Google Insights For Search Investor Sentiment And Stock Market Returns

The (time-series) frequencies of Google search terms that are possibly related to investor sentiment were transformed into a summary measure, using principal component analysis. This variable was then used in a vector autoregression framework in order to evaluate its in-sample forecasting power and to test for Granger causality between sentiment and stock market returns. In pseudo out-of-sample forecasting experiments, the summary measure's "real-time" performance was compared to an autoregressive model and to a random walk. While I found that investor sentiment Granger causes stock market returns and explains the time-series of stock market returns better than past returns alone, sentiment-based forecast models were not able to consistently beat the two benchmark models out-of-sample.