Evaluating Google Trends data to the task of predicting stock returns

• Ano de defesa: 2021
• Autor(a) principal: Oliveira, Daniel Cunha
• Orientador(a): Pereira, Pedro L. Valls, Fujita, Andre
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Palavras-chave em Português: Precificação de ativos; Previsão; Aprendizado de máquina; Séries temporais
• Palavras-chave em Inglês: Google Trends
• Link de acesso: https://hdl.handle.net/10438/31061

The problem of predicting financial assets returns is one of the main problems of the empirical finance literature. In particular one of it’s main challenges is to evaluate the usefulness of the so called alternative data to this task. One of the most common alternative datasets is Google Trends data which have gained popularity in recent years. In this work we want to evaluate the usefulness of this data to the task of predicting U.S. stock indices returns. To achieve this goal we break up the problem in two steps: first we employ feature selection methods, and second we employ forecasting models. We use 15 feature selection methods and 10 forecasting models to achieve this goal. In contrast to what the literature have found, we do not found evidence that the Google Trends data contributes to predict the returns of the stock indices in question. The conclusions seems to be robust across feature selection methods, forecasting models, accuracy and risk and return metrics.