Master Thesis
University of Amsterdam | Business economics: Finance
‘An Empirical Analysis on Internal Governance’
By Sjors van Kuik, 10274855
Under supervision of Tomislav Ladika
Statement of Originality
This document is written by Sjors van Kuik who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Abstract1
The theory of Internal Governance suggests that firms benefit when managers inhibit CEOs from destroying firm-value. I support this theory by using a unique proxy for the likeliness of its functioning: I show that when older CEOs sell firm equity while younger managers at the same firm do not, firms perform better in terms of stock return. I argue that Internal Governance should be especially strong under these circumstances, because compared to the CEO; managers now hold relatively more equity, while they also face longer time horizons at the firm. This should incentivize them to monitor and discipline the CEO, who has the ability to destroy the value of their shares and their future firm. Besides positive stock returns I find that acquisition expenses decrease for high Internal Governance firms, which could indicate that disciplined CEOs overpay less on acquisitions or cut bad investments. Because I cannot exclude other potential drivers for this pattern in stock return and find no clear channel through which performance increases, my analysis concludes that Internal Governance could have contributed to this outperformance.
Table of Contents
1. Introduction 5
2. Literature Review 8
3. Data and Empirical Model 10
3.1 Data Sources 10 3.2 Key Variables 11 3.3 Descriptive Statistics 12 3.4 Empirical Specification 13 4. Main Results 14 4.1 Results 14 4.2 Interpretation of Results 14 5. Robustness Tests 15
6. Potential Other Explanations 17
7. Shortcomings of the Analysis 18
8. Conclusion 21
References Figure 1 Tables 1-9 Data Appendix
5
1. Introduction
Public firms suffer from conflicts between those who lead them and those who own them (i.e. Jensen, 1997). Those who lead them ─ CEOs in particular ─ may strive to maximize their personal value rather than shareholder value. One form of an agency problem especially related to this paper is empire building: CEOs may use excess cash to overinvest rather than to pay out the shareholder (Jensen, 1997). Motivations to overinvest are to lead a larger company, which could be positively related to salary and arguably to social status. To align the incentives of both parties, top managers often receive equity in the form of shares and stock options as part of their annual salary. Now that their personal value is linked to shareholder value, executives are more likely to behave according to the preferences of the shareholder (i.e. Jensen & Meckling, 1976). However, if officially reported, executives can sell their firm equity and thereby lose these shared incentives. Besides granting equity payment, the board of directors is expected to reduce agency problems by monitoring the actions of management (i.e. Landier et al., 2012).
Acharya et al. (2011) suggest that other than by the board of directors, monitoring also occurs on a firm-level, between executives that see each other on a daily basis. If managers have both the capability and the incentives to monitor, they could potentially discipline a self-interested CEO. They call this mechanism Internal Governance (hereafter: IG). The main focus of this paper is to find support for this theory by showing that firms perform better when IG is more likely.
In expectation, strong IG should lead to a decrease in empire building and other forms of agency problems. In expectation, this should reduce wasteful investment, such as overpaying on acquisitions, but also increase dividend pay-outs and support stock prices. IG is likely to be more present in some firm-years than others, because incentives to monitor also differ per firm and over time. In this paper I estimate the likeliness of IG and test the following main hypotheses:
1. Firms characterised by high IG at time t should have higher stock returns from time t to t+1 than low IG firms.
2. Firms characterised by high IG at time t should increase their pay-out ratio more from time t to t+1 than low IG firms.
3. Firms characterised by high IG at time t should decrease their acquisition expenses ratio more from time t to t+1 than low IG firms.
6
What separates my paper from other research on this topic is that I have access to a unique proxy for IG; one that clearly motivates the monitoring incentives for managers. Ladika (2015), in his paper on equity replenishment, finds that after equity sales, boards only restore 7% of the amount of equity sold. This implies that an executive can enduringly lower its shared incentives by selling off equity. Ladika also finds that boards grant similar equity pay to frequently and infrequently selling executives. This results in executives at the same firm ending up with having very different incentives. The relationship with IG is that when managers subordinate to the CEO (hereafter: managers) do not sell equity, while the CEO does, they have a clear motive to monitor, because the CEO potentially destroys their value. In table 1 I show that there is a wide variation in equity sales, both on firm- and executive-level. In figure 1 I show that infrequent sellers collect more incentives over time than frequent sellers do. Both tables are replicated from Ladika (2015).
When the CEO sells and managers do not, at least two forces are likely to put pressure on stock returns. Firstly, because the CEO is selling part of his equity share, investors might recognize this as a bad signal and start selling too. This mechanism would have a negative effect on the stock price, but is likely to happen in the short-run. Secondly, IG is expected to increase and should positively affect stock returns in the longer-run2. This IG effect could work either directly, assuming that investors are aware
of the increase in monitoring, or indirectly, for example via increased earnings, increased pay-out or reduced wasteful investment. Of course, a combination of both direct and indirect effects is not excluded. It is not sure for what time-period IG will work, but I regard it as a timely process, because important decisions are normally not made on a daily basis. Because my data is yearly, the dependent variable is measured over the year after IG is recognized. In my robustness tests I also test for stocks returns during the three years after IG and find some evidence that the effect is still apparent but dies out.
An important measure I use in this paper is the ratio of equity held by the CEO relative to the equity held by all executives in the sample (CEO Equity Ratio). My main explanatory variable, labelled Internal Governance (IG), is defined as the CEO Equity Ratio multiplied by -1, so that IG increases when the CEO sells equity (ceteris paribus). Using this ratio of within firm holdings, I provide some exogenous variation to the analysis, because equity sales are arguably unrelated to firm performance when this ratio changes; in expectation, all top directors would trade in the same direction, given that their information is equal. To isolate only the portion of IG that is caused solely by equity sales, I run IV regressions using big sales3 from the past as an instrument. In my main analysis I use big sales two
years prior to the measurement of IG (and thus three years prior to the change in performance). In
2 Current research does not allow me to be more specific on the time-period, but based on my analysis IG seems to work at least up to three years.
7
table 3 I show that the instrument is both a valid predictor of IG (Column 1) and seems unrelated to the dependent variables (Columns 2-4).
Besides equity sales, I use the age difference between the CEO and the average manager as a proxy for IG. Explained by Acharya et al. (2011), IG should work best when the CEO faces a shorter horizon at the firm than the managers do. The rationale is that the CEO is then likely to care less about the long-run performance of the firm than the managers. I think that this horizon is best estimated using age variables: an older CEO is likely to retire soon and should therefore face a shorter horizon than the younger managers. For completeness, I also use Li’s (2014) No.2 specification, defined as the best paid non-CEO in the sample, and compute the age difference between the CEO and that particular manager. I divide both age difference measures over quartiles and compare the top quartile to the bottom quartile. As an alternative measure I use binary variables indicating whether the CEO or a certain manager is expected to retire soon (age 62-67).
