Title page
Contents
Foreword 2
1. Background 5
2. Measuring economic activity with Google Trends data 6
3. Extracting information from Google Trends data 7
Description of Google Trends data 7
Real time data vs non-real time data 7
Sampling noise 7
Downward trend 11
Seasonality 14
4. Computing ICT growth rates 15
5. Choosing the best model 17
6. Choosing hyperparameters 20
7. Standard errors 23
8. Brief overview of the nowcasting results 24
9. Concluding remarks 30
References 31
Endnotes 33
Table 3.1. Comparing sample variance and the variance of sample averages 8
Table 5.1. Comparing different statistical methods 17
Table 5.2. Comparison of different machine learning methods 18
Table 6.1. RMSE for models with one hidden layer 21
Table 6.2. RMSE for models with two hidden layers 22
Table 6.3. RMSE for different activation functions 22
Figure 3.1. Sampling noise distribution before and after correction 9
Figure 3.2. Average search index distribution before and after correction 10
Figure 3.3. Search index for the category "statistics" in Austria 11
Figure 3.4. Search index for the "statistics" category in Austria decomposed using a Hodrick-Prescott filter 12
Figure 3.5. Search index for the statistics category in Austria decomposed using a Lowess smoother 12
Figure 3.6. Search index for the "statistics" category in Austria decomposed using fixed effects 13
Figure 3.7. Search index for the "statistics" category in Austria: Monthly versus yearly average 14
Figure 4.1. Average observed ICT growth rate by year in OECD countries 15
Figure 8.1. Observed and predicted ICT sector growth rates, 2011-23 25