Product Market Competition, Executive Compensation, and CEO Family Ties

See among others Giroud and Mueller (2010) for the relationship between market competition and firm governance, or Bena and Xu (2017) for the relationship between competition and firm value.

Burkart et al. (2003) even envisage the risk that “over-monitoring” by the large shareholders may undermine the external manager’s initiative and effort.

For a thorough analysis of parental altruism and its consequences on corporate governance and firm value, see Lubatkin et al. (2005) and Karra et al. (2006).

Although the ownership stake is a large fraction of the family CEO’s wealth, the relative illiquidity of his shares makes the cash component of performance-related pay particularly incentivizing (McConaughy 2000).

CONSOB annual relations various issues (2017, p. 85).

At the end of 2016, 92% of the firms had adopted the last version of the Code. See, Assonime, “Corporate governance in Italy”, Note e Studi 18/2016.

CONSOB, Communication of February 24th, 2011.

Even though the application of the IFRS2 for the disclosure of executive stock-based compensation dates back to mid-2000s’ in the company reports, it was impossible to find these data consistently over time and across firms. Too many missing data prevented the construction of reliable time series in those years.

As a comparison, in 2017 imports and exports in the United States were 15.0% and 12.1%, respectively. World Bank national account, import and export data are available at https://​data.​worldbank.​org/​indicator/​ne.​imp.​gnfs.​zs and https://​data.​worldbank.​org/​indicator/​ne.​exp.​gnfs.​zs .

We also used the following alternative definitions of the dependent variable: (1) the log of total compensation; (2) the ratio of total compensation to total assets; and (3) the log of the ratio of total pay to total assets (regressed on the log of firm performance). All of the regressions provide similar results (which are available on request), in line with those that are obtained when we use the variable share of pay as a dependent variable.

Sensitivity is typically used to define the general relationship between managerial compensation and performance (when the relationship is between the log of pay and the log of performance it is called the elasticity). In our case, the relationship is between the variable component of the pay and performance, and we use the PPS expression to highlight the idea that managerial compensation is designed to respond to changes in firm performance.

Ideally, one would like to exploit a natural experiment or a shock to the competitive environment such as a sudden appreciation of the currency or an unexpected reduction in trade barriers (Cunat and Guadalupe 2005, 2009). Unfortunately, a shock such as this is not available for Italy from 2000 to 2017, because trade liberalization towards China and other Asian countries was a gradual process rather than a foreign trade shock. For this reason, we classify different competitive environments by using one-time, out-of-sample industry level variables (Type 1/Type 2, High/Low Imp-Pen, etc.). In fact, time-varying variables would raise simultaneity concerns because of the parallel evolution of competitive conditions and compensation policies as changes in import penetration or R&D or advertising intensity in the industry ultimately derive from firm-level decisions, which are, in turn, the firms’ and CEOs’ responses to continuously changing conditions.

In the Appendix, we report the correlation matrix in Table 16 and the Variance Inflator Factor (specifying individual and average values) in Table 15.

The dependent variable is bounded from 0 to 1. As a robustness check, we performed three different sets of regression for the main specification that investigate the differences between Type 1 and Type 2, High and Low Import penetration, and high and low domestic) concentration, using econometric methods that take into account the bounded nature of the dependent variable: (1) We estimated Tobit regressions; (2) we used the fractional response (probit) regression model (Papke and Wooldridge 1996), which is especially designed for models where the dependent variable is greater than or equal to 0 and less than or equal to 1, and relies on quasi-likelihood estimators; and iii) we used the transformation log[varshapay_cash/(1-varshapay_cash)] as a dependent variable, adding 0.001 to varshapay_cash when the variable was zero. The results were found to be always in line with those obtained using the fixed effects model: quantitatively and qualitatively. We thank one referee for suggesting these robustness tests.

CONSOB Regulation n. 11971, May 14, 1999.

Note that some firms changed ownership status over the sample period, so they appear both in the family and in the non-family subsamples, pro-rata.

We draw the data from multiple sources: Accounting and financial data are collected from three annual directories—Le Principali Società, Indici e Dati, and Il Calepino dell’Azionista—that are published by Mediobanca: a large Italian investment bank (www.mbres.it). Information about firms’ ultimate ownership, controlling share, corporate governance, family ties of the CEO, age, business activity and primary industry at the 3-digit NACE classification was obtained from annual reports and company websites, CONSOB and Borsa Italiana (Italian Stock Exchange market)’s websites, and Dun & Bradstreet.