In my main analysis I find positive future stock returns for high IG firms in the top quartile of the age difference distribution (relatively old CEO). This is consistent with hypothesis 1. An increase of 0.01 (1%) in Internal Governance seems to increase next year’s stock return by 0.87% in absolute terms4. This result is given in table 4. As for my second hypothesis, I find that firms with these same
characteristics seem to decrease their acquisition expenses ratio by 0.12% followed by a 1% increase in Internal Governance. This finding is consistent with hypothesis 2 and is given in Table 5. On the third hypothesis, whether high IG firms also distribute more to the shareholder, I find mixed results: the result is not significant for the older CEO subsample, but is significant for the subsample in which the CEO is expected to retire soon5. In the latter group, a 1% increase in Internal Governance should lead to
a 0.73% increase in pay-outs in the following year. This result is given in Table 6. In general I think that these results are quite substantial but become more unlikely as the change in IG grows larger.
I conduct several robustness tests that show consistency in my results. Firstly, I find that the results are unchanged when I measure the instrument at a different time. Secondly, I find that high IG firms also outperform in the longer run, but that the effect seems to be strongest for the first year. Lastly, I find that IG is probably not a proxy for external governance, because controlling for external governance does not take IG’s predictive power away.
I come up with potential other explanations that could fit the predicted pattern. One, managers increase their productivity when they hold more equity in the firm. To complete the pattern we observe; CEOs should delegate more responsibilities when they become older and young managers should be better able to handle this workload than older managers. Two, we observe a rebound in stock returns from an initial decrease. To complete the pattern, the initial decrease should only
4 All my results are measured in absolute terms. An increase of 0.01 (1%) in IG indicates that the CEO decreases its relative Equity Ratio by 0.01 (1%), so for example from 53% to 52%. An increase in return of 0.87% means for example an increase from 7% to 7.87%.
8
happen when CEOs are older than managers, possibly because investors relate selling rationale to age. Three, investors assume that an older CEO sells because he is planning for retirement and view this transition as favourable. Similar to IG, these explanations rely on assumptions and are therefore not necessarily more credible. The observed effect could be a combination of forces.
One limitation of my analysis is that I cannot control correctly for talent of the CEO, capability of the managers to monitor and managers’ contribution to value. Information on schooling or past performance per executive might partially solve this problem. Another limitation is that I consider only one year of sales, neglecting the history of selling behaviour in firms. Also, I do not differentiate between good and bad acquisitions. This can be solved by investigating post-acquisition returns.
The rest of the paper is structured as follows: Section 2 gives a brief literature review on research in this field. Section 3 describes my data and empirical model. Section 4 shows and interprets my results. Section 5 shows my robustness tests. Section 6 shows potential other explanations. Section 7 discusses the limitations of my study. Lastly, Section 8 concludes. References, figures, tables and information on variables can be found at the end of this paper.
2. Literature Review
In this section I discuss several papers that empirically analyse IG. Before I do, I should explain the common pattern we observe in research on this topic so far. Because we do not witness the monitoring intensity within firms, there is no evidence that directly supports the theory. Research so far has focused on finding indirect evidence: just as in this paper, research tries to find ways to separate the data into groups with supposedly high IG on one side and low IG on the other. When performance is higher for the high IG group, we conclude this could be due to IG. I believe that when research continues to find indirect evidence, it bundles up and the argument will become stronger.
To differentiate between high and low IG, research relies on the conditions given by Acharya et al. (2011). In explaining the mechanism, they describe a scenario that should allow IG to work best: besides the difference in horizon as described in section 1 of this paper, an important condition for IG is that both the CEO and the managers should contribute enough to value creation. The explanation is that when the CEO tunnels6 all assets away from the firm (contributing nothing to value creation), the
manager realizes that his effort is not resulting in a bright future for him, but merely enhances the welfare of the CEO. In this case he is likely to leave the firm or stop exerting effort. When it is the manager that contributes too little to value creation, the CEO will not be disciplined by his potential
6 Tunneling (or ‘self-dealing’) is considered another form of an agency problem. Morck & Yeung (2003) explain it as using pyramid structures to transfer value from one holding to another.
9
departure. A point somewhere in between will lead to an equilibrium in which the manager is both needed and is able to create value for the future. Furthermore, Acharya et al. argue that IG should work better for firms that compete in industries that rely heavily on human-capital.
Landier et al. (2012) are among the first to empirically analyse IG: they create a measure of how connected top managers are to the CEO and relate that to firm performance. Landier et al. define their main explanatory variable, ‘independently minded’ directors, as managers who joined before the current CEO was appointed. The idea is that those directors should impose more discipline on the CEO than the ones that joined later, hired by the current CEO. Their paper finds that having more independently minded directors should lead to higher future performance. Besides accounting measures, Landier et al. also consider shareholder returns attributed to acquisitions and find that they are larger for high IG firms. They also control for external governance measures and find that the result holds. Compared to my paper, this study is more advanced on its acquisition and external governance analysis. However, my main proxy for IG is very different from theirs because I establish incentives from a different perspective. Both our papers fall short concerning the ability of managers to monitor the CEO.
Aggarwal et al. (2013) use a measure based on the concept that IG works best when both the CEO and managers are important for value creation. As measuring value creation is problematic, the authors choose to use compensation instead. The relative value creation is measured by the CEO’s compensation relative to the managers’. When they test for the level of investment, Aggarwal et al. find that starting from a low point of relative contribution; firm performance is upward sloping in relative contribution, but will be downward sloping when higher levels of relation contribution are reached. Consistent with the theory, their research shows that firm performance is better for high IG firms. While measuring IG from a different perspective, our papers have in common that we both test for age difference between the CEO and subordinate managers. Aggarwal et al. find that the pattern is more visible when this age difference is greater.
Li (2014) also works with compensation and defines the No.2 executive as the highest paid non-CEO executive for each firm-year. Li measures the monitoring capability of the No.2 based on what he calls the GAP; a measure for the relative pay between the CEO and the No.2. He explains that a lower GAP is likely to be correlated with a higher monitoring capability for the No.2, because No.2 is most likely paid relatively more for his skill, knowledge and/or influence compared to firms with a higher GAP. Li develops several other measures for IG strength, such as whether the No.2 is also a director of the company board, whether the No.2 has a president title and, consistent with the analysis of Landier et al. (2012), whether the No.2 joined the firm before the current CEO. Li finds higher
10
performance for all of his IG measures7, consistent with the theory. He also mentions that IG is a
substitute for external governance mechanisms rather than an addition because he finds a negative correlation between the two. My paper does not support this result.
Analysing the situation between the CEO and the No.2 clearly tests the situation as sketched by Acharya et al. (2011), but in my opinion it is more meaningful to use information on all top executives; while it is reasonable that the No.28 succeeds the CEO and probably has strongest incentives to
monitor, we do not have enough information to decide on whether or not he dominates other managers in this process. For this reason I add more value to the analysis in which I compare the CEO to the average manager. Since Li suggests that further research should focus on incentives to monitor, I think that my research provides a modest contribution to the literature.
3. Data & Empirical model
In this section I describe my data sources and variables, as well as my empirical specification for the analysis.
3.1 Data Sources
The main dataset I use is the same as used in Ladika (2015). The dataset consists of all top executives that are both present in Compustat Execucomp & Thomson Insiders for the years 1996-2007. Compustat Execucomp gives us information on the total size of annual compensation as well as the fraction paid in firm equity9. The database also contains executive’s firm holdings at the end of the fiscal year,
executive tenure and executive age. Data in this database is available from 1992 onwards. Because data in Thomson Insiders is available from 1996 onwards, the sample starts from that year. Thomson Insiders has full information on firm equity sales and purchases by executives. The merged database thus has information on equity payment and net equity sales. To this main dataset, annualized stock returns are added. The monthly stock returns are collected from the CRSP database. Other control variables, such as earnings, asset size, pay-outs and acquisition expenses are derived from Compustat.
