Information, Search, and House Prices: Revisited

Their conclusion was also based on their finding that there was no price difference between first-time versus repeat buyers.

For example, there is no variable accounting for differences in perceived public school quality, which surely varies within a city of Glasgow’s size and has been shown to have a strong impact on housing price. For educational continuity, the moves of local residents may be constrained to market areas that would allow their children to remain in their current schools, even if these schools are of lower quality. In-migrants, on the other hand, have no such constraint and may therefore end up buying in better school zones.

Lambson et al. refer to the latter possibility the “anchoring hypothesis” based on the idea that buyers’ beliefs are anchored on what they have become accustomed to in the past.

For example, out-of-town buyers with school-age children will generally want to move into their new residence before a new school year begins. In-town movers, particularly those moving within the same school district, will not face such constraints on the search process.

See Turnbull and Sirmans (p. 548) for the derivation of this result.

This issue is investigated extensively by Harding et al. (2003), who suggest an innovative identification strategy for addressing the omitted variables problem.

As the identifier for the previous residence is only reported for buyers that have made use of portability, our sample is restricted to intra-Florida movers who previously owned a home in the state.

The use of tax assessors’ estimates of market value in lieu of including property characteristics in a hedonic price model was pioneered by Clapp and Giaccotto (1992).

County property appraisers rely on Computer Assisted Mass Appraisal (CAMA) programs that use information from recent sales to predict what each property would sell for on January 1 of the tax roll year. The Florida Department of Revenue annually audits each county’s market value estimates for accuracy.

We utilize the U.S. Office of Management and Budget’s November 2008 definitions for metropolitan statistical areas, accessible at the following URL: http://​www.​census.​gov/​population/​www/​metroareas/​metrodef.​html. Sales occurring in non-metropolitan areas are excluded from the analysis when using the NEWMARKET variable, as it is unclear how housing markets should be defined in non-metropolitan areas.

Such a relationship may arise, for instance, if transportation costs increase substantially with distance.

More information on the HUD Neighborhood Stabilization Data can be found at http://​www.​huduser.​org/​portal/​datasets/​NSP.​html.

Of the 2,000 census tracts included in our sample 685 had estimated foreclosure rates in excess of 8%. Dropping sales in these tracts eliminated 2,006 sales from the sample.

More information on the PLSS can be found at http://​nationalatlas.​gov/​articles/​boundaries/​a_​plss.​html. Each township/range/section combination typically corresponds to a one-mile by one-mile square. Although quite rare, it can be the case that the TRS code identifies a geographic area that is larger or smaller than one square mile. This might be the case, for instance, in sections near jagged borders (e.g., near the coast).

As a basis for comparison, TRS squares are generally smaller than census tracts. In suburban Miami-Dade County, for instance, there are approximately four TRS squares in a typical census tract.

All of the models reported here express the natural logarithm of the sales price as a function of the natural logarithm of each of the regressors. In another set of models, we estimated equations in which the natural logarithm of the sales price was expressed as a function of the test variables expressed in levels. The results from estimating these log-linear models were qualitatively similar to the results from the log-log specification. As both classes of models appeared to fit the data equally well, we focus here on the log-log results because the coefficients can be interpreted as elasticities.

In his discussant’s comments, Geoffrey Turnbull offered an alternative explanation for our results; namely, that they are due to local buyers more frequently purchasing “for-sale-by-owner” (FSBO) homes. The validity of this explanation hinges upon whether 1) local buyers do in fact more frequently purchase FSBO homes, and 2) FSBO homes, ceteris paribus, sell for less than homes listed with a broker. We could find no evidence on 1), but it does have intuitive appeal, since it is reasonable to believe that local buyers are under less time pressure to find a new home, and therefore can visit a larger number of prospective neighborhoods and discover more FSBO homes. Regarding 2), the evidence is highly mixed. In their review of the studies that have tested the hypothesis that broker-listed homes sell for more, Yavas and Colwell (1995) cite 3 studies whose results support the hypothesis and an equal number whose results did not support the hypothesis. Their own evidence also did not support the hypothesis; in fact, they found that the seller’s use of a multiple listing service reduced the sales price by 5.7%. The intuition underlying this result is that brokers search less when a the seller quotes a higher price because a higher price reduces the probability that a buyer will purchase the house. The most recent evidence on the hypothesis comes from Hendel et al. (2009), who find that listing the house shortens the time it takes to sell the house, but does not increase the final sales price.

Staying with the above example, if a buyer moves within the same block group, PREVIOUSPFOOT and CURRENTPFOOT are the same, which implies that ln(PFOOTRATIO_BG) is zero. It should be noted that these specifications do allow for anchoring effects to influence buyers that are moving within the same metropolitan area but across anchoring geographies. For instance, if a buyer is changing jurisdictions within the same metropolitan area, the NEWMARKET term will be equal to zero, but the ln(PFOOTRATIO_JURIS) term will likely be non-zero.

In addition to the specifications reported here, we also estimated models in which we: (1) allowed for the NEWMARKET term to vary across markets of different size and (2) allowed for interactions between the anchoring variables and the distance variables. For the latter set of models, the interaction terms were never statistically significant; in the former set of models there did not appear to be a systematic relationship between the NEWMARKET price premium and the size of the market into which the buyer relocated. These results are available from the authors upon request.

In all of the specifications reported in Table 5, intra-market movers serve as the reference category.

At the suggestion of one of the referees, we conducted two additional robustness tests. In the first of these tests, we reestimated the equations reported in Tables 6, 7 and 8 including, in addition to proxy variables for the buyer’s previous neighborhood, the same characteristics of the buyer’s current neighborhood. The second robustness test involved implementing the identification strategy outlined in Harding et al. (2003), in which the characteristics of the buyer’s previous and current neighborhood enter the regression equation in sums and differences. The results of these robustness tests, available upon request, produced coefficients on the anchoring and distance variables that were very similar in magnitude and significance to those reported in Tables 6, 7 and 8.