Daily Online Search Volume As A Timely Measure For Investor Attention

This paper studies if online searches for stock tickers can predict daily abnormal returns. Prior research finds that the number of searches for a stock ticker on Google is a proxy for investor attention and can predict weekly abnormal returns. In a sample from 2005 to 2010 of S&P500 constituents I make one-day-ahead forecasts using Google Search Volume (GSV) and benchmark models. Based on a Diebold-Mariano test and conditional test of predictive ability (Giacomini and White (2006)) I find that the GSV model significantly outperforms an AR(1) model. This holds for both in- and out-of-sample and for different estimation windows. However, an AR(1) model does not improve when GSV is added. I conclude that GSV has some power to predict abnormal returns, however only beats the worst performing benchmark model. This is evidence that GSV is less successful in predicting abnormal returns on a daily instead of weekly basis. This is in line with the notion that daily stock returns are notoriously hard to predict.