7 So higher performance when (i) the No.2 sits on the company board, (ii) the No.2 has a president title, (iii) the No.2 joined the firm before the current CEO, (iv) the GAP is smaller and (v) the age difference between the CEO and the No.2 is large.
8 Not necessarily the No.2 as described by Li (2014). The No.2 can be defined based on other criteria besides compensation, i.e. tenure.
9 Firms are required to provide information on compensation size and structure for the CEO and four other executives that receive the largest compensation. Ladika (2015) however notices that in 92% of cases, firms report this for five or more executives.
11
The sample ends at 2007 because the financial crisis unfolding in the years after might bias the result10. Firms that do not pay equity in three years straight are dropped from the sample. Before
losing executive specific information, the sample consists of 26,020 executives at 2,664 firms. The full regression dataset consists of 19,700 observations for 2,664 firms, averaging at about 7.4 years of data per observed firm.
3.2 Key Variables
While a detailed list of all used variables can be found in the appendix, I dedicate this sub-section to describing the most important variables for my analysis.
Internal Governance
This is the main explanatory variable used throughout my analysis. CEO Equity Ratio measures the ratio of firm equity held by the CEO relative to all firm equity held by executives in the sample. In my sample, the mean ratio held by the CEO is about 53%. Internal Governance (IG) is defined as this CEO Equity Ratio, multiplied by -1. In table 3, we see that as expected, sales by the CEO increase this ratio, while sales by subordinate managers decrease it. This means that for a firm where the CEO sells and managers do not, IG should increase.
Big sales
To explain the variation in IG in my IV regressions, I use big sales by the CEO and managers as an instrument. Big sales are those sales in which the particular executive sells more than 10% of its total stake in the firm.
Change in Pay-outs
Pay-outs are defined as the sum of total dividends and share repurchases, divided by earnings before interest expenses & taxes (EBIT). This variable is winsorized at the 5-95 level because of severe outliers. Change in Pay-outs is the value this year minus the value last year (absolute yearly change).
Change in Acquisitions
Acquisitions are defined as the expenses on acquisitions, divided by the firm’s total asset value. The change in acquisitions is winsorized at the 5-95 level. Change is again the absolute year-to-year change in the ratio.
10 The unfolding of the financial crisis could bias the result in numerous ways. A likely example is that executives might choose to postpone equity sales to post-crisis periods, when prices have recovered. It would however be interesting to find out whether IG works even better in those times of uncertainty, but with an IG measure related to equity sales this will probably not work.
12 Stock Returns
Besides pay-outs and acquisition expenses, my main dependent variable is the logarithmic transformation of stock return. Stock returns are annualized from monthly data and winsorized at the 5-95 level.
3.3 Descriptive Statistics
Table 2 presents descriptive statistics on my main variables, firm characteristics and executive characteristics.
Panel A shows how my main variables change over time. CEO Equity Ratio decreases on average by 0.2% a year, which is relatively modest, compared to the average level of 53% (Panel B). The median however is slightly positive at 0.5%. We observe this pattern because it is quite common that at least one of the managers sells while the CEO does not sell or sells little. This causes the ratio to increase most of the years but decrease more significant in a year that the CEO sells part of his larger stake. Furthermore, the bottom quartile of the distribution corresponds to a drop of over 4.3% and the top quartile to an increase of over 6.4%, showing that there is a wide-spread variation in sales over firm-years. The three dependent variables are given in their winsorized form. Stock returns average at 7% a year, pay-outs increase by 3.2% and acquisitions are stable. Average levels for these variables are given in panel B.
Panel C firstly shows that CEOs in the sample have been at the firm for about 8 years, while this averages 4.5 years for managers. This should not be confused with future tenure (or horizon), which in expectation is longer for managers, because the CEO is often older and close to retirement. This expectation is confirmed by the age difference of 3.4 years between the CEO and the average of subordinate managers. Interestingly, the sample contains a large variation in this number: one quarter of firm-years has a CEO that is at least 2 years younger than the average manager, while CEOs are more than 9 years older in the top quartile. Next, we notice that the No.2 ─ defined as the best paid manager, similar to Li (2014) ─ is on average 1 year older than the CEO. As mentioned in section 2, I prefer using full information on executives, so I consider the first age difference as more relevant. Yet, this measure can be very helpful to show consistency in results. In this Panel I also show that for an average firm-year 0.77 managers leave the firm and about one quarter of managers does a big sale. Finally, firm-yearly salaries of CEOs and managers are given in millions of dollars, together with the gap between them (again from Li).
Panel D shows that CEOs sell firm equity in half of the years, while they engage in big sales in 19% of years. This panel also shows the retirement variable, which is ─ similar to age ─ also linked to horizon. When CEOs are of age 62 to 6711, managers are likely to expect their retirement soon and, in
11 In the US, the retirement age gradually increases from 65 to 67. When CEOs are older than 67, I do not think that managers expect them to retire soon, because it is not clear at what age they will retire instead.
13
hypothesis, increase their monitor activity. 14.3% of CEOs in the sample are expected to retire. In 16.5% of firm-years at least one manager is expected to retire (age 62-67).
3.4 Empirical Specification
To test my main hypotheses, I should find the effect of equity sales on performance measures. Consider an OLS regression of stock returns on equity sales: I would face severe endogeneity problems, mostly because reversed causality is highly likely: executives are expected to sell when they expect deteriorating performance and not sell when they think that performance will pick up. To provide some exogenous variation, just like Ladika (2015), I examine equity sales relative to others at the same firm. Consider the measure Internal Governance (IG): when the CEO sells and other managers do not, this ratio will increase. Assuming that the CEO’s information is not superior to that of other executives at the firm, this increase is probably not linked to performance. The reason I am assuming this is because if the CEO would sell because he expects lower performance, other directors would most likely act in the same way. Since they do not sell, it is likely that either they have less information or that the CEO sells because of different reasons. The other way around, when the CEO does not sell while the managers do, this ratio will decrease. Because of the same rationale, it could very well be that this is not related to performance. When both the CEO and the managers sell/do not sell, Internal Governance should not be affected so much and therefore not bias the result.
The IG variable in its current specification is not only affected by equity sales, but also by equity payment and by executives leaving the firm. IG is only supposed to change after equity sales, because only in that case the particular executive deliberately chooses to change its incentives. To isolate the variation in IG that is caused by equity sales, I take an approach similar to Bertrand & Mullainathan (2001): in a two-stage procedure, I firstly perform a regression of the level of IG on Big Sales from two years earlier to obtain the predicted value of IG. In the second stage I perform a regression of the dependent variable measured one time-period later on the predicted level of IG. I choose to use big sales as an instrument for IG because it seems to do quite well in terms of validity and exogeneity: table 3 shows a significant correlation between past big sales and the level of IG (column 1), but also shows that these sales are probably uncorrelated to performance measures (column 2-4). The two-stage regression model has the following composition:
Stage 1:
௧ିଵ= +ଵ ௧ିଷ + ଶ ௧ିଷ
+ ଷ ℎ ௧ିଵ +
14 Stage 2:
௧= +ଵ ௧ିଵ + ଶ ℎ ௧ିଵ + (2)
In stage 1, Big Sale CEO is a binary variable that equals 1 when the CEO sells more than 10% of his holdings during that year and 0 otherwise. Big Sale Managers is the fraction of managers that sell more than 10% of their holdings. In both stages, Firm Characteristics are control variables such as the log of total asset value, Tobin’s Q, average executive tenure and whether executives were promoted during that year. The full list of control variables can be found in the tables. Performance in the second stage is either the log of stock returns, the change in pay-outs or the change in acquisition expenses. ߝ denotes the error term. All regressions control for industry and year fixed effects. For industry fixed effects I use the 48 Fama-French industries. Each regression also controls for possible clustering at the firm level.
