Attention and ownership structure : the effect of a share’s popularity on shareholder structure

(1)

1

Attention and Ownership Structure

The effect of a share’s popularity on shareholder structure

Quirijn Abel Knab

Supervisor: dr. F.S. Peters

Email: quirijn.knab@gmail.com

(2)

2

Statement of Originality

This document is written by Quirijn Knab who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is 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 completion of the work, not for the contents.

(3)

3

1. Introduction

In today’s world it is tempting to invest in stocks and shares instead of saving your money at a bank. With interest rates dropping below 1% on most savings accounts in the Netherlands more and more people are exploring the possibility for investing their savings money in shares. A question often asked is where to invest one’s money. People would the layman recognize a company named “Novisource N.V.” for instance? The hypothesis in this thesis is that individuals tend to invest in companies they know; large, attention grabbing companies. A possible effect of this tendency is that it is easier for these large, attention grabbing companies to exploit these individuals through poor corporate governance.

As Barber and Odean (2008) already have pointed out, amateur investors like the ones just described face major problems: they lack the knowledge, time and instruments to acquire the information needed for an investment decision. Therefore instead of choosing stock by carefully weighing their options, they buy a company’s share that grabs their attention by appearing in the news, or because the company produces products that everyone uses. Barber and Odean (2008) have researched this theory in their paper and confirmed it.

The aim of this thesis is to confirm this theory for the Netherlands. Because the rules and laws governing investment in the Netherlands are different, the outcomes of the results in this thesis are unsure. Moreover, a different, more inclusive dataset is available for the

ownership structures of companies incorporated within Dutch law. If Barber and Odean’s theory is confirmed, it could have large implications for corporate governance in the

Netherlands. If the theory is not confirmed, it contradicts earlier research and gives reason to take a closer look at the theory.

Barber and Odean’s theory is that because individual investors tend to buy shares in companies that are the center of attention, the ownership of these companies will be more dispersed - their shares will be held by a higher fraction of individual investors. Many researchers have already confirmed the theory that when a company is owned by a higher fraction of individual investors it is an opportunity for the directors of the company to exploit this situation in their favor. This means that the company is not run in the best interest of the owners of the company, the shareholders, which means that there is a case of bad corporate governance.

The expectation of the outcome of these regressions is that the amount of attention a firms gets explains its ownership structure. To be more specific, the more attention a firm

(5)

5 gets, the more dispersed the ownership of the firm is, because attention causes individual investors to buy the firms stock. Because this theory has already been confirmed in several other studies this hypothesis is not surprising. However, there are several differences between this thesis and existing literature. First of all, the dataset is different. Other research has focused mainly on US data, whereas here Dutch data is used. For the hypothesis this makes no difference; a logical expectation is that investors in the US behave in the same manner as investors in the Netherlands, whether individual or institutional. There are differences between the US and the Netherlands however. First of all the dataset used is smaller because there are fewer companies enlisted on the Dutch stock market. Secondly the companies included in the database are different. Because in research using a US database the luxury exists to include only the largest companies in the research (the S&P 500 for example) the dataset for the proxies for attention is more inclusive. The final difference between using a US dataset and a Dutch dataset is that the laws in both countries for trading in stock are different. One of the major differences is that in the Netherlands foundations owning a certain

company’s shares are allowed to certify these shares and sell these certified shares as certificates. The foundation and the company whose shares are owned by the foundation are in most cases closely related and the foundations are in most cases controlled by the same directors controlling the company.

The second difference between this thesis and existing research is that in this thesis several parameters are used as proxies for attention, whereas in other research mainly one is used. The parameters used as proxies for attention in this thesis are trading volume,

advertising expenses and the Search Volume Index. This should yield better outcomes. Lastly the method of measuring the shareholder structure is different from the existing literature. This difference will be discussed later in this thesis.

The contribution to the existing literature is clear. Because in this thesis several proxies for attention are used, instead of one, all proxies are tested. Moreover, the existing theories are tested for the Netherlands, which has some interesting differences from US data. Lastly, a more complete manner to measure shareholder structure is applied in the regressions. If the results show that under these circumstances there is a relation between attention and shareholder structure, further research could explore the effect this has on the corporate governance of the companies. If there is no relation between attention and shareholder structure follow up research could be needed to examine and refine the theory even more.

(6)

6 The methodology that is used is as follows. Using panel data to control for time and company fixed effects, ownership structure is regressed on attention. Because attention is difficult to grasp several proxies for attention are used. These are trading volume, advertising expenses and the Search Volume Index (SVI). The dataset for ownership structure is obtained from the Autoriteit Financiële Markten (AFM), which is the Dutch authority for financial markets. By manipulating this data the ownership structure can be assessed and quantified. The data on trading volume and advertising expenses are obtained through Datastream. The SVI is generated by hand using the data available from www.google.com/trends. Paragraph 3C will explain the mechanism behind this.

In the second section of this thesis the existing literature relevant to this topic is addressed. The main theories about what causes ownership structure and the relationship between ownership structure and attention are thoroughly examined. It also explain how this thesis precisely contributes to the existing literature. In the third section the methodology is discussed. In this section the dataset and model are addressed. In the fifth and sixth sections of this thesis the results are presented and the robustness is checked. In the results sections tables are provided with information about the regression. The economic consequences of the results are also explained. To conclude this section the results are placed into context within the existing literature. In the seventh and final section of this thesis a conclusion of the research is given. A brief summary is given and the limitations and implications of the research are discussed. Lastly some directions for future research are suggested.

2. Literature review

A. Shareholder Base

The main data of this thesis is that of the shareholder base. As per AFM (Autoriteit Financiële Markten) regulations, it is mandatory for shareholders to enter their shareholdings from up to 3% in NV’s, which are the Dutch equivalent of limited companies, there is a unique

possibility to map the shareholders holding a large proportion of the stock of publically traded companies. These are in most cases institutional shareholders and founding shareholders, or the relatives of founding shareholders. The Dutch government has made these regulations to improve financial transparency and protect shareholders. The register is publically available and can be found at www.afm.nl.

(7)

7 Because in most other countries these kind of registers do not exist in this manner it is interesting to see how other research on shareholder structure have estimated the shareholder base. Barber and Odean (2008) have attempted to estimate the difference between individual and institutional shareholders by comparing the investors of four different brokerages. These are a large discount brokerage, a small discount brokerage, a large retail brokerage and professional money managers. The assumption made is that small individual investors invest their money at the large discount brokerages, while institutional investors use the professional companies that prioritize quality. Barber and Odean estimate the shareholder base by looking at the amount of stock that is acquired at an attention day at a certain brokerage as compared to a non-attention day. Their results show that at an attention day more stock is bought at brokerages where individuals are assumed to invest their money, relative to the brokerages where professionals would go and vice versa.

Ding and Hou (2015) use the number of shareholders to measure the shareholder base. Although they do not give a precise explanation for why they use this proxy to measure the shareholder base, it seems that this method is inferior when investigating the effect of attention on corporate governance, because it does not say anything about which percentage of shares is owned by which shareholders, i.e. it is impossible to tell something about shareholder power. For instance, when a company has ten shareholders there is a possibility that each one owns 10% of the company, or that one shareholder owns 91% of the company and the other nine shareholders own only 1% of the company. In the latter example one shareholder has a lot of shareholder power and the other nine have almost none, while in the former example every shareholder has the same amount of shareholder power. This example is similar to the idea behind the Herfindahl index, which will be explained later on in this thesis.

The last method to discuss is the method Lou (2014) and Drake, Roulstone and Thornock (2012) use to measure the shareholder base. In Lou’s (2014) paper on the effect of advertising expenses on shareholder structure and Drake, Roulstone and Thornock’s (2012) paper on the Search Volume Index the CDA/Spectrum database is used from which they obtain holdings from financial institutions with greater than $100 million under management. This database exists because these kind of holdings need to be reported to the SEC. The reports are made quarterly or semiannually and holdings in common stock of more than 10,000 shares or $200,000 are to be disclosed. This method is very similar to the method used in this thesis. However, in the dataset in this thesis, shareholdings of founding shareholders or

(8)

8 relatives are also recorded, as opposed to the CDA/Spectrum database in which only holdings of large enough financial institutions are reported. Although the latter are not necessarily professionals, it is still important to include them in the dataset, because first of all they do not own the shares because they grab their attention and secondly they will monitor the firm differently from individual shareholders that do not have such a tight relation to the company.

B. Attention

This thesis tests the hypothesis that the amount of attention a share gets has an impact on the shareholder structure of the company that issued the shares. In past literature on finance the assumption was made that information is available and more importantly, usable for

everyone. Merton (1987) however showed that this is not the case. Because only certain types of information catch an investor’s attention, the rest of the information remains unused (Kahneman, 1973). It is interesting to find out which kinds of information reaches investors. When making decisions people value information that stands out and captures attention more than information that does not do this (Tversky and Kahneman, 1973, Grether, 1980).

Odean’s (1998) paper on overconfidence in market traders gives a precise description of the individual investors. The individual investor tends to base his opinion on information that is more accessible to him (Clark and Rutter, 1985) instead of dry statistical data (Tversky and Kahneman, 1973). Merton (1987) states very correctly that when an individual investor wants to invest in a company he must at least know of this company’s existence. Measures to

capture attention are discussed in the remainder of this section.

Barber and Odean (2008) propose a model in which information is limited for small and individual investors. Therefore these investors tend to focus on information that comes to their attention. This causes small and individual investors to buy attention-grabbing stocks, while professional investors seem less prone to make attention-driven purchases. The reason for this is that professional investors have more time to find the best stocks to buy, and also have more knowledge and better resources. Barber and Odean (2008) use abnormal trading volume as a measure for attention, because when there is a large amount of trade in a stock on a particular day, this should catch the attention of the small and individual investors. Secondly they use the fact of whether or not a firm is in the news as a measure for attention, because the news is an easy accessible source of information for small and individual investors. This will surely grab most of their attention. The results of Barber and Odean (2008) show that on high

(9)

9 volume days individual investors are net buyers. This suggests that individual investors’ buying decisions are driven by attention. Furthermore, when stocks are in the news it also has the effect of individual investors buying these stocks.

Lou (2014) finds that a company increasing its advertising expenses results in a rise in the amount of stock of the company bought by individual investors. Grullon, Kanatas and Weston (2004) get similar results as Lou. Their results suggest that a firm increasing their advertising expenses leads to their shareholdings becoming more scattered; the effect being larger for individual investors, but still positive for institutional investors. Although

advertising usually has the purpose of increasing awareness of the products of the company, it also has an effect on awareness of the shares of the company. The explanation for this is that when a company invests significantly in advertising, individual investors will have a good chance to get familiar with the stock and thus have a higher probability to invest in that stock. Lou (2014), agreeing with Barber and Odean, argues that because of the constraint individual investors have when searching for stocks to buy, advertisements are an easy, accessible way to get individual shareholders’ attention. Because the information individual investors have at hand is limited, they use information that comes easily to them to make their investment decisions. A company’s advertisements are an example of information that comes easily to the individual investor described in the introduction. As was stated earlier Lou uses a different method to measure institutional- and individual holdings from Barber and Odean. Lou (2014) finds that an increase in a company’s advertisement expenses of one standard deviation leads to an increase in individual shareholdings of more than 1,4% in that stock.

A more contemporary way to measure the amount of attention of investors a company attracts is to use the Search Volume Index (SVI). The SVI can be found at

www.google.com/trends and tells the relative amount of searches for a certain search term. The SVI is constructed by scaling all the search terms and, if necessary, computing monthly results into yearly data. In the rest of this paragraph and paragraph 3C I will explain the construction of the SVI more thoroughly; now I will continue explaining how the SVI is used in other research so far.

Because the internet has become a bigger factor in daily life, investors find using the internet an increasingly convenient tool when searching for news about the stock market. Research has shown that two weeks prior to an earnings announcement information demand for the company making the announcement rises. The peak of the information demand lies on the day the announcement is made. At this day the SVI is 13.2% higher than normal. After the

(10)

10 announcement it takes roughly two weeks for the SVI to return to normal levels again. This effect only arises when the announcement is expected. These findings of Drake, Roulstone and Thornock (2012) suggest that investors seek and process the information they find using Google before and after the announcement. Da, Engelberg and Gao (2011) show that looking at the SVI is an easy but convenient way to see whether individuals pay attention to stocks or not. Ding and Hou (2015) have used this index to see how much attention investors pay to S&P500 stocks and with that information they tested if it had any influence on the shareholder base. They found that a high SVI correlates with a broader shareholder base, thus confirming Barber and Odean’s (2008) theory that attention has an influence on the structure of the shareholder base. Da, Engelberg and Gao (2011) found similar results. Their results suggest that a rise in SVI is associated with a higher trading volume of investors. They also find that this effect is the largest in markets where less professional investors trade, suggesting that the effect is the largest for individual investors.

The advantage of using the SVI as a measure of attention over measures such as media coverage and advertising expenses is that the latter, so called passive attention measures, have showed in the past to be sometimes less significant in the results. Da, Engelberg and Gao (2011) explain this very well, stating that “a wealth of information creates a poverty of attention”. This implies that advertisements on television or articles in the newspaper do not necessarily imply that investors see or read these articles, meaning that passive attention measures are not necessarily reaching investors or playing a role in the decision-making process when buying stocks (Ding and Hou, 2015). When an investor searches on Google for information about stocks it means that he really wants to find investment information, which therefore is a direct way to measure investors’ attention (Da, Engelberg and Gao, 2011).

The SVI can be found at www.google.com/trends. When the term is entered for which the SVI is requested, Google Trends will return the results. Weekly data of searches made in Google is available, but when there have been few searches with a particular search term Google Trends provides monthly data. Da, Engelberg and Gao (2011) as well as Ding and Hou (2015) argue that the stock ticker is the most appropriate search term to use to gauge investor attention. They explain that an investor may either search for a company’s stock by using its name or its ticker. Because when the company’s name is used the searches for, for instance, information on the company’s products or job openings interfere with the searches from investors, using the company’s ticker is more suitable for this type of research (Da, Engelberg and Gao, 2011). Drake, Roulstone and Thornock (2012) agree with this, stating

(11)

11 that searching for a company’s ticker symbol gives better chances to find investment related searches.

They find that there is a positive relation between the SVI and stock return

magnitudes. This fits the theory that investors search for additional information when such news reaches them. Another relation they find is that when liquidity of a stock is low, investors search more for the stock. This means that when it is hard to trade in a stock investors search for more information about it on Google.

Because trading is often one of the things investors look at while analyzing their portfolio, this also is an indicator for investor attention. The rationality is that when investors have more attention for a stock they also trade more often in it, because of heterogeneous opinions among investors about the stock (Scheinkman and Xiong, 2003, Odean, 1998). When investors do not pay any attention to a stock they will also trade less in it (Hou, Xiong and Peng, 2009). Lo and Wang (2000) argue that trading volume is higher among large stocks, and because large stocks also have more attention, trading volume is a good proxy for this. Because returns of high volume portfolios are higher than returns of low volume

portfolios there is evidence of high volume stocks getting more attention than low volume stocks, when controlled for firm size (Chordia and Swaminathan, 2000). Moreover, Gervais, Kaniel and Mingelgrin (2001) find that when a stock has a large increase in volume, a month later the prices of this stock rise, suggesting that attention to this stock has increased. A stock’s trading volume is also thought to be correlated with news about that stock, because when investors get the news about the stock they are more likely to trade in it (Barber and Odean, 2008). Although they state that news does not necessarily mean that investors use it, significant news will cause investor opinions to be heterogeneous and trading volume will rise. As they also made clear individual investors are more likely to trade in attention grabbing stocks than professional investors.

An attention measure recently proposed is the dividend payout ratio. The economic theory behind this is that individual shareholders want to make high profits easily and therefore look at shallow data such as how many dividends they expect to receive. They also will not be trading their stock as actively as professional investors and they will therefore think they will enjoy their dividend rights. Literature on this topic is contradictory however. On the one hand Harada and Nguyen (2011) find that dividend policy in fact is correlated with a lower ownership concentration in Japan, on the other hand Short, Zhang and Keasey (2002) find an opposite relationship for the UK. They explain this reasoning that the UK deals

(12)

12 with a relatively high institutional investor ownership. These institutional investors demand higher dividends to monitor the board and reduce free cash flows available to management (Jensen, 1986). The dividend payout ratio could therefore either be an attention factor or a control variable, depending on the outcome of the results from this thesis.

3. Methodology and Data

A. Ownership Structure

In this thesis the relationship between attention and shareholder structure is tested. The data of the shareholder structure, is the main data used. This data is gathered from a database

provided by the Autoriteit Financiële Markten (AFM, 2015), which is the Dutch authority concerning financial markets. This database contains information about substantial ownership of shares in NV’s, which are the Dutch equivalent of limited companies, incorporated under Dutch law. To clarify, this means that this database not only contains information of Dutch companies but also of non-Dutch companies incorporated under Dutch law. These companies are mostly from other EU-member states because EU-law makes it possible for companies to establish a business using a legal framework of any member state.

When a NV’s shareholder passes a threshold of 3% ownership he or she needs to report this. A new notice has to be given every time the shareholder reaches, passes or undershoots another threshold. These thresholds are 3%, 5%, 10%, 15%, 20%, 25%, 30%, 40%, 50%, 60%, 75% and 95%.

Hence, to make this dataset ready for usage the dataset must be reduced to only the last recordings of a shareholder’s filings. When the dataset is manipulated in the right way, it will contain all the yearly substantial shareholdings of Dutch NV’s. This is done for the years 2007 until 2015. After the adjustment the dataset only includes shareholdings of common stock, which means that holdings of warrants, American depository receipts (ADR’s) and options are excluded. When a shareholders also owns other financial instruments next to his common shares these other instruments are left out of the data and his total stake in a company is adjusted so that it is correct. This makes sure that only the instruments that actually have voting power are present in the database. When shares are owned indirectly through the same company the stake is also adjusted. This leaves the dataset with 468 to 1134 owners of 117 to 143 companies varying from year to year. 1159 observations in total over

(13)

13 the 9 year period are available for the regression. The average of total ownership by

substantial shareholders is 57.38% with a standard deviation of 28.32%. As can be seen in table 1 the distribution of this total ownership is random and fluctuates between 0% and 100%.

B. Herfindahl Index

An interesting computation to make is the calculation of the Herfindahl index. Normally the Herfindahl index is used to compute the dominance of players in a single market. For instance when in the telecom market T-Mobile has a market share of 23%, KPN 12%, Vodafone 25% and the rest is held by smaller providers, the Herfindahl index is used to compute the intensity of the competition in the market. The Herfindahl index for this situation would then be:

23%2+ 12%2+ 25%2 = 12,98%

An example of where the Herfindahl index is used is by authorities to see if mergers or

acquisitions in a certain industry create unwanted environments for competition. This recently happened when Ryan Air wanted to take over Aer Lingus and this acquisition was prohibited by the authorities. As a rule of thumb, industries of which the Herfindahl indexes range from 15% to 25% are considered to be moderately concentrated industries whereas Herfindahl indexes above 25% are considered to be highly concentrated industries.

Using the Herfindahl index for the sample used in this thesis is supported by economic theory. This is illustrated using the following example. Imagine a company with ten

shareholders. When all ten are equally powerful and own 10% of the company six of them need to collaborate in a normal vote to create a majority. The Herfindahl index of this company is 10%, which is considered to be unconcentrated. Now imagine a company with again ten shareholders. One of them owns 63% while the other nine own 3%. Now the large shareholder is practically in control of the whole company because he can overrule all the other shareholders in a normal vote. The Herfindahl index of this company is 40.5% which is highly concentrated. Both companies are owned for the total 100% by shareholders owning a 3%+ share of the company, but the shareholder power in the latter company is higher, for the 63% shareholder that is. Although these examples are exaggerated, they appear to be present

(14)

14 in a less extreme form in the AFM dataset. Therefore the regressions are also run using the Herfindahl index as the dependent variable instead of the normal ownership concentration.

C. Search Volume Index

Because it is hard to measure attention in a quantitative manner, proxies are needed to approximate attention. These proxies are, as derived from the literature (section 2),

advertising expenses, volume of trade and the SVI. The latter is an interestingly constructed variable and special attention will be given to this proxy for attention.

The SVI can be found at www.google.com/trends. The output this engine gives when a search term is entered is as simple as it is elegant. It tells the reader how much there is sought for a certain search term throughout the years; in numbers and in the form of a graph. Two search terms can also be entered. For illustrative purposes I have entered the Dutch words for “swimming trunks” (“zwembroek” in Dutch) and “glove” (“handschoen” in Dutch) into the engine. The output is shown in figure 1 in appendix II.

One thing to note is that the peak searches for swimming trunks are always in the summer months June, July and August. The peak searches for glove on the other hand are always as expected during the winter, in November and December. Needless to say when it is summer because that is when they need new ones, because their old ones are worn out or not fashionable anymore. The same explanation goes for gloves: when it starts getting cold people start searching for gloves. This simple example shows how the SVI works.

Where other researchers such as Ding and Hou (2015) and Da, Engelberg and Gao (2011) have used a company’s stock ticker symbol as the search term to measure attention, I find that in this thesis this search term is not feasible. In contrast with the use of ticker

symbols of companies on the S&P500 or the Russel 3000, entering ticker symbols of NV’s in the Google search engine does not result in finding the company. The reason for this is that the ticker symbols of NV’s are in most cases also abbreviations that are used elsewhere and the NV’s from the dataset provided by the AFM are on average not large enough to dominate the searches for these stocks. This means that another means has to be used when using the SVI.

After several considerations the most obvious choice as a means to use for the SVI is the term ‘koers -company- ’. Ding and Hou (2015) explain that for using the SVI a search

(15)

15 term is needed that isolates searches of investors. When just the company name is used as search term the SVI of this search term also includes searches of e.g. people who search for the product of the company, a job at the company or media news of the company. Those are people who don’t necessarily want information about investing in the company. Therefore it is unfit to use the company name as the search term for the SVI.

To answer the question of which search term is fit to use to find the SVI of each company, the search term that is needed is the one which investors would use when they search for a particular company. Although it is not convenient that the search term consists of two words, I believe that the best search term that investors would use when searching for a company’s stock is ‘koers -company-’. ‘Koers’ is the Dutch word for direction but it is used in the investor world to tell what the price of a stock, i.e. shareprice, is and also to describe the path the share price of the stock has taken. An example of two search terms used in this thesis when entered into Google Trends is found in figure 2 in appendix II. Because the term ‘koers –company-’ is specific and commonly used in the investor world I think this is the best option to use as search term for the SVI.

The downside of using this search term however is that in Google Trends there have not been a lot of searches for this search term. On the one hand it is because the search term is complicated because it contains two words, on the other hand it is because firms in the AFM database tend to be smaller than firms in for instance the S&P500 used by Ding and Hou (2015) and therefore they are not well known by investors. Because in this sample there are a lot less results for the SVI the 0 values are not excluded from the database. Google trends returns a 0 value when a search term is rarely searched for, therefore it only seems fair not to exclude those terms. This is in contrast with the research by Ding and Hou (2015) and Da, Engelberg and Gao (2011).

As base search term for the SVI the term ‘koers ING’ is taken, because ‘koers ING’ is the search term that gives the highest SVI. All the other SVIs are scaled by the ‘koers ING’ SVI. From Google trends either monthly or weekly data is downloaded, depending on the availability. The SVIs are downloaded for the years 2006-2015. This data is averaged for every year to get the values that are needed. Only one SVI appears to be biased: the one of D.E. Master Blenders. When attempting to get the SVI for this company and therefore entering ‘koers DE’ into the Google search engine, Google also returns the stock price and path of the US company ‘Deere and Company’. All the other SVIs appear to be unbiased and can be used in this research. In total the dataset has 112 positive SVIs, 1046 SVIs with a zero

(16)

16 value and one biased SVI. This low amount of positive SVIs could cause biased results in the regression, an issue that will be discussed later.

D. Other variables and control variables

The rest of the data used in the regression is acquired from Datastream. When the yearly values are not available the data is averaged to get yearly values. This situation only occurred with the trading volume proxy. For all the other variables annual data is available. As the data of the shareholder structure is yearly, it is not possible to make use of the abnormal trading volume as used by e.g. Barber and Odean (2008) to measure shareholder attention, because daily shareholder structure data is needed then. A possible solution for this problem could be to generate standard deviations of the yearly trading volume as a proxy for abnormal trading volume standard.

The advertising costs are generated using an approximation by subtracting research and development and operating lease costs from the selling, general and administrative expenses made by a firm. The choice for using this approximation is supported by the definitions in Datastream. The reason for choosing this approximation is because a firm’s advertising costs as a raw number is rarely available in Datastream.

Furthermore, there is accounted for a for the Dutch market very specific phenomenon. In the Netherlands it is allowed to buy shares of a company and redistribute them again as certificates. These certificates do not have voting rights so they do not represent shareholder power. Therefore, the companies buying shares and redistributing them again are not

excluded from the dataset because they have, in most cases, a lot of shareholder power. These companies can be identified very easily from the AFM database, because they are always foundations (in Dutch “Stichtingen”) called “Administratiekantoor”, which means office of administration. Because these foundations do however have a significant impact on the shareholder structure, a dummy is created as a control variable. Moreover, Demsetz and Kehn (1985) argue that firm size influences ownership concentration. This is however not

necessarily an attention grabbing factor. Therefore data on this control variable is also gathered from Datastream and averaged because it was not available on a yearly basis, but only on a daily basis.

(17)

17 It is still ambiguous whether the dividend payout ratio is a measurement of attention or a control variable. On the one hand investors want tangible returns and dividends is something they would definitely look at. On the other hand, as Short, Zhang and Keasey (2002) already argued, this could also be something an institutional investors would demand for. When this is the case there is need for control and the variable also needs to be included. Furthermore, as standard control variables Tobin’s Q, profitability and property, plant and equipment (PP&E) are used. All control variables are scaled by assets.

E. The Model

In the regression ownership structure and the Herfindahl index are explained using proxies for attention. Because there is data for several companies over several years a perfect opportunity arises for the usage of panel data. Using panel data there will dealt with several cases of endogeneity constructing time fixed (λ_𝑡) and firm fixed (𝛼_𝑖) effects. An advantage of using panel data that the dataset is larger because data of several years is used, which makes the chance of collinearity smaller. Furthermore, using the time fixed effects variable the

regression will deal with omitted variables that differ across companies but are constant over time. The firm fixed effects variable does the same for variables that differ across time but are constant for the companies. This can be a very helpful instrument because in the period of the dataset a crisis took place. Using the time fixed effects makes sure that there we deal with omitted variables that were not observed but because of the crisis had an influence on ownership structure (Stock and Watson, 2003).

Because for not every observation all the data is available there is dealt with an

unbalanced panel. The standard errors in the regression are clustered. This means that they are robust both to heteroscedasticity and to correlation over time within an entity.

The first model will be as follows: 𝑂𝑤𝑛𝑒𝑟𝑠ℎ𝑖𝑝 𝑠𝑡𝑟𝑢𝑐𝑡𝑢𝑟𝑒_𝑖,𝑡

= 𝛽₁∗ 𝐴𝑑𝑣𝑒𝑟𝑡𝑠𝑖𝑛𝑔 𝐸𝑥𝑝𝑒𝑛𝑠𝑒𝑠_𝑖,𝑡 + 𝛽₂∗ 𝑇𝑟𝑎𝑑𝑖𝑛𝑔 𝑉𝑜𝑙𝑢𝑚𝑒_𝑖,𝑡+ 𝛽₃∗ 𝑆𝑉𝐼_𝑖,𝑡 + 𝛾₁∗ 𝑀𝑎𝑟𝑘𝑒𝑡 𝐶𝑎𝑝𝑖𝑡𝑖𝑙𝑖𝑧𝑎𝑡𝑖𝑜𝑛_𝑖,𝑡+ 𝛾₂∗ 𝐷𝑢𝑚𝑚𝑦 𝐹𝑜𝑢𝑛𝑑𝑎𝑡𝑖𝑜𝑛_𝑖,𝑡+ 𝛾₃ ∗ 𝑂𝑤𝑛𝑒𝑟𝑠ℎ𝑖𝑝 𝑆𝑡𝑟𝑢𝑡𝑢𝑟𝑒_{𝑖,𝑡−1}+ 𝛽₄/𝛾₄∗ 𝐷𝑖𝑣𝑖𝑑𝑒𝑛𝑑𝑠 + 𝛾₅∗ 𝑇𝑜𝑏𝑖𝑛′𝑠 𝑄 + 𝛾₆ ∗ 𝑃𝑟𝑜𝑓𝑖𝑡 + 𝛾₇∗ 𝑃𝑃&𝐸 + 𝛼_𝑖 + 𝜆_𝑡+ 𝜇_𝑖,𝑡

(18)

18 The expectations for this regression are that advertising expenses have a negative impact on ownership structure, because the more a company invests in advertising, the more it will be a company the public knows and small investors will be tempted to buy the company’s stock. The coefficient for trading volume is also expected to be negative, because when a stock grabs an investor’s attention the investor is more likely to trade in this stock. Because individual investors react strongly to attention grabbing factors, they are more likely to be net buyers and thus owners of stocks for which the trading volume is higher. For the SVI the expected coefficient is also negative. The SVI is as opposed to the other proxies for attention a direct measure of attention. The more investors search for stocks on Google the higher the SVI is. As attention is believed to have a positive effect on the scattered ownership of the shares a high SVI is expected to correlate with lower ownership structure or Herfindahl indexes.

As for the control variables, although a causal relationship is not necessarily needed, expectations are that market capitalization will have a negative coefficient as the larger a company is the smaller the ownership structure and Herfindahl index are thought to be. When a company is worth more on the market it is simply harder to own a larger share of this company. The dummy for when a foundation is a shareholder of the company is expected to have a positive coefficient as the foundations are thought to correlate with more institutional investors owning large shares of a company. Special attention goes out to the coefficient of the dividend yield, as economic theory supports both positive and negative coefficients.

4. Results

A. Simple regressions

In the first, simple regressions either the percentage of blockholder ownership or the Herfindahl index of blockholder ownership are regressed on either trading volume,

advertising expenses or the SVI. The expectations are as follows: the coefficient of trading volume, advertising expenses and the SVI are all expected to be negative. From the literature the suggestion is derived that the more there is dealt in a certain stock, the more dispersed the ownership of that stock is. The coefficient of advertising expenses is also expected to be negative, as more advertising is expected to create more awareness at an individual level, so

(19)

19 that ownership will become more dispersed. Lastly, the coefficient of the SVI is expected to be negative, because the more there is searched for a particular stock, the expectation is that more individuals will buy this stock.

As tables 2 to 7 show most of the results do not agree with the hypotheses. They can be summarized as follows. First of all, most of the coefficients are not even significant at a 10% level. Secondly in a lot of the regressions the signs of the estimations do not agree with the expectations. For example, in table 2 regression II the estimated coefficient is positive, while a negative effect of trading volume on blockholder ownership is expected. The results so far thus suggest a non-relation between blockholder ownership and the proxies, on an individual level that is.

For a more broad review of the results of the simple regressions I refer to the tables themselves. I want to add however that these results do not necessarily imply a non-relation between the proxies and blockholder ownership or the Herfindahl index of blockholder ownership, because these regressions are as the title of this paragraph suggests, ‘simple’. There are some other factors that need to be controlled for and these proxies can of course be combined in the regression. This will happen in the next paragraph.

B. Combined regressions

Tables 8 to 15 show the results for the combined regressions in which every combination of trading volume, advertising expenses and SVI is regressed on either the blockholder

ownership or the Herfindahl index of blockholder ownership. Also in tables 14 and 15 the dividend yield is added into the fifth regression, because according to the literature it might be a proxy for explaining blockholder ownership. It is interesting to see that the dividend yield in both tables appears to be both positive and significant, supporting the theory stating that dividend yield is mainly something institutional investors find important. This means that the dividend yield is a control variable.

Again most of the estimations are not significant and the signs of the coefficients are in roughly more than half regressions not as expected. The regressions in these tables are thus not economically relevant. However, because in these regressions there is no control for some basic factors, these outcomes are not unexpected. Therefore in the next section the results of the regressions with controls are discussed.

(20)

20

C. Controlled regressions

In tables 16 and 17 the control variables are added. Again the coefficients of the explanatory variables do not seem to be significant nor have the correct sign on a structural basis.

Therefore, the results from these complex regressions are not economically relevant.

Regressions with just one of the proxies for attention as they are seen in tables 1 to 7 but with the control variables were also generated. Because they did not yield any economically relevant results these tables are excluded from the thesis.

One thing that grabs the attention is the dividend yield. Again it seems to have a positive and significant relation with dependent variables. This suggests that institutional investors allocate more value to the dividend yield than individual investors. The ‘foundation dummy’ typical for the Dutch legal system is expected to have a significantly positive effect on the blockholder ownership, because when such a foundation is shareholder it typically owns a large part of the shares. In six of eight regressions this seems to be the case, although in only two of these six regressions the effect appears to be significant. In the other six regressions the estimations are not significant.

The lagged dependent variable appears to have a positive relation with the dependent variable, although this relation is only significant in the regressions with the blockholder ownership as the dependent variable. The expectation for market capitalization is that it has a negative relationship with the dependent variables. Although this only appears to be true in table 17, in both tables the estimations of this coefficient are not significant. This means that it is not economically relevant. The conclusion for these two tables is that adding the control variables does not make the proxies have a better economic relevance.

The only possible conclusion from all these regressions is that a company’s ownership structure cannot be explained with the used proxies: trading volume, advertising expenses and the SVI. This means that there is not economically significant relation between attention and blockholder ownership, while the expectations were that there was a relation. In the

(21)

21

5. Model tests

In this section the tests that are run to see if the regressions are correctly executed are discussed. First of all a Hausman test is run to see whether there is need for fixed or random individual effects in the panel data regression. In the Hausman test a consistent estimator is compared with an estimator that is efficient under the assumptions being tested (Stata, 2015). When the null hypothesis is rejected there is a difference between the two estimators and there is reason to believe that the assumptions on which the efficient estimator is based are biased. If this is the case, one should choose a fixed effects model. When the null hypothesis is not rejected the individual effects are random and follow and normal distribution. Because in the model used in this thesis the standard errors are clustered, a standard Hausman test does not give any answers in Stata. However when using a modified Hausman test the outcome is that the null hypothesis should be rejected. Therefore the choice to use fixed individual effects is made.

To see if time fixed effects are needed using a fixed effects model the “Test for linear hypotheses after estimation” is used. In this test the hypothesis tested stating that the time fixed effects are significantly different from zero. When the null hypothesis in this test is rejected the time fixed effects are significantly different from zero. The outcomes from this test suggest that the time fixed effects are significantly different from zero in every year in the model in this thesis.

Constructing a modified Wald statistic helps testing for grouped heteroscedasticity in a fixed effects model (Webel, 2011). Because the outcome from this test is that the standard errors are heteroscedastic, a solution for this problem needs to be found. Using robust standard errors is a possible solution however and therefore it is not a problem that the null hypothesis stating that the errors are homoscedastic is rejected.

6. Conclusion

In this thesis the relationship between attention and ownership structure is tested. The hypothesis is that when a company attracts more attention it also attracts more individual investors relative to institutional investors. This is tested using a dataset containing panel data. The regressions used therefore contain entity and/or time fixed effects. As proxies for

(22)

22 SVI. The dependent variable, the percentage of a company’s blockholder ownership, is

obtained through a dataset that can be found at the AFM. This dataset is modified to construct the blockholder ownership of each company for each year. As an alternative for the

blockholder ownership the Herfindahl index of this figure is also used. In the model there is also controlled for several factors such as market capitalization and a company’s dividend policy, because the existing literature suggests that this is needed.

The results from the regressions show however that there is no significant relation between the proxies for attention and a company’s ownership structure. The outcomes show that the relations between the proxies for attention and ownership structure are not always as expected and inconsistent. The independent variables also show large standard deviations meaning that their values are not very precisely estimated.

Interpreting these results tells us that in the Netherlands, no correlation is found between attention and ownership structure. While the expectations were that when a company attracts attention this leads to individual investors buying the shares of the company which means a more dispersed ownership, the results of this study suggest that this is not the case, and that in fact there is not relationship at all between attention and ownership structure. This also contradicts the existing literature. Both Barber and Odean’s (2008) suggestions about trading volume as well as Lou’s (2014) and Grullon, Katanas and Weston’s (2004)

suggestions about advertising results yield no correlation as proxies for attention with

ownership structure. Also the relatively unknown direct measure for attention, the SVI, from Da, Engelberg and Gao (2011) and Ding and Hou (2015) research shows no correlation with the proportion of individual investors in a company.

There are several reasons for these findings. One important difference with the results from the existing literature is that the results in this thesis are based on Dutch data, whereas in the existing literature mainly data from the US is used. The most important difference

between the US data and the Dutch data lays in the legal system. In the Netherlands companies are allowed sell their shares to an administrative office, which then certifies the shares, and sells these certificates to the public. These certificates carry no voting rights and are therefore powerless in a way. This system of keeping the voting rights inside the company (because mostly the foundation running the administrative office is controlled by the

company) by selling certificates is not used in the US. In the Netherlands certifying of shares is a common practice. In 129 of the 1159 observations a company has certified (a part of) his shares. The consequence of the presence of the certification of shares in the Netherlands is

(23)

23 that the ownership structure is biased towards a structure with more institutional holdings relative to US company structure.

An attempt to control for the presence of certification is made in the model by adding a dummy that is equal to 1 when an administrative office foundation is holding shares of a company and zero when it’s not. The expectation of adding this control variable to the regression is that when an administrative office foundation is present and the value of the dummy is equal to one it has a (large) positive effect on the proportion of institutional

holdings of a company. The reason for why this way of controlling is not yielding the desired results is that holdings of these administrative offices is also not constant. For example, in 2015 the lowest percentage held in a company by an administration office foundation is 4.14% whereas the highest percentage is 99.84%. The conclusion therefore is that the

presence of administrative office foundations is biasing the results, but difficult to control for. Another difference between the US dataset and the Dutch dataset is the size and especially the size of the companies it includes. Whereas the dataset in this thesis includes all listed companies incorporated after Dutch law, Ding and Hou (2015) use the S&P 500 index and Da, Engelberg and Gao (2011) even use the Russell 3000 index. Not only is their dataset larger, for the dataset used in this thesis only contains 143 companies maximum in 2015, the companies in the dataset used in this thesis are also a lot smaller in terms of market

capitalization as the S&P 500 and Russell 3000 index contain the largest 500 or 3000

companies of the US. This has implications for the regressions. For instance using the SVI in this thesis yields almost no results, mainly because there are only 112 out of 1159 positive results when retrieving the SVI. This is partly because overall less people are interested in dealing in Dutch companies than they are in US companies. But what has even larger consequences is the fact that the search term used in the other papers, i.e. a stock’s ticker symbol, is biased when applied in this thesis. When stock ticker symbols of Dutch companies are searched for in Google, in most cases the results do not show information about the stock searched for. Instead, Google returns other information because Dutch companies’ ticker symbols are in most cases abbreviations of words sought for more often than the ticker symbol. Therefore another search term to compute the SVI is used in this thesis, but because this search term is longer and less obvious it yields almost no results.

A reason more fundamental is that there is a difference in the way ownership structure is measured in this thesis as compared to other studies. In Ding and Hou’s (2015) paper ownership structure is measured using merely the number of shareholders. As was stated

(24)

24 earlier in this thesis this does not seem to be an appropriate measure to use in this thesis as in most cases this does not necessarily say something about shareholder power. In Barber and Odean’s (2008) paper they choose to look at the amount of stock that is bought and sold at two types of brokerages: brokerages where mostly institutional investors trade and brokerages where mostly individual investors trade. When a certain type of stock is bought more at a brokerage for individuals the assumption is made that the ownership of this stock becomes more dispersed. This is a very inventive way to measure ownership structure, however it cannot be applied in this thesis.

The method that corresponds the most with the method to measure shareholder ownership in this thesis is used in the papers of Lou (2014) and Drake, Roulstone and Thornock (2012). In their papers the ownership structure is calculated using the

CDA/Spectrum database in which institutions holding over $100 million make reports of holdings of more than 10,000 shares, or $200,000 of a certain common stock. The difference between their database and the AFM database is that in the AFM database large shareholdings of founding shareholders or shareholdings of relatives are also included. Although these are not necessarily institutional shareholders, these shareholders do own a large part of the company and can exercise their shareholder power. In my opinion the AFM database is the better basis for calculating the ownership structure for research where shareholder power is examined, because it includes more owners who can exercise their power. Therefore it is more inclusive.

Lastly, advertising expenses were not available in Datastream. Therefore an approximation for this proxy for attention had to be generated. While this approximation seems to make sense based on the information Datastream supplies it might not be as accurate as hoped for. Also because advertising expenses had to be generated a large part of the sample got lost, explaining why the sample of advertising expenses is a lot smaller than the other samples (see table 1). This limits the regressions in this thesis.

The results of this thesis have several implications. If they are to be true, it means that

attention has no effect on ownership structure. This means that individual shareholders do not discriminate between large attention grabbing companies and smaller not so well known companies. The fear when writing this thesis was that individual investors do discriminate between these types of companies. If this was the case, large attention grabbing companies

(25)

25 could abuse their power by exploiting individual shareholders and conducting bad corporate governance. This fear however seems to be unfounded. Moreover the results show that individual investors are not as unprofessional as thought. Apparently individuals also do their homerwork and do not only invest in the larger companies.

Lastly, this thesis seems to contradict other studies. Although using merely Dutch data brings along a lot of limitations which seem to bias the regression, I think a big and important difference between this thesis and the existing literature is the dataset used. Since a different dataset a different ownership structure is calculated by using a different dataset, the results are different from the results in existing literature. Because the ownership structure used in this thesis includes more owners with shareholder power it is better to use when doing follow up research about corporate governance. But, as was mentioned earlier, this thesis seems to imply that there is no reason to think that large companies have worse corporate governance because of the attention they attract. Therefore my suggestion for further studies is to examine the same premise in other countries.

(26)

26

References

AFM. (2015) Register substantiële deelnemingen en bruto shortposities. [Online] Available from:

http://www.afm.nl/nl/professionals/registers/alle-huidige-registers.aspx?type=%7B1331D46F-3FB6-4A36-B903-9584972675AF%7D. [Accessed: 6th April 2015].

Aguilera, R. V., Rupp, D. E., Williams, C. A., & Ganapathi, J. (2007). Putting the S back in corporate social responsibility: A multilevel theory of social change in

organizations. Academy of management review, 32(3), 836-863.

Barber, B. M., & Odean, T. (2008). All that glitters: The effect of attention and news on the buying behavior of individual and institutional investors. Review of Financial Studies, 21(2), 785-818.

Bauer, R., Braun, R., & Clark, G. L. (2008). The emerging market for European corporate governance: the relationship between governance and capital expenditures, 1997 2005. Journal of Economic Geography, 8(4), 441-469.

Chordia, T., & Swaminathan, B. (2000). Trading volume and cross-autocorrelations in stock returns. Journal of Finance, 913-935.

Clark, N. K., & Rutter, D. R. (1985). Social categorization, visual cues, and social judgements. European Journal of Social Psychology, 15(1), 105-119.

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

Demsetz, H., & Lehn, K. (1985). The structure of corporate ownership: Causes and consequences. The Journal of Political Economy, 1155-1177.

Ding, R., & Hou, W. (2015). Retail investor attention and stock liquidity. Journal of International Financial Markets, Institutions and Money, 37, 12-26.

Drake, M. S., Roulstone, D. T., & Thornock, J. R. (2012). Investor information demand: Evidence from Google searches around earnings announcements. Journal of Accounting Research, 50(4), 1001-1040.

Lou, D. (2014). Attracting investor attention through advertising. Review of Financial Studies, hhu019.

(27)

27 Gervais, S., Kaniel, R., & Mingelgrin, D. H. (2001). The high-volume return

premium. Journal of Finance, 877-919.

Gompers, P. A., Ishii, J. L., & Metrick, A. (2001). Corporate governance and equity prices (No. w8449). National bureau of economic research.

Grether, D. M. (1980). Bayes rule as a descriptive model: The representativeness heuristic. The Quarterly Journal of Economics, 537-557.

Webel, K. (2011). Greene, WH, Econometric analysis. Statistical Papers,52(4), 983-984 Grullon, G., Kanatas, G., & Weston, J. P. (2004). Advertising, breadth of ownership, and

liquidity. Review of Financial Studies, 17(2), 439-461.

Harada, K., & Nguyen, P. (2011). Ownership concentration and dividend policy in Japan. Managerial Finance, 37(4), 362-379.

Hou, K., Xiong, W., & Peng, L. (2009). A tale of two anomalies: The implications of investor attention for price and earnings momentum. Working Paper.

Jensen, M.C., 1986. Agency costs of free cash flow, corporate finance and takeovers. AEA Papers and Proceedings 76 (2), 323 – 329.

Kahneman, D. (1973). Attention and effort (p. 246). Englewood Cliffs, NJ: Prentice-Hall. Lo, A. W., & Wang, J. (2000). Trading volume: definitions, data analysis, and implications of

portfolio theory. Review of Financial Studies, 13(2), 257-300.

Merton, R. C. (1987). A simple model of capital market equilibrium with incomplete information. The Journal of Finance, 42(3), 483-510.

Odean, T. (1998). Volume, volatility, price, and profit when all traders are above average. The Journal of Finance, 53(6), 1887-1934.

Scheinkman, J. A., & Xiong, W. (2003). Overconfidence and speculative bubbles. Journal of political Economy, 111(6), 1183-1220.

Short, H., Zhang, H., & Keasey, K. (2002). The link between dividend policy and institutional ownership. Journal of Corporate Finance, 8(2), 105-122.

Stata. (2015) Hausman specification test. [Online] Available form:

http://www.stata.com/manuals13/rhausman.pdf. [Accessed: 4th January 2016].

Stock, J. H., & Watson, M. W. (2003). Introduction to econometrics (Vol. 104). Boston: Addison Wesley.

Tversky, A., & Kahneman, D. (1973). Availability: A heuristic for judging frequency and probability. Cognitive psychology, 5(2), 207-232.

(28)

28

Appendix I – Summary Statistics

Table 1 Summary statistics

In this table the data of the parameters used in the regressions is summarized. Column (I) gives the sample size; column (II) gives the sample mean; column (III) gives the sample standard deviation and in column (IV) and (V) the sample minimum and maximum values are given. Advertising expenses is in millions of euros.

(I) (II) (III) (IV) (V)

N mean sd min max

% Owned by Blockholders 1,159 0.57 0.28 0 1.0

Herfindahl Index of Blockholders 1,159 0.23 0.29 0 1.0

Trading Volume 1,126 952.59 3,159.42 0 32,619.0 Advertising Expenses 725 0.84 2.31 0 17.7 SVI 1,159 28.03 13.60 1 111.0 Dividend Yield 1,123 3.95 23.05 0 521.7 Market Capitalization 1,044 2,984.42 7,047.34 0 73,169.0 Dummy Foundation 1,159 0.13 0.33 0 1.0

One Year Lagged % Owned by Blockholders 1,158 0.57 0.28 0 1.0 One Year Lagged Herfindahl Index of Blockholders 1,158 0.23 0.29 0 1.1 Interaction SVI - dummy Foundation 1,159 3.71 11.39 0 110.2

Tobin's Q 919 1.84 35.24 -761 467.4

Profit 912 0.01 0.33 -7 0.5

PP&E 895 0.23 0.25 0 1.0

(29)

29

Appendix II – Figures

Figure 1 – Search Volume Index example

Blue line search term: zwembroek Red line search term: handschoen

(30)

30 Figure 2 – Search Volume Index as it is used in this thesis

In this graph the Search Volume Index (SVI) for the company called “Aegon” scaled to the SVI of the company called “ING” is illustrated. The graph shows the reader that for both search terms (‘koers Aegon’ and ‘koers ING’ respectively) a peak of investment related searches occurred around the year 2009. After this year the popularity of both the search terms decreased somewhat.

