Parity codetermination at the board level and labor investment efficiency: evidence on German listed firms

The German law on codetermination requires the firm to have 50% employee representatives on the supervisory board once the firm exceeds the threshold of 2000 domestic employees.

Hubbard (1998) and Stein (2003) provide surveys of this literature.

Studies on capital investment and market frictions are presented by Hubbard (1998) and Stein (2003).

There is evidence to suggest that information-sharing is positively related to the level of wages and benefits in both union and non-union businesses, but negatively related to profits and cash flows in non-union businesses (Kleiner and Bouillon 1988).

Employees or unions may increase unilateral advantages by, e.g., demanding higher wages and holding the level of employment constant, even if this endangers the firm’s economic position (Lazear and Freeman 1995).

To proxy for stock price informativeness, Ben-Nasr and Alshwer (2016) use the Probability of Information-based Trading (PIN) derived from Easley et al. (1996) market microstructure model.

Future research should seek to broaden our understanding of the determinants and effects of labor investment efficiency for multinational and non-U.S. samples.

For an overview of existing or non-existing codetermination at board level rights please refer to European Trade Union Institute’s website (ETUI 2014) or to Lopatta et al. (2018).

Some studies even suggest that one third participation is not strictly enforced in German firms (Boneberg 2009, 2010, 2016; Frankfurter Allgemeine Zeitung 2016). This may be because one-third participation applies only to smaller firms, which are often family-owned. These owners may refrain from sharing their powers with people outside their family (e.g., employee representatives) even if this jeopardizes firm performance (Cucculelli and Micucci 2008). To the contrary, other researchers argue that the share of firms with 500–2000 employees without one-third codetermination may also be partly due to reasons other than enforcement (for a discussion of these reasons see Dilger (2009)). Non-compliance with the legislation is therefore more likely to be detected for parity codetermination than for one-third codetermination. In order to reduce the risk of self-selection bias, we focus on parity codetermination, which we know from hand-collection, is enforced without a considerable delay.

In case no majority is achieved, labor representatives elect the deputy chair while shareholder representatives elect the chair (Section 27 (2) of the law on codetermination).

Employees who serve on the supervisory board can elect the members of the management board for at most 5 years, with the possibility of re-election [Section 84 (1) of the Stock Corporation Act (Aktiengesetz or AktG)].

The rather long-term orientation of employee representatives and the rather fixed compensation of employees may partly explain why not all previous studies have found a beneficial effect of labor representation. In line with Lin et al. (2018), we argue that employees have a different utility function than management and shareholders. Whereas management and shareholders rather symmetrically benefit from performance increases and decreases, employees merely benefit from performance increases and suffer directly from performance decreases due to job losses and restructuring programs. Hence, whereas management and shareholders would initiate most if not all projects with a positive net present value (also high-risk projects, as the expected returns are correspondingly high), it is plausible that employee representatives are not willing to initiate all positive (high-risk) net present value projects. At worse, this could harm firm value (and consequently firm performance, e.g., through negative market returns), as projects with positive net present value would potentially not be initiated. Given these considerations, we believe it is plausible to assume that the heterogeneity in the results on firm performance are rather due to different risk-return functions (symmetric vs. asymmetric relation) of managers and shareholders versus employees. In our setting, these varying risk-return considerations should not affect our results as we implicitly look at the effect of co-determination on labor investment efficiency for given performance levels (see Sect. 3).

This may be beneficial to the extent that firms in highly unionized industries are characterized by higher levels of information asymmetry and greater bargaining power on the part of unions (Frost 2000; Kleiner and Bouillon 1988; Scott 1994).

Both parity and one-third codetermination should have an effect on a firm’s information environment. However, we focus in our empirical setting only on parity codetermination and do not estimate effects for one-third codetermination as the latter would potentially cause severe methodological issues. The adaption of parity codetermination is an external effect that triggers the voice of labor due to legislation, which enables us to reduce the risk of self-selection bias. One may argue that missing enforcement sanctions are an issue. However, after hand-collection, we know that firms in our sample change to parity codetermination without a considerable delay. We believe that this is mainly reputation/attention-driven: larger and especially listed firms are more often in the spotlight of the media and the public. Non-compliance with the legislation is therefore more likely to be detected for parity codetermination than for one-third codetermination. Yet, the enforcement of the one-third codetermination around the threshold of 500 domestic employees is somehow questionable due to other issues that have been detected in prior research (see Sect. 2.2). Such non-compliance with codetermination legislation would ultimately result in a self-selection bias. Hence, by focusing on parity codetermination we test for the effect of firms switching from no or one-third codetermination to parity codetermination. We assume that the beneficial effect of parity codetermination should be more pronounced for firms switching from no to parity codetermination. However, we also believe that that the beneficial effects of a labor voice at board level (e.g., reduced agency conflicts and lower information asymmetries) should still exist for firms switching from one-third to parity codetermination, albeit be less pronounced. One beneficial effect of a switch from one-third to parity codetermination could be that the employees’ voice becomes stronger in boards discussions as they no longer constitute a minority. By contrast, focusing only on firms that switch from none to parity codetermination is not feasible due to a very small sample size. Given these arguments, we follow prior studies that also focus on parity codetermination and the related threshold of 2000 employees (Lin et al. 2018; Lopatta et al. 2018).

For instance, if labor resources are scarce (and hence more expensive) in a given year and industry, firms would be expected to hire fewer new employees as a reaction to sales growth.

For more details about why our control variables are expected to affect our dependent variable, we refer to Pinnuck and Lillis (2007), p. 1040. As an additional sensitivity test, we also include profit bins.

We test whether the loss bins are jointly significant. The (untabulated) results are similar.

Like U.S. firms, German firms are not required to publicly disclose any information about the union membership of their employees. In contrast to U.S. studies, Germany does not provide ratings about the unionization level per industry. However, using variables from the ASSET4 database allows us to control for firms’ union relations policies, which shows whether the management promotes a relationship between trade unions and employees.

The Hans Boeckler Foundation was established in 1954 and is located in Düsseldorf. It deals with co-determination and labor research and, on behalf of the Confederation of German Trade Unions, supports students (Hans-Böckler-Stiftung 2017).

To ensure correctness, we checked whether the firms actually (do not) comply with parity codetermination requirements for a random sample of 100 firm-years and did not find any false information.

The firms are assigned to the main Fama–French industry code (12 industries) according to their primary activities. We assign these industries using the industry indicators provided in COMPUSTAT GLOBAL.

For sensitivity, we rerun our regression Eq. (1) without SALES_GROWTH_it and ROA_it in time t. The effect of size is negative and significant at the 5% level, indicating firms that increase in SIZE have a lower percentage change in employees

Based on the expected values of net hire, we have split the sample into positive and negative deviations from the optimal value (zero). As we do not interpret values close to zero as over- or underinvestment, we do not include values close to zero to any of the two subsamples (defined as the absolute values lower than 1% of the median of our first-stage residuals).

For a more detailed overview of the German setting, refer to Sect. 2.2.

We do not control for the presence of works councils in our empirical model, as we did not find any sample firms without a works council, which is in line with the data used in similar studies (Lin et al. 2018).

We hand-collect data on the existence of collective bargaining agreements between unions and firms (hereafter labor agreements (German: “Tarifverträge”) and the name of the union from annual reports.

We also control for union strength as we expect larger unions to have a stronger negotiation position due to greater expertise and a stronger threat-potential. We posit that the strength of a union is positively correlated with size.

For further details, refer to the variable definition of STRONG_UNION_it in Appendix.

In another test, we also include STRONG_UNION_it and LABOR_AGREEMENT_it in Eq. (1) and use the residuals of the adjusted equation for the measurement of labor investment efficiency. When rerunning Eq. (2), the results remain significant and similar to our previous tests (results untabulated). For comprehensiveness, we do not include all tables for our sensitivity analyses in this paper. However, the untabulated results of all modifications listed in this section are available upon request.

First, we exclude SALES_GROWTH_it. This mitigates the risk that our significant results are biased by redundant control variables for sales growth, size and the growth effect in Eq. (1). In a second regression analysis, we include variables that control for the firms’ size in terms of the number of employees, namely LABOR_INTENSITY_it and EMP_SIZE_it in Eq. (1). Third, for taking the risk that our size variable drives the results in Eq. (2) into account, we replace SIZE_it with various alternative measurements such as percentile ranks of SIZE_it (hereafter: SIZE_R_it), or/and EMP_SIZE_it as well as SALES_GROWTH_it.

We modify the matching procedure in two ways. First, we extend the covariates by variables that reflect the firms’ average growth in sales and labor intensity. Second, we also include the third last year before the treatment effect. The results remain not only robust, but also very similar in terms of coefficients’ magnitudes and significances (untabulated).

In line with Lopatta et al. (2018), our matching approach ensures that each treated firm is paired with the nearest neighbor in the control group. Technically, we sort the list of control firms (Control Group I) in descending order by distance (in the propensity score) to the treated firm’s outcome in the propensity score that is based on the last 2 years’ average of LEV_it, ROA_it, SIZE_it. We keep only the first two firm-year observations to form the final matching pair.

For instance, larger firms may have more difficulties to find their optimal net hire level as their business operations may be more complex than those of smaller firms.

Similar patterns emerge for a second-order polynomial fit (quadratic function) and bin sizes of 100 employees.

Given that parity codetermination is strictly regulated in Germany, we assume that non-compliance with the parity codetermination rules by firms with more than 2000 employees is rather unlikely. We find no anecdotal or empirical evidence either in the academic literature or in the business press that firms voluntarily apply parity codetermination. However, we find very few observations where the firm has fewer than 2000 domestic employees and applies parity codetermination, which may be, for instance, due to time delays in the data provided. In these instances, our indicator variable for parity codetermination equals 1. Given these few exceptions, we cannot completely rule out that self-selection affects our results significantly although we believe it is rather unlikely. To exercise a high level of caution, we also run RDDs with a dummy if a firm is above the threshold of 2000 and polynomials of up to four dimensions. The results remain qualitatively unchanged.