Investors’ financial attention frequency and trading activity

Investors’ financial attention frequency and trading activity

Pacific-Basin Finance Journal 58 (2019) 101239 Contents lists available at ScienceDirect Pacific-Basin Finance Journal journal homepage: www.elsevie...

1MB Sizes 1 Downloads 77 Views

Pacific-Basin Finance Journal 58 (2019) 101239

Contents lists available at ScienceDirect

Pacific-Basin Finance Journal journal homepage: www.elsevier.com/locate/pacfin

Investors’ financial attention frequency and trading activity Wenwu Caia, Jing Lua,b, a b



T

School of Economics and Business Administration, Chongqing University, China Corporate Finance and Accounting Governance Innovation Institute, Chongqing University, China

ARTICLE INFO

ABSTRACT

Keywords: Securities service mobile applications Financial attention frequency Ostrich effect Trading

Based on data on users' daily adoption of securities service mobile applications, we measure investors' financial attention frequency, which reflects how often they use such apps, to obtain information on their frequency of opening these apps and online duration. We find that financial attention frequency shows a clear ostrich effect, suggesting that investors acquire financial information less frequently following periods of low market returns and high market volatility. In addition, it significantly promotes trading activity in the market. Further, this driving force remains after a series of robustness tests controlling for other market factors such as investor attention and sentiment and after addressing endogeneity concerns. Finally, financial attention frequency also increases individuals' net buying transactions.

JEL classification: G12 G14

1. Introduction With the rapid development and popularization of mobile Internet technology, smartphones have become the main tool for investors to obtain information and manage assets (Moore, 2013; Patel, 2014; Brown et al., 2015). Securities service mobile applications (SSMAs), which are built into smart mobile devices, integrate market dynamics, stock and news information, research reports, stock recommendations, and trading systems to provide investors with 24/7, low-cost, and efficient financial information as well as trading services. These popular SSMAs integrate demand for and the supply of financial information, making them the core intermediary of the new information era. Hence, their use reflects the way in which investors now often acquire financial information. However, do investors' decisions to obtain financial information through SSMAs change with market dynamics or conditions? Does the use of SSMAs affect investors' trading activities? And does it induce investors to buy more? We examine these issues based on the A-share market in China to better understand how investors make their information acquisition decisions through SSMAs and uncover the possible linkages between SSMA usage and stock market transactions, topics that have thus far received insufficient scholarly attention. This research is rooted in our understanding of investor attention. Kahneman (1973) and Hirshleifer et al. (2002) defined investor attention from the perspectives of psychology and economics, respectively and explained its limited characteristics. As the limitations of investor attention render attention itself expensive, rational investors face a trade-off between the attention cost and its benefits (Sims, 2003, 2006; Huang and Liu, 2007). Based on this cost/benefit trade-off, many researchers are interested in how investors allocate attention to different assets at the same time. For example, Peng and Xiong (2006) argued that investors pay more attention

Abbreviations: China Stock Market & Accounting Research, (CSMAR); Easy View Thousand Sails, (EVTS); Heteroskedasticity- and autocorrelationconsistent, (HAC); Instrumental variable, (IV); Securities company, (SC); Securities information service company, (SISC); Securities service mobile application, (SSMA); Shanghai Securities Composite Index, (SSCI); Two-stage least squares, (2SLS) ⁎ Corresponding author at: School of Economics and Business Administration, Chongqing University, China. E-mail address: [email protected] (J. Lu). https://doi.org/10.1016/j.pacfin.2019.101239 Received 19 March 2019; Received in revised form 26 October 2019; Accepted 29 October 2019 Available online 31 October 2019 0927-538X/ © 2019 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/BY-NC-ND/4.0/).

Pacific-Basin Finance Journal 58 (2019) 101239

W. Cai and J. Lu

to market and industry information than individual stock information. Barber and Odean (2008) stated that individuals are distracted by hundreds of options when they buy stocks. Only those attention-grabbing stocks such as stocks with high media coverage, unusually high trading volumes, and extremely high or very low daily returns can enter investors' stock selection. Seasholes and Wu (2007), Huddart et al. (2009), and Da et al. (2014) also provided evidence for the influences behind how investors allocate attention to assets as well as their different allocation decisions. Moreover, previous studies, mainly based at the market level, also explore how investors allocate their attention to a single asset or portfolio in a different period (or under different market conditions). For example, Hou et al. (2009) pointed out that investors pay less attention to the stock market in a bear market than in a bull market. Li and Yu (2012) found that nearness to the Dow 52-week high and Dow historical high attracts more investor attention than other traditional macroeconomic variables. Moreover, Yuan (2015) suggested that both the new-high and the new-low records of market indices attract investor attention. Some studies have argued that in periods of high market volatility, investors increase the frequency of information acquisition to reduce the uncertainty in their portfolios (Bank et al., 2011; Vlastakis and Markellos, 2012; Dimpfl and Jank, 2016). However, Karlsson et al. (2009) linked information acquisition decisions to the hedonic utility of information1 and pointed out that investors obtain psychological pleasure from good news. They further found an ostrich effect in investor attention; that is, individuals are more active in obtaining information after initial good news, whereas they reduce the receipt of bad news to avoid its negative psychological impact. Sicherman et al. (2016) further confirmed the ostrich effect in return and volatility data. In our research, the ostrich effect of investor attention predicts that investors use SSMAs less actively to obtain financial information following periods of low market returns and high market volatility. Studies that have discussed investor attention from different perspectives have all drawn the relatively consistent conclusion that financial attention drives investors' trading activity (Barber and Odean, 2008; Fang and Peress, 2009; Da et al., 2011; Yuan, 2015). Thus, the driving force of investor attention predicts that more frequent information acquisition through SSMAs leads to more active market trading behavior. Barber and Odean (2008) and Huddart et al. (2009) suggested that limited attention stimulates net buying by individual investors, implying that the more frequent use of SSMAs encourages individuals to buy more shares. The growing popularity of smart mobile devices has provided more convenient access to information for individual investors (Moore, 2013; Patel, 2014). Moreover, the use of the SSMAs built into these devices reflects investors' information acquisition activities or their sensitivity to financial market information. In this study, based on 51,962 daily user behavior observations of 183 SSMAs in Mainland China from 2016 to 2018, we construct two indicators reflecting investors' financial attention frequency from two aspects: frequency of use and online duration. We further examine the factors influencing investors' financial attention frequency and its impact on market trading activities. The following findings are reported. First, investors' financial attention frequency has a clear ostrich effect, indicating that investors obtain financial information less frequently in periods of low market returns and high market volatility. Second, investors' financial attention frequency plays a significant role in promoting overall trading activities in the market. This conclusion remains unchanged after a series of additional robustness tests controlling for other market factors (e.g., investor overconfidence, attention, and sentiment), measuring financial attention frequency based on SSMAs without actual trading systems, adopting financial attention frequency on Sundays, and conducting two-stage least squares (2SLS) regressions using users' behavior data on the most popular Chinese mobile social and shopping apps (i.e., WeChat and TaoBao) as instrumental variables (IVs). Third, financial attention frequency significantly increases the net buying transactions of individual investors, indicating that it has a greater impact on individual investors' buying decisions than selling. The study by Sicherman et al. (2016) is similar to our research. They investigated financial attention using data on investors' online account logins and found that daily account logins decrease after market declines and with changes to the volatility index (VIX). However, our research differs from theirs in the following aspects. First, compared with the data on 100,000 paperless accounts used by Sicherman et al. (2016), our data cover most investors who use SSMAs in the A-share market, which allows us to construct an overall attention frequency indicator for that market. Second, investors' financial attention is not only reflected in account logins, as suggested by Karlsson et al. (2009) and Sicherman et al. (2016), but also shown in other ways. In our research, investors actively or inadvertently access market information or portfolio-related information simply by opening a specific SSMA, which helps better define and measure investors' financial attention. Third, Sicherman et al. (2016) identified whether an investor has access to financial information related to his or her holding portfolio only by whether he or she logs into his or her account. Therefore, their financial attention indicator actually reflects how many investors log in. However, we measure the frequency of investors' financial attention instead of the quantity. Fourth, we find the driving force behind the effect of financial attention frequency on individuals' net buying. As investor attention has become a popular and attractive topic in recent years, our findings could contribute to the research in this field in the following aspects. First, the popularity of the mobile Internet makes it easier for investors to obtain financial information through the SSMAs embedded in smart mobile devices, which is rebuilding the information diffusion model of the financial market. However, these changes following the advancement of information technology have attracted insufficient research attention. To the best of our knowledge, this is the first attempt to use SSMAs, thereby providing a new research perspective on investor attention. Second, existing research has directly measured investor attention by depending on the Internet search volume index (SVI) proposed by Da et al. (2011). While the SVI reflects how many investors have paid attention, it cannot measure the frequency of investors' access to information. By contrast, we find that frequency of use and online duration can effectively measure

1 The literature on information-dependent and belief-dependent utility posits that information also has a hedonic impact on utility that exceeds the mechanical costs and benefits (e.g., Loewenstein, 1987; Caplin and Leahy, 2001; Brunnermeier and Parker, 2005).

2

Pacific-Basin Finance Journal 58 (2019) 101239

W. Cai and J. Lu

investors' financial attention frequency. This also reflects the use of these SSMAs, a new popular financial technology, by individual investors. Third, the validation of the ostrich effect is mainly based on data from investor login accounts (Karlsson et al., 2009; Sicherman et al., 2016). Our findings thus provide new evidence of the ostrich effect in investors' financial attention from the perspective of attention frequency as well as at the overall market level. Fourth, our findings explain the driving force of attention frequency on transactions, enriching the research on investor attention at the market level. The rest of this paper is organized as follows. Section 2 introduces the data and variables. Section 3 investigates the ostrich effect in investors' financial attention frequency. Section 4 explores the stimulating effect of financial attention frequency on trading activities. Section 5 examines the impact of financial attention frequency on individual investors' net buying transactions and Section 6 concludes. 2. Data Measuring the frequency of investors' financial attention or its sensitivity to financial markets is challenging. Fortunately, the development of big data technology allows us to track mobile users' SSMA usage and thus acquire user behavior data. Based on these SSMA usage data, we propose an effective method of measuring investors' financial attention frequency on a daily basis. 2.1. The Easy View Thousand Sails (EVTS) database Competition in the mobile app industry is becoming increasingly intensive. Most mobile app developers cooperate with big data companies to comprehend the development trends and user statuses of mobile apps in their fields. Yiguan Corp. is one of the leading big data analysis firms in China. It has built data analysis tools, products, and solutions based on massive digital user assets and effective algorithm models, which can help enterprises (i.e., app developers) efficiently manage their digital user assets, conduct better operations of products, upgrade their businesses to achieve sales growth, and evade significant business risk. Yiguan's major products include EVTS, Easy View Ark, and Industry Solutions. As of December 31, 2018, these products covered 2.38 billion smart terminals and 604 million personal users, accumulating data from > 50,000 mobile apps. In this study, our raw data come from the EVTS database of Yiguan Corp. By docking a software development kit with mobile app software developers, the EVTS database collects non-identity information about users' equipment (e.g., operating system, device type, device settings, IP address) and information related to the mobile app behavior of mobile terminal users (e.g., frequency of use and online duration) with developers' permission.2 The EVTS database divides these mobile apps into 45 application fields such as finance, sports, and mobile shopping. Then, it further subdivides the apps into 314 sub-industries in these 45 fields. For example, it subdivides mobile apps in financial services into 17 sub-industries including securities services, insurance services, mobile banking services, and lottery services. 2.2. Selection and statistics of the samples The data used in this study come from the securities service sub-industry in the financial services field of the EVTS database. We obtain users' behavior data from 307 mobile apps (including past data on newly entering as well as exiting mobile apps) involving securities services in Mainland China from January 1, 2016, to December 31, 2018. The data mainly include the name as well as the active number of users,3 average frequency of use, and online duration4 of each SSMA daily. Table 1 reports users' behavior and characteristics in relation to the top 15 SSMAs in Mainland China (i.e., those that have the highest number of daily active users on average). In Table 1, we distinguish SSMA developers into two types: third-party securities information service companies (SISCs) and securities companies (SCs). Further, the main functions of these SSMAs are classified as whether they (i) provide financial information, including real-time market dynamic information, individual stock financial information, and important market news; (ii) support online forums to facilitate online interactions among investors; and (iii) allow access to securities trading systems. In Table 1, the three most popular SSMAs, namely Flush (同花顺), East Money (东方财富), and Da Zhihui (大智慧), are all developed by third-party SISCs. These three SSMAs are free for investors. They not only have a higher average number of daily active users, higher frequency of use, and longer online duration than other SSMAs, but also integrate financial information, online forums, and trading systems (mainly docked with the trading systems of small SCs because large SCs have developed their own mobile apps). In particular, Flush has an average of 18.86 million active users daily, almost eight daily uses, and 0.62 h of daily online time. Other SSMAs are developed by the large-scale SCs in Mainland China for their own customers (except for Tencent Zixuangu (腾讯自选股), 2 Under the premise of the authorization of app developers, Yiguan Corp. uses software development kit technology to bury users' mobile devices and collect their behavior data about their use of mobile apps. When the user downloads the specific mobile app that docks the technology developed by Yiguan, their use of this app is monitored. This kind of monitoring is through the cooperation and license of app developers and is built into a data module of the mobile app. This means that users cannot evade such monitoring unless they do not use the app. Researchers can obtain more information and purchase access to the EVTS database from https://qianfan.analysys.cn. 3 When a user launches a mobile app, he or she is regarded as an active user of that app on that day. 4 According to the data acquisition rules of the EVTS, the measurement of an app's online duration only calculates the time of displaying a specific app on the home screen of the user's mobile device. This measurement does not consider the app's running time in the background.

3

4

1886.31 508.09 374.27 307.60 165.00 158.75 149.57 142.33 140.19 132.30 131.38 120.06 118.19 117.06 107.84

Flush East Money Da Zhihui Zhangle Caifutong Qingting Dianjin Yitaojin Guotai Junan Jin Taiyang e-Haitongcai Pingan Securities Galaxy Securities Zhiyuan Yihutong Zhongtai Qifutong Tencent Zixuangu Fangzheng Xiaofang

同花顺 东方财富 大智慧 涨乐财富通 蜻蜓点金 易淘金 国泰君安 金太阳 e海通财 平安证券 银河证券 智远一户通 中泰齐富通 腾讯自选股 方正小方

7.92 7.49 6.23 6.00 4.55 5.14 5.41 5.83 5.45 5.05 5.64 5.93 4.76 5.22 5.27

Average daily personal startup frequency (Times)

0.62 0.60 0.49 0.45 0.29 0.33 0.48 0.41 0.48 0.35 0.43 0.36 0.39 0.37 0.36

Average daily personal online length (Hours)

SISC SISC SISC SC SC SC SC SC SC SC SC SC SC SISC SC

Nature of developer (SISC or SC)

YES YES YES YES YES YES YES YES YES YES YES YES YES YES YES

Financial information

YES YES YES YES NO NO NO NO NO NO NO NO NO YES NO

Online forums

Major functions (YES or NO)

YES YES YES YES YES YES YES YES YES YES YES YES YES NO YES

Trading system

This table reports the user behavior statistics and app characteristics of the 15 most popular SSMAs in Mainland China. The research period is from January 1, 2016 to December 31, 2018. We distinguish SSMA developers into third-party SISCs and SCs. The main functions of these SSMAs include whether they provide financial information, support online forums, and allow access to securities trading systems.

Average daily app users (10 Thousand)

SSMAs name (in English)

SSMAs name (in Chinese)

Table 1 User behavior statistics and app characteristics of the 15 most popular SSMAs.

W. Cai and J. Lu

Pacific-Basin Finance Journal 58 (2019) 101239

Pacific-Basin Finance Journal 58 (2019) 101239

W. Cai and J. Lu

which is developed by Tencent Corp.). Such SSMAs are limited by the customer size of their securities brokerage businesses and do not provide online forum exchanges for investors. Hence, the number of users and user usage frequency are lower than those of the top three SSMAs. Overall, SSMAs are gradually becoming the main market information acquisition and trading tools for individual investors.5 To obtain more accurate data, we exclude observations with fewer than 100,000 daily active users because a newly released SSMA may result in serious statistical errors during test runs. Table 2 shows our sample selection process and the overall activity statistics of the SSMA category on trading days. Panel A in Table 2 shows the sample selection process. It is found that approximately 30 SSMAs have more than one million active users per day. These popular SSMAs are our main interest because they not only have a huge user base but also relatively mature technology. Therefore, compared with SSMAs that have fewer users, their operations and statistics are more continuous, thus providing a better sample for our research. In addition, > 57,000 observations have fewer than 100,000 daily active users. SSMAs that provide these observations have not only few users, but also a low market survival rate. Therefore, we exclude these data and finally obtain 51,962 daily user behavior observations from 183 SSMAs. Panel B in Table 2 shows the activity statistics of these SSMAs grouped according to the number of active users on trading days. Combined with Table 1, we find that the average number of active users of Flush alone accounts for 25.45% of the market; it also has the highest frequency of use and online duration daily. SSMAs with between 100,000 and 10 million active users account for 69.10% (10.63% + 33.56% +24.91%) of the total users in the market (attributable to the larger number of SSMAs), but their user usage frequency is much lower than that of the group with the highest active users. Panel C further shows the average daily activity of individual users for each SSMA in each group. For SSMAs with more than five million users, the average frequency of use per user daily is 8.02, which is greater than the overall average (6.29), while the average online duration is 0.69, which is longer than the overall average (0.52). In general, Table 2 suggests that although our basic observations come from 183 SSMAs, the top 30 SSMAs that have a higher number of and more frequent users contain most of the information we need. Furthermore, compared with the 39.34 million investors holding accounts in the Shanghai Stock Exchange at the end of 2017, the SSMAs in China's A-share market are popular. Therefore, the impact of the use of these SSMAs on the trading behaviors of investors cannot be ignored. 2.3. Investors' Financial attention frequency Although most SSMAs provide information services and support online transactions simultaneously, investors mainly use them to obtain market or portfolio information. First, investors' motivations to obtain information (whether market information or price information related to holding assets) are much greater than trading. By analyzing detailed panel data on 1,168,309 investors with defined contribution retirement accounts for 2007–2008, Sicherman et al. (2016) pointed out that “trading is very infrequent compared to logins.” According to their research data, 99% of individual investors log in on 507 days but only trade on 21 days. Second, frequent intraday trading is limited in the A-share market. Owing to the “T + 1” trading rule6 implemented in this market, investors cannot sell until the next trading day after buying shares. This means that investors in the A-share market are unlikely to conduct intraday high-frequency trading. Third, although the frequency of use of SSMAs may be affected by investors' trading behavior, the online duration of these apps is an accurate measure of information acquisition frequency. Because the submission of trading orders is instantaneous, it can be neglected compared with tracking dynamic market information. Therefore, we regard investors' use of SSMAs as one way in which they obtain information. The foregoing shows that we cannot simply use the daily number of active users of SSMAs to construct an index similar to the SVI. On the one hand, a user may use different SSMAs at the same time (e.g., he or she may use both a SSMA developed by a third-party SISC and a SSMA developed by a SC). On the other hand, as the mobile Internet is a new technology, the gradual popularization of smartphones and mobile apps is raising the number of active users over time. To minimize the impact of these factors on the accuracy of our indicators, we calculate average user activity weighted by the number of active users to represent the average financial attention frequency of investors as follows:

FreqSTtRaw =

FreqOTtRaw = Raw

Nt i=1 Nt i=1

Usersi, t Nt i=1

STi, t (1)

Usersi, t

Usersi, t Nt i=1

OTi, t (2)

Usersi, t

Raw

FreqSTt and FreqOTt represent the daily frequency of investors' financial attention based on frequency of use and online duration, respectively. Nt is the number of SSMAs on day t. Usersi,t represents the number of active users of SSMA i on day t. STi,t is the 5

Compared with the higher number of daily active users of these SSM apps, the numbers of A-share individual investors' holding accounts in the Shanghai Stock Exchange at the end of 2016 and 2017 were 37.40 million and 39.34 million, respectively, and it is permissible for an investor to have multiple accounts. 6 Here “T + 1” is the stock trading system in which stocks bought on the same day cannot be sold until the next trading day. “T” means the date of buying a stock and “T + 1” means the next day after buying the stock. Since 1995, the Shanghai Stock Exchange and Shenzhen Stock Exchange have implemented the “T + 1” trading system for all A-share trading. 5

Pacific-Basin Finance Journal 58 (2019) 101239

W. Cai and J. Lu

Table 2 Sample selection and user behavior statistics of SSMAs. Panel A: SSMA selection process App-group by user size

App-day(# of Observations)

Number of SSMAs (With Overlap)

> 10 Million 5 Million To 10 Million 1 Million To 5 Million 100 Thousand To 1 Million < 100 Thousand Total Baseline observations

721 490 8779 41,972 57,016 108,978 51,962

1 3 28 179 274 307 183 (Without Overlap)

Panel B: Daily total activity of SSM APP-group by user size

Total daily active users (10 Thousand)

Percent (%) Total daily startup frequency (10 Thousand Times)

Total daily online length (10 Thousand Hours)

> 10 Million 5 Million To 10 Million 1 Million To 5 Million 100 Thousand To 1 Million < 100 Thousand Total

1899.91 793.61 2504.94 1859.42 406.74 7464.62

25.45% 10.63% 33.56% 24.91% 5.45% 100%

1222.07 590.41 1167.93 667.74 170.52 3818.67

15,422.63 6275.91 14,754.3 9201.36 1909.27 47,563.47

Panel C: Average daily SSMA users' activity APP-group by user size

Average daily personal startup frequency (Times)

Average daily personal online length (Hours)

> 10 Million 5 Million To 10 Million 1 Million To 5 Million 100 Thousand To 1 Million < 100 Thousand Mean

7.95 8.08 5.69 5.14 4.61 6.29

0.62 0.76 0.43 0.40 0.41 0.52

This table shows the mobile app selection process and distribution of SSMA users' activity on trading days for each group during January 1, 2016 to December 31, 2018. We group these SSMAs according to their number of daily active users. Panel A shows the SSMA selection process. Panel B shows the daily activity of different groups. Panel C reports the daily average activity of each group per user.

average frequency of use (measured as the number of times the app is opened) per user for SSMA i on day t. OTi,t is the average online duration per user for SSMA i on day t. However, with the accumulation of experience, investors increasingly rely on SSMAs to obtain financial information and manage their investments. Therefore, to ensure the robustness of the proxy variables, we eliminate the time trends of the original indicators. Moreover, similar to the SVI, the financial attention frequency indices also have significant weekly and monthly effects and thus we also eliminate the influence of these two factors. Finally, we obtain two indicators of investors' financial attention frequency in terms of SSMA usage frequency and online duration, namely FreqST and FreqOT. More details can be found in Appendix Table A.1. Compared with existing non-technical indicators, the measurement of investor attention in our research has the following characteristics. First, FreqST and FreqOT reflect investors' active access to financial information. However, investor attention based on a media coverage measurement reflects investors' passive or indirect attention or access to information (Fang and Peress, 2009; Engelberg and Parsons, 2011; Hillert et al., 2014). Second, we measure investors' active attention in terms of frequency instead of quantity. Although the SVI, which is based on the search volume measurement of Internet search engines such as Google, also implies investors' active access to information (Da et al., 2011, 2015; Zhang and Wang, 2015), it provides more information on the number of investors actively searching for specific information or the scope of information dissemination and diffusion. Hence, the indicators of investors' financial attention frequency in this study explain investors' active information acquisition frequency or the energy that they spend on this. To some extent, it also suggests the sensitivity of investors to market dynamics. Third, our measurement of investors' financial attention frequency covers most individual investors in the A-share market. The online account login data used by Sicherman et al. (2016) is similar to our data source, which they defined as financial attention for acquiring asset information. However, the data used by Sicherman et al. (2016) not only fail to describe the frequency of investors' information acquisition (i.e., average number and duration of daily uses per user), but also contain only approximately 100,000 individual investor accounts. The big data feature in our study confirms the effectiveness of our indicators. 2.4. Other major variables and descriptive statistics The main dependent variable used in this study is market trading activities. Therefore, we first obtain the natural logarithm of 6

Pacific-Basin Finance Journal 58 (2019) 101239

W. Cai and J. Lu

daily dollar trading volume (Volume) and turnover (Turnover) of the A-share market from January 1, 2016, to December 31, 2018. Some studies of the relationship between volume and price in the stock market find a significant positive correlation among market returns, volatility, and trading volume (Chan and Fong, 2006; Fleming and Kirby, 2011; Rossi and de Magistris, 2013). Therefore, following Chordia and Subrahmanyam (2004), we first control for Rett-1, the previous market returns of the Shanghai Securities Composite Index (SSCI). In addition, we control for previous intraday realized volatility RVt-1.7 Here, RVt-1 is the standard deviation of the return square of the SSCI every five minutes per trading day. Both Rett-1 and RVt-1 thus reflect the short-term market conditions. Additionally, Odean (1999) and Dorn et al. (2008) found that individual investors behave more similarly to momentum traders. Hence, we also adopt CRt-1,t-30 and SDt-1,t-30 to represent the cumulative market returns and volatility in the medium term, respectively. Specifically, CRt-1,t-30 represents the cumulative market returns over the past 30 trading days and SDt-1,t-30 is the standard deviation of the market returns of the SSCI over the past 30 trading days. The data for these variables are from the China Stock Market & Accounting Research (CSMAR) database. More details on the variable definitions and calculations can be found in Appendix Table A.1. Table 3 shows the descriptive statistics and correlation analysis of trading, financial attention frequency, and market conditions. In Panel A, the mean of Turnover is 1.19%, suggesting that trading in the A-share market is active. In Panel B, the correlation coefficients of FreqST with Volume and Turnover are 0.352 and 0.122 (significant at the 1% level), respectively and the correlation coefficient between FreqOT and Volume is also significantly positive at the 1% level. This preliminary result verifies the significant and positive correlation between overall market trading activity and investors' financial attention frequency. In addition, there is no significant correlation between financial attention frequency and Ret or CRt-1,t-30. Moreover, the correlation between investors' financial attention frequency and previous realized volatility (RV) is negative and relatively small. Meanwhile, SDt-1,t-30 has a significantly negative relationship with financial attention frequency at the 1% level. This finding implies that market return volatility has an impact on financial attention frequency. More importantly, financial attention frequency shows a possible ostrich effect in that investors tend to obtain financial information less frequently in periods of low market returns and high market volatility. 3. Ostrich effect in financial attention frequency Karlsson et al. (2009) suggested that information acquisition and processing increases the psychological impact of information (i.e., information's hedonic utility), which further affects investors' decisions to acquire information. Karlsson et al. (2009) and Sicherman et al. (2016) further confirmed the ostrich effect in acquiring information, using investor account login data. The financial attention frequency indices constructed in this study reflect the behavior of investors obtaining financial information through SSMAs, including not only market dynamics but also their holding of (or potential holding of) portfolios. Therefore, we first examine the possible ostrich effect in financial attention frequency. As Sicherman et al. (2016) noted, a variety of public signals, especially on market returns may affect investors' decisions to pay attention to the financial market as well as their portfolios. Hence, we first investigate the ostrich effect following market returns over both short and long time horizons following Sicherman et al. (2016). They also found that financial attention shows an ostrich effect in return volatility. Therefore, we consider the impacts of short- and long-term market return volatility on investors' financial attention frequency. Table 4 shows the regression results using Newey–West standard errors (Newey and West, 1987).8 The optimal lag order is based on the automatic selection principle proposed by Newey and West (1994), that is, the integral part of 4*(730/100)2/9. Therefore, we choose 6 as the maximum lag order in all the regressions. Panels A and B report the ostrich effects of FreqST and FreqOT, respectively. We also control for the weekday effects with day-of-the-week dummy variables in all the regressions. M1 in Panel A regresses daily financial attention frequency measured by FreqSTt on previous market returns Rett-1 and the dummies NegativeRet_Dumt-i,t-j representing longer-term negative cumulative returns over the past i-th to j-th trading days. The ostrich effect predicts a positive coefficient of Rett-1 and negative coefficients of the down-SSCI dummy variables. In M1, the coefficient of Rett-1 is 0.036 at the 1% significance level and the down-SSCI dummy variables except NegativeRet_Dumt-16,t-30 have significantly negative impacts on FreqSTt. These findings support the ostrich effect hypothesis, suggesting that investors' frequency of using SSMAs increases with previous market returns and declines during down-SSCI periods. Moreover, the longer-term down-SSCI dummies have greater negative impacts on FreqST, implying that the ostrich effect induces long-term path dependence on the market index in investors' financial attention frequency. In M2 to M6 of Panel A, we further test the ostrich effect by gradually adding the short-term to long-term market return volatility indices.9 Specifically, we adopt previous daily realized volatility RVt-1, which is calculated using the high-frequency data of the SSCI every five minutes, and the dummy variables HighVol_Dumt-1,t-j representing high volatility periods. HighVol_Dumt-1,t-j takes the value 7 Andersen and Bollerslev (1998) confirmed that past foreign exchange trading volatility can be estimated using the sum of the squares of changes (“realized” volatility) in the data sampled every five minutes in one day. Andersen et al. (2001a), Andersen et al. (2001b), and Barndorff-Nielsen and Shephard (2002) also showed that realized volatility can accurately measure the historical return volatility of assets. 8 Since our tests are limited to certain time horizons, we run the regressions using Newey–West (1987) standard errors with 6 as the maximum lag order (Newey and West, 1994) considering both potential heteroskedasticity and autocorrelation. 9 We did not use the VIX to measure volatility following Sicherman et al. (2016). First, option trading in the A-share market is not as active as in the United States and short index options are heavily restricted by the government. Second, in June 26, 2015, China Securities Index Co., Ltd., released the only volatility index of the A-share market (Shanghai Securities 50ETF VIX). However, it stopped publishing the index in February 2018.

7

Pacific-Basin Finance Journal 58 (2019) 101239

W. Cai and J. Lu

Table 3 Descriptive statistics and correlation analysis. Panel A: Descriptive statistics Variable

N

Mean

SD

Min

Median

Max

Volume Turnover (%) FreqST FreqOT Ret (%) CRt-1,t-30 RV SDt-1,t-30

731 731 731 731 731 731 731 731

26.79 1.19 0.00 0.00 0.09 −0.01 0.87 0.92

0.27 0.36 0.61 0.09 1.04 0.06 0.56 0.53

25.96 0.56 −1.39 −0.13 −6.80 −0.30 0.26 0.30

26.81 1.15 −0.10 −0.01 0.13 0.01 0.72 0.72

27.53 2.73 1.65 0.27 4.65 0.12 4.06 2.82

Panel B: Correlation analysis Variable

Volume

Turnover

FreqST

FreqOT

Ret

CRt-1,t-30

RV

SDt-1,t-30

Volume Turnover (%) FreqST FreqOT Ret (%) CRt-1,t-30 RV SDt-1,t-30

1.000 0.897*** 0.352*** 0.167*** 0.050 0.319*** 0.045 −0.055

1.000 0.122*** 0.047 0.019 0.176*** 0.246*** 0.266***

1.000 0.724*** 0.030 0.046 −0.110*** −0.258***

1.000 0.030 0.011 −0.019 −0.136***

1.000 −0.070* −0.184*** −0.004

1.000 −0.412*** −0.577***

1.000 0.547***

1.000

This table shows the descriptive statistics and correlation analysis of the main variables. Panel A shows the descriptive statistics of the variables. Panel B reports their Pearson correlation coefficients. Volume is the natural logarithm of the daily trading volume of the A-share market. Turnover is the daily turnover of market value. FreqST and FreqOT represent daily investors' financial attention frequency. Ret is the daily returns of the SSCI. CRt-1,t-30 represents the cumulative market returns over the past 30 trading days. RV is daily realized volatility calculated using the high-frequency data of the SSCI every five minutes intraday. SDt-1,t-30 represents volatility over the past 30 trading days. *** and * denote statistical significance at the 1%, and 10% levels, respectively.

of 1 if the standard deviation of the market daily returns of the SSCI over the past j trading days (SDt-1,t-j) is higher than its median; otherwise, it takes 0.10 In M2, the coefficient of RVt-1 is −0.124 at the 5% significance level, suggesting that the frequency of using SSMAs decreases with an increase in realized volatility RVt-1. In M3 to M6, the high-volatility dummy variables HighVol_Dumt-1,t-j all have significantly negative coefficients at the 1% level, indicating that the average daily frequency of using SSMAs drops more following high-volatility periods. This finding further confirms the ostrich effect in financial attention frequency for volatility. Panel B of Table 4 further examines the ostrich effect in financial attention frequency measured by online duration. The regression results are consistent with those in Panel A. Overall, the results in Table 4 thus support the ostrich effect hypotheses of financial attention for both returns and volatility. That is, in periods of low cumulative returns and high volatility, investors obtain financial information less frequently through SSMAs. 4. Financial attention frequency and trading activity In this section, we focus on how investors' financial attention frequency affects market trading activity. Fig. 1 illustrates how financial attention frequency and trading volume change over time. Panels A and B show the daily distribution and 20-day moving average of Trading Volume compared with FreqST and FreqOT, respectively. Panel A shows that Trading Volume (shaded parts) and financial attention frequency (FreqST; blue line) fluctuate in a relatively consistent way in the short term. For example, in July 2016, 2017, and 2018, both variables show similar serrated fluctuations and also reach short-term lows in January 2017, February 2018, and September 2018. However, in the long run, FreqST has greater volatility. It suddenly increases to nearly 1.5 in April 2017, partly because the A-share market reached a new high in April 2017 after the stock market crash from July 2015 to January 2016. Conversely, compared with trading behavior, investors' use of SSMAs to obtain market information has greater flexibility. Not only is the use of SSMAs almost free, but it is more emotionally driven. In addition, the moving average of Trading Volume (orange line) and 20-day moving average of FreqST (red line) fluctuate more consistently, further indicating a relationship between investors' financial attention frequency and overall trading volume in the market. Panel B plots similar distributions for Trading Volume and FreqOT. Here, the fluctuation of FreqOT (blue line) clusters in the short term but jumps the long term, indicating that the time investors spend on information acquisition is relatively stable in the short term but can be stimulated by external factors in the long term. Overall, Fig. 1 preliminarily shows similar levels of volatility between investors' financial attention frequency and market transactions, which is clearer in the short term. 10

The regression results are consistent when we use the continuous variable SDt-1,t-j directly in the regressions. 8

Pacific-Basin Finance Journal 58 (2019) 101239

W. Cai and J. Lu

Table 4 Ostrich effects of financial attention frequency. Panel A: Ostrich Effects of FreqST Variables

Rett-1 NegativeRet_Dumt-2,t-15 NegativeRet_Dumt-16,t-30 NegativeRet_Dumt-31,t-50 NegativeRet_Dumt-51,t-250 RVt-1

FreqSTt M1

M2

M3

M4

M5

M6

0.036*** (2.64) −0.150* (−1.69) −0.109 (−1.27) −0.254*** (−3.01) −0.447*** (−4.39)

0.024 (1.64) −0.144* (−1.66) −0.122 (−1.42) −0.265*** (−3.13) −0.414*** (−4.07) −0.124** (−2.50)

0.028** (2.14) −0.093 (−1.21) −0.081 (−1.06) −0.277*** (−3.56) −0.237** (−2.35)

0.029** (2.12) −0.118 (−1.51) −0.054 (−0.69) −0.266*** (−3.39) −0.205** (−2.19)

0.026* (1.95) −0.081 (−1.02) −0.082 (−1.06) −0.244*** (−3.10) −0.212** (−2.25)

0.038*** (2.69) −0.206*** (−3.07) −0.151** (−2.21) −0.307*** (−4.52) −0.312*** (−2.64)

HighVol_Dumt-1,t-15 HighVol_Dumt-1,t-30

−0.447*** (−5.15)

HighVol_Dumt-1,t-50

−0.468*** (−5.11)

HighVol_Dumt-1,t-250 Weekday-calendar Constant Observations F-statistics

Yes 0.316*** (2.80) 730 5.387

Yes 0.427*** (3.08) 730 4.897

Yes 0.468*** (4.06) 730 6.708

Yes 0.471*** (3.97) 730 5.847

−0.466*** (−5.08) Yes 0.456*** (3.90) 730 6.002

−0.635*** (−8.15) Yes 0.682*** (7.26) 730 11.03

Panel B: The Ostrich Effects of FreqOT Variables

Rett-1 NegativeRet_Dumt-2,t-15 NegativeRet_Dumt-16,t-30 NegativeRet_Dumt-31,t-50 NegativeRet_Dumt-51,t-250 RVt-1

FreqOTt M1

M2

M3

M4

M5

M6

0.005** (2.43) −0.026* (−1.78) −0.021 (−1.42) −0.026* (−1.91) −0.030* (−1.80)

0.005** (2.41) −0.025* (−1.78) −0.021 (−1.42) −0.027* (−1.92) −0.029* (−1.79) −0.002 (−0.26)

0.004* (1.90) −0.017 (−1.30) −0.017 (−1.27) −0.030** (−2.30) 0.001 (0.08)

0.004* (1.89) −0.021 (−1.60) −0.013 (−0.98) −0.028** (−2.17) 0.004 (0.27)

0.004* (1.78) −0.016 (−1.21) −0.017 (−1.26) −0.025* (−1.93) 0.004 (0.25)

0.006** (2.31) −0.032** (−2.45) −0.026* (−1.94) −0.032*** (−2.60) −0.015 (−0.79)

HighVol_Dumt-1,t-15 HighVol_Dumt-1,t-30

−0.065*** (−4.38)

HighVol_Dumt-1,t-50

−0.065*** (−4.08)

HighVol_Dumt-1,t-250 Weekday-calendar Constant Observations F-statistics

Yes 0.039** (2.49) 730 3.179

Yes 0.040** (2.18) 730 3.115

Yes 0.061*** (3.83) 730 4.318

Yes 0.060*** (3.74) 730 3.994

−0.066*** (−4.12) Yes 0.058*** (3.70) 730 4.139

−0.069*** (−5.05) Yes 0.079*** (5.04) 730 5.157

This table shows the ostrich effects of financial attention frequency. Panels A and B report the regression results of FreqST and FreqOT, which represent daily investors' financial attention frequency. Rett-1 is the previous daily returns of the SSCI. NegativeRet_Dumt-i,t-j represents the negative SSCI cumulative return dummies over the past i-th to j-th trading days. RVt-1 is the previous daily realized volatility calculated using the highfrequency data of SSCI every five minutes intraday. HighVol_Dumt-1,t-j represent the high market volatility dummies for the previous j trading days. The weekday effects are controlled for in all the regressions. The t-statistics presented in parentheses below each estimated coefficient are reported based on regressions using Newey–West standard errors. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 9

Pacific-Basin Finance Journal 58 (2019) 101239

W. Cai and J. Lu

1000

1.5

900

1

800 0.5

700 600

0

500

-0.5

400 -1

300 200 1/4/2016

-1.5 7/4/2016

1/4/2017

7/4/2017

1/4/2018

7/4/2018

Trading Volume

Trading Volume_MA20

FreqST

FreqST_MA20

Panel A: Trading Volume and Financial Attention Frequency Measured By FreqST 1000

0.3

900

0.25

800

0.2 0.15

700

0.1

600

0.05

500

0

400

-0.05

300

-0.1

200 1/4/2016

-0.15 7/4/2016

1/4/2017

7/4/2017

1/4/2018

7/4/2018

Trading Volume

Trading Volume_MA20

FreqOT

FreqOT_MA20

Panel B: Trading Volume and Financial Attention Frequency Measured By FreqOT Fig. 1. Daily trading volume and financial attention frequency in the A-share market. This figure shows daily trading volume (left axis) and investors' financial attention frequency (right axis) from January 1, 2016 to December 31, 2018. Panels A and B depict the daily distribution and moving average of Trading Volume compared with FreqST and FreqOT, respectively. The shaded parts in both panels reflect Trading Volume (Unit: Billion RMB) in the A-share market and the orange lines depict the 20-day moving average value, Trading Volume_MA20. The blue lines in Panels A and B plot the daily distribution of FreqST and FreqOT, respectively and the red lines show their 20-day moving average values, FreqST_MA20 and FreqOT_MA20. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

To examine whether the frequency of investors' access to financial information through SSMAs affects their trading activities, we next adopt a series of regression analyses. Fig. 1 shows that FreqST and FreqOT have similar volatility characteristics to trading volume. In other words, there is a consistent long-term trend between the volatility of investors' financial attention frequency and trading volume. Therefore, to eliminate the impact of this trend, we first use the abnormal measures of financial attention frequency as well as trading to conduct regressions. Specifically, we use the current value of the variables minus their moving average over the past 20 trading days to calculate these abnormal measures. This approach not only removes much of the intermittent variation in these variables, but also is consistent with Da et al. (2011), who constructed their abnormal SVI from Google Trends data. Table 5 shows regression results of abnormal investors' financial attention frequency regarding the overall trading activity of the A-share market. We again run the regressions using Newey–West standard errors with 6 as the maximum lag order. We control for previous market returns (Rett-1) and volatility (RVt-1), and the medium-term cumulative market returns CRt-1,t-30 as well as past volatility SDt-1,t-30 over the past 30 trading days. Further, the weekday effects with day-of-the-week dummy variables are also controlled for in all the regressions. In Table 5, M1 to M4 show the influences of abnormal financial attention frequency on the abnormal value of the natural 10

Pacific-Basin Finance Journal 58 (2019) 101239

W. Cai and J. Lu

Table 5 Impacts of financial attention frequency on trading activity. Variables

ABVolumet M1

ABFreqSTt-1 ABFreqOTt-1

0.146*** (3.92)

Rett-1

ABTurnovert M2

0.656*** (2.66)

RVt-1 CRt-1,t-30 SDt-1,t-30 Weekday-calendar Constant Observations F-statistics

Yes 0.000 (0.01) 710 5.426

Yes 0.001 (0.04) 710 3.585

M3 0.152*** (4.28) 0.049*** (6.01) −0.034* (−1.68) 0.757*** (3.30) 0.067** (2.05) Yes −0.029 (−0.91) 710 11.22

M4

0.669*** (2.90) 0.052*** (6.22) −0.027 (−1.20) 0.702*** (3.05) 0.059* (1.77) Yes −0.028 (−0.86) 710 8.657

M5 0.134*** (3.50)

Yes 0.000 (0.01) 710 4.409

M6

0.712*** (2.64)

Yes 0.001 (0.03) 710 3.275

M7 0.135*** (3.76) 0.059*** (6.10) −0.023 (−0.93) 0.614* (1.90) 0.066 (1.54) Yes −0.039 (−1.00) 710 8.468

M8

0.708*** (2.77) 0.061*** (6.36) −0.018 (−0.69) 0.569* (1.77) 0.060 (1.38) Yes −0.039 (−0.97) 710 7.083

This table reports the regression results of investors' abnormal financial attention frequency on the overall trading activity of the A-share market. ABVolume is the abnormal value of the natural logarithm of daily trading volume. ABTurnover is abnormal daily turnover. ABFreqST and ABFreqOT represent daily investors' abnormal financial attention frequency. We use the current value of the variables minus their average over the past 20 trading days to calculate these abnormal variables. Ret is the daily returns of the SSCI. RV is daily realized volatility calculated using the highfrequency data of the SSCI every five minutes intraday. CRt-1,t-30 represents the cumulative market returns over the past 30 trading days. SDt-1,t-30 represents volatility over the past 30 trading days. The weekday effects are controlled for in all the regressions. The t-statistics presented in parentheses below each estimated coefficient are reported based on regressions using Newey–West standard errors. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.

logarithm of daily trading volume Volumet. In M1, the coefficient of ABFreqSTt-1 is 0.146 at the 1% significance level, suggesting that an abnormal frequency of the use of SSMAs ABFreqSTt-1 significantly stimulates overall market trading volume. M2 shows the impact of ABFreqOTt-1 on abnormal trading volume. Similarly, the coefficient of ABFreqOTt-1 in M2 is also significantly positive (0.656) at the 1% level. Moreover, even if we control for previous market returns Rett-1, realized volatility RVt-1, cumulative market returns CRt-1,t30, and past volatility SDt-1,t-30 in M3 and M4, the coefficients of ABFreqSTt-1 and ABFreqOTt-1 remain positive at the 1% significance level. This finding indicates that for each additional unit of ABFreqSTt-1, the abnormal trading volume increases by 16.42% (e0.152–1), whereas it increases by 6.92% (e0.0669–1) for every 0.1-unit increase in ABFreqOTt-1. M5 to M8 show the influence of abnormal financial attention frequency on abnormal daily market turnover (%). ABFreqSTt-1 and ABFreqOTt-1 also have significantly positive coefficients. In particular, in M7 and M8, each additional unit increase in ABFreqSTt-1 and a 0.1-unit increase in ABFreqOTt-1 lead to increases of 14.45% (e0.135–1) and 7.34% (e0.0708–1) in ABTurnovert, respectively. Table 5 preliminarily proves that investors' financial attention frequency stimulates their trading activity. If we understand the financial attention frequency index from the perspective of investor attention, it reflects the average number of times investors pay attention to the market or the time cost they spend obtaining market information every day. Therefore, this finding also validates the driving force of investor attention on trading (Barber and Odean, 2008) from the perspective of attention frequency. To verify the robustness of the financial attention frequency effect, we further conduct a series of robustness tests. 4.1. Other market factors In the previous regressions, we mainly controlled for previous market returns Rett-1, intraday realized volatility RVt-1, cumulative returns CRt-1,t-30, and past volatility SDt-1,t-30. Although these factors reflect the main market information, some variables might still be missing from these models. When these missing variables are related to investors' financial attention frequency, they can lead to inconsistencies in the estimation results, which can seriously affect their credibility. Therefore, we add further control variables on other market factors to ensure the robustness of our results. First, Barber and Odean (2001) argued that the Internet provides vast amounts of information that reinforces investors' prior beliefs about asset values and causes them to become overconfident when choosing stocks and other securities, leading to more active trading. According to Barber and Odean (2001), acquiring a large amount of information through SSMAs can aggravate investors' overconfidence. Gervais and Odean (2001) showed that the past success of market information judgments can cause investors to become overconfident, thus encouraging them to overtrade. This results in abnormal increases in trading volume at the market level and raises stock prices. Deaves et al. (2010) argued that the behavior caused by investor overconfidence can be expressed by the relationship between past returns and trading volume. Chuang and Lee (2006) divided changes in market volume into those related to investor confidence and those unrelated to investor confidence. Therefore, we measure and control for daily investor overconfidence (OVER) using regression analysis following Chuang and Lee (2006). 11

Pacific-Basin Finance Journal 58 (2019) 101239

W. Cai and J. Lu

In addition, investors' financial attention frequency, as measured in this study, actually reflects investor attention, which also drives investors' trading. Intuitively, when they pay more attention to the financial market, investors obtain information more frequently. Thus, we also control for the impact of investor attention. Specifically, we adopt the daily Baidu search volume index11 (BSVI) of the three major indices12 in the Chinese A-share market as a proxy for investor market attention following Da et al. (2011) and Zhang and Wang (2015). Shiller (2000) suggested that the news media play an important role in the formation and promotion of market trends and attention. Other studies have also confirmed such a media-driven effect hypothesis (Tetlock, 2007; Fang and Peress, 2009; Garcia, 2013). Hence, we also control for media coverage in the market, MEDIA, which reflects the number of news reports on the three major indices of the A-share market reported by major Internet media. In summary, BSVI measures investors' active acquisition of information, while MEDIA reflects investors' passive access to information. Moreover, Da et al. (2011) and Joseph et al. (2011) found that online searches can be an effective proxy for investor sentiment. When investor sentiment is high, investors more actively obtain market information, leading to an increase in trading volume. Baker and Wurgler (2006) constructed a widely accepted monthly investor sentiment index. Unfortunately, their sentiment index cannot measure the daily change in investor sentiment. Brown and Cliff (2004) suggested that the advance/decline line can describe daily investor sentiment and Wang et al. (2006) suggested that the ARMS index, created by Richard Arms, also reflects daily sentiment. Therefore, we adopt the advance/decline ratio (ADR) and ARMS as proxy indicators of daily investor sentiment. ADR is a positive indicator of investor sentiment, whereas ARMS represents its reverse indicator. Appendix Table A.1 provides the definitions, calculations, and sources of the variables. Table 6 reports the regression results after controlling for these other market factors. We again run the regressions using Newey–West standard errors with 6 as the maximum lag order. All the control variables are added with their lag terms. As before, the weekday effects with day-of-the-week dummy variables are controlled for in all the regressions. The coefficients of ABFreqSTt-1 and ABFreqOTt-1 in M1 and M2 (0.138 and 0.619, respectively) are significant at the 1% level, indicating that even when factors such as investor overconfidence, attention, and sentiment are controlled for, investors' abnormal financial attention frequency also significantly increases trading activity at the market level. Similarly, in M3 and M4, ABFreqSTt-1 and ABFreqOTt-1 also have significantly positive coefficients of 0.118 and 0.654 at the 1% level, respectively, further indicating that attention frequency has a significant stimulating effect on trading. In addition, OVERt-1 shows a positive relationship with trading activity, which is consistent with the findings of Gervais and Odean (2001). Moreover, the coefficients of ADR, a proxy for investor sentiment, are significantly positive at the 1% level in M1 to M4, whereas the coefficients of ARMS, the reverse index, are significantly negative at the 1% level, illustrating the role of investor sentiment in promoting market transactions (Da et al., 2011; Joseph et al., 2011). 4.2. Direct measure of investors' financial attention frequency To better understand the driving forces behind trading, we also provide evidence based on the direct measure of daily financial attention frequency. Table 7 shows the regression results with 6 as the maximum lag order. To save space, the control variables reflecting other market factors shown in Table 6 are not reported. As before, the weekday effects with day-of-the-week dummy variables are controlled for in all the regressions. In Table 7, M1 and M2 test the influence of financial attention frequency on trading volume, while M3 and M4 test the influence on the turnover rate. In M1 and M2, the coefficients of FreqSTt-1 and FreqOTt-1 are 0.185 and 0.613, respectively, significant at the 1% level. Similarly, FreqSTt-1 and FreqOTt-1 also have significantly positive impacts on the turnover rate in M3 and M4. Table 7 confirms the driving forces of financial attention frequency on trading. 4.3. Financial attention frequency based on apps without trading systems Because some of the SSMAs used thus far support both financial information services and securities trading systems, the financial attention frequency indices may be mixed with the noise of real trading behavior, which naturally increases market trading activity. To eliminate the impact of the actual transaction process, we thus remeasure financial attention frequency based on SSMAs without trading systems. By downloading all 307 SSMAs, we first manually identify whether they support securities trading systems. We then remove the 168 SSMAs that support real stock trading systems from the sample, leaving 139 SSMAs13 with 57,779 daily observations. Finally, we remeasure financial attention frequency using these data consistent with Eq. (1) and Eq. (2). Table 8 reports the regression results using Newey–West standard errors with 6 as the maximum lag order. Again, the weekday effects with day-of-the-week dummy variables are controlled for in all the regressions. 11 Because Google withdrew from Mainland China in March 2010, we built the investor attention index using search volumes from the Baidu search engine following Zhang and Wang (2015). The Baidu search engine is the largest in China. As of December 2016, its brand penetration in Mainland China reached 82.9%. 12 The three main indices of China's stock market are the SSCI (code: 1A0001), Shenzhen Securities Component Index (code: 399001), and Growth Enterprise Market Index (code: 390006). 13 These apps that do not support securities trading systems are mainly used to gather financial information, simulate transactions, and use the open-account services of SCs.

12

Pacific-Basin Finance Journal 58 (2019) 101239

W. Cai and J. Lu

Table 6 Controlling for other market factors. Variables

ABVolumet M1

ABFreqSTt-1 ABFreqOTt-1 Rett-1 RVt-1 CRt-1,t-30 SDt-1,t-30 OVERt-1 BSVIt-1 MEDIAt-1 ARMSt-1 ADRt-1 Weekday-calendar Constant Observations F-statistics

0.138*** (3.91) 0.018 (1.64) −0.011 (−0.46) 0.651*** (2.87) 0.055 (1.51) 0.681*** (2.69) −0.037 (−0.57) −0.006 (−0.81) −0.068*** (−3.11) 0.054*** (3.49) Yes 0.432 (0.58) 710 11.43

ABTurnovert M2

0.619*** (2.85) 0.015 (1.38) −0.006 (−0.23) 0.585*** (2.64) 0.041 (1.12) 0.768*** (3.02) −0.032 (−0.47) −0.003 (−0.39) −0.074*** (−3.48) 0.062*** (4.12) Yes 0.366 (0.47) 710 9.664

M3 0.118*** (3.24) 0.016 (1.21) 0.006 (0.19) 0.516* (1.66) 0.056 (1.25) 0.646** (1.99) −0.058 (−0.70) −0.009 (−0.94) −0.081*** (−2.82) 0.076*** (3.49) Yes 0.671 (0.71) 710 7.754

M4

0.654*** (2.74) 0.013 (1.04) 0.010 (0.31) 0.466 (1.52) 0.046 (1.02) 0.720** (2.22) −0.056 (−0.66) −0.006 (−0.66) −0.085*** (−3.11) 0.082*** (3.90) Yes 0.637 (0.66) 710 6.991

This table reports the regression results of investors' abnormal financial attention frequency on the abnormal overall trading activity of the A-share market when controlling for other market factors. ABVolume is the abnormal value of the natural logarithm of daily trading volume. ABTurnover is abnormal daily turnover. ABFreqST and ABFreqOT represent daily investors' abnormal financial attention frequency. Ret is the daily returns of the SSCI. RV is daily realized volatility calculated using the high-frequency data of the SSCI every five minutes intraday. CRt-1,t-30 represents the cumulative market returns over the past 30 trading days. SDt-1,t-30 represents volatility over the past 30 trading days. OVER reflects the degree of investor overconfidence. BSVI is the natural logarithm of the daily total Baidu search volume index about market indices. MEDIA is the media coverage index about market indices. ARMS is the Arms index representing the reverse indicators of investor sentiment. ADR is the advance/decline ratio, which also represents daily investor sentiment. The weekday effects are controlled for in all the regressions. The t-statistics presented in parentheses below each estimated coefficient are reported based on regressions using Newey–West standard errors. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. Table 7 Evidence based on daily investors' financial attention frequency. Variables

Volumet M1

FreqSTt-1 FreqOTt-1 Controlst-1 Weekday-calendar Constant Observations F-statistics

0.185*** (7.78) Yes Yes 20.168*** (23.23) 730 16.53

Turnovert M2

0.613*** (3.42) Yes Yes 20.947*** (18.60) 730 8.843

M3 0.168*** (6.82) Yes Yes −10.180*** (−10.47) 730 27.39

M4

0.607*** (3.51) Yes Yes −9.512*** (−8.39) 730 19.65

This table reports the effects of daily investors' financial attention frequency on overall trading activity in the A-share market. Volume is the natural logarithm of daily trading volume. Turnover is daily turnover. FreqSTt-1 and FreqOTt-1 represent investors' previous financial attention frequency. To save space, the control variables reflecting other market factors shown in Table 6 are not reported. The weekday effects are also controlled for in all the regressions. The t-statistics presented in parentheses below each estimated coefficient are reported based on regressions using Newey–West standard errors. *** denotes statistical significance at the 1% levels.

13

Pacific-Basin Finance Journal 58 (2019) 101239

W. Cai and J. Lu

Table 8 Financial attention frequency measured based on the SSMAs without trading systems. Panel A: The driving effects on abnormal trading activity Variables

ABVolumet M1

ABFreqSTt-1⊥ ABFreqOTt-1⊥ Controlst-1 Weekday-calendar Constant Observations F-statistics

0.093*** (3.62) Yes Yes 0.239 (0.31) 710 10.34

ABTurnovert M2

0.557*** (2.71) Yes Yes 0.409 (0.53) 710 9.574

M3 0.084*** (3.01) Yes Yes 0.505 (0.53) 710 7.322

M4

0.594*** (3.03) Yes Yes 0.683 (0.71) 710 7.287

Panel B: The driving effects on overall market trading activity Variables

Volumet M1

FreqSTt-1⊥ FreqOTt-1⊥ Controlst-1 Weekday-calendar Constant Observations F-statistics

0.145*** (7.12) Yes Yes 19.381*** (21.47) 730 20.34

Turnovert M2

0.473*** (3.47) Yes Yes 21.018*** (18.88) 730 9.410

M3 0.137*** (6.75) Yes Yes −10.955*** (−10.94) 730 33.07

M4

0.526*** (4.10) Yes Yes −9.499*** (−8.51) 730 21.40

This table reports the regression results based on financial attention frequency measured using the SSMAs without trading systems. (AB)Volume is the (abnormal) value of the natural logarithm of daily trading volume. (AB)Turnover is daily (abnormal) turnover. (AB)FreqST⊥ and (AB)FreqOT⊥ represent daily investors' (abnormal) financial attention frequency measured using the SSMAs without trading systems. To save space, the control variables reflecting other market factors shown in Table 6 are not reported. The weekday effects are also controlled for in all the regressions. The tstatistics presented in parentheses below each estimated coefficient are reported based on regressions using Newey–West standard errors. *** denotes statistical significance at the 1% levels.

Panel A in Table 8 reports the regression results for abnormal trading activity. Both the financial attention frequency indices measured using apps without trading systems, namely ABFreqSTt-1⊥ and ABFreqOTt-1⊥, have positive impacts on abnormal trading at the 1% significance level. Panel B in Table 8 shows the impacts on overall market trading. In M1 to M4, both these indices, FreqSTt-1⊥ and FreqOTt-1⊥, have significantly positive coefficients at the 1% level. This finding implies that the frequency of use and online duration of these SSMAs without trading systems also have a positive relationship with market trading activity. 4.4. Financial attention frequency on Sundays Investors' financial attention frequency, or the use of SSMAs, might be closely related to the most recent market events such as changes in trading volume. While trading volume is a reference point for investment decision making (Barber and Odean, 2008), Hou et al. (2009) found that it also reflects attention to a certain extent, and they thus used trading volume and the turnover rate as proxy variables for investor attention. Therefore, to reduce the endogenous impacts of trading volume and financial attention frequency, we test the Sunday effect.14 Specifically, we examine whether financial attention frequency of the last non-trading day before the first trading day of a week significantly increases trading activity on the next trading day. Table 9 shows the regression results using Newey–West standard errors. Here, the optimal lag order is 4 (the integral part of 4*(152/100)2/9) according to Newey and West (1994). Here, FreqSTSunday and FreqOTSunday represent daily investors' financial attention frequency on the last non-trading day before the first trading day of a week. M1 to M4 explore the impacts of FreqSTSunday and FreqOTSunday on trading volume. The coefficients of FreqSTSunday in M1 and FreqOTSunday in M2 are 0.093 and 1.684, respectively, significant at the 1% level. In addition, after controlling for the other market factors in M3 and M4, FreqSTSunday and FreqOTSunday again show significantly positive relationships with Volumet. M5 to M8 14

Since the last non-trading day before the first trading day of most trading weeks is Sunday, we call this the Sunday effect. 14

Pacific-Basin Finance Journal 58 (2019) 101239

W. Cai and J. Lu

Table 9 The driving force of financial attention frequency on Sundays. VARIABLES

Volumet M1

FreqSTSunday FreqOTSunday Controlst-1 Constant Observations F-statistics

0.093*** (3.86) No 26.497*** (284.50) 152 14.93

Turnovert M2

1.684*** (3.10) No 26.331*** (164.11) 152 9.620

M3 0.111*** (3.44) Yes 25.401*** (12.79) 151 13.09

M4

1.168*** (2.74) Yes 22.561*** (13.91) 151 8.303

M5 0.193*** (6.59) No 0.573*** (5.88) 152 43.37

M6

3.534*** (4.97) No 0.220 (1.14) 152 24.74

M7 0.146*** (3.81) Yes −3.486** (−2.23) 151 30.43

M8

1.679*** (3.55) Yes −7.087*** (−5.65) 151 19.00

This table reports the impact of investors' financial attention frequency on Sundays on the overall trading activity of the A-share market. Volume is the natural logarithm of the daily trading volume of the A-share market. Turnover is daily turnover. FreqSTSunday and FreqOTSunday represent daily investors' financial attention frequency of the last non-trading day before the first trading day of a week. To save space, the control variables reflecting other market factors shown in Table 6 are not reported. The t-statistics presented in parentheses below each estimated coefficient are reported based on regressions using Newey–West standard errors. *** and ** denote statistical significance at the 1% and 5% Andersen et al., 2001a, 2001b; Andersen and Bollerslev, 1998; Baker and Wurgler, 2006; Bank, M et al., 2011; Barber et al., 2009; Barber and Odean, 2001, 2008; Barndorff-Nielsen and Shephard, 2002; Brown et al., 2015; Brown and Cliff, 2004; Brunnermeier and Parker, 2005; Caplin and Leahy, 2001; Chan and Fong, 2006; Chordia and Subrahmanyam, 2004; Chuang and Lee, 2006; Da et al., 2011, 2014, 2015; Deaves et al., 2010; Dimpfl and Jank, 2016; Dorn et al., 2008; Engelberg and Parsons, 2011; Fang and Peress, 2009; Fleming and Kirby, 2011; Garcia, 2013; Gervais and Odean, 2001; Hillert et al., 2014; Hirshleifer et al., 2002; Hou et al., 2009; Huang and Liu, 2007; Huddart et al., 2009; Hvidkjaer, 2008; Joseph et al., 2011; Kahneman, 1973; Karlsson et al., 2009; Lee and Radhakrishna, 2000; Li and Yu, 2012; Loewenstein, 1987; Moore, 2013; Newey and West, 1987, 1994; Odean, 1999; Patel, 2014; Peng and Xiong, 2006; Rossi and de Magistris, 2013; Seasholes and Wu, 2007; Shiller, 2000; Sicherman et al., 2016; Sims, 2003, 2006; Tetlock, 2007; Vlastakis and Markellos, 2012; Wang et al., 2006; Yuan, 2015; Zhang and Wang, 2015 levels, respectively.

investigate the effects on Turnovert. M5 and M6 show that the coefficients of FreqSTSunday and FreqOTSunday (0.193 and 3.534, respectively) are significant at the 1% level; in M7 and M8, similar effects are still significant. In summary, by investigating the extent to which investors' financial attention frequency on non-trading days' affects trading on future trading days, the empirical results in Table 9 validate the driving force behind financial attention frequency on trading, excluding the influence of current trading volume. 4.5. IV estimation In the previous subsections, we adopted Sunday's financial attention frequency as well as financial attention frequency without trading systems to reduce the impact of endogeneity. However, these two methods caused us to lose a number of sample observations. Moreover, we could not exclude the possibility that financial attention frequency on Sundays or without trading systems is not influenced by trading on the previous trading day. To further mitigate the impact of endogeneity, we use an IV estimation. Specifically, we adopt data on users' daily behavior on WeChat,15 the most popular social networking service, and TaoBao,16 the most popular mobile shopping site in Mainland China, as the two IVs. From the EVTS database, we obtain the average number of users using the WeChat and TaoBao apps every day from January 1, 2016, to December 31, 2018, excluding the weekly and monthly effects and time trends. Then, we attain daily user activity, WeChat_Activity and TaoBao_Activity, which represent the trend of smartphone usage and popularity that influences the use of SSMAs. However, the tens of millions of securities investors are insignificant compared with the nearly one billion active users of Taobao and WeChat, implying that the use of SSMAs does not affect the trend of mobile apps represented by WeChat_Activity and TaoBao_Activity. Therefore, these two IVs are exogenous relative to the capital market. Table 10 shows the 2SLS regression results. We use a heteroskedasticity- and autocorrelation-consistent (HAC) weighting matrix with 6 lags as the maximum lag order. All the regression models passed weak identification tests and overidentification tests. We also control for the other market factors as before as well as weekday effects with day-of-the-week dummy variables. Panels A and B in Table 10 show the first-stage and second-stage 2SLS regression results, respectively. In Panel A, both WeChat_Activity and TabBao_Activity are positively related to ABFreqST and ABFreqOT. This means that the overall popularity of smartphones affects investors' use of SSMAs. Moreover, as presented in Panel B, both ABFreqSTt-1 and ABFreqOTt-1 show significant positive influences on abnormal market trading, which is consistent with our earlier results. To summarize, the results in this section show that investors' financial attention frequency, as measured by the use of SSMAs, 15

The WeChat mobile app software was developed by Tencent Holdings Co., Ltd., the largest Internet company in China. It is the most popular instant messaging mobile app in the mainland market with over one billion users. It supports sending voice messages, videos, pictures, and text and users can chat in groups. It consumes only a small amount of communication traffic and is suitable for most smartphones. 16 TaoBao was founded by Alibaba in May 2003. It is a popular online shopping and retail platform in China, with nearly 500 million registered users, > 60 million visitors per day, and > 800 million online goods sales per day, selling an average of 48,000 goods per minute. With the penetration and popularization of the mobile Internet, its transactions from mobile devices account for > 90% of total sales. 15

Pacific-Basin Finance Journal 58 (2019) 101239

W. Cai and J. Lu

Table 10 Results using the IV method. Panel A: The first stage regression results of 2SLS VARIABLES

WeChat_Activityt TaoBao_Activityt Controlst Weekday-calendar Constant Observations F-statistics

ABFreqSTt

ABFreqOTt

M1

M2

0.043*** (3.06) 0.182*** (3.36) Yes Yes −0.697 (−0.71) 710 4.91

0.004* (1.67) 0.029*** (4.00) Yes Yes −0.143 (−1.31) 710 2.16

Panel B: The second stage regression results of 2SLS Variables

ABVolumet

ABTurnovert

M1 ABFreqSTt-1 ABFreqOTt-1 Controlst-1 Weekday-calendar Constant Observations chi2-statistics

M2

0.271** (2.54)

2.346** (2.34) Yes Yes 0.688 (1.14) 710 117.8

Yes Yes 0.607 (1.08) 710 133.8

M3 0.212* (1.75) Yes Yes 0.794 (1.08) 710 89.32

M4

1.943* (1.72) Yes Yes 0.877 (1.15) 710 86.84

This table shows the results using the IV method. Panels A and B report the first- and second-stage 2SLS regression results, respectively. We take WeChat_Activity and TaoBao_Activity, which represent the daily frequency of use of the WeChat and TaoBao apps, respectively, as the IVs. ABFreqST and ABFreqOT represent daily investors' abnormal financial attention frequency. ABVolume is the abnormal value of the natural logarithm of the daily trading volume of the A-share market. ABTurnover is daily abnormal turnover. To save space, the control variables reflecting other market factors shown in Table 6 are not reported. The weekday effects are also controlled for in all the regressions. The t-statistics in Panel A and z-statistics in Panel B presented in parentheses below each estimated coefficient are reported based on HAC standard errors. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.

stimulates them to trade more actively. In addition, such an effect remains unchanged when controlling for other market factors, examining the effects based on a direct measure of financial attention frequency, measuring financial attention frequency on Sundays or with SSMAs without trading systems, and adopting WeChat_Activity and TaoBao_Activity as the IVs in 2SLS regressions. 5. Financial attention frequency and individual investors' net trading The indices used thus far mainly reflect the frequency of individual investors' access to market information. Naturally, we are also interested in whether financial attention frequency affects the trading behaviors of individual investors. Lee and Radhakrishna (2000) suggested that trading volume can be used to distinguish between individual and institutional investors. Barber et al. (2009), Hvidkjaer (2008), and Yuan (2015) found that a small capital flow represents the overall order flow of individual investors. Therefore, to examine the impact of financial attention frequency on individuals' trading behaviors, we obtain daily trading data classified by order size from the Choice database of Eastmoney Co., Ltd.17 Trades of < 40,000 yuan (RMB) are defined as small trades (SmaOrder), trades between 40,000 yuan (RMB) and 200,000 yuan (RMB) are defined as medium trades (MidOrder), trades between 200,000 yuan (RMB) and 1 million yuan (RMB) are defined as large trades (LrgOrder), and those of > 1 million yuan (RMB) are defined as super-large trades (SlrgOrder). SmaOrder and MidOrder are more likely to come from individual investors, while LrgOrder and SlrgOrder are more likely to be initiated by institutional investors. Appendix Table A.1 presents the definitions, calculations, and sources of these variables. Table 11 reports the impact of (abnormal) financial attention frequency on (abnormal) net purchases by order size. The regression results are reported using Newey–West standard errors with 6 as the maximum lag order. Again, we control for other market factors 17 East Money (东方财富) Co., Ltd. is a leading third-party financial services company in Mainland China. Its self-developed Choice database docks the trading systems of the Shanghai Stock Exchange and Shenzhen Stock Exchange.

16

Pacific-Basin Finance Journal 58 (2019) 101239

W. Cai and J. Lu

Table 11 Financial attention frequency and investors' net buying. Panel A: The relationships between abnormal financial attention frequency and investors' net buying Variables

ABSmaOrdert M1

ABFreqSTt-1 ABFreqOTt-1 Controlst-1 Weekday-calendar Constant Observations F-statistics

0.401*** (3.04) Yes Yes 5.235* (1.96) 710 6.533

ABMidOrdert M2

1.739* (1.95) Yes Yes 5.032* (1.88) 710 4.566

M3 0.104*** (2.96) Yes Yes 0.929 (1.28) 710 6.242

ABLrgOrdert M4

0.589** (2.33) Yes Yes 0.902 (1.25) 710 5.682

M5

ABSLrgOrdert M6

−0.209** (−2.57) Yes Yes −3.081* (−1.91) 710 5.788

M7

−0.961* (−1.82) Yes Yes −2.985* (−1.85) 710 4.193

M8

−0.296*** (−3.44) Yes Yes −3.083* (−1.73) 710 6.575

−1.367** (−2.11) Yes Yes −2.949* (−1.66) 710 4.944

Panel B: The driving effects of financial attention frequency on investors' net buying Variables

SmaOrdert M1

FreqSTt-1 FreqOTt-1 Controlst-1 Weekday-calendar Constant Observations F-statistics

0.195*** (2.78) Yes Yes −5.756** (−2.21) 730 3.505

MidOrdert M2

0.470 (0.92) Yes Yes −4.782* (−1.83) 730 2.797

M3 0.078*** (3.62) Yes Yes −2.404*** (−3.18) 730 6.731

LrgOrdert M4

0.432*** (2.93) Yes Yes −2.228*** (−3.01) 730 6.171

M5 −0.138*** (−3.40) Yes Yes 6.268*** (3.99) 730 5.200

SLrgOrdert M6

−0.373 (−1.29) Yes Yes 5.610*** (3.54) 730 4.328

M7 −0.134*** (−2.65) Yes Yes 1.892 (1.07) 730 3.460

M8

−0.530 (−1.45) Yes Yes 1.400 (0.80) 730 2.752

This table shows the impacts of investors' financial attention frequency on their net buying. The dependent variables are (AB)SmaOrder, (AB) MidOrder, (AB)LrgOrder, and (AB)SlrgOrder, which represent the (abnormal) daily net buying of small orders, medium orders, large orders, and super-large orders, respectively. Trades of < 40,000 yuan (RMB) are defined as small trades, trades between 40,000 yuan (RMB) and 200,000 yuan (RMB) are defined as medium trades, trades between 200,000 yuan (RMB) and 1 million yuan (RMB) are defined as large trades, and those > 1 million yuan (RMB) are defined as super-large trades. (AB)FreqST and (AB)FreqOT represent investors' (abnormal) daily financial attention frequency. To save space, the control variables reflecting other market factors shown in Table 6 are not reported. The weekday effects are also controlled for in all the regressions. The t-statistics presented in parentheses below each estimated coefficient are reported based on regressions using Newey–West standard errors. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.

as well as weekday effects with day-of-the-week dummy variables. Panel A (B) shows the driving forces behind financial attention frequency on investors' net buying based on abnormal (direct) measures. In Panel A, the coefficient of ABFreqSTt-1 in M1 is significant (0.401) at the 1% level, indicating that financial attention frequency, measured using abnormal frequency of use, significantly increases the abnormal net purchases of small orders. Similarly, the coefficient of ABFreqOTt-1 in M2 is 1.739 at the 10% significance level. In addition, in M3 and M4, the coefficients of ABFreqSTt-1 and ABFreqOTt-1 are significant (0.104 and 0.589, respectively), suggesting that financial attention frequency also plays a significant role in promoting the net purchases of medium orders. On the contrary, in M5 to M7, the coefficients of ABFreqSTt-1 and ABFreqOTt-1 are significantly negative, indicating that abnormal financial attention frequency is negatively related to the net buying transactions of institutional investors. Similar findings are shown in Panel B. Table 11 shows that investors' financial attention frequency significantly stimulates individual investors to make small and medium orders. If we understand the economic implications of the use of SSMAs from the perspective of investor attention, our findings are also consistent with those of Barber and Odean (2008), Da et al. (2011), and Brown et al. (2015). In addition, the findings in Table 11 suggest that the indices of investors' financial attention frequency constructed in this study do reflect the frequency of individual investors' acquisition of market information rather than acquisition by institutional investors. 6. Conclusion The popularity of the mobile Internet is causing individual investors to increasingly rely on mobile devices such as smartphones to acquire information and manage assets. As a financial technology that provides direct access to individual investors, SSMAs play a crucial role in this process. In this study, based on daily user behavior data from the EVTS database on the use of SSMAs in Mainland China, we construct investors' financial attention frequency indices that reflect investors' access to financial market information and 17

Pacific-Basin Finance Journal 58 (2019) 101239

W. Cai and J. Lu

we make the following findings. First, there is a clear ostrich effect in the decisions of investors to use SSMAs to obtain financial information. That is, investors' financial attention frequency is lower in periods of low market returns and high volatility. In addition, investors' financial attention frequency significantly increases market trading activity. This effect remains significant irrespective of whether we measure financial attention frequency based on frequency of use or online duration. We also conduct robustness tests that control for other market factors, examine the effects based on a direct measure of financial attention frequency, measure financial attention frequency on Sundays, and explore SSMAs without trading systems. All the empirical results remain consistent with our previous conclusion. Finally, we find that financial attention frequency also significantly increases individual investors' net buying transactions, measured by small and medium-sized orders. As the only research that links the use of SSMAs with capital market trading, the presented findings enrich the body of knowledge on investor attention and investor trading and better explain the impact of smartphones, especially mobile apps, on investors' information acquisition and trading decisions. Scholars ought to pay close attention to the tremendous changes that mobile apps are bringing about in the capital market. However, this study also has some limitations. First, because of technical constraints, no daily data on SSMAs from before January 1, 2016 were available, thereby limiting our research. Second, our direct measure of SSMA users' activity, which can only be measured at the market level, confines our research to analyses of the overall market. The advancement of technology in this big data era will improve the availability of high-frequency, accurate, and diversified data on the behaviors of SSMA users. Further, more statistics on the click-through rates of investor-selected stocks, searches in SSMAs, and click-through rates of news in SSMAs will become available.18 All these data will further enable us to study more directly and deeply the impact of investors' information acquisition on the capital market, thus further promoting the development of the behavioral finance literature. Acknowledgements Jing Lu gratefully acknowledges the financial support from the National Natural Science Foundation of China (No: 71973018 and 71373296), and the Fundamental Research Funds for the Central Universities (No: 2018CDJSK02PT10). Appendix Appendix Table A.1

Variable definitions and data sources Variable

Definition

Data sources and calculation formulas

Trading Volume

The daily dollar trading volume of the Ashare market The natural logarithm of the daily dollar trading volume of the A-share market Daily turnover Daily investors' financial attention frequency measured using frequency of use

Data Source: CSMAR database.

Volume Turnover (%) FreqST

Data Source: CSMAR database. Data Source: CSMAR database. Data Source: Yiguan Corp. We first calculate the raw data from

FreqSTtRaw =

Nt i=1

Usersi, t

STi, t Nt i = 1 Usersi, t

. N is the total number of SSMAs used on day t. Users

represents the number of active users of SSMA i on day t. ST is the average frequency of use per user for SSMA i on day t. Then, using the regression

FreqSTtRaw = FreqOT

Daily investors' financial attention frequency measured using online duration

+

0

4 i = 1 i DumWeekdayi

+

11 j = 1 j DumMonthj

+

1 Tt

+ t , we exclude the time

trend (T), weekly (DumWeekday), and monthly (DumMonth) effects and measure FreqST as εt. Data Source: Yiguan Corp. We first calculate the raw data from

FreqOTtRaw =

Nt i=1

Usersi, t

OTi, t Nt i = 1 Usersi, t

. N is the total number of SSMAs used on day t. Users

represents the number of active users of SSMA i on day t. OT is the average online duration per user for SSMA i on day t. Then, using the regression

FreqOTtRaw =

Ret (%) RV

CRt-1,t-j

The daily market return of the SSCI Daily realized volatility calculated using the high-frequency data of the SSCI every five minutes intraday

0

+

4 i = 1 i DumWeekdayi

+

11 j = 1 j DumMonthj

+

1 Tt

+ t , we exclude the

time trend (T), weekly (DumWeekday), and monthly (DumMonth) effects and measure FreqOT as εt. Data Source: CSMAR database. Data Source: CSMAR database. Formula: RVt =

48 2 i = 1 ri, t .

Ri,t represents the return (%) of the

SSCI every five minutes corresponding to the trading time of four hours per trading day in the A-share market. The cumulative market returns of SSCI over Data Source: CSMAR database. Formula: CRt−1, t−j = ln Pricet−1 − ln Pricet−j, where Price is the past j trading days the closing price of the SSCI on day t.

(continued on next page)

18

Indeed, these technologies have been implemented in many commercial fields such as mobile loans, shopping, news, and videos. 18

Pacific-Basin Finance Journal 58 (2019) 101239

W. Cai and J. Lu

Appendix Table A.1 (continued) Variable

Definition

SDt-1,t-j

The standard deviation of market daily returns of the SSCI over the past j trading days The negative SSCI return dummy for the previous trading day The negative SSCI cumulative return dummy

NegativeRet_Dumt-1 NegativeRet_Dumt-i,t-j

HighVol_Dumt-1,t-j AB'X'

High market volatility dummy for the previous j trading days The abnormal measure of variable X

OVER

The degree of investor overconfidence

BSVI

The natural logarithm of the daily total Baidu search volume index about market indices The media coverage index about market indices The Arms index representing the reverse indicators of investor sentiment

MEDIA ARMS

ADR FreqSTSunday (FreqOTSunday) FreqST⊥ (FreqOT⊥)

WeChat_Activity TaoBao_Activity SmaOrder (Billion MidOrder (Billion LrgOrder (Billion SLrgOrder (Billion

RMB) RMB) RMB) RMB)

The advance/decline ratio representing daily investor sentiment Daily investors' financial attention frequency measured using frequency of use (online duration) on the last non-trading day before the first trading day of a week. Daily investors' financial attention frequency measured using the frequency of use (online duration) of SSMAs without trading systems. Daily frequency of use of the WeChat mobile app Daily frequency of use of the TaoBao mobile app The daily total volume of net buying of small orders The daily total volume of net buying of medium orders The daily total volume of net buying of large orders The daily total volume of net buying of super-large orders

Data sources and calculation formulas Data Source: CSMAR database. Formula: SDt

1, t j

=

1 j

1

j i=1

(Ret

t j

1 CRt 1,t j j

)

2

Data Source: CSMAR database. A dummy variable that takes the value of 1 when the previous Ret is negative; otherwise, it takes 0. Data Source: CSMAR database. A dummy variable that takes the value of 1 when CRt−i, t−j = ln Pricet−i − ln Pricet−j is negative; otherwise, it takes “0”. Here CRt-i,t-j represents the cumulative return during the previous ‘i'th to ‘j'th trading days. Data Source: CSMAR database. A dummy variable that takes the value of 1 when SDt-1,t-j is greater than its median; otherwise, it takes “0”. 1 20

Formula: ABXt = Xt

20 i = 1 Xt i .

Data Source: CSMAR database. Calculated following Chuang and Lee

(2006):.Volumet =

0

+

p j = 1 j Rett j

+

t

=[

0

+ t] +

p j = 1 j Rett j

= NONOVERt + OVERt .

Data Source: Baidu Corp. Data Source: Baidu Corp. Data Source: CSMAR database. Formula: ARMSt =

# Advt / AdvVolt , # Dect / DecVolt

where #Adv, #Dec, AdvVol,

and DecVol, respectively denote the number of rising stocks, number of declining stocks, trading volume of rising stocks, and trading volume of declining stocks. Data Source: CSMAR database. Formula: ADRt = ln (1 + # Advt/ # Dect), where #Adv and #Dec, respectively denote the number of rising stocks and declining stocks. Data Source: Yiguan Corp. Weighted by the number of users similar to FreqST (FreqOT).

Data Source: Yiguan Corp. Weighted by the number of users similar to FreqST (FreqOT).

Data Source: Yiguan Corp. Data Source: Yiguan Corp. Data Source: Choice Database. Trades of < 40,000 yuan (RMB) are defined as small trades. Data Source: Choice Database. Trades between 40,000 yuan (RMB) and 200,000 yuan (RMB) are defined as medium trades. Data Source: Choice Database. Trades between 200,000 yuan (RMB) and 1 million yuan (RMB) are defined as large trades. Data Source: Choice Database. Trades > 1 million yuan (RMB) are defined as super-large trades.

References Andersen, T.G., Bollerslev, T., 1998. Deutsche mark–dollar volatility: intraday activity patterns, macroeconomic announcements, and longer run dependencies. J. Financ. 53 (1), 219–265. Andersen, T.G., Bollerslev, T., Diebold, F.X., Ebens, H., 2001a. The distribution of realized stock return volatility. J. Financ. Econ. 61 (1), 43–67. Andersen, T.G., Bollerslev, T., Diebold, F.X., Labys, P., 2001b. The distribution of realized exchange rate volatility. J. Am. Stat. Assoc. 96 (453), 42–55. Baker, M., Wurgler, J., 2006. Investor sentiment and the cross-section of stock returns. J. Financ. 61 (4), 1645–1680. Bank, M, Larch, M., Peter, G., 2011. Google search volume and its influence on liquidity and returns of German stocks. Financ. Markets Portfolio Manag. 25 (3), 239–264. Barber, B.M., Odean, T., 2001. The internet and the investor. J. Econ. Perspect. 15 (1), 41–54. Barber, B.M., Odean, T., 2008. All that glitters: the effect of attention and news on the buying behavior of individual and institutional investors. Rev. Financ. Stud. 21 (2), 785–818. Barber, B.M., Odean, T., Zhu, N., 2009. Do noise traders move markets? Rev. Financ. Stud. 22 (9), 151–186. Barndorff-Nielsen, O., Shephard, N., 2002. Econometric analysis of realized volatility and its use in estimating stochastic volatility models. J. R. Stat. Soc. 64 (2), 253–280. Brown, G.W., Cliff, M.T., 2004. Investor sentiment and the near-term stock market. J. Empir. Financ. 11 (1), 1–27. Brown, N.C., Stice, H., White, R.M., 2015. Mobile communication and local information flow: evidence from distracted driving laws. J. Account. Res. 53 (2), 275–329. Brunnermeier, M.K., Parker, J.A., 2005. Optimal expectations. Am. Econ. Rev. 95 (4), 1092–1118. Caplin, A., Leahy, J., 2001. Psychological expected utility theory and anticipatory feelings. Q. J. Econ. 116 (1), 55–79. Chan, C.C., Fong, W.M., 2006. Realized volatility and transactions. J. Bank. Financ. 30 (7), 2063–2085. Chordia, T., Subrahmanyam, A., 2004. Order imbalance and individual stock returns: theory and evidence. J. Financ. Econ. 72 (3), 485–518. Chuang, W.I., Lee, B.S., 2006. An empirical evaluation of the overconfidence hypothesis. J. Bank. Financ. 30 (9), 2489–2515.

19

Pacific-Basin Finance Journal 58 (2019) 101239

W. Cai and J. Lu

Da, Z., Engelberg, J., Gao, P.J., 2011. In search of attention. J. Financ. 66 (5), 1461–1499. Da, Z., Gurun, U.G., Warachka, M., 2014. Frog in the pan: continuous information and momentum. Rev. Financ. Stud. 27 (7), 2171–2218. Da, Z., Engelberg, J., Gao, P.J., 2015. The sum of all FEARS investor sentiment and asset prices. Rev. Financ. Stud. 28 (1), 1–32. Deaves, R., Luders, E., Schroder, M., 2010. The dynamics of overconfidence: evidence from stock market forecasters. J. Econ. Behav. Organ. 75 (3), 402–412. Dimpfl, T., Jank, S., 2016. Can internet search queries help to predict stock market volatility? Eur. Financ. Manag. 22 (2), 171–192. Dorn, D., Huberman, G., Sengmueller, P., 2008. Correlated trading and returns. J. Financ. 63 (2), 885–920. Engelberg, J.E., Parsons, C.A., 2011. The causal impact of media in financial markets. J. Financ. 66 (1), 67–97. Fang, L., Peress, J., 2009. Media coverage and the cross-section of stock returns. J. Financ. 64 (5), 2023–2052. Fleming, J., Kirby, C., 2011. Long memory in volatility and trading volume. J. Bank. Financ. 35 (7), 1714–1726. Garcia, D., 2013. Sentiment during recessions. J. Financ. 68 (3), 1267–1300. Gervais, S., Odean, T., 2001. Learning to be overconfident. Rev. Financ. Stud. 14 (1), 1–27. Hillert, A., Jacobs, H., Muller, S., 2014. Media makes momentum. Rev. Financ. Stud. 27 (12), 3467–3501. Hirshleifer, D., Lim, S.S., Teoh, S.H., 2002. Disclosure to a Credulous Audience: The Role of Limited Attention. SSRN. https://ssrn.com/abstract=604143. Hou, K., Xiong, W., Peng, L., 2009. A Tale of Two Anomalies: The Implications of Investor Attention for Price and Earnings Momentum. SSRN. https://ssrn.com/ abstract=976394. Huang, L., Liu, H., 2007. Rational inattention and portfolio selection. J. Financ. 62 (4), 1999–2040. Huddart, S., Lang, M., Yetman, M.H., 2009. Volume and price patterns around a stock’s 52-week highs and lows: theory and evidence. Manag. Sci. 55 (1), 16–31. Hvidkjaer, S., 2008. Small trades and the cross-section of stock returns. Rev. Financ. Stud. 21 (3), 1123–1151. Joseph, K., Wintoki, M.B., Zhang, Z.L., 2011. Forecasting abnormal stock returns and trading volume using investor sentiment: evidence from online search. Int. J. Forecast. 27 (4), 1116–1127. Kahneman, D., 1973. Attention and Effort. Printice-Hall, Englewood Cliffs, NJ. Karlsson, N., Loewenstein, G., Seppi, D., 2009. The ostrich effect: selective attention to information. J. Risk Uncertain. 38 (2), 95–115. Lee, C.M., Radhakrishna, B., 2000. Inferring investor behavior: evidence from TORQ data. J. Financ. Mark. 3 (2), 83–111. Li, J., Yu, J., 2012. Investor attention, psychological anchors, and stock return predictability. J. Financ. Econ. 104 (2), 401–419. Loewenstein, G., 1987. Anticipation and the valuation of delayed consumption. Econ. J. 97 (387), 666–684. Moore, E., 2013. Smartphones mean never having to worry about your portfolio. Financ. Times September 19. Newey, W.K., West, K.D., 1987. A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent covariance matrix. Econometrica 55 (3), 703–708. Newey, W.K., West, K.D., 1994. Automatic lag selection in covariance matrix estimation. Rev. Econ. Stud. 61 (4), 631–653. Odean, T., 1999. Do investors trade too much? Am. Econ. Rev. 89 (5), 1279–1298. Patel, S., 2014. Should you trade stocks on your iPhone? Wall Street J April 13. Peng, L., Xiong, W., 2006. Investor attention, overconfidence and category learning. J. Financ. Econ. 80 (3), 563–602. Rossi, E., de Magistris, P.S., 2013. Long memory and tail dependence in trading volume and volatility. J. Empir. Financ. 22, 94–112. Seasholes, M.S., Wu, G., 2007. Predictable behavior, profits, and attention. J. Empir. Financ. 14 (5), 590–610. Shiller, R., 2000. Measuring bubble expectations and investor confidence. J. Psychol. Financ. Markets 1 (1), 49–60. Sicherman, N., Loewenstein, G., Seppi, D.J., Utkus, S.P., 2016. Financial attention. Rev. Financ. Stud. 29 (4), 863–897. Sims, C.A., 2003. Implications of rational inattention. J. Monet. Econ. 50 (3), 665–690. Sims, C.A., 2006. Rational inattention: beyond the linear-quadratic case. Am. Econ. Rev. 96 (2), 158–163. Tetlock, P.C., 2007. Giving content to investor sentiment: the role of media in the stock market. J. Financ. 62 (3), 1139–1168. Vlastakis, N., Markellos, R.N., 2012. Information demand and stock market volatility. J. Bank. Financ. 36 (6), 1808–1821. Wang, Y.H., Keswani, A., Taylor, S.J., 2006. The relationships between sentiment, returns and volatility. Int. J. Forecast. 22 (1), 109–123. Yuan, Y., 2015. Market-wide attention, trading, and stock returns. J. Financ. Econ. 116 (3), 548–564. Zhang, B., Wang, Y., 2015. Limited attention of individual investors and stock performance: evidence from the ChiNext market. Econ. Model. 50, 94–104.

20