Friday, April 24, 2009

Treasury Methodology Begs Essential Questions

From today's Treasury methodology paper:

To guide estimation, the BHCs were provided with a range of indicative two‐year cumulative loss rates for each of the 12 loan categories for the baseline and more adverse scenarios. [Banks] were permitted to submit loss rates outside of the ranges, but were required to provide strong supporting evidence, especially if they fell below the range minimum. The indicative loss rate ranges were derived using a variety of methods for predicting loan losses, including analysis of historical loss experience at large BHCs and quantitative models relating the performance of individual loans and groups of loans to macroeconomic variables.

The 12 categories are as follows:
  • Mortgages
    • prime
    • Alt-A
    • Subprime
    • Closed-end junior liens
    • Home equity loans/lines
  • Commercial Real Estate
    • Construction
    • Multifamily housing
    • Non-farm, nonresidential
  • Consumer
    • CreditCards
    • other (auto, RV, boats)
  • Commercial and Industrial Loans
  • Other

Note the fine distinctions on the past 2 banking crises: commercial (1990) and residential real estate (2007+). You can always count on regulators to protect us against the most recent credit events. Odds that the next banking crisis is centered in some subcategory of 'other': 98.76%.

The big question is, given unemployment rises by 3%, GDP falls by 4%, and housing falls another 20%, what are the effects on all these categories? The white paper does not say. What is the Treasury's base assumption for the average asset in these 12 categories? Who knows!

Ask yourself how much money your business would lose if GDP fell by 4%. Do you know with any certainty? Of course not, no one does. It will fall, but any exact number is spurious precision. Their estimates have large standard errors. They may be decent estimates, but this is the Federal Government operating in a very politicized environment so I'm not going to give them the benefit of the doubt. Their reluctance to say what these numbers are, on average, for the 12 lending areas, suggests they are not confident enough to defend them.

You might as well say, the number is Y=a*b*X, where a=3.141592654 and b=2.71828182. Gee, you have two fancy numbers in there to many digits, but as "X" is unknown, it really does not clarify anything.


J said...

Say you were hired to carry out some undefined "stress test" and your political bosses demand fancy numbers with many decimals (98.76% will not satisfy them). You know that the stress thingie is an exercise aimed at increasing the public's confidence in the banks.

Could you do a better job?

J said...

Josh Rosner, an analyst at Graham Fisher & Co.: “The most significant numbers provided by the Fed in the paper appear to be the page numbers.”