Network Analysis of Search Dynamics: The Case of Stock Habitats

There is an increasing attention in information systems to be able to predict outcomes using search frequency on popular portals. This growing literature focuses on revealing demand patterns of individual assets (e.g., products, stocks, services). However, users typically search many different assets together (e.g., correlated searches) and leave a digital footprint, which can help provide insights on the behaviors of a group of assets. Furthermore, such group behavior can be used to predict outcomes (e.g., demand, stock returns) in the future. We analyze the underlying behavior of distinct subnetworks formed by correlated user searches for multiple items in the stock market and use such information for return prediction. Using cosearch data for stocks from Yahoo! Finance, we find 50–79 search clusters representing 230–349 stocks among Russell 3000 stocks at different time periods. These clusters reveal interesting habitats where the returns of stocks within a cluster tend to comove after controlling for known determinants of comovement. When a stock enters (departs) a cluster, the focal stock return comoves (detaches) with the cluster returns. Thus, search cluster–based habitats reveal aggregate investment preferences and are more granular than fundamental-based habitats. We find that search-based habitats can also improve return predictability of related stocks. This paper was accepted by Chris Forman, information systems .