by Calculated Risk on 2/17/2011 06:09:00 PM

The National Association of Realtors (NAR) responded today to recent stories about the NAR overstating sales. This was in response to Calculated Risk posts in January (here, here and here), and to CoreLogic's release on Tuesday. Barry Ritholtz picked up on the story this morning.

From the NAR: Existing Home Sales: Benchmarking FAQ

Home sales measurements have shown to differ at times as reported by NAR and some outside organizations. Here are the facts and background you need to know.

How are NAR home sales computed?

NAR collects sales data from numerous MLSs, with a reporting sample of about 40 percent. If data computes to be a 5 percent increase from one year ago then we say home sales rose 5 percent from one year ago.

What about 5 million home sales? An increase of 5 percent is understood, but how is 5 or 6 million home sales computed?

A base figure is used from Census 2000 where one can compute how many homes were bought. If you recall there was a long-form of Census which asked questions about whether you moved or not and whether or not you bought a home. Based on this, one knows that 5.2 million existing homes were sold in 2000. Note that this benchmarking process does not use any data from MLSs. Hence, it is considered clean. With this base figure, we then apply the percent changes to sales obtained from MLSs. So if MLSs data addition says a 5 percent increase then we would say there were 5.4 million home sales.

How can NAR sales data drift away from true measure?

It is not definitive if NAR data has a measurable drift other than normal small statistical noise that may arise from not using all MLSs and from any data entry error or local MLSs sending wrong data to NAR. In statistics, one just assumes the positive and negative noises cancel each other out. It is however possible for this statistical noise to drift mostly in one direction and hence cumulatively add up over many years. In our last benchmark in year 2000, we found the reported home sales had a 13 percent upward drift compared to what Census data implied. NAR then revised the past 1990s data to match up with the Census data.

How are other home sales data computed?

Most other data comes from courthouse recordings. Generally, they make some assumptions about noncovered areas. Because of improved electronic recordings, they claim to capture more data and more quickly than in the past.

When will the new benchmarking take place?

A: In 2010 Census, a long-form questionnaire was not used. Therefore, the Census no longer asked about whether people moved and bought a home. So another brand new benchmarking process is needed. NAR has already been in contact with all key housing economists in the industry and government agencies and a few in the academia about finding a new benchmarking process. We expect a new clean, agreed-upon benchmark figure by the summer of this year.

In addition, we will be determining a new way to re-benchmark on a more frequent basis, possibly annually, to lessen any drift that can accumulate over time. This frequent re-benchmarking, rather than waiting every 10 years, is needed since the Census no longer collects a long-form questionnaire. With benchmarking, we will be working with various outside housing economists to develop a new-agreed upon method.

So the revisions should come this summer. I expect the NAR to revise down sales for the last 4 or 5 years, with significant downward revisions for 2009 and 2010 (CoreLogic estimated 15% to 20%).