*4.3. Robustness Checks for Endogeneity*

In this section, we tend to identify the causal drive of the pension incentive to the hedging activity. As in most research in business, endogeneity may cause a concern in this study. While, by and large, the relation between pension incentive and hedging is well supported by theoretical work, it is relatively challenging to capture the causal effect because compensation contracts and hedging practices can be determined by unobservable firm-related and other factors.

We employ the approach of instrumental variables (IV) regression, where two instruments (firm age and CEO age) are used for each of the pension incentive measures in regressions.<sup>4</sup> In particular, treated as endogenous variable, pension incentive is regressed on a set of all other explanatory variables plus firm age and CEO age as instruments in the first stage. Predicted values of pension incentive are computed from the first-stage regression and used in the second-stage regression as specified in Equation (1).

We report the results in Table 5. In Model 5 of Panel A, as the dependent variable (hedging dummy) and the endogenous independent variable (the dummy variable of CEO pension relative leverage greater than one) are both dichotomous variables, we apply the Bivariate Probit model. Specifically, we simultaneously estimate a system in which the first and second stage regressions are both Probit models (Greene 2004). For other models of the hedging dummy regressions, we apply the IV-Probit model in which the first stage is an OLS (Ordinary Least Squares) model and the second stage is a Probit model. For the models in Panel B where the dependent variable is hedging notional amount scaled by the firm size (a continuous nonnegative variable), we apply IV-Tobit model as a regular IV regressions with truncated continuous dependent variable, namely, the first stage is an OLS model and the second stage is a Tobit model. Panel B of Table 5 reports the second stage results of IV regressions on hedging dummy and hedging notional amount, respectively. Consistent with our conjecture, the positive relation between hedging and pension incentive remain robust after controlling for potential endogeneity. Meanwhile, we find that the impact of equity-based incentives also shows evidence consistent with theoretical predications. For example, in Model 5 of Panel A (the hedging dummy regression), the coefficient of Vega incentive relative to Delta incentive is negative and significant the 5% level. In the regressions of hedging notional amount, the coefficient of this variable is found negative and significant at the 1% level in Models 5 of Panel B.



<sup>4</sup> We recheck the results by trying the lagged variable approach, with which the endogenous variable of pension incentive is lagged to help in mitigating the potential bias of endogeneity. We observe the positive relation between hedging activity and pension incentive unchanged.


**Table 5.** *Cont.*

This table reports the results of the robustness checks based instrumental variables (IV) models. The dependent variables are hedging dummy in Panel A and are hedging intensity (the ratio of total notional amount of hedging positions scaled by the book value of total assets) in Panel B. The two-stage regressions with a Probit model in the second stage are applied on Panel A and the two-stage regressions with a Tobit model in the second stage are applied on Panel B, respectively. We control for industry effects by using the Fama-French 48-industry classification and control for time effects by using year dummies. The robust standard errors are used in all models in Panel A to calculate Z-Statistics or T-Statistics that are reported in parentheses below estimates. \*, \*\* and \*\*\* denote significance at the 10%, 5% and 1% levels, respectively.

Roberts and Whited (2013) suggest that the IV regression model is only as good as the choice of instrumental variables. We select the instruments by considering both its relevance with pension incentives and irrelevance with hedging. More specifically, the relevance rule requires a non-zero correlation between an instrument and the endogenous variable of pension incentive, and the irrelevance rule indicates that the selected instrument must not be correlated with the unobserved determinants of hedging. Existing literature on pension benefits suggests that CEO age and firm age are closely related to the granting of retirement plans (Liu et al. 2014; Cassell et al. 2012; Sundaram and Yermack 2007). The firm age is used as one of instrumental variables due to the business life-cycle reason. Firms usually do not offer pension or other fringe benefit plans when they start in business, and gradually adopt them afterwards. Brown and Medoff (2003) document that the firms that have been in business longer are more likely to implement pension plans. With the consideration of little conclusive theory in classical economics for the connection between hedging activity and biological age, CEO age is used as another variable in the set of instruments. However, we are aware of some evidence in the recent literature that reports the impact of executives' age on corporate polices. For example, Serfling (2014) suggests that younger CEOs generally have a more aggressive leadership style. Jenter and Lewellen (2015) and Yim (2013) show that younger CEOs are more likely to engage in risky acquisitions. Croci et al. (2017) find a high likelihood of being a hedger from senior CEOs. When biological age connects to executives' psychological or physiological situation, the CEOs' age may not be able to introduce exogenous variations to capture the causal impact of pension incentive on firms' hedging policy.5

But to be more prudent, we use the econometric methods to test the verification of IV model and instruments. We employ the Smith-Blundell test of exogeneity to test the null hypothesis that there is no serious endogeneity in the model, and the Amemiya-Lee-Newey over-identification test to test the null hypothesis that there is no over-identification problem. Across the models, we find that the null hypothesis of Smith-Blundell test of exogeneity cannot be rejected in all Models except Model 4 for the hedging propensity and hedging intensity, indicating that a problem of endogeneity is not extensively present. Meanwhile, from models 1 through 4 for both the hedging propensity and hedging intensity regressions, the p-value in the Amemiya–Lee–Newey over-identification test is much greater than 10%, indicating that the instruments are chosen appropriately and there is no over-identification problem in our models. To summarize, the robustness checks confirm that the positive impact of pension incentive on hedging activities detected in baseline models is unlikely to be driven by the endogeneity issue.

#### *4.4. The Role of Pension Incentive in Di*ff*erent Governance Environment*

From the above empirical analyses a significant and positive relation between hedging and pension incentive has been documented in baseline and robustness models, but another important question, which relates to a considerable debate regarding the nature of incentive design as aforementioned, is how this positive relation varies in the different situations of corporate governance. As a component of compensation contract, pension incentive itself is closely associated with governance mechanism. Lee and Tang (2011) document that CEO pension benefits are associated with corporate governance. They find that entrenched managers, proxied by a small board, CEO duality, and protection from anti-takeover provisions, are more likely to obtain higher compensation of pension benefits. Halford and Qiu (2012) test whether firms that are likely to face more severe agency problems of debt

<sup>5</sup> With this caveat in mind, we seek the alternative instruments by trying the federal and state personal tax rates (Anantharaman et al. 2014; Kim and Lu 2011), as one can reasonably assume that those highest-paid CEOs would have different preference towards the compensation packages when they are subject to the different personal income tax brackets. In addition, we also utilize a regulatory reform, Pension Protection Act in 2006, which attempts to strengthen the pension system, protects retirement accounts and makes pension benefits more attractive. As the legislation provides greater incentives to firms and employees in investing in pension plans, this exogenous change of law should have no implicit impacts on the hedging decision. Although the tests by using the above alternative sets of instruments show the qualitatively consistent results, we acknowledge that the potential issue of endogeneity cannot be entirely resolved given the lack of perfect instruments, and therefore the causality results based on the IV models should be interpreted with caution.

provide more debt incentives. Inconsistent with the agency theory's prediction they find evidence that firms with lower default risk use more pension plans, and little evidence to support the hypothesis that pension incentive is used to alleviate the agency costs of debt.

In Section 2, our second hypothesis (H2) predicts the different relations between corporate hedging and CEO liability-based compensation which are conditional upon the strength of corporate governance. Recall that our first hypothesis proposes a positive relation between hedging and pension incentive. However, the optimal contracting hypothesis predicts that the positive relation should be more significant for firms with strong corporate governance than for firms with weak governance. Thus, in this section we test how the influence of pension incentive on hedging is contingent on the mechanism of corporate governance. In Table 6, we present the regressions of three baseline models but in the subsamples categorized by the median value of board independence.

Following the same steps, we report the incremental effect on the probability of hedging for a one-standard-deviation change in the continuous explanatory variables (or for a change from zero to one for the dummy explanatory variables). Furthermore, we apply the Chow test to examine the difference in coefficients between the two subsamples. Panel A reports the results for the hedging propensity regressions. In particular, the influence of pension incentive on hedging is more pronounced for the firms in the above-median group than in the below-median group (0.069 vs. 0.04 in Model 1 and Model 2). In other words, the effect of pension incentives on hedging is more prominent for firms with strong shareholder power. The differences are also economically significant based on the predicted probabilities. Specifically, Model 5 and Model 3 show that when CEOs' wealth leverage is greater than firms' leverage, it will cause a 44% (=0.287/0.650) rise in predicated probability of hedging for the firms in the above-median group of independent board directors, but only a 16% (=0.124/0.783) rise for the firms in the below-median group of independent board directors. Panel B of Table 6 reports the results based on the notional value of hedging, further confirming the finding.

Similarly, we conduct the subsample comparison by splitting firms into groups with the median value of blockholder ownership and report the results in Table 7. In Models 1 and 2 of Panel A, we find that a one-standard-deviation increase in pension incentive in firms with higher blockholder ownership will increase 7.3% of hedging probability, more than double the increase (2.8%) in the subsample with lower blockholder ownership. A similar difference with statistical significance is also found in Panel B of Table 7. For example, the coefficient of the dummy variable of CEO pension relative leverage greater than 1 is substantially larger in the subsample with low blockholder ownership than that in the subsample with high blockholder ownership (0.028 vs. 0.016 in Model 5 and Model 6) with significance at the 5% level. As a result, the above evidence lends strong support for the optimal contracting hypothesis (our second hypothesis). The pension incentive is governed to expose more influence on executives' decisions, and this risk-reducing function is induced more when firms have strong governance, but does not fully perform in the case of weak governance.



