The elusive ESOP—productivity link

Journal

of Public

Economics

52 (1993) 273-283

North-Holland

The elusive ESOP-productivity Evidence

from U.S. firm-level

link

data

Subal C. Kumbhakar Department of Economics, Uniuersity

of Texas at Austin, Austin, TX 78712, USA

Amy E. Dunbar Division of Accounting and Information Systems, University of Texas at San Antonio, San Antonio, TX 7828S, USA Received July 1990, final version

received December

1991

This paper investigates whether employee participation in ownership or profit-sharing in publicly held firms through an ESOP or profit-sharing plan was positively associated with productivity measures. The sample consists of firms that adopted such plans during 1982 through 1987. Production function models augmented with age variables for ESOP and profitsharing plans were estimated using panel data-procedures. The productivity effect increased with the age of the ESOP at the rate of 1.8 to 2.7 percent per annum and with the age of the profitsharing plan at the rate of 3.9 to 4.6 percent per annum.

1. Introduction Productivity issues are becoming increasingly important in the United States. According to Alan Blinder, ‘the nation’s chief economic malady today is not the budget deficit, the trade deficit, inflation, or poverty. It is our miserably slow pace of productivity improvement’ [Blinder (1989, p. lo)]. An increase in productivity is associated with a more efficient utilization of inputs, which could result from a number of sources including increased effort from labor, better management and organization, and improved technology. Because labor constitutes the major share of production costs, an increase in labor productivity is likely to exert greater influence on productivity (output per labor). Thus, a policy increasing labor productivity by a certain percentage is more effective than a policy raising capital productivity, for example, by the same percentage. Labor productivity can be raised through increased work effort. There is, however, no unique way of increasing work Correspondence to: S.C. Kumbhakar, TX 78712, USA. 0047-2727/93/$06.00

0

1993-E]

Department

of Economics,

sevier Science Publishers

University

of Texas,

B.V. All rights reserved

Austin,

214

XC. Kumbhakar and A.E. Dunbar, ESOP-productivity link

effort. While factors such as job tenure, unemployment rate, and extent of supervision affect work effort, changes in the compensation system may also influence labor-use efficiency. Employee participation in ownership and profits through retirement plans is an example of the latter. Prior research indicates that profit sharing does increase productivity. The evidence linking employee stock ownership plans (ESOPs) and productivity is much weaker. Because an ESOP is a retirement plan that invests primarily in employer securities, it links the employees’ retirement wealth to the firm’s stock performance.’ Employees with an equity stake may be motivated to work harder, with the result that productivity will increase. According to a Government Accounting Oflice survey [GAO (1986)], 70 percent of the firms adopting ESOPs expect improved productivity. The rest of the paper investigates the link between ESOPs and productivity. In addition, the link between profit-sharing plans and productivity is confirmed. The specific issue is whether employee participation in ownership or profits in publicly held firms that adopted an ESOP during the period 1982 through 1987 is positively associated with productivity measures. We estimate firm-level production functions augmented with variables representing the age of the ESOP and profit-sharing plan. The production functions are estimated using panel data procedures that accommodate both firm- and time-specific effects. This paper is organized as follows. Section 2 describes prior ESOPproductivity research. Section 3 discusses some models that capture productivity effects in different forms. The sample is described in section 4. Section 5 describes the estimation procedures and presents the results. The final section summarizes the paper. 2. Previous research The productivity effects of alternative compensation systems have been analyzed in detail in Blinder (1990). The profit-sharing literature surveyed by Weitzman and Kruse (1990) supports the view that profit-sharing increases productivity.’ Although the theoretical arguments supporting increased productivity from profit sharing are applicable to ESOPs, Conte and Svejnar (1990) concluded from their survey of the ESOP literature that the evidence for a positive and statistically significant ESOP effect is mixed. There are

‘Other retirement plans can also invest in stock of the employer. All retirement plans are either defined benefit or defined contribution (which includes ESOPs and profit-sharing plans). Investments of defined benefit plans in employer securities are limited to 10 percent of plan assets. There is no limit for defined contribution plans. %ee Wadhwani and Wall (1990) for the effects of profit-sharing on employment, wages, stock returns, and productivity based on firm-level data from the United Kingdom.

SC. Kumbhakar and A.E. Dunbar, ESOP-productivity link

215

fewer studies (relative to the profit-sharing literature) examining the ESOP productivity effect. The general conclusion of two ESOP-productivity studies that did not use panel data procedures to control for firm- and time-specific effects [GAO (1987) and Conte and Svejnar (1988)] is that employee stock ownership does not appear to lead to productivity gains unless coupled with participation in decision-making. Bloom (1985) used both cross-sectional and first-difference time-series regressions, but found no significant ESOP effect. Kruse (1989) used panel regressions and concluded that the productivity effects of ESOPs were not as consistently positive as those of prolit-sharing plans and almost never statistically significant. Bloom (1985) and Kruse (1989) used the Department of Labor (DOL) Form 5500 tapes, which are also used in this study. GAO (1987) used survey data in addition to Internal Revenue Service data. Conte and Svejnar relied only on survey data. The survey studies were able to incorporate additional measures of the participation variable beyond a binary variable for employee ownership. For example, GAO considered the degree of ownership, employee control, and employee participation in decision-making. Because our study relies on publicly available data, it cannot control for the level of participation in decision-making. This information must be obtained directly from the firm. This study, however, emphasizes the model specification to separate ESOP effects, if any, from several other factors that might mask the true ESOP effect. First, we control for firm- and time-specific effects. The firm-specific effects capture heterogeneity of the firms in the sample. The time-specific components capture the effects of such factors as economic fluctuations and technical changes on productivity. Second, we allow the factor shares (elasticity of inputs) to change over time. Changes in factor shares capture non-neutral technical change. Finally, instead of using a dummy variable for ESOPs we use age of ESOPs as the relevant variable capturing the ESOP effect on productivity. This age variable allows ESOP effects to change according to the duration of the plan. Some of these controls cannot be used if panel data are not available. For example, ESOP effects in a cross-section data model cannot control for firmand time-specific effects. Similarly, the fact that some firms have adopted ESOPs for a longer period of time and therefore might have a different effect from others cannot be captured. The ESOP dummy captures the mean effect of ESOPs of different ages. Failure to control one or more of these factors, however, might bias the estimates of the ESOP effect.3 This research uses panel data and thus can control each of these.

‘Some of these features are included in previous studies but not all of them at the same time. Failure to control all of these factors might lead to inconsistent estimates of capital and labor coefficients.

216

XC. Kumbhakar

and A.E. Dunbar, ESOP-productivity

link

3. The model We specify the production y=

f(K 9,

technology

as

(1)

where Y is output, K is capital, L is labor, and f( .) is the production function. Labor is usually measured by hours worked or number of workers. In the efficiency wage literature [see Akerlof and Yellen (1986)] a distinction is made between physical labor and work effort (unobserved) so that the labor input in the production function should be measured in efficiency units. Because labor in efficiency units is the product of physical labor and work effort, the relevant question for this research is whether a change in the compensation system (e.g. adopting an ESOP or protit-sharing plan) is associated with increased work effort. Such effort might come from ‘ . . . working harder, working smarter, taking initiative, and taking advantage of unforseen opportunities of time and place. . . . [A] worker will work harder and produce more output under profit sharing than under a rigid wage system because he or she has some stake in the outcome’ [Weitzman and Kruse (1990, p. 97)]. Because the ESOP is required to invest primarily in the stock of the that tying the fortunes of labor and capital employer, ‘. . . the argument together might improve productivity has been applied to ESOPs, just as it has to profit sharing’ [Blinder (1989, p. 7)]. Increased productivity might be forthcoming from psychological satisfaction (because of ownership), information sharing with the managers, and less supervision. Ownership in a company might also give employees some incentive to use capital and other inputs more efficiently (less waste). These factors suggest that adoption of an ESOP or a profit-sharing plan might increase efficiency of both capital and labor inputs. Thus, we can write the production function in (1) as

where 0, and 8, are efficiency factors associated with capital and labor, respectively. The formulation of the production technology in (1) and (2) is not very helpful for further analysis unless some specific functional form is assumed. We start with the Cobb-Douglas form of f( .) in (1) and specify 8, and 9, as f3,=exp(flKEESOP+PKPPS) and 8,=exp(fl,,ESOP+fi,,PS), where ESOP and PS are dummy variables for adoption of ESOPs and profit-sharing plans. With these specifications, the production function in (2) can be written as

S.C. Kumbhakar and A.E. Dunbar, ESOP-productivity link

271

In x, = cl0 + CIKIn Ki, + c1~In.& + BEESOPi* + 0pPSit + sir,

(3)

Model I

where Eif = /ii + A, +

vjf

(4)

and

Subscripts i and t denote firm (i= 1,. . . , F) and time (t= 1,. . . , T), respectively, sit is the random term which consists of (i) a firm-specific component, pi, (ii) component, vit, a time component, I,, and (iii) a firm- and time-specific assumed to be independently and identically distributed over time and across firms with zero mean and constant variance. We extend the above model by allowing labor and capital elasticities (Q and c(J to vary over time. To avoid imposition of any a priori restrictions on their time behavior we specify ~(~(t) and OIL as

%(t)= Y&L

(Q(t)=YL.Pt,

(6)

where D, is a dummy variable which equals one for year t and 0 These specifications allow shares of capital and labor to change due to changes in economic conditions (not under the control of This phenomenon of time-varying factor shares can be viewed neutral technical change. eE, the coefficient of ESOPi,, captures the effect of adopting an productivity. It can be interpreted as the difference in mean (log) the ESOP adopters over non-adopters, namely

8, = E(ln Y 1ESOP

adopters)

- E(ln Y ( ESOP

non-adopters),

otherwise. over time any firm). as a nonESOP on output of

(7)

everything else being the same. tip, the coefticient of PSI,, can be similarly interpreted. It is not possible, however, to identify whether the productivity gain is the result of improved efficiency of labor or capital or both. One major problem with the above model is that eE and 0, remain unchanged with the duration (age) of the plan. It is very likely that the productivity effect for firm i with a three-year-old plan at time t is different from firm j with a one-year-old plan even though everything else is the

278

S.C. Kumbhakar and A.E. Dunbar, ESOP-productivity link

same.4 More generally, productivity may increase or decrease with the duration of the plan. In view of the above problem we extend the basic model outlined in (3) and (4) to accommodate age of the plans in measuring the effect of ESOP and PS on productivity. For this we define the age variables as ESOPageit = 1: = ,, ESOP,, and PS - ageit = 2: = 0 PS,, and rewrite eq. (3) as Model II

In x, = Or,+ UKIn Ki, + aL In Lit + y~$, In Ki, + yLtD,In Li, + BEESOP- ageit + B,PS-

age,, + pi + 2, + Vi*.

(8)

The models specified in (3) and (4) as well as in (8) tit into the standard panel data model. If the firm- and time-specific effects are assumed to be random, one gets the random effects (RE) model. On the other hand, the fixed effects (FE) model is obtained when ,ui and 1, are treated as fixed. Following the panel data literature, we assume E(pi) = E(I,) =0 in the RE model. Because the common intercept a0 is present in (8), all of the parameters in pi and R, cannot be identified in the FE model unless C pi =C ~,=O. If the non-adopters (control group) and the adopters (treatment group) are otherwise identical, the coefficient on the ESOP-age variable captures the ESOP effect.5 It can be interpreted as the difference in mean (log) output of ESOP-adopters (of a certain age group) over ESOP non-adopters. That is, for example, when ESOP-age= 1, 19~= E(ln x, 1ESOP - age = 1) - E(ln x, 1ESOP - age =O),

(9)

everything else being the same. Thus, the effect of a two-year-old ESOP is 2eE. Alternatively, with an increase in the age of the ESOP by one year, output increases by 100 percent of eE. The coefficient on the PS-age variable can be similarly interpreted.6 4. The sample

This research uses a data base that was constructed

by matching publicly

4Thus, if one uses cross-section data [e.g. Bloom (1986)], the ESOP effect is confounded with (i) the effects of plans of different ages, and (ii) firm effects which arise due to heterogeneous firm sizes. It can, therefore, be argued that the cross-section ESOP effect in the Bloom study is not the ‘true’ ESOP effect but a combination of(i) and (ii) above. ‘The control group consists of the observations for each firm for the years prior to the year of adoption of either the ESOP or the profit-sharing plan. 6Kruse (1989) used a panel data model where the trend variable is interacted with the ESOP and PS dummy variables. These interaction terms are different from our ESOP-age and PS-age variables because adoption of these plans occurred at different points in time.

219

SC. Kumbhakar and A.E. Dunbar, ESOP-productivity link Table SIC code composition

1

of 1981 Annual

Compustat

firms.

Annual

firms

ESOP

adopters

Profit-sharing

Category

SIC code

No.

%

No.

%

No.

%

Agriculture Mining Construction Manufacturing Transportation Communication Utilities Wholesale Retail Finance Insurance Real estate Investment Services

l-9 lo-14 15-17 20-39 4&47 48 49 5&5 1 52-59 60-62 63-64 65 6669 7Ck89

2 69 20 816 45 26 191 78 99 172 54 36 46 110

0.1 3.9 1.1 46.3 2.6 1.5 10.8 4.4 5.6 9.8 3.1 2.0 2.6 6.21

0 0 1 35 1 0 2 5 4 0 0 1 0

0 0 1.9 64.8 1.9 0 3.1 9.2 1.4 0 0 1.9 0 9.2

0 3 0 30 2 2 18 6 5 1 0 0 0 1

0 4.3 0 43.5 2.9 2.9 26.1 8.7 7.3 1.4 0 0 0 2.9

1,764

100.0

54

100.0

69

100.0

Total

adopters

held corporations on the Annual Compustat tapes with Form 5500 information. Form 5500 is the annual return filed by an employee benefit plan with 100 or more participants. IBM compatible tapes reflecting the information from the Forms 5500 tiled by plans for 1981-1987 were obtained from the DOL. The Compustat firms were merged with the DOL firm filings using the firm’s employer identification number.’ The plans of these firms were examined to identify any ESOP or profit-sharing plan adoption during the period 1982-1987. Any firm whose other plans were not consistent throughout the period was not considered. 79 ESOP and 99 profit-sharing firms were initially identified. After eliminating firms that were missing pertinent Compustat data, the final sample consisted of 54 ESOP firms and 69 prolitsharing firms, for a total of 123 firms each observed for a period of 7 years. Other retirement plans were constant on a within-firm basis throughout the sample period. The motivation for constant plans is to ensure that the adoption or termination of another plan does not affect the estimation of an ESOP or profit-sharing effect. Essentially the adopters sample constitutes a within-firms design with each firm serving as its own control, allowing the estimation of a productivity effect relative to a non-plan period for the same firm. Table 1 provides the industry composition of the Annual firms and the ‘Compustat provides a maximum of 20 years of annual data for 2,452 firms. Firms that were added to Compustat subsequent to 1981 were not considered because the panel data procedures require observations in every year, 1981-1987. Of the 1,764 firms that were listed on Compustat since 1981, 1,390 firms were matched with at least one DOL tiling.

280

S.C. Kumbhakar and A.E. Dunbar, ESOP-productivity link

sample. To determine if the sample is representative of the overall population of Annual firms, the ratio of total firms in each industry to the total number of firms should be compared with the ratio of sample firms in each industry and with the total firms in the sample. For example, consider the manufacturing industry: 46.3 percent of the Annual firms compared with 64.8 percent of the ESOP Annual firms and 43.5 percent of the profit-sharing Annual firms are in the manufacturing industry. This study represents the dependent variable, output, by net sales. The independent variable, capital, is represented by either gross or net property, plant, and equipment (GP or NP). Because of the difficulty of determining which representation is most appropriate, both models are estimated twice, once with each of the two variables for capital. The independent variable, labor, is represented by the number of employees (EMP). This is the only publicly available information regarding this variable for purposes of estimating production functions at the firm level across industries.

5. Estimation

and empirical results

Models I and II are estimated under both the FE and the RE approaches. The RE model assumes pi and Izi to be random variables with zero means and constant variances independent of vit and other regressors in the model.8 Because only the variances of pi, A,, a:, and c: are estimated, the RE model has the advantage of estimating a smaller number of parameters when compared with the FE model. But the advantage of the FE model is that ,U~and II, are not assumed to be independent of the regressors. The computational procedure for estimating the parameters in the FE model is straightforward. When considering both the firm- and time-specific effects, it is necessary to find the means of the time-series observations separately for each firm, cross-section means for each year as well as the overall mean. Each variable is then transformed by subtracting the appropriate time-series mean, cross-section means, and adding the overall mean. Finally, the ordinary least squares (OLS) method is applied to the transformed data. Estimates of the firm- and time-specific components (pi and 2,) are then recovered from the residual means over time for each firm and across firms for each year, respectively. Alternatively, one could use (N- 1) firm and (T- 1) year dummies together with an intercept and the appropriate regressors in the OLS regression. There are several methods that can be used to estimate the random effects model. We used the SAS procedure, ‘PROC TSCSREG’, which is based on the Fuller and Battese (1974) procedure. Each model is estimated using two ‘This model is also referred to as an error components model because sum of either two or three random components of variation.

the residual

error

is the

SC. Kumbhakar and A.E. Dunbar, ESOP-productivity link

281

different measures of capital, namely gross property, plant, and equipment (GP) and net property, plant, and equipment (NP). These alternative capital specifications check the robustness of the ESOP effects. The results from Model II (after deleting the interaction terms with tvalues less than unity) are reported in table 2.9 Several important features are revealed. First, the capital and labor elasticities changed over time. Capital elasticities (cc,+cc,,D,) declined significantly during 1986 and 1987. The same is true for labor elasticities (cr,+cr,,D,), which declined up to 1986 and then increased in 1987. The Cobb-Douglas models with constant capital and labor elasticities over time are rejected in both the FE and the RE specifications. Second, the coefficient on the ESOP --age and PS- age variables are positive and statistically significant in all models. This finding suggests that (1) both ESOP and profit-sharing plans increased productivity, (2) the productivity effects increased with the age of the plans, and (3), although the impact of the ESOP and profit-sharing plans on the growth rate of output varies from the FE to the RE model, the overall conclusion is that each of these plans affected productivity positively. The rate of output growth for profit-sharing firms is about 2 percent higher than that of ESOP firms, everything else being the same. This result is in agreement with some of the previous studies. It can be seen from the FE models in table 2 that the coefficient estimates are very robust to the choice of alternative measures of capital (GP vs. NP) except in their own coefficients. The growth rate of output increased at the rate of 1.8 and 1.9 percent per annum with an increase in the age of the ESOP when capital is measured by GP and NP, respectively. For firms with profit-sharing plans, the output growth rate is 3.9 and 3.8 respective to GP and NP as measures of capital. The share of labor declined from 59 percent in 1981 to 56 percent in 1986 and then increased to about 63 percent in 1987. On the other hand, the share of capital declined substantially in 1986 and 1987. The results are similar in the RE models. Both the ESOP-age and PS-age effects are somewhat larger in the RE models compared with their FE counterparts. Labor elasticities are generally the same, but the capital elasticities are slightly larger compared with their respective values in the RE models. In summing up the results from the four models in table 2, we conclude that there is a positive and significant effect of ESOP and PS no matter which model is chosen. 6. Conclusion This

paper

9To economize

addresses

the

issue

space we are not reporting

of whether

employee

the results from Model I.

participation

in

S.C. Kumbhakar and A.E. Dunbar, ESOP-productivity link

282

Table 2 Model II: Fixed and random effects. Parameter estimates of firm productivity model. Dependent variable = In (net sales). Independent variables In(GP)

Fixed effects’

Random

0.191 (0.026)

In(NP)

effects

0.239 (0.022) 0.167 (0.022)

0.204 (0.019)

In(EMP)

0.592 (0.032)

0.595 (0.032)

0.600 (0.027)

0.616 (0.026)

ESOP - age

0.018 (0.009) 0.039 (0.010)

0.019 (0.008)

0.024 (0.008)

0.038 (0.010)

0.046 (0.010)

0.027 (0.009) 0.046 (0.010)

PS - age In(GP)*D6

-0.041 (0.008)

- 0.035 (0.008)

In(GP)*D7

- 0.070 (0.01 I)

-0.061 (0.011)

In(NP)*D6

- 0.038 (0.008)

- 0.033 (0.008)

In(NP)*D7

- 0.060 (0.010)

-0.053 (0.010)

In(EMP)*D2

-0.014 (0.010)

-0.012 (0.010)

-0.012 (0.010)

-0.010 (0.010)

In(EMP)*D3

- 0.025 (0.010)

- 0.022 (0.010)

-0.021 (0.010)

-0.018 (0.010)

In(EMP)*D4

-0.033 (0.011)

- 0.032 (0.010)

- 0.028 (0.010)

- 0.026 (0.010)

ln(9MP)*D5

-0.033 (0.011)

- 0.032 (0.011)

- 0.029 (0.010)

- 0.028 (0.010)

In(EMP)*D7

0.041 (0.014)

0.032 (0.014)

0.038 (0.014)

0.032 (0.014)

0.255

0.250

0.03 1

0.029

0.019

0.019

2 UP c: 2 cv

0.019

0.019

N = 123, T = 7 for all models. Standard errors in parentheses. “These models have fixed firm- and time-specific effects which estimated using (N - 1) firm and (T - I) time dummies.

are

ownership or profits in publicly held firms through an ESOP or protitsharing plan was positively associated with productivity measures. The empirical evidence from both the fixed and random effects models support the hypothesis that there is a positive and statistically significant relationship between the presence of an ESOP or a profit-sharing plan and the productivity of a firm. The productivity effect increased with the age of the

XC. Kumbhakar and A.E. Dunbar, ESOP-productivity link

283

plan. Thus, the results uniformly support the hypothesis of a positive productivity effect. The results of this research do not support the arguments of those who predict that participation in ownership and profits leads to inefficiencies. Instead, the contention of those who predict that participation in ownership and profits leads to efftciency gains was uniformly supported. References Akerlof, G. and J. Yellen, 1986, Efficiency wage models of the labor market (Cambridge University Press, Cambridge). Blinder, A., 1989, Want to boost productivity? Try giving workers a say, Business Week, 17 April, p. 10. Blinder, A., 1990, Paying for productivity (The Brookings Institution, Washington, DC). Bloom, S., 1985, Employee ownership and firm performance, Doctoral Dissertation presented to the Department of Economics, Harvard University, Cambridge, MA. Conte, M. and J. Svejnar, 1988, Productivity effects of worker participation in management, profit-sharing, worker ownership of assets and unionization in U.S. firms, International Journal of Industrial Organization, 139-151. Conte, M. and J. Svejnar, 1990, The performance effects of employee stock ownership plans, in: A. Blinder, ed., Paying for productivity (The Brookings Institution, Washington, DC) 143-181. Fuller, W.A. and G.E. Battese, 1974, Estimation of linear models with crossed-error structure, Journal of Econometrics 2, 67-78. General Accounting Office (GAO), 1986, Employee stock ownership plans: Benefits and costs of tax incentives for expanding stock ownership (US General Accounting Office, Washington, DC). General Accounting Ollice (GAO), 1987, Employee stock ownership plans: Little evidence of effects on corporate performance (US General Accounting Offtce, Washington, DC). Kruse, D., 1989, Profit-sharing and productivity: Microeconomic evidence, Mirneo. Wadhwani, S. and M. Wall, 1990, The effects of profit-sharing on employment, wages, stock returns and productivity: Evidence from UK micro-data, Economic Journal 100, 1-17. Weitzman, M. and D. Kruse, 1990, Profit sharing and productivity, in: A. Blinder, ed., Paying for productivity (The Brookings Institution, Washington, DC) 95-141.