA-rated agency MBS traded at a 4% repo haircut in June 2007, just before the fatal shift in eurodollars. By June 2009, that same MBS would find instead a 10% haircut. That was a massive change, which caused selling to beget selling and so on and so on. For unrated agency MBS, again prime, the haircut that in June 2007 was 10% had moved to 30% or even 100% by 2009. And that was not the full extent of the collateral/liquidity problem, either, as certain strains of even prime MBS collateral became non-negotiable on any haircut terms."
Collateral had been one of the protocol "layers" that had allowed the eurodollar system, wholesale finance, to undertake such massive expansion especially in the later 1990's and middle 2000's. The haircut adjustments were very much like "burstiness" suddenly being exhibited in a formerly stable network of collateral flow and trading (rehypothecation as another protocol "layer" within that collateral layer). Once the network function was destabilized, there was no saving it. The eurodollar system as a set of standards remained constant, but we found out in close experience how unstable it could become once perturbed (and I will note, quite superfluously, that there were fractals all through that, too).
I have described at various points in the past and recent past the occurrence of negative swap spreads (and now their re-occurrence). Interest rate swaps are again something like network protocols whose end result is quasi-monetary. In a simple, stylistic example we can picture a bank with a single security holding, say $100. It has simple modeled characteristics of 2% expected daily variance; thus its VaR would be $2. If the market for that security suddenly hit a rough spot and the expected daily volatility rose to 3%, the bank can either absorb that increased VaR or if its risk parameters do not allow it, the bank will have to go into the market to hedge that additional $1 VaR away.
Negative swap spreads present a couple of difficulties, including right at the front introducing modeled uncertainty since swap spreads (the relation of interest rate swap prices/rates to the same maturity US treasury bond or note) are never supposed to be negative. Worse, however, the very appearance of negative swap spreads suggest that the "supply" of hedging capacity in the form of IR swaps is likely limited and quite severely. In the example of the preceding paragraph, if this shortage is severe enough (meaning either too costly or, in extreme cases, totally unavailable; which was the 2008 experience with credit default swaps in the same context) the bank has no choice but to sell some or even all (depending on further liquidity) of its holding.
Thus, the various factors that go into that one single choice, to hold some security or to just sell it (true liquidity), is governed by a multi-dimensional system where various individual facets of it at any one time can act like currency and money. In practice, it is far, far more complicated as whole financial portfolios lead to unbelievably complex situations and needs (including the incestuous nature of it all where "money dealers" are also system participants). It was eurodollar system's inherent flexibility, like TCP/IP's, that allowed it to become so and remain so, but only as long as there was/is determined effort to maintain it.
In the past year and a half, nearly two years, there has been increasing disruption traced to the Asian nodes of the eurodollar network. I have called this the Asian "dollar" simply to make a distinction within the eurodollar system. The primary focal point of disorder is traced to China, no surprise, but increasingly in recent months we are finding Japan.
The Bank of Japan has been at the forefront of destroying market function for more than just this latest "cycle." It pioneered ZIRP and QE at a time when Western economists and policymakers thought such extraordinary efforts ridiculous and unrealistic. It never works, of course, including its latest flameout in QQE that was meant to be so large and shocking that it could not have possibly failed. Despite obvious flaws in its theory, it was promised as such and then dutifully described in the media as "stimulus" each and every time said "stimulus" failed to stimulate.
Instead, Japanese "markets" are increasingly left barren and stripped. Because QQE did not work, the Bank of Japan has since added negative interest rates (at the end of January), amplifying a really breathtaking destruction of Japanese money markets that had already been in process because of the country's broken bond markets traced back to QQE. An article published by Bloomberg a few days ago describes the extent, as interbank volume in Japan reached a record low on March 31. Izuru Kato, president of Totan Research in Tokyo was quoted (rightfully) bashing Japan's central bank, "Among central banks, the BOJ is the one that destroys functioning markets the most."
Another interbank functional layer, the country's call market, dropped by 50% in February from January when NIRP was announced. The call market is a vital interbank conduit for setting not just borrowing costs and estimating potential liquidity and its volume, but in carrying those estimates into the rest of the wholesale financial standards - including those undertaken by Japanese banks in eurodollar activities. Just two weeks ago, the Japanese government bond market was hit with a "buying panic" that sent already negative JGB yields even more negative. It was, in my view, feedback from both negative basis swaps and increasing uncertainty and volatility having unseen effects upon total financial function. In other words, what are Japanese banks doing about rising volatility/uncertainty especially in funding, and how might that impact VaR, vega and other risk parameters that we are not privy to?
Japanese banks have been among the heaviest suppliers of eurodollar resources dating back to very near its origins. When the Fed was forced to "bail out" the world with dollar swaps in late 2008 (I use quotes here because it was another abject failure in an unending string of them), the Bank of Japan was the second largest recipient after the ECB. During the Asian flu of 1997 and 1998, Japanese banks begged the Fed for "dollars" then, too.
If the Bank of Japan is undercutting its own markets and banking system, then we cannot think of that as individually a problem for Japan alone. To circle back again to the networking analogy, it is as if there are huge disruptions emanating from one of the eurodollar network nodes that is likely causing feedbacks of disruption (burstiness) all through the rest of the network system. As this one node destabilizes, it tends to destabilize others that then act up in systemic fashion that can cascade into complete network outage (global liquidations; two so far).
That is just the more recent of glitches that have been on display over the past few months where "calm" had seemingly returned. Dating back to last year, viewing yen and yuan in concert, it seems as if Chinese difficulties were causing Japanese banks to rethink their funding of the Chinese eurodollar "node." The expansion of the negative basis swap since then and the further insanity of the BoJ might have already risked turning a derivative funding issue (that is, from China alone) into one that is now self-contained (and self-reinforcing) to Japan as separate from its still-ongoing China "dollar" exposure. The Chinese yuan exchange to the dollar has been stable throughout (though undoubtedly due to artificial interference, which will bring about its own reckoning) while the Japanese yen has not been; big time.
In all of this, yuan, yen and not a dollar in sight, we are describing the function of a system of linked networks that contribute shared financial resources all around the world. They take the word "dollar" because of both tradition and how that started the common language of the shared standards. Like the internet is a system of customized data flow, the eurodollar is a bespoke system of financial flow. Unfortunately, when there is disruption it is likewise shared and transmitted. In that respect, the eurodollar protocols are nothing like TCP/IP, which has been nurtured and even corrected when needed by its own participants and operators. In the eurodollar system, those tasks have been centralized to agencies who remain purposefully unaware of its very existence. They see trillions of dollars in bank reserves and think no dollar shortage could ever be possible. Bank reserves are so Arpanet.