The Blessing of Dimensionality in Forecasting Real House Price Growth in the Nine Census Divisions of the U.S.

Abstract

This paper analyzes whether a wealth of information contained in 126 monthly seriesused by large-scale Bayesian Vector Autoregressive (LBVAR) models, as well as FactorAugmented Vector Autoregressive (FAVAR) models, either Bayesian or classical, can proveto be more useful in forecasting the real house price growth rate of the nine censusdivisions of the United States, compared to the small-scale VAR models, that merely usethe house prices. Using the period of 1991:02 to 2000:12 as the in-sample period and2001:01 to 2005:06 as the out-of-sample horizon, this study compares the forecastperformance of the alternative models for one-to-twelve months ahead forecasts. Basedon the average Root Mean Squared Error (RMSEs) for one-to-twelve months aheadforecasts, the findings reveal that the alternative FAVAR models outperform the othermodels in eight of the nine census divisions.