CPDO’s – Yelling Fire in a Crowded Theater

Posted by on Nov 12, 2012 in The T-Report | No Comments

Ratings and Rating Agencies

Before going into the details of how CPDO works, it is worth talking about rating agencies, what they are, and what they do.

I think in most cases they do a good job, yet I would never rely on them. I think their “single name” issuer ratings have had a pretty decent history. This is the part of the business where they tend to get access to management and information that isn’t always generally available. They have missed some fraud (but they aren’t alone in that). They have some quirks where they tend to be reluctant (in my opinion) to have companies cross from Investment Grade to High Yield. The fact that being investment grade, or high yield, or even AAA is so important, is something we need to revisit.

Before moving on to the structured side of the business, it is worth spending a couple more minutes on the corporate side. What is the rating of a holding company versus an operating company? What is the difference in rating between senior secured and unsecured debt? Such simple questions have no simple answer. All of the rating agencies attempt to rate “probability of default” rather than “probability of loss” for their corporate issues. They can argue that holding companies are more likely to default than operating companies, or that secured debt is less likely to be defaulted on. Many investors like secured debt of junk bond issuers (leveraged loans) because they feel the rating precludes a lot of investors from buying, but that the rating overstates the risk if the collateral is good. I’m not sure there is a right answer here, but the distinction of what they are rating becomes apparent as soon as you move into the structured world.

In the structured ratings world, most ratings are tied to probability of loss. Various scenarios are run and some expected loss and stress loss is calculated and that is mapped to a rating. It is very methodical, generally is conservative, but here loss severity plays a key role in the rating. It is more explicit than on the corporate ratings. That is somewhat irrelevant to the overall argument, but worth noting because it does highlight the general differences between structured ratings – mechanical based on no specific additional knowledge, and corporate ratings – personal and subjective with additional company specific knowledge.

The structured side of the business has had far more problems. It is here that CPDO, Leveraged Super Senior, Super Senior, Sub-Prime, CDO Squared, ABS of ABS, SIV’s, etc. were created. Some of the problems were very technical in nature (the following is my opinion of some of those issues)

  • “thin” slices have big loss severity if there are defaults, and that was a problem in CDO squared, and in the legendary ABX BBB trades. Pools of BBB all tended to default together, and because each tranche in the underlying deal was thin, if it experienced defaults it was unlikely to recover anything
  • BBB tranches that never had better returns than the “equity” below them
  • Liquidity seems to have constantly been overestimated, which was a driving force in SIV’s, Leveraged Super Senior, and ultimately played a role in CPDO. This risk, or at least the margin calls associated with it came back to haunt them on the corporate side as it was the downfall of AIG and the monolines.
  • A reliance on ratings as a predictor of default risk rather than spread, when in many cases, spread was showing a much higher risk, that often proved to be the correct indicator, particularly in the short term

Some things that went wrong just went wrong. For all the complaints about sub-prime, it was hard to find many investors who though housing prices could go wrong. So that was a mistake, but it was hard to see at the time.

So, in general I’m no enamored with ratings, but don’t think they are to blame for everything, but in the case of CPDO, as you will see, I thought, even at the time, that the ratings made no sense and were likely to cause problems.

You Can’t Yell Fire In a Crowded Movie Theater

There are limitations of free speech. There are certain things that you cannot say and expect protection. If you could say whatever you wanted to, there wouldn’t be proceedings related to libel and slander. So free speech does have its limits. Before getting into the specifics of why CPDO in particular seemed so bad, there have always been some things that make me question the validity of the free speech argument:

