The MARS Algorithm in the Spatial Framework: Non-Linearities and Spatial Effects in Hedonic Models

Abstract

Multivariate Adaptive Regression Spline (MARS) is a simple and powerful non-parametric technique that automatizes the selection of non-linear terms in regression models. Non-linearities and spatial effects are natural characteristics in numerous spatial hedonic pricing models. In this paper, we propose using the MARS data-driven methodology combined with the Instrumental Variables method in order to account for potential non-linearities and spatial effects in hedonic models. Using a large data set of more than 6,000 dwellings in Hamburg and about 17,000 in St. Petersburg, we confirm the presence of both effects (non-linearities and spatial autocorrelation). The results also show that there is a non-linear effect of the prices of neighboring houses on the price of each house. High prices for neighboring houses have a lower impact on the house price than low prices of neighboring houses. Finally, an extensive Monte Carlo exercise evaluates the ability of MARS to incorporate the correct spatial spillover terms in spatial regression models simultaneously including at same time non-linear effects.