Now I know AI's readers love to dive deep into the data, so here is Moody's default rates from 1960-2006.
The tiny blue lines are actual investment-grade defaults. That never gets above 0.51%. I've only included that because if I didn't, someone would ask. The relevant number for IG investors is the "All-Rated" number, which is in red. See, if you buy an investment-grade bond, odds are good that it would get downgraded to junk before it actually defaults. So looking at IG defaults only sort of hides the reality. The All-Rated number basically shows you defaults as a percentage of all bonds. Now that probably overstated IG defaults, but better to be conservative in a study like this.
Anyway, I've highlights three spikes: 1970, 1990, and 2001. This supports the idea that defaults tend to be centered around periods of poor economics. Not surprising.
My analysis was based on a 4.6% 10-year cumulative default rate. This is based on the default rate for bonds which were rated Baa at the beginning of any 10-year period within Moody's study period. Given that none of the actual spikes in my graphic are over 4.6% (and that the graphic includes all bonds), I think my spike analysis is valid. Its at least a fair illustration.
So anyway, I've asked the Anonymous poster to elaborate on his study, because it sounds like he's done some good work, and yet come to a conclusion that's radically different from mine.
If anyone would like an Excel copy of Moody's 2006 default study, e-mail me. accruedint *at* gmail.com.
Hey AI, thanks for your interest. As you mentioned if we use IG only (from Moodys) then we do get around a 30-40bps spread expectation but I preferred to use the 'All' as it better reflects the static nature of the CDX indices (unmanaged!). The approach is simple enough - generate rolling 10Y curves of actual default rates, push this into a standard CDS pricer as the hazard rate process with an adjustment for the recovery based on the relationship between default and recovery (this adds some spread). Then price a 3,5,7, and 10Y CDS at par based on this hazard rate and recovery.
ReplyDeleteThis gives a recessionary peak spread between 120 and 180bps depending on severity and recovery expectations. 140 is 40% recovery and previous two recessions. Lots of holes can be poked in it but it gives me a sense of what is fair especially for an index such as CDX that is more crossover than good IG. Certainly IG9 with CFC/RESCAP etc. should be wider even with its shorter maturity. This is also a real-world spread NOT a risk-neutral spread which would inevitably be wider.
I think your 34bps is good floor on 'current' IG spreads (given the weak outlook for the economy and rising default expectations - on the latter point, regress the Senior Loan Officer survey against default rates for some fun). Plenty of stuff in IG CDS land trading at these levels - FNM 5Y at 45bps! (don't like the short but do like the senior-sub decompression).
Have you looked at the MCDX which launched today - AA/AAA muni CDS index trading around 45bps last run i saw. compare that to the IG CDX at 95bps - 50bps differential from AA/AA muni to BBB corporate - i say spreads decompress further - any thoughts?
There is a post coming on the MCDX. I think fair value is a good bit lower than the initial coupon of 35bps. My bet would be that technically, there will be more buyers than sellers initially, pushing the spread wider for a while.
ReplyDeleteBank of England numbers are a lot closer to 30-40bp than they are to 145bp for pure default risk, although default risk uncertainty takes it up a bit higher.
ReplyDeleteI blogged about it.
James: That's a great piece. Thanks for linking it.
ReplyDeleteAre you comparing the 4.6% 10y cumulutive default rate from your modeal to the annual default rates for all rated bonds from the chart? Or are the rates in the chart also 10y rates? Seems like these are apples and oranges unless I'm misunderstanding.
ReplyDeleteNo the chart is actual default rates year-by-year. It is a little bit apples and oranges. I just trying to say that my 4.6% peak default scenario was within the bounds of reason.
ReplyDeleteIn response to Mr. Hymas comment, i would add that it is a well known fact that Merton-style structural models (even KMV's empirically-adjusted EDF) severely understate higher quality (read IG) spreads. They are most effective at crossover and below and only then should be very carefully calibrated. The best way to use the structural models is as a ranking rather than absolute pricing model (trust me - i've tried pretty much every adjustment there is). Mathematically, the Merton model has MANY weak assumptions and from a trader's perspective the maturity at which its inputs are most accurate (less than 1Y) is the area when the math of the model (both asymptotic convergence to zero and lack of uncertainty) breaks down and the marketing which we trade is the least liquid.
ReplyDeleteI love structural models, use them all the time for capital structure arb or relative-value but do not use them for 'judging' real default risk premia. There is some good research on the strengths/weaknesses of these approaches - search SSRN for Kay Giesecke or Lisa Goldberg.
This is a fascinating discussion, thanks for the forum.
On MCDX, agreed, technicals will keep it wider than intrinsics but i just cant help but feel 50bps has to decompress to reality.
The BoE approach ascribes the difference between the spread calculated from default probability and the spread observed in the market to liquidity and other concerns. This strikes me as being a lot more realistic than attempting to account for every beep in terms of default probability.
ReplyDeleteSeems to me that a reasonably good qualitative ranking is the best that we can possibly hope for when attempting to forecast the future and - if it's, say, 60% accurate - is well worth the effort.
Any references you could dig up comparing ex ante default probability with actual subsequent experience over a significant dataset would be greatly appreciated.
I see 3 spikes. A small statistical sample from which to draw extrapolations. May be interesting to expand the sample set going back further or using other economies.
ReplyDelete