Title: Spatio-temporal diffusion of US house prices with unknown spatial weights
Authors: Achim Ahrens - Heriot-Watt University (United Kingdom) [presenting]
Abstract: The aim is to examine the relationship between house prices and their fundamental drivers, such as real per capita income, interest rates and inflation, using a panel data set of US states over the sample period 1976Q1-2014Q4. The issue of spatial, or cross-sectional, dependence is addressed using a novel estimation method based on the Lasso estimator, a well-established regularisation technique. The vast majority of the spatial econometric literature assumes that the spatial weights matrix, which determines interaction effects between units, is known and, in practice, spatial weights are typically specified on an ad hoc basis using observable distance measures. The proposed method allows for unknown interaction effects and employs the Lasso estimator to appropriately control for spatial effects under the identification assumption of sparsity. The methodology is of more general interests for large $T$ panels where spatial dependence is an issue or where the interest lies in modelling spatio-temporal diffusion processes. Estimation results show a long-run relationship between house prices and real per capita income. The most striking result, however, are strong and complex spatial effects that would not have been captured with standard specifications of the spatial weights matrix such as the binary contiguity matrix.