For a complete list of Beginners posts, see Financial Crisis for Beginners.
This is more of an advanced beginners topic – I already covered CDOs (collateralized debt obligations) in my first Beginners article – but I imagine that most of our readers are already familiar with structured products. At least, many people know that first a bunch of securities are pooled together, and then they are “sliced and diced,” in the common media parlance I find incredibly annoying. But Joshua Coval, Jakub Jurek, and Erik Stafford have a new paper, “The Economics of Structured Finance,” which does a brilliantly clear job of describing what these securities are and why they were so widely misunderstood, with the results we all know.
The paper is 27 pages long, not counting references, tables, and figures, and if you are comfortable with probabilities and follow it carefully you can understand everything in it. I will provide a summary to whet your appetite. I am not going to use numerical examples because the examples they use throughout their paper are so good.
The key to CDOs is that they could be used to manufacture AAA-rated securities out of underlying securities (like mortgages) that were not even close to AAA. (“AAA” is a bond rating, meaning that the security in question had about a 0.02% chance of defaulting in a given year.) This is well known. But although these new, synthetic securities had expected default rates comparable to traditional AAA-rated securities, they had other properties that were unlike their traditional brethren, having to do with (a) correlations between the underlying assets and (b) sensitivity to underlying default rates. (a) is the probability that, if one mortgage inside a pool defaults, the other mortgages will also default; (b) is the degree to which small changes in those default rates can affect the expected value of the manufactured AAA securities. This meant that these CDOs were much more sensitive both to errors in estimating their characteristics, and to macroeconomic changes, than most people realized.
If you didn’t follow that I’ll go over it again more slowly.
In a simple, “pass-through” securitization, each investor in the pool of mortgages has an equal claim to the mortgage payments. Therefore, the expected loss for each security is exactly the same as the average default rate of the mortgages.
In a CDO, the investors have unequal claims. By creating some junior tranches (“tranche” is French for “slice,” in case you were wondering) that absorb the first losses, you create a large senior tranche that is buffered and suffers no losses until all of the junior tranches are completely wiped out. This is why the senior tranche can get a AAA rating; the estimated chances are pretty low that enough people will default to wipe out the junior tranches. In a CDO-squared, you take some of the junior tranches of ordinary CDOs, pool those, and then create tranches out of that pool. The amazing thing is that you can then create not only a senior tranche but a mezzanine (middle) tranche of your CDO-squared that has the expected default rate of a traditional AAA bond, even though it is made out of junior tranches. (There are very clear examples of all of this in the paper.)
However, this only works well if the default probabilities of the underlying mortgages are not highly correlated. Assume in an extreme case that defaults among the underlying mortgages are perfectly correlated: either none default or they all default. In this cases, the tranches do nothing for you: if all the mortgages default, then the senior tranche gets wiped out along with the junior tranches.
Furthermore, the performance of AAA-rated tranches is highly sensitive to the default rates of the underlying mortgages. Conceptually, this happens because the amount of protection provided by the junior tranches is not that much bigger than the expected default rate; so if the actual default rate is just a little higher than expected, a much larger proportion of the protection will get eaten up. In the example in the paper, an increase in defaults from 5% to 7.5% can knock a AAA-rated tranche of the CDO-squared down to a BBB- rating.
The conclusion is probably apparent to many readers at this point. The underlying mortgages were more highly correlated than people thought, both because they often came from the same types of developments in the same regions (California, Nevada, Florida), but also because everything in the economy became highly correlated. And as the economy got worse, default rates climbed higher than estimated based on historical data, because all the historical data came from a period when housing prices only went up. While this would only have a “linear” impact on a simple pass-through securitization (double the defaults, double the losses), it had a “non-linear” impact on AAA tranches of CDOs, and especially of CDOs-squared.
Finally, there is one more misunderstood characteristic of CDOs. Securities with the same expected payoffs, and hence the same rating, can have different characteristics. In particular, they can differ in their degree of correlation with the rest of the economy. The authors cite catastrophe bonds (which default only, for example, if a hurricane hits South Florida) as securities that are uncorrelated with the economy. Because of their lack of correlation, they are more desirable than other securities with the same rating, and hence have lower yields (higher prices). Senior tranches of CDOs are just the opposite: they only go bad if the economy as a whole goes bad; that is, they are highly exposed to systemic risk, which almost by definition is difficult to quantify. Because of this high degree of correlation, investors should have demanded higher yields (lower prices). But because investors by and large thought that all AAA securities were comparable, they didn’t demand high enough yields, and the issuers (investment banks) made the difference.
The bottom line is that all AAA securities are not created equal – even if they have the same estimated probabilities of default. And treating AAA tranches of CDOs and CDOs-squared as if they were AAA corporate bonds played an important role in the growth of the structured finance market and, as a result, the overall asset bubble that is collapsing around us.
Update: A reader points out that CDOs generally do not contain mortgages, but instead already contain subordinated (not senior) tranches of mortgage-backed securities or of other CDOs. And the mortgages inside those mortgage-backed securities are often subprime. Because these tranches are themselves unique, it is even harder to project the cash flows coming into a CDO. A mortgage-backed security backed by a pool of plain vanilla mortgages, by contrast, would be relatively easier to value and less subject to price fluctuations.
By James Kwak