Banking crises are correlated with business cycles, and business cycles are correlated with volatility. Thus, one could say they have too often gone 'short volatility'. But that is rather incidental, not an explicit strategy. The real question is why their portfolios tend to experience highly correlated declines that threaten their solvency every 15 years or so (just within the US: 1819, 1837, 1857, 1873, 1893, 1907, 1901,1929, 1966, 1970, 1975, 1981, 1990, 2008). My theory, based on Batesian mimicry, is here.
As to how to ameliorate if not eliminate these crises, Cowen supports Kevin Drum's argument:
The way to do it is with very simple, very blunt leverage restrictions that apply to all financial actors over a certain size: banks, insurance companies, hedge funds, private equity, you name it. If you have assets over, say, $10 billion, then the rules kick in. Strict leverage limits (say, 10:1 or maybe 15:1) based on conservative notions of both assets and capital would be a pretty effective bulwark against excessive risk taking but wouldn't seriously interfere with the basic asset allocation function of the financial industry.
I do like the way his solution actually makes sense, in that you know what he is saying. You can listen to some people talk for an hour and realize they are merely for more and better regulation, without any specifics. But, the reason why prominent people are vague is because then you aren't wrong, like Drum and Cowen are in this instance. The problem with this great idea is that if you make leverage the key point, regulators will focus on that, and so banks will just invest in riskier assets.
Crises are often compared to drinking binges, where excessive exuberance is followed by hangovers. Following that analogy, if you try to say, 'we don't want people drinking to excess, so we will stop everyone at 6 drinks', well then, they will switch from beer to wine, whiskey, or grain alcohol.
But the drinking problem is potentially soluble because you can measure alcohol, and say allow people one 12 ounce beer, one 5 ounce glass of wine, etc. Basically, the key unit is 'ethanol, which has measurable properties. In contrast, we don't know what 'risk' is. Beta? Volatility? Skew? Get back to me on that. You can't regulate what you can't define.
What about leverage as a proxy for this unmeasurable risk? As Proshares is showing with the Ultra (2x leverage), UltraPro (3x leverage), and short products, a security can have double or triple leverage behind it, have positive or negative exposure, and still be called a 'stock'. Similarly, a bank with 10:1 ratio but plenty of ninja loans had a lot more risk than a bank with 20:1 leverage but all their mortgages had 20% down (the bad old days per Alicia Munnell). Facility risk (eg, loan-to-value) is part of the problem, so too is obligor risk (eg, credit and bureau scores), and so the multiheaded beast grows, with risk hiding from any one metric that can be applied across any large financial institution. This, alas, brings forth many demographics, all who want unsecured lending, which has a populist appeal.
The key is that drinkers engage in moderation only when they realize hangovers are not worth any temporary high. Unless they believe that in their hearts, they will get around your regulations the same way college kids--who generally are below the 21 year old drinking age--tend to get around restrictions promoting their sobriety. Any top-down rule to prevent excess will simply waste time because you can hide the leverage at the other end, say be investing in assets that are leveraged a lot, or who have suppliers who implicitly leverage them (as in the dot-com bubble). Such rules might even make things worse by giving people a false sense of security if nothing bad happens for 10 years, as often is the case.
I used to be head of 'economic risk capital allocations' for a bank, and we had very low risk for mortgages. I left before the madness started, but I can see how it morphed because it would be easy for the business line managers pushing product to point to historical losses in mortgages and say they are basically riskless (eg, the Stiglitz and Orzag analysis). Now, some smart people (eg, Greg Lipmann, Andy Redleaf, Peter Schiff, among many others) saw the past data were not relevant once you start lending to people with no money down, or people with no documents, and that depending on collateral prices rising basically was a game of musical chairs. But within large organizations like banks and GSEs these people were demoted, as happened to David A. Andrukonis, the risk manager at Fannie Mae who was fired for getting in the way of Bill Syron's $38MM windfall. A big idea that is plausible and has many beneficiaries is very hard to resist in real time, and such ideas don't end via argument, but rather conspicuous failure.
So, define risk--not its correlates like leverage, but actual risk--first. Make sure it isn't backward-looking, looking at the past couple of crises, but by say anticipating the next bubble in muni debt or education. Unless you can convince people that such risks are real, any leverage rule will be made irrelevant via the creativity of people designing contracts taking into account the letter of the law. That's really hard, you might say, and we have to do something now. Doing something is not better, unless you think pandering to the mob is a primary objective. If risk management were merely following some simple asset-to-liabilities test, someone would have figured that out by now.