Skip to main content

View Diary: Scientists *Prove* Toxic Assets are Impossible to Regulate (268 comments)

Comment Preferences

  •  Not quite (12+ / 0-)

    The paper does not say that there are no good CDOs or that the risk of CDOs cannot be determined under any circumstances.  What is says is that CDOs are vulnerable to undetectable manipulation -- that you can't tell in computationally tractable time whether the underlying assets were non-randomly distributed among a non-trivial number of CDOs -- whether someone intentionally made some bad CDOs by choosing to concentrate the lemons in them.  If you could be certain a priori that the underlying assets were randomly distributed, then (modulo other considerations, such as mistaken assumptions about the independence of the underlying asset prices) the risk of the CDOs could be computed.

    In other words, the solution to the problem can be seen as providing a trustworthy mechanism that guarantees random distribution of the underlying assets into the pool of CDOs.  Whether that is itself a tractable problem is left as an exercise for the reader.

    •  Were we able to exercise prior control (4+ / 0-)
      Recommended by:
      DBunn, Akonitum, NCrissieB, MichaelNY

      on the construction of CDO/Ss, wouldn't the problem of undetectable manipulation thereafter remain? Your conditional - "If you could be certain a priori" - is so stringent as to be unlikely to be satisfied in the real world by regulatory agencies whose ability to protect the public is limited at best and highly subject to prevailing political winds. Even if your stringent condition were in theory satisfiable, how would buyers be able to reassure themselves that it had been actually implemented? Would potential buyers really be in any better position to know the quality of the CDO/S on offer before them?

      American democracy: One dollar, one vote. See? Equality!

      by psnyder on Sun Oct 18, 2009 at 12:34:11 PM PDT

      [ Parent ]

      •  Sure, guaranteeing randomness isn't easy (5+ / 0-)
        Recommended by:
        barath, psnyder, knocienz, NCrissieB, MichaelNY

        But it may not be computationally hard the way ex post regulation of CDOs is.

        My points are two.  First, that the paper does not show that CDOs themselves are inherently evil or the source of evil, but rather that the evil of claiming random distribution of risk in a CDO pool when, in fact, the distribution is non-random passes through and is hidden by the CDOs.  Second, that guaranteeing random distribution defeats the evil, even while leaving the CDO market intact.

        Is guaranteeing a random distribution of the underlying asset risk any more tractable than regulation that depends on ex post detection of CDO manipulation?  There is reason to be hopeful that the answer is yes.  Trustworthy randomness is not just a problem for CDOs, but for many problems in information security.  As such there are already some fairly strong solutions, and there will continue to be improvements.  Guaranteeing that CDOs were built with trustworthy randomness would essentially be a matter of feeding a list of underlying assets into a regulator-audited and regulator-secured black box that would spit out CDOs with a certificate of randomness, a seal of approval.  Trusting the CDO to be free of manipulation then becomes a matter of trusting the certificate and the black box process.  Establishing and maintaining that trustworthiness is certainly non-trivial, but at least in concept it is more tractable than is ex post regulation of CDOs.

        •  Maybe... (2+ / 0-)
          Recommended by:
          Foodle, NCrissieB

          Though there are plenty of protocols to do joint randomness generation and to do secure multiparty computation (and lots of work on joint-shuffling for voting mixes and the like), it would be adding complexity to the situation, which we've found is perhaps not the direction we want to go in.

          It could work, but maybe that's looking at the solution from too CS centric a viewpoint...I wonder if we could really expect regulators to understand the processes for randomness generation, etc.

          •  It would be adding steps to the process (3+ / 0-)
            Recommended by:
            barath, NCrissieB, MichaelNY

            but it would actually be reducing complexity in the sense that information would be destroyed -- it wouldn't be possible for the CDO creator to retain privileged information about the distribution of lemons.

            •  Turning the CDO market into a mutual fund? (3+ / 0-)
              Recommended by:
              Foodle, NCrissieB, MichaelNY

              Given an accredited random distributer, what is the distinction between the different resulting CDOs? In theory, with sufficiently large input, the different groups should have a gain/loss distribution that would be determined by entropy equations.

              In that case, why even distribute them into different CDOs? Your payoff distribution would be entirely based upon the criteria for entering the pool prior to the randomization and breaking them into different groups after randomization would be worthless.

              Effectively, you would be investing in a subsection of the mortgage market that was used to create the pre-randomization pool.

