Forecasting House Prices in Italy

Forecasting house prices is a difficult task given the strong relationship between real estate markets, economic activity and financial stability, but it is an important one. This paper evaluates the out-of-sample forecasting performance of various models of house prices in a quasi-real time setting. Focusing on Italy, we consider two structural models (using simultaneous equations) and a Bayesian VAR and compute both conditional and unconditional forecasts. We find that the models perform better than a simple autoregressive benchmark; however, the relative forecast accuracy depends on the forecast horizon and also changes over time. For the full sample period the simultaneous equation model, which takes into account credit supply restrictions and real estate taxation, shows the best performance measured in terms of root mean squared forecasting error (RMSFE). In the first part of the sample (2005-2010), medium-term forecasts of house prices greatly benefit from conditioning on the evolution of households' disposable income, whereas from 2010 onwards the path of the stock of mortgages becomes important