I emailed Mike Peterson, the editor of Creditflux, to ask him about the "risk
weight" figures we say yesterday. (If you recall, the risk weighting
on synthetic CDOs went up from 2.89 in 2005 to 3.45 in 2006, but it wasn’t clear
what that meant.) Mike explained:
We risk weight each synthetic CDO tranche by multiplying it by a factor depending
on its seniority. So anything between 0% and 3% is multiplied by 20, anything
between 15% and 100% is mutliplied by 0.4, etc. It’s a crude version of a
delta.
Is that clear? Let me try to explain. Synthetic CDOs are structured to behave
just like normal CDOs, where you bundle up a bunch of bonds and then tranche
them. The top-rated tranches get paid first. Once they’re paid, the next tranche
down gets paid. And so on.
The way that Mike is measuring the seniority of these tranches is by looking
at something called the "attachment and detachment points". He explains:
Any tranche with [loss] attachment and detachment points between 0% and 3%
of the portfolio. If you attach at 0% and detach at 3% you get hit from the
moment of the first loss in the portfolio and by the time losses get to 3%
you are finished.
A first-loss tranche such as that one counts as an equity tranche, and carries
a risk weight of 20 in Mike’s scheme. On the other hand, if you only start suffering
losses once 15% of the portfolio has been wiped out, then you’re safe as houses,
and Mike assigns your tranche a risk weight of just 0.4.
So now we’re a bit closer to understanding what those risk-weight numbers mean.
Let’s say you took a barbell strategy on a synthetic CDO, buying an almost-risk-free
tranche from 15% to 100%, and an equity tranche from 0% to 3%. Then the average
risk weight would be (85*0.4+3*20)/88 = 1.07. That compares, I think, to a risk
weight for the all the tranches combined of 1. (I do hope that Mike will correct
me if I’m wrong here!) In other words, if you have all of the very safest tranches,
plus the entire equity tranche, you’re still taking more risk overall than is
in the structure as a whole – because you take all of the first loss.
Now the clever thing about synthetic CDOs, unlike normal CDOs, is that you
don’t need to actually sell every tranche from 0% all the way through to 100%.
And in fact there’s really little if any demand for the safest tranches of synthetic
CDOs, since it’s possible to get similar returns with better liquidity elsewhere
in the fixed-income world. So increasingly people have been buying just the
riskier tranches of these animals, with detachment points in the single digits.
And that explains the average risk weight going up to 3.5. But remember that
equity is 20, so 3.5 is still relatively safe, and I’m not sure I’d
describe it, as the FT did, as "ultra-risky".
So 60% of the risk is in the equity tranche? And another 34% (0.85*0.4) is in the supersenior? Leaving an average of 0.5 for the levels in between? So if 5% to 15% is 0.4, then 3 to 5 averages 1, and 0 to 3 averages 20? Doesn’t this seem a bit dramatic?
Is this independent of the quality of the underlying credits? Or are these mostly sold on investment grade credits, and that’s what’s being quoted here?
Really, really, really good question. For which I don’t have an answer, beyond this email from Mike: