Stress Testing Now is a Really Bad Idea

Last week the market stumbled, especially financials. I think a lot of that was because people are afraid of Geithner's potential stress test. Today, I read two eminent people calling for stress testing. First, Andrew G Haldane, Executive Director of Financial Stability at the Bank of England. As Haldane notes:

First, setting the stress scenario. The key elements here are devising a multi-factor risk scenario that is sufficiently extreme to constitute a tail event. Arguably, designing such a scenario is better delegated to the authorities than to individual firms, in part because they ought to be more immune to disaster myopia. In its Financial Stability Report (FSR), the Bank describes such stress scenarios.

What is this stress test in the FSR, I wonder? I check it out online, and see that it seems to merely extrapolate losses on mortgages and corporate bonds. Now, with mortgages one can estimate based on the recent past, but with all of the mortgage proposals out there, this is all a huge guess, because they have the potential to make the current situation much worse via giving mortgage holders the incentive to default and then apply for federal aid. After all, a $1000 bounty is being proposed for such restructurings, in which case a new wave of mortgage defaults will occur as people rush to fit into the Feds new guidelines for restructuring. As to the corporate default rate extrapolations, there aren't any good aggregate corporate default rate models. I was at Moody's, and the model is pretty bad, merely because you really do not have a lot of data, and its basically like business cycle forecasting, which has thus far failed to anticipate every major economic event since Moses (not that they are stupid, it's a hard problem). Speculative default rates peaked in 1970, 1990, and 2002 at around 9%, so I would simply stick with that (it was 15% in 1933, however), though this is not mentioned in the report so I don't know what they are using. As per the investment grade default rate, those peak around 0.4% since the Great Depression. But is not clear the potential stress test would differentiate between speculative and investment grade loans. A distinction with a difference, one of many in this issue.

But this is a small part of many bank's portfolios. At KeyCorp, where I was head of capital allocations, most of our exposure was to consumer credit, and non-rated corporate debt, and the fluctuations there were milder, historically. What stress will you apply there? If they apply the speculative or mortgage default stress to these books, I imagine most banks are insolvent, but what does that prove? If a sector of the financial system is experiencing the one in 100 year flood, do you then assume every sector has the same 1 in 100 year flood simultaneously? Haldane does not specify this important issue. And many thought the invasion of Iraq was poorly planned.

Nouriel Roubini and Mathew Richardson are for instant stress tests to determine which banks live:

First -- and this is by far the toughest step -- determine which banks are insolvent. Geithner's stress test would be helpful here. The government should start with the big banks that have outside debt, and it should determine which are solvent and which aren't in one fell swoop, to avoid panic.

No details on the stress test are given. I presume he figures a 'stress test' is pretty straightforward. It is not. Any quick, nationwide stress test will be arbitrary. Regulators do not have a lot of experiencing assessing economic risk capital to 'good' assets. Historically, they merely check leverage ratios (tier 1, tier 2, total), and then make sure various level of bad loans are in their proper buckets (OAEM, Substandard, Doubtful, Loss).

So, a test might be something like, assume all assets can lose 10% of their value. This necessarily penalizes banks with lots of high quality assets, because it treats BB rated bonds like AAA rated bonds, and even with the recent surprises, one should not assume every liability has the same probability of going into default. A stress test should apply the stress based on their historical loss rates, looking at the institution and industry-wide data, but that would generate a large amount of differentiation and complexity, and this would generate a lot of haggling over important, real, differences of opinion, and sheer politics as some use legitimate issues as cover for naked opportunism. It is inherently complicated, and even though we would like a solution quickly, it is counterproductive to attempt a quick solution in this situation.

Now, doing this quickly would stop the contagion issue, but only by being arbitrary. I would take my chances with contagion vs. randomly culling the herd via some poorly thought-out stress test.


And who are these guys? Dr. Doom, Nouriel Roubini, is a man who always sees disaster looming (that picture is from his own homepage, I guess it's his cheeriest photo). He forecasts disaster perpetually from an infinite number of sources, and his mechanism hardly relied on the mortgage sector ex ante, so I think it's safe to say he is just a broken permabear clock who now has the correct time. Andrew Haldane, in his speech, critiques the financial community for missing the crisis, but since 2005, he headed the Bank of England's Systemic Risk Assessment work, responsible for the Bank’s new quantitative risk assessment framework and its six-monthly Financial Stability Report. I do not see why he exempts himself from this failure, or why he feels confident that an instant bank decimation scheme based on his trenchant understanding of the financial sector is better than simply leaving the patient alone.

There's a lot of talk about hubris driving this crisis. I think the hubris of these schemes is at least on that level. Just remember, no matter how bad things are, they can always get worse, and when dealing the the collapse of complex systems no one fully understands, attempting a quick fix is almost certainly going to make things worse.