4. Main Results
In this section I describe my main results from following the procedure as described above. In sub-section 4.1 I present the effect of IG on stock returns, the change in acquisition expenses and the change in pay-outs. In sub-section 4.2 I will give an interpretation of the economic magnitude of the results.
4.1 Results
Table 4 shows whether IG affects stock returns. A positive coefficient in columns (3) and (5) would support my first hypothesis that IG increases stock return. The coefficients of 0.87 and 0.92 respectively support this hypothesis. Columns (2) and (4) show whether equity sales also affect performance when IG conditions are not optimal. This does not seem to be true. Column (1) shows that for the full sample, equity sales alone do not change performance.
Table 5 shows whether IG affects acquisition expenses. A negative coefficient in columns (3) and (5) would support my second hypothesis that acquisition expenses decrease when IG is stronger. The coefficients of -0.12 and -0.15 support this hypothesis. Columns (2) and (4) point out that equity sales, in weak IG conditions, do not seem to affect acquisitions. For the full sample (1), we observe no effect either.
15
Table 6 shows whether IG affects the pay-out ratio. A positive coefficient in columns (3) and (5) would support the third and final hypothesis that pay-outs are higher when IG is strong. The insignificant coefficient in (3) rejects this hypothesis. Because the coefficient in (5) is both sizable (0.73) and significant, results seem mixed for pay-outs. When IG conditions are weak, equity sales do not seem to lead to changed pay-outs.
4.2 Interpretation of Results
Stock returns seem to increase by about 0.87% in absolute terms for each percentage point absolute increase in IG12. Because IG is defined as -1 times the CEO Equity Ratio, this finding corresponds to a
0.87% increase in returns after a reduction of 1% in the CEO Equity Ratio13. In my opinion this
estimate is quite substantial, especially for larger increases in IG. I believe that if IG is able to increase performance, it can only do this up to a certain level. At that certain level the CEO’s actions are already limited and the marginal benefit of monitoring should be small. Because currently we do not have an estimate for this level, I do not attach too much value to the coefficient itself, but more to the sign in front of it.
After a 1% absolute increase in IG, the acquisition ratio decreases by 0.12% in absolute terms. To put this into perspective, the average acquisition/assets ratio is 3% (table 2). In relation to the average level, this estimate is quite substantial. For larger increases in IG this estimate makes no sense: we assume that non-profitable acquisitions are cut first, which is a good thing. But at some point only profitable investments should remain and cutting those will not increase performance.
The pay-out ratio ─ considering only the signiicant estimate in column (5) ─ should increase by 0.73% after IG increases by 1% in absolute terms. Compared to the average pay-out ratio of 18% this is again a generous estimate. Because the coefficient in (3) is insignificant, coefficients on pay-out should be handled with extra care.
In the following section I show my robustness tests, in which I aim to find out how sensitive the stock return result is to changes in specification.
5. Robustness Tests
In this section I discuss my three main robustness tests. I discuss the implications of using a slightly different instrument, I test for performance in the longer run and I include a measure of external governance to my analysis. Taken together, these robustness tests add stability to my main result by showing that high IG firms outperform low IG firms under alternative specification.
12 Because the relationship is Log-Level for the regression of stock returns on IG, a 0.01 increase in IG corresponds to an approximate β1 % increase in stock returns.
16 Instrument: Sales from Year t-2
In this robustness test I show that my main result is unchanged, and stronger, when I use a slightly different instrument. For this analysis I also use big sales to explain the variance in IG, but I will now switch to sales made one year later than those in the main analysis. In table 3 we should now look at the rows Big CEO Sale (t-2) and Big non-CEO Sale (t-2) to observe instrument validity and exogeneity. Because these sales are made just one year before we measure the level of IG, the instrument has higher explanatory power than the one used in the main analysis (column 1). The concern is that exogenous variation might decrease, because these sales are now measured only two years prior to the dependent variables. From columns 2-4 however, it seems that exogenous variation may not be lower, because none of the dependent variables seem related to sales. I find that, with a coefficient of 1.64, the effect of IG on return is now even larger (Table 7). This could mean that (a) the IG effect is actually stronger than estimated in the main analysis and is now estimated by a better instrument, or (b) the IG effect is overestimated here because the instrument provides less exogenous variation. Despite the analysis from table 3, I am more concerned about endogeneity problems using this instrument and have therefore decided not to use it in the main analysis. Nevertheless, since the coefficient is both positive and significant in this test, I show that the main result is unchanged.
Average Future Returns
In this robustness test I show that my main result is unchanged, but weaker, when I test for average stock returns during the three years from t to t+2. Compared to the coefficient of 0.87 when testing for year t alone, the coefficient is 0.39 in this analysis (Table 8). This implies that after a 1% increase in IG, given the high age difference, firms outperform by 0.39% in each of the three years after. While it is unclear for what time IG works14, it is likely that the effect is strongest in the first year and then dies
out slowly, given that the CEO does not engage in other big sales. This is exactly what we observe: because the total effect for the three years is approximately 3 times 0.39, or 1.17%, the first year seems to represent around three quarters of this number.
External Governance
In this robustness test I show that my main result is unchanged when I include a measure of external governance to the analysis. When I include the Gompers-Ishii-Metrick (GIM) index (Gompers et al., 2003) as a control variable to the regressions, my result is not affected (Table 9). The reason that I include such measure is that Landier et al. (2012) consider that IG may be a proxy for external governance. If that is true, the GIM measure should take the explanatory power away from the IG measure, making it insignificant. However, consistent with Landier et al., this does not happen in my
14 The duration of IG should differ greatly per firm and situation because it is likely to correlate also with psychological factors and personalities. With this remark I mean that we are also unaware of a general guideline that averages the effect.
17
analysis, indicating that IG is not a proxy for external governance but rather a mechanism on its own. I choose to use the GIM index because it is frequently used in Corporate Governance literature and includes a large amount of external governance indicators, yet I am aware that it does not account for all external governance forces.
As a second test I have split up the sample in terms of high versus low GIM to test whether the IG effect is stronger when external governance is weaker (high GIM). This does not seem to be true, indicating that, as contrary to Li (2014), IG seems not to be the less expensive substitute for external governance.
6. Potential Other Explanations
So far, we observe that after the CEO sells equity and managers do not, performance increases when the CEO is older than the managers. This pattern clearly fits in the IG theory because it gives managers great incentive to monitor: the CEO deliberately chooses to untie his incentives from firm performance and should have little career concerns, while at the same time managers should be very affected by firm performance, both in terms of career opportunity and in terms of equity stake. To determine the likeliness that IG causes this outperformance, we should consider alternative explanations that fit this pattern. This section describes these potential other explanations.
Managers Increase Productivity
One potential explanation is that managers increase their productivity. Consider the situation in which the CEO does not sell: when managers sell, the IG measure is lower than when they do not sell. If managers do not sell, their welfare is more related to firm performance. This should cause them to put in more effort, leading to increased performance. Because not selling leads to a higher IG measure and to putting in more effort, high IG could positively correlate with the amount of effort managers put in. At this point the selling part is explained, but the age difference is not. It is likely that managers are more able and more willing to work hard when they are younger. However, I have also tested on a sub-sample of young managers, irrespective of the CEO’s age, and found no evidence of increased returns here. What would explain this pattern is that older CEOs tend to pass on more of their responsibilities to the younger, more productive managers.
Another closely related explanation is that managers expect the CEO to retire soon and therefore put in more effort to potentially get promoted. In this case the non-selling part could be explained by managers showing commitment to the company and trust in future performance.
To test whether this increased productivity explanation is more likely than IG, I have run similar regressions as in my main analysis, but now using only sales by managers as an instrument. I
18
do not find the same pattern in these regressions, indicating that CEO sales are important as well. Because of this result, I believe that IG or a combination of IG and higher productivity is more likely than this explanation on its own. It would be helpful if there was data on the amount of effort that managers put in, like the amount of work hours per week or some kind of evaluation by the CEO for example.
We Observe a Rebound
Another possible explanation is that we observe a rebound in returns. In the introduction I have explained that CEO sales might be interpreted as a bad sign by investors, leading to a decrease in return. This decrease in return is likely to happen in the very short-run, most likely when the news reaches the markets. If investors realise that profitability does not decrease in the period after the sales, indicating that the immediate response was too dramatic, they should be more willing to buy. This buying pressure will lead to higher returns and may just be the returns that we observe. Again, the age difference between the CEO and the managers remains unexplained. One reason could be that this initial decrease only occurs when the CEO is older. This assumes that investors believe that the rationale of CEO sales is linked to their age: when CEOs are younger, sales might indicate that they need liquidity to support their living, like moving to a larger house, rather than because they fear lower performance. Further testing would be necessary to validate this point.
Investors Expect Retirement of CEO
Connected to the previous point about the rationale behind sales, investors might assume that given these conditions, it is likely that the relatively old CEO prepares for retirement by exchanging equity for cash. This explanation requires that investors on average view CEO successions as favourable, perhaps because the new CEO is expected to be more productive or to implement important changes. In my analysis I do control for CEOs leaving at year t, but not for the fact that investors believe that the CEO will retire soon.
7. Shortcomings of the Analysis
Omitted variables
I try to limit the omitted-variable bias by using appropriate control variables and by controlling for industry- and time specific factors. I also correct for possible clusters at the firm level. Still, there are certain factors that I cannot account for correctly:
19
• Talent of the CEO: talented CEOs, aware of their skill, are probably less likely to sell equity, as they are able to increase its value over time. Having a talented CEO should also lead to higher firm performance. Because this omitted variable is negatively related to IG and positively to the dependent variable, the coefficient on IG is likely to be underestimating15 the true population
parameter. Although I believe that with the available data, salary might be the best predictor of talent, I do not regard it as a great estimate. Researchers like Bebchuk & Fried (2003) also argue that powerful CEOs can influence the pay process to be better off. This result would reduce the correlation between salary and talent. When I do correct for the CEO’s salary, my estimate becomes slightly less significant and smaller. If we find a better measure for talent, which is for example based on profitable decisions made in the past, we can reduce the omitted-variable bias in this relation. Because this omitted variable probably imposes a negative bias to the coefficient on IG, it would most likely not change my main result that high IG firms outperform.
• Monitoring Capability: in my analysis I only account for managers having an incentive to monitor the CEO, not for them being capable to do so. If, for example, the CEO is severely overpaying on an acquisition but the managers are not sure about the fair value, they will not try to cancel the deal. In that case the bottleneck is not the willingness to monitor, but the ability to monitor. I believe that being capable to monitor comes in two forms: one, having sufficient (firm-specific) knowledge, two, having a certain power or authority to influence the CEO. Li (2014) accounts for capability by calculating the salary GAP between the CEO and the No.2 manager, but as in the previous point, I am not convinced that salary is a great indicator of either talent or knowledge. Consistent with Bebchuk & Fried (2003), salary could however be related to power or authority. Other measures of monitoring capability could perhaps be related to education.
• Contribution to Value: another condition for IG is that the CEO should fear that the manager withdraws his contribution, i.e. leaves the firm. According to Acharya et al. (2011), The CEO will not be disciplined by a manager that contributes relatively little. Again, because contribution to value is kind of similar to capability to monitor and talent, I believe that other variables could be better predictors than salary. I think that a measure related to past performance would do better.
15 Because the model does not include ‘Talented CEO’, the effect will be disguised in IG. If IG goes up, Talented CEO goes down, increasing return by the positive effect of IG minus the negative effect of Talented CEO, resulting in a lower estimate.
20 One Year of Sales
My analysis builds on equity sales in just one year and neglects the history of equity sales at the firm. Imagine firm A where the CEO sells every year versus firm B where the CEO sells for the first time. There are two important differences between the firms: (a) CEO sales could be more meaningful for firm B, because they are an exception and (b) the CEO Equity Ratio should be lower for firm A, which should result in a higher IG measure. The problem is that I neglect both of these differences in my analysis, because IG is predicted by sales from only one year. Controlling for these problems is hard under the current specification, as sales from those earlier years are probably not very well predictors of the IG measure16. Moreover, many instruments would have to be used, which could possibly bias the
result.
I have tried to address this issue by running OLS regressions on groups of firms based on past selling characteristics17. I fail to find evidence that high IG groups outperform low IG groups in this
analysis. In order to further support the IG theory, future research should focus on using the longer-term selling characteristics of firms.
No Clear Channel
In my main result I show that acquisition expenses decrease for high IG firms. I assume that this result corresponds to positive performance, up to a certain level, because bad acquisitions should be cut first. If, however, the firm invests wisely and does not overpay on acquisitions at all, cutting investment would not be a good thing as it would inhibit growth. Without this context, my result on acquisitions is rather dubious. Papers like Landier et al. (2012) solve this issue by looking at the returns that follow from acquisitions. I think that a similar test would be very relevant for my analysis18. Because the
result on acquisitions is quite dubious, my analysis does not come up with a clear channel through which IG should increase performance.
16 See table 3 for the decline in validity from t-1 to t-3. This pattern is likely to continue from t-3 onwards. 17 I have used various specifications for the groups, but they all came down to the following: Group A: CEO/Managers both sell. Group B: CEO sell/Managers do not (High IG). Group C: Managers sell/CEO does not (Low IG). Group D: CEO/Managers both do not sell. My specifications differed in time as well as in amounts sold. 18 I did not carry out this test because I chose to spend my time focusing on returns.
21 8. Conclusion
During my research I find that firm performance increases after relatively old CEOs sell part of their firm equity. This situation should coincide with high levels of Internal Governance because it provides managers with an incentive to monitor the CEO, as they have reason to believe that he worries less about future firm performance. I find that followed by an increase of 1% in IG, stock returns increase by 0.87% in the following year. I do not prove that this increase in performance is caused (strictly) by Internal Governance, because firstly the mechanism remains an assumption and secondly, other explanations fit the observed pattern. Compared to my alternative explanations the theory does seem reasonable, because each seems to rely quite heavily on assumptions. I also find that high Internal Governance firms decrease their acquisitions expenses compared to lower Internal Governance firms. This could imply that CEOs are more disciplined and take on less value-destroying investments. Because I do not prove that these acquisitions strictly have negative effects on performance, this analysis does not show the channel through which IG is expected to increase performance. Future research should try to find measures that estimate the CEO’s talent, the managers’ capability to monitor and the managers’ contribution to value. If a similar approach to the one in this paper is used, I recommend that one uses the equity dispersion within firms over a longer time horizon than one year. Altogether I think that this research adds to the argument of IG, but that future research on this topic is highly relevant as a lot remains uncertain.
References
Acharya, V. V., Myers, S. C., & Rajan, R. G. (2011). The internal governance of firms. The Journal of Finance, 66(3), 689-720.
Aggarwal, R. K., Fu, H., & Pan, Y. (2013, December). An empirical investigation of internal governance. In AFA 2011 Denver Meetings Paper.
Bebchuk, L. A., & Fried, J. M. (2003). Executive compensation as an agency problem (No. w9813). National Bureau of Economic Research.
Bertrand, M., & Mullainathan, S. (2001). Are CEOs rewarded for luck? The ones without principals are. Quarterly Journal of Economics, 901-932.
Gompers, P. A., Ishii, J. L., & Metrick, A. (2003). Corporate governance and equity prices. Quarterly Journal of Economics 118, 107-155
Jensen, M. C. (1997). Eclipse of the public corporation. Harvard Business Review (Sept.-Oct. 1989), revised.
Jensen, M.C. and Meckling, W.H. (1976), Theory of the firm: managerial behavior, agency costs and ownership structure, Journal of Financial Economics, Vol. 3 No. 4, pp. 305-60. Ladika, T. (2015). Do Firms Replenish Executives’ Incentives After Equity Sales? University of
Amsterdam Working Paper.
Landier, A., Sauvagnat, J., Sraer, D., & Thesmar, D. (2012). Bottom-up corporate governance. Review of Finance, rfs020.
Li, Z. F. (2014). Mutual monitoring and corporate governance. Journal of Banking & Finance, 45, 255-269.
Morck, R., & Yeung, B. (2003). Agency problems in large family business groups. Entrepreneurship Theory and Practice, 27(4), 367-382.
Figure 1: Effect of Equity Sales on Total Holdings over Time
This figure shows the effect of equity sales on total equity holdings for executives with different selling patterns. A frequent seller is defined as an executive that sells on average more than 100% of his equity payment and an infrequent seller is one that sells less than 20% of his equity payment. To find this average sales measure, the ratio is first determined on a yearly basis: what fraction of payment does the executive sell. These are then averaged over the executive’s tenure. The blue (bottom) line shows how incentives change over tenure for a frequent seller. The red line shows how these average holdings change for an infrequent seller. The sample consists of all top executives in the intersection of Compustat Execucomp & Thomson Insiders for the years 1996-2007. Firms that do not grant equity pay in three years straight are dropped from the sample. Executives that are not present in the sample for at least five years are dropped.
1 1 .0 5 1 .1 1 .1 5 1 .2 L o g T o ta l In ce n ti v e s 1 2 3 4 5 6 Years of Tenure
Table 1: Descriptive Statistics on Equity sales
This table provides descriptive statistics on equity sales. Panel A shows the fraction of executives that sell for a given firm-year. The table makes a separation between ‘All Equity Sales’ and ‘Big Equity Sales’. Big Equity Sales are sales that represent more than 10% of an executives’ stake in the firm, whereas All Equity Sales is any sale value that is different from zero. Panel B considers the executive level rather than the firm level and shows the fraction of years that executives sell during their tenure. The sample consists of all top executives in the intersection of Compustat Execucomp & Thomson Insiders for the years 1996-2007. Firms that do not grant equity pay in three years straight are dropped from the sample. Executives that are not present in the sample for at least five years are dropped.
Panel A: Firm Level
All equity sales Big equity sales
Fraction of selling executives: No. of firm-years % No. of firm-years %
None 5.070 25,7 8.963 45,5
Fewer than 50% 5.344 27,1 6.482 32,9
At least 50% 6.402 32,5 3.496 17,7
All 2.884 14,6 759 3,9
Panel B: Executive Level
All equity sales Big equity sales
Fraction of selling years: No. of execs % No. of execs %
None 810 10,9 2.032 27,4
Fewer than 50% 2.696 36,3 3.942 53,1
At least 50% 3.397 45,8 1.403 18,9
Table 2: Descriptive Statistics for Firm-Years in the Regression Dataset
This table provides descriptive statistics for the dataset used in the regression analysis. CEO Equity Ratio is the fraction of equity held by the CEO relative to all executives in the sample. Big sales are sales representing more than 10% of an executives’ stake in the firm. Log Stock Return is the natural logarithm of 1 plus the fractional, annualized stock return, winsorized at the 5-95 level. Change in Pay-outs is the absolute change in total pay-outs from t-1 to t, defined as the fraction of dividends plus share repurchases over earnings before interest expenses & taxes (ebit), winsorized at the 5-95 level. Change in Acquisitions is the absolute change in acquisition expenses over total assets from t-1 to t, winsorized at the 5-5-95 level. The construction of the variables in Panel B is identical to those of Panel A, but considers the average ratio instead of the average yearly change in this ratio. The sample consists of all top executives in the intersection of Compustat Execucomp & Thomson Insiders for the years 1996-2007. Firms that do not grant equity pay in three years straight are dropped from the sample. Information on stock returns is derived from CRSP. Information on equity sales is derived from Thomson Insiders, information on executive tenure, age and salary is derived from Compustat Execucomp and information on earnings and asset value is derived from Compustat. An executive that does not return in the sample next year is marked as a leaving executive, given that this firm-year is available in my sample. A retirement CEO/manager is one that is of age 62-67. No.2 is the highest paid non-CEO in the sample. 25% and 75% respectively give the 25th and 75th percentile of the distribution. All variables besides Log Stock Return, Change in Pay-Outs and Change in Acquisitions are given non-winsorized. ‘Yes’ for the binary variables means that the variable equals 1, ‘no’ means it equals zero.
