In Depth Analysis: CalculatedRisk Newsletter on Real Estate (Ad Free) Read it here.

## Wednesday, September 12, 2007

### Ready, Set, Reset

by Tanta on 9/12/2007 09:30:00 AM

I saw another media piece on ARM resets this morning. The last time we posted on ARM resets, there came to pass some confusion about the differences among the various published numbers. My very simple purpose today is to help everyone understand why you can, legitimately, get very differing numbers, and what questions you should ask of any data so that you can understand what you’re being told.

It comes down to several questions: Are we using originations data or outstandings data, and if the latter, from what point in time? Are we looking at all ARMs, or just securitized ARMs? (Do note that investment bank sources generally focus on securitized ARMs only, because the performance of securities is their concern, not necessarily the performance of all mortgage loans.) Are we looking only at first reset, or at all resets? What prepayment and cumulative foreclosure assumptions are we using?

Here’s a very concrete example to flesh out the issues. You have a hypothetical 2/28 ARM portfolio of \$1.2 million original balance. It contains 12 \$100,000 loans, one originated per calendar month of 2005. Each loan will have a first rate adjustment in each calendar month of 2007. The “12-month reset projection” for this pool, considering only the first adjustment, is very simple: each month, 1 loan resets, for a dollar amount of \$100,000 per month or \$300,000 per quarter.

But what if you do not limit yourself to just the first reset? The 2/28 will, if it does not prepay, reset every six months after the first reset. If we assume no prepayment, then, and include subsequent adjustments, we get 1 loan resetting in January-June, but 2 loans resetting each month from July-December. Starting in July, there is 1 loan hitting its first reset and 1 loan hitting its second reset. If you simply counted resets, you would show 2 loans in July-December, for a balance of \$200,000 per month. If you tried to total up the monthly balances for a year, you’d end up showing \$1.8 million in resets on a \$1.2 million portfolio of loans. You could say, in a certain context, that \$1.8 million in resets are scheduled for 2007, but that is not saying that \$1.8 million worth of loans are “at risk.”

And, of course, not every loan will survive on the books after its first adjustment. It could pay off voluntarily (refi, home sale) or involuntarily (short sale, foreclosure). If you wanted to take a vintage of originations and project out a reset schedule, you would have to make projections of prepayment and default. If you started with current outstandings, you would already have your prior prepayments and defaults removed from your pool, but you would still have to project these into the future, unless your goal was a “what if” scenario that involved no loan paying off or defaulting until its reset date.

Even if you wanted to do that, there’s no reason to assume that all reset-related defaults will be due solely to the effect of the first adjustment. It is the most wicked reset for the borrower, but the ugly fact of the 2/28 ARM is that borrowers who survive the first adjustment, possibly just barely, will get another smaller one in six months, and then another one in another six months, until the loan reaches either fully-indexed (then-current 6-month LIBOR plus margin) or its lifetime cap (usually start rate plus 6.00 points). Given the depth of the teaser discounts, the hefty margins, and the movement in LIBOR since these loans were originated, there is no reason to think many of them won’t keep adjusting upward every six months for two years until they hit indexed or capped. So the borrower who just barely survived the first reset might go down at the second one. The borrower who more comfortably survived the first reset might go down at the third one. There is a point to “cumulative” projections of resets.

However, you would still have to adjust these numbers further. You would also project index values forward (to guess when caps will come into play and loans would stop adjusting), and you would have to take into account varying margins. I could assume for our hypothetical pool that all loans have the same margin, but in the real world they don’t.

You will, therefore, see differing presentations of reset volume, and those differences may have a lot to do with prepayment speed assumptions, underlying index movement assumptions, or the weight of caps and margins in a particular pool of loans. That does not mean that someone is lying to you, although you may or may not find the underlying assumptions reasonable (assuming you can figure out what they are).

Today, Reuters reports this:

About \$75 billion in adjustable-rate U.S. mortgages are going to reset in the fourth quarter, most of which will emerge next month. Of the loans resetting, around 75 percent are subprime mortgages.
As far as I can determine, this \$75 billion number includes only the first reset of any ARM (the date on which it changes from “fixed to floating” rate), based on Q207 securitized outstandings, and has no prepayment adjustments. If you assume even conservative prepayment speeds, the actual number of resets will be lower. However, if you “add back” subsequent adjustments for loans that survived their first adjustment, the raw number of resets is higher. The Bank of America chart CR posted several weeks ago shows securitized plus non-securitized, which is why it has such large numbers compared to the Reuters number. I believe, but cannot verify, that it also includes only the first adjustment.

There is no “right number.” There is only a number in context.