Forecasting economic variables using google trends

This thesis examines the ability of Google Trends (GT), i.e. the free public tool for obtaining data regarding web search activities, in forecasting economic variables of Turkey by using different query selection methods. It reports whether internet search activity improves forecasting of tourism demand, unemployment, and car sales by comparing forecast errors of models with and without GT variable. For each variable of interest, three models are constructed. The first, baseline model, is estimated using only the past values of selected variables. The second model, in addition to the past values of the economic variables, integrates additional regressor constructed from the simple query selection method, where search indices of single keywords related to each of the selected variables are obtained. The third model follows the same approach as the second, but with different query selection method, where composite search index using Principal Component Analysis (PCA) is constructed from the large number of queries related to our economic variables. All models are estimated using the Autoregressive Integrated Moving Average (ARIMA) methodology. Forecast comparisons indicate that models with GT information outperformed the basline models in most of the forecasting experiments. When it comes to performance of the two query selection methods, composite search index, on average, provides better forecasts. The study's results could be beneficial for the policy makers and other stakeholders as selected variables play important role for the Turkish economy. GT offers a new way of tracking economic behavior at almost zero cost. Furthermore, they have ability to get real-time insights regarding economic decisions.