Forecasting housing investment

This study applies a model averaging approach to conditionally forecast housing investment in the largest euro area countries and the euro area. To account for substantial modelling uncertainty, it estimates many vector error correction models (VECMs) using a wide set of short and long-run determinants and selects the most promising specifications based on in-sample and out-of-sample criteria. Our results highlight marked cross-country heterogeneity in the key drivers of housing investment which calls for country-specific housing market policies. A pseudo out-of-sample forecast exercise shows that our model averaging approach beats a battery of ambitious benchmark models, including BVARs, FAVARs, LASSO and Ridge regressions. This suggests that there is ample scope for model averaging tools in forecast exercises, notably as they also help to reduce model uncertainty and can be used to assess forecast uncertainty.