Statistical relationship between internet data indicators and the Brazilian stock exchange

This work presents a statistical analysis between indicators obtained from internet data in Portuguese - news sentiment and Google Trends - and data on the Brazilian stock market through Spearman's rank correlation coefficient. The methodology used to collect, pre-process and obtain each indicator is detailed. Data from the years 2019 to 2021 were obtained. For the sentiment analysis of the news, a CNN model (Convolutional Neural Network) was adopted, which obtained an F1-score of 96%. As a result, some interesting correlations were obtained, among which, an inverse correlation characterized as “moderate” (according to the Cohen scale) between news sentiment and adjusted closing price in 2019; between search volume and closing price, a negative and “very large” and a positive and “large” correlation between trade volume and search volume. In both 2020 and 2021, negative coefficients defined as “large” were found, taking into account the closing price and trading volume.

I. Introdução

With the ease of internet access provided by a time when technology advances more and more rapidly, more than 5 billion users¹

https://www.internetlivestats.com/internet-users/. Access date: July 28, 2022

generate progressively larger amounts of data through various sources such as searches in search engines, social networks, applications, among others. Therefore, both companies and government agencies and individuals have awakened to the importance of using this information in a strategic way.