Friday, February 27, 2009

ISDA Open Sources CDS Model, Issues Challenge To White Hats Everywhere

In a historic event that went largely unnoticed, last night ISDA disclosed it is open sourcing JP Morgan's legendary CDS Standard Model which it got ownerships of on January 29, thereby issuing a challenge to the global community of financial white hats to decompile the code and figure out just what the voodoo is the wizards in JP Morgan's Quantitative Research group have put together over the ages.

The premise for open sourcing is "is to enhance transparency and to optimize use of standard technology for CDS pricing, thereby helping promote the development of the credit derivatives industry as a whole. We are very pleased to be able to provide this service freely to the entire industry." This occurs on the back of the decision to appoint Markit as administrator of the code, several weeks ahead of the March 20 transition of CDS trading as a standardized contract (i.e. pts upfront on every trade). ISDA has announced it will also disclose model inputs such as recovery value and yield curves in due course. Curiously, ISDA will also launch an "online discussion form, which will provide for community input."

That last bit should make people very, very nervous. ISDA is in effect disclosing the CDS model is imperfect, and likely has inaccuracies and inefficiencies. That would make sense: after all it was initially the rushed product of a few finance/math Ph.D. who with the input of a few credit traders had to cobble up something that they could use to give to accounts, who would in turn use it to trade CDS and pay JPM the bid/offer spread. Over time, overcaffeinated analysts had to make adjustments on the fly, likely with no second set of eyes supervising, due to the escalating complexity of the code. And as any first year IBanking analyst knows, taking a complex model and layering more complexity to it increases the likelihood of terminal errors in output delivery to a near certainty.

So white hats the world over are (or if they are not, should be) decompiling the code and trying to find where JPM goofed, in order to claim a stake in the financial hacker pantheon. After all the incentive to do so is huge - the most recent indication of the total gross notional size of CDS outstanding was roughly $25 trillion. Discovering an error that would lead to even a 1% mistake in CDS model output would imply a $250 billion correction to global P&Ls! And being all too aware of subconscious optimism bias, it is likely that the mistake will not be in the P category.

All those who can't wait to start going thru the code line by line can find it here.
Sphere: Related Content
Print this post


Gentlemutt said...

Net pnl on that $25 trillion is roughly what?

Assume a negative surprise of 5-10% on the net, not on the notional, methinks. Still bad, granted, for those on the sorry end.

Anonymous said...

I looked at the code here - there are only 51 files of source code, totaling less than 1 MB of code. Having worked with a lot of legacy code, I think the likelihood that all of the code hasn't been extensively checked is low given the small code base.

One of the files is called ""badday.c" and has a function, JpmcdsBad2GoodBadDayList, that "converts a bad day to a good one (if it's in list)." We could all use that!

Mike Sankowski said...

Also, CDS are derivatives. I think every L has to have a P somewhere. Not that it would be pretty for the L side...

Conrad said...

I recently came across your blog and have been reading along. I thought I would leave my first comment. I don't know what to say except that I have enjoyed reading. Nice blog. I will keep visiting this blog very often.