Forecasting housing starts

This paper develops a multi-equation econometric model of the U.S. housing sector and compares its forecasting performance to that of a time-series (ARIMA) model. The essential issue explored is wether a faithful modeling of relationships suggested by economic theory constitutes a real advantage in forecasting. The econometric model has a stock-flow structure with cost and credit availability variables emphasized in the housing supply equation. We conclude that unaided one-period ahead econometric forecasts are superior to the time-series forecasts. But, due to bias in the former, prediction by the ARIMA model turns out to be better for longer period forecasts. This disadvantage in the econometric model can be overcome by judgmental adjustments while the same is not necessarily true for the ARIMA model.