Nowcasting unemployment using mixed data sampling and Google Trends data

In this thesis I focus on the analysis and prediction of unemployment rates in the Czech Republic. During analysis of the labour market, a commonly encountered problem is that data describing the current state of the market are available with almost a year-long delay. For this reason, I analyse the possibility of using data from Google Trends for nowcasting current unemployment rates. Because Google Trends data are available at a higher frequency than unemployment data itself, I use MIDAS regression, which allows for such a connection without the need to aggregate the higher frequency data, therefore preventing the associated unwanted loss of potentially useful information. For the purpose of comparing the prediction quality, I create ARIMA and ARIMAX models, which however do not use data from Google Trends. After comparing the prediction models, I then determine the MIDAS model to be the most suitable.