Forecasting stock market movements using Google Trend searches

In the following paper, we seek to evaluate the predictive capabilities of internet search data. While past studies have proposed using Google Trends as an effective proxy for investor attention, we re-evaluate this idea in the context of a Granger causal framework. We apply the Kaplan-Meier estimator to quantify the level of persistence in lagged correlations between the search volume series and the directional movements in the S&P 500. We find that the directional movement of the S&P 500 from changes in the search volume series is dependent on the specific term being searched for, and by extension, the sentiment of the term itself. We hypothesize that while Google Trends is a valid measure of investor attention, the signals derived from changes in search volume is conditional upon the sentiment inherent to the search terms. Using the terms that are persistently found to be Granger causal with the index, we propose several generalized linear models for forecasting the probability of positive or negative directional movements, and propose a trade strategy from the generated forecasts, resulting in a 40% outperformance of a traditional buy-and-hold strategy in our testing period.