The usefulness of Google Trends data in macroeconomic forecasting: evidence from Finland

No Thumbnail Available

URL

Journal Title

Journal ISSN

Volume Title

School of Business | Master's thesis

Date

2023

Major/Subject

Mcode

Degree programme

Economics

Language

en

Pages

61 + 9

Series

Abstract

This thesis looks into Google Trends data, its special features and uses in macroeconomic forecasting. Evidence is gathered from the literature to assess whether including Google Trends data in forecasting models improves their predictive accuracy. Google Trends data has many particularities, but nevertheless all papers included in the analysis find evidence in favor of Google Trends data being at least somewhat useful. Due to the differing modeling choices and sometimes conflicting results, it is not possible to draw exact conclusions on the conditions under which Google Trends data is useful. The hypothesis of the usefulness of Google Trends data is also tested empirically in Finnish context by estimating a dynamic factor model and by comparing the out-of-sample forecasting performance of models with and without Google Trends data. Inclusion of Google Trends data is found to not improve the forecasting performance significantly. Especially the poor nowcasting performance of the Google Trends data augmented models is surprising considering the previous literature. A robustness check suggests that the nowcasting ability of the tested models depend on the phase of the economic cycle, the models perform better during economic expansion and recovery and considerably worse during recessions and economic slowdowns. The nowcasting ability of the model is also found to increase the further in the quarter the nowcast is produced. It is not conclusive which search term selection method is optimal

Description

Thesis advisor

Kitti, Mitri

Keywords

forecasting, nowcasting, Google Trends, dynamic factor model

Other note

Citation