In Search Of Alpha-Trading On Limited Investor Attention

H/T AbnormalReturns

Konstantin Storms

WHU – Otto Beisheim School of Management

Julia Kapraun

WHU – Otto Beisheim School of Management

Markus Rudolf

WHU Otto Beisheim Graduate School of Management

November 3, 2015

Abstract:

In this study we develop a trading strategy that exploits limited investor attention. Trading signals for US S&P 500 stocks are derived from Google Search Volume data, taking a long position if investor attention for the corresponding security was abnormally low in the past week. Our strategy generates 19% average annual return and thereby outperforms a simple market buy-and-hold strategy. After controlling for the well-known risk factors, a significant alpha (abnormal return) of 10% p.a. remains. Returns are sufficiently large to cover transaction costs.

In Search Of Alpha-Trading On Limited Investor Attention

Researchers as well as investment professionals are continuously searching for promising trading strategies. While academia is primarily interested in understanding the dynamics of capital markets and the behavior of its acting investors, investment professionals like hedge fund managers rather yearn for abnormal returns and desire to outperform relevant benchmarks.

Goals of trading strategies are multifarious. While some simply aim at diminishing portfolio risk, e.g. by shifting into certain asset classes at the occurrence of predefined events, others seek to better diversify an existing portfolio by adding securities or varying portfolio weights according to certain signals. Some other strategies advance more aggressively and actively select under-priced stocks.

There are almost countless sources of trading signals. Popular examples include past returns, which are applied in momentum trading (e.g. Jegadeesh and Titman, 1993; Chan et al., 1996) or contrarian strategies (e.g. De Bondt and Thaler, 1985, 1987).1 Also market news or investor sentiment are proven sources of trading signals (e.g. Uhl et al., 2015; Da et al., 2015; Tetlock, 2007). And truly extraordinary triggers like Super Bowl results (Krueger and Kennedy, 1990) or weather forecasts (Hirshleifer and Shumway, 2003) have been successfully applied in the never ending attempt to beat the market.

In this study we focus on a novel and by now little considered source of trading signals. The behavior of internet users, who conduct online searches for certain key words, contains valuable information and often has direct implications for real world developments. Fortunately, Google makes such behavior of its users publicly available and provides data when and how often certain terms are searched for. The power of this “Google Trends” data and its correlation with real world phenomena has been shown in many studies. For instance, Google Search Volume (GSV) data can be used to predict the diffusion of influenza in the US (Polgreen et al., 2008; Ginsberg et al., 2009) or to forecast car sales (Varian and Choi, 2009). For a financial application, Da et al. (2011) propose GSV as a reliable measure of retail investor attention. They highlight if users search certain stocks on the internet, they definitely pay attention to it, having implications for the return of those stocks. Building on these insights, Vozlyublennaia (2014) and further Storms et al. (2015) emphasize that low retail investor attention as measured by GSV is associated with less efficient asset prices. If investor attention is low, firm related information are not sufficiently incorporated into security prices. As a consequence, stocks experiencing low investor attention should be suitable for an active stock selection strategy, as they can be expected to be inefficiently priced.

Indeed, few studies successfully apply GSV in order to detect trading signals. Preis et al. (2013) suggest a GSV approach, for which search terms related to financial turmoil like “debt” or “risk” can be used as a warning indicator of declining stock markets. As the authors emphasize, their approach can be applied in a trading strategy in order to shift asset classes whenever Google searches warn of a crisis. The study, however, does not provide tools for active stock selection. Kristoufek (2013) develops a trading strategy for US Dow Jones securities based on GSV. Main focus of his study is yet on risk diversification. His strategy alternates portfolio weights of the 30 Dow Jones stocks depending on past attention. Weight is increased if attention was low in the last week, as risk is hypothesized to be lower for these stocks. But also this study does not give explicit guidance for stock picking and does not adjust for the known risk factors. Hence it lacks an assessment regarding the strategy’s ability to create abnormal return (alpha). Bank et al. (2011) develop an active stock selection strategy based on GSV for German stocks. However, the authors receive trading signals from “large or low signed changes” in Google Search Volume, and, therefore, do not explicitly focus on the low attention stocks. Also their search term approach using firm name is rather noisy when capturing investor attention.