Thursday, October 09, 2008

Overselling the Subprime Problem

These days the US Presidential race is getting down to the final days - less than a month left. As with most Presidential races, it boils down to 'the economy, stupid.' Two months ago, talking about the indifferent economy mainly dealt with oil prices, and those big, bad guys, the "oil speculators."

That was a joke of a bogeyman, and now, a few months later, when oil is down to less than $90, no one is talking about those speculators - now those folks who bought when oil was $140 are no longer evil speculators, just bad investors. Then, the mainstream business media didn't really explain how oil "speculation" could have a negative long-term impact on things.

These days, the monster under the bed is subprime mortgages. You know, those Wall Streeters and their financial engineering, their "securitizations" - my gosh, what a long word, it must be a complex and difficult to understand concept, I mean, it even has a 10-point Scrabble tile in it, it must be nefarious!

The popular understanding of securitization is some sort of fancy slice-and-dice that turns bad assets (subprime debt) into multiples of good assets, but this understanding is just wrong.

Everyone knows what a mortgage is, and most understand correctly that subprime mortgages means loans to people who, on average, will default on their loans more often than the regular ("prime") market. To offset the fact that they will default more often, you as the bank charge a higher interest rate. Makes sense, and simple enough.

Making a loan and pricing the loan (i.e. setting the level of interest to charge) is a forward looking bet. All else being equal, the price you charge for a loan depends on how likely the borrower will pay you back. So how do you predict whether the borrower will pay you back? You use things such as credit scores, and divide borrowers into subprime and prime markets.

But what if you can make a backwards looking bet? What if you can build a time machine, so you don't need to predict whether someone will pay you back (regardless of their credit score), you'll actually know if they will pay you or not?

If you can know the future, you can make money. This is what securitization tries to do.

A few decades back some financial types and lawyers (I'm assuming) got another simple idea: let's build something close to a time machine.

How do you do this?

First, let us assume that 25% of all subprime loans will default. That sounds bad, but that also means 75% will pay you.

So if you make one loan, you have a 75% chance of getting 100% of your money and a 25% chance of getting 0% of your money back (disregarding partial payments prior to default).

Second, we would quickly realize that if you make 100 loans (or 1,000 or 1M, etc.), then you know that statistically you will get 75% of your money back.

So how do I build this (imperfect) time machine with these realizations? Well, you first bundle up a lot of subprime loans to get statistics on your side (law of large numbers and everything), and then you divide up the loans into discrete packages.

So we bundle up 100 subprime loans and then divide them into four packages, each package having the rights to payments from 25 loans. And now, to sprinkle on the (imperfect) time machine dust, we say that the first package will be entitled to the first 25 loans out of the 100 that gets paid off, the second package will be entitled to the next 25, and so on.

This is the financial engineering magic - instead of having 100 random subprime loans with an expected value of 75%, we can bundle up the loans and then divide them up into smaller packages *with conditions attached* to turn the 100 loans into four packages, the first three having an expected value of 100% and the last package of 25 loans having an expected value of 0%.

If you are an investor buying package 1 or package 2, you'll feel pretty confident of getting repaid. If you are buying package 3, statistics say that you will get repaid 100% but you are a little bit uncomfortable.. so Wall Street gets an insurance company, say AIG, to guarantee that, out of those 25 loans, AIG will pay after the first 5 defaults. The expected default rate for package 3 is 0%, so AIG is comfortable in accepting an insurance premium to write this type of insurance - after all, statistically and if the assumptions hold up, it will never have to pay out on the insurance. Now, with the insurance, the investor is comfortable that it will get repaid and therefore will buy package 3 at a price that makes Wall Street money.

That, simply put, is what securitization does. The idea is that simple. And because the financial and legal types like to think of themselves as geniuses, they call these packages "tranches."  So is this creating value out of crap? Financial mumbo-jumbo, a slice-and-dice that ends up shredding investors?

No, it's just unlocking hidden value. 100 random subprime loans can (and are) worth more when you package them into tranches. There are a lot of things in life that are worth less than the sum of its parts.  Look at a car junkyard - it's a business only because a car parted out is worth more than the whole car.  So too subprime loan, parted out through securitization, are worth more than the original loans.

There is a lot of talk about Wall Street securitizing loans, and making all these fancy and exotic investment vehicles which then turned out to be worthless, but I just don't buy it.  No matter how many packages you divide a bundle of loans into, you cannot create more than the original 100 loans, so financial engineering doesn't expand the universe of loans (i.e. doesn't expand the universe of risk), it just divides it up into smaller packages.

So the subprime collapse only happens when reality is different from assumptions. We assumed a 25% default rate. What happens when it is actually 35%? Well, tranche 1 and 2 are still the same (65 loans get repaid, and 1 and 2 "take up" only 50 loans). Tranche 4 is still the same, it's still crap (0 loans get paid).

For tranche 3, the investors expected to get fully repaid (i.e. 25 good loans), but with a 35% default, there are only 15 good loans left (65-50 = 15).  AIG, our insurer, also expected 25 good loans remaining when it insured against 5 bad loans.

With a 35% default rate, the tranche 3 investors will collect on the 15 actual good loans and the 5 insured loans from AIG, and will take a loss of 5 loans.  AIG will take a loss of 5 loans against their income from the insurance premium.

So is there a subprime problem?  Yes and no. Yes, in reality default rates have likely been higher than the assumed rates in the models used to price the tranches, but the failures are not due to problems with black magic. It's a simple error in assuming default rates by investors and insurers.

Investors and insurers only lose more money than expected when the default rates are higher than expected.  Outside of "naked" insurance policies (aka naked Credit Default Swaps), there is no expansion of risk in securitization.  AIG lost money because it did not assess the risk in providing insurance properly.

And the answer is "No" because, for all the negative press on the subprime market, it is surprisingly not that bad.  While the data is dated, as of last year the total dollar amount of outstanding subprime loans is 1.3 trillion, with a default rate of 14.5%, and the average loan amount being $180k.

That means that about $200B worth of loans is in default. So why is there a $700B bailout?

1 comment:

Anonymous said...

Actually, at the time of the bailout, the number was 144 billion for ALL mortgages 90 days overdue. To pay them off, not just bring current.

http://www.federalreserve.gov/releases/z1/