Friday, April 24, 2009

Proposal done

Phew. I always feel silly when I have writer's block. I know that there is no real obstacle to putting words on the page and yet they still don't get on the page. But deadlines have a clarifying effect.

I did have a substantive breakthrough, which helped. (I'm not sure that much of the breakthrough is actually in the proposal, but it helped me see where I am going better.) Let me see if I can articulate it here.

Some of my current research is about how reporting fraud screws with competitors--they think that the fraudulent firm is doing better than it is, which sets the bar higher for the returns needed at a competitor for that competitor to believe that their work is...competitive. The excellent Simon Angus and I have built a cool NetLogo model that shows that the dynamics created by fraudsters put down amongst credulous competitors are wacky and a bit scary. Wacky, because our measure of the economy's health veers around wildly, in non-monotonic ways, and scary because it spends time in very bad territory. Sometimes it equilibrates there.

The basic take-away is that put a large pinch of misinformation in a system without a fast discipline source, and all hell breaks loose. And because (as I say in my proposal) fantasy can easily out-compete reality, you can get a quick convergence on whatever the fantasy is pushing.

Now to the current crisis.

Think of "the market" as a vast complicated machine full of intersecting cogs. Each cog is a market for a particular thing or suite of things. There are at least 3 or 4 cogs worth focusing on: the market for "safe" investments writ large, the market for structured finance instruments (an overlapping set with the safe investments), the market for credit ratings, and the market for banks (and insurance companies, and other big, important financial hubs). Each of these suffered from the misreporting dynamic we have modeled in my current work. Note that I am not accusing anyone of criminal fraud--that isn't necessary for this metaphor to work, only that there was systematic misreporting occurring due to bad models, poor accounting standards, etc.

Start with the narrowest market, that for structured finance instruments. Coval et al in their article in the current edition of Journal of Economic Perspectives demonstrate that comparatively small errors in estimating default risk of the underlying assets get amplified significantly in estimating the risk of a CDO. This amplification of error doesn't happen with traditional pass-through instruments--ones that, while they package large numbers of mortgages or other debt, don't have complicated orders of payoffs but instead just give you the right of a certain fraction of the cash flow--so CDO's risk estimates will systematically be much further off the mark (in one direction or the other) than those of other instruments. Within the universe of possible CDOs, some bundles of debts will appear much riskier than they really are, because the error estimating the risk of the underlying asset is off in same direction by a little bit, and others will appear much safer than they really are, again due to the error in estimating the risk. But take two potential types of debt one could bundle, one that gets estimated as riskier than it is and one that gets estimated as safer than it is--in all likelihood only the one that appears safer will get bundled, because it looks like such a great deal. So if there were a type of debt that had performed badly in the years for which we have data, but would for some sort of structural reason start performing better more recently, this debt will not get bundled. It will look really risky! Those instruments that are being misestimated in an optimistic direction are going to be the instruments that become the most popular, because they look so good.

Why doesn't the market self-correct? Well, it is ... right now. And that self-correction isn't much fun. Market discipline only kicks in when the mistake is laid bare, and when you are talking about a mistake that only becomes obvious in a severe downturn, you have to wait for that downturn for the market to work its magic. And in the interim, you get the kind of market convergence on the mistake that my fraud model predicts.

The same dynamic that pits different types of CDOs against each other happens in the next market--the market for AAA-rated investments. Since the type of risk that CDOs face is not addressed by the ratings agencies (and they are committing those estimation errors), CDOs look great--they turn shit into gold and there is money left over for higher returns than T-bills. The same market correction problem occurs, because a lot of people make a lot of decisions before any information about the mistake being made comes to light. So we get a market of AAA-rated stuff that gets dominated by unsafe CDOs.

Now turn to the ratings agencies. Estimating the risk of CDOs is really hard. When CDOs were first created, the agencies used pretty different models to estimate their risk, and therefore, the same product might get widely divergent ratings from different agencies. Now, a word about how agencies compete. They make money by charging the companies they rate, so they want to do as many ratings as possible (comparable to selling lots of stuff in a more traditional retail business) but the price they can charge is related to the reliability of their ratings. The reason I can't go up to Moody's and pay them to stamp AAA on my forehead is because people would stop taking a Moody's AAA very seriously, and therefore the value of having that AAA goes down, and therefore the price Moody's could charge for the priviledge would also go down. So reputation acts as a constraint on their behavior.

So back to CDOs. Assuming all three agencies have comparable credibility, the packager of a CDO is going to buy the rating from the company that is willing to give the CDO the highest rating. So the model that produces the most AAA ratings is going to get a lot of business. And since none of those CDOs were performing badly in the boom years, the model that was saying AAA was maintaining its credibility. So those models that weren't turning out AAA were just losing their agency money, but to no credibility gains. So the models the agencies use converged on models that gave AAA. (And there is another paper I want to write about reputation-dependent oligopolies. If one cartel-izes quality, not quantity, the reputation constraint stops maintaining high quality products....)

So that is another, inevitable market effect that puts us in the worst possible place.

Finally, there is the market for bank stocks. Banks have to hold a lot of their capital in AAA-rated stuff so that they can leverage their capital as much as possible. And a bank that chooses to buy lots of CDO AAA crap is going to earn much higher returns on that crap than a bank that holds their AAA stuff in T-bills. So the bank that is taking the risky, ultimately bankrupting stuff is going to appear to be doing much better than the smart, safe bank. And so the smart safe bank is very likely to copy the strategy of the stupid bank.

And here we are in 2009 with the economy down the toilet.

Now I just have to find a way to prove all of this, and then write it up sans bathroom vocabulary....

No comments:

Post a Comment