Building a Better House Price Index With MLS Data

Two Federal Reserve economists have developedrna home price index which they claim operates in a more real time environmentrnthan those indices currently accepted as the industry standard.  Rather than base their index on sales prices,rnElliot Anenberg and Steven Laufer have constructed theirs using a repeat-salesrnapproach but relying on listing data.  Using Data on Seller Behavior to Forecast Short-run House Price Changes was published as part of the Fed’s Financial andrnEconomics Discussion Series.</p

The paper notes that home price changes have consequencesrnfor the economy as they affect both household wealth and an owner’s ability tornborrow.  But information on those prices,rnunlike that of other assets such as stocks, are reported with a significantrntime lag.  The Case-Shiller house pricernindex contains information from several months earlier but has an immediaterneffect on home building related stock prices and it is likely that the pricesrnindexes also have an impact on individual homeowners, policy makers, lenders,rnand others.  By using information that was availablernat the time of therncontract negotiations, the authors designed their list-price index to mitigaternthis information friction.</p

The authorsrnattribute the time lag in producing the traditional indices to a lack ofrnincentive for buyer or seller to publicize the price once they have negotiatedrnan agreement.  Even once the price isrndisclosed at closing there is typically another lengthy delay before the publicrnrecord becomes available.  In contrast,rnbefore the contract is signed the seller has a strong incentive to broadcastrnthe current asking price as a marketing tool. rnThus information on listing prices is disseminated through platformsrnsuch the Multiple Listing Services (MLS) in real time.  Once a sale agreement is reached thosernlistings are removed.  The authorsrntheorize that using the information on home prices before those homes arerndelisted could allow them to learn about the level of sale prices much earlierrnthan what is currently available.  </p

The authorsrndeveloped a new house price index reproducing the Case-Shiller repeat-salesrnindex, substituting sales prices with an estimate based on the final listrnprices of all homes that are delisted. rnKey to the methodology is associating each delisting with the mostrnrecent prior sale of that property, creating a pair of observations analogousrnto a pair of repeat sales in the Case-Shiller and other similar indices.  This allows a timelier index of price trendsrnwhile maintaining the most attractive feature of other indices; their abilityrnto control for changes in the mix of homes sold over time by partialing out arnhouse-specific effect from each price.  </p

Testing their theory was complicated because thernsale-to-list price ratio varies, both in the cross section and across time andrnbecause many delistings are done for reasons other than a sale.  The authors found that some of thesernvariations could be explained by other observable information about sellerrnbehavior such as the time on market (TOM) and the history of list pricernchanges.  This information was used tornadjust the final list price up or down.  </p

The index was tested by using micro data from three largernmetropolitan areas, Phoenix, Seattle, and Los Angeles, over the period 2008-2012.  They found their index could account forrnhetroskedastic errors (i.e. homes with a longer interval between sales shouldrnbe downweighted because the likelihood of unobserved changes to house qualityrnare higher).  It could also account forrnvalue weighting – that more valuable homes comprise a larger share of a realrnestate portfolio and thus their appreciation/depreciation rates should be givenrnmore weight.  The index also accuratelyrnforecasts the Case-Shiller index several months in advance, outperformsrnforecasting models that do not use listings data, and for the one metropolitanrnarea in which data on futures contracts are available, outperforms the market’srnexpectations as inferred from prices on Case-Shiller future contracts. </p

Thernsecond set of data was micro data on home listings with dates from which can bernderived the TOM.  There is no data tornindicate the reason for delisting or whether, if it were delisted because of arnsale, no information on the terms of that sale. rnThe listing also includes the specific property address and somerninformation on the home’s characteristics.  rnEach home was linked by address to its previous sales record.</p

All three cities in the sample experienced significant declinesrnin house pricesrnduring the beginning of the samplernperiod, although the magnitudernof the decline varied considerably across cities   The sample period also covers time duringrnwhich the homebuyer tax credit was in effect and the 2012 beginning of pricernrecovery in the cities, all three of which are covered by the Case-Shillerrn20-City Index.  The authors identifiedrnnearly a million properties that were delisted during the sample period andrnwhich they could link to a previous transaction record. A majorityrnof listings are delisted without a list price change. The median TOM is between one and two months.rnMany delistings are relisted soon after delisting: 20 percentrnof delistings are relisted within less than a month and 17 percent of are relistedrnbetween 2 and 6 months later. Many of these relistings may be due to sales agreements that fall throughrnbecause a mortgagerncontingency fails or an inspection fails.</p

The authors derivedrntwo index models.  In the first, whichrnthey called the simple-list price index they used the same regression equationrnas Case-Shiller except for the months where that index was not yet availablernand they substituted sales prices with the final list prices of delistings thatrnwere expected to close in a month.  Forrnthe previous sale, they used the house price level calculated from the transaction data alone rather than re-estimating it using both transactions and listings data.rn</p

They foundrnthat, despite the extreme changes in housing market conditions over the samplernperiod the sale-to-list price ratio fluctuated within a band of only several percentrnbut that variation does appear to be correlated with the house price cycle;rnperiods of rising prices tend to have high sales-to-list price ratios.  Another potential source of bias was therninclusion of all delistings rather than just those that led to sale.  Delistings that led to sales tended to havernlower list prices and the magnitude of that price difference was negatively correlated with the house price cycle. This share is also volatilernover time, with hotter markets being associated with a higher probability of sale, suggesting that including all delistings, rather than only the ones that result in sales, will bias the index duernto selection. </p

On average,rnthere is a delay of about six weeks between delisting and closing. The distribution of delays does not change much over time. This suggests that the assumption of a time-invariant distribution seems very reasonable, especially since the index isrncalculated as a movingrnaverage of the previous three months.</p

The second model, the adjusted listrnprice index, attempted to eliminate problems with including all delistings by deliveringrnpredictions for how outcomes would vary by applying observable listingrnvariables such as time on the market and the list price history. That modelrnattempts to describe the behaviorrnof a homeowner trying to sellrnher house. It had to take into accountrnvarious factors that might influence the outcome such as the value the sellersrnplace on not selling and staying in the home which may arise from factors suchrnas employment opportunities or changes in the family’s social or financialrnsituation; time constraints such as the start of the school year or a closingrndate on a trade-up home purchase.</p

The authors considered the ability ofrneach index to forecast the Case-Shiller HPI at various time horizons – i.e. thernnumber of weeks from the date of the last observed listings data until the endrnof the month they were trying to forecast.  At longer horizons an increasing share of thernsales are from properties which have not yet observed delistings.  However, even five months into the futurernthey found their index still had significant predictive power which occursrnbecome sore transactions take a significant amount of time to close and becausernthe smoothing process causes sales that close in a given month to affect thernindex for the two subsequent months as well. </p

The adjusted list price Indexrnperforms well, even at 12 weeks.  Notrnsurprisingly performance improves as more listings information about the monthrnbecomes available.  Even the SimplernList-Price Index performs well although the adjusted index delivers improved performancernof about 20 percent.</p

They authors stress that their samplernperiod covers one of the most volatile time periods in housing history andrnPhoenix, one of the most volatile sub-markets. rnThe fact that our index performs so well during this time period gives us confidence that performance would be as good, or possibly even better, out of sample.”

All Content Copyright © 2003 – 2009 Brown House Media, Inc. All Rights Reserved.nReproduction in any form without permission of MortgageNewsDaily.com is prohibited.