Browsing by Author "Widgrén, Joona"

The internet is a popular channel for finding information. The search queries entered into a search engine contain a huge amount of data, but can it be used in economic forecasting? This thesis investigates if Google searches observe the changes in the Finnish housing market. The focus is this thesis is in housing price and home sales forecasting. Google search data is collected from Google Trends. Google Trends provides data describing the popularity of search queries. Google Trends data is updated every day and thus its publishing frequency is much higher in comparison with the official housing market data. The difference in publishing frequency can help to predict changes in housing markets before the official data is revealed. To evaluate the usefulness of Google data a simple model is extended with the Google search index. The forecasting ability of the simple model and the model with Google searches are then compared. Both models are used to forecast the current values of housing market indicators as well as forecasting near-future values. Furthermore, the Granger causality test is employed to investigate if Google searches are useful in forecasting housing market variables. The robustness of the results is studied using the fixed effects model. Also, housing price changes are forecasted as a robustness check. The results suggest that Google searches are useful in forecasting the Finnish housing market. Adding Google searches to a simple housing price forecasting model improves the accuracy of the contemporaneous forecast by 7.5 percent on average. Google searches improve contemporaneous home sales forecast by 15.9 percent on average. Also, the Granger causality test suggests that Google searches are useful in forecasting home sales. The findings are not as clear for Granger causality between Google searches and housing prices. The Granger causality test results suggest that Google searches could be useful in forecasting the current housing prices but not future values. The results also suggest that Google searches improve the near-future forecasts of both indicators.