« Chutzpah | Main | Reformers, Take 2 »

September 21, 2008

Comments

I have a question, no doubt a dumb one. We keep hearing that the reason this stuff is so toxic is that no one knows how much of it is good debt and how much bad. It seems to be that if any accounting standards at all are in use, it should be possible to tie each security to the pool of mortgages it came from, analyze how many of them are in default, and come up with a sensible valuation for it. If no accounting standards were used, why aren’t all these people in jail for fraud?

Mike,

the only barrier to that being true (beside the actual possibility of fraud and/or crap accounting, as you note) is the absolutely stupefying amount of information such a global sussing out of mortgage pools one would have to sort.

The labor expenditure alone that is required to actually valuate these assets based on the default levels of the original borrowing population would make many of these assets unprofitable, if such a task were to be undertaken.

Influence "pedaling"? What the hell is that?

Mike Shilling.

The problem is mistated.

There is one type of mortgage security that, no matter how exotic, no matter how many itterations removed from the loan pool upon which it was ultimately based you can take a look at it on Bloomberg and be able to drill down to the underlying.

But looking at the actual pool and the number of currently good vs in default loans contained within tells you nothing about the security's value. Valuation, especially on mortgages is forward looking. It depends upon assumptions of future default rates, future prepayment speeds, and future interest rates. Prior to this crisis the PSA had decades of statistics useful in profiling a pool of loans and prediciting its default and prepayment bahavior under a wide variety of interest rate scenarios. But the model is broken as defaults are exceeding the parameters allowed for by the predictive model. That is why they cannot be valued going forward. Give the pools another two or three years of seasoning and the models will work in the new default data and we will be able to once again value the securities with confidence.

Other 'mortgage related' assets exist that are not tracable to any specified pool of mortgages. These may be dirivative of the performance of a mortgage index, or of a pooled liability insuring mortgages, or whatever else one can think of creating.

The valuation of the second type of security can not be done with any confidence until the confidence is restored in the valuations of the first type of security.

And we won't have confidence until a track record is re-established on recent mortgages pools that is reliably predictive going forward.

The problems which the GSE's have now are a symptom, not a cause.

The big problem in the mortgage market is in the subprime and Alt-A pools, which is to say the people who were not able or willing to obtain GSE loans, and with the way that securitization of MBS broke the feedback relationship between loan origination decisions and investor risk - it is both easy and tempting to make bad decisions approving loans when you are playing with somebody else's money, even worse if you earn your salary via sales commissions.

Arguing over the history of GSE regulation is a red herring.

We should be looking at failures with regard to the so called shadow banking system (IB's and hedge funds), and why they were allowed to use such dangerously high levels of leverage without adequate regulatory restraints, and at loan origination decisions in the subprime and Alt-A pools.

Thanks, ken, that helps.

Let me be sure I get it:

Laymen like me would imagine that people playing with billions of dollars would use a robust model with input variables for, well, things that vary, like changes over the term of mortgages in the price of the underlying real property, especially shortly before balloon payments or rate increases are due, or for changes in borrowers' average ability to repay. For example, if I were considering buying mortgages from Scranton, PA in the 1980s, I would want to know what effect closing steel mills and coal mines would have on the borrowers' income and the ability to re-sell. Right now, higher defaults seem to be due mostly to (a) more loans to what were traditionally considered poor-risk borrowers, (b) the correction to the price of the underlying homes (this includes the vast amount of fraud), and (c) more whole-value ARMs, which (surprise, who'd'a thought the levies would breach) stand even less chance of repayment after the price correction than traditional instruments. But all of those boil down to: the borrowers are less able to repay than we thought, and the price went down. So, we think, why should it be so hard to at least make a much better prediction?

But, if I understand you correctly, the models are actually much more simple and rigid than that: they assume that a big enough pool of mortgages will smooth out those variables, so there are no inputs of that sort. That does sound like the kind of idiocy financiers usually commit if left alone too long.

If that's it, though, why is it so hard to make an equally rigid, simple model that comes a lot closer to the reality? Prediction is a mug's game, but we have a LOT of data by now on how many more mortgages are defaulting than we expected. There's nothing wrong with the basic concept that a big pool smooths out local variation, the problem (I think?) was that the model made rosy assumptions about the economy as a whole. We have to change that sort of input quickly in all sorts of government budget calculations, don't we know how?

BTW, sheer curiosity, and feel free to tell me it's none of my beeswax, but I gather you are a securities trader? A year or two ago, did you or your colleagues see the problem with mortgage-backed securities? How did you deal with that as to your own customers?

But the model is broken as defaults are exceeding the parameters allowed for by the predictive model. That is why they cannot be valued going forward. Give the pools another two or three years of seasoning and the models will work in the new default data and we will be able to once again value the securities with confidence.

This is the same imbecilic argument that created the mess in the first place.

The people creating the models used naive statistical techniques to value the securities. They made assumptions of the possible range of market conditions based on a few years or decades of data. Of course, as always happens in these situations, events occurred that were so unexpected that their models simply assumed they would not occur. And then we had a financial crisis.

Now, ken tells us, we just need 2 or 3 more years data. Then the naive statistical models can be updated, and we can go on as before. Until the next unexpected event occurs. Probably within 5 or 10 years. At that point, the people who created the models will feign innocence again and we can have yet another financial crisis.

2 or 3 more years of data on interest rates, foreclosure rates, housing price changes, and market volatility will not give you a bounded range for what those parameters will be the year after. They will not give you true probabilities for what those parameters will be. Nor will 100 years data, nor will 1000.

They made assumptions of the possible range of market conditions based on a few years or decades of data. Of course, as always happens in these situations, events occurred that were so unexpected that their models simply assumed they would not occur.

Worse, to a significant extent the universal reliance on such models *created* the crisis. A beautiful example of the craziness- the idea that foreclosure rates in Indiana and New Mexico weren't strongly correlated (so an investment based on both is usually safe from both failing). Which might be true under normal conditions, but believing this led to the abnormal conditions by extending credit to anyone who could sign their name.

But I dont think that this invalidates statistical modeling as you say. When the US Great Plains were first settled extensively, there was an extended period of high rainfall over the region. People acted as if this would last forever, and this naturally led to tragedy.
But that doesn't mean that a very long-term view of climate isn't possible, just as it's possible for a reasonable model of mortgage valuation to be constructed. The flaws were with the way the models were constructed (eg paying fees to Moody's to 'buy' AAA ratings, and then pretending that those AAA ratings were indicative of the reliability of the product), not with the general principle of modeling.

Crafty trib,

I don't mean to mislead anyone into thinking I am a securities trader when I talked about 'we' being able to value securities, etc. I was just trying to be inclusive, as in all of us, that's all.

The sketch I gave of three variables is not necessarily all encompassing. Whatever other factors the statisticians find significant are also taken into account. They do in fact take into account just about everything you can imagine.

But here's the thing. When pricing a security every one of the variables is negotiable between buyer and seller. Each variable effects the price in a different way. So the variables are there, but sometimes when the pressure is on it is not possible to come to an agreement about what factor to use in one or more of the variables.

When traders are under pressure they are more likely to have weak convictions about pricing variables, after all they are in a rush, so they fade the bid to build a cushion against error.

Give the markets enough time to calm down, take the pressure off, and calm rational negotiations between buyer and seller can take place and then reasonable valuations can and will be set.

Also note that McCain didn't cosponsor S. 190 until May 2006, well after it had already died in committee. Not only did he not champion it, he didn't even mention it until it was already dead.

The comments to this entry are closed.