  • The ratings agencies can earn the Nationally Recognized Statistical Rating Organization (NRSRO) designation from the SEC. The SEC not “Better Homes and Gardens” awards this designation. That seems like it should mean something. The NRSRO designation isn’t easy to get as I can’t get it, Goldman Sach’s research department can’t get it, and even Egan Jones had some recent troubles with the SEC on parts of their designation.
  • Ratings from NRSRO impact regulatory capital. Banks can choose to use internal models, but the fact that regulators and regulations ranging from NAIC (U.S. Insurance companies) to Basel (for global banks) allow ratings from NRSRO to affect regulatory capital seems also to mean they have a higher standard (in addition to laziness on the part of regulators)
  • Many investments are required to hold assets of a certain rating quality. This is particularly noticeable at the AAA level and the investment grade versus high yield level. While many of these decisions were done of their own choice (including money market funds and many government run investments) it would seem that they looked to the regulators use of rating agency ratings as an indication that they were doing the correct and “prudent” thing.
  • You cannot mislabel products. You can’t purposely mislead people. Nutella got in trouble for being a “healthy” choice, in spite of the fact that it was chocolate. They argued over what healthy meant. But in general, people are protected from sellers of a product making unsubstantiated claims in an effort to sell something. You can call something AAA to sell it and that is fine, but label a sugar pill as a diet supplement and you can face fines?
  • Then there is product liability. My old Saturday Night Live skit about “bag o glass” as a kid’s toy is still my favorite. You can’t sell shards of broken glass as it’s dangerous. Heck, you can’t serve someone coffee that is too hot in case they spill it on their lap, but a horribly devised product, that was obviously wrong, and would not have passed back testing, is fine?
  • You cannot falsely yell “Fire” in a crowded movie theater as it serves no public purpose and is dangerous.

Why CPDO Looked Attractive (the Infomercial)

Before going into the problems of CPDO, it is worth looking at why it looked attractive and why it seemed to work.

We will look at a simplified version as it explains most of the concepts without losing too much.

The first thing to remember is investment grade CDS spreads were about 40 bps for 5 year CDS at the time. So you could earn 0.4% per annum for selling investment grade CDS on average. Bonds traded almost as tight as the market was doing well. Sub-prime wasn’t an issue, AIG was alive and well, and you could borrow as much as you wanted in the repo market to fund bonds. There were banks, and even hedge funds, that ran “basis” books in the billions and even tens of billions of dollars. Credit was cheap.

So let’s create a CDS index, call it CDSI. It has 10 names (the real ones had 125 names, but this is easier to explain with a smaller index).

You could sell protection on this index and earn 0.4%. But let’s say you are a “cash” investor, so you need a funded or “structured” note. You could buy one linked to the CDSI and receive LIBOR + 25 on a BBB rated note.

  • It is only LIBOR + 25 because the entity selling the note charges LIBOR – 5 bps on the funding, and the dealer is trying to make 10 bps on the index. Not exact but some variation of that is at work.
  • It is rated BBB or maybe BBB+ because the WARF (weighted average ratings factor) of the 10 names gives that rating.

Now along comes your friendly neighborhood structured salesperson and tells you they can offer you a structure note that pay LIBOR + 150 and is AAA rated. WOW!

So in an environment where you can earn L+25 on a BBB structured note linked to the index, here is a product that can pay you L+150 and is AAA rated! And wait, not just the coupon, not just the rating, but if the deal performs well, you get a kicker at the end! So, the coupon, the rating, and an extra payment at maturity, but wait, that’s not all! This is all based on the same index that the other unleveraged structured note is base on! Impossible you, but no it isn’t. All that was missing was the ginsu knives.

How CPDO Worked (The Alchemy)

So how is this miraculous trade created? How do you transform L+25 BBB notes into L+150 AAA notes with upside?

The deal references the CDSI index (in this case our hypothetical 10 name index).

It uses 15 times leverage. So for every $10 million of exposure taken via the note, the investor takes risk to $150 million of CDSI.

The immediate and obvious question is how do you take something that is BBB and leverage it 15 times and turn it into AAA?

This is where it is both clever and complicated, and also where the mistakes come in.

