Too Central to Fail

A lot of attention has been put on “too big to fail,” the idea that big is risky. What really matters in a complex network system, however, is not bigness per se but connection centrality. In a network the liabilities of institution A become the assets of institution B whose own liabilities become the assets of institution C. An institution with high connection centrality can spread distress throughout a large portion of the network.

Inspired by Google’s PageRank, the authors of a new paper create DebtRank, a measure of connection centrality. The vertical axis in the following diagram shows DebtRank (centrality) the horizontal axis asset shows size relative to the total network and the color indicates fragility/leverage. Institutions such as Wachovia, RBS and Barclays were relatively small but because of their centrality and fragility they imposed big risks on the system.

You can find the paper here but do check out the web page of the author group which includes much more material including these animations. Mark Buchanan over at Bloomberg also offers useful comment.

One point to note is that the authors calculated centrality using ex-post data from the Fed. Using this measure, DebtRank clearly signaled danger prior to the crisis and did so earlier than other metrics. In order to do this in real time, however, much more transparent and timely data would be necessary. The fact that centrality doesn’t correlate all that well with bigness, however, indicates that without this data the problem of monitoring risk is even more difficult than it appears.


Fascinating research and a great idea.

This is the kind of thing where I wish it had been thought of during the brief moment when people were actually angry at the banks, and legislation might have actually had a chance of being passed. Figure out what data you'd need to construct a "systemic risk rating" (or "DebtRank") for each large bank, and mandate that the bank's computers funnel that information into a government computer in real time so that the regulator can calculate the number. And then when a bank's DebtRank gets too high, split that mofo up into smaller banks.

Also, hire smart people to work for the government to constantly update the DebtRank algorithm so you don't get regulatory capture and Goodhart's Law can't catch up to the measure.

sigh a guy can dream, right?

"...centrality doesn’t correlate all that well with bigness"

Seems like a pretty positive correlation to me.

If a bank's big yet not central enough; doesn't that hint it is under-diversified?

Luis Amaral of Northwestern has also published interesting research on connectivity in epidemics and metabolism. I wonder what his computational approach would find for finance.

Debtrank is an excellent measure of exposure, however is it a good forecaster of danger? I think it needs to be combined with the banks' capital ratio, less well capitalized banks have been a good indicator of failure.

What really matters as the long-term liability in a complex human network is--
"The Iron Law of Oligarchy", known for over a century.

The fundamental connection-centrality factor is institution leadership/management.

All forms of human organization will eventually and inevitably develop into oligarchies.

Oligarchy results from the technical indispensability of leadership in the network {organization}... and, the overwhelming tendency of the leaders to manipulate & consolidate the network for their own
self-interests. Risks to the network/organization are much greater than the personal risks to the leaders, as they manipulate a complex system to their personal advantage.

Notice many Big-Bankers going to prison or living in homeless shelters ??

(..this "debt-rank" stuff seems to be nonsense, overall)

The oligarchy comments is interesting. If you look at the history of each Chinese dynasty, it seems to be a perfect model of how oligarchies take over institutions and then fall (along with the institutions).

Even educated Westerners have a weaker understanding of this, because until 1945 competition between states tended to keep the development of oligarchies within states in check to some extent. There was extensive growth of oligarchies in the 18th century, though probably not as much as in Qing China at the same time. Late Chinese style dynastic stasis happened in Ottoman Turkey. These exceptions prove the rule.

Would be interested in seeing this data and cart posted over a period of time, say each year since 1994, to observe changes in leverage and centrality over time

The system was stable from 30s to 80s by restricting the operations of banks and placing the Fed and FDIC at the center of everything.

The premise was the government was too big a risk, so banks should be free to interact however the market led them because the market is always able to integrate all the information better than technocrats.

Information has value, so keeping it secret is key to profit.

Now the banks were led to places that made no sense after the facts are revealed.

In the end, stability was maintained by putting the Fed and FDIC at the core of the system.

The Fed and FDIC even became the core of the shadow banking system to save it from what was never supposed to happen because the shadow banking system turned into demand deposits just as the opponents of allowing retail money market funds predicted, and the proponents promised would never happen. Millions of individuals not only were unable to evaluate the assets of the shadow banks holding their deposits, they weren't able to see the assets because those holding represented competitive advantage.

The past fifty years suggest to me a long argument like calling for the removal of all regulation of roads. Get rid of speed limits, drivers licenses, vehicle design requirements. Let the drivers decide how to drive like they want knowing everyone else is free to do whatever they want, and everyone on the roads knows you die if you crash. But then we discover that buses have been hired by passengers based on the drivers being experts, only to find the bus drivers argued they should not only get really high pay for their skills, but also ejector seats because their skills are too scarce to risk dying in a crash.

As buses got larger and the excess cost of lighting led to shedding all lights, a massive pileup with all the drivers ejecting leaves the mess for the government to deal with, doing triage on the passengers and removing the wreckage while a fight takes place over who is to blame and how to prevent it happening again.

A faction is blaming the passengers because the passengers trusted the drivers which meant the buses just got larger, so individuals should take responsibility and drive solo, and if a bus runs you over, well its your fault. (Right now, the Wall Street wonks are saying all the individuals who see the roads as too dangerous and so they walk cross country are missing out, but they just figure the roads are rigged so only the bus drivers survive.)

Another faction is suggesting that the records of the drivers leading up to the crash can be used to selectively remove the dangerous drivers before they crash.

But while not perfect, the one rule for all regardless of skill: speed limits, drivers tests, vehicle specs, and police patrols enforcing the rules is able to limit the carnage. And if failures occur, the government is at the center of dealing with it.

"speed limits, drivers tests, vehicle specs, and police patrols enforcing the rules is able to limit the carnage. "

This is an assumption, and considering this kills more people than anything comparable I wouldn't assume too strongly.

This touches on really important issue.

Much of this "connectedness" surely reflects repos--a major source of funding for banks. Repos are collateralized short-term loans, & were major source of contagion in "financial crisis cuz securitized mortgae assets (MBAs, CDOs, CDO-Squared, &c) were used as collateral. The proximate cause of crisis was freeze-up of repo market that reflected lack of transparency--not just value of collateral, but chain of counter-party "connectedness", e.g., maybe I trust counter-party's collateral, but don't know quality of it's counterparty's collateral.

This is an emergent "network effect" not predictable from analysis of banks in aggregate. But can be addressed by eliminatating the unobservable chains of connectedness thru clear--as for (most) CDS swaps in Dodd-Frank.

Since repos functionally same as OTC swaps, except for matter of legal ownership, why not just require banks to actually use swaps & require clearing? That way, the potential "network effect" effect in contagion could be minimized....

One hypothesis about over-connectedness leading to failure would be that increasing concentration, leads to economic/systemic homogeneity. This is driven by the trade off of efficiency (ROEs) at the expense of systemic resilience. Eventually the overly homogenous systems fail to have the adaptive resilient capacity to respond to shocks or the engage in unsustainable activities such as confusing an assets' price with its value. this is fairly common in risk taking activities, reading (Kindelberger's manias, panics and crashes or Devil take the Hindmost) shows how common the pattern of systemic collapse and rebirth is in debt markets over the centuries.

Comments for this post are closed