Forecasting U.S. Unemployment Using Google Trends

Abstract: Thesis aims to analyze whether Google Trends data can be used as a leading predictor to forecast U.S. unemployment rate. To test this I selected benchmark ARIMA models based on Box-Jenkins (1976) methodology and in-sample performance (measured by information criterion). Augmented these models by adding explanatory exogenous variables: Initial jobless claims (IC) and Google Trends Index (GI). Then created out-of-sample forecasts and evaluated whether models including GI outperformed benchmark models. Performed several robustness checks to ensure validity of the results. Models which included Google Trends data, indeed, outperformed benchmark models. Although improvements are rather modest. The "best" model included both IC and GI showing that both are useful leading indicators and contain information which does not fully overlap.