Forecasting the US Real Private Residential Fixed Investment Using Large Number of Predictors

This paper employs classical bivariate, factor augmented (FA), slab and spike variable selection (SSVS)-based, and Bayesian semiparametric shrinkage (BSS)-based predictive regression models to forecast the US real private residential fixed investment series over an out of sample period of 1983Q1 to 2011Q2, based on an in-sample of 1963Q1-1982Q4. Both large-scale (with 188 macroeconomic series) and small-scale (20 macroeconomic series) FA, SSVS and BSS predictive regressions, besides 20 bivariate regression models, are used in order to capture the influence of fundamentals in forecasting residential investment. We evaluate the ex post out-of-sample forecast performance of the 26 models using the relative average Mean Square Error for one-, two-, four- and eight-quarters-ahead forecasts and test their significance based on the McCracken (2004, 2007) MSE-F statistic. We find that, on average, the SSVS-Large model is the best amongst all the models. We also find that one of the individual regression models (based on house for sale as a predictor, H4SALE) performed best at the four- and eight-quarters-ahead horizons. Finally, we use these two models to predict the relevant turning points of the residential investment, via an ex ante forecast exercise from 2011Q3 to 2012Q4. The SSVS-Large model forecasts the turning points more accurately, though the H4SALE model did better towards the end of the sample. Our results suggest that it is best to consider economy-wide factors, in addition to specific housing market variables, when evaluating the real estate market.