Employee capitalism or corporate socialism? Broad-based employee stock ownership

Transcription

1 Employee capitalism or corporate socialism? Broad-based employee stock ownership E. Han Kim and Paige Ouimet 1 Abstract Adopting employee share ownership (ESO) plans leads to an increase in firm value only when the plan is small, i.e., less than 5% of outstanding shares. When the plan is larger, there is no value gain. This inverse U-shaped relation between shareholder value and ESO size is robust to firm fixed effects and controls for possible selection biases. Large ESOP adoptions are followed by substantial increases in employee compensation, whereas small ESOPs show no such increases. Compensation increases are smaller when firms have high leverage, which also counteracts the value negating impact of large ESOPs. We conclude that the adoption of an ESOP is generally followed by performance gains, most of which accrue to employees (shareholders) when employees have substantial (small) control rights. November 8, 2007 JEL classification: G32, M52, J54, J33 Keywords: ESOPs, 401-K Plans, Employee Incentives, Worker Compensation 1 Both are at Ross School of Business, University of Michigan, Ann Arbor, Michigan 48109, with addresses of ehkim@umich.edu and pshelby@bus.umich.edu, respectively. We are grateful for helpful comments/suggestions by Sreedhar Bharath, Amy Dittmar, Francine Lafontaine, Margaret Levenstein, Randall Morck, Clemens Sialm, Jagadeesh Sivadasan, and seminar participants at INSEAD, University of Michigan, and University of Oxford. An earlier version was presented at International Conference on Human Resource Management in Banking Industry sponsored by Korea Institute of Finance. We acknowledge financial support from Mitsui Life Financial Research Center. 1

2 Broad-based employee share ownership (ESO) is an important economic phenomenon. The two most common types of plans which encourage ESO are Employee Stock Ownership Plans (ESOPs) and 401-K plans in which an employer contributes its own stock. According to the National Center for Employee Ownership, ten and one-half million employees participated in 9,650 ESOPs, with combined assets over $675 billion at public and private firms in The corresponding numbers for ESO through 401-Ks are four million participants in 2,200 plans with $75 billion in assets. Both of these plans show an increasing long-term trend; the NCEO estimates the number of participants in ESOPs was one-quarter million in 1975, five million in 1990, and over ten million in ESO through 401-Ks has also become increasing popular since the 1990s. There are a number of studies documenting worker productivity increases subsequent to adoption of ESO plans (Jones and Kato, 1995; FitzRoy and Kraft, 1987; and Beatty, 1995). Studies also show that the announcement of ESOP adoptions, not implemented under takeover pressure, is met with a positive stock price reaction (Gordon and Pound, 1990; Chang and Mayers, 1992; Chaplinsky and Niehaus, 1994; and Beatty, 1995). However, there is little evidence on how ESO plans affect employee compensation. The effect on employee compensation is an important issue, not only because it has employee welfare implications, but also because any change in employee compensation has a direct implication on firm valuation and shareholder value. The issue is especially relevant because share ownership may give employees a stronger bargaining position in compensation negotiations. A typical ESO bestows not only cash flow rights, but also voting or some other forms of control rights to employees. As the size of ESO increases, greater cash flow 2

3 rights may lead to greater productivity gains through improved team effects and collective employee behavior, while greater control rights may lead to higher employee compensation. It is not clear how these two effects offset each other in affecting the shareholder value. This paper attempts to address these questions by investigating how ESO plans affect firm valuation and employee compensation. To illustrate the pertinent issues, define π, V, and C as the value of operating performance gains due to ESO, the change in shareholder value, and the change in costs accompanying ESO, respectively, such that V = π - C. Then, V 0, if C π. The shareholder value effect, V, may depend on the size of ESO. When the size is small, giving employees negligible control rights, most of π may accrue to shareholders. When the size of ESO is sufficiently large to allow employees non-trivial control rights, employees total compensation and benefits (TCB) may increase. When TCB is positive, it will increase C and decrease the fraction of π accruing to shareholders. However, V needs not be smaller because π may be greater due to greater cash flow effects on worker productivity. If the incremental π is bigger than TCB, V will increase as the size of ESO increases; otherwise, an increase in ESO size will decrease V. There is no obvious theoretical prior on which outcome is more likely. Only data can tell. Thus, we conduct an empirical investigation of how the effects of ESO on firm valuation and employee compensation vary with the size of ESO. A positive TCB may materialize if a firm encourages a large ESO hoping for greater improvement in team effects and collective worker behavior, endowing employees with substantial control rights and strengthening their bargaining power in future wage negotiations. Alternatively, managers expecting hostile takeover threats may 3

4 intentionally provide workers substantial voting rights as part of a management-worker alliance, as in Pagano and Volpin (2005). They develop a management-labor collusion model, in which managers pay workers above-market wages to entice them to vote with management. Either way, employee compensation will increase. An undesirable, yet possible outcome associated with substantial employee control rights may occur if TCB > π such that V < 0. This represents the danger of value destroying corporate socialism. When employees possess sufficient control rights, they may extract unearned compensations and benefits at the expense of other stakeholders, increasing the firm s marginal costs and eroding growth opportunities. Such firms will invest less, suffer poor performance, and be valued lower. Faleye, Mehrotra, and Morck (2006) document such phenomena for firms with large ESO. However, they also raise the possibility that poorly managed and badly performing firms may establish large ESO plans to share their misfortune with employees. Thus, whether large ESO indeed leads to value-destroying corporate socialism warrants further investigation. We find that on average firms electing to encourage ESO (ESOPs) realize a 6% (9%) increase in firm valuation, relative to the industry median, where our definition of ESO plans is inclusive of both ESOPs and ESO through 401-K plans. The positive valuation effect persists even when comparing future effects to a baseline estimated two years after an ESOP adoption. For a small ESO plan (ESOP), defined as less than 5% of the outstanding shares, the valuation increase is even higher, estimated at 11% (12%). These estimates, however, represent the upper tail of the distribution of possible gains associated with ESO, as firms not promoting ESO are likely to anticipate smaller gains. 4

5 To examine the relation between value gains and the size of ESO, we define large ESO plans as those estimated to control more than 5% of the shares outstanding. In contrast to small ESO plans, large plans show no positive valuation effects. This difference in valuation effects is related to employee compensation and benefits (TCB). Large ESOPs are followed by significant permanent increases in TCB following their establishments, whereas small ESOPs are not. When TCB reported before the adoption of large ESOPs are compared to those reported by the same firm at least five years later, TCB increases an average $1,038 per employee, which translates to an average aggregate labor cost increase of over $30 million per firm in 2006 dollars. This increase in labor cost is well over the mean value of ESOP shares, $13 million in 2006 dollars, allocated each year for our sample of firms. The inverse U-shaped relation between employee ownership and firm value is not inconsistent with the Himmelberg, Hubbard, and Palia (1999) evidence of no relation between managerial share ownership and firm value with firm fixed effects. Our results are distinct from any underlying relation between managerial share ownership and firm value. The relation between ESO and firm value is significant only for those plans that distribute shares without a bias in favor of management. Our results also do not appear to be an endogenous result of firms choosing to implement an ESO plan. We estimate all regressions controlling for firm fixed effects. Thus, differences in time-invariant characteristics between ESO firms and a control group of firms do not affect our findings. We carefully consider several other possible selection biases: private information, confounding policy changes, and a spurious correlation 5

6 caused by a positive relation between firm performance and new share issues. Our results do not support any of these selection stories as an explanation for our results. Our findings raise the issue of why management would adopt large ESO plans, given the greater shareholder value associated with smaller plans. One possible explanation is the management-labor collusion theorized by Pagano and Volpin (2005), whereby management bribes workers with above-market wages in return for their cooperation in fending off takeover bids. Bribes should be more prevalent in states with business combination statutes (BCS), because ESO plans are more effective deterrents against hostile takeover attempts in BCS states. Consistent with this prediction, compensation increases following the adoption of large ESOPs are concentrated among firms incorporated in BCS states. This finding adds to the Bertrand and Mullainathan (1999, 2003) finding of significant compensation increases following the enactment of BCS; namely, the compensation increases they document occur mainly through establishment of large ESOPs. Thus, the managers pursuit of quiet lives described by Bertrand and Mullainathan seems to be facilitated by management-labor collusion. We also consider cross-sectional variation in compensation increases by firm characteristics. Bronars and Deere (1991) argue that high financial leverage weakens unions bargaining power because of the threat of possible bankruptcy. To test for a similar effect, leverage is interacted with the presence of an ESOP. We find that high leverage reduces employee compensation increases following large ESOPs. Leverage also weakens the negating valuation impact of large ESOPs, which is expected given leverage s counter-balancing effect on compensation. The leverage effect on valuation appears to be unrelated to the unique tax benefits permitted to leveraged ESOPs. When 6

7 we compare valuation impacts between leveraged and unleveraged ESOPs, the tax benefits appear too small to be detected by our regressions. Finally, we examine what happens when financially constrained firms adopt ESOPs that may be designed as a substitute for cash wages. We do not find any significant changes in firm valuation or employee compensation following such ESOPs. Our documented valuation and compensation effects seem to be concentrated at financially unconstrained firms. The rest of the paper is organized as follows. Section I briefly describes a celebrated case of ESO and surveys the literature. Section II describes the data. Empirical results and tests on selection biases are presented in Section III, followed by additional robustness checks in Section IV. Section V concludes. I. Effects on employees and shareholders The case of United Airlines illustrates the potential benefits and costs of ESO. In July of 1994, United employees who agreed to significant wage concessions were granted equity shares with an aggregate claim to 55% of the firm s cash flow rights. The plan was implemented to catalyze a cultural shift and transform United into a more competitive firm. Employees were also given the right to elect three members to United s twelvedirector board, thereby bestowing employees with substantial influence in electing the CEO. This influence was apparent in the selection of the first CEO brought in after the deal, Gerald Greenwald, who had a long history of being labor-friendly. The first few years under ESO appeared successful. Employee absenteeism and turnover declined, while United s stock performed better than the market. 7

