Goldman, the SEC and performativity

April 26, 2010

There has been no shortage of blog commentary on the recent civil suit by the SEC against Goldman Sachs. In my view, the case also makes clear the value of Donald MacKenzie’s view of the credit crisis, and the notion of performativity. In a recent and ambitious article, MacKenzie has argued that the credit crisis was caused by the counter-performativity of the models employed to rate mortgage derivatives. The Goldman episode provides clear proof of the value of these ideas, and the need for sociologists to engage with them.

Let us first review the case. In essence, the SEC’s claim is that Goldman stuffed its mortgage derivatives with bad mortgage bonds. It did so to please an esteemed client who wanted to bet against the housing bubble, John Paulson. And having done so, it concealed it from its other clients – the ones who were betting that the securities were sound. These clients, the unlucky European banks IKB and ABN Amro, bought the securities, lost millions and subsequently went bankrupt.

A mystery remains
For all the light on the ugly side of human nature that the case seemingly sheds, one critical mystery remains. Why did the Europeans buy these derivatives? Implicitly, the financial press has presented these bankers as naïve or, in the extreme, irrational. In my view, however, the answer is a different one and has to do with performativity.

The concept of performativity speaks to the relationship between economics and the economy: economists create models that investors use, shaping prices in ways that make the models more accurate. The opposite process is also possible. A model is counter-performative when its use makes the model less accurate. MacKenzie (2009) argues that the reverse dynamic is what happened in the credit crisis. The models used by investors to value mortgage derivatives made these models a less accurate representation of value. In fact, MacKenzie does not just refer to models, but also to indexes, cultural norms and organizational arrangements.

MacKenzie (2009) focuses on the use of the rating system to value credit derivatives such as collateralized debt obligations. The use by the credit rating agencies of a fixed set of characteristics, clearly specified and communicated to investors through software packages such as S&P’s Evaluator, made it possible for investors to game the system. It worked as follows: every security has multiple qualities and characteristics. Knowing that only a few of them are quantified and used to value the credit derivatives, the arrangers of these derivatives took bonds of mortgages that were on the whole close to worthless but that looked good from the narrow standpoint of the credit rating. Investors could naturally choose to evaluate these mortgage derivatives themselves. But given their sheer complexity, the speed at which they had to respond to bid-lists (a few hours), and the fact that credit rating had worked perfectly well in the past, few of them chose to do this.

Why buy a toxic asset?
This, in short, is why IKB and ABN Amro bought mortgage derivatives. The very mortgage derivative that led to the problem, ABACUS, was rated as a safe security. These companies were credible, at the time. Past precedence was reassuring. They went ahead.

Interestingly, this is also why John Paulson shorted those same mortgage derivatives. He knew the game that some issuers were playing. He reputedly asked Goldman to put together a security with bonds that excluded the most honest actors. The goal was to explicitly create something that looked fine under the rating system, but that in fact was worthless. And then, bet against it.

One security, two valuation systems — both of them seemingly legitimate. And two opposite views about the same security: the European banks in favor, Paulson against. The crucial detail that might perhaps have given the game up to the Europeans – that John Paulson was taking the opposite side — was apparently concealed by Goldman.

What do we learn? It is unclear, at this point, whether the SEC will be able prove its case in court, or how will Goldman’s reputation survive the case. But what is certain is that the European banks were not irrational or naïve. They were unaware of a complex process of counter-performativity that would continue to march on… to the point of bankrupting Lehman, Bear, RBS and a long list of others. The Goldman episode and the credit crisis illustrate the effectiveness of MacKenzie’s analysis, of performativity, and the intellectual potential of the social studies of finance.

16 Responses to “Goldman, the SEC and performativity”

  1. jck Says:

    “toxic asset” with the benefit of hindsight, they were not considered toxic then.

  2. danielbeunza Says:

    Well, what’s interesting here is that some people at Goldman already thought they were toxic then (e.g. “Fab” Fabrice”) whereas some others might not.

    In other words, whereas I grant you that we should be biased by hindsight, our job does not finish there. The key point raised by the hindsight issue is that there was a diversity of perspectives, and that’s what’s so difficult to understand for economists and (apparently) journalists. Hence the “email game” that we are now witnessing, with Goldman and the SEC selectively releasing emails that attempt to show that either no-one at Goldman thought there was a bubble; or that everyone did.

    The reality is, organizations respond to uncertainty by having diverse points of view. Should we then be outraged that some folks at Goldman sold CDOs while others shorted them? I think it depends on whether Goldman sold the CDOs by endorsing them as a corporation.

    Academic institutions understand this, and do not attach the organization’s view to that of it’s employees. The LSE, for example, does not argue that it stands by my research. But banks appear to try and keep the pretence of having one single view on things. Which perhaps allows them to charge extra. And then gets them in trouble.

  3. jck Says:

    Very true, there is always a diversity of perspective but most banks have a policy of forcing their employees to sing the party line, that is expressing the official view of their analysts to the customers, not their own opinion.
    From memory, GS chief economist [Jan Hatzius] became bearish on subprime well after that deal, so the party line at the time of the Abacus deal was either neutral or positive (don’t remember exactly)
    Fab is in trouble if he represented his own opinions as if they were GS.
    Otherwise he is in the clear, what he said to his girl friend is irrelevant.

  4. joseossandon Says:

    Hi Daniel,

    I think another striking element in this controversy is how ” risk ratings” are seen now as “agents” in the press (see for instance: Following Mackenzie we can say this corresponds to a wider topic in the sociology of knowledge, but not just about “knowledge cultures”, but about how rankings transform what they touch. In this sense, it seems this case is not so different to other areas, like university rankings studied -among other- by Espeland & Sauder.

