Matt Levine, newly arrived at Bloomberg View, has a very smart response to my post about those weird jumbo mortgage rates. Levine comes up with two reasons why jumbo rates might be lower than the rates on loans which can be sold to Frannie, and both of them are entirely plausible.
The first is that credit risk on conforming mortgages doesn't simply disappear just by dint of those mortgages being sold to Frannie (OTCQB:FNMA). The agencies need to charge a fee to cover the credit risk on the mortgages that they're buying, and that fee is going to find its way, one way or another, into the yield on conforming mortgages. Since it stands to reason that the credit risk on conforming mortgages is greater than the credit risk on jumbo mortgages (on the grounds that rich people, in general, are more creditworthy) then it similarly makes sense that the all-in yield on conforming mortgages might be higher too.
Levine's second hypothesis is related to the fact that bonds in general, and agency bonds in particular, are instruments which are marked to market daily - while jumbo loans are long-term assets which can sit on a bank's balance sheet for decades, through many credit cycles. If a bank buys mortgage bonds, notes Levine, then it is obliged to mark them down, taking a hit to its P&L, if and when mortgage rates rise. Actual mortgages, on the other hand, not being marked to market, never need to suffer such markdowns. And that makes them more attractive. Conforming mortgages will always end up being priced off the market, and during times when banks expect interest rates to rise, the market might well be quite expensive, compared to loans which are designed to be held to maturity.
Both of Levine's ideas are pretty good ex post explanations for why jumbo mortgage rates might be lower than the yield on agency bonds. But Levine - along with Matt Yglesias, and myself - failed to ask the most obvious question: are jumbo mortgage rates, in actual fact, lower than the rates on conforming mortgages? It certainly looks that way, in the WSJ article. But then I got a very interesting email from Keith Gumbinger, of mortgage-rate information service HSH.com.
The WSJ's Nick Timiraos cites some data about the spread between the two rates from HSH, but his chart uses information from the Mortgage Bankers Association. And what Timiraos never says is that if he just stuck to HSH data throughout, the spread would never have turned negative.
Part of the problem is that there's no such thing as a simple, commodity mortgage. These things are all pretty complex beasts, and most of them include the borrower paying "points" up front, which need to be converted into percentage points using a rule of thumb like four "points" = 1 percentage point. According to HSH's data, jumbo mortgage rates were actually closer to 4.86% last week, a full 15 basis points more than the WSJ's 4.71% figure.
On top of that, according to HSH, the MBA figures used by the WSJ show conforming mortgage rates about 12-15bp higher than the figures coming out of both HSH and Freddie Mac (OTCQB:FMCC).
In other words, use a different dataset, and you don't see the crossover phenomenon at all. Gumbinger has a few ideas about why the MBA's data might be such an outlier; one reason, he says, is that the MBA counts all mortgages of more than $417,000 as jumbo mortgages, even when they're in markets like New York and Los Angeles where mortgages can be conforming when they're as large as $625,500. Additionally, he says, the MBA rates reflect actual quoted mortgage rates to borrowers - which means that if borrowers are becoming less creditworthy for whatever reason, the rise in conforming rates might not really be comparing apples with apples. For a borrower of given creditworthiness, mortgage rates won't have risen as much as the WSJ chart shows.
All of which is symptomatic of a broader truth: the minute you start seeing headlines about some yield trading through some other yield, be suspicious. Fixed-income instruments, be they bonds or loans, tend to be extremely illiquid, and although a chart can make it seem as though they're priced and traded to the nearest basis point, in reality their pricing is much, much fuzzier than that. Outside Treasury bonds and the primary market, it just doesn't make sense to talk about "the yield" on a certain credit, or "the price" of a certain bond: when there are generally dozens of different instruments outstanding, none of which trade very much, the price and the yield are pretty much whatever you want them to be. You might be able to get a vague idea of the range in which they've been trading, but we're not talking about things like stocks, here, where the price at any given point in the day can be nailed down to the nearest cent.
A more realistic chart, then, from the WSJ, would have shown a wide and fuzzy green line, and a wide and fuzzy blue line, and those two lines coming closer to overlapping a little bit. In fact, all charts of secondary-market credit spreads should look like that. (Mortgage rates are primary market, not secondary market, but the principle is the same.)
In a ZIRP world, it's easy to get excited about a couple of basis points here or there. But before you do, remember the error bars. Because yields on any credit product are ultimately Heisenbergian: the closer you look, the harder they are to nail down.