HOW WEB SEARCH ACTIVITY EXERT INFLUENCE ON STOCK TRADING ACROSS MARKET STATES?

Transcription

1 Association for Information Systems AIS Electronic Library (AISeL) PACIS 2014 Proceedings Pacific Asia Conference on Information Systems (PACIS) 2014 HOW WEB SEARCH ACTIVITY EXERT INFLUENCE ON STOCK TRADING ACROSS MARKET STATES? Tzu-Lun Huang National Sun Yat-sen University, Kuan-Ling Lai National Sun Yat-sen University, Miao-Ling Chen National Sun Yat-sen University, Follow this and additional works at: Recommended Citation Huang, Tzu-Lun; Lai, Kuan-Ling; and Chen, Miao-Ling, "HOW WEB SEARCH ACTIVITY EXERT INFLUENCE ON STOCK TRADING ACROSS MARKET STATES?" (2014). PACIS 2014 Proceedings This material is brought to you by the Pacific Asia Conference on Information Systems (PACIS) at AIS Electronic Library (AISeL). It has been accepted for inclusion in PACIS 2014 Proceedings by an authorized administrator of AIS Electronic Library (AISeL). For more information, please contact

2 HOW WEB SEARCH ACTIVITY EXERT INFLUENCE ON STOCK TRADING ACROSS MARKET STATES? Tzu-Lun Huang, Department of Finance, National Sun Yat-sen University, Kaohsiung, Taiwan, R.O.C., Kuan-Ling Lai, Department of Finance, National Sun Yat-sen University, Kaohsiung, Taiwan, R.O.C., Miao-Ling Chen, * Department of Finance, National Sun Yat-sen University, Kaohsiung, Taiwan, R.O.C., miaoling@mail.nsysu.edu.tw Abstract With the popularity of the network, the influence of web search activity on capital market has become a central issue in finance. Although a growing body of literature finds that the online search volume helps to explain stock trading activities, few studies take the impacts of market states into account. To fill up the gap, this study explores the influence of market states on the relationship between web search activity and capital market from the perspective of stock returns. The evidence shows that web search activity proxied by the weekly Google search volume drives stock returns in the regular and up market states, although investors exhibit the same behavior to search for information via the Internet across time. The results generally support the attention effect of online search activity on stock returns. However, the reluctance to buy stocks in the bearish market due to the fear of bearing substantial systematic risks may limit the attention effect and weaken the link between web search activity and stock returns. Keywords: Web search activity, Market state, Google search volume, Stock return, Investor attention 1. Introduction It is no doubt that people nowadays utilize the Internet to gather and process information before making decisions. In stock market, the importance and implication of investors web search activity for trading decision-making has been explored by finance and economic literatures. Specifically, studies employ the Google search volume data and find a significant influence of web search activity on stock investors trading behavior (Bank et al., 2011; Da et al., 2011; Vlastakis and Markellos, 2012; Vosen and Schmidt, 2012; Aouadi et al., 2013; Ding and Hou, 2013; Siganos, 2013; Zhang et al., 2013). In the pioneering study of Da et al. (2011), the Google search volume is proved to drive * Corresponding author: Miao-Ling Chen, Department of Finance, National Sun Yat-sen University, No.70, Lien-Hai, Rd, Kaohsiung City 80424, Taiwan (R.O.C.). Tel: Fax: address: miaoling@mail.nsysu.edu.tw

3 stock returns as Barber and Odean s (2008) attention theory predict. Barber and Odean (2008) argue that investors consider purchasing only stocks that have caught their attention since investors have difficulty in searching the thousands of stocks they can potentially buy. Their attention-driven buying behavior subsequently causes positive price pressure of the attention grabbing stocks. By proxying the Google search volume of stock ticker as the attention measure, Da et al. (2011) show a positive link between Google search volume and stock returns, confirming the attention theory of Barber and Odean (2008). The following studies also show that the Google search volume is positively correlated with stock returns (Bank et al., 2011; Zhang et al., 2013; Vozlyublennaia, 2014). However, investors are shown to have different sentiments and inconsistent trading patterns during stock market booms and recessions (Baker and Wurgler, 2007; Chung et al., 2012; Stambaugh et al., 2012; Garcia, 2013; Gong et al., 2014). For example, Garcia (2013) indicates that investors use different decision-making rules during good times and bad times, because human behavior is significantly different in times of anxiety and fear versus periods of prosperity and tranquility. Vlastakis and Markellos (2012) also indicate that the effect of Google search volume on market volatility becomes stronger in high return market states, implying the influence of market states on the link between web search activity and capital market. It thus raises a question that does the web search activity have the same effect on stock returns under different market states? To address this question, we follow previous studies (Da et al., 2011; Vlastakis and Markellos, 2012; Vosen and Schmidt, 2012; Aouadi et al., 2013; Siganos, 2013; Zhang et al., 2013) in proxying web search activity on the weekly Google search volume Index (hereafter, SVI) and classifying three different market states, i.e. the up, regular and down state based on the approach of Ryan and Taffler (2004) and Vlastakis and Markellos (2012). Unlike most previous studies focusing on the American and European stock markets (Bank et al., 2011; Da et al., 2011; Vlastakis and Markellos, 2012; Aouadi et al., 2013; Ding and Hou, 2013; Gwilym et al., 2013; Siganos, 2013), we examine the case of the Taiwan stock market. Taiwan is one of the top 10 densely populated countries in the world and the Internet users in Taiwan are roughly 75% of the population. 1 The substantially higher population density with higher Internet penetration rate makes the more active web search behavior and more rapid information transfer in Taiwan stock market. Moreover, individual investors account for roughly 90% of all stock trading volume in Taiwan (Barber et al., 2009). Since individual investors are more likely to search financial information related to a stock in Google than institution investors (Da et al., 2011), the large trading volume of individual investors suggests a material influence of Google search volume on Taiwan capital market. We hope to complement prior work by providing the related evidence from Taiwan stock market where individual investors are the majority participants. Our study also contributes to the emerging literature on the influence of Google search activity 1 According to the release of Internet World Stats ( there were 17,530,000 Internet users in Taiwan, representing 75.4% of the population at mid-year The Council for Economic Planning and Development (CEPD) also noted that Taiwan s Internet penetration rate is the fourth highest in Asia.

4 on capital market. Using SVI of stock ticker or company name, studies find a positive link between SVI and stock returns (Bank et al., 2011; Da et al., 2011; Zhang et al., 2013), price increase in target firms (Siganos, 2013), market index returns (Vozlyublennaia, 2014), stock trading volume and volatility (Vlastakis and Markellos, 2012; Aouadi et al., 2013), as well as shareholder base (Ding and Hou, 2013) and stock liquidity (Bank et al., 2011; Ding and Hou, 2013). Dzielinski (2012) also construct a novel measure of economic uncertainty based on SVI data of the word economy and show its predictive power to stock return and volatility. We extend these studies by taking different market states into account to empirically test the influence of Google search activity on stock market under the cyclical stock market conditions. Our findings show that the market states do not alter investors behavior to search for information via the Internet. However, the overall stock returns, volatility, turnover rate in the up markets are significantly higher than those in the down markets, suggesting that investors are more active when the stock market is bullish. When we turn to examine the effect of web search activity on stock returns, which is the factor that investors care about most when evaluating their investments, we find a significantly positive association between SVI and stock returns in the regular and up market states. According to the attention theory of Barber and Odean (2008), stocks with more investor attention are likely to experience positive price pressures. However, since investors use different decision-making rules during good times and bad times (Garcia, 2013), we argue that the fear and the pessimism may induce reluctance to buy stocks in the bearish market and thus limit the attention effect on stock returns in the down markets. Our robust analysis further reveals that the attention effect on stock returns is absent during the 2008 Financial Tsunami and the 2011 European Sovereign Debt Crisis. In summary, our findings indicate that the market state plays a role in affecting the relationship between web search activity and stock returns, although the investors may exhibit the same behavior to search for information via the Internet across time. The rest of the paper is organized as follows. The next section provides the literature review of attention effect. Section 3 details the data and summary statistics. Section 4 and Section 5 present empirical findings and robustness analyses. The final section offers concluding remarks. 2. Literature Review: The Effect of Investor Attention Two principal schools of thought have addressed the economic effect of investor attention on stock returns. Merton (1987) argued that firms which capture less investor attention have to offer higher stock returns so that investors can be compensated for imperfect diversification due to incomplete information. The other school of thought postulates an opposite effect of investor attention. According to the attention theory of Barber and Odean (2008), individual investors are net buyers of attention-grabbing stocks; an increase in investor attention leads to net buying and thus results in positive price pressure. In this context, Google search volume on the Internet serves as a direct measure of investor attention 2 (Da et al. 2011); therefore, a dramatic increase in search 2 Da et al. (2011) provide a comprehensive analysis to support it.

5 volume can be viewed as a boon in investor attention that pushes the stock prices up. A couple of empirical studies have confirmed the strong impacts of Google search volume on stock returns. Using a sample of Russell 3000 stocks, for example, Da et al. (2011) found that an abnormal increase in SVI predicts higher stock prices in the next two weeks. Similarly, Bank et al. (2011) found evidence that an increase in SVI is associated with temporarily higher future returns. 3. Data Following previous studies (Da et al., 2011; Vlastakis and Markellos, 2012; Vosen and Schmidt, 2012; Aouadi et al., 2013; Siganos, 2013; Zhang et al., 2013), we use SVI as a proxy of web search activity. Since data on SVI is not available prior to 2004 from Google Insights for Search, the sample is limited to the period from January 2004 to October In contrast to Da et al. (2011) employing specific ticker for stocks as search identifier, we focus on the search frequency of each firm s abbreviated name given by the TWSE. 3 We do this for two reasons: firstly, these abbreviated company names are not only commonly seen on the mass media and official websites, but also used as stock symbols when investors trade in the Taiwan stock market. Therefore, investors tend to use these abbreviated company names as keywords when searching on Google. Second, owning to the intrinsic attribute of Chinese language, most of the stock names are compounding words, indicating that they have exclusive meanings (Zhang et al., 2013). We focus on all listed common stocks ever traded on the Taiwan Stock Exchange (TWSE) during this period and obtain a sample of 866 common stocks. For each stock in our sample, we draw the corresponding time series of its SVI from Google Insights for Search. Since Google Insights designates a certain threshold of traffic for search terms, we obtain SVI data only on 805 common stocks, of which 542 are at weekly frequency. To avoid arbitrariness and assure the reliability of the analysis, we also remove stocks whose abbreviated firm names have a generic meaning or sound indistinguishable. For example, we remove the SVI data of stock abbreviated as Greater Taipei, which exhibits a geographical meaning. Another example is Elite for stock code 2331, which also possesses multiple meanings. As a result, we collect a total of 217,175 firm-week observations. To measure stock trading activity and to construct traditional asset pricing models, we collect all relevant data include trading volume, trading volume in dollars, stock turnover rate, market return, closing price, risk free rate, market capitalization, book values, and equity values from the Taiwan Economic Journal database (TEJ). We follow the conventional approaches to evaluate asset pricing factors. Take the momentum factor for example. We rank all stocks with a formation period of six months and then group them into three portfolios based on stock performance. The top 30% of stocks constitute the winner portfolio, while the 30% worst ones form the loser portfolio. Return on the winner portfolio minus return on the loser portfolio yields the momentum factor. For the risk-free rate we use the one-month money market rate. 錯誤! 找不到參照來源 provides the summary statistics. 3 For each firm s abbreviated name given by the TWSE, please see

6 variable mean std. dev. min median max skewness kurtosis N Variables Related to Investor Attention SVI ,175 Variables Related to Stock Characteristics FirmRet ,175 Vol 25,099 55, ,754 2,227, ,175 VolNTD 869, , , ,600, ,175 Turnover ,175 Table 1. Descriptive statistics. SVI is the aggregate search frequency from Google Trends based on the official abbreviation of company name; FirmRet is the stock return during week t; Vol is the number of shares traded volume of stock; VolNTD is the Trading volume in New Taiwan Dollars (NT Dollars), calculated as the number of shares traded volume multiplied by the respective price; Turnover is the fraction of shares traded relative to the number shares outstanding. 4. Empirical Results 4.1 Stock Characteristics across different Market States We first start by examining the stock characteristics across three different market states. In Table 2, we conduct t-tests to compare a few essential stock characteristics and to see whether stock trading in the up markets differs from that in the down markets. In line with previous studies (Ryan and Taffler, 2004; Vlastakis and Markellos, 2012), we define three market states on the basis of the distribution of weekly market returns. 4 Specifically, weeks in which the absolute return of the market lie within one standard deviation of the mean are classified as the regular market states, while those in which the absolute return of the market greater (less) than one standard deviation of the mean are defined as the up (down) market states. Panel A lists some statistics for each market state. The regular state in which market returns vary from -1% to 1% accounts for 74.57% of our sample. Panel B shows the comparison results of the average value of stock characteristics for each market state. We find that, except for SVI, the overall stock returns, volatility, turnover rate in the up markets are significantly higher than those in the down markets. Because previous studies regard these three variables as essential measures of trading activity in the capital market (Karpoff, 1987; Lo and Wang, 2000; Chordia et al., 2001; Chordia et al., 2007), the significantly higher value in the up markets suggests that investors are more active when the stock market is bullish. It is noteworthy that the t-test fails to distinguish the level of web search activity measure by 4 We divide market states into the up, regular, and down states rather than the two market states used by Ryan and Taffler (2004) and Vlastakis and Markellos (2012) in order to have a clear comparison of stock characteristics across different market States.

7 SVI across market states. Vlastakis and Markellos (2012) find that the information demand proxied by SVI increases significantly during periods of higher returns; however, our result shows that investors in Taiwan have a stable information demand across time. The different findings between this study and Vlastakis and Markellos (2012) may be attributed to the sample characteristic since they use data for the 30 largest stocks traded on NYSE and NAXDAQ, while we examine the whole stock market of Taiwan. Although market states do not dramatically alter investors behavior to search for information via the Internet, investors are reluctant to trade in bad times duo to the fear of losing more money. Consequently, trading activity in the capital market appears to be more active in the up markets than in the down markets. Panel A: Information about Each Market State Market States Down Regular Up Std. Dev. (-, -1) [-1, 1] (1, ) Min Max Number of firms Observations 30, ,939 24,402 Panel B: Stock Characteristics across Market States Market States VARIABLES Down Regular Up t-value p-value FirmRet it Mean *** 0.00 Median Vol it Mean 23,738 24,633 29, *** 0.00 Median 5,940 6,824 7,474 VolNTD it Mean 834, , , *** 0.00 Median 129, , ,900 Turnover it Mean *** 0.00 Median SVI it Mean Median

8 Table 2. Stock characteristics across different market states. Three market states are defined following the approach of Ryan and Taffler (2004) and Vlastakis and Markellos (2012). Variables are defined as previous. *, **, *** denote 10%, 5% and 1% significance levels. 4.2 Effect of web search activity on stock returns across different market states To further examine whether the different market sates influence the effect of web search activity on the stock returns, we conduct a panel regression analysis for each subsample under three market states. Specifically, we use the SVI variable to proxy for web search activity and regress the weekly stock returns on it. Control variables include typical asset pricing factors such as market returns ( t), size (S B t ), book-to-market (H L t ), and momentum ( O t), which come from three traditional asset pricing models: the market model, the Fama and French (1996) three-factor model, and the Carhart (1997) four-factor model. Moreover, we include firm fixed effect in all panel regressions with the following models: it = b 0 + b 1 SVI it + b 2 t + c i + u t + ε i,t. (1) ε i,t. it = b 0 + b 1 SVI it + b 2 t + b 3 S B t + b 4 H L t + c i + u t + (2) it = b 0 + b 1 SVI it + b 2 t + b 3 S B t + b 4 H L t + b 5 O t + c i + u t + ε i,t. (3) The results are reported in Table 3. In each specification, the dependent variable it is always return premium, calculated as the stock return minus the risk-free rate. Specifications (1)-(3), (4)-(6), and (7)-(9) correspond to the down, regular, and up market states, respectively. The coefficients on SVI it are significantly positive in specifications (4)-(9), showing that the web search activity causes positive price pressure in the regular and up market states. Moreover, the values of coefficients on SVI it are larger in the up states than those in other states, suggesting that the influence of web search activity on stock returns become stronger in the bulls. Da et al. (2011) prove that SVI is the direct measure of investor attention and can drive stock returns as Barber and Odean s (2008) attention theory predict. However, the fear and the pessimism may induce reluctance to buy stocks in the bearish market and thus limit the attention effect on stock returns in the down markets. Therefore, our findings generally support attention theory and previous studies (Bank et al., 2011; Da et al., 2011; Zhang et al., 2013; Vozlyublennaia, 2014) but to some degree reveal the role of market state in the attention effect on stock returns.

9 Variable Market States Down Regular Up Std. Dev. (-, -1) [-1, 1] (1, ) (1) (2) (3) (4) (5) (6) (7) (8) (9) Intercept 0.833*** 0.207* 0.193* *** *** *** ** * (0.111) (0.106) (0.108) (0.023) (0.024) (0.024) (0.137) (0.135) (0.136) SVI it *** 0.004*** 0.004*** 0.006** 0.005** 0.005** (0.002) (0.002) (0.002) (0.001) (0.001) (0.001) (0.002) (0.002) (0.002) MktRetRf t 1.252*** 1.088*** 1.084*** 0.971*** 0.989*** 0.987*** 1.112*** 1.111*** 1.101*** (0.021) (0.019) (0.018) (0.013) (0.013) (0.013) (0.032) (0.032) (0.032) SMB t 0.463*** 0.462*** 0.439*** 0.433*** 0.527*** 0.523*** (0.019) (0.019) (0.015) (0.015) (0.021) (0.021) HML t 0.190*** 0.188*** 0.061*** 0.056*** (0.018) (0.018) (0.019) (0.019) (0.019) (0.020) MOM t *** (0.014) (0.007) (0.017) it Observations 30,834 30,834 30, , , ,939 24,402 24,402 24,402 R-squared Number of firms Table 3 Effects of web search activity on stock returns across different market states. Three market states are defined following the approach of Ryan and Taffler (2004) and Vlastakis and Markellos (2012). FirmRet is the stock return during week t; SVI is the aggregate search frequency from Google Trends based on the official abbreviation of company name; MktRet is the Market Return. Robust standard errors are given in parentheses. *, **, *** denote 10%, 5% and 1% significance levels. 5. Robust analysis 5.1 Alternative definition of market states In this section, we examine whether our inference is sensitive to the alternative definition of market states. We use one standard deviation as a unit to divide the whole sample and investigate whether our results are robust to the dividing points between market states. Table 4 presents the regression results for each subsample which is divided by the standard deviation of market returns. The range value of standard deviation for each subsample increases monotonically with the specification number. Going from specification (1) to (7), we can find that coefficients on SVI it are not statistically significant at the 5% level until we reach specification (5), whose lower bound is located around zero. The finding implies that web search activity exert its influence on stock returns only when the market goes up. Moreover, its influence grows stronger particularly when the market

10 performs unusually well, as shown in specification (6). However, coefficients on SVI it lose its statistical significance again in specification (7), which might be due to insufficient observations. Overall, the estimation results shown in Table 4 are consistent with the evidence presented in Table 3. Therefore, the inference that the effect of web search activity on stock returns is stronger in the up markets than in the down markets is robust to our definition of market state. VARIABLES Market States (Std. Dev.) [-, -3) [-3, -2) [-2, -1) [-1, -0) [0, 1) [1, 2) [2, 3) (1) (2) (3) (4) (5) (6) (7) Intercept *** 1.895** 0.762*** *** (6.230) (0.832) (0.202) (0.047) (0.043) (0.210) (1.642) SVI it * *** 0.009*** (0.007) (0.005) (0.003) (0.001) (0.001) (0.002) (0.008) MktRetRf t 3.608*** 1.318*** 1.225*** 0.978*** 0.942*** 1.305*** 0.707*** (0.616) (0.117) (0.048) (0.025) (0.020) (0.050) (0.232) SMB t 0.409*** 0.571*** 0.528*** 0.448*** 0.423*** 0.507*** 0.495*** (0.026) (0.033) (0.024) (0.017) (0.0156) (0.023) (0.061) HML t *** 0.152*** 0.066*** 0.047** (0.142) (0.034) (0.024) (0.020) (0.021) (0.025) (0.046) MOM t 0.309*** *** *** 0.054*** ** (0.068) (0.023) (0.019) (0.011) (0.010) (0.016) (0.060) it Observations 2,816 6,773 21,245 64,459 97,480 21,044 2,874 R-squared Number of firms Table 4 Robust analysis: alternative definition of market states. Variables are defined as previous. Robust standard errors are given in parentheses. *, **, *** denote 10%, 5% and 1% significance levels. 5.2 Sub-period analysis In this section, we repeat the regression analyses year by year. This analytic procedure enables us to observe the influences of web search activity on stock returns during financial crises. Table 5 presents the regression results for each year of the sample period. The results indicate that the influences of web search activity on stock returns are absent during the 2008 Financial Tsunami and the 2011 European Sovereign Debt Crisis. The results are quite consistent with our expectation that the attention effect tends to be insignificant during bad times. Furthermore, coefficients on SVI are larger in bullish markets (as those in 2009 and 2010) than in bearish markets (as those in 2008 and 2011). The results are in line with our previous finding that the effect of web search activity on stock returns is influenced by market states.

11 VARIABLES FirmRetRf t Year (1) (2) (3) (4) (5) (6) (7) (8) (9) Intercept *** *** *** *** * *** (0.064) (0.061) (0.080) (0.099) (0.105) (0.165) (0.130) (0.144) (0.113) SVI it ** 0.007** 0.012*** *** 0.024*** 0.010* 0.014*** (0.002) (0.002) (0.003) (0.004) (0.004) (0.006) (0.005) (0.006) (0.004) MktRetRf t 0.985*** 0.938*** 1.030*** 0.987*** 1.106*** 1.005*** 1.060*** 1.058*** 1.035*** (0.019) (0.027) (0.021) (0.018) (0.017) (0.020) (0.018) (0.018) (0.022) SMB t 0.348*** 0.328*** 0.440*** 0.543*** 0.550*** 0.495*** 0.438*** 0.587*** 0.594*** (0.020) (0.023) (0.025) (0.023) (0.024) (0.026) (0.022) (0.029) (0.029) HML t ** *** 0.257*** 0.062*** 0.074*** 0.077*** (0.022) (0.030) (0.030) (0.031) (0.020) (0.025) (0.026) (0.024) (0.024) MOM t *** *** 0.134*** *** *** (0.018) (0.021) (0.015) (0.027) (0.018) (0.015) (0.016) (0.022) (0.020) Observations 22,187 22,999 23,873 23,698 24,626 24,694 25,404 26,536 23,158 R-squared Number of firms Table 5 Sub-period analysis. Variables are defined as previous. Robust standard errors are given in parentheses. *, **, *** denote 10%, 5% and 1% significance levels. 6. Conclusions The objective of this study is to examine the influence of market states on the relationship between web search activity and capital market from the perspective of stock returns. We use Google search volume as a proxy of web search activity and classify the sample period into up, regular and down market states based on the approach of Ryan and Taffler (2004) and Vlastakis and Markellos (2012). Consistent with previous studies (Bank et al., 2011; Da et al., 2011; Zhang et al., 2013), we find that Google search volume drives stock returns in the regular and up market states although it does not change significantly under different market states. Da et al. (2011) employ Barber and Odean s (2008) attention theory to explain the positive price pressure of attention-grabbing stocks. We argue that the fear and the pessimism may induce reluctance to buy stocks in the bearish market and thus limit the attention effect on stock returns in the down markets. In conclusion, our results suggest that the public index of Google search volume may provide company managers an easy access to observe the changes in investor attention and the possible shifts in stock price when the stock market performs well.

12 References Aouadi, A., Arouri, M., and Teulon, F. (2013). Investor attention and stock market activity: Evidence from France. Economic Modelling, 35, doi: /j.econmod Baker, M., and Wurgler, J. (2007). Investor sentiment in the stock market. Journal of Economic Perspectives, 21(2), Bank, M., Larch, M., and Peter, G. (2011). Google search volume and its influence on liquidity and returns of German stocks. Financial markets and portfolio management, 25, Barber, B. M., Lee, Y. T., Liu, Y. J., and Odean, T. (2009). Just How Much Do Individual Investors Lose by Trading? Review of Financial Studies, 22(2), doi: /rfs/hhn046 Barber, B.M., and Odean, T. (2008). All that glitters: the effect of attention and news on the buying behavior of individual and institutional investors. Review of Financial Studies, 21, Carhart, M. M. (1997). On persistence in mutual fund performance. Journal of Finance, 52(1), Chordia, T., Huh, S. W., and Subrahmanyam, A. (2007). The cross-section of expected trading activity. Review of Financial Studies, 20(3), doi: /rfs/hhl014 Chordia, T., Roll, R., and Subrahmanyam, A. (2001). Market liquidity and trading activity. The Journal of Finance, 56(2), Chung, S. L., Hung, C. H., and Yeh, C. Y. (2012), When does investor sentiment predict stock returns? Journal of Empirical Finance, 19, , doi: /j.jempfin Da, Z., Engelberg, J., and Gao, P. J. (2011). In search of attention. Journal of Finance, 66(5), doi: /j x Fama, E. F., and French, K. R. (1996). Multifactor explanations of asset pricing anomalies. Journal of Finance, 51(1), Garcia, D. (2013). Sentiment during Recessions. Journal of Finance, 68(3), Gong, B., Lei, V., and Pan, D. (2013). Before and after: the impact of a real bubble crash oninvestors trading behavior in the lab, Journal of Economic Behavior & Organization, 95, Karpoff, J. M. (1987). The relation between price changes and trading volume: a survey. Journal of Financial & Quantitative Analysis, 22(1), Lo, A. W., and Wang, J. (2000). Trading volume: definitions, data analysis, and implications of portfolio theory. Review of Financial Studies, 13(2), Ryan, P., Taffler, R.J., (2004). Are economically significant stock returns and trading volumes driven by firm specific news releases? Journal of Business Finance & Accounting, 31, Siganos, A. (2013). Google attention and target price run ups. International Review of Financial Analysis, 29(0), doi: Stambaugh, R. F., Yu, J. F., and Yuan, Y. (2012). The short of it: investor sentiment and anomalies.

13 Journal of Financial Economics, 104(2), Vlastakis, N., and Markellos, R. N. (2012). Information demand and stock market volatility. Journal of Banking & Finance, 36(6), doi: /j.jbankfin Vosen, S., and Schmidt, T. (2012). A monthly consumption indicator for Germany based on Internet search query data. Applied Economics Letters, 19(7), doi: / Vozlyublennaia, N. (2014). Investor attention, index performance, and return predictability. Journal of Banking & Finance, 41(0), doi: Zhang, W., Shen, D. H., Zhang, Y. J., and Xiong, X. (2013). Open source information, investor attention, and asset pricing. Economic Modelling, 33, doi: /j.econmod