Blue line search term: Koers ING Red line search term: Koers Aegon

(31)

31

Appendix III – Results

A. Simple regressions

Table 2 Trading Volume on % Blockholder Ownership

This table shows the output of the regression that has trading volume as an independent variable and the percentage of blockholder ownership as the dependent variable. Trading volume is one of the proxies that is used for attention. Trading volume is in millions of stock traded. Robust t-statistics in parentheses: *** p<0.01, ** p<0.05, * p<0.1.

Dependent variable: % Blockholder Ownership

(I) (II) (III) (IV) OLS Entity FE Year FE Entity+Year FE Regressor Trading Volume -600.82 367.30 -586.11 765.24 (-0.57) (0.64) (-0.55) (1.23) Constant 58.12*** 57.13*** 58.11*** 48.23*** (24.18) (96.71) (24.03) (34.12) Additional Information

Panel data No Yes Yes Yes

Entity FE No Yes No Yes

Year FE No No Yes Yes

Model statistics

Observations 1,041 1,041 1,041 1,041 R-squared 0.01 0.80 0.05 0.85 Number of entities 137 137 137 137

(32)

32

This table shows the output of the regression that has trading volume as an independent variable and the Herfindahl Index of blockholder ownership as the dependent variable. Trading volume is one of the proxies that is used for attention. Trading volume is in millions of stock traded. Robust t-statistics in parentheses: *** p<0.01, ** p<0.05, * p<0.1.

Dependent variable: Herfindahl Index Blockholder Ownership (I) (II) (III) (IV) OLS Entity FE Year FE Entity+Year FE Regressor Trading Volume 517.47 615.51 523.20 714.35 (0.42) (1.09) (0.42) (1.18) Constant 22.22*** 22.12*** 22.22*** 19.52*** (8.28) (38.03) (8.24) (15.82) Additional information

(33)

33

Table 4 Advertising expenses on % Blockholder ownership

This table shows the output of the regression that has advertising expenses as an independent variable and the percentage of blockholder ownership as the dependent variable. Advertising expenses is one of the proxies that is used for attention. Advertising expenses is in millions of euros. Robust t-statistics in parentheses: *** p<0.01, ** p<0.05, * p<0.1.

(I) (II) (III) (IV) OLS Entity FE Year FE Entity+Year FE Regressor Advertising Expenses 0.22 2.86*** 0.19 1.06 (0.16) (3.27) (0.14) (1.06) Constant 55.35*** 53.13*** 55.37*** 48.30*** (19.79) (72.48) (19.71) (43.88) Additional information

Observations 717 717 717 717

R-squared 0.00 0.88 0.03 0.90

(34)

34

Table 5 Advertising Expenses on Herfindahl Index

This table shows the output of the regression that has advertising expenses as an independent variable and the percentage of Herfindahl Index of blockholder ownership as the dependent variable. Advertising expenses is one of the proxies that is used for attention. Advertising expenses is in millions of euros. Robust t-statistics in parentheses: *** p<0.01, ** p<0.05, *p<0.1

Dependent variable: Herfindahl Index Blockholder Ownership (I) (II) (III) (IV) OLS Entity FE Year FE Entity+Year FE Regressor Advertising Expenses 1.78 0.67 1.78 0.61 (1.17) (0.70) (1.16) (0.48) Constant 21.00*** 21.93*** 21.00*** 20.33*** (7.08) (27.38) (7.05) (19.67) Additional information

R-squared 0.021 0.922 0.023 0.923 Number of entities 114 114 114 114 Robust t-statistics in parentheses

(35)

35

Table 6 SVI on % Blockholder Ownership

This table shows the output of the regression that has the SVI as an independent variable and the percentage of blockholder ownership as the dependent variable. The SVI is one of the proxies that is used for attention. More information about the SVI can be found in paragraph 3C and figures 1 and 2. Robust t-statistics in parentheses: *** p<0.01, ** p<0.05, * p<0.1.

(I) (II) (III) (IV) OLS Entity FE Year FE Entity+Year FE Regressor SVI -0.03 0.16** -0.08 0.01 (-0.23) (2.49) (-0.63) (0.26) Constant 58.20*** 52.89*** 59.74*** 48.41*** (14.20) (29.44) (14.24) (27.40) Additional information

(36)

36

Table 7 SVI on Herfindahl Index

This table shows the output of the regression that has the SVI as an independent variable and the Herfindahl index of blockholder ownership as the dependent variable. The SVI is one of the proxies that is used for attention. More information about the SVI can be found in paragraph 3C and figures 1 and 2. Robust t-statistics in parentheses: *** p<0.01, ** p<0.05, * p<0.1.

Dependent variable: Herfindahl Index Blockholder Ownership

(I) (II) (III) (IV)

OLS Entity FE Year FE Entity+Year FE

Regressor SVI -0.09 0.00 -0.10 -0.027 (-0.78) (-0.00) (-0.90) (-1.00) Constant 25.72*** 23.30*** 26.17*** 21.55*** (6.59) (41.60) (6.61) (25.32) Additional information

Observations 1,159 1,159 1,159 1,159

R-squared 0.00 0.88 0.01 0.88

(37)

37

B. Combined regressions

Table 8 Trading Volume & Advertising Expenses on Blockholder Ownership

This table shows the output of the regression that has trading volume and advertising expenses as the independent variables. The percentage of blockholder ownership is the dependent variable. Both trading volume and advertising expenses are in millions. Robust t-statistics in parentheses: *** p<0.01, ** p<0.05, * p<0.1.

(I) (II) (III) (IV) OLS Entity FE Year FE Entity+Year FE Regressor Trading Volume -810.18 -123.20 -708.79 61.34 (-0.81) (-0.38) (-0.71) (0.25) Advertising Expenses 1.18 2.83*** 1.04 1.07 (1.17) (3.30) (0.98) (1.05) Constant 54.68*** 52.48*** 54.69*** 47.55*** (18.87) (65.80) (18.75) (40.62) Additional Information

R-squared 0.01 0.87 0.03 0.89

(38)

38

Table 9 Trading Volume & Advertising Expenses on Herfindahl Index

This table shows the output of the regression that has trading volume and advertising expenses as the independent variables. The Herfindahl index of blockholder ownership is the dependent variable. Both trading volume and advertising expenses are in millions. Robust t-statistics in parentheses: *** p<0.01, ** p<0.05, * p<0.1.

Dependent variable: Herfindahl Index Blockholder Ownership (I) (II) (III) (IV) OLS Entity FE Year FE Entity+Year FE Regressor Trading Volume 199.11 74.73 207.25 98.09 (0.20) (0.83) (0.20) (0.92) Advertising Expenses 1.73* 0.69 1.72* 0.61 (1.76) (0.72) (1.72) (0.47) Constant 19.83*** 20.88*** 19.83*** 19.15*** (6.33) (23.81) (6.31) (17.35) Additional Information

R-squared 0.03 0.92 0.03 0.92

Attention and ownership structure : the effect of a share’s popularity on shareholder structure