The Predictive Power of Google Searches in Forecasting US Unemployment

We suggest the use of an index of Google job-search intensity as the best leading indicator for the US unemployment rate. After selecting the best model in sample at each forecast origin, we perform a deep out-of-sample forecasting comparison among models that adopt such a leading indicator, the more standard initial claims or alternative indicators based on consumers’ and employers’ surveys. Models augmented with Google data outperform traditional ones, with their relative performance improving with the forecast horizon. Such models provide more accurate forecasts compared to the Survey of Professional Forecasters, models based on labor force flows and nonlinear models. Google-based models predict particularly well at the turning point taking place at the start of the Great Recession, while their relative predictive ability stabilizes afterwards.