Steven R. Dunbar
Department of Mathematics
203 Avery Hall
University of Nebraska-Lincoln
Lincoln, NE 68588-0130
Stochastic Processes and
Advanced Mathematical Finance
A Model of Collateralized Debt Obligations
Mathematically Mature: may contain mathematics beyond calculus with proofs.
How can you evaluate cumulative binomial probabilities
when the value of is large, say and the value of is small, say ?
To illustrate the previous sections altogether we will make a simple binomial probability model of a ﬁnancial instrument called a CDO, standing for “Collateralized Debt Obligation”. The market in this derivative ﬁnancial instrument is large, in 2007 amounting to at least $1.3 trillion dollars, of which 56% came from derivatives based on residential mortgages. Heavy reliance on these ﬁnancial derivatives based on the real estate market contributed to the demise of some old-line brokerage ﬁrms such as Bear Stearns and Merrill Lynch in the autumn of 2008. The quick loss in value of these derivatives sparked a lack of economic conﬁdence which led to the sharp economic downturn in the fall of 2008 and the subsequent recession. We will build a “stick-ﬁgure” model of these instruments, and even this simple model will demonstrate that the CDOs were far more sensitive to mortgage failure rates than was commonly understood. While this model does not fully describe CDOs, it does provide an interesting and accessible example of the modeling process in mathematical ﬁnance.
Consider the following ﬁnancial situation. A lending company has made mortgage loans to home-buyers. We make two modeling assumptions about the loans.
Let be the number of loans that default, resulting in a total proﬁt of . The probability of or fewer of these 100 mortgage loans defaulting is
We can evaluate this expression in several ways including direct calculation and approximation methods. For our purposes here, one can use a binomial probability table, or more easily a computer program which has a cumulative binomial probability function. The expected number of defaults is , the resulting expected loss is and the expected proﬁt is .
But instead of simply making the loans and waiting for them to be paid oﬀ the loan company wishes to bundle these debt obligations diﬀerently and sell them as a ﬁnancial derivative contract to an investor. Speciﬁcally, the loan company will create a collection of 100 contracts also called tranches. Contract 1 will pay dollar if of the loans default. Contract 2 will pay dollar if or of the loans defaults, and in general contract will pay dollar if or fewer of the loans defaults. (This construction is a much simpliﬁed model of mortgage backed securities. In actual practice banks combine many mortgages with various levels of risk and then package them into derivative securities with diﬀering levels of risk called tranches. Each tranche pays out a revenue stream, not a single unit payment. A tranche is usually backed by thousands of mortgages.)
Suppose to be explicit that of the loans defaults. Then the seller will have to pay oﬀ contracts through . The loan company who creates the contracts will receive dollars from the loans that do not default and will pay out dollars. If the seller prices the contracts appropriately, then the lender will have enough money to cover the payout and will have some proﬁt from selling the contracts
For the contract buyer, the contract will either pay oﬀ with a value of or will default. The probability of payoﬀ on contract will be the sum of the probabilities that or fewer mortgages default:
that is, a binomial cumulative distribution function. The probability of default on contract will be a binomial complementary distribution function, which we will denote by
We should calculate a few default probabilities: The probability of default on contract 1 is the probability of defaults among the loans,
If , then the probability of default is . But for the contract 10, the probability of default is . By the 10th contract, this ﬁnancial construct has created an instrument that is safer than owning one of the original mortgages! Because the newly derived security combines the risks of several individual loans, under the assumptions of the model it is less exposed to the potential problems of any one borrower.
The expected payout from the collection of contracts will be
That is, the expected payout from the collection of contracts is exactly the same as the expected payout from the original collection of mortgages. However, the lender will also receive the proﬁt of the contracts sold. Moreover, since the lender is now only selling the possibility of a payout derived from mortgages and not the mortgages themselves, the lender can even sell the same contract several times to several diﬀerent buyers if the proﬁts outweigh the risk of multiple payouts.
Why rebundle and sell mortgages as tranches? The reason is that for many of the tranches the risk exposure is less, but the payout is the same as owning a mortgage loan. Reduction of risk with the same payout is very desirable for many investors. Those investors may even pay a premium for low risk investments. In fact, some investors like pension funds are required by law, regulation or charter to invest in securities that have a low risk. Some investors may not have direct access to the mortgage market, again by law, regulation or charter, but in a rising (or bubble) market they desire to get into that market. These derivative instruments look like a good investment to them.
