NOWCASTING THE SWEDISH UNEMPLOYMENT RATE USING GOOGLE SEARCH DATA

Abstract: In this thesis, the usefulness of search engine data to nowcast the unemployment rate of Sweden is evaluated. Four different indices from Google Trends based on keywords related  to unemployment are used in the analysis and six different regARIMA models are  estimated and evaluated. The results indicate that the fit is improved for models when data  from Google Trends is included. To evaluate the nowcast ability of models, one-step-ahead  predictions are calculated. Although the prediction error is lower for the models with data  from Google Trends, Diebold-Mariano tests do not indicate that the predictions are  significantly better compared topredictions from a model without data from Google Trends.  It is therefore concluded that one cannot state that data from Google Trends improves  nowcasts of the unemployment rate of Sweden. Additionally, predictions are calculated for  longer forecast horizons. This analysis indicates that Google search data could be useful to  forecast the unemployment rate of longerforecast horizons.