Google Trend Index as an Investor Sentiment Proxy in Cryptomarket: Nonlinear Relationships With Cryptomarket and Predicting Bitcoin Returns With Machine Learning Approach

Abstract

This study investigates the utility of Google trend indices as proxies of investor sentiment, examining their relationships with cryptocurrency market prices and their potential for return prediction. Employing several nonlinear econometric models including the momentum threshold autoregressive AR (MTAR), Kapetanios, Shin, and Snell, and exponential smooth transition autoregressive vector error correction model, the research the relationships between Google trend indices and BTC prices. Additionally, the study evaluates the performance of three developed artificial neural network models in predicting bitcoin returns based on investor sentiment derived from Google trend indices. The findings highlight that the MTAR model effectively captures significant relationships between the variables studied. However, predicting bitcoin returns remains challenging due to their typically small values, which represent the changes between observation points.