If rebundling mortgages once is good, then doing it again should be better! So now assume that the loan company has loans, and that it divides these into groups of each, and creates contracts as above. Label the groups Group 1, Group 2, and so on to Group 100. Now for each group, the bank makes 100 new contracts. Contract 1.1 will pay oﬀ $1 if 0 mortgages in Group 1 default. Contract 1.2 will pay $1 if 0 or 1 mortgages in Group 1 default. Continue, so for example, Contract 1.10 will pay $1 if mortgages default in Group 1. Do this for each group, so for example, in the last group, Contract 100.10 will pay oﬀ $1 if mortgages in Group 100 default. Now for example, the lender gathers up the contracts , one from each group, into a secondary group and bundles them just as before, paying oﬀ dollar if or fewer of these contracts defaults. These new derivative contracts are now called collateralized debt obligations or CDOs. Again, this is a much simpliﬁed “stick-ﬁgure” model of a real CDO, see . Sometimes, these second-level constructs are called a “CDO squared” . Just as before, the probability of payout for the contracts is
and the probability of default is
For example, with and then . Roughly, the CDO has only of the default probability of the original mortgages, by virtue of re-distributing the risk.
The construction of the contracts is a convenient example to illustrate the relative values of the risk of the original mortgage, the tranche and the second-order tranche or “CDO-squared”. The number 10 for the contracts is not special. In fact, the bank could make a “super-CDO ” where it pays $1 if of the -tranches in group fail, even without a natural relationship between and . Even more is possible, the bank could make a contract that would pay some amount if some arbitrary ﬁnite sequence of contracts composed from some arbitrary sequence of tranches from some set of groups fails. We could still calculate the probability of default or non-payment, it’s just a mathematical problem. The only questions would be what contracts would be risky, what the bank could sell and how to price everything to be proﬁtable to the bank.
The possibility of creating this “super-CDO ” illustrates one problem with the whole idea of CDOs that led to the collapse and the recession of 2008. These contracts quickly become confusing and hard to understand. These contracts are now so far removed from the reality of a homeowner paying a mortgage to a bank that they become their own gambling game. But the next section analyzes a more serious problem with these second-order contracts.
Now we investigate the robustness of the model. We do this by varying the probability of mortgage default to see how it aﬀects the risk of the tranches and the CDOs.
Assume that the underlying mortgages actually have a default probability of , a 20% increase in the risk although it is only a increase in the actual rates. This change in the default rate may be due to several factors. One may be the inherent inability to measure a fairly subjective parameter such as “mortgage default rate” accurately. Finding the probability of a home-owner defaulting is not the same as calculating a losing bet in a dice game. Another may be a faulty evaluation (usually over conﬁdent or optimistic) of the default rates themselves by the agencies who provide the service of evaluating the risk on these kinds of instruments. Some economic commentators allege that before the 2008 economic crisis the rating agencies were under intense competitive pressure to provide “good” ratings in order to get the business of the ﬁrms who create derivative instruments. The agencies may have shaded their ratings to the favorable side in order to keep the business. Finally, the underlying economic climate may be changing and the earlier estimate, while reasonable for the prior conditions, is no longer valid. If the economy deteriorates or the jobless rate increases, weak mortgages called sub-prime mortgages may default at increased rates.
Now with we calculate that each contract has a default probability of , a increase from the previous probability of . Worse, the 10th CDO-squared made of the contracts will have a default probability of , an increase of over ! The ﬁnancial derivatives amplify any error in measuring the default rate to a completely unacceptable risk. The model shows that the ﬁnancial instruments are not robust to errors in the assumptions!
But shouldn’t the companies either buying or selling the derivatives recognize this? A human tendency is to blame failures, including the failures of the Wall Street giants, on ignorance, incompetence or wrongful behavior. In this case, the traders and “rocket scientists” who created the CDOs were probably neither ignorant nor incompetent. Because they ruined a proﬁtable endeavor for themselves, we can probably rule out malfeasance too. But distraction resulting from an intense competitive environment allowing no time for rational reﬂection along with overconﬁdence during a bubble can make us willfully ignorant of the conditions. A failure to properly complete the modeling cycle leads the users to ignore the real risks.
This model is far too simple to base any investment strategy or more serious economic analysis on it. First, a binary outcome of either pay-oﬀ or default is too simple. Lenders will restructure shaky loans or they will sell them to other ﬁnancial institutions so that the lenders will probably get some return, even if less than originally intended.
