Social media index as proxy of the Brazilian consumer confidence index

Abstract

In this paper we formalize a Social Media Index (SMI) based on sentiment analysis performed on Twitter messages publicly available in the Portuguese language. This (possibly high-frequency) index shows the ability to detect events that are relevant to the social discord. Changes in the SMI are compared with changes with the monthly-aggregated Brazilian Consumer Con dence Index (CCI) in a 84-month window. The correlation r between the two indexes shows great dependence on the choice of lexicon used to evaluate sentiment (0.59 < r < 0.75). Granger-causality tests show that it is more likely that changes in social media sentiment precedes changes in Consumer Con dence than vice-versa. Furthermore, post-selection of Twitter messages based on selected keywords has the power to improve the association between SMI and CCI up to r = 0.81. On the one hand, our ndings suggest that the SMI and CCI are in uenced by the same set of emotions. On the other hand, the SMI has the ability to be published beforehand and with rather higher frequency.