The determinants of managerial ownership and the ownershipperformance

Transcription

1 The determinants of managerial ownership and the ownershipperformance relation Student name: Huib Raterink Administration number: Faculty: Economics and Management Department: Finance Supervisor: dr. A. Manconi Date:

2 1

3 Acknowledgements This research could not have been realized without the help of others. I would especially like to thank my thesis supervisor, dr. A. Manconi, for giving me valuable feedback week after week and for staying positive at all times. I would also like to thank Marloes Coppoolse for her never-ending optimism and support, and for putting up with me the past year. Lastly, I would like to thank my parents and Max for helping me wherever they could. 2

4 Table of contents 1. Introduction Hypothesis development Determinants of managerial ownership The ownership-performance relation Data and methodology Data Econometric problem Methods Empirical results Determinants of managerial ownership Determinants of firm performance: fixed effects estimation Determinants of firm performance: IV estimation Determinants of firm performance: sample splits Conclusion References Appendix

5 1. Introduction Being the CEO of Apple from 2000 until 2011, the late Steve Jobs earned a yearly salary of only $1. 1 While that may seem strange, other CEOs have had $1 salaries as well. Some examples include Google s Eric Schmidt 2, Facebook s Mark Zuckerberg (as of 2013) and Mr. Jobs s successor at Apple Tim Cook. Obviously, these men are compensated in other ways. 3 In their extensive study on CEO pay, Frydman and Jenter (2010) show that stock-based compensation like restricted stock grants and option grants have become increasingly important in the composition of CEO pay. Indeed, Mr. Cook was granted 1,000,000 Restricted Stock Units (RSUs) as a promotion and retention reward when he was appointed. 4 A natural question, then, is: why not simply pay these executives with salary? The answer lies in agency theory. Berle and Means (1932) were the first to note that the separation of corporate ownership and control results in a conflict of interests between shareholders (the principal) and managers (the agent). Unlike shareholders, managers do not necessarily want to maximize firm value. Instead, they may want to engage in empire building, spend money on pet projects that do not create value or consume corporate assets. Because shareholders cannot perfectly observe the actions of managers, there is a need for incentive alignment in order to reduce agency costs. A well known solution is to make managers themselves owners of the firm by giving them an equity share. This is exactly the reasoning behind Mr. Job s compensation package as formulated by Apple: While Mr. Jobs served as CEO, the Company believed that his level of stock ownership significantly aligned his interests with shareholders interests; therefore, his total compensation consisted of $1 per year. - Apple s 2012 Proxy Statement While the theory described above may sound perfectly reasonable, in the end I am interested in whether executive ownership really is effective in increasing firm performance. This is the main topic my thesis. More specifically, I replicate and extend the research of 1 Source: Apple Inc. proxy statements. 2 Source: Google Inc. proxy statements. 3 For instance, Steve Jobs was granted options worth over 500 million USD in 2001 and restricted stock grants worth 75 million USD in 2003 (source: Compustat ExecuComp). 4 Based on the adjusted closing price of Apple stock on the day of his promotion (August 24, 2011), this reward was worth 374,920,000 USD. 4

6 Himmelberg, Hubbard and Palia (1999), hereafter referred to as HHP. My main research question is as follows: Is there an empirical link between managerial equity ownership and firm performance? An impressive body of literature on the relationship between executive ownership and firm performance already exists. The need for further examination stems from the fact that the empirical evidence has not led to consensus on the matter. On the contrary, there is a wide variety in observed ownership-performance relations. I rely on HHP s methodology because they point out a serious flaw in many other studies. According to HHP, ownership and performance are partly determined by common characteristics. Failing to control for these characteristics would result in spurious correlations. Addressing this endogeneity problem of (unobserved) omitted variables is at the heart of their study. Before I focus on the ownership-performance relation, I first examine the determinants of executive ownership. Whereas Jensen and Meckling (1976) associate low levels of ownership with suboptimal compensation design, HHP argue that the compensation contracts observed in the data are endogenously determined by the contracting environment, which differs across firms in both observable and unobservable ways. Put differently, the severity of the moral hazard problem will differ across firms and as a result, some managers may need less equity for incentive alignment. Apart from gaining insight into how equity compensation is determined, the variation in ownership explained by unobservable, time-invariant factors informs us on the possible endogeneity of ownership in regressions on firm performance. As an extension to the HHP investigation, I try to identify an ownership-performance relation by separating firms on the basis of share liquidity, agency costs and institutional ownership. I hypothesize that the relationship is stronger for firms whose shares are relatively liquid because managers will prefer liquid shares to illiquid ones. Agency costs give an indication of the scope for moral hazard and could therefore weaken the relationship. Lastly, institutional investors may want to actively monitor managers because of their large equity holdings, thereby strengthening the relationship. My findings are not in favor of using managerial equity as a mechanism for incentive alignment. Like HHP, I find no convincing evidence for an ownership-performance relation, after controlling for unobserved heterogeneity. The only evidence I find for a correlation is tentative at best, and results from splitting the sample on the basis of share liquidity. This finding suggests that managers may indeed be more motivated to maximize shareholder value 5

