Identification

Certain issues addressed in corporate governance codes in some jurisdictions may be regulated in binding laws in others. Comparing the regulatory responses to such issues would not render particularly meaningful results since no conclusions could be drawn from the fact that a country did not emulate a best practice standard. It is therefore important to focus on mechanisms that will typically be insulated from the codified corporate law of a country and dealt with predominantly by soft law. In addition, the relevant issues should lend themselves to prescriptive, detailed regulation, as opposed to regulation by means of general guidelines or broad standards, in order to be amenable to coding.

Arguably, both conditions are satisfied by rules that structure the board of directors of listed companies. While basic tenets of board structure regulation, notably the distinction between one-tier and two-tier boards, are generally laid down in binding legislation, the detailed aspects of board structure, such as the number of independent directors or the requirement to establish board committees, are commonly not found in the codified law. These requirements depend on changing characteristics of the company and the market for non-executive directors. To give just a few examples, a small- or medium-sized company may find it unnecessarily cumbersome and inefficient to establish separate nomination, remuneration and audit committees. In smaller economies, the market for non-executive directors may not be sufficiently deep to allow companies to find qualified individuals meeting a long list of independence requirements. In some companies, it may be particularly important to retain firm-specific expertise and hence allow retiring executives to continue to serve in a non-executive capacity on the board of directors or supervisory board. The comply-or-explain principle of corporate governance codes is more appropriate to regulate these issues than binding law. Indeed, board structure regulation is a centerpiece of all codes analyzed in the sample, and often the relevant recommendations contain detailed prescriptions of how boards should be structured. Other issues addressed by corporate governance codes are either less well insulated from the binding company law or formulated in open-ended terms less amendable to quantification. For example, executive remuneration is increasingly regulated in binding acts, which impose extensive disclosure obligations and more recently also shareholder approval requirements, and provisions dealing with shareholder engagement or internal control commonly only establish general guidelines.

It is widely accepted in corporate governance theory (and also by some policy makers) that effective boards should be structured so as to contain a strong non-executive and independent element on the board, avoid a concentration of power at the helm of the company by requiring that the roles of chairman and CEO be separated, and delegate issues that involve particularly pronounced conflicts of interest to independent committees, namely succession planning, responsibility for the review of internal control procedures and the appointment of the external auditor, and remuneration decisions (see, e.g., Mallin, Reference Mallin2015, and Commission Recommendation 2005/162/EC). These considerations of what constitutes good governance have been translated into six elements of board structure regulation: (1) the requirement that at least half of the board shall consist of non-executive directors; (2) a majority of the non-executive directors must be independent; (3) independence is defined prescriptively in the form of an enumeration of factors that must be satisfied, most importantly not having been an executive director or employee of the company in a senior management position for a specified number of years, receiving additional remuneration from the company, representing a major shareholder, or having a significant business relationship or family ties with the company; (4) the roles of CEO and chairman shall not be exercised by the same individual; (5) the CEO shall not become chairman upon retirement as chief executive (cooling-off period); and (6) companies shall establish nomination, audit and remuneration committees composed of a majority of independent non-executive directors (Table 1.A).

These best practice standards form the six variables that measure changes in regulation and determine the value of a second set of variables, called ‘diffusion variables’, which count how many standard setters have adopted a similar formulation of each of the six standards (Table 1.B). The variables are coded by assigning a value between 0 and 1, as set out in Table 1.A, with higher values representing a comparatively more stringent formulation (from the perspective of the executive directors or corporate insiders). A code provision is ‘similar’ to the definition of the best practice standard used for purposes of comparison (the most stringent alternative of the definitions given in Table 1.A), if the value of the respective variable is equal to or greater than 0.75. The dichotomous indictor variable ‘adoption of benchmark rule’ takes the value 1 if a standard setter adopts a provision similar to the definition, and the dichotomous indictor ‘regulatory innovation’ takes the value 1 if a standard setter changes the code so that the provision is more similar to the comparator definition than before. These dichotomous indicators are used as dependent variables in the empirical models in Section 4.

While the regulatory variables chosen here operate generally independently of the corporate law in place in the respective jurisdiction, certain interdependencies between soft law and the binding legislative framework are unavoidable. Most importantly, it is clear that the six elements of board structure regulation are not equally applicable to one-tier and two-tier boards. In jurisdictions with a two-tier board structure, executive and non-executive directors are by definition separated and the board ultimately responsible for monitoring, the supervisory board, consists entirely of non-executive directors, who may or may not be independent. Similarly, the positions of chairman of the supervisory board and chairman of the executive board (CEO in Anglo-American parlance) cannot be exercised by the same individual. Thus, elements (1) and (4) are omitted from the analysis if the legal system employs the two-tier board model. The coding protocol contains further explanations, including references to statutory sources where appropriate.Footnote ⁴ Panel A of Table 2 reports summary statistics. In the first row, ‘board structure’ is based on the mean of the scores for the six (or four, as applicable) elements of board structure received by the codes in the sample. The other rows report summary statistics for the six individual elements of board structure as well as the ownership variables, Panel B correlation coefficients, and Panel C mean values of the regulatory variables broken down by country.

Reductionism

The second methodological challenge, here called the ‘reductionism problem’, needs to be addressed by all legal research using quantitative methods. Norms are often not phrased in a binary way, but may provide for a catalogue of exceptions or allow market actors to modify or disapply the norm contractually. In addition, a norm may be susceptible to more than one interpretation, and a universally accepted interpretation ensuring legal certainty may not exist. A binary or otherwise not sufficiently nuanced coding system could not capture these aspects and would, therefore, be prone to measurement error and lead to a loss of information. On the other hand, a more gradual quantification necessarily requires judgments to be made that will, to some extent, be conditional on the interpretation found most convincing. Scholars generally do not go so far as to conclude that quantitative techniques are inappropriate in legal scholarship because of these reasons, but they caution against oversimplification and the blind faith in the comparability of any aspect of a legal system (Epstein and Martin, Reference Epstein, Martin, Cane and Kritzer2010; Michaels, Reference Michaels2009; Siems, Reference Siems2005b). The key, it seems, is to be aware of the trade-off between accuracy and objectivity inevitably involved in the quantification of social phenomena that are open to interpretation and rational disagreement about their meaning (Gompers et al., Reference Gompers, Ishii and Metrick2003).

In order to make allowance for this trade-off, I propose a two-step approach to coding. The first step consists in a mechanical application of a simple metric that is usually composed of not more than three levels:Footnote ⁵ a lower bound reflecting the absence of any meaningful regulation of the relevant issue (in which case the variable equals zero), an upper bound reflecting the most stringent regulation practically relevant (the variable equals one), and one intermediate stage reflecting best practice standards that impose meaningful constraints but fall short of the most stringent formulations available (the variable equals 0.5). In this first step, any ambivalence in the formulation of the best practice standard, exemptions or the possibility to disapply the rule, for example, upon a motion by shareholders, are disregarded.

In a second step, the value derived in the first step is adjusted if the mechanical quantification does not accurately reflect the true import of the rule. The adjustment is confined to two scenarios that relate to the strictness and precision of the provision. The value is either adjusted if the rule is formulated in an optional way and provides for exemptions or allows non-compliance outside the comply-or-explain principle, i.e. the company is in full compliance with the code, and accordingly is not required to publish an explanation, although the provision is not applied, or the provision is phrased so ambivalently or in such general terms that precise behavioral prescriptions cannot be derived from it. In both cases, the first-step value is adjusted as explained, for each case of adjustment, in the coding protocol.Footnote ⁶

It may be useful to explain the operation of the two-step analysis with the help of an example. I take element (4), separation of CEO and chairman of the board, and analyze how the UK Corporate Governance Code of 2012 and the Italian Corporate Governance Code of 2011 correspond to the metric developed here. The two code provisions are as follows:

The UK code provides that there should be a clear division of responsibilities at the head of the company and the roles of chairman and chief executive should not be exercised by the same individual. This falls squarely within the upper bound of the definition of element (4). The provision is unambiguous and does not contain any qualifications or exemptions. Therefore, the UK Corporate Governance Code receives the value 1 for ‘separation chairman/CEO’.