              •  Yup (3+ / 0-)
                Recommended by:
                barath, knocienz, NCrissieB

                I agree as far as I can see, and I must confess to not really understanding the utility of the practice of dividing the pool up into multiple CDOs instead of just selling shares in one big CDO created from the entire pool.  Of course, the random function for distributing all of the pool assets into a single CDO is also satisfyingly simple, easily audited, and very trustworthy.

                •  Different investers want different things (1+ / 0-)
                  Recommended by:

                  Some want high risk/higher return. Others want low risk/lower return. Without this function, they become pointless.

                  CDO's were not used much until David Li came up with his Gaussian copula models for pricing.

                  Investment banks jumped on them since they assumed the formula was valid and would reduce risk. This diary shows that they are not and, more importantly, can never be.

                  •  Tranches (1+ / 0-)
                    Recommended by:
                    Claudius Bombarnac

                    Risk/return can be differentiated by tranches within a single CDO.  The risk/return needs and desires of different investors don't demand dividing the asset pool into multiple CDOs.

                    •  You are right to a point... (0+ / 0-)

                      But not all CDOs are created equal and this is my point.

                      A senior tranche in one CDO can be quite different from that in another CDO depending on the overall quality of the underlying assets, cash flow, yield, management, and a plethora of other factors.

                      This gives a much greater range of risk/return options/features.

                      I can't picture having a single CDO created from one gigantic pool and simply divided into 4 tranches. It is not the way the free market works.

    •  Another possibility (1+ / 0-)
      Recommended by:

      is simply to require that all tapes describing the assets underlying these derivatives are encoded in a standard and machine-readable form.  If you can get all of the information into a database, you can check for yourself on a variety of properties such as the geographic, demographic, and property attributes of each mortgage.  If they are well-distributed, then congratulations, you have a chance of losing your shirt honestly.  If they turn out to all be two-year old 3,000 sq. footers made by a fraudulent builder in Detroit ... good luck.

      But, I think the end message is the same - always require the fundamentals to be as exposed as possible.  That way there is at least the chance to read the fine print.

      •  Sure, but (2+ / 0-)
        Recommended by:
        OdinsEye2k, NCrissieB

        if you are going to force that level of information awareness on potential CDO buyers, then those buyers may as well become direct lenders.  The whole point of a CDO is to be able to extract low-risk investment opportunities that don't require the level of knowledge of risk in the underlying asset that direct lending does.

      •  wrong conclusion (0+ / 0-)

        The issue described here persists even if all information is available, and the root or the issue is how the liabilities and payoffs to different tranches are defined.

        However, I am not sure if the paper describes the "real" problem.  In my opinion, the real problem is the following: well designed derivatives allow to trade "risk" with correct prices if the participants in the market exercise due dilligence.  However, the attention of the participants is diverted from the analysis of the underlying facts (e.g. how long one can keep selling mortgages in markets where buyers cannot afford them without those markets crashing) to the analysis of the model.

        For example, suppose that a car is worth 1000 when it functions and 0 when it does not (so-called lemon) and ca. 20% or cars in a pool of used cars are those that will pretty soon stop functioning.  To sell the cars we issue warranty contracts.  Warranty contracts are pooled and securitized.  Now we have pay someone 250k to provide financial guarantee for  1000 cars -- giving that someone a profit of 50k and removing the risk from our books.  So far so good.

        But suppose that the problem with cars is not that some have hidden defects, but that most of them are insufficiently resistant to corrosion.  More precisely, the problem emerges when salt is applied to roads for to long period of time in a single winter.  Cars survive first and second winter with few defects.  The fee to guarantee 1000 cars drops to 100k.  Then comes a harsh winter and someone who pocketed 100k has 500k liabilities.

        Which would be a "personal problem", except that that "someone" did it 1000 times over, and  most banks gave risk-free loans with cars on warranty as collaterals etc.

    •  I agree with your analysis. (1+ / 0-)
      Recommended by:

      However, the problem then becomes how to regulate a private market such that sellers have no choices in what underlyings (of a given type) that they bundle into CDOs/CDSs. As you note in comments below, that is a non-trivial problem because in a private market clever people can usually find ways to cheat.

      The problem here is that if this paper is correct, buyers can't use statistical analysis of CDO/CDS performance to identify fraudulent sellers. The computation required to detect lemon-loading fraud will take longer than the lifetime of the instrument and the statute of limitations. Why would any buyer take that risk?

Subscribe or Donate to support Daily Kos.

Click here for the mobile view of the site