We thank one referee for the suggestion to use the ratio of variable pay to total pay as the dependent variable. Results are also available, on request, for alternative definitions of the pay variables, such as: the log of total compensation; the ratio of total pay to total assets; and the log transformation of that ratio. In the end, we present the results that are based on the ratio of variable pay to total compensation because this variable is more easily interpreted and also because two variants are readily accommodated: one that includes only the cash component: Varshapay_cash; the other that adds the equity-based component: Varshapay_eq.

Data became available in 2005, after the European Commission issued Recommendation 2004/913/CE about executives’ role and compensation policy.

For a robustness check, we also use the log of tenure to account for non-linearity.

There is a large corporate finance literature about how to measure firm performance when estimating pay-performance sensitivity. See among the others: Murphy (1985), Jensen and Murphy (1990), Bertrand and Mullainathan (2001), Cunat and Guadalupe (2009), Croci et al. (2012), Cai et al. (2013).

We borrow from Rajan and Zingales (1995) the idea to use a classification that employs out-of-sample (or out-of-country) data to proxy for industry characteristics. This strategy is meant to reduce endogeneity concerns because—although the sectoral R&D and advertising intensities of Italian and UK economies are likely correlated—it cannot be claimed that the UK intensities are determined or influenced by Italian companies’ strategic decisions about R&D and advertising investments. Furthermore, in Italy there are no reliable statistics to measure the R&D and advertising intensity at the industry level.

At the bottom of the table, we report the F-tests that verify the separability of the full sample of non-family firms into two sub-samples of Type 1/Type 2 and high/low import penetration industries. The reported tests ensure that the sample-specific variables in a fully unrestricted model statistically differ, thus supporting the choice of using two sub-samples.

When we include only the linear term, its coefficient is insignificant; this is true also when we use the logarithmic transformation.

We also tested a specification with the log transformation of the ratio of debt to total assets, but we found that the coefficient is insignificant in all columns. The results are available on request.

As anticipated in Sect. 4, we performed a robustness test of the results on the full sample (as opposed to separate sub-samples) with the use of a semi-continuous variable for import penetration, which is constructed to capture four different thresholds of import penetration intensity. We found that the result on the higher PPS of family CEOs hold in this specification: \(Varshapay_{it} = \alpha + \beta_{1} \left( {ROA_{it - 1} } \right) + \beta_{2} \left( {FAMCEO_{it - 1} } \right) + \beta_{3} \left( {ROA_{it - 1} } \right)x\left( {IMP\_PEN_{it - 1} } \right) + \beta_{4} \left( {ROA_{it - 1} } \right)x\left( {FAMCEO_{it - 1} } \right)x\left( {IMP\_PEN_{it - 1} } \right) + \sum \beta_{j} {\mathbf{X}}_{jt - 1} + \mu_{i} + \lambda_{t} + \varepsilon_{it}\). The results show that, while the β₁ coefficient is positive but insignificant and the β₃ coefficient is insignificantly negative, the β₄ coefficient is positive and significant: The PPS of Family CEOs in industries with higher import penetration is significantly higher, in line with the evidence from the separate sub-samples in Table 5. The results are available on request. Unfortunately, we could not repeat the full sample analysis with a continuous or semi-continuous version of the R&D and Advertising intensity data, because they are unavailable.

The results hold when we estimate a model with the continuous variable—CR5—for the full sample. We find that on average PPS is lower where concentration is higher, but not for family CEOs, as the variable share of family CEOs’ pay is higher when the Cr5 is higher. The results are available on request.

The results show that, in line with the previous evidence, the variable share of total pay is significantly related to MTB in Type 2 and high import penetration industries and—different from Table 5—significant in less-concentrated industries. Varshapay_cash is lower for family CEOs as well as for CEOs who also hold the Chair position. However, we found no differences between family and non-family CEOs, so we report the results without the interactions. More generally, it is not surprising that, in Italy, CEO pay is more sensitive to an accounting measure of performance than to a market-based one, because the stock market is still thin and relatively underdeveloped.

As we indicated in footnotes 14 and 18, we also performed a battery of robustness checks that use alternative definitions of the dependent variable—log of total pay; the ratio of total pay to total assets; and its log transformation—and alternative estimation methods: Tobit regressions; and fractional response probit regression model. Comfortingly, the results confirm the evidence in Tables 5 and 6, regardless of the definitions and the estimation methods.

The results for the homogenous and low import penetration industries were insignificant (as they were in Tables 4, 5, and 6) and are available on request.