Justin Fox argues that the financial sector needs to be "compartmentalized":
Why Bankers Need to Be Put Into Little Boxes, by Justin Fox: There's a beguiling little moment in the financial-crisis documentary Inside Job where hedge fund billionaire George Soros describes the principles of oil tanker design. If a tanker consisted of one big tank of oil, the sloshing liquid would soon capsize the vessel, Soros explains. So tankers are comprised of lots of smaller, separate tanks, which keeps the sloshing in check and the ships afloat.
Financial markets are like that, Soros goes on. If they're compartmentalized, the risk of crisis is much lower than if all sorts of financial products and institutions are allowed to mix together in a giant sloshfest.
It's a nice analogy. That doesn't mean it perfectly describes the workings of financial markets (it's an analogy), but it certainly gets at some aspects not hinted at in the general equilibrium model that long dominated financial economics — in which more "complete" and intertwined financial markets are supposed to lead to better economic outcomes. To mainstream economists the Glass-Steagall Act that separated the banking and securities industries looked like a competition-restricting, innovation-damping anachronism. To those knowledgeable about oil tankers, its repeal in 1999 must have been far more disturbing. ...[I]t is a sector where ... compartments — serve a clear purpose.
I agree to a point, but not completely. When I think of an interconnected network that can spread problems from institution to institution, there are two possible scenarios. First, think of a drop of poison in the ocean. The ocean is so big that even a powerful poison can be neutralized as it spreads. After it spreads, the amount of poison in, say, any given cubic foot is too small to even be noticed, let alone do damage. In this case, you do not want to have the network compartmentalized (If we confine the poison to one cubic foot through compartmentalization, it will remain toxic to anything living within that space, by allowing it to spread you avoid this damage). This is much like traditional financial risk sharing where large individual losses are spread throughout the system so that the losses to any one person are tiny.
But now think of a poison that acts more like an infection. As it spreads it does not become less toxic, it continues to be lethal to anyone who comes in contact with it. In this case, you want to break the network connections -- i.e. compartmentalize -- as soon as possible to prevent the spread of the lethal infection. You may even want to have the compartments set in advance if you cannot sever the ties fast enough.
Prior to the financial crisis we emphasized the first scenario, the one where risks are spread throughout the world through financial intermediation. But we got hit by the second -- a falling domino type problem that spread from institution to institution through extensive network connections, and it was lethal to those who came in contact with it (though in some cases the government stepped in with an antidote).
I think that most of the time connectivity that spreads losses throughout the system acts like the first scenario - -it takes a potentially lethal hit to an individual and dilutes it until it is rendered relatively harmless. But when other types of poisons hit the system (i.e. very large shocks), we have to be able to break the network connections almost instantaneously to prevent the spread of a lethal infection. If we cannot do that, then we need to have the compartmentalization in place a priori (e.g. through a Glass-Stegall type firewall). The hard part, of course, is determining which connections are okay to leave in place and which ought to be severed in advance to prevent problems. That's something regulators will have to struggle with. But neither a fully connected networks nor a fully compartmentalized networks are optimal in all cases, and we need to find the right balance.