Nonlinear Forecasting of Exchange Rate Volatility Using Google Search

Abstract

The purpose of this research is to widen previous studies by looking at the relationship between the Google Search Volume Index: GSVI and foreign currency rates volatility. We also test whether Google search queries can provide effective and efficient ways to predict exchange rates. Firstly, we compare the effective prediction of the traditional model and the new model using GAM method, which is allowed both linear and non-linear relationship in the model. Finally, we employ Granger Causality test to test if GSVI is causal to exchange rate volatility. The results show that the GAM model is more efficient than the classical linear model and it has significantly increased the forecasting potential. However, the intention of adding investor attention variables is insufficient to prove that Google trend factors can improve forecast accuracy.