8 When Greenwald retired after four years, employees supported and subsequently elected James Goodwin as CEO. Shortly thereafter, the pilot union began contract renegotiations with their hand-picked CEO. Unsurprisingly, the pilots walked away with a generous deal. They received pay raises of 22% to 29%, becoming the highest paid pilots in the industry. After September 11, 2001, the situation at United became particularly dire. Goodwin wrote an open letter to employees, acknowledging current troubles and asking for help. Presumably, this letter was directed at the mechanics union, which was about to begin contract negotiations. In response, the mechanics union demanded Goodwin s resignation and two weeks later he did. As before, John Creighton, the next CEO approved by the employee-owners, had a reputation for being laborfriendly. The mechanics union entered negotiations and, once again, a United union left the bargaining table with a favorable deal. After this labor agreement, United became the most expensive airline to operate on a per mile basis. The firm declared bankruptcy in December A. Employee compensation As the United Airlines case illustrates, when an ESO plan gives employees sufficient control rights, workers may use their influence to extract higher compensation and benefits. Employee control rights may exacerbate the managerial tendency to acquiesce to worker demands for higher wages when managers desire a quiet life as in Bertrand and Mullainathan (1999, 2003). Large employee ownership may also indicate worker-management collusion as theorized by Pagano and Volpin (2005). Powerful employees with control rights may induce management to shift its allegiance to workers 8

9 as suggested by Atanassov and Kim (2007). A likely result of any of these propositions is higher employee compensation. ESO also cause employees to hold less diversified portfolios and have liquidity concerns. A risk-adverse employee will value a cash equivalent higher than an undiversified stake in her employer s securities. ESOP shares cannot be sold until employees leave the company, with the exception of diversification requirements triggered at 55 or 60 years of age. In 401-K plans, employees are relatively free to sell their company shares, but they face penalties and tax withholdings if they withdraw the proceeds prematurely. In equilibrium, such risks and costs should be priced in the form of higher employee compensation. B. Firm performance The higher costs due to increases in employee compensation may be offset by higher worker productivity due to improved incentive and team effects, tax benefits, and the conservation of cash at financially constrained firms. B.1. Productivity gains The most often stated objective of ESO is to increase firm value by improving employee incentives. Shareholders typically do not monitor non-managerial employees; instead, they delegate the monitoring to management, agents themselves vulnerable to their own incentive problems. As a supplement to delegated monitoring and to better align employee incentives with shareholder values, firms may encourage ESO as an incentive device. However, individual workers may feel they have little impact on stock price, raising doubt on the ability of ESO to alter individual behavior in tasks requiring additional effort or sacrifice. 9

10 Collectively, however, important benefits may arise if ESO provides a proper group-based incentive. Kandel and Lazear (1992) argue that free-rider problems can be mitigated by orientation and indoctrination of new employees about workplace norms, which creates a work environment where peer pressure enforces the group-based incentive. FitzRoy and Kraft (1987) and Blasi, Conte, and Kruse (1996) also argue that group-based incentive schemes such as ESO induce co-monitoring, reducing costly monitoring by managers. Moreover, Jones and Kato (1995) argue that ESOPs induce employees to develop a sense of identity and loyalty to their company; participate more actively in productivity-enhancing activities, such as quality-control circles; and increase the quality of decision making. These arguments are consistent with the claims often made by firms initiating ESOPs that ESO improves team work by fostering a culture of employee involvement. ESO also may help prevent value loss due to labor disputes. Cramton, Mehran, and Tracy (2007) develop a model in which share ownership by unionized workers creates incentives for unions to refrain from costly strikes. In addition, granting stocks to employees may help attract employees with more optimistic views of the firm when potential employees have differing beliefs about future prospects of the firm (Oyer and Schaefer, 2005). This will have a positive effect on firm performance if optimistic workers are more productive, with spillover effects onto their co-workers. 2 These theoretical arguments on productivity are supported by Jones and Kato (1995) who document that an ESOP adoption in Japan leads to a 4-5% increase in 2 Furthermore, more optimistic employees may value shares higher, possibly enabling the firm to hire workers at below-market wages. As pointed out by Bergman and Jenter (2007), however, it requires bounded rationality on the part of employees because employees are able to buy publicly traded shares at market prices. 10

11 productivity, starting about three years after the adoption. They report that 91% of all firms listed on Japanese stock markets had an ESOP in Japanese ESOPs do not provide tax benefits and most shares are allocated to non-executive employees. 3 These results are remarkable because the typical Japanese ESOP is allocated 1% or less of outstanding shares, demonstrating that even very small ESO plans generate substantial productivity gains. In addition, FitzRoy and Kraft (1987) find that profit sharing and workers capital ownership have positive effects on factor productivity for a sample of metal working firms in West Germany. Although there are no comparable studies on worker productivity for U.S. firms, Beatty (1995) finds an increase in sales in the two years after the adoption of an ESOP. B.2. Tax effects ESOPs are often established through a trust which borrows money to buy company stock. Over time, the company repays the loan taken by the trust which, in turn, allocates its shares to employee accounts. These loan payments (interest and principle) are treated as wages and, thus, are tax deductible, within certain payroll limits. Tax benefits unique to leveraged ESOPs arise when dividends paid to stocks, held by the trust, are used to pay down debt. These dividends are effectively deducted twice from the firm s taxable income, once as wages and then again as interest payments. 4 B.3. Cash conservation Core and Guay (2001) find stock option plans for non-executive employees are often used at firms which appear cash-constrained. Likewise, issuing stocks through ESO 3 See an earlier study by Jones and Kato (1993). 4 Prior to November 1989, banks received a tax break to fund leveraged ESOPs, which led to below market interest rates on these loans. The dividend deduction became effective in

12 plans may be the result of financially constrained firms substituting stocks for cash wages. Such ESO plans are likely to lower cash wages without changing total employee compensation. While the decision to substitute equity for cash wages may be optimal for firms facing costly external financing, it is doubtful that such plans will have as strong an uplifting effect on employee morale, team effects, and collective behavior. C. Hypotheses To summarize, we predict that, all else equal, when the size of ESO is small, firm valuation will increase, with little changes in employee compensation. When the size of ESO is sufficiently large to enhance employee bargaining position in wage negotiations, employee compensation will increase. However, a larger ESO also gives employees greater cash flow rights, which may lead to a greater operating performance gains. Thus, the net impact of a larger ESO plan on firm valuation is ambiguous. Only data can tell. II. Data We make use of two data sources on employee ownership, one that we created and the other provided by the U.S. Department of Labor (DOL). The first database, referred to as the ESOP database hereafter, includes data on ESOPs at US public firms from 1980 through This data is hand collected using Factiva. For each year, we search Factiva using the terms ESOP and employee stock ownership plan. We read all articles and note the first date a firm is mentioned as having an ESOP. We identify 756 unique public firms with ESOPs over the sample period. Of these firms, we drop 35 firms with total assets less than $10 million in 2006 dollars. The lack of press coverage on such small firms makes it likely that we missed other similar- sized firms with ESOPs, wrongly 12

13 identifying them as non-esop firms. This potential error is important as our control group is derived from firms in Compustat without identified ESOPs. With the remaining 721 ESOP firms, we run additional Factiva searches using the firm s name and employee stock to locate further information on each firm s ESOP. 5 When available, we record information on whether the ESOP was funded with debt, the ESOP initiation date, the total cash flow and voting rights of the employees, and whether the ESOP was initiated in response to a takeover threat. 6 We are able to identify the year of the ESOP initiation for 360 unique firms. The ESOP database is then matched to Compustat and Center for Research in Security Prices (CRSP) databases. Our second dataset begins in 1992 and extends through 2004 and includes all ESO through company-sponsored plans, including ESOPs and 401-K plans, as reported on Internal Revenue Service (IRS) Form 5500 files. Obtained from the Department of Labor (DOL), we refer to this dataset as the DOL database. All firms with employee retirement programs are required to file Form We estimate employee ownership using the dollar value of the plan s investment in their company stocks. This database is also matched to Compustat and CRSP databases. The inclusion of the 1980s is a key advantage of the ESOP database. To investigate effects of an ESOP implementation, we need to estimate a baseline, or pre- ESOP firm characteristics, requiring firm-year observations prior to the ESOP initiation. 5 In a few cases, this additional search led us to identify the presence of an ESOP in an earlier year. We exclude these observations because their inclusion introduces a survivorship bias. Information about an ESOP may not have been discovered in our first search process if the firm was small and received limited press coverage. When the firm becomes more profitable and grows larger, press coverage becomes more likely, increasing the probability we observe the ESOP. This could cause a positive correlation between observed ESOPs and firm performance. 6 If a firm underwent a bankruptcy or was dropped from Compustat for a year or more, we assume the ESOP was terminated unless other information is present. We also assume ESOP shares are common voting, unless otherwise specified. 13

14 The DOL database begins coverage in 1992, a serious drawback because over 80% of our ESOP sample was implemented before We, therefore, conduct primary tests using the ESOP database. The DOL database is used to obtain additional information unavailable in our ESOP database, 7 and as a robustness check. Table 1, Panel A, presents summary statistics of both databases including the number of new ESOP adoptions, observation counts, and average employee ownership percentages. 8 The ESOP database identifies 5,872 firm-year observations between 1980 and 2004 with an average employee ownership of 8.23% (the median = 5.30%). 9 The DOL database identifies 5,387 firm-year observations for ESOPs between 1992 and 2004 with an average employee ownership of 9.03% (the median = 4.98%). 10 The DOL database also shows 13,065 firm-year observations of employee ownership not attributed to an ESOP, which we attribute to 401-K plans (see Rauh, 2006). These plans report a substantially smaller average employee ownership of 3.16% (the median = 1.23%). Panel B, of Table 1, provides summary statistics of the relevant firm level variables. The first column details the control group. It summarizes characteristics of pooled time-series observations in Compustat that meet the following criteria: (1) we do 7 Factiva searches are inadequate to identify information on other forms of employee ownership, such as 401-K plans. Thus, when using the ESOP database and comparing firms with ESOPs to firms without ESOPs (the control group), other unobserved forms of employee ownership may or may not be present, adding noise to our tests. 8 There are 957 firm-year observations in our ESOP database (16% of the sample) for which we confirm the presence of an ESOP but do not have adequate information to estimate the percentage of shares held by the ESOP. Because detailed information must be filed in proxy statements when ownership exceeds 5%, we infer from this lack of information that the percentage ownership is less than 5%. 9 The reported means and medians exclude firms that are identified as having ESOPs but do not provide information on the percentage of shares held by employees. As mentioned above, these firms are likely to have employee ownership less than 5%. Thus, the actual mean and median are likely to be smaller. 10 The DOL database covers more firms with ESOPs than our ESOP database for several reasons. First the DOL database does not exclude very small firms, as we do with the ESOP database. Second, the DOL database includes KSOPs, which is a hybrid ESOP K plan, in most aspects resembling a 401-K plan, but with the tax advantages of an ESOP. These plans are rarely reported in newspaper articles. Finally, we may have simply missed firms with ESOPs that are included in the DOL database. 14

15 not confirm an ESOP, (2) the firm has total assets greater than $10 million in 2006 dollars, and (3) the firm has more than 3 years of Compustat data-- our empirical design requires a minimum of 4 years data to estimate ESO effects. The second column describes firms in Compustat for which we identify an ESOP. The third column details firms with large ESOPs that are estimated to control more than 5% of the outstanding common shares. The median and mean ESOP share ownership for this group of firms are 11.00% and 14.10%, respectively. We choose a demarcation point of 5% to separate small versus large ESOPs because it seems a reasonable tradeoff point between what is large enough for the employee-shareholders to exert influence on corporate decisions without making the sample size of large ESOP firms too small. The 5% is also close to the sample median and represents the triggering point at which firms must disclose the size of employee ownership in their proxy statements. Comparing Column 1 with Columns 2 and 3 reveals that firms with ESOPs tend to be larger, more capital intensive, more profitable, less R&D intensive, and more highly levered than firms without ESOPs. Columns 4 and 5 provide the summary statistics for firms in the DOL database, exhibiting similar patterns. III. Empirical results In this section we first estimate the relation between firm value and the presence of ESO, followed by an examination of possible selection biases. Then, we investigate the effects of ESOP adoption on employee compensation, the disciplining role of financial leverage, and alternative benefits to ESOPs not associated with incentives. A. Relation between firm valuation and ESO 15

16 We estimate the relation between firm value and employee ownership by regressing industry adjusted Q on indicator variables for the presence of ESO plans. Our dependant variable Q it is estimated as the market to book value ratio of firm i at time t minus the (2-digit SIC code) industry median market to book value ratio at time t. Our approach is similar to that of Himmelberg et al. (1999). We assume Q depends on the presence of an ESOP, observable firm characteristics, and unobservable firm characteristics. We include firm fixed effects to control for time-invariant unobservable firm characteristics. We also control for time series patterns with year fixed effects. To control for observable firm characteristics, we include the log of total assets (normalized in 2006 dollars), the R&D expenditures to sales ratio, the capital expenditures to assets ratio, and financial leverage. ESOPs are often initiated as leveraged, which adds to a firm s total debt, although any increase can be neutralized by rebalancing other debt obligations. We also control for firm age. Because the indicator variable for an ESO plan is comparing industry adjusted Q following the adoption of the plan to an earlier period, the coefficient on the indicator variable could pick up an age factor, if industry adjusted Q changes with firm age. We define Age as the difference between the current year and the first year the firm is included in Compustat. 11 A.1. ESOPs We first estimate the base regression using the ESOP database over the period 1980 to Column 1 of Table 2 shows that the presence of an ESOP is associated with a statistically positive increase in industry adjusted Q. This positive correlation between employee ownership and Tobin s Q is in sharp contrast to Himmelberg et al. (1999), who find no relation between managerial ownership and Tobin s Q, after the 11 Because Compustat data is not available prior to 1950, the oldest firm in our sample is

17 inclusion of firm fixed effects. One possible explanation for our strong positive association between employee ownership and firm value is the greater time series variation in ESOP holdings at the individual firm level (see Zhou (2001)). The average (median) within-firm standard deviation in employee ownership is (0.012) for firms with ESOPs in our sample, whereas the standard deviation in CEO ownership in Execucomp is (0.003). The difference in within-firm variations between employee and CEO ownership is even more dramatic if we consider an ESOP indicator, which has an average (median) within-firm standard deviation of (0.476), for firms that have an ESOP at one point during the sample period. In addition, we have over 200% more usable observations than Himmelberg et al. Column 1 also shows that the coefficients of the control variables are consistent with our expectations. Larger firms are valued less, firms with more R&D investment and capital expenditures are valued more, and financial leverage is positively related to firm valuation. The negative coefficient on Age indicates that new firms have high valuations, possibly reflecting large growth options. The positive coefficient on Age squared suggests that as firms invest in these growth options, the relative value of the firm declines at a decreasing rate. Viewed in the context of a difference-in-difference approach, the firm fixed effect picks up the before period and ESOP captures the average change experienced in the after period. This interpretation motivates two refinements to our sample. First, the estimate of the value of the firm before an ESOP, as captured in the firm fixed effects, proxies for the expected value of the firm in future years, had the ESOP not been adopted. Thus, we only include those firm-year observations beginning five years prior to the 17

18 ESOP adoption, for the set of firms that later adopt ESOPs. Second, we exclude the year of the announcement of ESOP adoption and the year after, because it may take time for effects associated with the ESOP implementation to be observed, as in Jones and Kato (1995). The sample with these modifications is used in Column 2 and in all subsequent regressions. As a comparison of Columns 1 and 2 indicates, the results are quite robust to the sample refinements. To give a sense of the economic magnitude of the result in Column 2, we note that the median value of Tobin s Q in our sample is Thus a coefficient of implies that the presence of an ESOP is associated with a firm valuation increase of 8.67%, relative to the industry median. The coefficient in Column 3 of implies a firm valuation increase of 11.72% associated with small ESOPs. If ESOPs are implemented to maximize shareholder value, then firms which choose not to encourage ESO may do so in anticipation of benefits being outweighed by costs. Thus, these estimates likely represent the upper tail of the potential distribution of shareholder value creation if all firms were to adopt ESOPs. In Column 3, we include an indicator variable to capture ESOPs with more than 5% of the company shares, ESOPg5. The results show that the positive relation between firm value and an ESOP diminishes if the ESOP controls more than 5% of the outstanding shares. Because the coefficients on the ESOP indicator variables are cumulative, the combined coefficient on ESOPg5 is , or To determine if large ESOPs are associated with an overall firm value effect, we enter 18

19 ESOPg5 alone in Column 4 and find an insignificant coefficient. The relation between firm value and ESO seems to be inverse-u shaped. 12 It is possible that this inverse-u shaped relation is driven by a similar inverse U- shaped relation between managerial share ownership and firm value, as documented by Morck, Schleifer, and Vishney (1988) and McConnell and Servaes (1990), among others. However, Himmelberg et al. (1999) show that this relation with managerial ownership disappears with the inclusion of firm fixed effects. Since our regressions also include firm fixed effects, it is unlikely our results are confounded by managerial share ownership. Nevertheless, we check the robustness by separating ESOPs into high and low levels of managerial participation by using information available in the DOL database regarding the nondiscriminatory coverage requirement (NCR). 13 In general, if a plan satisfies the NCR, the plan does not allocate highly compensated employees, presumably management, a disproportionate fraction of the ESOP shares. In Column 5, we re-estimate the regression by separating the sample according to whether or not the NCR is satisfied, which is virtually time-invariant. 14 The coefficient on ESOP biased towards highly compensated employees is negative and significant 12 A potential source of noise is the age of ESOPs. A firm that initiated an ESOP 15 or 20 years ago has probably not allocated any new shares during the past few years, making the effect of the ESOP weaker in the later years. In an unreported regression we add a new indicator variable, 10 years after the ESOP initiation, which assumes the value of 1 if the firm has had the ESOP for 10 years or more. This variable also is interacted with ESOPg5 to allow the variables ESOP and ESOPg5 to capture the effect of the ESOP within the first 10 years of adoption. Neither the new variable nor the interaction term is significant. The results are virtually identical. 13 Specifically, the test takes into account differences between the coverage ratios for highly and non-highly compensated employees, the percentage of total employees covered, and the compensation of employees covered by the plan as compared to employees excluded by the plan. For more information, see the IRS instructions for completing form 5500 available at The definition of highly compensated employee is contained in Code section 414(q), as amended by section 1431 of SBJPA, those regulations under section 414(q) that reflect current law, and Notice 97-45, I.R.B The mean (median) variance of an indicator variable on whether an ESOP satisfies the NCR is (0.000). 19

20 relative to ESOP. Likewise, the coefficient on ESOPg5 biased towards highly compensated employees is positive and significant relative to ESOPg5. Estimated separately, the coefficients on ESOP and ESOPg5 at firms where the plans are biased toward highly compensated employees are not significantly different from zero. Because whether a firm s ESOP meets the NCR is likely a choice made by the firm, we can only conclude that the inverse U-shaped relation between firm valuation and ESOPs is driven by firms which do not give a disproportionate share of ESOP stocks to managers. A.2. Employee stock ownership through 401-K plans As a robustness check, we repeat the regression analyses with the DOL database in Table 3. In Column 1, we create a new indicator variable, Employee ownership, which assumes the value of 1 if the firm has any type of ESO. Employee ownershipg5 indicates an ESO estimated to control more than 5% of the shares outstanding. The results are consistent with those in Table 2. In Column 2, we distinguish between ESO attributed to an ESOP and not attributed to an ESOP. Because most of the latter type is associated with 401-K plans, we create an indicator variable 401-K, which assumes a value of 1 if the firm has positive ESO not associated with an ESOP. 401-Kg5 indicates the presence of ESO greater than 5% of the shares outstanding. The coefficients on the ESOP and 401- K variables suggest an inverse U-shaped relation for both types of ESO. In Column 3, we include only ESOPg5 and 401-Kg5. The results show that unlike large ESOPs, large ESO through 401-K plans is associated with an overall negative firm valuation. One possible explanation relates to the stylized fact that shares in an ESOP cannot be sold at employees discretion, whereas there are no such restrictions on stocks held in a 401-K plan. Perhaps, the ability to sell stocks in 401-K plans weakens the 20

21 benefits arising from employee incentive and team effects such that the costs and risks associated with large employee control rights dominate the benefits. An alternative explanation for the difference is that the ESOP sample is concentrated in the mid-1980s to early-1990s, whereas most 401-K ESO plans appear after the mid-1990s. It could be that employees have learned over time how to better coordinate their control rights to their maximum advantage. 15 In sum, we find a positive association between employee share ownership and firm value. This positive relation is limited to relatively small ESO, and disappears when the size of ESO becomes large. Our result is not driven by high levels of managerial participation in ESOPs. Finally, there is a difference between large employee ownership through an ESOP and a 401-K plan. Large ESOPs have no association with firm value, but large ESO through 401-K plans seem to be associated with an overall negative firm valuation. So far, we have been careful to describe our results as evidence of a relation between firm value and ESO. We will next make the argument, supported by additional empirical evidence, that these results are consistent with a causal relationship. B. Selection bias To argue for a causal relation between ESOPs and firm value, we first rule out alternative explanations of our findings based on selection biases. The most common selection story argues firms that select the treatment in our case, establish employee 15 It also could be due to employees selling their 401-K shares in their own company stock to lock in the gains after positive stock-price performance and/or to firms offering more employee shares to 401-K accounts when their stock appears to be under-valued. Such behaviors would result in smaller (greater) employee ownership in 401-K plans during years of high (low) stock market valuations. Unfortunately, we are unable to test this hypothesis with our data because we observe only net changes in holdings, not the separate components of new share allocations and employee share sales. 21

22 stock ownership plans are inherently different from firms not selecting the treatment. Such differences may involve firm characteristics which are stationary or evolving over time. Because the inclusion of firm fixed effects in all of our regressions controls for stable firm characteristics, we concentrate on possible selection biases which depend on time-varying characteristics. B.1. Private information story We first consider a story where a firm has private information, predicting higher future permanent profits, and decides to reward its employees through an ESO plan. In such a case, we would expect to find a positive association between firm value and an ESO plan. Industry adjusted Q will increase when the market learns the positive information following the adoption of the ESO plan. We control for this possible selection bias by relying on a stylized fact that ESOPs are often implemented over time. For example, in leveraged ESOPs, a trust temporarily holds all ESOP shares and allocates the shares to employee accounts over time, with regulation dictating that all shares must be allocated within ten years. 16 Some nonleveraged ESOPs also allocate shares over time, especially when firms purchase shares on the open-market. We expect these ESOPs to have relatively modest effects on employee behavior in the first few years following initiation. 17 In addition, Jones and Kato (1995) find productivity increases start to take effect about three years after the adoption of ESOPs for Japanese companies. Thus, we use the second full fiscal year after the announcement of the ESOP as the baseline year (or the before period in the 16 There are no specific legal requirements regarding who can vote ESOP shares held by a trust and, thus, firms are free to set their own rules at the ESOP initiation. The two most common approaches are to vote the shares 1) according to management s preferences or 2) in an identical proportion to the votes cast by employees holding allocated ESOP shares. 17 Data limitations prevent us from identifying specific ESOP observations which allocate shares over time. 22

23 difference-in-difference approach.) We expect these ESOPs to have additional effects in later years, relative to the baseline of the 2 nd year. In Table 4, we use this alternative baseline with our ESOP database. We recode the firm-year observation which falls two fiscal years after the adoption of an ESOP (and thus was originally coded as having an ESOP) as not having an ESOP. We also drop all earlier observations of ESOP firms. Now the ESOP indicator variable in Column 1 will pick up differences in firm characteristics between this second year after an ESOP initiation and the subsequent years. We also exclude firm-year observations after an ESOP is terminated to ensure that our baseline is not picking up post-termination effects. 18 The results reported in Table 4, Panel A, are remarkably similar to their counterparts in Table 2 (Columns 2 and 3). Because the private information is not likely to remain private for more than two years, these results contradict the selection story based on private information held at the time of initiation. We observe no attenuation of the ESOP effect with the alternative baseline. On average, the market seems to have underestimated the positive effect of small ESOPs. Perhaps, this is not too surprising. The complexity of an ESOP makes it difficult to assess its impact in the earlier years, because many details about share allocation and workers response to share ownership are still unknown. A worker s behavioral reaction to becoming an employee owner is highly individual-specific. It is difficult to predict how different individual reactions translate into group behavior in the workplace. The 18 There are 61 ESOP terminations (269 firm-year observations) in our ESOP database. Terminating an ESOP is a complex legal procedure. The firm must be able to legally justify why the ESOP was valueincreasing for the firm in the past but is now value-decreasing; otherwise, it is open to lawsuits from ESOP holders and shareholders. Thus, it is more common to freeze-out an ESOP. A freeze-out is usually not announced officially and thus is hard to identify. In our sample, firms which are electing to freeze-out their ESOP will still be recorded as having an ESOP, which is literally true because the ESOP still exists. There are some firms that have rolled up their ESOP into a 401-K plan. Such 401-K plans may still be recorded in our database as an ESOP, which is not completely off-base because they still represent ESO. 23

24 same can be said about how the newly acquired voting rights or the prospect of getting more will affect employee influence on corporate decisions because the influence depends greatly on their ability to coordinate their votes and how successful shareholders are in countering it. These uncertainties require time to resolve, and the market will reassess the firm value as shares are actually allocated and observable actions materialize. Other information-based selection biases follow if an ESOP is implemented to deter hostile takeovers or to conserve cash. The announcement of an ESOP initiation may reveal private information that the management is concerned about possible takeover bids or about the availability of funds to satisfy wage obligations. This information is likely to be reflected immediately in stock prices, not two fiscal years after initiation. Thus, they are also contradicted by the results in Panel A, of Table 4. Furthermore, even if the market slowly compounds this new information, these stories are unable to explain our non-linear findings. For example, if a signal for potential hostile bids is value-enhancing, it may explain the higher firm value associated with small ESOPs, but cannot explain the negative effect of large ESOPs. Alternatively, if it signals that management is more interested in private benefits of control, it may explain the value negating effect of large ESOPs, but cannot explain the value increases with small ESOPs. B.2. Correlation between firm performance and issuance of new shares In this section we explore a bias which may arise in computing the percentage of shares held by ESOPs, estimated as the number of shares controlled by the ESOP divided by the number of shares outstanding. While the numerator is often constant for a number of years, the denominator fluctuates as the firm issues new shares or repurchases outstanding shares. If high performance firms issue new shares to expand their operations 24

25 or acquire other firms, the denominator will increase, making the relative size of ESOPs smaller. Conversely, poorly performing firms may repurchase outstanding shares because they lack good investment opportunities, making the size of ESOPs larger. To control for this possible spurious correlation between firm performance and the size of ESOPs, we create two additional variables to capture changes in (splitadjusted) shares outstanding. The first variable, share difference 1 y, is estimated as (current shares outstanding - shares outstanding from one year prior) / shares outstanding from one year prior. Share difference 5 y is estimated in a similar manner but with a five year change to shares outstanding. We revert to the original baseline as in Table 2, and compare changes following ESOP adoptions to firm-years which precede the ESOP. Column 3 of Table 4 reports the results with these additional control variables. We observe a positive relation between the one-year share difference and firm value. However, controlling for changes in shares outstanding does not affect our principal finding of a positive coefficient on ESOP and a negative coefficient on ESOPg5. Estimating share difference 5 y requires a minimum of five years of data on shares outstanding in CRSP, which may introduce a new bias by limiting the sample to older firms. Thus, we repeat the regression with a modified definition of share differences. In cases where there is inadequate data to estimate share difference 5 y, we instead code the difference in shares outstanding as zero. This captures the intent of our control. For firms without an earlier time period, we should anticipate no legacy effects from earlier time periods. The results are reported in Column 4. As with Column 3, we find a positive coefficient on ESOP and a negative coefficient on ESOPg5. In unreported tests, we estimate various alternative regressions, including interactions between the change in 25

26 shares outstanding and the ESOP variables. We consistently find a positive and significant coefficient on ESOP and a negative and significant coefficient on ESOPg5. B.3. Confounding policy changes Another alternative story for our findings is that at the time of ESOP initiation, the firm implements other policy changes that affect firm value. To explain the results in Table 4, Panel A, the effects of these policy changes must be slow changing and continue to impact stock prices two fiscal years after the ESOP initiation. An example would be a management team that believes a hostile takeover bid is looming and the best defense is to maximize firm value by increasing efficiency. As a precautionary move, the firm also implements an ESOP. If this were the case for a large portion of our sample firms, we would observe a positive correlation between an ESOP and firm value. Although this story may explain some of our results, it is contradicted by our overall results. First, an efficiency-based story must not only explain the increase in firm value associated with small ESOPs, but also the declining value associated with large ESOPs. To the best of our ability, we cannot think of an efficiency-based story that can explain both. Second, one way to increase efficiency is to cut employee compensation. In the following section we examine employee compensation changes subsequent to ESOP initiations. We find no evidence of a decrease in employee compensation. C. Employee compensation In this section we test the prediction that employees with sizable control rights are more successful in extracting higher total compensation and benefits (TCB). We estimate TCB as the log of the ratio of total compensations and benefits (in thousand dollars, normalized to 2006 dollars) to the number of employees. Because personnel 26