Wednesday, February 25, 2009

Geithner's Stress Test Revealed

The crack team at the Treasury, headed by veteran enterprise-wide capital allocations specialist Tim Geithner, revealed their stress test. The following scenarios are modeled:

The stress test assumes an unemployment rate averaging 8.9% in 2009 and 10.3% in 2010, a 3.3% contraction in gross domestic product in 2009 and home-price declines of another 22% in 2009 and 7% in 2010.

They then merely see how the various bank exposures are affected by these assumptions, including their ability to raise new capital, and 'other risks that are not fully captured in regulatory capital calculations.'

Well, that sounds reasonable if I didn't know anything about banks, but alas, you only fully realize The Experts are usually talking out of their backsides when you hear one talk about something near and dear to you, and note they have not a clue what they are talking about. Then, an epiphany: perhaps it's the same when experts discuss the Gaza strip, health care, everything? In my twenties I felt I didn't understand anything. Now I realize the experts don't know much about their expertise, so it's nothing to get too worked up about. Geithner is an expert in the Robert Rubin sense, who wasn't concerned with things like Citi's asset composition, but knows about getting access to the big politicos that affect essential regulation (isn't that what managing a bank is all about?).

Banks do not have an asset labeled GDP, Unemployment, or Homes. They merely have assets with various uncertain correlations to these factors. Some correlations are truly causations, as in mortgages and home prices, but even there, the connection between one and the other is very complicated, depending a multitude of characteristics in addition to these assumptions (vintage of the loan, its Loan-to-value, the credit score, income and wealth of the borrower). In other cases, such as looking at loans to shipbuilders, the connection is far removed.

If you think finding correlations between asset prices and basic assumptions is easy, consider that Peter Schiff, someone pounding the table that home prices were overvalued because of reckless underwriting standards. His portfolios actually performed horribly last year, because he assumed that this assumption implied a weaker dollar and the US equity market would underperform. You can be totally right on assumptions, but the complexity of the economy renders that insight irrelevant.

Or consider that business cycle forecasting is so fruitless because the particular drivers of each recession are different. The Stock-Watson leading economic indicators model failed to predict the 1990–1991 recession, and an updated version of the model (one that would have caught the 1990 recession) then failed to predict the 2001 recession.

Stock and Watson discuss this failure and argue that it is hard to predict recessions because each is caused by a unique set of factors. For instance, housing and durable goods consumption was strong preceding and throughout the 2001 recession, because the decline was focused on high technology manufacturing. By contrast, in the 1990–1991 recessions, housing and durable goods spending slowed considerably. As Stock and Watson say, “Without knowing these shocks in advance, it is unclear how a forecaster would have decided in 1999 which of the many promising leading indicators would perform well over the next few years and which would not.” Thus, note the stress test primarily involves housing, because that has fallen the most in the past couple of years. Three years ago, it probably would have included a big decline in internet stocks, whereas in 1980 an inflation or oil shock. One wonders what kind of trouble such thinking can lead to, this extreme reliance on the past couple years in forecasting the next couple years...

So, assuming you know where a few pieces of a complex economy are going is a very thin basis for any meaningful stress test. What are the implications for things like one's small business credit lines, credit cards, or airline leases? If you are using real data, you will probably have only 3 recessions at most in your historical analysis, and chances are, the coefficients on many asset classes will be insignificant, but all will be highly uncertain. Just yesterday, I was remarking on an article that blamed the subprime crisis in large part on people assigning too much credence in uncertain correlations. The correlations in a copula are infinitely more precise than any correlation with GDP, Housing prices, and Unemployment. Thus, the results of this exercise is so imprecise it can hardly be the basis for any action. Hopefully, the Treasury shrewdly recognizes this and are actually targeting a placebo effect.

People have to realize that banks are not portfolios of market traded bonds, with lots of data that make for some sort of fat-tail adjusted Value-at-Risk, or beta, meaningful at the bank holding company level. Further, most banks are not investment banks. A standard bank has most of its book not marked to market, its loans highly illiquid, and the best data one has is limited, multidimensional, noisy, has selection biases, and refers to a market with changing parameters (see mortgage underwriting innovations in the 1990's).

Geithner noted he will wrap this up by April. Given the absurdity of this exercise, they should shoot for Friday and save everyone a lot of time. It won't be any more accurate by taking two months.


Anonymous said...

Right on. I too find myself hoping these bizarre programs are just ruses to fight people's crises of confidence. Actually the first couple of housing interventions seemed to be exactly that, e.g. pp5 of (

# 56 said...

Agreed. What would a real stress test look like? Is it even possible to produce meaningful results? Given TG's assumptions and recent asset back performance I wonder if a couple quants, in a week's time, couldn't come within a rounding error of what the Treasury will take months to produce. The April deadline is simply a continuation of the kick the can approach. Then there's the reasonable question, is the "stress test" a thinly veiled attempt at window dressing?