Monday, November 10, 2008

AIG Shows Guts of CDO portfolio


I'm not sure if they have published this before, but if you take a peek at appendix A (under the Credit Presentation link), you get to see what's inside some of the CDOs that carry AIG's default protection.

It is amazing (to those not in the know) the range of collateral that is in some of these. There is no general 80/20 rule obviously at work (80% good stuff, 20% quasi-speculative stuff). I'm not sure if that means anything, other than a lot of confidence in modeling the underlying risk of instruments and less attention to broad, "commonsense" ways of mitigating loss (through apportionment).

Only $16.4B of the original "AAA" underlying portfolio amount of $55B remains rated "AAA", less than what is now "junk" ($18B).

CDO36 had 53% "CDO", which I take to mean AIG were plenty comfortable with so-called "CDO-square" risks (and how little look-through transparency this kind of report might yield for that). Indeed, they model these in their "stress tests" by calling this collateral "inner CDOs", as I understand it, to designate those "held inside". (I cannot judge whether their assumptions are aggressive or not for "inner CDOs").

As I understand it, the "attach" and "detach" points are meant to indicate the "stop" and "start" default-loss levels between which a credit insurance writer is 'on the hook'. Given the wide range of collateral, it's almost impossible, it seems, to make a general, eyeball characterization about just how optimistic they were in setting these cutoffs when they wrote the CDS.

They report $120B less in net exposure than gross exposure on the residential mortgage CDS portfolio - it's not clear to the untrained what they are netting out, exactly, collateral or offsetting CDS.

Projecting their net exposure is difficult (there are "2a-7 puts", afterall ...). Some of the CDS they wrote to help "regulatory capital" requirements appear like they could be in a steady amortization (and thereby, risk reduction). There was a $6.2 billion - billion - "currency fluctuation" in the multi-sector portfolios net ... don't think you know everything.

Their loss severity on prime, for the same vintage, is half what it is on sub-prime. Not sure why that makes sense, off hand (isn't a default, a default, a default? Do prime have that much more equity on average for the vintages in question?)

They have $31 billion so far in cumulative mark-to-market losses (which they hope will reverse, no doubt) and $12 billion in guesstimated, heavily qualified "stress test" risk in the book(s). Here are two quotes that give you pause that the losses will reverse:

To trigger a default on the most secure subprime RMBS debt - rated AAA, and structured with a typical 18 per cent attachment rate - foreclosure rates would have to reach the 30 per cent.-FT Alphaville, November 7th, 2007

Standard & Poor's Ratings Services' current ratings on the 2006 subprime U.S. residential mortgage-backed securities (RMBS) vintage reflect, in part, an assumed loss severity of 45%. That assumption includes our estimate that, based on our views on the current housing market environment, foreclosure costs and market value declines account for losses of 26.3% and 18.7%, respectively, in the loan balances of these mortgages. We project a 19% aggregate loss for subprime mortgages backing U.S. RMBS sold in 2006 based on assumptions for foreclosure and loss severity. The 19% assumption is the product of projected foreclosures of approximately 42% and the assumed loss severity of 45%.-May 7th, 2008

Paulson's Cash-n-Carry guy isn't the one buying up the CDOs, it's the Fed using a Fed financed SPV, this time. Is this purchase predictable? Preventative? Probably. Could it facilitate an unwind of this credit protection sink-hole? Yes... Could it facilitate the Fed, as counterparty, dropping collateral requirements? Yes... Still, an unlevered $30B looks like just a fig newton, clocking, as it does, at just 1/4 of the multi-sector portfolio's net exposure...

Anyway, if any reader wants, I found an open-source model for risks that I think I'll have a look at, time permitting.

No comments: