• No results found

The effect of investor attention on the AEX stock returns, trading volume and volatility

N/A
N/A
Protected

Academic year: 2021

Share "The effect of investor attention on the AEX stock returns, trading volume and volatility"

Copied!
34
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

The effect of investor attention on the AEX stock returns, trading

volume and volatility

University of Amsterdam Amsterdam Business School BSc Business Administration

Mitzy Rachel van Gelder 11884436

BSc in Business Administration: Finance Track Finance

University of Amsterdam Amsterdam Business School Date of submission: June 2020

(2)

Acknowledgment

I would like to thank my family and friends for their enormous support through writing this thesis. I would like to thank my supervisor Dr. J.J.G. Lemmen for supporting and guiding me through this period.

Statement of Originality

This document is written by Mitzy Rachel van Gelder who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no

sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of

(3)

Abstract

This thesis studies the effect of investor attention, expressed in Google search volume, on stock returns, trading volume and volatility of firms on the Amsterdam Exchange Index. The results indicate that there is a positive relation between investor attention and stock returns. Google search volume is not only positively related to volatility, but it can also predict volatility. These findings contribute to the current literature due to the novel sample, which captures Dutch investors.

Keywords: Google Trends – Investor attention – Stock returns – Volatility – Trading Volume

(4)

Table of contents Acknowledgment ... 2 Abstract ... 3 Table of contents ... 4 1 Introduction ... 5 2 Literature Review ... 7 2.1 Investor Attention ... 7

2.2 Stock Prices and Return ... 9

2.3 Trading Volume ... 11

2.4 Volatility ... 12

3 Data ... 13

3.1 Google Search Volume ... 13

3.1.1 Abnormal Search Volume Index ... 15

3.2 AEX stock Data ... 16

3.2.1 Stock prices and abnormal returns, trading volume and volatility ... 16

4 Methods ... 18

5 Results ... 21

5.1 Descriptive statistics and correlations ... 21

5.2 Main analysis ... 22

6 Discussion ... 27

7 Conclusion ... 29

8 References ... 30

(5)

1 Introduction

In today’s world everyone finds themselves multitaskers. However, according to Kahneman (1973), attention is a limited cognitive resource. Thus, doing more than one task together is less simple than assumed. For an investor to decide in which stocks to invest, there is a lot of choice. There is a lot of information available on the stock market, but there is only a limited set of stocks and information an investor is able to evaluate, due to the investor’s scarce attention (Peng & Xiong, 2006). Due to the constraints of attention, it is almost impossible for an individual investor to consider every available stock. Investors are more likely to buy stocks that caught their attention, resulting in investors to make attention-driven investment decisions (Barber & Odean, 2008).

The way investors receive information has changed over the last two decades, due to the contemporary digital environment. There has been a transformation from traditional media institutions to internet media, and as a result even more information has become available, but also accessible to more people (Van Couvering, 2008). In 2010, 92% of Dutch companies listed on Euronext use internet as their main source to attract investors and build or strengthen relationships with investors (Bollen, Geerings & Hassink, 2010). These firms use it as a platform to publish reports, hold press releases and sometimes even direct contact in order to capture investors’ attention.

Due to the digitalization, online search engines are an unmissable part in our digital age, as they help people navigate through all available media and information. In January 2020, Google is still the search engine with the biggest market share, namely 87% (Clement, 2020). Google’s CEO Eric Schmidt said himself ‘Google is simply an aggregator of

information’ (Sullivan, 2006). It is common for investors to use search engines in order to

find information. Da, Engelberg and Gao (2011) have proven that Google Search Volume (GSV) of stock tickers can be used as a direct measure of investor attention.

One assumption under the capital asset pricing model is that the market is efficient (Sharpe, 1964; Lintner, ,1965). As a result, information must be known to everyone. This assumption is only applicable in theory, as investors face limited attention in reality, making it impossible to be exhaustively informed. Studies have shown that investor attention has an effect on the valuation of assets, proving the inefficiency of the markets (Merton, 1987).

(6)

Research have reported that Google Search Volume Index (SVI), has an effect on stock’s returns (Da et al., 2009; Bijl et al., 2016). This supports Barber and Odean’s (2008) price pressure hypothesis. Other researches have also focused on other effects, which have reported significant relations with investor attention, namely trading volume and volatility (Aoudi, Arouri & Teulon, 2013; Bank, Larch & Peter, 2011).

This research will contribute to past literature focused on the relation between investor attention and stock returns, trading volume and volatility. Previous studies have focused on data sets based on stock markets such as the U.S., Germany, France and more (Da et al., 2011; Bank et al., 2011; Aouadi et al., 2013). However, some stock markets have not been studied yet. This research aims to partly fill this gap by using a new data set which focuses on the Dutch stock market, more specifically the Amsterdam Exchange Index (AEX). By employing a new stock market and more recent dataset, this research is aiming to show support on literature, and to find out whether past findings can be generalized to the Dutch stock market.

Therefore, the following research question has been formulated: ‘What is the effect of

investor attention, expressed in the Google Trends search query volumes, on AEX stocks returns, trading volume and volatility?’

First, a univariate analysis is conducted in which the stocks are sorted in search volume. Mean differences for return, trading volume and volatility are then tested. After, a multivariate analysis is done using time series regressions with fixed effects in order to find the relation between return, trading volume, volatility with abnormal SVI.

This study confirms that there is a positive and significant relation between investor attention and return, and thus supports findings from Barber and Odean (2008) and Da et al. (2011). Moreover, a significant and positive relation is reported between Google search volume and volatility, confirming to previous findings from Vlastakis and Markellos (2012) and Aouadi et al. (2013).

This thesis is set up as follows. Chapter 2 provides a literature review. Chapter 3 describes the data. Chapter 4 presents the method. Chapter 5 presents the results. Chapter 6 provides a discussion of the results and chapter 7 concludes the thesis.

(7)

2 Literature Review

There have been different studies in the past focusing on the impact of investor attention on stock’s returns, volatility and trading activity. Google SVI as a proxy for investor attention is still a relatively new topic. As a result of the novelty of the topic, researches are exploring different markets, and the different effect GSV has on these markets and the market activity. However, there is still an ongoing debate on how GSV affects the stock market.

2.1 Investor Attention

Most financial models are built on assumptions of efficient markets and complete

information (Sharpe, 1964). According to Merton (1987), however, the cognitive rationality of investors is not sufficiently captured in the financial models with frictionless markets, and appear to be more complex. As a result, Merton (1987) introduced investor recognition, implying that the investors’ attention affects stock’s valuation and liquidity. When studying investor attention, it should be recalled that attention is a limited cognitive resource, as shown by Kahneman (1973). It implies that attention and effort are closely related. Meaning that when attention is given to one matter, other matters cannot receive the attention (Peng & Xiong, 2006). This is applicable to investors, as there are more matters to pay attention to than available cognitive resources. Building on this theory, there are two effects of investor attention.

First, because of the scarcity of attention and broad availability of information, individual investors have to process information efficiently. Peng and Xiong (2006) argue that limited investor attention has an effect on the investor’s learning process, but that most investors face this problem. They hypothesize that it is more common for investors to process market or industry-wide information, rather than company specific information, due to the scarcity of investor attention. As a result, Peng and Xiong (2006) argue that this, together with overconfidence, affects return.

Second, investor attention affects the price pressure hypothesis, argued by Barber and Odean (2008). This implies that investors often show attention-grabbing buying behavior, meaning that investment decisions are affected by attention catching stocks (Barber & Odean, 2008). They hypothesize that investors will only buy stocks that they have paid attention too,

(8)

from. As a result, Barber and Odean argue that the increase in investor attention shows a temporary rise in a stock’s price. Barber and Loeffler (1993) argue that the short-term price change is caused by naïve investors, who believe they can profit from information that is publicly available.

It is of more recent studies since the last two decades to see whether investor attention has any effect on asset pricing models (Barber & Odean, 2008; Da et al. 2011; Bank et al., 2011; Bijl et al., 216). The price pressure hypothesis by Barber and Odean (2008) contradicts the Efficient Market Hypothesis (EMH) as constructed by Fama (1970). The EMH states that in a strong form of the hypothesis, all public and private information is projected in stock prices. In the weak form of EMH, all historic information is incorporated in stock prices. According to this theory, information is incorporated in the stock price, assuming the whole market informed, no profits can be made (Fama, 1970).

When a company is in the news, or a remarkable event around a stock happens, investors are likely to pay attention to it. As a result, they are also likely to buy a stock, as investors are net buyers of stocks that catch investors’ attention (Barber & Odean, 2008). However, according to EMH, this should not impact the stock price, and if it does, it should be permanently, as the news should already be incorporated in the price. It means that if an event is remarkable, the information it gives should not have anything to do with the future performance of a stock (Barber & Odean, 2008). However, the price pressure hypothesis states that it will cause a temporary price rise.

Past studies have used different proxies for investor attention, such as news events (Yuan, 2008; Fang & Peress, 2009), advertising costs (Grullon, Kanatas & Weston, 2004) and abnormal returns (Barber & Odean, 2008). The previous mentioned proxies measure investor attention in an indirect way, as it is uncertain whether investors are actually paying attention. For example, it is uncertain whether investors have read or seen certain news events or advertisements. As a solution, Da et al. (2011) have successfully implemented Google search frequency (SVI) as a direct measure for investor attention. They claim that it captures the attention of retail investors in a timely way. Da et al.’s theory of investor attention supports Barber and Odean’s price pressure theory (2008). After Da et al.’s study, many new researches have implemented SVI as a proxy for investor attention, examining different markets, and different impacts investor attention has.

(9)

2.2 Stock Prices and Return

According to the price pressure theory, an increase in investor attention should show an increase in stock prices (Barber & Odean, 2008). Da et al. (2011) support this theory, as their study shows that if SVI increases, stock prices increase for the first two weeks. The

outperformance of stocks is only for two weeks, after which prices move back again. Merton (1987) argues that there is a negative relation between stock return and investor attention. Assuming not all information is available, thus in an inefficient market, Merton argues that an investor should be compensated for the non-diversifiable, but idiosyncratic, risk. Building on these assumptions, an increase in investor attention is expected to show a permanent decrease in future returns (Bank et al., 2011), which is supported by Fang and Peress (2009). Table 1 summarizes important researches on the effect of investor attention on stock prices and return, which mainly use SVI as a proxy for investor attention.

Table 1 shows that there are many studies who tested the price pressure theory and used SVI as a query for investor attention, initiated by Da et al. (2011). Findings from both Bank et al, (2011) and Takeda & Wakao (2013) support Da et al.’s findings. Additionally, there are other researches supporting these findings, such as Joseph, Wintoki & Zhang

(2011). Fang and Peress (2009) have found opposite results, namely a negative relation which supports Merton (1987). One noticeable concern is that SVI is not used as a measure for search volume, instead media coverage is used as an indirect measure for investor attention (Fang & Peress, 2009). On the other hand, Bijl et al. (2016) have used SVI as proxy, and still found a negative relation between SVI and return. They state that this could be due to the recent nature of the dataset, as information is possibly included faster in prices.

Another trend in past studies is that the size of firms shows a negative moderating effect on the relation between investor attention and stock returns. Individual investors suffer more from information asymmetry, due to an individual’s limited attention. On the other hand, professional, or institutional, investors often use more analyses and data, making them more informed on the market. Moreover, larger firms’ stock outstanding often consist of more institutional investors, and a smaller fraction of individual investors. Therefore, the relation between SVI and stock prices should be stronger for small firms than for big firms, as it captures more individual investors rather than institutional investors. This is supported by Fang and Peress (2009), Da et al. (2011) and Bank et al. (2011).

(10)

Based on previous studies that show that investor attention with SVI as a proxy has a positive effect on stock’s prices and return the following hypothesis are formed:

(11)

2.3 Trading Volume

Barber and Odean (2008) argue that abnormal trading volumes can be observed as an attention-grabbing event. However, it is not a direct measure for investor attention. Nevertheless, Barber and Odean state that, due to the price pressure theory, if an investor pays attention to a stock, it is likely to have a higher trading volume (Takeda & Wakao, 2013). Moreover, trading volume is suitable to measure if search volume, or investor attention, has an effect on trading activity (Bank et al., 2011). Table 2 summarizes previous studies on the effect of investor attention on trading volume.

Table 2 shows that there is a consistent positive correlation between SVI and trading volume in past studies. Based on this, the following hypothesis is formed:

(12)

2.4 Volatility

French, Schwert and Stambaugh (1986) confirm that there is a positive relation between stock return and stock market volatility. Assuming investors are risk-averse, there is also a historical trade-off between stocks with higher and lower return, as they are accompanied with higher or lower risk, or volatility, respectively (Berk & DeMarzo, 2017). Building on the theory of investor attention, Vlastakis and Markellos (2012) suggest that as investors are more risk averse, they demand more information.

Dimpfl and Jank (2016) argue that there is a relation between volatility and individual investors, which are also depicted as the uninformed investors. In turn, Google searches are argued to represent the investor attention of uninformed investors (Da et al., 2011).

Moreover, Black (1986) argued that noise is connected to an increase in volatility (Black, 1986; Dimpfl & Jank, 2016). Building on those findings, De Long et al. (1990) argued that there is an excess volatility on stock prices, if there is trading activity of uninformed traders who create noise. Table 3 displays studies that have tested whether SVI affects volatility, as SVI is representative for uninformed traders.

Table 3 shows different studies that found a relation between SVI and volatility. As a result, the following hypothesis is formed:

(13)

3 Data

The data for this research is gathered from Google Trends, Yahoo! Finance and FactSet. Companies that were listed on the AEX stock between January 2015 and January 2020 have been obtained through FactSet, resulting in 40 companies. Data on these companies listed on the AEX in the entire sample period between 2015 and 2020 are being gathered in this research, in order to avoid survivorship bias (Da et al, 2011). Some companies have been removed due to incomplete information on stock data and insufficient google search queries, which will be elaborated later. Stock information is obtained from Yahoo! Finance, which includes daily stock prices and trading volume.

3.1 Google Search Volume

Investor attention is measured with Google Search Volume (GSV). GSV is obtained through Google Trends, a service from Google which gives insights into search queries on Google’s search engine (Google Trends). Google trends provides historical data, called ‘non-real-time data’, from searches from 2004 until 36 hours before present time (Google Trends). The frequency of the data differs; hourly data is visible for one day, daily data for a 90-day period, weekly data is visible for 5-year periods and monthly data for longer periods since 2004 (Google Trends). As the number of Google searches per day is extremely large, Google trends only uses a sample of their searches. It is ensured that the sample being used represents the total number of searches. However, it is not an exactly representative to reality, for

example, unpopular search terms with very low volume are shown as ‘0’. Even though in reality someone might have searched the term in the assigned period. This is due to the normalization of the data to the time and location of a search query. The data is transformed into relative numbers by Google Trends, on a scale ranging from 0 to 100, in order to better compare the terms. The data provided on a search query can be seen as a relative number to total searches of the same time period and location. Thus, when total search number on a query is equal to 90, it means that it exhibits a very high search volume relative to other searches in the same time period and location (Kim et al, 2019). Finally, the data from Google Trends is anonymous.

Past studies have mostly used weekly data as a measure of investor attention, often for a period no longer than 5 years (Da et al., 2011; Aouadi et al., 2013; Takeda & Wakao, 2013; Kim et al., 2019). In this research weekly data is retrieved for a period of 5 years, namely

(14)

location. Past researches have also employed data filtered by the same country as the studied index (Da et al., 2011; Preis, Moat & Stanley, 2013). By not applying the filter, the data would contain more noise, as some words could have a different meaning in another

language. Thus, irrelevant searches are likely to be connected to a query. Additionally, due to home bias it is more likely for Dutch investors to be interested in Dutch company and its corresponding name (Bodie, Kane & Marcus, 2018). As a result, the Netherlands is used as a filter on Google Trends to capture Dutch investors.

If an investor wants to obtain information on a stock, it is rational for an investor to search for either the corresponding stock ticker or the company name. Thus, determining what search query to choose for the analysis is extremely important. Da et al. (2011) have argued that stock tickers should be used to identify search frequency, as it contains less noise than search query of a company name. On the other hand, Dimpfl and Jank uses the market index as search queries for investor attention rather than company-specific queries, for noise reduction. Company names as search query appears to contain most noise. However, multiple studies have proven th searches on company names are also an appropriate way for obtaining search frequencies (Bank et al., 2011; Vlastakis & Markellos, 2012; Preis et al., 2013; Aouadi et al. 2013; Bijl et al. 2016). Moreover, Google Trends shows that stock tickers as query have low search volumes in The Netherlands, which would lead to omitting many companies. Thus, even though company names as search terms have more noise, it is a more appropriate way to measure search volume of Dutch investors in this research, as it is less likely that Dutch investors search for stock tickers.

Each company’s name has been used as mentioned on Yahoo! Finance. Then, legal terms, such as ‘NV’, have been removed, as it appeared to have less search frequency on Google Trends when included in the search term. Companies for which their names are also their product or website, such as ‘Heineken’, ‘ING’ and ‘Philips’ have been removed. For such companies, it is likely that most search queries are unrelated to investor attention.

Omitting these variables ensures less noise in the data. Moreover, companies with zero search volume in two consecutive weeks are omitted, and company names which have a double meaning are all deleted.

(15)

3.1.1 Abnormal Search Volume Index

To create a measure of investor attention, GSV is transformed into an appropriate variable for the analysis. Following Da et al. (2011), the following formula is used, which creates the abnormal search volume index (ASVI). In the formula, the logarithm of SVI in week t is subtracted by the logarithm of the median of the SVI from the 8 weeks before t (Da et al. 2011).

𝐴𝑆𝑉𝐼!,# = log(𝑆𝑉𝐼#) − log [𝑀𝑒𝑑(𝑆𝑉𝐼#$%, … , 𝑆𝑉𝐼#$&)] (1)

Following Fang and Peress (2008), Bank et al. (2011) and Takeda and Wakao (2014), for every week the 21 firms are divided into three equal parts each week, based on the

abnormal search intensity of that previous week. As a result, there are three quantiles that are regarded as three portfolios. The first portfolio (Q1) contains firms with the smallest ASVI of the previous week per week, portfolio 2 (Q3) contains firms with medium ASVI of the previous week and portfolio 3 (Q3) contains the firms with the highest ASVI in the previous week.

(16)

3.2 AEX stock Data

AEX stock data has been retrieved from Yahoo! Finance, for the period between January 2014 and January 2020. The AEX is the most common stock index of The Netherlands and consists of the 25, based on market capitalization, biggest companies listed on the

Amsterdam stock exchange (Euronext Amsterdam).

Due to the fact that some companies were only listed on the AEX after 2015, and that some were delisted before 2019, some companies are excluded. Thus, stock data is obtained for 21 companies from Yahoo! Finance. Specifically, data of stocks’ weekly trading volume, as well as open price and (adjusted) close price is gathered. A list of included companies and their Google search term is shown in appendix A.

3.2.1 Stock prices and abnormal returns, trading volume and volatility

Stock prices, returns, trading volume and volatility is obtained through data from Yahoo! Finance. Adjusted stock prices, which is corrected for dividends and splits, were used to obtain daily stock prices. From the daily stock prices, the stock’s daily return is computed by dividing the adjusted close price of day d by the adjusted close price of the previous day’s

𝑅!,'= ln ( (!

(!"#) (2)

adjusted close price 𝑃'$%, and then in the natural logarithm of this. The weekly stock prices are also obtained by selecting the adjusted close price of the end of each week, which is mostly on Friday. In some occasions there is no trading on Friday and/or Thursday, in that case the last trading day of the week is used. Thus, to calculate weekly return, the natural logarithm is taken of stock price of week t for firm i divided by the stock’s adjusted close price of the week before.

𝑅!,# = ln ( ($,&

($,&"#) (3)

Abnormal returns are calculated with the Fama & French 3 factor model (Fama & French, 1993). This model is derived from the Capital Asset Pricing Model (CAPM) by

(17)

Sharpe (1964), in which excess return is calculated as shown below, where 𝑟!,# is the stock

return, 𝑟),#the risk-free rate, 𝛽% shows the sensitivity

𝑟!,#− 𝑟),# = 𝛼 + 𝛽%;𝑟*,#− 𝑟),#<+𝜀!,# (4)

of the stock to the market, and ;𝑅*,# − 𝑅),#< represents the market risk premium. Building on CAPM, Fama and French (1993) have added two more factors. The first additional factor is the SMB (small minus big), which controls for the size risk that is expressed

𝑟!,# − 𝑟),# = 𝛼 + 𝛽%;𝑟*,# − 𝑟),#< + 𝛽+𝑆𝑀𝐵 + 𝛽,𝐻𝑀𝐿 + 𝜀!,# (5)

in market capitalization. The third factor is HML (high minus low), which controls for value risk, as expressed in book to market value. The data for each of the three factors is derived from the Kenneth R. French Data Library, as there was no access to data based on the Dutch market. These risk factors act as control variables.

Following Bijl et al. (2016) and Kim et al. (2019), abnormal trading volume (ATV) is the measure variable for trading activity. The average of trading volume from the year before is subtracted from the weekly trading volume, then this is divided by the standard deviation of trading volume of the preceding year.

𝐴𝑇𝑉!,# =

-.$,&$'(#∑'($)#-.&"#

0*+,&"#, (6)

Poon and Granger (2003) calculate volatility assuming that mean return is zero, to ensure a more accurate estimation, which is also used by Bijl et al. (2016) and Kim et al. (2016). Thus, volatility is measured with daily returns 𝑟',! . By taking the square root of the

𝑆𝐷!,# = C∑2 𝑟'+

!3% (7)

sum of squared daily returns per week, the weekly volatility is measured (Andersen et al., 2016; Dimpfl & Jank, 2016).

(18)

4 Methods

Data for 21 companies listed on the AEX is collected with their appropriate Google search volume and stock data, which is used to create variables, such as ASVI, ATV, R and SD. The sample consists out of weekly data from 15March 2015 until the 22 December 2019 for 21 companies, making it a panel dataset. An overview of the companies in the sample and the search term used is presented in Table 1 of Appendix A.

First, a univariate analysis is conducted. Stock are sorted into three portfolios, as explained in section 3.1.1, based on the previous week ASVI to explore the relations between ASVI and return, trading volume and volatility (Bank et al, 2011; Takeda & Wakao, 2014). For each portfolio the average of the weekly return, trading volume and volatility is

computed. Then, comparisons of means are conducted with a statistical t-test in order to examine whether the data supports the price pressure hypothesis from Barber and Odean (2008). Thus, for the mean-comparison test, portfolio 1 (Q1) is subtracted from portfolio 3 (Q3), to find if the differences between the variables of Q3 and Q1 are significantly different from zero.

Then a multivariate analysis is conducted in order to find statistical evidence for the relations between search volume and (abnormal) return, trading volume, and standard deviation. First, the abnormal return for Q1, Q2 and Q3 is estimated with the Fama-French three-factor model, visible in eq. 5. Weekly average excess return for each portfolio, expressed in time series, is regressed on the Fama-French (1993) three-factor model. The intercept, or alpha (𝛼), represents the abnormal return. Joseph, Wintaki and Zhang (2011), Bank et al. (2011) and Takeda and Wakao (2014) all treated 𝛼 as abnormal return of the stocks or portfolio. Thus, to estimate significance of abnormal returns for each portfolio, a time series regression analysis is conducted based on the model from eq. 5. Finally, a portfolio is formed. This portfolio is based on two different strategies. Bank et al. (2011) have conducted zero-investment strategy in which they go long in stocks with high search intensity and short in stocks with small search intensity. Their results supported past results from Barber and Odean (2008) and Da et al. (2011). As this strategy also complies with the study’s hypotheses, a strategy is formed, which goes long in Q3 and short in Q1. Moreover, a second portfolio is formed according to a contrarian strategy which goes long in Q1 and short in Q3. This strategy exploits the market’s overreaction to news (Lakonishok, Vishny &

(19)

Shleifer, 1993). To examine if there is a significant abnormal return, the portfolios’s excess return is regressed on the Fama-French three-factor model (Fama & French, 1993).

Finally, for analyzing the relation between ASVI and stock’s returns, a panel

regression with fixed effects is conducted for the entire panel dataset. CAPM (Sharpe, 1964) and Fama-French (1993) three-factor model are used as models. In addition to the models,

ASVI is added as a variable. The risk factors act as control variables. In the following model,

the excess stock return is regressed against ASVI ( 𝐴𝑆𝑉𝐼!,#) as independent variable and market risk premium (𝑟*,# − 𝑟),#), SMB (𝑆𝑀𝐵#) and HML (𝐻𝑀𝐿#) as control variables. The

lagged excess return (𝐸𝑅!,#$%) is added to the model to account as an additional control

variable, to decrease omitted variable bias.

𝑅!,#− 𝑅),# = 𝛼 + 𝛽%𝐴𝑆𝑉𝐼!,#+ 𝛽+;𝑟*,# − 𝑟),#< + 𝛽,𝑆𝑀𝐵#+ 𝛽4𝐻𝑀𝐿#+ 𝐸𝑅!,#$%+ 𝜀!,#(8)

Moreover, to examine the forecasting power of the variables and to maintain a robust analysis, an additional regression is conducted with lagged variables (Stock & Watson, 2015). This is presented in the following model.

𝑅!,#−𝑅),# = 𝛼 + 𝛽%𝐸𝑅!,#$%+ 𝛽+𝐴𝑆𝑉𝐼!,#$%+ 𝛽,;𝑅*,#$%− 𝑅),#$%< + 𝛽4𝑆𝑀𝐵#$%+

𝛽5𝐻𝑀𝐿#$%+ 𝜀!,# (9)

Past studies have examined stock trading activities in different ways. For example, Bank et al. (2011) study the relation of SVI and trading activity using trading volume, stock turnover rate and illiquidity. Then, they concentrate on the effect of SVI on illiquidity with a multivariate analysis. Due to the simplicity of this study there will be a focus on stock

volume only. However, in Bank et al.’s multivariate analysis, they employ a panel regression with entity- and time-fixed effects. Even though illiquidity is the dependent variable in their regression, other studies have used similar methods to test the effect of investor attention on trading activities. For example, Kim et al. (2019) have also used panel regression, but with only entity fixed effects. Both Bank et al. (2011) and Kim et al. (2019) use lagged variables, and similar independent variables. As a result, the following

(20)

𝐴𝑇𝑉!,# = 𝛽6+ 𝛽%𝐴𝑇𝑉!,#$%+ 𝛽+𝐴𝑆𝑉𝐼!,#$%+ 𝛽,𝑅!,#$%+ 𝛽4𝑆𝐷!,#$% + 𝜀!,# (11)

two models are formed, where 𝐴𝑇𝑉!,# is the dependent variable and depicts the abnormal

trading volume. In model 10, abnormal trading volume 𝐴𝑆𝑉𝐼!,# acts as the independent

variable and return 𝑅!,# and volatility 𝑆𝐷!,# act as controls. Finally, the lagged variable of ATV is included as an additional control variable (Kim et al., 2019). Eq. 11 contains the same variables as eq. 10, but lagged. The variables are lagged in order to test the forecasting ability of the model (Kim et al, 2019). The results are concluded from panel data regressions with fixed effects.

Volatility is modeled in a similar way to abnormal trading volume, due to the control variables. Aouadi et al. (2013) and Kim et al. (2019) both performed time series regression to study the effect of investor attention on volatility, which is represented by the standard deviation of the stock’s returns as calculated in eq. (7). Following Kim et al. (2019), the following two models are formed, in which eq. 13 contains the lagged variables of eq. 12.

𝑆𝐷!,# = 𝛽6+ 𝛽%𝐴𝑆𝑉𝐼!,-+ 𝛽+𝑅!,#$%+ 𝛽,𝐴𝑇𝑉!,#+ 𝛽4𝑆𝐷!,#$%+ 𝜀!,# (12)

𝑆𝐷!,# = 𝛽6+ 𝛽%𝑆𝐷!,#$%+ 𝛽+𝐴𝑆𝑉𝐼!,#$%+ 𝛽,𝑅!,#$%+ 𝛽4𝐴𝑇𝑉!,#$% + 𝜀!,# (13)

Thus volatility (𝑆𝐷!,#) is the dependent variable and represents the standard deviation of stock

i return at time t. 𝐴𝑆𝑉𝐼!,#($%)is the (lagged) independent variable which represents investor

attention, 𝑆𝐷!,#$%is the lagged volatility. Finally, (lagged) return (𝑅!,#($%)) and (lagged)

abnormal trading volume (𝐴𝑇𝑉!,#($%)) are added as control variables. Again, the results are concluded from panel data regressions with fixed effects.

Finally, robust standard errors are used for every analysis for more reliable results. (Stock & Watson, 2015).

(21)

5 Results

5.1 Descriptive statistics and correlations

Table 5 shows the descriptive statistics for the assessed variables. ASVI is based on eq. 1, weekly return is based on eq. 3, and ATV and volatility are based on eq. 6 and 7 respectively. ASVI and weekly return both show a mean close to zero (=.001). ATV and volatility have a higher mean, namely 4.741 and 2.213 respectively. Moreover, ASVI, weekly return and ATV are positively skewed, meaning that the right-side tails are fatter. It should be noted that skewness for weekly return (=1.528), ATV (=16.000) are higher than 1 and volatility (= -3.337) lower than -1. These values are all significant (p < 0.05). However, according to Field (2017), values are often significant if the sample is large. As a result, the central limit

theorem still holds (Field, 2017). Moreover, a high kurtosis indicates heavy tails. This can indicate the presence of outliers.

Table 6 shows the correlation between the variables. For most variables the

correlation is close to zero. However, volatility and ASVI (r =.190) and volatility and weekly return (r = .224) show higher correlations. Moreover, there is a negative correlation between weekly return and abnormal trading volume (r = -.0066). In accordance to our hypotheses, ASVI is related with weekly return, ATV and volatility.