7 if they can easily sell their shares. Regarding the determinants of ownership, I reject HHP s statement that observable firm characteristics are strong predictors. However, I corroborate that managerial equity stakes are largely determined by unobserved heterogeneity at the firm level. This thesis makes three contributions to the literature. First, it examines the ownership-performance relation using a very extensive data set (22,463 firm-year observations) consisting of recent data ( ). Second, it shows that the variables used by HHP to instrument executive ownership are actually not strong instruments. Third, it documents how controlling for mediators such as liquidity can help uncover evidence of an ownership-performance relation. The remainder of this thesis is organized as follows. In section 2, I discuss the relevant literature and develop my hypotheses. Section 3 describes our data, discusses the endogeneity issue in an analytical framework, and describes the tests we run. In section 4, I present my empirical results and interpret them. Section 5 concludes. 2. Hypothesis Development 2.1 Determinants of managerial ownership While not directly investigating the determinants of managerial ownership, Demsetz and Lehn (1985) show that firm risk has a significant and positive effect on the degree of ownership concentration. According to them, firms that transact in markets characterized by stable prices, stable technology, stable market shares, and so forth are firms in which managerial performance can be monitored at relatively low cost. In less predictable environments, however, managerial behavior simultaneously figures more prominently in a firm s fortunes and becomes more difficult to monitor. Extending this argument to the determination of executive ownership, the riskier the firm, the higher the equity stake that should be given to managers (all else equal). However, there is also a potential offsetting incentive effect since more equity means more idiosyncratic risk that cannot be diversified away. HHP seek to find which of these effects prevails by controlling for other variables proxying for the scope for moral hazard. These include firm size, capital intensity, R&D expenses, advertising expenses, market power and investment rate. They find that idiosyncratic stock price risk is actually associated with lower levels of managerial ownership. Moreover, observable firm characteristics in the contracting environment strongly predict managerial equity levels and in 6

8 predicted ways, i.e. consistent with the principal-agent model. Using the same set of explanatory variables as HHP, I test if these logical relations still hold based on recent data and therefore formulate the following hypothesis: H1 Managerial ownership is determined by observable characteristics in the contracting environment in ways that are consistent with the principal-agent model. 2.2 The ownership-performance relation As mentioned in the Introduction, there is an impressive body of literature on this relation. Mørck, Shleifer and Vishny (1988) consider a cross-section of fortune 500 firms and estimate a piecewise-linear relation between board ownership and Tobin s Q, the most commonly used measure of firm valuation/performance. They do this because they suspect that managerial ownership affects firm valuation differently dependent on the level of equity holdings. More specifically, two effects are distinguished: the convergence-of-interests and the entrenchment effect. The latter is explained as follows: A manager who controls a substantial fraction of the firm s equity may have enough voting power or influence more generally to guarantee his employment with the firm. [ ] With effective power, the manager may indulge his preference for non-value maximizing behavior. This theory is more or less supported by their results, as the relation is increasing for ownership levels between 0% and 5%, decreasing between 5% en 25% en increasing beyond 25%. The Mørck et al. study was later replicated by McConnell and Servaes (1990), Cho (1998) and Holderness, Kroszner and Sheehan (1999), providing mixed results. 5 In addition, McConnell and Servaes (1990) use a quadratic specification of ownership and find an inverted U-shaped relation. Their evidence suggests that the entrenchment effect dominates the incentive alignment effect when managerial equity stakes reach 40% to 50%. An important caveat to the above findings is that ownership is treated as an exogenous explanatory variable in regressions for firm performance. Jensen and Warner (1988, p. 11) note that this assumption is likely to result in spurious correlations because executive ownership and firm performance may be influenced by common characteristics. HHP give some really intuitive examples. For instance, a firm that has access to a superior monitoring technology will not need to give its managers high levels of ownership to align incentives. At the same time, there is less scope for managers to engage in non-value maximizing activities, 5 The relation for the 0-5% range of ownership is corroborated by all, but not for the other ranges. 7

9 resulting in a higher valuation. Not accounting for this characteristic in the firm s contracting environment leads to a negative ownership-performance relation, which is obviously misleading. Hence, many researchers claim that ownership should be treated as endogenous. This is what Hermalin and Weisbach (1991) do and they also find a significant nonmonotonic relation. Other studies accounting for endogeneity (like Demsetz and Villalonga (2001) and HHP), however, show that ownership fails to explain variations in firm performance. Clearly, the literature does not provide a common view on the ownership-performance relation. Using Tobin s Q as measure of firm performance and building on the theory of Mørck et al., I test the following hypothesis: H2 Managerial ownership affects firm performance in accordance with the theoretical notions of incentive alignment and managerial entrenchment. Because the literature shows that it is generally hard to even identify a relationship between managerial equity ownership and firm performance, I formulate three hypotheses concerning specific circumstances that may strengthen or weaken the relation. The first is related to the liquidity of firms shares. Everything else equal, managers will prefer liquid shares to illiquid shares as they know they can easily convert these into cash at going market prices. I therefore expect liquid shares to provide managers with stronger incentive effects. Another reason why liquidity may affect the ownership-performance relation is that firms whose shares are relatively liquid are probably more transparent than others; traders know their information to be reliable and trade accordingly. Being more transparent, managers will be more inclined to pursue value maximization. I thus hypothesize: H3 The more liquid a firm s shares are, the more positive is the ownership-performance relationship. Besides share liquidity, agency costs may also affect the ownership-performance relationship. High agency costs indicate a strong need for monitoring because there is a scope for managers to use corporate assets for their own benefit. The greater this scope, the less effective an equity stake in the firm is in aligning incentives with shareholders (all else equal). 8

10 My fourth hypothesis is thus: H4 The lower a firm s agency costs are, the more positive is the ownership-performance relationship. Most of the studies mentioned earlier in this section do not solely focus on managerial ownership but on ownership structure as a whole. Some, like Demsetz and Villalonga (2001), include the fraction of shares owned by a firm s largest shareholders in their regressions for firm performance because it may influence the relation we try to identify in this thesis. The reasoning is straightforward: large shareholders have a greater incentive to monitor executives compared to smaller investors (who typically also lack the required expertise). Since institutional investors generally hold large amounts of shares, I hypothesize: H5 The higher the concentration of institutional ownership is, the more positive is the ownership-performance relationship. On a final note, addressing another issue of endogeneity, reverse causality, is not within the scope of this thesis. Studies like Kole (1996), Cho (1998) and Demsetz et al. (2001) claim that their evidence suggests that firm performance actually causes managerial ownership, i.e. executives are rewarded with shares of stock if they do well. While I do not ignore their findings, there at least seems to be some room for interpretation as HHP point out so I leave this issue for future research. 3. Data and methodology 3.1 Data The sample used in this study consists of all Compustat ExeCucomp firms for which additional information on firm characteristics (accounting information) is available in the CRSP/Compustat Merged data set over the period The ExecuComp data set contains information on various types of executive compensation, including the percentage of total shares owned by executives. 6 It is from this figure that I construct my various ownership variables. The CRSP/Compustat Merged data set contains information on balance sheet and 6 Excluding still to be exercised options. 9

11 income statement items, which I need for examining the determinants of managerial ownership and as controls in regressions for firm performance. I retrieved holding period returns and information on share liquidity from CRSP. Finally, Thomson Reuters provides me with measures of institutional ownership that I use for a sample split. Table 1 in the Appendix provides a detailed description of how I constructed each variable and where I retrieved the (underlying) variables from. When I merge all the data into one file, I end up with a sample that is, to my knowledge, much larger than any sample previously used in studies on this particular topic. In total, I have 22,463 firm-year observations on 2,559 firms over a period of 11 years. To put this in perspective, HHP, Mørck et al. (1988) and Demsetz et al. (2001) examine 600, 500 and 223 firms, respectively. Moreover, these and several other studies only consider specific subsets of firms. For example, HHP and Demsetz et al. (2001) require their sample to have no missing observations on certain variables while Mørck et al. (1988) restrict themselves to examining Fortune 500 firms. Since I do not impose any such restriction on my observations, I would argue that besides being more extensive, my sample is also more random because the sample selection bias is less severe. Lastly, the data I am using has a panel structure, i.e. it contains repeated observations over the same units (in this case: firms). This allows me to control for unobserved firm heterogeneity in our regressions with firm-fixed effects. Section 3.2 will elaborate on this. To get a first impression of the relation between managerial ownership and firm performance, the scatter plot in figure 1 displays all observed combinations of the fraction of shares owned by managers and Tobin s Q, a market-to-book ratio which proxies for firm performance. Like all of my other variables, fraction of ownership and Tobin s Q have been censored at the 1 st and 99 th percentiles to prevent distortion of our estimates. Note that I cannot make any inferences on the ownership-performance relationship based solely on figure 1 since both ownership and performance may be influenced by common characteristics (this is related to the endogeneity problem described previously). Our models described in section 3.3 will try to control for these characteristics. 10

12 0 5 Tobin's Q Figure 1: Scatter plot of Tobin s Q on the fraction of shares owned by managers Fraction of shares owned by managers.4 Figure 1 does not show a clear correlation between managerial ownership and firm performance. With a bit of imagination, one can discern a somewhat downward sloping relation in the scatter plot. This observation would not correspond to the theory discussed in section 2 since I expect managerial ownership to positively affect firm performance, at least at lower levels of ownership. Then again, when looking at the lowest levels of firm performance at a given level of managerial ownership, performance seems to be slightly increasing with ownership. I rely on OLS and instrumental variables models to determine the exact relationship (if there is one). Table 2 in the Appendix reports means, medians, standard deviations, minima and maxima of our dependent and independent variables. A comparison with HHP would be very informative here but unfortunately, they do not report summary statistics in their study. In any case, our average fraction of executive ownership of 2.74% is well below the averages found in other studies. For example, Mørck et al. (1988) and McConnell and Servaes (1990) report average ownership stakes of 10.6% and 11.84%, 7 respectively. It is unlikely that this difference is attributable to the nature of our sample since McConnell and Servaes (1990) also consider a large sample consisting of firms from various industries. A more logical 7 Average executive ownership for their 1986 sample. 11

