Work detail

Google Econometrics: Predicting Bond Prices

Author: Mgr. Marek Krečmer
Year: 2018 - summer
Leaders: prof. Roman Horváth Ph.D.
Consultants:
Work type: Finance, Financial Markets and Banking
Masters
Language: English
Pages: 107
Awards and prizes:
Link: https://is.cuni.cz/webapps/zzp/detail/185543/
Abstract: The thesis analysed whether it is possible to improve on time-series forecasting
models used to predict prices and volatility of government bonds by adding
online search data. Previous research showed that Google trends data are an
useful source of an additional information which could improve various forecasting
or nowcasting models. Our research expanded the area into government
bonds and tested if the Google trends data could be of any use on this kind of
data as well.
We have analysed most of the government bond tenors of all the main
English speaking countries and the Czech Republic and focused on one-dayahead
forecasting of yields and weighted volatility. To forecast the next day
values, we have set up ARIMA-GARCH, GARCH(1,1), AR(1), mean, median
and lagged values and compared their performance with the realized values. In
addition, we have set-up augmented versions of ARIMA-GARCH, GARCH(1,1)
and AR(1) that included online search data. The subsequent findings can be
sometimes inconclusive but we have observed quite significant improvements
for some of the models and tenors of United States, United Kingdom and
Australian government bonds.
We have arrived at the conclusion that Google trends data could be used
to improve some of the models. It is also possible that the usability depends
on some kind of a minimal number of searches that is higher than the minimal
number used by Google to make its’ index greater than zero.

27

November

November 2022
MonTueWedThuFriSatSun
 123456
78910

111213
14151617181920
21222324252627
282930    

Partners

Deloitte

Sponsors

CRIF
McKinsey
Patria Finance
Česká Spořitelna
EY