(22)

5.2 Main analysis

To assess the stock’s average characteristics based on portfolio Q1, Q2 and Q3, a univariate analysis is first conducted. Table 7 shows the average characteristics for Q1, Q2 and Q3. For Q3, the average return, ATV and volatility are highest in comparison to Q2 and Q1. For Q1, the average return, ATV and volatility is lowest. Moreover, a t-test is conducted to assess whether the difference between the firm characteristics of Q3 and Q1 is significant. Even though the differences are positive, they are all not significantly different from zero (p > .05).

Next, a regression analysis is performed and presented in table 8. Each portfolio’s excess return is regressed against the market premium, SMB and HML. The 𝛼 represents the abnormal return for each portfolio, and identifies whether the abnormal return is significantly different from zero. For Q1, Q2 and Q3, the intercept is -.022, -.021 and -.020 respectively. They are all negative but significant (p < .01), implying that that the abnormal return is significantly different from zero. Even though the difference is small (.002), the abnormal return for Q3 is higher than the abnormal return for Q1 and Q2. The coefficient for market risk premium for Q1, Q2 and Q3 are (almost) equal to .01 and are significant (p < .01). The coefficient for SMB is only significant for Q2 (𝛽+= .009, p < .05). The coefficient for SMB for Q1 and Q3 both equal .0002 and is not significant (p > .1). Further, the coefficient 𝛽, for

(23)

Finally, there are two portfolios presented in table 8, Q3 – Q1 and Q1 – Q3, in which

Q3 – Q1 goes long in stocks weekly sorted in Q3 and short in stocks weekly sorted in Q1 and

vice versa. Q3 – Q1 (𝛼 = -.02, p < .01) and Q1 – Q3 (𝛼 = -.023, p < .01) both have a negative intercept, which represents the portfolio’s abnormal return. The abnormal returns of the two portfolios are significantly different from zero. Additionally, Moreover, the coefficients for the three risk factors are all insignificant.

For further analysis for the relation between ASVI and excess return panel data regression with fixed effects has been performed for the complete dataset based on eq. 8 and eq. 9. The estimates of the coefficients of both models are shown in table 9. Model 1 in table 9 depicts the CAPM and abnormal search volume index ASVI as an added explanatory factor, or independent variable, for return. The coefficient for ASVI equals .015 and is significant (p < .05), implying that an increase in ASVI has a positive effect on return. Moreover, the intercept (𝛼= -.022) and the market risk premium (𝛽%= .01) are both significant at the 1%

significance level. After the addition of SMB and HML as risk factors, ASVI remains equal, and R2 increases by .01, indicating that the variance of return explained by the model has improved. Finally, when the lagged excess return is added as visible in model 3, ASVI remains equal again. R-squared increased again, by .027.

(24)

In models 4 and 5 the lagged variables were used, to test if the variables have

predictive power on excess return. In both models, 𝑆𝑀𝐵, 𝐻𝑀𝐿 and ER were significant (p < .01). However, the variable of main importance, namely ASVI, did not show any significant effect on return in model 4 and 5. It is noticeable that, despite being insignificant, the coefficients for ASVI were negative in model 4 and 5. The R-squared in both models are considerably lower than the previous models, indicating that less of the return’s variance is explained by the independent variables.

Due to the results reported in table 8 and 9, it can be concluded that there is a

significant relation between ASVI and excess return. Negative abnormal returns were shown in table 8, which contradicts the hypothesis. On the other hand, the returns were still higher when ASVI was higher. Moreover, the results from table 9 showed that there was a significant positive relation between ASVI and return. Thus, hypothesis 1 is accepted.

The results of the analysis on the relation between GSV and trading volume are presented in Table 10. Panel data regressions with fixed effects are performed, which test the explanatory power of ASVI on ATV in models 1 to 4 and the predictive power of ASVI on

ATV in models 4 and 5. Model 1 only shows control variables R and SD as independent

variables. Both coefficients of the control variables showed insignificant returns. The independent variable ASVI is added in model 2, to test if there is any model improvement. However, all coefficients are insignificant (p > .1). R-squared remains equal, meaning that

(25)

only .1% of the variance in ATV is explained by the independent variables. In model 3 the lagged ATV is added as a control variable, and shows significant results (p < .01). The addition of this control variable also increases the R-squared considerably by 82.3%. In model 4, ASVI is included to see if there is any model improvement. Noticeably, the

coefficient of ASVI is significant at the 10% significance level, which is identified as a weak significance level for this study. R-squared remains equal to model 5, indicating that the inclusion of ASVI to model 6 does not lead to a better explanation of the variance of return. The coefficients reported in model 5 and 6 only show significant results for the lagged ATV. Indicating that ASVI does not forecast ATV.

The null hypothesis states that ASVI has no effect on ATV, or more generally that investor attention has no effect on trading volume. As a result of the weak relation between

(26)

Lastly, the results of the panel data with fixed effects regression conducted to study the relation between volatility and GSV is reported in table 11. SD is the dependent variable for each model. Model 1 depicts the control variables R and ATV as independent variable, and both showed to be significant (p < .01). Model 2 shows a slight improvement over model 1, due to the addition of ASVI. All coefficients are significant (p < .01) and R-squared showed an increase of 3.2%. The lagged variable of ASVI is added as a control variable in model 3 and is also significant (p < .01). Finally, the complete model is tested in 4. There is no increase in R-squared from model 4 to model 3. However, ASVI is positive and significant (p < .01), indicating that when ASVI increases by one unit, SD increases by .093. In addition, the coefficient of lagged ASVI is significant (p < .05) and positive in model 6. As R-squared increases by .1% from model 5 to model 6, it can be concluded that ASVI can predict future SD.

(27)

6 Discussion

This section further discusses the relation between stock return, trading volume, volatility and investor attention, by providing interpretations, implications, limitations and

recommendations.

This study focused on investor attention with Google search volumes as a proxy. A main focus was on the effect Google search volumes have on AEX stock prices, trading volume and volatility. The results confirm that Google search volume has a positive and significant effect on stock prices and standard deviation. The study does not show a significant relation between Google search volume and trading volume.

Firstly, in accordance with Bank et al. (2011) and Takeda and Wakao (2014), stocks with the highest Google search volume in the previous week scored higher in return,

abnormal trading activity and volatility than stocks with lowest search volume in the same week. However, these differences were not significant, thus no implications can be made. Both studies employed a larger sample containing considerably more stocks than this study, which could explain why the results are not significant. Kim et al. (2019) studied a smaller stock index in Norway and did not find a significant relation between stock return and Google search volume. Moreover, many studies used either the US stock indices for their sample, or other larger stock markets, such as Japan (Takeda & Wakao, 2014) and Germany (Bank et al., 2011).

The trading strategy suggested a negative abnormal return. The abnormal return of both portfolios is significantly different from 0, implying that the portfolios performed worse than expected, estimated from risk factors. However, the negative abnormal return could be due to other reasons. For example, the data for the risk factors are based on the U.S. market, rather than the Dutch market. Country-specific economic circumstances, such as inflation, could make the U.S. data less suitable for the Dutch market, possibly explaining the negative abnormal return. Again, it can also be due to the smaller firm size, as the portfolios consisted of far less stocks than Bank et al. (2011).

The study’s multivariate analyses showed significant results in accordance with Barber and Odean’s (2008) price pressure theory as hypothesized, implying that there is a

(28)

positive relation between investor attention and AEX stock returns. Contradicting, the forecasting model showed a negative, but insignificant, relation between Google search volume and future return, in accordance to Bijl et al.’s (2018) suggestions. The positive and significant relation between search volume and volatility of the AEX stocks were in line with the hypothesis, supporting Aouadi et al. (2013) and Dimpl and Jank (2015).

In contrast to the hypothesis, no relation was found between trading volume and Google search volume. Again, this can be due to the smaller stock market in The

Netherlands. Moreover, many studies suggest a moderating effect of firm size. This could be a possible explanation, due to the different firms on the AEX, where some are much larger than others.

The conclusions on return and Google search volume build on the theory of Barber and Odean (2008) and Da et al. (2011). It provides new inside by employing a different stock market, namely the Amsterdam Exchange Index. Conclusions on Google search volume and volatility are in accordance with existing evidence from Black (1986), Vlastakis and

Markellos (2012), Dimpl and Jank (2016). On the other hand, no relation was found between trading volume and Google search volume, unlike the study’s expectations and past studies.

One limitation of this study is the smaller sample in comparison to other studies, which mostly employ the U.S. stock market. Moreover, the data on Google search volume was limited to Dutch investors, targeting only a small group of people. As a result, this could lead to low Google search volume, which could influence the results. Also, different

measures for Google search volume, trading volume and volatility could improve the robustness of the study, and provide more possible insights.

Lastly, suggestions for future studies are, while still focusing on the Dutch stock market, considering a larger group of investors, so that there is an improvement in Google search volume data. Moreover, adding variables such as firm size can provide possible better insights into the relationship between AEX stock returns, trading volume, volatility and Google search volume. Moreover, considering the temporary nature of the relationship between AEX stock returns, trading volume, volatility and Google search volume, studying the long-term relation could provide more in-depth insights of the effect of investor attention on stock valuation.

(29)

7 Conclusion

Individual investors are faced with a wide availability of information but only a limited amount of attention to give. Da et al. (2011) have employed a novel direct measure of investor, namely Google search volume. This research aimed to study the effect of Google search volume on stock prices, trading volume and volatility. Most papers study the U.S. stock market (Da et al., 2011). However, this study focuses on the Dutch stock market, which creates new insights into the external validity of the price pressure theory. Based on an empirical analysis, it can be concluded that Google search volume affects stock prices and volatility. The findings suggest a positive relation between investor attention and AEX stock returns, and investor attention and volatility. No significant relation was found between GSV and trading volume. By employing different analyses more insight is gained into the panel dataset. First, portfolios are formed based on intensity on GSV, on which an investment strategy is analyzed. Then, by conducting panel data regressions with fixed effects, insight is provided into the entire dataset, from which findings are concluded. While significant relations are found, some questions are raised on the sample, and omitted variables, such as firm size. Further research is needed to determine whether the results also apply to investors outside of the Netherlands, or to find a possible moderating effect of firm size. However, the findings of this study are, even when focusing on a novel stock market, in accordance with conclusions from Barber and Odean (2008) and Da et al. (2011).

(30)

8 References

Aouadi, A., Arouri, M., & Teulon, F. (2013). Investor attention and stock market activity: Evidence from France. Economic Modelling,35, 674-681.

doi:10.1016/j.econmod.2013.08.034

Bank, M., Larch, M., and Peter, G. (2011). Google Search Volume and its

influence on liquidity and returns of German stocks. Financial Markets and Portfolio Management, 25(3), 239-264.

Barber, B., & Loeffler, D. (1993). The "Dartboard" Column: Second-Hand Information and Price Pressure. The Journal of Financial and Quantitative Analysis, 28(2), 273-284. doi:10.2307/2331290

Barber, B. M., & Odean, T. (2012). All that glitters: The effect of attention and news on the buying behavior of individual and institutional investors. The Handbook of News Analytics in Finance,173-210. doi:10.1002/9781118467411.ch7

Berk, J. B., & DeMarzo, P. M. (2017). Corporate finance. Boston: Pearson.

Bijl, L., Kringhaug, G., Molnár, P., & Sandvik, E. (2016). Google searches and stock returns. International Review of Financial Analysis,45, 150-156.

doi:10.1016/j.irfa.2016.03.015

Black, F. (1986). Noise. The Journal of Finance, 41(3), 528-543. doi:10.1111/j.1540- 6261.1986.tb04513.x

Bodie, Z., Kane, A., & Marcus, A. J. (2021). Investments. New York, NY: McGraw-Hill Education.

Bollen, L. H. H., Geerings, J. & Hassink, H. F. D. (2003) Investor relations on the Internet: a survey of the Euronext zone, European Accounting Review, 12(3), 567-579, DOI: 10.1080/0963818032000083531

(31)

Bordino, I., Battiston, S., Caldarelli, G., Cristelli, M., Ukkonen, A., & Weber, I. (2012). Web Search Queries Can Predict Stock Market Volumes. PLoS ONE, 7(7).

doi:10.1371/journal.pone.0040014

Clement, J. (2020, March 25). Search engine market share worldwide 2019. Retrieved from https://www.statista.com/statistics/216573/worldwide-market-share-of-search-engines/

Coval, J. D., & Moskowitz, T. J. (1999). Home bias at home: Local equity preference in domestic portfolios. The Journal of Finance, 54(6), 2045-2073

Da, Z. Engelberg, J., and Gao, P. (2011). In search of attention. The Journal of Finance, 66 (5), 1461-1499

Euronext Amsterdam - AEX. (n.d.). Retrieved June 27, 202, from https://live.euronext.com/en/markets/amsterdam

Fama, E. F. (1970). Efficient Capital Markets: A Review of Theory and Empirical Work. The

Journal of Finance, 25(2), 383. https://doi.org/10.2307/2325486

Fama, E. F., & French, K. R. (1993). Common risk factors in the returns on stocks and bonds. Journal of Financial Economics, 33(1), 3-56.

doi:10.1016/0304-405x(93)90023-5

Fang, L. H., & Peress, J. (2008). Media Coverage and the Cross-Section of Stock Returns. SSRN Electronic Journal. doi:10.2139/ssrn.971202

Field, A. (2017). Discovering statistics using IBM SPSS statistics: North American edition. sage

French, K. R., Schwert, G., & Stambaugh, R. F. (1987). Expected stock returns and volatility. Journal of Financial Economics, 19(1), 3-29.

(32)

Grullon, G., Kanatas, G., & Weston, J. P. (2002). Advertising, Breadth of Ownership, and Liquidity. SSRN Electronic Journal. doi:10.2139/ssrn.304240

Joseph, K., Wintoki, M. B., & Zhang, Z. (2011). Forecasting abnormal stock returns and trading volume using investor sentiment: Evidence from online search. International

Journal of Forecasting, 27(4), 1116-1127. doi:10.1016/j.ijforecast.2010.11.001

Kahneman, D. (1973). Attention and effort. Englewood Cliffs: Prentice-Hall.

Kim, N., Lučivjanská, K., Molnár, P., & Villa, R. (2019). Google searches and stock market activity: Evidence from Norway. Finance Research Letters,28, 208-220.

doi:10.1016/j.frl.2018.05.003

Lakonishok, J., Vishny, R., & Shleifer, A. (1993). Contrarian Investment, Extrapolation, and Risk. doi:10.3386/w4360

Lintner, J. (1965). The Valuation of Risk Assets and the Selection of Risky Investments in Stock Portfolios and Capital Budgets. The Review of Economics and Statistics, 47(1), 13. doi: 10.2307/1924119

Long, J. B., Shleifer, A., Summers, L. H., & Waldmann, R. J. (1990). Noise Trader Risk in Financial Markets. Journal of Political Economy, 98(4), 703-738.

doi:10.1086/261703

Merton, R. C. (1987). A Simple Model of Capital Market Equilibrium with Incomplete Information. The Journal of Finance, 42(3), 483–510. doi: 10.1111/j.1540-6261.1987.tb04565.x

Peng, L., & Xiong, W. (2006). Investor attention, overconfidence and category learning.

Journal of Financial Economics, 80(3), 563–602. doi: 10.1016/j.jfineco.2005.05.003

Preis, T., Moat, H. S., & Stanley, H. E. (2013). Quantifying Trading Behavior in Financial Markets Using Google Trends. Scientific Reports, 3(1). doi:10.1038/srep01684

(33)

Sharpe, W. F. (1964). Capital Asset Prices: A Theory of Market Equilibrium under Conditions of Risk. The Journal of Finance, 19(3), 425. doi: 10.2307/2977928

Stock, J. H., & Watson, M. W. (2015). Introduction to Econometrics

Sullivan, D. (2006). Search Engine Strategies Conference. Retrieved from https://www.google.com/press/podium/ses2006.htm

Takeda, F., & Wakao, T. (2013). Google Search Intensity and Its Relationship with Returns and Trading Volume of Japanese Stocks. SSRN Electronic Journal.

doi:10.2139/ssrn.2332495

Van Couvering, E. V. (2008). The History of the Internet Search Engine: Navigational Media and the Traffic Commodity. Web Search Information and Science and Knowledge

Management, 177-206. doi: 10.1007/978-3-540-75829-7_11

Vlastakis, N., & Markellos, R. N. (2012). Information Demand and Stock Market Volatility. SSRN Electronic Journal. doi:10.2139/ssrn.1558434

Yuan, Y. (2008). Attention and trading. Financial Institutions Center, Wharton School, University of Pennsylvania.

(34)

Referenties

GERELATEERDE DOCUMENTEN

According to Chuang and Susmel (2011), the Taiwanese stock market shows evidence that individual investors trade with more overconfidence than institutional

(2011), the correlations of SVIs downloaded at different points of time are greater than 97%. Therefore, the effect of different download time can be ignored. And the maximum

The descriptive data of the variables, in Tables 1, 2, 3 and 4 show that the Internet crisis and financial crisis periods are different than the whole sample period

Where the turnover ratio is measured as the absolute trading (numerator) volume divided by the number of total outstanding shares (denominator) Following Table

- In 2015 “practically all” teachers of the participating schools are sufficiently able to provide differentiation in their teaching and deal with individual

45 Nu het EHRM in deze zaak geen schending van artikel 6 lid 1 EVRM aanneemt, terwijl de nationale rechter zich niet over de evenredigheid van de sanctie had kunnen uitlaten, kan

We implemented an algorithm based on a machine-learning approach to learn the level of engagement of the audience in such a way that it can be possible to measure

Maar het antwoord dat het meeste voor komt, is dat spiritualiteit iets is waar ze altijd mee bezig zijn en niet iets is ‘wat je er extra bij doet’.. Evelien zegt bijvoorbeeld dat