Base Correlation Curve for CDO's

SurferD · April 16th, 2004, 9:59 am

Hi all,does anyone have experience using base correlations (i.e. correlations quoted for 0-3, 0-6, 0-9, etc. then a 3-6 is a 0-6 minus a 0-3, etc.)?Regards,SurferD

Jet · April 16th, 2004, 11:33 am

I heard that the detailed technical paper and a sample spreadsheet on base correlation by JPM will be published soon (hopefully today or early next week).

Yeren · April 16th, 2004, 11:39 am

Letest issue of "Derivative Week" has a paper called:Introducing Base CorrelationThis may be related to your topic.Yeren

hojdard · April 18th, 2004, 4:37 pm

Hi SurferD,I think the paper was already published about a month ago (at least that's when I've read it).The idea is not completely new - some houses use it for hedging and MtM for some time. However this is the first paper I know that went public. The methodology can have a significant impact on Greeks compared to implied correlation approach (3-6, ...) that are being market standard. I am afraid this topic is yet too new/proprietary to be discussed openly on the forum...;-(Rgds,

quantman · April 20th, 2004, 9:16 am

Here's JP Morgan paper on base correlation, guys !Hope this helps !

leemcg · April 22nd, 2004, 12:11 pm

HelloI am the primary author of the stuff on Base Correlations at JPMorgan, and I would _appreciate_ a wider discussion. We wrote this because we believe that a market standard is essential for this market to grow. It was thrown out to start discussions with clients and other dealers, not as something proprietary to us.The document attached above is a long client friendly piece which includes introduction to correlation. I have attached a subsection which is much shorter and focuses just on Base Correlations which I think may be more appropriate to this audience.On the subject of a spreadsheet and model doc, we have a small problem here because we work for such a large firm. Basically, we have to get (clients) to sign an "as is" form before they recieve the model (just says that it's for information and education only etc.). I think it should be okay to send this out to people who aren't clients, but need to check.Feel free to dive in with comments/questions/criticism.RegardsLee

quantman · April 22nd, 2004, 12:18 pm

Thanks for your feedback, Lee !I'd really be interested to look at a spreadsheet explaining the methodology, it could be insightful.I'll hope your management will be fair enough to authorize you to share that piece of work for educationnal purposes !Regards.

leemcg · April 22nd, 2004, 12:27 pm

It's not so much management, but legal and compliance.The spreadsheet is _only_ for educational purposes, but we need to get people to sign that they understand that first. I'll see what I can do.RegardsLee

SurferD · April 22nd, 2004, 3:48 pm

Your help is highly appreciated! I have a simple question: how arbitrary is it to start with 0-3 and not with say 30-100 (assuming it would be observable)?Regards,SurferD

hojdard · April 22nd, 2004, 4:19 pm

1. 30/100 is observable: but with much less frequency than 0/32. you want to use correlations in an arrea where some trading happens (0->22 let say) 3. and finaly it is arbitrary - so it is just a suggested convention: note that if you have 30/100 you have 0/30 so e.g. you can work out 3/6 as either 3/100 - 6/100 or as 0/6 - 0/3.Cheers

leemcg · April 22nd, 2004, 4:31 pm

It is all about liquidity and standardisation. This process only really works if there is price discovery from the market, rather than model derived prices.In this case, whilst you could come to any dealer and get a price on any tranche, there are certain tranches that have become market standard, where liquidity is concentrated. These happen to be 0-3, 3-6, 6-9, 9-12, 12-22 in most of the world, and 0-3, 3-7, 7-10, 10-15, 15-30 in the US. These are entirely arbitrary. Interesting though that even on different indices (iBoxx vs DJ TRAC-X) the same attachment points are used.Choosing to start at 0, and not 100 is another arbitrary choice - we have done this because the 22-100 tranche (or 30-100 in US) is not typically actively quoted. (Note that the compound correlation on this tranche, were it quoted, is a further nail in the coffin of that methodology)Hope that helpsLee

smellslikekimchi · April 22nd, 2004, 10:45 pm

Is this the right way to approach pricing off-the-run tranches?1. Interpolate correlations for 0-start% and 0-end% tranches2. Calculate the difference in the expected loss of those two tranches using the interpolated correlations3. Find the correlation that makes the expected loss of (start%-end%) equal to the diff in expected loss4. Using this correlation, calculate the break even coupon for (start%-end%)Thanks Lee, your version of defining the correlation skew seems to be much more intuitive.

leemcg · April 23rd, 2004, 6:33 am

Actually its slightly easier as there is no need for stage 3. Once you have the expected loss for the specific off-the-run tranche you are looking at, this can be used directly to calculate the coupon.In other words, the correlation input helps us build the distribution of losses and hence expected loss for a tranche.RegardsLee

SurferD · April 23rd, 2004, 8:35 am

I have another simple question: do you expect the base curve to move around when spreads get wider/tighter? Thanks a lot for your help,SurferD

leemcg · April 23rd, 2004, 9:12 am

That's the next big question we are working on. Unfortunately there just isn't enough real data (that is tranche spreads and Base Correlations from supply and demand based prices, not models).On the limited data we have, there seems very little relationship, but since August last year, spreads have moved very little.RegardsLee