Panel A: Main Variables
N Mean Standard deviation 25% Median 75%
Change in CEO Equity Ratio 14,947 -0,002 0.198 -0.043 0.005 0.064
Log Stock Returns 17,549 0.070 0.470 -0.151 0.105 0.325
Change in Pay-outs 17,469 0.032 0.365 -0.049 0 0.101
Change in Acquisitions 15,902 0.000 0.056 -0.007 0 0.006
Panel B: Firm Characteristics
N Mean Standard deviation 25% Median 75%
CEO Equity Ratio 17,979 0.53 0.24 0.38 0.53 0.70
Pay-out Ratio 18,092 0.18 11.15 0 0.14 0.43
Acquisition Ratio 16,800 0.03 0.07 0 0 0.03
Panel C: Director Characteristics
N Mean Standard deviation 25% Median 75%
Tenure CEO 12,651 8.26 7.38 3 6 11
Tenure non-CEO 19,699 4.62 2.49 2.83 4.2 5.83
Age CEO 17,209 55.35 7.46 50 55 60
Age non-CEO 12,276 52.42 6.94 48 52 56
Age CEO - average Age non-CEO 11,357 3.37 9.80 -2 3.5 9
Age CEO - No.2 Age 11,350 -0.90 11.66 -7 0 5
Number of non-CEOs leaving 17,036 0.77 0.93 0 1 1
Fraction of non-CEOs selling big 16,264 0.24 0.32 0 0 0.5
Salary CEO (M$) 17,979 660 363 415 600 850
Salary non-CEOs (M$) 18,433 343 169 231 304 414
Salary Gap (M$) 17,861 316 257 165 290 434
Panel D: Binary Variables
N Yes % No %
CEO sells 19,700 9,929 50.4% 9,771 49.6%
CEO sells big 15,539 2,890 18.6% 12,649 81.4%
CEO is expected to Retire 17,209 2,455 14.3% 14,754 85.7%
Table 3: Effect of Past Equity Sales on Internal Governance & Current Performance
This table investigates whether big sales by CEOs and non-CEOs at year t-3 (instruments in later IV regressions) are (1) related to Internal Governance at t-1 (instrument validity) and (2) unrelated to the dependent variables: Log Stock Return, Change in Pay-outs & Change in Acquisitions at year t (instrument exogeneity). Big sales are sales representing more than 10% of an executives’ stake in the firm. Log Stock Return is the natural logarithm of 1 plus the fractional, annualized stock return, winsorized at the 5-95 level. Change in Pay-outs is the absolute change in total pay-outs from t-1 to t, defined as the fraction of dividends plus share repurchases over earnings before interest expenses & taxes (EBIT). The fraction is winsorized at the 5-95 level. Change in Acquisitions is the absolute change in acquisition expenses over total assets from t-1 to t, winsorized at the 5-95 level. The sample consists of all top executives in the intersection of Compustat Execucomp & Thomson Insiders for the years 1996-2007. Information on stock returns and volatility is derived from CRSP. Stock Volatility measures the standard deviation of the stock over the past 48 months. Other control variables, such as net income, asset value and information to calculate Tobin's Q are obtained from Compustat. Tobin's Q is the market value of equity plus book value of liabilities, divided by the book value of equity plus book value of liabilities. Average Executive Tenure measures for what time the average executive in the sample has been part of the firms' top management. Change in Net Income/Assets is the absolute change in Net Income divided by Total Assets from t-2 to t-1. Resigned as CEO equals 1 if the CEO leaves his/her position and 0 otherwise. Promoted to CEO/Chair/C-Suite equals 1 if a certain executive is promoted to the particular position and 0 otherwise. Each column (1 to 4) is an OLS regression; (1) tests for instrument validity & (2-4) test for instrument exogeneity on the different dependent variables. Each regression controls for clusters at the firm level. Industry Fixed Effects are based on the 48 Fama-French industries. t-statistics are in parentheses and significance is measured at the 10%, 5% and 1% level and indicated with *, **, *** respectively.
Instrument Validity Instrument Exogeneity
Dependent Variable: Internal Governance (t-1) Log Stock Return (t) Change in Pay-outs (t) Change in Acquisitions (t)
(1) (2) (3) (4)
Big CEO sale (t-1) 0.05*** 0.00 0.02* 0.00
(7.97) (0.10) (1.89) (0.72)
Big CEO sale (t-2) 0.05*** -0.01 -0.01 -0.00
(8.07) (-0.71) (-0.98) (-0.24)
Big CEO sale (t-3) 0.03*** 0.02 -0.00 -0.00
(5.16) (1.24) (-0.25) (-0.84)
Big non-CEO sale (t-1) -0.06*** 0.00 0.01 0.00
(-6.21) (0.27) (0.87) (1.08)
Big non-CEO sale (t-2) -0.05*** -0.01 0.00 0.00
(-6.31) (-0.58) (0.13) (0.39)
Big non-CEO sale (t-3) -0.03*** 0.02* 0.01 0.00
(-4.24) (1.68) (0.53) (0.47)
Log Assets (t-1) 0.01** -0.01*** 0.00* -0.00***
(2.54) (-4.60) (1.87) (-6.31)
Tobin's Q (t-1) 0.00 -0.01** 0.00 0.00***
(0.65) (-2.52) (0.78) (3.04)
Log Stock Return (t-1) 0.01 -0.09*** 0.04*** 0.00
(1.54) (-5.24) (3.55) (1.00)
Log Stock Return (t-2) 0.01 -0.03** 0.02 -0.00
(0.89) (-2.15) (1.52) (-0.13)
Stock Volatility -0.22*** -1.07*** 0.04 -0.02**
(-2.66) (-8.95) (0.68) (-2.30)
Change Net Income/Assets (t-1) 0.01 0.10 -0.02 0.01*
(0.59) (1.28) (-0.90) (1.77)
Average Executive Tenure (t-1) -0.02*** 0.00 0.00 -0.00
(-8.63) (1.55) (0.61) (-1.31) Resigned as CEO (t-1) 0.27*** 0.01 0.02 -0.00 (14.12) (0.36) (0.66) (-0.41) Promoted to CEO (t-1) 0.06*** -0.01 -0.00 0.00 (4.57) (-0.59) (-0.00) (1.10) Promoted to Chair (t-1) 0.01 -0.03 -0.04 0.00 (0.47) (-0.98) (-1.53) (0.51) Promoted to C-Suite (t-1) -0.00 0.04*** 0.00 -0.00* (-0.02) (2.88) (0.40) (-1.70) Constant -0.55*** 0.25** -0.04 0.03** (-5.43) (2.02) (-1.29) (2.00)
Year Fixed Effects Yes Yes Yes Yes
Industry Fixed Effects Yes Yes Yes Yes
N 8266 8378 7482 6842
Table 4: IV Regressions on Stock Return
This table measures the effect of IG on stock returns. In these IV regressions, the variation in Internal Governance at t-1 is predicted by big sales from CEOs and non-CEOs at time t-3. The dependent variable, Log Stock Return at the year t, is the natural logarithm of 1 plus the fractional, annualized stock return. Big sales are sales representing more than 10% of an executives’ stake in the firm. The sample consists of all top executives in the intersection of Compustat Execucomp & Thomson Insiders for the years 1996-2007. Firms that do not grant equity pay in three years straight are dropped from the sample. Information on stock returns and volatility is derived from CRSP. Stock Volatility measures the standard deviation of the stock over the past 48 months. Other control variables, such as net income, asset value and information on Tobin's Q are obtained from Compustat. Tobin's Q is the market value of equity plus book value of liabilities, divided by the book value of equity plus book value of liabilities. Average Executive Tenure measures how long the average executive in the sample has been part of the firms' top management. Change in Net Income/Assets is the yearly change in Net Income divided by Total Assets. Resigned as CEO equals 1 if the CEO leaves his/her position and 0 otherwise. Promoted to CEO/Chair/C-Suite equals 1 if a certain executive is promoted to the particular position and 0 otherwise. Each column (1 to 5) is an IV regression; (1) tests the full sample, (2) tests firms in which the age difference between the CEO and the average manager is small (bottom quartile), (3) tests firms in which the same age difference is large (top quartile), (4) and (5) are similar to (2) and (3) but use the age difference between the CEO and the highest paid non-CEO. Each regression controls for clusters at the firm level. Industry Fixed Effects are based on the 48 Fama-French industries. t-statistics are in parentheses and significance is measured at the 10%, 5% and 1% level and indicated with *, **, *** respectively.
