Selecting sensitive web info via conditional probabilities to model economics and financial variables

Abstract

In this paper, we propose a methodology to identify relationships between web data and social/economic variables, such as inflation. Our method enables the selection of relevant time series from a large data sample by employing a criterion based on a few hypotheses regarding their dynamics. Specifically, we examine the correlation between web activities and the dynamics of two macroeconomic variables: the unemployment rate and US automotive sales. We demonstrate how changes in the search volume of specific keywords, as measured by corresponding Google Trends data, are reflected in the underlying dynamics of these variables. The findings presented in this paper, along with the versatility of our approach, suggest the potential extension of this study to other economic variables.