Housing Prices: Comparing Shiller and HPI
Shiller's model is an improvement over the HPI but for practical purposes and year over year comparisons, both models seem fatally flawed.
Eyes are focused on the upcoming Office of Federal Housing Enterprise Oversight (OFHEO) housing price report due out on Thursday to hopefully tell us what we already know: home prices are falling.
For reference purposes the fourth quarter 2006 headline from OFHEO was U.S. House Price Appreciation Rate Steadies. Yes, the word appreciation was still in the headline. Here is a snip:
"The rate of home price appreciation in the U.S. remained steady in the fourth quarter of 2006, extending a general trend of deceleration begun earlier in the year. Home prices, based on repeat sales and refinancings, were 1.1 percent higher in the fourth quarter than they were in the third quarter of 2006. This is slightly above the revised growth estimate of 1.0 percent from the second to the third quarter..."
Does anyone possibly believe that housing prices in general were still rising at the end of 2006? I don't. I doubt even David Lereah believes that.
Shiller Reports First Price Drop Since 1991
According to the Shiller Index U.S. home prices fell for first time since 1991:
"U.S. home prices dropped 1.4% in the first quarter compared with a year earlier, the first year-over-year decline in national home prices since 1991, according to the S&P/Case-Shiller index released Tuesday..."
Here are a few of snips about the HPI (Housing Price Index) as computed by the OFHEO.
Each quarter, Fannie Mae and Freddie Mac provide OFHEO with information on their most recent mortgage transactions. These data are combined with the data of the previous 32 years to establish price differentials on properties where more than one mortgage transaction has occurred. The data are merged, creating an updated historical database that is then used to estimate the HPI.
The HPI is a weighted, repeat-sales index, meaning that it measures average price changes in repeat sales or refinancings on the same properties.
The HPI is based on repeat transactions. That is, the estimates of appreciation are based on repeated valuations of the same property over time. Therefore, each time a property "repeats" in the form of a sale or refinance, average appreciation since the prior sale/refinance period is influenced.
To be eligible for inclusion in the indices, a house must be a single-family dwelling. Condominiums and co-ops are specifically excluded. Houses included in the indices must also have two or more recorded arms-length sale transactions. As a result, new construction is excluded.
High Turnover Frequency
Data related to homes that sell more than once within six months are excluded from the calculation of any indices. Historical and statistical data indicate that sales made within a short interval often indicate that one of the transactions 1) is not arms-length, 2) precedes or follows the redevelopment of a property, or 3) is a fraudulent transaction.
Time Interval Adjustments
Sales pairs are also weighted based on the time interval between the first and second sales. If a sales pair interval is longer, then it is more likely that a house may have experienced physical changes. Sales pairs with longer intervals are, therefore, given less weight than sales pairs with shorter intervals.
Shiller attempts to adjust for time factors by giving more weighting to repeat sales (as long as they are at least six months apart) as the following chart shows.
Shiller goes on to make a lengthy mathematical presentation of his model but there is theory and there is practice.
His model shows that prices in Tampa are down 3%, Phoenix down 3%, and prices rose 1% in Miami. That is what the data shows (in theory). In practice I do not buy it. There is simply no way home prices rose in Miami in 2006 nor is there any way prices in Tampa or Phoenix are down as little as reported.
What Went Wrong?
Supposedly the index's main advantage is that it only looks at repeat sales. Perhaps that is fine in a steady market with frequent repeat sales, where the rate of appreciation is uniform, but how often do those conditions exist? How long does a blowoff top spike keep influencing the data? And what happens at major turning points or when repeat sales take longer and longer?
Based on the above weightings chart, a home that resells with an interval of 10 years has a weighting of 0.8 while one that sells in six months (but not sooner) has a weighting of 1. Let's consider a house that sold in 2002 for $200,000. It sells in 2007 for $250,000. Is that a gain? How much? What if that house would have sold for $400,000 in 2005 and $300,000 in 2006 based on comparables? There is no way to account for this in his model because it looks at only the seen. Also unseen are potential upgrades such as new kitchen, incentives, etc. that were done to make the home "more sellable." Extensive upgrades should be factored in but they are not.
Both indices exclude new home sales even when the homebuilder is still building the same model year after year (at now dramatically reduced prices). With homebuilders discounting new homes by hundreds of thousands of dollars and 20% or more (not counting incentives) it is absurd to believe (or even report) that Miami home prices are rising by 1% or Tampa prices are falling by a mere 3%.
But I am not just talking Florida. Homebuilders nationwide have been reducing prices dramatically in nearly every market, even more so when one includes incentives. Try telling the guy trying to bail on his condo in any city anywhere that prices are down a mere 1.4% last year. See what kind of reaction you get. Ooops, condo prices are not included in Shiller's index either.
Also note that neither model looks at incentives and incentive have been massive lately (but nonexistent at the blowoff peak in 2005).
The differences between Shiller and the HPI seem to be that Shiller does not look at refinancings but the HPI does, and the HPI has a price cutoff based on Fannie Mae (FNM) conforming loans at $417,000 but Shiller does not. Otherwise they both seem to have a fatal flaw in assuming that there was an increase in price over a five year period (on a house that last sold in 2002) when in fact there may have been huge unrealized gains three years and a huge unrealized loss in year five.
By attempting to eliminate distortions caused by optimistic appraisals and evaluation of comparables, other serious flaws in the computations were introduced. While it's true that the faster the housing turnover the less the bias (barring fraud), does a difference in weightings of 1.0 vs. 0.8 over a six month period vs. a 10 year period properly account for that bias?
More importantly, for long timespans it is not what happened but how it happened that is important. A model that looks at a sales price (and a sales price only) from 10 years ago vs. today cannot properly account for year to year variations in price. Prices may have gone up five years were flat three years and fell for two years but if the price appreciated over 10 years the model is going to be biased towards a year over year gain when in fact there may have been a loss two consecutive years.
It will be interesting to see if the OFHEO tells us what everyone already knows: home prices are falling. But what the index will not show is by how much. The methodology of looking only at previous sales is simply too flawed. Shiller's model is an improvement over the HPI but for practical purposes and year over year comparisons, both models seem fatally flawed.
Copyright 2011 Minyanville Media, Inc. All Rights Reserved.
Daily Recap Newsletter