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

Tuesday, December 13, 2011

Lawler on NAR Revisions for 2007 through 2011

by Calculated Risk on 12/13/2011 03:35:00 PM

CR Note: Economist Tom Lawler first noted that the NAR appeared to be overestimating sales in 2009. In late 2010 other economists approached NAR with questions. From Nick Timiraos at the WSJ in Feb 2010: Home Sales Data Doubted

Several economists approached NAR late last year [late 2010] with questions about its modeling. NAR economists promised to study the issue during a December conference call that included economists from the Mortgage Bankers Association, Fannie Mae, Freddie Mac, the Federal Reserve, the Federal Housing Finance Agency and CoreLogic.
In January, I hinted about this meeting and that there would be significant downward revisions. The revisions will be announced on Dec 21st.

From economist Tom Lawler: NAR to Release Existing Home Sales Revisions this Month

The National Association of Realtors yesterday sent out a media advisory [announcing] that it would release its benchmark revisions to its existing home sales estimates on December 21st. Here is what the NAR sent out:
The National Association of Realtors will release benchmark revisions to existing-home sales at the December 21 lock-up news briefing for the November report, which will begin 30 minutes before the normal briefing at 9:30 a.m. for monthly data. All data can be released at 10:00 a.m. EST.

Although there are downward revisions for total sales in recent years, there is little change to previously reported monthly comparisons or characterizations based on percentage change. There is a comparable downward revision to unsold inventory, so there is no change to relative month’s supply. Also, there is no change to median home prices.

An up-drift in sales projections developed over time between the fixed model for calculating sales rates and the actual marketplace, including growth in multiple listing service coverage areas, geographic population shifts, a decline in for-sale-by-owner transactions, some new-home sales trickling into MLS data and some individual sales being recorded in more than one MLS. Divergence of the data with other housing data metrics began in 2007, so revisions for 2007 through the present will be released.

NAR began to capture a larger share of actual transactions than was assumed in the calculation model based on the 2000 Census; resolving these issues has taking longer than anticipated in the absence of decennial data from the U.S. Census Bureau, which are no longer collected. Other major statistical series such as Gross Domestic Product and employment figures go through comparable periodic benchmark revisions to produce the most accurate data possible; the new benchmark process will permit much more frequent revisions.

NAR began its normal process for benchmarking sales at the beginning of this year in consultation with outside housing market experts. Data for the new benchmark was presented to and discussed with representatives of organizations including the Federal Reserve Board, Department of Housing and Urban Development, Freddie Mac, Fannie Mae, Mortgage Bankers Association, National Assocation of Home Builders, CoreLogic, etc.; and some individual economists.

Normal annual revisions will be released with January existing-home sales on February 22, 2012. Those revisions are expected to be minor and will fine-tune the data back though 2007.
While the NAR did not hint at the magnitude of the downward revisions, the “consensus” is that 2010 existing home sales will be revised downward by about 13% or so (yup, there’s a “consensus” for everything!).

In the past the NAR has done benchmark revisions based on decennial Census data (for owner-occupied transactions) and the decennial Residential Finance Survey (for “investor/vacant home” transactions, with this latter estimate being pretty “squishy"). Because of data availability lags, past benchmark revisions have not been released until MANY years after the end of a decade. The benchmark revisions for 1999 were not released until February 2005, and resulted in decline in estimated existing SF home sales of about 11%.

This decade, of course, the NAR could not continue with the same methodology, as the decennial Census no longer included the so-called “long form” with the data needed for the benchmarking (the “long form” is used in the ACS).

What drove the ACCELERATED benchmarking, however, were reports indicating that in many parts of the country existing home sales based on publicly recorded sales were showing SUBSTANTIALLY lower sales [than] the NAR estimates.

E.g., long-time readers will remember that I first started writing about this trend back in 2009, with my first piece focused on the widening “gap” between the NAR’s estimate of existing home sales in California and Dataquick’s data on “arms-length” existing home sales based on deeds recorded in the Golden State. DQ’s coverage of transactions in California is pretty complete, and its process for weeding out “non-arms-length” transactions seems pretty robust.

This issue got more media attention, however, when CoreLogic wrote a piece in its February 2011 “U.S. Housing and Mortgage Trends” report indicating that its data on existing sales based on public records covering “over 80%” of the US housing market strongly suggested that the NAR’s existing home sales estimates during the housing downturn were significantly overstated.

Many analysts were hoping that the NAR’s new methodology would be based on publicly recorded transactions, and apparently the NAR’s staff actually did explore this avenue. Rumor has it, however, that the new “benchmark” revisions will NOT be based on publicly recorded transactions – in part, apparently, because data coverage in many states is not comprehensive; data quality in many states/counties is poor; AND there are disparities among various private vendor estimates of sales based on publicly-recorded transactions.

An alternative for the NAR would be to use an approach similar to its old “Census/RFS” approach, but instead (1) to use the American Community Survey data for owner-occupied existing transactions; and (2) to make “crude” assumptions about turnover rates (and use some American Housing Survey data (ick!) to “guesstimate” transactions on investor/vacant homes.

Any approach, however, will result in a material reduction in estimated sales over the last few years – though the result will still be estimates and not actuals.

CR Note: This was from economist Tom Lawler.