Nowcasting Swedish Private Consumption With Google Search Data

by Lauri Tuomisto

Institution: University of Helsinki
Year: 2015
Keywords: Taloustiede
Record ID: 1145249
Full text PDF: http://hdl.handle.net/10138/153285


Nowcasting is generally defined as the prediction of the present. Nowcasting is required in economics as many economic variables have a publishing lag and the publishing frequency of the variables can be lower than desired. Therefore it is often necessary to estimate the current values of economic variables through nowcasting models. The aim of the thesis is to find out if Swedish private consumption can be nowcasted more accurately with models that include Google search data as explanatory variables. Traditionally private consumption is nowcasted with consumer confidence index and in the thesis models based on consumer confidence index are compared to models based on Google search data. Google search data for the thesis is downloaded from Google Trends, which is a service provided by Google Inc. through which the public can access the search word statistics of each country. The Google search data is available on a weekly frequency, which in this thesis is aggregated into monthly frequency. The Swedish private consumption data for the study is downloaded from OECD Statistics. The private consumption data from OECD is available on a quarterly frequency, which is disaggregated into monthly frequency using the Chow-Lin procedure and Swedish retail trade data as an indicator series. The Swedish retail trade data is downloaded from the Swedish Statistics and the Swedish consumer confidence index from the National Institute of Economic Research. All the time series used in the nowcasting models are on a monthly frequency and cover a period from January 2005 to June 2014. The main methods used in the thesis include principal component analysis, which is applied for the Google search data and dynamic linear regression based on which the nowcasting models are formed. Two types of nowcasting models for the private consumption are formed in the thesis. Google Trends service uses an algorithm to categorize the search words automatically based on the subject and the first type of the models uses these automated search categories as explanatory variables in the nowcasting model. The second type uses the most popular search words as explanatory variables. Only search categories and search words related to private consumption are included in the models. The models based on search words are found to be more accurate compared to the models based on search categories. However, the model based on consumer confidence index is still found to produce more accurate nowcasts compare to the model including Google search words. In an in-sample nowcast the model including both Google search words and consumer confidence index as explanatory variables slightly outperforms the model including only consumer confidence index. The performance of the nowcasting model including both Google search words and consumer confidence as explanatory variables is then tested in an out-of-sample nowcast in which case the model performs worse compared to the model including only consumer confidence index. Based on the results of the thesis it can be concluded…