The US Real Estate Market is arguably the largest financial market in the world. For the past several years there have been over $2 trillion of new mortgages. Further, for the majority of people in the U.S. their house, and the equity they have in it, represents their single largest investment and the vast majority of their wealth. As a result, changes in housing prices can have enormous impact on the economy resulting in changes in credit, wealth, purchasing patterns, employment and most major aspects of the economy.
Given the size of the market and the significant impact that it has on the economy, it is surprising that there does not exist a definitive measure of housing prices. Currently, the two primary indices of residential prices which are often referenced in the press are the House Price Index produced by the Federal Housing Finance Agency (FHFA) and the S&P Case-Shiller Index™. The FHFA index is the same index that was previously produced by the Office of Federal Housing Enterprise Oversight (OFHEO). The other marker of residential price changes that is often referenced is the median sales price by region during the most recent period. It is produced by the National Association of Realtors (NAR). All of these indexes are based on the houses that have sold in the most recent period and ignore the houses that are not sold.
The challenge facing anyone attempting to produce a residential property index is that houses, unlike securities or consumer products, are each unique and are traded infrequently. For example, the often quoted advice as to the three most important factors to consider when buying a house, location, location and location, is generally ignored when the above indices are computed. Houses from undesirable areas are lumped with houses from desirable areas and no adjustment to the makeup of the pool is made to the indexes. As a result, it is often unclear whether the change in the index is a result of a true change in price or rather a change in the mix of properties sold.
When an index such as the consumer price index (CPI) is created, the prices for the same basket of goods is compared from one period to the next so that the price change observed is not impacted by the mix of products being purchased. This cannot be done for housing since the same house is not sold repeatedly every month. Therefore, alternative approaches must be used. The FHFA and S&P Case-Shiller indexes are both repeat sales indices1. This approach tries to measure the overall price changes by looking at all of the houses sold in each time period and matching them with their previous sales price. That will create a ratio of price change for each house that sells. Then, an estimate is made of what overall price change best fits all of the ratios for each period and those estimates become the index.
There are numerous well documented problems with this approach. The most obvious is that it assumes that the house does not change over time. In other words, if a house purchased ten years ago sells this month, the underlying assumption is that it has not been modified, that the location is equally desirable to what it was ten years ago, that financing is equally available, that the seller's financial condition is similar, that all aspects of the house are in equally good condition, and that tastes for that type of house are unchanged in ten years. These are obviously very strong and generally incorrect assumptions. It may be that, across many properties with a large enough geographical area, things may average out. However, as the geographical region becomes more focused, that becomes less likely.
Another problem with this type of index is that it does not take into account the mix of properties. For example, when the recent collapse of the secondary market significantly reduced the number of jumbo mortgages that lenders were able to offer, the mix of properties selling in any higher home value region was suddenly impacted. In communities where most homes met conforming limits there was much less impact. This results in over or understating the price changes for the region.
Finally, even assuming that the above issues were not significant, there is another feature of a repeat sales index that makes it undesirable as a basis for a tradable derivative. Since the only houses that are used to compute the index are ones that sold in the current period, these only represent a small percentage of the total housing stock and, therefore cause the index to be somewhat volatile. This often results in estimates that are unstable in that an index reported in the current period is subject to significant revision in the next period and, in fact, for several periods in the future2. Therefore, it is difficult to clear a market against an index that may be significantly revised in the future.
The NAR median price report also provides an ambiguous measure of the market since it is simply the median price of the properties that sold in the current period in a particular geographical region. There are many market and institutional reasons why the houses that sold in a particular period are not representative of the market as a whole.
This all leads to confusion about what is going on in a market. Different indexes even result in contradictory conclusions. These discrepancies result from the issues mentioned above, along with the fact that each of the indexes uses different data to compute their measures. For example, the FHFA index has historically restricted itself to data from conforming mortgages that Fannie Mae or Freddie Mac purchased while the S&P/Case-Shiller index includes data for a broader range of mortgages by using public record data. The NAR data comes from the data reported by their realtor network.
One possible approach to address this dilemma is to consistently use all of the housing stock for each period. To do this it is necessary to have a market price for each house in each period. Since all the properties do not sell, a model must be created to price all of the major attributes of a house (i.e. location, gross living area, age, lot size, bedrooms, bathrooms, etc.) and then compute an estimated market price for each property. The value of each of the attributes is estimated based on the observed properties that do sell over an extended window. This approach allows all of the properties to be included in the index which then provides a stable broad based index for each period. Models of this type are referred to as hedonic models3 and have been used for many decades to estimate the value of a property. The limitation to using this approach for an index has been the general lack of availability of characteristic information about residential properties4.
FNC has developed a hedonic index based on the data collected from public records and blended with data from appraisals. The addition of the appraisal data provides the physical property characteristic data that is often missing from public records. The procedures used to create the index are described in, Dorsey, Hua, Mayer, and Wang (2010) “ Hedonic versus repeat-sales housing price indexes for measuring the recent boom-bust cycle” Journal of Housing Economics 19 (2010) 87–105.
The methodology for a repeat sales index is described in Baily, M., R. Muth and H. Nourse (1963), “A Regression Method for Real Estate Price Index Construction,” Journal of American Statistical Association 58, 933-942. and Case, K.E. and R.J. Shiller (1987) “Prices of Single-Family Homes since 1970: New Indexes for Four Cities,” New England Economic Review (September/October), 45-56.
A discussion of the substantial revisions that can occur can be found in Clapp, J.M. and C. Giaccotto (1999) “Revisions in Repeat-Sales Indexes: Here Today, Gone Tomorrow?” Real Estate Economics 27:1, 79-104. and a discussion of why such indices are not appropriate to be used for home equity insurance or real estate futures can be found in Clapham, E., Englund, P., Quigley, J.M. and C.L. Redfearn (2006) “Revisting the Past and Settling the Score: Index Revision for House Price Derivatives,” Real Estate Economics 34, 275-302.
One of the first papers describing hedonic models was Griliches, Z. (1961) “Hedonic Price Indexes for Automobiles: An Econometric Analysis of Quality Change,” in The Price Statistics of the Federal Government. Washington D.C. Government Printing Office.
A discussion of the data limitations for hedonic models can be found in Rappaport, J. (2007) “A Guide to Aggregate House Price Measures,” Federal Reserve Bank of Kansas City Economic Review Second Quarter 2007, 41-71.