Regarding my previous post on CPDOs, the delightfully tart Felix Salmon gets delightfully tart with, er, me. He writes:

It would seem that Waldman is rather smarter than anybody at any credit-rating agency.

Now I'm hardly as bright as your average bread mold, so Salmon is being a bit tough on the credit agencies here. Plus, the last thing I meant to do was accuse the rating agencies of stupidity. On the contrary, rating agencies are being quite as clever as the investment-bank CPDO issuers. They are both, in my opinion, playing the same game, which I'll call "Keynesian sound banking", after the Great Man's famous quote:

"A 'sound' banker, alas, is not one who forsees danger and avoids it, but one who, when he is ruined, is ruined in a conventional and orthodox way along with his fellows, so that no one can really blame him." [1]

Or we might refer to this as "Waldmann's Rule" &mdash not my rule, for heaven sake I don't deserve such a thing — but after Robert Waldmann, for what Brad DeLong learned from him:

What I learned from Robert Waldmann: Almost no professional portfolio manager worries about the lower tail, because if you are in the lower tail the whole world has gone to hell in a handbasket and people have other, more important things to worry about than whether one's portfolio manager had appropriately hedged whatever risk is now roosting on the roof.

My claim is only that the same logic applies to credit-rating agencies and banks. I'm certain the best and brightest at the rating agencies thought of everything I thought of. Rating agencies earn revenue (from the issuing investment banks) when they get to rate a booming new class of credits. Rating agencies get egg on their face if an issue they rate highly defaults. But if that happens in the context of a widespread credit event? Well then it's like Condi Rice and the World Trade Center. Who could possibly have foreseen terrorists flying planes into buildings!

Here's Salmon:

One of the things which makes the CPDO model so robust is that the riskiest risk that it's taking is six-month investment-grade credit risk. Since it's pretty much unheard-of for a company to go from investment-grade to default in less than six months, the rating on the CPDO can be very high. What's more, the CPDO, because it has leverage to spare, can continue to pay out its coupon even if that kind of default does happen.

Model risk is precisely the possibility that "the pretty much unheard-of" occurs. Plus, lightning fast "gap risk" defaults aren't required to break CPDOs. A sequence of general credit-quality deteriorations over several rebalancings of the CPDO portfolio would be sufficient even without default.

During a widespread generalized credit event, or following a sequence of periods during which credit conditions continued to deteriorate rather than reverting to mean, CPDOs would no longer have "leverage to spare". "Leverage to spare" is what Brian Hunter at Amaranth had for the first few meters as the bottom dropped out on the natural gas market, in a pretty much unheard of collapse.[2] Any strategy that involves continually increasing leverage to cover losses is exposed to multiple adverse movements in sequence. That is why the strategy is broadly referred to as "gambler's ruin". Gambler's ruin can be a rational and very profitable strategy, but the whole game turns on the precise likelihood of long series of adverse events. Estimating that likelihood requires a model, and getting the model even a little bit wrong can be the difference between sure profit and sure ruin.

But, Salmon demurs...

[R]atings agencies try very hard to understand every single way in which the model might break, and then stress-test the model under precisely those conditions.

He's simply wrong here. If ratings agency tried to take into account every way their models might break, they would be unable to rate. Ratings agencies are quite aware that there are questionable baseline assumptions they have to make in order to come up with a model at all. They publish their assumptions, and leave it to investors to accept their models or not. Here are some snippets ABN/AMRO's presentation of Surf 100, the first CPDO (now a venerable month or so old):

• Current modelling assumptions are unlikely to be consistent with actual performance of CPDO
• Key modelling assumptions are set out in S&P/Moody's base case assumption

[...]

S&P base case assumptions

...[10,000] Monte Carlo simulations ...Expected defaults produced by CDO Evaluator 3.0 ...Initial portfolio spread of 32bps with a volatility of 15%, meanreversion (MR) = 40bps at the end of year 1, and MR = 80 at the end of year 10

Moody's base case assumptions

...[10,000] Monte Carlo simulations ...Expected defaults produced by CDOROM

Here's Fitch [3]:

In recognition of the sensitivity of credit CPPI and CPDO to spread widening and volatility, Fitch models spread path as follows:

• exponential Vasicek model;
• parameters based on stressful historical periods;
• back-testing on historical data;
• spread jumps incorporated if necessary

In all cases, the ratings agencies are being very honest with us. They are pointing out the limitations and assumptions of their models and tests. In no case do these tests qualify as "every single way in which the model might break", and the rating agencies don't pretend that they do. Why does Salmon?