The Italian code is different in several respects. First, it is phrased in more ambiguous terms. The concentration of the roles of chairman and CEO is not prohibited, but the code holds that it is ‘appropriate’ to avoid it. Code Principle 2.P.5 and Criteria 2.C.3 envisage a situation where the company decides not to separate the two offices. However, they intend to ensure transparency and the existence of an appropriate counterweight in the form of a ‘lead independent director’ in case the same person acts as CEO and chairman. Finally, the comments, which are included in the code to clarify the principles and criteria, explicitly acknowledge that the accumulation of the roles of chairman and CEO may be beneficial and satisfy ‘valuable organizational requirements’. Is this a prohibition of the concentration of corporate offices that warrants quantifying the Italian code pursuant to the upper bound of the definition of element (4) (the roles of CEO and chairman must not be exercised by the same individual)? Arguably, it is a prohibition because companies must give reasons in the corporate governance report if they concentrate the offices, which is in line with the essence of corporate governance codes. However, the quality of the prohibition is different from that of the UK code. Clearly, this is a difference that cannot be captured by a simple binary metric and mechanical coding. At this point, the second, qualitative stage of coding becomes relevant.

The Italian code is an example for an ambivalent rule. It merely speaks of the ‘appropriateness’ to separate the roles of chairman and CEO, but does not establish a clear precept by means of a deontic statement, and acknowledges the usefulness of combining the roles of chairman and CEO. Furthermore, it makes provision for the delegation of executive powers to the chairman within the framework of the code, i.e. the company is technically in compliance with the code if the offices are combined, even though transparency is required by virtue of the code.Footnote ⁷ Therefore, it would be inappropriate to give the Italian code the same value as a code that prohibits the accumulation of the roles strictly and unambiguously, as the UK code. On the other hand, the code provides for two substitute mechanisms if management powers have been delegated to the chairman. The board has to disclose the reasons for this organizational choice in the corporate governance report, and it shall designate an independent director as lead independent director, who ‘represents a reference and coordination point for the requests and contributions of non-executive directors’.Footnote ⁸ Thus, the Italian code is also not comparable to corporate governance codes that fail to impose clear and unconditional precepts and do not provide for substitute mechanisms, for example, codes that merely ‘encourage’ the separation of offices.Footnote ⁹ It stands between these two archetypical provisions – the UK rule and the ‘encouragement’ formulation – and should accordingly receive a penalty reflecting this intermediate position.Footnote ¹⁰ This example shows that the coding even of provisions pursuing supposedly simple regulatory goals can involve difficult questions of interpretation. Only a detailed and transparent interpretation of norms, following commonly accepted canons of construction, can resolve such ambiguities.

Model specification

Proportional hazard models are a widely used method to estimate the rate of diffusion of innovations as a function of a set of predictors, including the diffusion of ‘regulatory’ innovations. For example, Ramirez et al. (Reference Ramirez, Soysal and Shanahan1997) apply event history analysis to the spread of women's suffrage rights and Wotipka and Ramirez (Reference Wotipka, Ramirez, Simmons, Dobbin and Garrett2008) to the ratification of international treaties. This article is concerned with a similar diffusion phenomenon, the spread of corporate governance institutions over time among standard setters based in different countries. An appropriate event history method is the Cox regression model, which has the advantage of not requiring any distributional assumptions, only the assumption that the effect of predictors, in principle, is constant over analysis time (proportionality assumption). A Cox regression estimates the function:

$$\begin{equation*} h\left({{t_{ij}}} \right) = {h_0}\left({{t_j}} \right){e^{\beta {X_{ij}}}}, \end{equation*}$$

where h(t_ij ) is the hazard that individual i will experience the event at time t_j ; h ₀(t_j ) denotes the baseline hazard, i.e. the hazard when all predictors are 0; and X_ij is a vector of covariates, which are allowed to vary between individuals and over time. Here, in order to test the three hypotheses formulated above, three sets of predictors are included as covariates: diffusion, ownership and legal origins variables. The diffusion variables were described in Section 3.1 above. I have ownership data on shareholdings by foreign investors and domestic institutional investors as a percentage of total equity of listed companies for most country/year combinations when corporate governance codes were adopted. Historical ownership data on the size of the largest voting blocks is fragmentary and not always comparable across countries. Therefore, in order to measure ownership concentration, I include a dummy variable that equals 1 if the median largest voting block is equal to or larger than the median value of all countries in the sample according to the most contemporaneous data available. Legal origins distinguish between the four legal families common law, German, French and Scandinavian legal origins. As additional controls, I create dummy variables that distinguish between one-tier and two-tier board systems, jurisdictions that require employee representation at board level, and the adoption of codes before and after two major corporate governance reforms entered into force that could have had an impact on the activities of standard setters: the Sarbanes–Oxley Act in the United States in 2002 and Commission Recommendation 2005/162/EC in the EU in 2006.Footnote ¹¹

The relevant event is defined in two alternative ways. In the first specification of the model, it is the adoption of a rule that corresponds to the benchmark formulation of the six elements of board structure given in Table 1.A. Since it is quite common that corporate governance codes are ambivalent or in other ways fall short of the formulation of the six components of board structure that is used for coding, the dichotomous indicator scores one if the code provision takes a value of at least 0.75. Once the event occurs, the code issuer no longer contributes to the risk set. In the second specification, subjects can experience recurrent events, which are defined, for each component of board structure, as an increase in the value of the respective variable compared to the previous code adopted by the same issuer.Footnote ¹² In both models, code issuers were right censored if they had not experienced the event by the end of 2012.Footnote ¹³

A problem frequently discussed in the diffusion literature is the interdependence between different actors that consider whether to adopt an innovation (so-called Galton's problem). Unless it can be assumed that the actors are influenced solely by factors that relate to the inherent characteristics of the innovation and the problem it is intended to address, rather than the decisions of others (Braun and Gilardi, Reference Braun and Gilardi2006), models need to take account of possible relationships of mutual influence between actors. In statistical terms, failure to do so might result in an omitted variable bias. Jahn (Reference Jahn2006) gives an overview of the solutions to this problem that have been developed in the literature. Here, we are particularly interested in the reaction of standard setters to the degree of international diffusion that a best practice rule has already achieved. In other words, the goal is to examine specifically the effect of a factor that is not related to corporate governance problems faced by the legal system in question, but captures the process of diffusion itself. Therefore, I model the hazard rate of adoption of a best practice rule as a function of prior adoptions of the same rule by other standard setters.

Two further considerations are important for a correct specification of the model. It is possible that the influence of the international spread of corporate governance innovations is not linear. For example, the probability that a corporate governance committee adopts a foreign regulatory innovation may increase initially as the innovation gains international acceptance. However, once it has become widely known because it has been adopted by a sufficient number of countries, say six or seven out of the sample of 23, the marginal effect of each additional adoption may be decreasing. The effect of diffusion would therefore resemble the well-known S-shaped form of other diffusion processes (Rogers, Reference Rogers2003). Intuitively, it could be said that the corporate governance committee has had the opportunity to consider the adoption of the rule and decided that it was not suitable, given the country's economic situation, legal system, or other reasons. In that case, it is maybe still possible, but less likely that the further spread of the regulatory innovation will prompt the committee to change its mind.

Table 2.

Summary statistics

Similarly, it is possible that the effect of the international spread of a regulatory innovation is not constant over time. Once an innovation has been pioneered by one country, it can be expected that corporate governance committees in other countries, especially those belonging to the EU or EEA and, consequently, being in fairly constant exchange, discuss the innovation and decide relatively quickly whether to adopt it or not. If this is correct, we would expect to see the adoption of regulatory innovations, once they have been promulgated by the forerunners in the corporate governance movement, by committees set up in other countries during the first years of their existence. As time passes after the establishment of the committee, it may become less likely that they will revisit well-known questions and adopt corporate governance institutions with a considerable delay. Thus, the impact of the diffusion of regulatory innovations may decrease over time while a committee is in existence and deals with rule changes.

In order to investigate these possibilities, I compute the martingale residuals for a null model to examine the functional form of the continuous predictors. The residuals indicate that the diffusion predictors are not linear, whereas a plot of the residuals against the other continuous predictors does not show any discernible pattern. Second, I test the proportionality assumption both graphically by examining the scaled Schoenfeld residuals and log cumulative hazard functions, and by regressing the Schoenfeld residuals on time. I do not reject the proportionality assumption, except in one model: the effect of the diffusion of the benchmark definition of independence is not constant, but rather decreases over time. These findings suggest that a quadratic term, ‘diffusion squared’, needs to be included in the equations assessing benchmark rules regarding the number of non-executive directors, independent directors, the separation of chairman and CEO, and board committees, and an interaction term with time in the equation dealing with the diffusion of the benchmark formulation of independence.Footnote ¹⁴