Dependent Variable: Log Stock Return (t)
(1) (2) (3) (4) (5)
Full Sample Young CEO Relative to Managers
Older CEO Relative to Managers
Young CEO relative to No.2
Older CEO relative to No.2 Internal Governance (t-1) 0.04 -1.46 0.87** -0.29 0.92* (0.15) (-0.64) (2.02) (-0.43) (1.77) Log Assets (t-1) -0.01*** -0.02 -0.00 -0.01 -0.01 (-3.65) (-0.94) (-0.10) (-1.32) (-0.92) Tobin's Q (t-1) -0.01** -0.02 0.02* -0.02** 0.02* (-2.36) (-0.94) (1.80) (-2.07) (1.93)
Log Stock Return (t-1) -0.09*** -0.07 -0.18*** -0.05 -0.20***
(-5.37) (-0.92) (-3.73) (-1.23) (-4.80)
Log Stock Return (t-2) -0.04** -0.04 -0.08** -0.01 -0.10***
(-2.40) (-0.55) (-2.31) (-0.39) (-2.98)
Stock Volatility -1.02*** -1.30* -0.61* -1.21*** -0.60*
(-7.86) (-1.73) (-1.73) (-2.80) (-1.95)
Change Net Income/Assets
(t-1) 0.10 0.39*** 0.12 0.36*** 0.17*
(1.26) (3.72) (1.29) (2.90) (1.90)
Average Executive Tenure (t-1) 0.00 -0.02 0.01 -0.00 0.01
(0.83) (-0.54) (1.37) (-0.43) (1.03) Resigned as CEO (t-1) 0.01 0.27 -0.50** 0.04 -0.41* (0.12) (0.42) (-2.37) (0.18) (-1.82) Promoted to CEO (t-1) -0.02 -0.01 -0.04 0.03 -0.13* (-0.75) (-0.15) (-0.53) (0.70) (-1.77) Promoted to Chair (t-1) -0.03 0.02 -0.00 -0.03 0.02 (-0.87) (0.20) (-0.05) (-0.41) (0.36) Promoted to C-Suite (t-1) 0.04*** 0.03 0.04 0.02 0.04 (2.82) (0.70) (1.26) (0.62) (1.18) Constant 0.06 -0.12 0.31 0.04 0.46 (0.30) (-0.44) (0.65) (0.22) (0.75)
Year Fixed Effects Yes Yes Yes Yes Yes
Industry Fixed Effects Yes Yes Yes Yes Yes
N 8232 1306 1300 1335 1238
Table 5: IV Regressions on Change in Acquisitions
This table measures the effect of IG on the change in Acquisition Expenses. In these IV regressions, the variation in Internal Governance at t-1 is predicted by big sales from CEOs and non-CEOs at time t-3. The dependent variable, Change in Acquisitions, measures the absolute change in Acquisition Expenses/Total Assets from t-1 to t. It is winsorized at the 5-95 level. Big sales are sales representing more than 10% of an executives’ stake in the firm. The sample consists of all top executives in the intersection of Compustat Execucomp & Thomson Insiders for the years 1996-2007. Firms that do not grant equity pay in three years straight are dropped from the sample. Information on stock returns and volatility is derived from CRSP. Log Stock Return is the natural logarithm of 1 plus the fractional, annualized stock return, winsorized at the 5-95 level. Stock Volatility measures the standard deviation of the stock over the past 48 months. Other control variables, such as net income, asset value and information on Tobin's Q are obtained from Compustat. Tobin's Q is the market value of equity plus book value of liabilities, divided by the book value of equity plus book value of liabilities. Average Executive Tenure measures how long the average executive in the sample has been part of the firms' top management. Change in Net Income/Assets is the absolute change in Net Income divided by Total Assets from t-2 to t-1. Resigned as CEO equals 1 if the CEO leaves his/her position and 0 otherwise. Promoted to CEO/Chair/C-Suite equals 1 if a certain executive is promoted to the particular position and 0 otherwise. Each column (1 to 5) is an IV regression; (1) tests the full sample, (2) tests firms in which the age difference between the CEO and the average manager is small (bottom quartile), (3) tests firms in which the same age difference is large (top quartile), (4) considers firms where the CEO is expected to retire soon (age 62-67) while none of the managers are expected to retire, (5) considers firms where one particular manager is expected to retire soon (62-67), while the CEO is not expected to retire. Each regression controls for clusters at the firm level. Industry Fixed Effects are based on the 48 Fama-French industries. t-statistics are in parentheses and significance is measured at the 10%, 5% and 1% level and indicated with *, **, *** respectively.
Dependent Variable: Change in Acquisitions (t)
(1) (2) (3) (4) (5)
Full Sample Young CEO Relative to Managers
Older CEO Relative to Managers CEO is Expected to Retire Soon Manager is Expected to Retire Soon Internal Governance (t-1) -0.03 0.31 -0.12** 0.15 -0.15** (-1.03) (0.74) (-1.97) (1.38) (-2.23) Log Assets (t-1) -0.00*** 0.00 -0.00** -0.00 -0.00 (-4.93) (0.10) (-2.50) (-0.14) (-1.30) Tobin's Q (t-1) 0.00*** 0.00 0.00 -0.00 0.00* (3.23) (0.94) (1.49) (-0.25) (1.96)
Log Stock Return (t-1) 0.00 0.00 0.01 0.01* 0.01
(1.34) (0.33) (1.45) (1.91) (1.00)
Log Stock Return (t-2) 0.00 -0.01 0.00 -0.01 -0.01
(0.24) (-0.94) (0.22) (-0.91) (-1.06)
Stock Volatility -0.03** 0.06 -0.08 0.01 -0.15**
(-2.36) (0.41) (-1.62) (0.10) (-2.38)
Change Net Income/Assets (t-1) 0.01* 0.03 0.00 0.03 -0.00
(1.77) (1.19) (0.03) (0.76) (-0.01)
Average Executive Tenure (t-1) -0.00 0.01 -0.00* 0.00 -0.00
(-1.46) (0.86) (-1.83) (1.00) (-1.21) Resigned as CEO (t-1) 0.01 -0.09 0.05* -0.04 0.04 (0.69) (-0.79) (1.65) (-1.30) (0.79) Promoted to CEO (t-1) 0.01 0.03 0.00 0.01 0.03* (1.28) (1.13) (0.08) (0.85) (1.67) Promoted to Chair (t-1) 0.00 0.00 -0.02 0.00 -0.03 (0.61) (0.13) (-0.79) (0.01) (-1.31) Promoted to C-Suite (t-1) -0.00* -0.01 0.00 0.01 0.00 (-1.69) (-0.97) (0.60) (0.82) (0.33) Constant 0.01 -0.00 -0.04 0.07* -0.02 (0.42) (-0.10) (-0.94) (1.78) (-0.64)
Year Fixed Effects Yes Yes Yes Yes Yes
Industry Fixed Effects Yes Yes Yes Yes Yes
N 6718 1057 1088 555 581
