Spatially Varying Regression Over Irregularly Shaped Regions: Application to a Hedonic House Price Model

Arnab Bhattacharjee (Heriot-Watt University, UK)

In many applications interest lies in spatially varying regression slopes. This paper considers a hedonic house price model for the Aveiro urban housing market in Portugal a coastal area divided by lagoons and rivers, and with natural holes and irregular boundaries. Important policy and business questions relate to spatially varying implicit price of living space and predicted house prices, which are our main objects of inference. We develop a new spatial regression techniques based on two dimensional local polynomials that are suitable for irregular domains and computationally feasible in large spatial dimensions. Our model incorporates spatial variation in slopes and spatial dependence in regression errors, as well as measurement errors. Distance decay in spatial autocorrelation is unknown up to finite parameters. Application to the Aveiro housing market in Portugal provides exciting new inferences on the value of living space and price predictions.