  5. danielbeunza Says:

    Jose — excellent NYTimes article, thanks. Your comment also opens up the issue of how performativity and reactivity (Espeland and Sauder’s concept) are different and similar.

    I’d say that it is one thing to have a ranking (which allows for prioritizing) and it is quite a different thing to have a model that allows you to calculate. In the case of MacKenzie (2009), the debate is particularly acute because he has expanded the study of performativity beyond formulae and models, to include valuation cultures. So your point is well taken.

    Is the 2008 credit crisis a case of performativity, or reactivity? One way to think about it is to conceive of the ways in which a “meta-device” is different from a ranking. More thoughts welcome…

  6. joseossandon Says:

    Hi Daniel. I read their paper long ago but I think they take the concept of “reactivity” from C. Heimer’s work on risk (which can be even closer to this case!). I understand your point that rankings and models should be distinguished, however, I would not say that calculation is a monopoly of models. At least in the sense given to calculation by Callon, Cochoy and Muniesa, we could say that what rankings do is that they enable (quantitative) comparison (or commensuration following Espeland again, or investments of forms in the sense given long ago by Thevenot). Empirically it also seems complicate to disentangle them. A good case is Poon’s paper, where it FICO (as she suggests a central device in the mortgage crisis) is a model that ranks. (I tried to organize the discussions about rankings in an old post called “sociologia de los rankings” (sorry it is in spanish) that can be found here:

  7. Juan Espinosa Says:

    Hi José and Daniel:

    You have a very interesting conversation here.
    I had been interested in classifications for a time.
    For me a very good resource is the Bowker and Star’s book: “Sorting things Out”

    In this book the authors explore the effects of classifications and what happens to our thinking as a result of our classifications. This also resonates with some work of Ian Hacking on the history of statistics…

    By the way, thanks for the post in Spanish José. Very interesting one!

    I hope that this could ad some value to this interesting post.

  8. joseossandon Says:

    Hi Juan. I totally agree, the literature on classifications can be quite helpful here. A great paper that connects nicely many of these issues is: “Lists come alive” by Leyshon & Thrift.

  9. marthapoon Says:

    “One security, two evaluation systems…” but placed in the legal context, faced with a third. What will be just as interesting is to see how the legal process values or re-evaluates securities through its own forms of rationality.

    I was delighted to see, for example, that the SEC have enrolled FICO, a financial actor of sorts, as a witness in their case. The complaint issued on April 16th remarks that “Paulson’s selection criteria favored RMBS that included a high percentage of adjustable rate mortgages, relatively low borrower FICO scores, and a high concentration of mortgages in states like Arizona, California, Florida and Nevada that had recently experienced high rates of home price appreciation.”

    The very method that permitted the precise selection of loans as a form of financial action, has now been turned into evidence of misconduct in a legal suit.

  10. danielbeunza Says:

    Juan and Jose — thrilled to see such interest among Spanish speakers. i might have to start posting in Spanish soon…

    Martha: there is very good reason why the selection criteria is now evidence of misconduct. These were “undercover” selection criteria that the buyers of Abacus did not know about.

    It would be very interesting to have a visualization that shows how “bad” CDOs can get the same rating as “good” CDOs.

  11. Juan Espinosa Says:

    By the way, and about the language. I received the next information from Spain:

    I Encuentro ANT: Presente y futuro de la teoría del actor-red

  12. Yuval Millo Says:

    Great post! And, as we say in these parts, the story is another fine example of ‘what you measure is what you get’!

  13. zsuzsannavargha Says:

    Martha, Daniel, Yuval et al,

    To play with properties of securities that will get the targeted rating—that’s become a “normal” part of assembling structured products, as the NYT article points out (thanks Jose). Ranking transforms what it touches: especially if the credit rating software allows users to see the input and instantly its outcome, then design (selection of securities) and outcome (rating) are not successive phases of producing financial value but a back-and-forth process of estimation.

    The difference from everyday practice was that Paulson allegedly knew how to select good-looking yet low-performing mortgages.

    I agree with Daniel—we (and the courts) should see not only what made the CDOs bad (FICO, state, etc.) but also what made them good at the same time.

    Mortgage brokers engage in reverse engineering, too. Knowing what parameters count, and what figures to plug in for the right result is a common feature of consumer credit applications. As standard practice, consumers and representatives of financial firms participate in this game together when they are trying to arrive at a suitable loan using pre-approval software. Brokers figure out by trial and error what constellation of properties will get mortgages eventually approved by the bank. The difference from Paulson’s skill is that these approvable properties aren’t monitored or predicted by the brokers, whether the customers will default or not.

    So, mortgage sellers could another instance of counterperformativity because they don’t commonly bet on their customers defaulting (they might not care or there is no construction with which to make the bet). It is the working of the bank’s credit assessment model-as-it-should that brings about less accurate results, instead of intentionally putting together worthless securities. Opinions?

    Finally, an interesting aspect of the Goldman litigation document that Martha linked above: the communication between the third-party assessment agency (ACA), Goldman, and Paulson on which securities to include (pp. 9-11). This was a very iterative process where the parties proposed partial replacements of the portfolio, with the understanding (not always open justification) that the new portfolio would still produce the targeted rating. It’s not clear to me how or why they reached agreement on the final selection. But to pick up a previous thread, this gradual and collective nature of assembling deals is completely missing from behavioral descriptions of decision-making.

  14. danielbeunza Says:

    Zsuzsi — yes. What I would love to see is a visualization that shows the value of a CDO from different perspectives.

    Here’s a clip on Paulson from Greg Zuckerman, the author of the excellent book on him.

  15. danielbeunza Says:

    This is very interesting — consistent with the “two perspectives on valuation” view, though not the “ratings” vs. “real value” version that I put out above. The way it’s written, we have to just trust Duffie that this is how it is. Hopefully more evidence will come up.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: