Ownership structure and stock return volatility

Transcription

1 Ownership structure and stock return volatility Master s Thesis Financial Economics Bas Janssen (s ) Abstract This paper reviews whether ownership structures affect stock return volatility by trading volumes or by a more information based approach. A larger investor base potentially leads to a more accurate price signal, hereby affecting stock return volatility. Our examination distinguishes by incorporating a large variety of ownership structure measurements, including measurements related to ownership concentration as well as measurements related to the size of the investor base. Results are in line with recent literature regarding this topic and confirm the belief of a channel related to trading volumes. Nijmegen School of Management Supervisor: dr. J. Qiu

2 Table of Contents Introduction... 3 Literature overview... 5 Data and methodology Empirical analysis Conclusion Appendix Bibliography

3 Introduction At portfolio level, volatility is the simplest measure of risk. Therefore, stock return volatility is at the centre of asset pricing research and literature about stock return volatility is voluminous (Zhang, 2010). Important findings relate to certain trends in individual stock volatility over time (Schwert, 1989). Campbell et al. (2002) find strong evidence for a positive trend in idiosyncratic firm-level volatility in late 20 th century. While Zhang (2010) finds a declining trend in average stock return volatility in , in the second half of 2007 stock return volatilities started to rise again. Fluctuations in volatility will be there in the future and linked to these developments will be the question what causes volatility. The relationship between ownership structures and volatility is a relatively new paradigm advocated through the increased share of institutional ownership in the stock market. The size of the investor base is part of this paradigm and is linked to volatility over the years (Wang, 2007; Zhang, 2010; Jankensgård & Vilhelmsson, 2016). This led to the reasoning that an increase in the size and diversity of the investor base will lead to lower volatility, more investors provide more information content which on its turn will lead to a more accurate stock price (Wang, 2007). However, on the other hand more investors can lead to a higher volatility through a trading channel (Zhang, 2010). This reasoning is natural to think about. Price changes occur due to trading volume, more trading volume would lead to higher volatility (Karpoff, 1987; Gallant, Rossi, & Tauchen, 1992; Zhang, 2010). These contradicting theories, on the one hand the trading-channel and on the other hand information provision, caused a recent conjecture in literature to emerge. This conjecture is tried to solve by Jankensgård & Vilhelmsson (2016), who mainly focused on the reasoning that an increase in the size and diversity of the investor base would lead to lower volatility. However, opposite results are found and additional analysis led them to believe that a trading channel is more important (Jankensgård & Vilhelmsson, 2016). By looking at the size and diversity of the investor base Jankensgård & Vilhelmsson (2016) were the first that empirically tested whether improved information content would lead to lower volatility. However, the size and diversity of the investor base are not the only dimensions of corporate ownership structures (Weimer & Pape, 1999). Ownership concentration measures are linked to volatility as well (Clark & Wójcik, 2004; Ezazi, Sadeghi, Alipour, & Amjadi, 2011). Based on the same reasoning: more concentrated corporate ownership will lead to higher stock price s volatility (Clark & Wójcik, 2004). Since, Jankensgård & Vilhelmsson (2016) do not use any of these ownership concentration measures their analysis is inconclusive. 3

4 Moreover, it is important to note that the investor base is dependent on country-specific determinants (La Porta, Lopez-de-Silanes, Shleifer, & Vishny, 2000). Since Jankensgård & Vilhelmsson (2016) only take Swedish listed firms into consideration, their findings won t be conclusive. This research will pay in particular attention on these shortcomings. The relationship between ownership structures and individual stock s volatility will be analysed by using the size and diversity of the investor base as well as ownership concentration measures. A profound empirical analysis is conducted by using several proxies for ownership concentration and the size of the investor base. Moreover, this paper analyses firms listed in The Netherlands and firms that are listed in the United States. Although some initial evidence is found that the size of the investor base contributes negatively to share price volatility, more profound analysis led to the belief of a trading channel interpretation. Through trading volume, the degree of ownership concentration will contribute negatively to stock price volatility. A larger investor base will lead to more trading and an accompanied higher volatility in the stock. The idea of a more accurate price signal due to the presence of a large investor base can be rejected based on findings in our dataset. The next chapter will deal with literature regarding these two seemingly contradicting theories. Moreover, this chapter will state different measures regarding ownership structure and their implications. Finally, the upcoming chapter will provide us with hypotheses that are constructed based on literature. Before illustrating the main empirical results, the chapter data and methodology further elaborates on the methodology and dataset used. Logically, the paper ends with a conclusion and some potential improvements and limitations. 4

5 Literature overview CORPORATE GOVERNANCE Analysing the relationship between ownership structure and price s volatility can be linked to studies in corporate governance. Ownership structure is a characteristic of corporate governance according to the taxonomy of Weimer & Pape (1999), ownership concentration is herein an important dimension (Weimer & Pape, 1999). The relationship between ownership structure and firm performance is in particular often analysed. Research based on this relationship dates back to Berle and Means (1932) who suggest an inverse relationship between diffused ownership and firm performance. This result is directly opposed by Demsetz and Lehn (1985). Their data lend no support; no significant relationship is found between accounting profit and ownership concentration (Demsetz & Lehn, 1985). Morck et al. (1988) re-examine the relationship and find no significant linear regression. However, significant nonlinear specifications are found, indicating an optimum between management ownership and firm performance (Morck, Shleifer, & Vishny, 1988). After Morck et al. (1988) several articles have tried to confirm a relationship between firm performance and ownership structure. However, viewed in totality no strong evidence was given in favour of such a relationship. Demsetz and Villalonga (2001) try to complete the analysis by taking into account the two dimensions of ownership structure given by Weimer & Pape (1999): ownership concentration and the identity of the shareholders. No evidence is found to support the relationship. Moreover, evidence from previous studies in favour of a relationship between ownership structure and firm performance is said to be biased, due to ignorance of the endogeneity of ownership structure (Demsetz & Villalonga, 2001). By for example management compensation in stock options, firm performance can influence ownership structure. This problem of endogeneity is originally shown in Demsetz and Lehn (1985). Demsetz and Lehn (1985) primary concern was not to test the relationship proposed by Berle and Means (1932), but to analyse the determinants of ownership concentration. Suggestion made is that the noisier (less predictable) a firm s environment is, the more difficult it is to monitor the managerial behaviour as a shareholder (Demsetz & Lehn, 1985). Therefore, concentrated ownership should be find in firm s that operate in an unpredictable environment. This noisy, unpredictable environment can be reflected in a higher volatility of stock market returns (Demsetz & Lehn, 1985). Thus, firms showing a high volatility in stock market returns should have a higher level of ownership concentration. Most studies analysing the ownership structure in relationship with stock price s volatility are influenced by the work of Demsetz and Lehn (1985). This 5

6 reasoning allows the problem of reverse causation to occur. When trying to link ownership concentration to stock price s volatility one should take the problem of endogeneity into account. This is emphasized by Demsetz & Villalonga (2001) arguing that reverse causation is possible and can lead to biased results in (among others) research by Morck et al. (1988). STOCK PRICE VOLATILITY AND OWNERSHIP STRUCTURE Investor-base broadening effect Clark and Wójcik (2004) find empirical evidence for a positive relationship between ownership concentration and stock price volatility in German listed firms. Moreover, they find that this ownership structure is region specific which leads to the implication that investment managers must take geography seriously (Clark & Wójcik, 2004). Different measures are used for ownership concentration as well as for volatility, which allows a decent and robust analysis of the conceptual idea (Clark & Wójcik, 2004). Their argumentation behind the positive relationship between ownership concentration and stock price volatility is reasoned on basis of information circulation:.. the degree of ownership concentration in particular, directly affect the circulation of information crucial for the assessment of firm market value. If the circulation of information is primarily internal, as we expect in a company with higher concentration of ownership, outside investors and agents trading its stocks are likely to be more uncertain about the true value of the company. Less agreement on firm value and the potential for insider trading would lead to a more volatile daily stock price (Clark & Wójcik, 2004, p.916). This argumentation is initiated by Merton (1987), given incomplete information, there will be a beneficial effect on the stock price when the amount of investors in a particular stock increases. Underlying argument is that each investor will specialize in information collection for a given firm, this well in the end reduce information asymmetries. Clark & Wójcik (2004) and Wang (2007) extent on this theory by arguing that increasing the amount of investors in a firm will reduce volatility. Each investor has only partial information about the firm, however the available information will become more accurate when the number of investors grows. Jankensgård & Vilhelmsson (2016) refer to this theory as the investor base-broadening effect: an increase in the number and diversity of the investor base will lead to lower volatility of the share price. Wang (2007) uses this reasoning to predict that the presence of foreign investors will decrease the volatility in emerging markets. Furthermore, Ruben et al. (2009) use the same reasoning with respect to institutional investors; institutional investors should improve the information content of the price due to their relatively higher financial sophistication. 6

7 Trading channel In literature about volatility, the link between prices and trading is referred to as the trading channel; when trading occurs, prices tend to move (Karpoff, 1987; Gallant, Rossi, & Tauchen, 1992; Zhang, 2010). As Karpoff (1987) shows in a survey of literature linking trading volume to price changes, this relationship deserves examination on its own. Volatility arises because of changing trading volumes, moreover trading volume is positively related to volatility (Karpoff, 1987). Stocks that are characterized by a small number of investors will have a lower liquidity (Ginglinger & Hamon, 2007). By means of this argumentation more investors (lower ownership concentration) would lead to more volatility in the share s price. This argumentation is in contradiction with the investor-base broadening effect. Enzazi et al. (2011) find empirically a positive relationship between the ownership concentration measured in holdings of the largest shareholders in percentage and share price volatility, providing evidence for the existence of such a trading channel. However, no relationship is found between the ownership concentration measured in holdings of the five greatest shareholders and share price volatility (Ezazi, Sadeghi, Alipour, & Amjadi, 2011), therefore no undisputed evidence is offered that ownership concentration and volatility are related. Moreover, they do not test the trading channel directly by incorporating trading volume as a variable in their analysis. Same holds for the analysis of Alzeaideen and Rawash (2014) who empirically try to link ownership concentration to share price volatility in the Jordanian stock exchange. The relationship between ownership concentration and share price volatility seems to be more evident in the Jordanian stock exchange, a positive relationship is found between ownership concentration and share price volatility by using the percentage ownership of the largest shareholder as measurement of ownership concentration as well as using the percentage ownership of the largest five shareholders as measurement (Alzeaideen & AL-Rawash, 2014). More recently, Jankensgård & Vilhelmsson (2016) tested the investor base-broadening effect. Results indicated the contrary; support was found in favour of the trading channel approach (Jankensgård & Vilhelmsson, 2016). A separate regression with trading volume as dependent variable indeed indicate support for the existence of such a trading channel. However, Jankensgård & Vilhelmsson (2016) use only the direct size and diversity of the investor base as measurements. In their analysis of ownership structure actual measurements of ownership concentrations do not occur (Jankensgård & Vilhelmsson, 2016). This brings us to the problem of different ownership structure measurements. 7

8 OWNERSHIP STRUCTURE MEASURES As described, effects of differences in ownership structures are already tested for years. Most commonly studied is the relationship between corporate performance and ownership structure. Empirical results linking this relationship are however mixed and offer no clear guidance for corporate governance (Yasser & Mamun, 2017). Part of the problem are the different dimensions in which ownership concentration can be tested, causing a lot of variation in measurements of ownership concentration. The question is whether simple measures of ownership concentration are sufficient to capture the role of ownership, or if more complicated measurements based on power indices are needed (Overland, Mavruk, & Sjögren, 2012). Moreover, Overland et al. (2012) distinguish three major reasons why results differ between studies that try to link ownership concentrations to corporate performance: - Contextual settings differ, example will be the legal background that matters. - Data quality differs, disclosure of ownership information is limited. - Methodology differs, distributions of ownership concentration measures actually differ (Edwards & Weichenrieder, 2004). Results of Overland et al. (2012) indicate that the choice of the ownership concentration measure matters. It is key to look at what measure fits to your research question. Most studies analysing the effects of ownership concentration use relatively simple measures relating to ownership concentration. Most simple measure used will be the share held by the largest shareholder (Thomsen & Pedersen, 2000). Additional, the combined fraction of shares held by a number of the largest owners is often used as a proxy for ownership concentration (McConnell & Servaes, 1990; Demsetz & Villalonga, 2001; De Miguel, Pindado, & Torre, 2004). Herfindahl indices originally constructed to measure market concentration are also used as a proxy for ownership concentration within a company (Demsetz & Lehn, 1985; Leech & Leahy, 1991). The Herfindahl index is calculated by summing the squared percentage of shares held by each shareholder. Advantage of such an index is that it includes all shareholders in a single concentration measure (Overland, Mavruk, & Sjögren, 2012). On the other hand, more advanced power indices based on game theory are used to measure ownership concentration as well (Zingales, 1994; Rydqvist, 1996). These more theoretically elaborated power indices are suitable if the aim of research is focussed on the shareholder conflict dimension (Overland, Mavruk, & Sjögren, 2012). Overland et al. (2012) argue that if the research aim is focussed on a monitoring dimension simpler measures that emphasize the largest owner(s) are preferred. 8

9 Jankensgård & Vilhelmsson (2016) tested whether improving the quality of the price signal (information) leads to lower volatility. Due to their focus on this investor-base broadening effect, they measured the variety in corporate ownership by looking at how many investors a firm has and the diversity of this investor base (Jankensgård & Vilhelmsson, 2016). Mainly used are the number of shareholders that have an ownership stake larger than 0.1% and the micro-float, defined as the fraction held by investors with a stake smaller than 0.1%. No previously mentioned ownership concentration measures are used, therefore their analysis is inconclusive when trying to link ownership structure to share price s volatility. HYPOTHESIS DEVELOPMENT Based on this theoretical background, we will develop the hypotheses that will be tested empirically. First of all, the investor-base broadening effect brings us to the next hypotheses: Hypothesis 1a : The degree of ownership concentration within a stock will positively affect the stock s price volatility Hypothesis 1b : The size of the investor base within a stock will negatively affect the stock s price volatility The trading-channel interpretation predicts opposite findings. If an opposite relationship is found in testing hypothesis 1a and 1b, findings suggest a trading-channel interpretation. It will be interesting to further test whether the trading channel will be valid for our sample by including trading volume as dependent variable: Hypothesis 2a : Trading volume will positively effect stock s price volatility Hypothesis 2b : The degree of ownership concentration within a stock will negatively affect trading volume Hypothesis 2c : The size of the investor base within a stock will positively effect trading volume. According to these hypotheses, the relationship between ownership structure and share price s volatility will be tested by using the degree of ownership concentration and the size of the investor base. Further elaborations on the exact variables used to test these hypotheses will be stated in the next chapter. 9

10 Data and methodology SAMPLE AND DATA Data will be retrieved from Eikon/Thomson Datastream. Detailed information regarding shareholder details will be obtained from the Shareholder s History Reports available on Eikon. Research period consists of the years Data will be included from in total 150 firms, 75 Dutch listed firms as well as 75 firms listed in The United States. For the Dutch listed firms the sample consists of the 25 firms listed on the AEX, 25 firms listed on the AMX and the 25 firms listed on the AScX. For the US firms the sample consists of 25 firms listed on the S&P 500 and 50 firms listed on the S&P SmallCap 600. Due to these characteristics of the dataset, large as well as relatively smaller firms are incorporated. Moreover, not only firms from one single country are included, which allows a more robust analysis of the relationship between ownership structure and share price volatility. The shareholder s history report obtained from Eikon contain detailed information regarding outstanding position, equity assets, investor type and sub-type, investment style and origin of the investor. Based on the Thomson Reuters Business Classification (TRBC) industry composition of the sample is as follows: 6% energy, 9% basic materials, 15% industrials, 11% cyclical consumer goods & services, 9% non-cyclical consumer goods & services, 18% financials, 12% healthcare, 15% technology, 4% telecommunications services, 1% utilities. METHODOLOGY As mentioned, one of the most important findings related to individual stock s volatility is a trend over time (Schwert, 1989; Campbell, Lettau, Malkiel, & Xu, 2002; Zhang, 2010). It therefore might be interesting to test whether a trend in volatility exists in our dataset. Moreover, it might be important to consider the problem of endogeneity discussed in the literature overview and to test which specific model has to be used to analyse the longitudinal observations. In general, our examination will consist of a panel data methodology. The derived model will use volatility as a dependent variable and ownership structure as an independent variable. Ownership structure will be analysed in ownership concentration measures as well as measures that indicate the size of the investor base in a stock. See overview below for further clarification of the variables and measurements used. 10

11 Type Name Symbol Measurement Dependent variable Dependent variable Independent variable Control variable Share price volatility VOLA Standard deviation in returns Trading volume VOLU Number of shares traded, annualized LARG % holdings of largest Ownership shareholder concentration FIVE % holdings of largest 5 shareholders HERF Herfindahl-index MICRO Combined % holdings of shareholders with ownership stake below 0.1% Total amount of TOTA shareholders Micro-float: amount of shareholders with MICR ownership stake below Investor base size 0.1% relative to total amount of shareholders DIFF Amount of shareholders with ownership stake above 0.1% Size SIZE Total assets Earnings uncertainty EARN Actual earnings/total assets Dividend policy DIVI Dummy indicating whether firm pays dividend or not Leverage LEVE Value of debt/total assets. Table 1. Overview variables/measurements 11

12 Dependent variables Volatility will be determined by the annualized total volatility, measured based on the standard deviation in log returns. Trading volume will be defined as the number of shares traded for a stock during that particular year. Independent variables The analysis is not based on the shareholder conflict dimension in which Overland et al. (2012) state more advanced power indices based on game theory might be necessary to measure ownership concentration. We simply want to empirically analyse the relationship between ownership structure and individual stock s volatility by using ownership concentration as a dimension. Overland et al. (2012) describe this as the monitoring dimension and the relatively simpler concentration measures are therefore sufficient. The percentage holdings of the largest shareholders, the percentage holdings of the five largest shareholders combined and the Herfindhal-index will be used as measures of ownership concentration. The Herfindhal-index is calculated by summing the squared percentage holdings of each shareholder. These measurements are in line with previous literature using simple ownership concentration measures (Overland, Mavruk, & Sjögren, 2012). Moreover, the combined percentage holdings of shareholders with an ownership stake below 0.1% will be taken into account as well (MICRO). One could interpret this variable MICRO as the relative importance of investors with an ownership stake below 0.1%. If this variable is low, the importance of these investors as a group is low. The size of the investor base within the listed stocks will be measured by using data regarding the number of unique shareholders a firm has. To increase the robustness of the analysis we will do an analysis with the amount of investors with ownership stake above 0.1% (DIFF). Micro-float will be defined as the amount of investors with an ownership stake smaller than 0.1% relative to the total amount of shareholders (MICR). Thus, three variables will be used for the amount of shareholders: total number of shareholders (TOTA), micro-float (MICR) and the amount of shareholders with stake above 0.1% (DIFF). In this way, analysis can lead to a robust testing of hypothesis 1b. Note that the variable MICRO captures the combined percentage holdings of shareholders with an ownership stake smaller than 0.1%, while MICR is defined as the percentage of shareholders with an ownership stake below 0.1% (relative to total amount of shareholders). 12

13 Control variables Several firm-specific control variables will be integrated in the regression analysis. Firm s size will be used as this came forward as an important determinant of volatility in literature (Campbell, Lettau, Malkiel, & Xu, 2002; Rubin & Smith, 2009; Jankensgård & Vilhelmsson, 2016). Size will be measured using total assets. Even though Shiller (1991) already observes that stocks are way too volatile to be explained by earnings and dividends only, it is important to include such fundamentals in the examination of volatility. Uncertainty in earnings is important in explaining return volatility (Wei & Zhang, 2006). We will try to capture the uncertainty in future cash flows by including the actual earnings/total assets as a control variable. Actual earnings tend to be correlated with the volatility of earnings and will therefore be used as a proxy for earnings uncertainty similar as Jankensgård and Vilhelmsson (2016) did. Firm s that pay no dividend have more volatile returns (Pástor & Pietro, 2003). Jankensgård & Vilhelmsson (2016) confirm that a firm s dividend policy is of significant explanatory power on volatility. Most crucial finding of Ruben & Smith (2009) is that the sign of the correlation between institutional ownership and volatility depends on whether the firm pays dividend or not. These findings, provide a motive to include the dividend policy of a firm as a control variable. Leverage is another factor repeatedly coming back as determinant of volatility of cash flows to equity investors (Campbell, Lettau, Malkiel, & Xu, 2002; Pástor & Pietro, 2003; Rubin & Smith, 2009; Jankensgård & Vilhelmsson, 2016). Reasoning is that when leverage increases, stockholders will bear a greater share of the total cash-flow risk of the firm, stock s volatility will increase accordingly (Black, 1976; Campbell, Lettau, Malkiel, & Xu, 2002). The variable leverage will be defined as value of debt divided by total assets. Hypothesis testing In general form the estimation specification to test for hypothesis 1a and 1b will look like: Volatility = α + β 1 OwnershipStructure + β 2 Size + β 3 Earnings + β 4 Dividend + β 5 Leverage + ε (1) By using the several measures for ownership structure and estimating these specifications separately we hope to identify a robust relationship between ownership structure and volatility. As shown in specification 1, the estimation will include size, leverage, earnings uncertainty and the firm s dividend policy as control variables. 13

14 Hypotheses 2 relate to a trading-channel interpretation. This implicates that shareholder ownership affects volatility through trading volumes. By including a variable trading volume we can test whether our ownership structure measures effect trading volume. First of all, it is important to check whether these trading volumes actually effect volatility in our dataset. Therefore, specification 2 will be tested first. Volatility = α + β 1 TradingVolume + β 2 Size + β 3 Earnings + β 4 Dividend + β 5 Leverage + ε (2) This specification includes trading volume as an explanatory variable of volatility. According to Jankensgård and Vilhelmsson this is unsatisfactory since: both are natural left-hand side variables that are determined by more fundamental ones (Jankensgård & Vilhelmsson, 2016, p. 21). We argue however, that it does not make sense to test the trading channel interpretation with respect to ownership measures without testing whether trading volume actually effects volatility in our dataset. Specification 3 will actually test whether the trading channel interpretation is valid for our ownership measures. In this estimation trading volume will be included as the dependent variable and will be explained by the same set of independent variables as in our main analysis of volatility. Results of this specification allows us to test hypothesis 2b/2c and potentially provides evidence in favour of a trading-channel interpretation. Trading Volume = α + β 1 OwnershipStructure + β 2 Size + β 3 Earnings + β 4 Dividend + β 5 Leverage + ε (3) 14

15 Empirical analysis DESCRIPTIVE STATISTICS The summary statistics for the variables can be found in table 2 in the appendix. With a total number of 149 listed firms and a time span of 7 years, the maximum amount of observations in our dataset would be Due to missings, most variables only have around 900 observations. Useful variations are existing in the several measures for ownership concentration and amount of shareholders. For example, the micro-float exhibits a minimum of 0 % and a maximum of %. This means that there is a firm in our dataset with no shareholders holding a stake below 0.1% and a firm with 96.57% of its shareholders consisting of shareholders with a stake below 0.1%. The control variable earnings uncertainty displays a negative minimum, note that this is possible cause earnings uncertainty is defined as the net income divided by total assets. Net income can be negative during a year, therefore resulting in a negative earnings uncertainty. The scale of size and trading volume are altered by taking the natural logarithm. The variable dividend policy is incorporated as a dummy, indicating whether the firm pays dividend or not in a particular year. From the 912 observations in this dividend dummy, 72% is indicating that the firm is paying dividend and 28% of the observations are not paying dividend. The control variable leverage exhibits a variation between 0 % and % with a mean value of 22.4%. There are thus firms in our dataset with no value of debt at all, indicating an absence of financial leverage. Correlations between the several measurements used for ownership structure are shown in table 3. As one would expect the different measurements for ownership structure show clear correlation with each other. Large correlations are existing between LARG, FIVE and HERF. Moreover, MICRO has a high correlation with TOTA, MICR and DIFF. The measurements related to the amount of shareholders show a high correlation with each other as well. These findings are not problematic, since these measurements are used as supplement to each other. They will be used in separate regressions. According to findings of Overland et al (2012) the measure used for ownership structure matters for inference. One can t substitute these several measures based on their correlation coefficients due to differences in distributional properties. Including the several measurements is therefore of added value to the analysis, it will increase the robustness of our findings. 15

16 INVESTOR BASE BROADENING EFFECT Since we have no variables in our model that remain stable over time, expectation is that a fixed effects model will be most appropriate to analyse the longitudinal data. Fixed effects model is indeed proven to be most appropriate by comparing the coefficients of a fixed-effects model and a randomeffects model using Hausmann specification tests. Results of specification 1 can be found in table 4. The specification is estimated several times, each time with a different measure of ownership structure. Hypothesis 1a can be rejected based on findings of model 1, 2 and 3. Hypothesis 1a states that the degree of ownership concentration within a stock will positively affect the stock s price volatility. This is the expectation based on the investor base broadening effect. Therefore, positive relationships are expected between the percentage holdings of the largest shareholder (LARG), the percentage holdings of the largest five shareholders (FIVE), the Herfindahl index (HERF) and share price volatility on the other hand. Findings presented in table 4 show no significant relationship between these variables, although the relationship seems to be positive. The combined percentage holdings of shareholders with an ownership stake below 0.1% shows a negative, significant estimator. If this group does have a higher ownership stake, the ownership concentration within that stock will be lower. Therefore, model 4 provides us with evidence in favour of hypothesis 1a. Share price volatility becomes lower, the larger the combined stake held by investors with an ownership stake below 0.1%. In general, we can conclude that some evidence is found in favour of hypothesis 1a but these findings lack persuasiveness. Model 5, 6 and 7 are specificated to test for hypothesis 1b: the size of the investor base within a stock will negatively affect the stock s price volatility. Model 5 shows us that the total amount of shareholders within a stock (TOTA) has a negative significant effect on share price volatility. The higher the amount of shareholders the lower share price volatility, which will be in favour of the investor base broadening effect. Note that the coefficient is rather small , however this makes sense cause TOTA is the total amount of shareholders and varies within a range of (see table 2). This result is confirmed by model 6 and 7. Model 6 incorporates the micro-float: the amount of shareholders with an ownership stake below 0.1% as a percentage of the total amount of shareholders (MICR). This variable MICR shows again a negative, significant estimator with share price volatility. Same holds for the variable DIFF which is defined as the amount of relatively larger shareholders (stake exceeding 0.1%). Therefore, evidence is found in models 5, 6 and 7 supporting hypothesis 1b: the size of the investor base within a stock will negatively affect the stock s price volatility. 16

17 All in all, when using measures related to the size of the investor base (TOTA, MICR, DIFF) clear evidence is found to believe that the size of the investor base in a stock will negatively affect the stock s price volatility, providing evidence in favour of the investor base broadening effect. However, no comprehensive evidence is found in favour of the investor base broadening effect when testing the ownership concentration measures in hypothesis 1a (LARG, FIVE, HERF, MICRO). Thus leading to mixed results, the investor base broadening effect is not well supported. Opposite findings that potentially indicate a trading-channel interpretation aren t found as well in the analysis of ownership concentration measures. The findings related to the control variables seem to be consistent with the expectations. Size is an important determinant of volatility. Large firms have significantly lower volatility in our dataset. Moreover, earnings uncertainty (EARN) has a large explanatory power. Firms which have high earnings realization are associated with lower volatility. The negative coefficient of earnings uncertainty (EARN) is 1%-significant at each model. Findings related to dividend policy seem to be in line with literature as well, however profound evidence is not present. Firms that pay dividends are associated with a lower volatility, however only in some models the coefficient is significant at a 10% level. Profound evidence might not come forward because only 28% of our observations are not paying dividend. The control variable leverage (LEVE) defined as the value of debt divided by total assets does not hold any explanatory power in our dataset. One would expect higher leveraged firms to have a higher volatility (Black, 1976; Campbell, Lettau, Malkiel, & Xu, 2002). Nevertheless, our findings are in line with Wei & Zhang (2006), who do not find any significant relationship between leverage and stock return volatility. TRADING CHANNEL The trading-channel interpretation predicts exactly the opposite of the investor base broadening effect. As mentioned, there is no indication for a trading-channel interpretation in our results of hypothesis 1. However, it will be interesting to further test the trading channel in our dataset. First of all, to test this interpretation it is important to provide evidence that the trading volume actually positively influences the share price volatility. This relationship is tested by specification 2. Results are shown in table 5 and exhibit evidence that trading volume significantly effects stock return volatility. Therefore, hypothesis 2a is supported in our dataset. Since trading volume is transformed by the natural logarithm we can interpret the results as a 1% increase in trading volume would lead to a percentage increase of in volatility. To further test this trading channel interpretation, the same ownership measures will be used with trading volume as dependent variable in specification 3. Recall that the trading channel refers to the 17

18 finding that firms with more concentrated ownership display lower liquidity (Ginglinger & Hamon, 2007). According to this interpretation negative estimators are expected for the percentage holdings of the largest shareholder (LARG), the combined percentage holdings of the five largest shareholders (FIVE) and the Herfindahl-index (HERF). Table 6 shows the results of these specifications, we observe indeed significant negative estimators for these three measures indicating that ownership concentration has a negative effect on trading volume. The combined percentage holdings of shareholders with ownership stakes below 0.1% (MICRO) does not have a significant effect on trading volume. Therefore, these specifications confirm hypothesis 2b. The degree of ownership concentration within a stock negatively effects trading volume. Hypothesis 2c which states that the size of the investor base within a stock will positively effect trading volume, will be tested by the total amount of shareholders in a stock (TOTA), the amount of relatively larger shareholders (DIFF) and the percentage of small shareholders in a stock (MICR). Therefore, positive estimators are expected for these variables. Findings in table 6 indicate a significant, negative relationship between the total amount of shareholders in a stock and trading volume. The higher the total amount of shareholders, the lower the trading volume will be. This would reject hypothesis 2c. However, on the other hand findings indicate a significant, positive relationship between the amount of relatively larger shareholders (DIFF), the percentage of small shareholders (MICR) and trading volume. Analysis by using these measures would fail to reject hypothesis 2c. Results related to hypothesis 2c are therefore mixed, no clear guidance can be derived. The amount of relatively larger shareholders does indeed contribute positively to the trading volume, while the total amount of shareholders does negatively contribute to trading volume. Potentially, this indicates that the group of investors with small ownership stakes do not have a positive effect on trading volume. However, the variable MICR indicates that this is at the same time the case. Results related to hypothesis 2c are therefore contradicting and offer no clear guidance. We can conclude based on these findings that in our dataset clear evidence is found in favour of an investor base broadening effect when using measures related to the size of the investor base in a stock, while clear evidence is found with respect to a trading channel when incorporating simple ownership concentration measures. The control variables related to volatility seem to hold high explanatory power on trading volume as well. Notable are the variables SIZE and EARN which contribute positively to trading volume but negatively to volatility. Dividend has a negative relationship with both volatility and trading volume, however the relationship with trading volume seems to be more robust. The degree of financial leverage (LEVE) again shows no significance at all with trading volume. 18

19 ROBUSTNESS CHECK In the correlation matrix in table 7, one can observe that the variables MICRO, TOTA and MICR do have a high correlation with SIZE and the variables MICRO, TOTA, MICR and DIFF have a high correlation with VOLU. These relatively high correlation coefficients can be problematic, since these are incorporated as explanatory variables. It would therefore be difficult to fully attribute the observed effect to the ownership structure variables. However, to overcome this issue size cannot just be ignored. In our regression, size came forward as an important determinant of volatility. Simply, removing the control variable size would therefore lead to a large omitted variable bias in our estimates. To overcome this problem, results are tested while using SIZE and VOL untransformed. In this way, correlation is no concern anymore as shown in table 8. Regression results are robust to using untransformed variables, results by using untransformed variables are similar. Jankensgård and Vilhelmsson (2016) use one period lags for their independent variables, reasoning is that the variable volatility is estimated based on daily stock returns while the explanatory variables are measured at year-end. In our dataset, the ownership structure measures are measured at yearend as well. Data in the shareholder history reports reflect the situation at year-end. Moreover, control variables are based on year-end observations. Therefore, it might be interesting to lag our independent variables similar to Jankensgård and Vilhelmsson (2016). Results of these lagged regressions are interesting. Previously, hypothesis 1a was rejected because no significant, positive relationship was observed. However, when lagging the independent variables by 1-year, significant negative relationships are established between the ownership concentration measures and share price volatility. The more concentrated the ownership, the lower the volatility will be. This is exactly the opposite hypothesis 1a predicts. It provides us with evidence in favour of the trading-channel interpretation. See table 9 for these regression results when lagging the independent variables 1 year. Moreover, before these lagged estimates evidence failed to reject hypothesis 1b. Models 5, 6 and 7 in table 9 show no significant results anymore when using the lagged variables, therefore rejecting hypothesis 1b. Concluding, when using lagged variables no evidence is found at all for the investor base broadening effect. Findings even indicate a trading channel interpretation. Further tests of this trading channel interpretation are estimated with lagged variables as well. These results are similar to previous estimations, findings for hypothesis 2a, b and c are robust to the lagged variables approach. In general, these additional tests led to an interesting insight: initial evidence in favour of the investor base broadening effect does not seem reliable, while robustness tests provide us with even more evidence to believe a trading channel approach in our dataset. Which means that the volatility 19

20 in a stock will increase by the size of the investor base and accordingly more concentrated ownership will lead to lower volatility. 20

