Intermediate Treasury rates have pulled back from the brink, with the yield on the 10-year Treasury dropping by more than 20 basis points from its highs.  While the Fed's program of "qualitative easing" did not result in huge purchases, it appears that the possibility of large-scale buying by the Fed did give yields a sufficient downward nudge.  It is interesting that Primary Dealer positions in Treasuries with 6-11 year maturities peaked on April 29th at $9.9 billion, which was coincidental with the rise in the 10-year yield to its peak of 3.34% on May 7th.  While it briefly moved past the 3.30% level, it did not stay there long enough to trigger the type of duration-shedding that might have further weighed on rates.
However, it is noteworthy that mortgages began to trade much "longer" as Treasury rates rose, even though MBS' model durations did not change significantly.  As the 5-year Treasury yield rose from 1.70% on April 15th to a peak of 2.17% on May 7th, the "empirical duration" of Fannie 4.5s rose from roughly 2.0 on April 15th to over 3.0 by Monday, a 50% increase.  (Empirical durations are a way of measuring the relative price volatility of an asset by using a linear regression of its daily price changes versus daily changes in yield of a benchmark.  The measure I quoted above compared price changes to yields on the 5-year Treasury over rolling 20-day periods.)  Over the same period, the durations being generated by valuation models did not change significantly; the option-adjusted duration generated by Yield Book for Fannie 4.5s increased by only 13%.  This is largely because these applications are driven by mortgage rates (which were fairly steady over the period in question) rather than Treasury yields.

It is interesting that investors began to treat MBS as a longer-duration asset in the selloff, even though most models did not validate the lengthening.  This has a number of implications.  Investors appear to look for longer MBS durations as a visceral reaction to the Treasury selloff, rather than something driven by their models.  It also suggests at least some of the backup in rates was attributable to MBS-related duration shedding.  On the other hand, the large-scale portfolio restructurings that would have pushed yields another 20-40 basis points higher did not take place because of the lack of validation from the models, mainly because primary mortgage rates defied the upward trend in yields and stayed around or below 5%.
I've noted in a number of columns that the MBA's refi index has recently reported some strange numbers, and have been asked for some clarification.  To do that, I'd like to take a step back and discuss what the index is, how it's compiled, and what it means.

The MBA (Mortgage Bankers Association, for the uninitiated) began putting out the refi index in 1990, responding to investors' requests for more data on the short-term prepayment outlook.   This took on greater urgency in the late 1980s, after the creation of "stripped mortgage-backed securities."  The refi index is part of the MBA's weekly application survey.  On a weekly basis, a select group of lenders sends in the count and dollar value of all loan applications taken for the prior week, broken out for conventional and government-backed programs.  Lenders also send other data, such as the percentage of loans taken for refinancings and for adjustable-rate loans.  The indices are then generated using the data supplied by the lenders.

The MBA calculates the indices by comparing the weekly loan totals to the number of applications taken for the week ending March 16, 1990 (which represents the base level of 100 for all the indices).  For example, if 500 applications were reported for the week ending 3/16/90 and 5000 applications were reported for a particular week, the index for that week would be 1,000.  While all indices are reported as being seasonally adjusted (along with the non-seasonally adjusted numbers), only the overall ("composite") and purchase indices are substantially revised using seasonal adjustment factors.  The seasonal adjustments for the refi index consist solely of accounting for holidays.  Interesting, most holidays (even where virtually all banks are closed for business) are accounted for by adjusting the day-count for the week by ½ day.  The choice of day-count by the MBA for the refi index can be rather arbitrary, and creates numerous distortions, particularly during seasons that have many holidays. Lenders only report "retail" originations, as the MBA has traditionally viewed application data for third-party lending to be unreliable.

While the MBA's indices are useful tools in gauging short-term activity levels and trends, there are a number of factors that make them somewhat unreliable as long-run indicators.  The most important factors distorting the indices are changes in both the lender population and the composition of loans products over time.  This is especially true over the past few years.  For example, during the period from 2002-2006, a lot of lending volumes gravitated from large depositories to both mortgage bankers and subprime lenders.  The subprime lenders (i.e., New Century, Ames, etc.) were never included in the survey.  In addition, large lenders such as Countrywide segregated their subprime lending from their retail channels, which meant that their subprime production was not included in the data submitted to the MBA.   This means that the index consistently understated overall activity during that period, at least by comparison to earlier periods.  Moreover, the demise of subprime and alt-A lending combined with the recent industry consolidation makes comparing activity levels over the periods quite misleading.

A recent change has also impacted probably impacted the indices.  The lending activities of Countrywide were officially absorbed into Bank of America Home Loans as of April 27th.  Interesting, the refi index dropped about 22% for the week ending April 24th, even though mortgage rates were steady to slightly lower over the period.  This suggests that Bank of America (one of the country's top three mortgage lenders) made some internal reporting changes that sharply reduced the number of applications being reported to the MBA.  In turn, this would have resulted in a drop in the index.  This means that individuals trying to judge lending activity have to account for this substantial drop as a one-time event caused by extraneous factors, as the drop did not reflect a major decline in activity levels.