  • Excess Carry. The first thing to note is that the index earns 40 bps (or did at the time). When you multiply that by 15, you get 600 bps of income to play with. If the dealer takes 100 bps as fee, and another 100 bps comes out as various deal fees, you are still left with 400 bps. But the coupon on the note is only 150 bps. So there is 2.5% per annum of excess carry. This money is meant to accumulate and protect against future losses. This part makes some sense. It basically relies on the fact that few investment grade companies default, so saving up some income and using that as a form of “portfolio” insurance makes sense on the surface. There are lots of reasons that it doesn’t stand up to scrutiny but that is for later.
  • Spread Not Default Risk. The other innovation was that the deal didn’t just sell protection on the “on the run” index. It wasn’t a strategy of selling CDSI and sitting on the trade for 5 years. The trade was to “roll” the protection into the new “on the run” index. Since the index, when it is created, has to be investment grade, it meant, that you only had “6 months of jump to default risk”. The indices rolled every 6 months. So you sell protection today on CDSI 1 and in 6 months would buy that back and sell protection on CDSI 2. Since CDSI 2 had to be only investment grade names, you could argue, that you were only taking 6 months of “default risk”. The risk that an investment grade company would default within 6 months of having that rating. Technically correct. This let the rating agencies rely on credit spreads, rather default risk, and is where it all went wrong.
  • Steep Curves. The first erroneous assumption with model was that the CDSI indices would be reasonably “steep”. They assumed that if you sold one Index at 60 bps, the new one would be at a similar or even wider level. This is important but horribly wrong assumption for two reasons we will go into later. In the theoretical model world, this was a huge assumption that benefitted CPDO. Say you wrote $150 million of protection at 40 bps and 6 months later sold it at 50 bps. You lost 10 bps. The duration of the index was about 4.25. So that 10 bp loss turns into 0.425% loss. With 15 times leverage, the note would have lost 6.375% (0.1% * 4.25 * 15). At the same time, the note was accruing Excess Carry and in 6 months would have made 1.25% of carry. So the mark to market on the note would be 94.875% (100% – 6.375% + 1.25%). It is important to note how critical the assumption of steep curves is. Now the model would assume the new index is also at 50 bps. So it earns 10 bps more than before. That means the excess spread goes from being 2.5% per annum to 4% per annum (2.5% + 0.1% * 15). So Steep Curves and Excess Carry work hand in hand to absorb losses in a spread widening world according to the model.
  • Mean Reversion. The next element was a “mean reversion”. They argued, reasonably well, that spreads had a tendency to mean revert. I will give that to them. I think many things tend to mean revert. The big question is what do they revert to? A near term mean? A longer term mean? That can make a difference. It plays a role here too, particularly in the model. I cannot remember the exact mean reversion, but I think they assumed spreads were already tight, so the reversion would be to something higher, let’s say 60 bps. On the surface that seems bad, but it isn’t according the model. If you did a trade at 40 bps and it went to 60 bps you would have a mark to market loss bigger than the one in the example above. But your Excess Carry would be even bigger and you would make more. It got really bizarre once you start going wider than the mean reversion level in scenarios. The trades would do great, because they would get even more excess carry for a period of time, and then make big gains as the CDSI levels reverted and went back from say 80 to 70 to 60.
  • 90% Loss Trigger. Here is the final piece, the piece that no one thought could happen, but did. The problem with running something 15 times leveraged is that you either need to floor the loss to the structured note. The structured note can only lose $10 million. With a $150 million of exposure, the loss could in theory be $150 million. So the note “triggers” if the note is ever only worth 10% of notional. In theory you would need the CDS spreads to go out to about 180 bps (or 140 wider for this to occur). 1.4% spread move * 4.25 duration * 15 leverage = 90%. That is a big spread move. Many saw it as near impossible. In fact with Excess Carry and the way duration works for CDS, it would have had to be an even bigger spread move as a one-time gap. But it is this trigger risk that could knock you out of your trade. 10 year CPDO’s were even riskier because they could withstand less spread widening.

Something So Complex Went Wrong So Simply

After taking a couple pages of the complexity required to understand the model and why it worked (or at least in the right ballpark) you will see how it fell apart so quickly, and frankly simply, and beyond frankly, how it would have failed in 2001 to 2002, just 5 years earlier. That is what is so alarming about the product.

