没有合适的资源?快使用搜索试试~ 我知道了~
2019_It depends on when you search(使用“搜索量”来量化“用户注意力”)
1.该资源内容由用户上传,如若侵权请联系客服进行举报
2.虚拟产品一经售出概不退款(资源遇到问题,请及时私信上传者)
2.虚拟产品一经售出概不退款(资源遇到问题,请及时私信上传者)
版权申诉
0 下载量 170 浏览量
2022-07-10
09:00:41
上传
评论
收藏 905KB PDF 举报
温馨提示
试读
29页
2019_It depends on when you search(使用“搜索量”来量化“用户注意力”)
资源推荐
资源详情
资源评论
2019-2976-FIN
1
It Depends on When You Search!
1
2
3
Internet search has been shown to have strong predictive power for stock prices, company
4
sales, and the spread of influenza. Following studies that propose search frequency as a
5
more direct and timely measure of investor attention, we explore the heterogeneity in the
6
search data and address several questions in the current literature. Using the daily search
7
volume index (SVI) from Google Trends for the S&P 500 stocks, we distinguish Internet
8
searches conducted during weekends from those conducted during weekdays and show
9
that weekend search can better capture retail investors’ attention. First, we find that
10
reduced market response to Friday earnings announcements happens only when the
11
weekend search (a proxy for attention) is low, contributing to the ongoing debate on the
12
cause of this phenomenon. Second, we find that weekend search can help to predict
13
returns for large-cap stocks, whereas the current literature shows that weekly search can
14
predict returns only for smaller firms. Furthermore, we find that the weekend search
15
measure from the Chinese Baidu Index can predict CSI 300 stock returns. These results
16
are robust when we include holidays or reorder the days in a week. Finally, we explore
17
periods with high investor sentiment and periods with high market volatility. We find that
18
search during “busy” weekends is less effective as a measure of investor attention in stock
19
return predictability.
20
21
Keywords: Google Trends, Internet Search, Retail Investor, Stock Return, S&P
22
500, Weekends
23
24
25
Introduction
26
27
The search engine is the most commonly used product on the Internet for
28
individuals to directly acquire information. .The search volume generated by users
29
is also widely used as a valid and timely proxy for predicting sales (Geva et al.
30
2017), the spread of influenza (Ginsberg et al. 2009, Lazer et al. 2014), and stock
31
prices (Da et al. 2011) in various fields of academic research. These studies argue
32
that search is a revealed attention measure: if you search for something with
33
Google, you are undoubtedly paying attention to it. The successful utilization of
34
search volume in financial markets greatly benefits both industry and academia.
35
Search volume, unlike social media, which usually contains content (text),
36
ratings (scores), and posters’ information (demographics), usually includes just the
37
search frequency (volume indices) and a time label (time generated) (Brynjolfsson
38
et al. 2016). Previous studies mainly focus on weekly or monthly changes in the
39
indices and have not explored their heterogeneity. In particular, the time label
40
associated with search volume remains unexplored. It is unclear whether searches
41
conducted on different days of a week, such as weekdays and weekends, contain
42
equal amounts of information. If the answer is that the amounts of information are
43
2019-2976-FIN
2
not equal, the question is, can we improve the predictive power of search volume
1
by taking advantage of the time label to refine the measure?
2
In this paper, we argue that Internet searches under different circumstances do
3
not reflect equal amounts of attention. This is especially true when there is an
4
overabundance of information, which can lead to constrained attention paid to a
5
particular event (Liu and Ye 2016, Hirshleifer et al. 2009). As is often stated in the
6
literature on attention, a wealth of information creates a poverty of attention. We
7
distinguish the aggregate Internet searches conducted during weekends from those
8
conducted during weekdays because weekdays typically have an overload of
9
information in comparison to weekends. The stock markets are closed over the
10
weekends. During trading sessions on weekdays, investors have a limited amount
11
of time to digest information when prices are moving in real time. Intra-week
12
information seasonality is also an issue since announcements and news releases
13
mostly occur during weekdays. Moreover, investors’ cognitive ability and
14
disposition usually improve during weekends, since they have more private time
15
and space and are away from the pressure and distraction of the workdays, all of
16
which lessen the constraint on the attention allocation (Ryan et al. 2010, Geurts et
17
al. 2003).
18
To construct search indices on weekdays and weekends, we first download the
19
search volume from Google Trends on a daily frequency from 2004 to 2012. For
20
each company in the S&P 500 index, we aggregate the daily data from each week
21
into weekday searches and weekend searches.
1
This process creates two weekly
22
series for each stock. Following Da et al. (2011), we compute the abnormal SVI
23
(ASVI) by subtracting the lagged log (SVI) from the current week log (SVI).
24
ASVI takes into account the week-to-week variation in the total SVI. We compute
25
the weekday ASVI, denoted by ASVI
1-5,
and the weekend ASVI, denoted by
26
ASVI
67
,
separately. Intuitively, ASVI
1-5
and ASVI
67
reflect the percentage change
27
of SVI over previous weeks. The contemporaneous correlation between ASVI
1-5
28
and ASVI
67
is 0.406 in our sample. Since our interest is in the difference between
29
the weekday and weekend search, we subtract ASVI
1-5
from ASVI
67
to compute
30
the incremental component, denoted by ASVI
d
. We find that in our sample, the
31
contemporaneous correlation between ASVI
d
and the total ASVI measure by Da et
32
al. (2011) is rather low at 0.02, which indicates that the information content in the
33
incremental component is very different from that in the ASVI measure in Da et al.
34
(2011).
35
With our constructed series of weekday and weekend search, we first study a
36
puzzling question surrounding Friday earnings announcements in Dellavigna and
37
1
We focus on the S&P 500 mainly because of the limited availability of the daily SVI. Except for
large stocks, the daily SVI is missing for many small stocks. In fact, Cziraki et al. (2017) find no
evidence of an empirical relation between SVI and future abnormal stock returns when using S&P
500 stocks. However, after differentiating stock search volume in weekends from weekdays, the
predictive power of ASVI in our study sharply increases and becomes significant among S&P 500
stocks.
2019-2976-FIN
3
Pollet (2009), which shows investors react less to Friday announcements than
1
announcements on other weekdays. It is unclear whether this reaction is due to
2
investors’ inattention during weekends or the fact that Friday earnings
3
announcements that elicit weak responses often occur on Fridays. Our strategy is
4
to investigate whether the immediate market responses are correlated with
5
weekend search activity. To the extent that weekend search is a good proxy for
6
retail investor attention, we expect that the response is stronger with more positive
7
weekend ASVI and weaker with more negative weekend ASVI. The results of our
8
analysis indeed suggest that the Friday earnings announcements effect is stronger
9
(weaker) during weekends with low (high) ASVI.
2
10
Our second exercise is to explore the stock return predictability of weekend
11
search. Da et al. (2011) find that a higher ASVI (using total search volume in a
12
week to calculate the abnormal search volume index) leads to higher returns in the
13
following weeks. Their sample is based on the Russell 3000, and the predictability
14
exists mostly for small-cap stocks. We explore whether distinguishing search time
15
can help to improve the predictability for large-cap stocks. The firms in our sample
16
are among the largest in the S&P 500, and we confirm that there is no
17
predictability of total ASVI and ASVI
1-5
on future DGTW (Daniel et al. 1997)
18
abnormal returns. In sharp contrast, we find that ASVI
d
has strong predictive
19
power for future DGTW abnormal returns in the subsequent week. For example, a
20
one standard deviation increase in ASVI
d
is associated with 1.17% increase in the
21
stock return in the subsequent week. The above results are robust when we include
22
holidays or reorder the days in a week.
23
Our finding that weekend searches can predict the subsequent week’s returns
24
for the largest firms is novel to the literature. Large stocks typically receive the
25
most coverage from financial analysts and the news media. Therefore, it is striking
26
that we can still find predictability for the stocks with the largest market caps by
27
refining the measure of investor attention.
28
Further, we extend our results to the emerging Chinese stock markets using
29
the Baidu search index. The Chinese stock markets have developed into one of the
30
largest and most active markets in the world.
3
Baidu is a dominant Internet search
31
engine and provides SVI that resembles Google Trends. The Baidu’s SVI for
32
stocks has another advantage as the stock tickers in China are unique, which
33
further reduces the noise in the data.
4
The Chinese stock markets have a higher
34
2
Note that throughout the paper, the “Friday earnings announcements effect” or the “Friday
announcements effect” refers to the empirical fact that the immediate market response to Friday
announcements is weaker than non-Friday announcements. Therefore, a stronger Friday
announcements effect is equivalent to a weaker market response to Friday announcements.
3
A new world record for one day’s trading volume was created in the Chinese stock market on
December 5, 2014: 1.0741 trillion RMB.
4
Chinese stock tickers are six digits and uniquely defined. Unlike US stock markets, searching for a
six-digit stock code in Baidu is almost identical to searching for the corresponding stock with very
2019-2976-FIN
4
proportion of retail investors than the US stock markets, so we expect to find a
1
stronger impact of limited investor attention. Our findings with Chinese stocks are
2
indeed consistent with and stronger than what we find with US stocks.
3
Finally, we explore variations in weekend search that can further strengthen
4
our argument for using weekend search over weekday search to measure investor
5
attention. We argue that not all weekends are equal and some weekends are “busy”
6
with an overabundance of information and events. During such busy weekends,
7
investors’ cognitive abilities are restrained, and weekend search can be a noisy
8
measure of investor attention as well. We analyze periods with high investor
9
sentiment and periods with high market volatility and find that search during busy
10
weekends is less effective as a measure of investor attention in predicting stock
11
returns.
12
13
14
Background and Hypothesis Development
15
16
Information Technology, Information Environment, and Attention Measures
17
18
In this era of big data, information plays an important role. The wisdom of
19
crowds based on massive information has 1great power and possibility in
20
influencing the financial market. For example, to companies, the use of social
21
media is associated with firm equity value, and the transformative power of social
22
media is crucial to company development (Luo et al. 2013). To investors, the use
23
of Twitter can significantly reduce information asymmetry and increase the
24
liquidity of stocks (Blankespoor et al. 2013, Das and Chen 2007). With respect to
25
the stock market, Wikipedia can improve the information environment in the
26
financial market and moderate the timing of managers’ voluntary disclosure of
27
companies’ bad news (Xu and Zhang 2013).
28
The empiricists in financial economics face a substantial challenge in testing
29
theories of limited attention because direct measures of investors’ attention are
30
difficult to obtain. Instead, researchers have resorted to indirect proxies for
31
attention, such as extreme returns (Barber and Odean 2008), trading volume
32
(Barber and Odean 2008, Gervais et al. 2001, Hou et al. 2009), news and headlines
33
(Barber and Odean 2008, Yuan 2015), advertising expenses (Chemmanur and Yan
34
2009, Grullon et al. 2004, Lou 2014), and price limits (Seasholes and Wu 2007).
35
These proxies are indirect because of the reliance on the critical assumption that a
36
stock’s extreme return, turnover, or coverage in the news media should guarantee
37
investors’ attention. This assumption can be especially problematic when
38
investors’ attention is too scarce to cover all of these events or when attention is
39
not associated with these events.
40
little noise. For instance, searching for 601857 -- the ticker for China National Petroleum Corp -- in
Baidu shows trading and financial information for this company at the top of the result list.
2019-2976-FIN
5
The development of the Internet and information technology not only
1
facilitates information acquisition but also enables real-time recording of users’
2
searching activities. Numerous types of information about searches conducted by
3
users through search engines can be recorded, including the search location, time,
4
and frequency. Taking advantage of these features, Da et al. (2011) introduce the
5
SVI from Google Trends as a timely and efficient proxy for measuring individual
6
investors’ attention.
5
Compared with traditional indirect measures, the SVI has the
7
following advantages. First, Internet users commonly use a search engine to
8
collect information; search engines such as Google and Baidu continue to be
9
favorites in the United States and China. Indeed, up to now Google has accounted
10
for more than 70% of all the search queries performed in the United States. The
11
search volume reported by search engines is thus likely to be representative of the
12
Internet search behavior of the general population. Second, and more critically, a
13
search is a direct measure of attention: if you search for a stock in Google, you are
14
undoubtedly paying attention to it. Third, Google Trends or Baidu Index offers
15
timely (weekly/daily) SVI data for most keywords, which can effectively measure
16
the real-time attention of investors. Therefore, the aggregate search frequency in
17
Google Trends or Baidu Index is a direct and unambiguous measure of attention.
18
These advantages have led to the widespread adoption in empirical studies of the
19
SVI as a measure of investors’ attention.
20
21
Can Weekend Search Explain the Friday Earnings Announcements Effect?
22
23
We first reexamine the well-known findings on Friday earnings
24
announcements. Dellavigna and Pollet (2009) document that the immediate stock
25
response to earnings announcements on Friday is 15% weaker than on other
26
weekdays, and they attribute this effect to limited investor attention during
27
weekends. However, a recent study by Michaely et al. (2016) argues that the
28
previous findings are driven by selection bias that the firms that make
29
announcements on Fridays are different from the rest, and they find no evidence
30
that investors pay less attention to Friday announcements after correcting for
31
selection bias.
32
To the extent that weekend search volume captures retail investor attention,
33
we should be able to identify the channel through which investor attention affects
34
the market response. Specifically, we can distinguish the weekends with high and
35
low retail investor attention and examine whether the Friday announcements have
36
different effects on stock returns across high and low ASVI weekends. If we found
37
no difference, our results would be consistent with Michaely et al. (2016) that
38
investor attention plays no role in the Friday earnings announcements effect. On
39
the other hand, if we find that the Friday announcements effect is more
40
5
Ben-Rephael et al. (2017) also extend this approach to capturing institutional investor attention
using Bloomberg view volume.
剩余28页未读,继续阅读
资源评论
博士僧小星
- 粉丝: 1711
- 资源: 5876
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
最新资源
- 农村信用社联合社计算机信息系统投产与变更管理办.docx
- 农村信用社联合社计算机信息系统数据管理办法.docx
- 利用SPSS作临床效度分析线上计算网站介绍-医学研究部统计谘.(医学PPT课件).ppt
- 利用Zabbix监控mysqldump定时备份数据库状态.docx
- 利用计算机解决问题的基本过程.doc
- 化工铁路通信工程总结.doc
- 北京大学网络教育软件工程作业.docx
- 医药公司(连锁店)计算机操作规程未新系统的自行按照旧制修改-新系统过制的编号加修模版.doc
- 医药公司(连锁店)计算机系统操作规程模版.doc
- 医药连锁门店计算机系统的操作和管理程序未新系统的自行按照旧制修改-新系统过制的编号加修模版.docx
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功