Some Thoughts on the Mortgage Crisis, Performativity and General Linear Reality

First, before I go any further, I should say that I really don’t understand macroeconomics or the recent economic crisis. It’s a goal for the next year, but for the moment I’m working on a very piecemeal understanding, the kind you get from trying to puzzle through the financial stories in the New York Times occasionally. So, I very much appreciated two things I came across recently. The first is this hilarious analysis (made more hilarious by it being British) of the mortgage crisis that I highly recommend if you have 10 minutes:

I love how the new financial models created new markets for derivatives and trenches and all sorts of fancifully named packages of mortgages. Finance theory created new markets for new things that could not have existed before.

The second is an excellent NYT Magazine piece (from a couple weeks ago) on the mortgage crisis and in particular the role of rating agencies like Moody’s in the whole mess. I’ve read accounts of the importance of credit agencies in the crisis before, but none that were as well-written.

Let me sample a few interesting quotes:

Structured finance, of which this deal is typical, is both clever and useful; in the housing industry it has greatly expanded the pool of credit. But in extreme conditions, it can fail. The old-fashioned corner banker used his instincts, as well as his pencil, to apportion credit; modern finance is formulaic. However elegant its models, forecasting the behavior of 2,393 mortgage holders is an uncertain business. “Everyone assumed the credit agencies knew what they were doing,” says Joseph Mason, a credit expert at Drexel University. “A structural engineer can predict what load a steel support will bear; in financial engineering we can’t predict as well.”

There’s some interesting things going on here with a move towards a ‘rationalized’ quantitative model, and the analogy to structural engineering. I love how the old banker is cast as inelegant, and basically irrational – acting on tradition and emotion. But the elegance of the new system obviously had dangers. But should not models for risk be precisely designed to take into account the possibility of “extreme conditions”?

Moody’s rated three-quarters of this C.D.O.’s bonds triple-A. The ratings were derived using a mathematical construct known as a Monte Carlo simulation — as if each of the underlying bonds would perform like cards drawn at random from a deck of mortgage bonds in the past. There were two problems with this approach. First, the bonds weren’t like those in the past; the mortgage market had changed. As Mark Adelson, a former managing director in Moody’s structured-finance division, remarks, it was “like observing 100 years of weather in Antarctica to forecast the weather in Hawaii.” And second, the bonds weren’t random. Moody’s had underestimated the extent to which underwriting standards had weakened everywhere. When one mortgage bond failed, the odds were that others would, too.

Basically, the financial wizards at Moody’s assumed that the individual mortgages in the C.D.O. packages would behave as independent trials of a random experiment – if you take 2000 mortgages where each lender seems to have a 5% chance of defaulting based on their past credit history, then you guess about 100 will default and you price the C.D.O. accordingly. Monte Carlo simulations help you figure out the variation on this – how likely is it that 150 or 200 will default? Etc. But why would we assume independence? The mortgages were drawn from across the country, which should help to eliminate regional shocks, but did no one imagine that there might be national level shocks that would affect everyone at once? I know absolutely nothing about financial modeling, so maybe there is some control for this that the NYTimes folks did not go into. But it sounds like what happened was not simply that “the mortgage market changed” making the past a poor indicator of the present, but that in particular, the cases were not independent in the slightest (“When one mortgage bond failed, the odds were that others would too.”). Were the financial wizards at Moody’s caught up in General Linear Reality* so strongly as to think of mortgages as completely independent?

Anyway, enough with the mini-rant. Watch the bit and read the article.

* General Linear Reality is a model of thinking about the world that assumes that the world acts like the assumptions of the general linear model in statistics. Basically, variables are well-defined (“race” means the same thing at all times and places and has a consistent effect on some outcome, say), cases are independent, etc. Assumptions that were needed to make mathematical models tractable become ways of understanding the world.

Edit: Some more on the subject via Economist’s View from this article, Blame the Models

The main problem with the ratings of SIVs was the incorrect risk assessment provided by rating agencies, who underestimated the default correlation in mortgages by assuming that mortgage defaults are fairly independent events. Of course, at the height of the business cycle that may be true, but even a cursory glance at history reveals that mortgage defaults become highly correlated in downturns. Unfortunately, the data samples used to rate SIVs often were not long enough to include a recession.


In physics the phenomena being measured does not generally change with measurement. In the finance that is not true. Financial modelling changes the statistical laws governing the financial system in real-time. The reason is that market participants react to measurements and therefore change the underlying statistical processes. The modellers are always playing catch-up with each other. This becomes especially pronounced when the financial system gets into a crisis.

This is a phenomena we call endogenous risk, which emphasises the importance of interactions between institutions in determining market outcomes. Day-to-day, when everything is calm, we can ignore endogenous risk. In crisis, we cannot. And that is when the models fail.