Let’s go back to our 10 name index. Say each name is trading at 40 bps. The index trades at 40 bps. Now let’s say 1 name has a lot of trouble (10% of the portfolio). Whether it is fraud like WorldCom or Enron, or fear like Dynergy, or overbuilding like Nortel, or writing too much protection like MBIA or a dumb business like an airline, or auto finance (to name a few real world examples). Let’s just say 2 names get into serious trouble during that period. They don’t default (I will give the model that one). But let’s say they trade at 40 points up front. So the CDS goes from nice and tame at 40 bps to 40% up front. That would imply bonds gapping down to 60%. The list of names that have seen that is pretty long – I think I forgot to mention Xerox and Conseco in the list of earlier examples.

In this environment, let’s say the average name somehow remained unchanged at 40 bps (unlikely but makes it easier). So 9 out of the 10 names remain unchanged, but 1 name lost 40%. That is a 4% loss on the index. That means the index would have a spread of about 135 bps. No default! Remember, we kept the rating agency assumption of no default, but what they missed was how much a name can move. That is the driver, and again, all you have to do is go back to 2001 and 2002 and see how quickly names can default. It also means this note would have a mark to market of about 40% now (a loss of 4% on the index * 15 is big). If you think 10% of the index universe gapping wider is high, maybe it is, but it happened in the early 2000′s and then again in 2007 and 2008.

But what about the excess carry, reversion to the mean, and steep curves? If that name was removed because it was no longer investment grade, a reasonable assumption, then the new index or CDSI 2 would be at 40 bps. We made the assumption names had stayed at 40. In any case, the key is that the names that come out, especially if downgraded, see a spike in spread, and the CPDO can never earn that back.

Now at a mark to market of 40% of par, and an index at 40 bps, the structure can only sustain a 30% move before triggering. Now that is only a 50 bp move and wouldn’t even sustain 1 name going to “points up front” in a meaningful way.

The Flaws that were Obvious and Should Have Been Caught

  • Low 6 month jump to default risk, doesn’t mean 5 year CDS can’t jump by significant amounts. Time and again, companies, particularly highly leveraged ones, or frauds, have seen gaps in CDS that preceded actual default (or in some cases, recovery).
  • When names gap wider on being downgraded, and you are forced out of those names via the roll, you accentuate the problem, ensuring that you are exiting at worst possible levels and receiving new names that cannot compensate for those losses
  • Curves are steep in good times, but inverted in stress, so even if a name stays in, you may sell protection in the new index at tighter levels than where you are buying it back in the old index
  • The trade itself caused CDS to go to abnormally tight levels, so any mean reversion should have accounted for that (ie, would revert wider because CPDO itself was causing the tightening), and it should have accounted for the spread widening that would be attributed to any potential unwind or even fear of CPDO unwinding which was real (and worked in conjunction with the aforementioned leveraged super senior).

In the end, the rating agencies did a horrible job of assessing spread gap risk. Too much focus on default risk, too little emphasis on correlation, particularly in bad times when it tends to go to 1, and complacency on things like steep curves and mean reversion.

At the time, we went back, and tried to use some combination of TRACERS (the first IG CDS index, which was a Morgan Stanley product) and CDX 1 to see what CDX 0 and CDX -1, etc. would have looked like. What names would have been in, and based on bond pricing and what historical CDS was out there, how it would have performed. It would not have survived 2001-2002. How can any product be AAA, when less than 5 years before it would have defaulted?

What Next?

Who knows, but feel free to contact us, as are happy to discuss our views and thoughts and go into more details.

If you can get your hands on an old Barclay’s report, it is worth reading. They tried to “break” CPDO and couldn’t and concluded that it was robust. The problem was they just tweaked the model with different levels of steepness, volatility, mean reversion, etc., but didn’t take the time to go through the common sense approach of would it have survived recent past, and analyzing the real risk, which was, short term spread movement risk, and names dropping out at the wides.

E-mail: tchir@tfmarketadvisors.com

Twitter: @TFMkts