"Just Google it": Forecasting Norwegian unemployment figures with web queries

Abstract

This thesis explores whether online search queries, represented by Google search queries, contain information useful in forecasting short term unemployment figures in Norway or not. Based on earlier work utilizing online web queries this should be possible, even in small countries. Looking at job search theory supplied with intuition, words from the Norwegian Welfare Administration (NAV) and counseling from the Language Council of Norway we create four Google Indicators that we add to baseline models to check if this reduces the forecasting error (RMSE) of the models. Our findings supports our hypothesis, that Google search contain information useful when predicting short term changes in unemployment. Our top performing model improves the forecasting accuracy compared to its baseline model by 18.3% on average over twelve months. Our best models also outperform the leading indicator “published job advertisements”. These are remarkable results given the noise in our data.