Forecasting house prices using online search activity

We show that Google search activity is a strong out-of-sample predictor of future growth in U.S. house prices and that it strongly outperforms standard predictive models based on macroeconomic variables as well as autoregressive models. We extract the most important information from a large set of search terms related to di§erent phases of the home search process into a single Google-based factor and then use it to predict movements in future house prices. At the onemonth forecast horizon, the Google factor delivers an out-of-sample R2 -statistic of about 50% for the aggregate U.S. market over the period 2009-2018. We show that the strong predictive power of Google search activity holds for longer forecast horizons, for various house price indices, for seasonally unadjusted and adjusted data, and across individual U.S. states.

Download PDF