21 Conclusion The dataset used in this empirical analysis consists of some crucial elements to determine a relationship between ownership structures and stock return. It incorporates firms listed in the United States as well as firms listed in The Netherlands. Moreover, it does include firms from different exchange indexes leading to a larger variety in firms included. Ownership structure is analysed by ownership concentration measures and measures related more directly to the size of the investor base. Due to this variety in used measurements, a profound analysis of the relationship between share price volatility and ownership structure emerged. Initial analysis found evidence in favour of an investor base broadening effect when using measures related to the size of the investor base and evidence in favour of a trading channel approach when incorporating ownership concentration measures. By making the same regressions with lagged independent variables, evidence in favour of the investor base broadening effect does not seem reliable. The evidence in favour of the trading channel approach holds under lagged independent variables and does even seem to improve. Using lagged independent variables does seem to be legit due to the difference in time measurement of the dependent and independent variables. Finally, we can conclude that the volatility in stock returns will increase by the size of the investor base and accordingly more concentrated ownership will lead to lower stock return volatility. Explanation is given in the form of trading volumes, the larger the size of the investor base the more trading will occur, subsequently leading to larger volatility. The theorem that more investors provide more information content which on its turn will lead to a more accurate stock price doesn t hold in our dataset. Results are therefore confirming Jankensgård and Vilhelmsson s (2016) finding that a larger investor base is associated with higher volatility instead of lower volatility predicted by a more accurate price signal. 21

22 Appendix Variable Observation Mean Std. Dev. Minimum Maximum Volatility (VOLA) Trading Volume* (VOLU) Ownership concentration % holdings of largest shareholder (LARG) % holdings of largest 5 shareholders (FIVE) Herfindahl Index (HERF) Combined % holdings of micro-float (MICRO) Amount of shareholders Total amount shareholders (TOTA) Micro-float (MICR) Amount of shareholders with stake > 0.1% (DIFF) Control variables Size* (SIZE) Earnings uncertainty (EARN) Dividend Policy (DIVI) Leverage (LEVE) Table 2. Descriptive statistics. * transformed with natural logarithm 22

23 LARG 1 2.FIVE HERF MICRO TOTA MICR DIFF Table 3. Correlation matrix measures ownership 23

24 Intercept *** (5.13) LARG (0.05) Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model *** (4.75) FIVE (1.12) *** (5.12) HERF (0.33) *** (3.49) MICRO *** (-4.52) *** (4.09) TOTA *** (-3.05) *** (4.51) MICR ** (-1.98) *** (4.83) DIFF ** (-2.35) SIZE *** (-3.36) EARN *** (-8.34) DIVI * (-1.82) LEVE (-0.06) *** (-3.21) *** (-8.27) * (-1.68) (-0.07) *** (-3.34) *** (-8.35) * (-1.80) (-0.08) (-1.31) *** (-8.20) (-1.35) (-0.16) ** (-2.18) *** (-8.50) (-1.62) (0.18) ** (-2.37) *** (-8.22) (-1.58) (-0.16) *** (-2.72) *** (-8.10) (-1.50) (-0.14) Observations R-sq (overall) Table 4. Model estimations ownership structure, dependent variable = Volatility - t-statistics are in parentheses - significance at 1%, 5% and 10% level is indicated by ***,**,*, respectively 24

25 Trading-channel interpretation Intercept (1.34) VOLU *** (10.99) SIZE *** (-5.22) EARN *** (-10.64) DIVI (-1.38) LEVE (-0.53) Observations 895 R-sq (overall) Table 5. Model estimation trading volume, dependent variable = Volatility - t-statistics are in parentheses - significance at 1%, 5% and 10% level is indicated by ***,**,*, respectively 25

26 Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Intercept *** (10.03) *** (11.06) *** (9.85) *** (8.89) *** (7.10) *** (9.87) *** (10.39) LARG *** (-4.58) FIVE *** (-6.86) HERF *** (-3.92) MICRO (-0.39) TOTA *** (-9.26) MICR *** (2.77) DIFF *** (6.41) SIZE *** (4.83) *** (4.41) *** (4.86) *** (4.81) *** (7.92) *** (3.68) *** (3.59) EARN *** (4.99) *** (4.16) *** (4.99) *** (4.56) *** (4.48) *** (4.36) *** (3.93) DIVI ** (-2.27) *** ** (-2.20) * (-1.74) (-1.15) ** (-2.16) *** LEVE (0.71) (-2.95) (0.57) (0.79) (0.52) (1.49) (0.66) (-2.91) (0.72) Observations R-sq (overall) Table 6. Model estimations ownership structure, dependent variable = Trading Volume - t-statistics are in parentheses - significance at 1%, 5% and 10% level is indicated by ***,**,*, respectively 26

27 SIZE 1 2.VOLU LARG FIVE HERF MICRO TOTA MICR DIFF Table 7. Correlation matrix transformed variables SIZE and VOLU SIZE 1 2.VOLU LARG FIVE HERF MICRO TOTA MICR DIFF Table 8. Correlation matrix untransformed variables SIZE and VOLU. 27