Knitting Multi-Annual High-Frequency Google Trends to Predict Inflation and Consumption.

Russia's Ruble during the onset of the Russian invasion of Ukraine in early 2022: The role of implied volatility and attention

“…First, as argued by Chan (2020); Bleher and Dimpfl (2021), in one batch, Google Trends allows one to retrieve a maximum of 270 consecutive observations that are scaled from 1 to 100. Concatenating these series into one group (which is unfortunately often used) would lead to jumps in the break-points because of the different scales of each batch, see Bleher and Dimpfl (2021).…”

“…First, as argued by Chan (2020); Bleher and Dimpfl (2021), in one batch, Google Trends allows one to retrieve a maximum of 270 consecutive observations that are scaled from 1 to 100. Concatenating these series into one group (which is unfortunately often used) would lead to jumps in the break-points because of the different scales of each batch, see Bleher and Dimpfl (2021). A better approach might be to employ logarithmic differences before concatenation; however, Bleher and Dimpfl (2021) shows that this approach can also lead to changes in the underlying distributional properties.…”

“…Concatenating these series into one group (which is unfortunately often used) would lead to jumps in the break-points because of the different scales of each batch, see Bleher and Dimpfl (2021). A better approach might be to employ logarithmic differences before concatenation; however, Bleher and Dimpfl (2021) shows that this approach can also lead to changes in the underlying distributional properties. Bleher and Dimpfl (2021) recommends using a regression-based procedure, and another viable alternative might be the procedure used by Chronopoulos et al (2018).…”

“…Bleher and Dimpfl (2021) recommends using a regression-based procedure, and another viable alternative might be the procedure used by Chronopoulos et al (2018). Our approach is more similar to that of Bleher and Dimpfl (2021) and even more so to that of Kristoufek (2015) as we do not use a regression. The similarity lies in the fact that for each term, we download search volume indices for each trading window 7 t in a batch covering 4−days, where consecutive 4−day batches overlap across 1-day, i.e.…”

“…Attention can be estimated from different sources, like news articles, Wikipedia page reads, Twitter, or Google search volume indices. The latter reference is considered to be a good proxy for information demand Bleher and Dimpfl (2021), as Google searches are highly popular in many parts of the world. In this study, we also rely on Google searches that are well-established in the finance literature (e.g.…”