A spatial hedonic model application of variance function regression to residential property prices in Beijing

Abstract

Variance function regression incorporates a novel method of residual analysis that should be of interest in spatial modeling. The method is a two part regression: one for the conditional mean, which is a standard regression, and one for the conditional variance, which is estimated from the residuals of the initial regression. The method is briefly described and then applied hedonic models of residential real estate prices in Beijing, using both a metropolitan area sample and a smaller sample of the district of Chaoyang. As expected the results generally indicate that distance to Tiananmen Square is a powerful predictor of residential real estate prices in Beijing. That distance effect is fully encompassed in the mean structure of a regression model. However, other variables, such as the age of an apartment block have price effects that are revealed fully only when the conditional variance is modeled as well.