13 explanation is that these studies are relatively old, dating back to the late 1980s/early 1990s, and firms capital stocks have grown in size since then. In any case, the average fraction of executive ownership has gone down percentagewise even though absolute values of managerial ownership have risen over the years (Frydman and Jenter, 2010). Regarding Tobin s Q, my observed average of 1.82 is higher than those found in older studies. 8 However, these values are still within one standard deviation of my average (1.13) so the difference is not statistically significant. 3.2 Econometric problem To get a better understanding of why I use the models described later in this section, I first look at the endogeneity problem in an econometric way by considering the analytical framework developed by HHP. This framework is based on the assumption that management compensation contracts, designed by firm owners (shareholders), include shares of equity for the purpose of aligning managers interests with those of the shareholders. I also assume that other means by which agency costs can be reduced are exhausted so that managerial equity stakes must address the residual agency costs. Now, let be the fraction of shares owned by the managers of firm at time and let and be vectors of observable and unobservable characteristics (unobserved firm heterogeneity) for firm at time. is determined by observable firm characteristics ( ) such as the scope for moral hazard and firm risk but also by unobservable firm characteristics ( ) such as a firm s monitoring technology and intangible assets. Note that the unobservable firm characteristics are considered to be timeinvariant, i.e. they are relatively constant for each firm. This assumption allows me to use fixed effects estimation which I will elaborate on later in this section. Putting the above in a simple model yields: (1) where represents an error term which captures all unobserved factors that are not correlated over time. Once managers know what share of equity they are given, they set their effort level,, accordingly. This effort level will also be influenced by the firm s observable and 8 For example, Mørck et al. (1988) report an average of 0.85 while Demsetz and Villalonga (2001) find an average of

14 unobservable characteristics. For example, managers of firms with a superior monitoring technology will put in more effort, everything else equal. Hence the following linear relation: (2) where is an error term. I now have the ingredients to construct a relation for firm performance, which is captured by Tobin s Q, a measure of firm value. I assume it depends on managers level of effort and observed and unobserved firm characteristics. Hence: (3) with being the error term. Inserting equation (2) into (3) allows managerial ownership,, to be put back into the equation for firm performance: (4) Simplifying equation (4) results in the regression specification used by Mørck et al. (1988) to estimate the effect of ownership on performance with ordinary least squares (OLS): (5) where is the simplified version of. The problem with equation (5) is that the coefficients on managerial ownership and observed firm characteristics can only be estimated consistently if the error term is uncorrelated with both and. Since the choice of how much shares to give to managers is based partially on unobserved firm characteristics, cov(, ) 0 and thus equation (5) cannot be estimated consistently with OLS. In order to remedy the problem formulated above, I control for unobserved firm heterogeneity by including a firm fixed effect for each firm. This is essentially a firm-specific dummy that purges the effect of from the error term, thereby making OLS a suitable estimator again. 13

15 3.3 Methods I estimate equation (6) to test my first hypothesis ( Managerial ownership is determined by observable characteristics in the contracting environment in ways that are consistent with the principal-agent model ): ln 1 & & (6) where and represent firm and time, is a firm fixed effect and is an error term. The fraction of managerial ownership is captured in a logarithmic function to be able to express changes in the explanatory variables into percentage changes in ownership. All of the explanatory variables, except for, are proxies for the scope for discretionary spending or, more generally, for the scope for moral hazard. Important to note here is that I cannot assume the error term to be independently and identically distributed since a) there is a potential for serial correlation because of the panel structure of my data (cov(, ) 0) and b) the variance of the error term is likely to differ across firms (var σ 2, i.e. heteroskedasticity). To deal with this problem, I use robust standard errors which are clustered around firms (Peterson, 2009). Like all other regression models discussed in this section, save the ones on sample splits, I also estimate equation (6) for specifications with pooled data and industry fixed effects. This will tell us something about the extent to which managerial ownership is endogenously determined by the contracting environment. I discuss the expected relations between executive ownership and the explanatory variables, borrowed from HHP, below. The effects of and, measured as the natural log of sales and the natural log of sales squared, on managerial ownership are unclear a priori. I include them nonetheless because Kole (1995) claims that differences in firm size can account for disparities between studies concerning the ownership-performance relation. HHP offer arguments in favor of a positive as well as a negative relationship. Firm size could positively affect managerial stakes because it is harder to monitor managers of large firms and because large firms are likely to 14

16 hire more skilled managers who demand a higher share of ownership. An argument in favor of a negative relationship is that large firms enjoy economies of scale in monitoring managers. I include to allow for nonlinear relations. The effect of firm-specific risk, captured by, on managerial stakes is also not evident a priori. On the one hand, the riskier a firm, the easier it is for managers to engage in non-value-maximizing activities while staying unnoticed, as argued by Demsetz and Lehn (1985). HHP favor a second interpretation. They argue that the added risk to managers portfolios from owning volatile stocks is not easily hedged (firm-specific risk cannot even be hedged at all). Therefore, the risk inherent to firm stock may not give managers the desired incentive beyond a certain level of ownership. is simply a dummy variable that allows me to account for the possibility that firms for which can be calculated ( = 1) 9 are inherently different from firms for which it cannot. and, the ratios of hard capital to sales and hard capital to sales squared (to allow for nonlinear relations), represent a firm s easier to monitor assets. Hard capital is defined here as a firm s property, plant and equipment. These are tangible, long-term assets and thus easily monitored. One would therefore expect to be negatively related to managerial ownership. I expect, &, and to be positively related to managerial ownership. I measure a firm s market power by using the ratio of operating income to sales, the rationale being that firms with more market power are generally more profitable. Managers of powerful firms will be less disciplined by market forces, resulting in a greater need for incentive alignment. This need will also be present for firms that spend relatively large amounts on research and development (R&D) and advertising since both types of spending are vulnerable to managerial discretion. The dummy variables & and are included to distinguish between reporting and non-reporting firms. My last measure for the scope for discretionary spending,, is given by the ratio of capital expenditures to hard capital. This ratio proxies for the link between high growth and opportunities for discretionary projects. 9 A minimum of 20 daily returns on a yearly basis is used as a requirement to calculate a meaningful Sigma. 15

17 In addition to the dependent variable based on the fraction of managerial ownership, I also use the log of average equity holdings per manager, calculated as the value of common shares outstanding times the fraction of shares owned by managers divided by the number of managers, as dependent variable for examining the determinants of executive ownership: ln (7) where is the shorthand notation for the observable explanatory variables in equation (6). Equation (7) serves as an extra test for H1 since I expect to see roughly the same results. The advantage of using average ownership over the fraction of ownership is that people are ultimately motivated by money rather than by the fraction of shares they own. Moreover, takes into account the number of managers (typically 5 per firm) while does not. To test my second hypothesis ( Managerial ownership affects firm performance in accordance with the theoretical notions of incentive alignment and managerial entrenchment ) I estimate the impact of managerial ownership on firm value, represented by Tobin s Q, in several ways. First, I use two specifications of managerial ownership in a regression for Q while controlling for unobserved firm characteristics: (8) (9) McConnel and Servaes (1990) use the quadratic specification in their study as in (8) while Mørck et al. (1988) use the spline specification as in (9) to allow for different slope coefficients for different regions of ownership levels (see table 1 in the Appendix for the interpretation of 1, 2 and 3). These specifications may allow me to identify when, if at all, the entrenchment effect dominates the incentive effect. I also estimate equations (8) and (9) using only the managerial ownership variables and the non-investment set of observable variables. Once again, I estimate all specifications for pooled data, industry effects and fixed effects, resulting in a total of eighteen regressions. An alternative approach to solving the endogeneity problem is instrumental variables estimation. A good instrument of and is correlated with these variables (the instrument should be relevant) but not with the error term in the regression and not with itself (the instrument should be exogenous). If the instrument does affect performance by itself, then it 16

18 should be included in the Q regression to prevent biasedness of the estimators. The idea here is that the instrumental variable affects firm performance only through its effect on managerial ownership, resulting in better estimates for and. I use the same variables as HHP to instrument the ownership variables, namely,, and 10. Applying the two stage least squares (2SLS) method results in the following model: (10) (11) (12) where, and are constants, is the full set of control variables, is the set of control variables without,, and, is the firm fixed effect and, and are error terms. Equation (10) and (11) make up the first stage of the 2SLS method that provides me with the fitted values of the endogenous ownership variables needed for the second stage, denoted by equation (12). A first-stage F-test will make clear whether my instruments are weak or strong based on the variation they explain in and while a Hansen s J-test informs me about the exogeneity of the instruments. Again, I also estimate the IV model using pooled data and industry fixed effects. Lastly, I try to identify an ownership-performance relation by splitting the sample based on firm liquidity, agency costs and institutional ownership. Firm liquidity is proxied by share turnover and two differently computed bid-ask spreads, agency costs are proxied by the ratio of cash to assets and by the ratio of SG&A expenses to sales, and institutional ownership is proxied by the fraction of shares held by institutional investors and by institutional blockholders (see the table 1 in the Appendix for more details on these variables). To test H3, H4 and H5, I run the same regression as in equation (8) on samples split at the median of the variable of interest for each year and for the entire sample period. In addition, I also split the samples into quantiles for each year and for the entire sample period. This makes for a total of 56 regressions (considering I only use firm fixed effects specifications here). Then I compare the coefficients on ownership for the top and bottom 20% of the sample, expecting a greater difference when comparing the top 50% against the bottom 50%. I formally test for the difference between coefficients with a Chow test. 10 There are reasons to believe that these are correlated with ownership but not with performance nor with omitted variables (HHP, p. 379). 17

19 4. Empirical results 4.1 Determinants of ownership Table 3A in the Appendix reports the estimated coefficients on the determinants of the fraction of executive equity ownership. 11 Turning first to the estimates of the pooled model in column (1), I observe that only R&D expenditures have a significant (and unexpected) impact on managerial stakes, apart from the coefficient on which is only marginally significant. This is quite a surprising result, considering that HHP mention that the fraction of executive ownership is affected by nearly all regressors for pooled data. Moving to column (2) where I control for unobserved industry heterogeneity by including industry fixed effects, I observe that the coefficient on Size 2 is no longer significant while the effect of R&D expenses persists, although it is smaller. In terms of economic significance, a one-standarddeviation increase in R&D expenses leads to an average decrease in ownership stakes of 2.81%, 12 which is a modest effect. Moreover, the coefficient on has now become marginally significant and, like the one on &, with a sign that runs counter to my theoretical predictions. Finally, column (3) reports estimates for the specification that includes firm fixed effects. Now, increases in R&D expenses no longer negatively affect ownership stakes (suggesting a correlation with unobserved firm characteristics) while the effect of a firm s investment rate has become slightly more pronounced in terms of statistical significance. The economic impact of a one-standard-deviation increase in is a decrease of 1.88% 13 in managerial stock holdings, a negligible effect. A possible explanation for this negative relation is that capital expenditures are investments in fixed assets and, as I have argued before, these are easily monitored. The estimates in Table 3A provide little support for my first hypothesis. Table 3B reports estimates for the determinants of managerial ownership when ownership is measured as the log of average equity holdings per manager. Again, column (1) reports the estimates of the pooled model while columns (2) and (3) control for industry and firm fixed effects, respectively. Although columns (1) and (2) suggest that there is a significant relationship between some observable characteristics and ownership, only the effect of idiosyncratic firm risk represented by a survives the inclusion of firm fixed 11 Note that our dummy variables SigmaDummy, R&DDummy and AdvertisingDummy are omitted from all regressions because of collinearity

20 effects. This again provides evidence for a correlation between the observed and unobserved characteristics. Consequently, estimates of the model with firm dummies will be more reliable than those of the other models. In terms of economic significance, a one-standard-deviation increase in will result in a decrease average managerial stock holdings of 26.07% 14, which is considerable but not implausibly high. This finding is in line with the interpretation that rewarding managers with risky firm stock comes at the cost of less portfolio diversification and therefore may not lead to the desired incentive effects. Everything considered, the results presented in Table 3A and 3B do not support the hypothesis that observable characteristics in firms contracting environments determine executive ownership in ways that are consistent with the principal-agent model. In fact, ownership is not even affected at all by nearly all variables. This contradicts the findings of HHP. Because ownership is largely explained by unobserved factors, it is essential to control for unobserved firm heterogeneity when regressing firm performance ( ) on ownership ( ) Determinants of firm performance: fixed effects estimation Moving on to the determinants of, table 4A and 4B report, respectively, the estimated coefficients for the quadratic and spline specifications of ownership. Turning first to column (1) of table 4A, both and have a statistically significant effect on when neither observable firm characteristics nor any fixed effects (except for year effects) are controlled for. This finding corresponds to those of studies that fail to account for the endogeneity problem. If the observed estimates for and in column (1) were trustworthy, it would imply an inverted U-shape relation between and with an inflection point at an ownership level of 21.59%, 15 a finding that is perfectly in line with my second hypothesis (the entrenchment effect supposedly dominates the incentive effect when the fraction of ownership exceeds 21.59%). In terms of economic significance, a one-standard-deviation increase in would result in an average increase of 6.98% 16 in. However, introducing industry effects in column (2) results in an that is only marginally significant and its significance is lost altogether when I control for unobserved firm heterogeneity in column (3). Controlling for the full set of control variables (columns (4) to (6)) and for the non-investment set of control variables (columns (7) to (9)) also yields ownership coefficients that are practically never statistically different from zero. I conclude from Table 4A that ownership is indeed an This is calculated by taking the first order derivative of with respect to and solving for

21 endogenous variable in regressions for firm performance and that and are actually not related, even though studies based on cross-sectional data claim the opposite. To my support, HHP find roughly the same significance pattern for the quadratic specification of. The reported estimates from the spline specification of ownership in Table 4B provide even less evidence for an ownership-performance relation. Only 1 is statistically significant in the pooled model without any other controls (column (1)). Its coefficient of indicates an increase of 2.73% 17 in firm performance when ownership below a level of 5% is raised by one standard deviation, an economic effect that can hardly be called significant. The inclusion of industry effects, firm effects and observable contracting determinants in columns (2) through (8) results in insignificant ownership coefficients, apart from the one on 2 in column (7). Although it is not obvious why there is some significance here, I know this estimate to be unreliable because it is not robust to the inclusion of industry and firm effects. In conclusion, Table 4A and 4B cast serious doubt on empirical findings from studies that fail to control for unobserved heterogeneity, be it at the industry level or at the firm level. Both specifications of ownership are not even robust to the inclusion of only observable firm characteristics. Something else worth mentioning is that practically all ownership variables have the predicted signs, even though they are not statistically significant. For example, nearly all estimations in Table 4A signal an inverted U-shaped pay-performance relation (corroborating McConnell and Servaes (1990)) while the signs on 1, 2 and 3 in Table 4B suggest that increasing ownership up to a level of 25% increases performance and decreases it thereafter Determinants of firm performance: IV estimation Can I now safely conclude that managerial stakes do not influence performance? No, because the panel data method used previously might not be the best method to estimate the real population parameter on ownership. For instance, fixed effects estimation does not solve the problem of time-varying omitted variables (Wooldridge, 2009). Table 5 reports instrumental variable estimates for and with,, and 18 as instruments. As usual, column (1) includes no fixed effects apart from year effects, column (2) includes industry effects and column (3) includes firm effects. One immediately notices the extremely high (and insignificant) estimates and corresponding standard errors on and Which is again dropped in both stages of 2SLS because of being collinear to one of the other explanatory variables. 20

22 . This is what happens when the instrumental variables explain only a small fraction of the variation in the instrumented variables (Wooldridge, 2009, p. 511 and p. 515). Our first-stage F-test statistics confirms the weak correlation between the instruments and the ownership variables. According to Stock and Watson s rule of thumb (Stock and Watson, 2011), this statistic should be at least 10 in order to reject the hypothesis that the instruments are jointly zero. Clearly, I cannot reject this hypothesis for any of the estimations reported in Table 5. Moreover, one could have already predicted the above findings simply by looking at Table 3A: none of the instruments are significantly related to the fraction of executive equity holdings. Even though I cannot reject the exogeneity of my instruments (the p-value of a Hansen s J-test is above 0.10 in each column), the ones used by HHP are certainly not suitable for this particular sample. Because the other explanatory variables do not satisfy the requirements for an instrumental variable either, IV estimation does not help identify the effect of ownership on performance Determinants of firm performance: sample splits Turning to the last possibility of detecting an ownership effect, Table 6 reports results from splitting the sample on the basis of the liquidity of a firm s shares. Three proxies for liquidity are used: share turnover, a bid-ask spread based on end-of-month closing prices and a bid-ask spread based on daily closing prices. Column (1) reports estimates for firms which have a share turnover above the median share turnover, which is calculated for each sample year. Column (2) reports estimates for firms which have a share turnover below the median share turnover. Unobserved firm heterogeneity is controlled for in both columns, which is also true for columns (3)-(6). The coefficient on is marginally significant in both columns, with a positive sign for firms that have a share turnover above the median and with a negative sign for firms that belong to the below-median group. In line with our predictions, the observed signs suggest that managers who are actually able to sell their shares at market prices are more motivated to maximize shareholder value than managers who are rewarded with less liquid stock. The negative coefficient in column (2) even indicates that compensation in the form of illiquid firm stock has an adverse effect on firm performance. Columns (3) and (4), which report estimates for firms that face below and above median end-of-month bid-ask spreads, 19 respectively, support the previously mentioned findings. The coefficients on and imply an inverted U-shaped relation between managerial stock holdings and performance 19 Recall that low bid-ask spreads are associated with liquid shares. 21

23 with an inflection point at an ownership level of 17.53% for firms in the above-median group. In addition, the impact of a one-standard-deviation increase in results in a 7.70% 20 change in, which is large enough to be economically significant. In columns (5) and (6), however, the ownership coefficients are no longer significant and this is reflected by the p-value of the Chow test. Whereas the ownership variables were statistically significantly different before (p-values below or practically below 0.10), I cannot reject the hypothesis that the ownership coefficients are different for columns (5) and (6). This also holds for all estimated coefficients from regressions on samples split at the median of each liquidity variable calculated for the entire sample period. The same is true when all previously described regressions are done for the top and bottom 20% firms in terms of share liquidity. Although this casts some doubt on the results reported in Table 6, 21 separating firms by share liquidity does seem to be a potential starting point on the basis of which an ownership-performance relation can be identified. I therefore do not accept, nor reject H3. Regarding all regressions for samples split on the basis of variables proxying agency costs and institutional ownership, I did not include any tables because I nowhere find significant ownership coefficients. Therefore, there is no evidence that supports H4 and H5. 5. Conclusion This thesis replicates and extends the study by Himmelberg, Hubbard and Palia (1999), who found little evidence that executive equity holdings and firm performance are related. Given that my research is based on recent data ( ) and a very extensive data set compared to other research in the field, I first show that managerial stakes are hardly determined by observable factors in the contracting environment. As a result, I reject the hypothesis that managerial ownership is determined by observable characteristics in ways consistent with the principal-agent model. Instead, I find that the variation in ownership is largely determined by unobserved factors at the firm level. This signals the need for controlling for unobserved firm heterogeneity in the regression for firm performance (represented by Tobin s Q). Failing to do so would result in inconsistent estimates for the explanatory variables. I also find that the effect of executive ownership on performance is not / Since we expected that the difference between ownership coefficients would be more pronounced when comparing the top and bottom 20% firms. 22

24 robust to the inclusion of industry and firm dummies, thereby casting serious doubt on the findings of studies that fail to account for unobserved heterogeneity. As an alternative to fixed effects, I use instrumental variables to control for the endogeneity of ownership in the regression. My results show that the instruments used by HHP are certainly not suitable for my sample. Again, I do not find evidence for a correlation between the two variables of interest. Extending the HHP study, I try to identify an ownership-performance relation by separating firms on the basis of share liquidity, agency costs and institutional ownership. Hypothesizing that the relation is stronger for firms with liquid shares, low agency costs and high institutional ownership, I only find some tentative evidence for a correlation when our sample is split on the basis of liquidity. This suggests that managers are more motivated and thus perform better when they are rewarded with liquid stocks instead of illiquid stocks. All in all, my results provide little evidence by which to reject the hypothesis that managerial equity ownership and firm performance are unrelated. However, I do not ignore the fact that certain aspects of my thesis are being questioned by other researchers (and maybe rightly so). For example, Demsetz and Villalonga (2001) provide arguments against the use of Tobin s Q as a proxy for firm performance 22 and point out that managers do not necessarily share common interests. Zhou (2001) even devotes an entire article to the HHP study, criticizing the use of within variation for resolving the issue at hand. Additionally, I noted earlier that fixed effects estimation does not solve the problem of time-varying omitted variables (Wooldridge, 2009). Lastly, my thesis does not address the possibility that managerial ownership is actually determined by firm performance as argued by Kole (1996), though HHP claim that their results allow for a different interpretation. Future research could focus on a variety of issues pertaining to the ownershipperformance relation. Instrumental variable estimation still seems like a viable method for identifying a relationship, the difficulty being finding proper instruments. One should look for policy or regulatory changes that do not affect firm value but do affect ownership on a crosssectional level. Another avenue of inquiry is exploring other factors that strengthen or weaken the relationship. For example, CEOs are more likely to influence firm value than COOs (Chief Operations Officers) which may be reflected by different ownership coefficients. Finally, stock options may be used as alternative mechanism for incentive alignment. 22 As well as against its most logical replacement, accounting profit. 23

25 6. References Berle, A., Means, G., The Modern Corporation and Private Property. New York: Harcourt, Brace and World. Cho, M Ownership structure, investment, and the corporate value: an empirical analysis. Journal of Financial Economics, vol. 47, pp Demsetz, H., Lehn, K., The structure of corporate ownership: causes and consequences. Journal of Political Economy, vol. 93, pp Demsetz, H., Villalonga, B., Ownership structure and corporate performance. Journal of Corporate Finance, vol. 7, pp Frydman, C., Jenter, D., CEO Compensation. Annual Review of Financial Economics, vol. 2, pp Hermalin, B., Weisbach, M., The effects of board compensation and direct incentives on firm performance. Financial Management, vol. 20, pp Himmelberg, C.P., Hubbard, R.G., Palia, D., Understanding the determinants of managerial ownership and the link between ownership and performance. Journal of Financial Economics, vol. 53, pp Holderness, C., Kroszner, R., Sheehan, D., Were the good old days that good? Evolution of managerial stock ownership and corporate governance since the great depression. Journal of Finance, vol. 54, pp Jensen, M., Meckling, W., Theory of the firm: managerial behavior, agency costs and ownership structure. Journal of Financial Economics, vol. 3, pp Jensen, M., Warner, J., The distribution of power among corporate managers, shareholders, and directors. Journal of Financial Economics, vol. 20, pp Kole, S., Managerial ownership and firm performance: incentives or rewards? Advances in Financial Economics, vol. 2, pp McConnell, J., Servaes, H., Additional evidence on equity ownership and corporate value. Journal of Financial Economics, vol. 27, pp Morck, R., Schleifer, A., Vishny, R., Management ownership and market valuation. Journal of Financial Economics, vol. 20, pp Petersen, M.A., Estimating standard errors in finance panel data sets: comparing approaches. Review of Financial Studies, vol. 22, pp Wooldridge, J.M., Introductory Economics. Fourth edition. South-Western. Zhou, X., Understanding the determinants of managerial ownership and the link 24