Heteroscedasticity in hedonic house price models

Abstract

This paper extends the work of Goodman and Thibodeau which examines heteroscedasticity in hedonic house price models. It investigates whether heteroscedasticity is related to other factors as well as age, since correcting for only one factor when several are involved can worsen the estimates. A detailed specification is used for the hedonic modelling since omitted variables can give the appearance of heteroscedasticity. The paper uses data from over 1400 sales in Stoke-on-Trent in 1994 and finds heteroscedasticity related to both age and the external area of the property. It is shown that this was not due to outliers in the data. Use of estimated generalized least squares is shown to remove the heteroscedasticity. The estimates provided by this technique are shown to give forecast errors that may have a smaller standard deviation than ordinary least squares estimates.