I wrote that...

CPDOs appear to violate the core constraint of finance, the no arbitrage rule. If the ratings are accurate, selling short a portfolio of ordinary AAA debt and purchasing a portfolio of CPDOs would be a perfect arbitrage, earning risk-free profit for the arbitrageur with no net outlay of capital. Either the CPDO opportunity must be transient (because the number of issues that can be synthesized is limited, or because CPDO and AAA yields will soon converge), or the ratings must be wrong. Or else the wizards at ABN have invented an infinite free-money machine for well-placed arbitrageurs, the financial equivalent of a perpetual motion machine...

Salmon responds, and but again mistakenly.

And as for Waldman's ratings arbitrage, where you go short French sovereign debt and go long CPDOs, yes, it does exist – but it's not "the financial equivalent of a perpetual motion machine". Rather, it's just another carry trade. CPDOs are much less liquid than French government bonds, so they should carry a yield premium. Plus, the carry trade can move against you: if the price of CPDOs falls while the price of French government debt rises, you take a mark-to-market loss. And finally, the trade isn't very profitable in any event, since you have to borrow those French bonds somewhere, and the repo rate isn't likely to be much less than the extra spread you're getting on the CPDO.

If indeed a CPDO AAA is statistically indistinguishable from an ordinary AAA, and if indeed a CPDOs consistently earn spread above ordinary AAA debt, then this would not not an ordinary carry trade. Going short a particular issue and long some CPDO would be a carry trade, as there would be price risk related to idiosyncracies of the two securities. But if a diversified portfolio of CPDOs (presuming the asset class takes off) behaves identically to a diversified portfolio of other AAA debt, then highly creditworthy financial institutions (not you, me, or your cousin's small hedge fund) would indeed have a perfect arbitrage, until the spread between CPDO and AAA debt converges. This won't happen, because it is a carry trade, there are different risk profiles to a diversified portfolio of CPDOs and a diversified portfolio of other AAA issues, and that difference is... CPDOs are riskier! And that's exactly my point. CPDOs are risky issues that earn risky spreads, but look for bureaucratic purposes like conventionally "risk-free" debt.

I should comment on my use of the word "risky", both in this and the previous piece. Salmon takes me to task...

First, on a factual level: Waldman says that the proceeds from CPDOs go into "a leveraged portfolio that includes high yield, risky debt" – which isn't true if by "high yield, risky debt" you mean sub-investment-grade debt. The portfolio is all investment-grade; it just isn't AAA.

I don't mean, and didn't previously mean, that current CPDOs are taking on "junk bond" risk. One could be forgiven for thinking my use of "high yield" implied that. I should have said "higher yield". By "risky", I mean non-AAA, since AAA is the category which is generally treated as nearly free of credit risk. The debt behind current CPDO issues is concentrated at the low-end of investment grade.

Finally, back to Salmon:

But if you're still not comfortable with that kind of risk, no one's forcing you to take it.

When banks use novel structured products to take on more risk than bank regulators anticipated, I am being forced to take that risk. We all are. Bank regulation is a complicated subject, and I don't claim that existing bank regulation is anywhere near optimal. I don't disagree with Salmon's contention that while CPDOs might let banks game things a bit, this new gaming is far harder than it was under Basel I. There's a cat and mouse game being played, between regulators getting tighter and banks getting cleverer, and that's perfectly ordinary in its way.

I bother to write about this stuff not because I am interested in the arcane details of a structured credit designed especially for bank investors. I write because I think the game is going awry, the odds of systemic crisis at the collapse of a credit bubble are growing, and CPDO-based regulation skirting is the latest little crack in the dam. Reasonable people can disagree. Salmon clearly does, and he's reasonable even if he is delightfully tart. But you can't pretend that Moody's has worked it all out and we can rest comfortably. There is no adequate model here. There is human judgment. Me, and Paul Volker, and Robert Rubin, a lot of us are worried. And those other guys are much smarter than bread mold.

By the way, if I were Goldman Sachs, I would short dollar-denominated CPDOs and purchase US Treasury debt. CPDOs aren't really financial perpetual motion machines. They just get to look like it, for about two seconds.

Update: Felix responds.