PREDICTING HOUSE PRICES USING MULTIPLE LISTINGS DATA

It is often necessary to accurately predict the price of a house betwe en sales. One method of predicting house values is to use data on the characteristics of the area's housing stock to estimate a hedonic regr ession, using ordinary least squares (OLS) as the statistical techniqu e. The coefficients of this regression are then used to produce the pr edicted house prices. However, this procedure ignores a potentially la rge source of information regarding house prices-the correlations exis ting between the prices of neighboring houses. The purpose of this art icle is to show how these correlations can be incorporated when estima ting regression coefficients and when predicting house prices. The pra ctical difficulties inherent in using a technique called kriging to pr edict house prices are discussed. The article concludes with an exampl e of the procedure using multiple listings data from Baltimore.