Using median (or mean) house prices as one big disadvantage: they are extremely sensitive to changes in the underlying distribution.

For example, maybe in 2006 60% of the homes were 3br, 30% were 4br, and 10% were 5br. Fast forward to 2007, and let's pretend that 40% of sales were 3br homes, 20% 4br, and 40% was 5br. Undoubtedly, the average selling price would have increased from 2007, but can we say that prices of the individual homes increased? There is no way for us to tell. It could be that the increase in the average sales price was due to the shift towards more expensive 5br homes (that perhaps actually decreased in value).

That's where the Case-Shiller method comes in (although I understand that OFHEO uses a similar method). CS looks at pairs of sales. In other words, it calculates the change in value for a single house over time. The index for a particular time is a function of the average price change, as opposed to the change in the average price (see example in above paragraph).

OK, so it's a great method. But it's a real pain in the ass to mine county records by hand. So, here I'm going to introduce a new method, which norms the data based upon the 2006 county assessment. Why norm to the county assessment? Because, like Case-Shiller, it is gives us an apples-to-apples comparison: I can compare the sale price of a house in 1990 to the 2006 assessment. Likewise, I can compare the sales price in 2010 against the 2006 assessment.

Note on Fairfax assessments: unlike other regions of the country, Fairfax homes are assessed every year based on recent sales of comparable homes. If your neighbor just sold their home for $800k, that means that your identical home will be assessed around $800k next year. Granted, any assessment, like any appraisal, has some error, but it is a reasonable baseline.

Below, I present you the normed sales trends for Oakton and McLean, which are two of the wealthiest areas in the region. But first, a little about the methodology.

The homes are all 4br SFH homes. The list was first gathered using Redfin (www.redfin.com) to gather all homes sold in the last 36 months (back to early 2005 or so). Sales from before that time were gathered from the Washington Post's real-estate sales database.

Tax information was gathered from the Fairfax County online database (http://icare.fairfaxcounty.gov/).

Homes excluded:
Sales to relatives
Teardowns, or other instances where the property was substantially improved
Sales of properties split into multiple parcels
Foreclosure and REO sales to the bank

Homes included:
'Normal' arms-length sales
Relocations
REOs, foreclosure, short sales when the sale is to a new owner. That is, it does not include the sale back to the bank.




As can be seen for both McLean and Oakton, the best fitting straight line doesn't describe the data (R2 ~ .01), whereas the polynomials account for 26-33% of the variation.  You can also see that most homes in those two areas are starting to sell below their 2006 tax assessment, whereas at the peak in 2006 most sold for substantially above the 2006 assessment.