The assumption of a uniform probability of default is too simple by far. Lenders make some loans to safe and reliable home-owners who dutifully pay oﬀ the mortgage in good order. Lenders also make some questionable loans to people with poor credit ratings, these are called sub-prime loans or sub-prime mortgages. The probability of default is not the same. In fact, rating agencies grade mortgages and loans according to risk. There are 20 grades ranging from AAA with a 1-year default probability of less than through BBB with a 1-year default probability of slightly less than to CC with a 1-year default probability of more than . The mortgages may also change their rating over time as economic conditions change, and that will aﬀect the derived securities. Also too simple is the assumption of an equal unit payoﬀ for each loan.
The assumption of independence is clearly incorrect. The similarity of the mortgages increases the likelihood that most will prosper or suﬀer together and potentially default together. Due to external economic conditions, such as an increase in the unemployment rate or a downturn in the economy, default on one loan may indicate greater probability of default on other, even geographically separate loans, especially sub-prime loans. This is the most serious objection to the model, since it invalidates the use of binomial probabilities.
However, relaxing any assumptions make the calculations much more diﬃcult. The non-uniform probabilities and the lack of independence means that elementary theoretical tools from probability are not enough to analyze the model. Instead, simulation models will be the next means of analysis.
Nevertheless, the sensitivity of the simple model should make us very wary of optimistic claims about the more complicated model.
This section is adapted from a presentation by Jonathan Kaplan of D.E. Shaw and Co. in summer 2010. Allan Baktoft Jakobsen of Copenhagen suggested some improvements and clariﬁcations in December 2013. The deﬁnitions are derived from deﬁnitions at investorwords.com.. The deﬁnition of CDO squared is noted in [1, page 166]. Some facts and ﬁgures are derived from the graphics at Portfolio.com: What’s a CDO.  and Wall Street Journal.com: The Making of a Mortgage CDO., 
That is, the expected payout from the collection of contracts is exactly the same the expected payout from the original collection of mortgages. More generally, show that
Suppose that you have a large population that you wish to test for a certain characteristic in their blood or urine (for example, testing athletes for steroid use or military personnel for a particular disease). Each test will be either positive or negative. Since the number of individuals to be tested is quite large, we can expect that the cost of testing will also be large. How can we reduce the number of tests needed and thereby reduce the costs? If the blood could be pooled by putting a portion of, say, samples together and then testing the pooled sample, the number of tests might be reduced. If the pooled sample is negative, then all the individuals in the pool are negative, and we have checked people with one test. If, however, the pooled sample is positive, we only know that at least one of the individuals in the sample will test positive. Each member of the sample must then be retested individually and a total of tests will be necessary to do the job. The larger the group size, the more we can eliminate with one test, but the more likely the group is to test positive. If the blood could be pooled by putting samples together and then testing the pooled sample, the number of tests required might be minimized. Create a model for the blood testing cost involving the probability of an individual testing positive () and the group size () and use the model to minimize the total number of tests required. Investigate the sensitivity of the cost to the probability .
Suppose you are taking a test with 25 multiple-choice questions. Each question has 5 choices. Each problem is scored so that a correct answer is worth 6 points, an incorrect answer is worth 0 points, and an unanswered question is worth 1.5 points. You wish to score at least 100 points out of the possible 150 points. The goal is to create a model of random guessing that optimizes your chances of achieving a score of at least 100.
 The Wall Street Journal. The making of a mortgage cdo. http://online.wsj.com/public/resources/documents/info-_flash07.html?project=normaSubprime0712&h=530&w=980&hasAd=1&settings=normaSubprime0712, 2008. Accessed August 25, 2010.
 Portfolio.com. What’s a c.d.o. http://www.portfolio.com/interactive-_features/2007/12/cdo. Accessed on August 25, 2010.
I check all the information on each page for correctness and typographical errors. Nevertheless, some errors may occur and I would be grateful if you would alert me to such errors. I make every reasonable eﬀort to present current and accurate information for public use, however I do not guarantee the accuracy or timeliness of information on this website. Your use of the information from this website is strictly voluntary and at your risk.
I have checked the links to external sites for usefulness. Links to external websites are provided as a convenience. I do not endorse, control, monitor, or guarantee the information contained in any external website. I don’t guarantee that the links are active at all times. Use the links here with the same caution as you would all information on the Internet. This website reﬂects the thoughts, interests and opinions of its author. They do not explicitly represent oﬃcial positions or policies of my employer.
Information on this website is subject to change without notice.
Steve Dunbar’s Home Page, http://www.math.unl.edu/~sdunbar1
Email to Steve Dunbar, sdunbar1 at unl dot edu
Last modiﬁed: Processed from LATEX source on September 8, 2016