Forecasting Housing Prices under Different Submarket Assumptions

This research evaluated forecasting accuracy of hedonic price models based on a number of different submarket assumptions. Using home sale data for the City of Knoxville and vicinities merged with geographic information, we found that forecasting housing prices with submarkets defined using expert knowledge and by school district and combining information conveyed in different modeling strategies are more accurate and efficient than models that are spatially aggregated, or with submarkets defined by statistical clustering techniques. This finding provided useful implications for housing price prediction in an urban setting and surrounding areas in that forecasting models based on expert knowledge of market structure or public school quality and simple model combining techniques may outperform the models using more sophisticated statistical techniques.