Forecasting House Prices in the 50 States Using Dynamic Model Averaging and Dynamic Model Selection

Abstract

We examine house price forecastability across the 50 states using Dynamic Model Averaging and Dynamic Model Selection, which allow for model change and parameter shifts. By allowing the entire forecasting model to change over time and across locations, the forecasting accuracy improves substantially. The states in which housing markets have been the most volatile are the states where model change and parameter shifts have been the most needed.

Keywords: Forecasting housing markets, 50 states, Kalman filtering methods, Model change, Parameter shifts, Boom-bust cycle, Model averaging, Model selection

JEL Classification: C51, C52, G1, E3