*JRFM* **2020** , *13*, 24

industry effects by using the Fama-French

in parentheses below estimates. \*, \*\* and \*\*\* denote significance at the 10%, 5% and 1% levels, respectively.

 48-industry and control for time effects by using year dummies. The robust standard errors are to calculate Z-Statistics or T-Statistics reported



 Tobit models are used in Panel B where the dependent variable is the ratio of total notional amount of hedging positions scaled by the book value of total assets. We controlindustry effects by using the Fama-French 48-industry and control for time effects by using year dummies. The robust standard errors are to calculate Z-Statistics or T-Statistics reportedparentheses below estimates. \*, \*\* and \*\*\* denote significance at the 10%, 5% and 1% levels, respectively.

in

#### **5. The Influence of Pension Incentive on Value of Corporate Hedging**

Tan and Young (2016) show that executives motivated by the long-term incentive pay, such as retirement plans may engage in behaviors unfavorable to shareholders. To provide a novel insight into the role of pension incentive in terms of value creation, we adopt Faulkender and Wang's (2006) methodology to assess the marginal value of hedging as a function of CEO pension incentives. We first treat hedging motivated by pension incentive as a beneficial driver to shareholders. Therefore, in addition to the firm characteristics specified in Faulkender and Wang (2006), we include the change of hedging activity in the regression model to quantify the value of hedging:

$$\begin{split} r\_{i\downarrow} - R\_{i\downarrow}^{B} &= \gamma\_0 + \beta \ast \Delta H d\mathcal{g} e\_{i\downarrow} + \gamma\_1 L\_{i\downarrow} + \gamma\_2 \frac{\Delta \mathbb{C}\_{i\uparrow}}{M\_{i\downarrow-1}} + \gamma\_3 \frac{\Delta \mathbb{E}\_{i\downarrow}}{M\_{i\downarrow-1}} + \gamma\_4 \frac{\Delta N \mathbb{A}\_{i\downarrow}}{M\_{i\downarrow-1}} + \gamma\_5 \frac{\Delta \mathbb{E}\_{i\uparrow}}{M\_{i\downarrow-1}} + \gamma\_6 \frac{\Delta \mathbb{I}\_{i\downarrow}}{M\_{i\downarrow-1}} \\ &+ \gamma\_7 \frac{\Delta D\_{i\downarrow}}{M\_{i\downarrow-1}} + \gamma\_8 \frac{N \mathbb{F}\_{i\downarrow}}{M\_{i\downarrow-1}} + \sum\_{t=1}^{10} Y \text{arc}\_{t} + \varepsilon\_{i\downarrow} \end{split} \tag{2}$$

where the dependent variable *ri,t* is the annual excess equity return calculated from a firm *i*'s stock return over year *<sup>t</sup>* − 1 to year *t,* net of *R<sup>B</sup> r*,*t* , the return of Fama and French (1993) size and book-to-market matched portfolio from year *t* − 1 to year *t*. Δ*Xi,t* is the notation for the one-year change of variable *X* for firm *i* over year *t* − 1 to year *t*, i.e., *Xt* − *Xt*−1; Δ*Hedge,t* represents the hedging propensity (*HP*) or change of hedging intensity (*HI*). *Mi,t* is market value of equity at time *t*; *Ci,t* is cash plus marketable securities; *Ei,t* is earnings before extraordinary items plus interest, deferred tax credits, and investment tax credits; *NAi,t* is total assets minus cash holdings; *RDi,t* is research and development expense (which is set to zero if missing); *Ii,t* is interest expense; *Di,t* is common dividends; *NFi,t* is total equity issuances minus repurchases plus debt issuances minus debt redemption; *Li,t* is the ratio of long-term debt plus debt in current liabilities divided by the market value of assets at time *t*. The regression results of Equation (2) are reported in Table 8. We use three measures of pension incentive to separate the sample.

**Table 8.** The impact of pension incentives on the value of hedging.



**Table 8.** *Cont.*

This table presents the OLS regressions of excess stock returns on CEO pension incentive and the changes in firm characteristics, including the explanatory variables from Faulkender and Wang (2006) specification augmented with the measures of pension incentive, hedging variables, and governance variable. Panel A examines the propensity of hedging (*HP*) and Panel B examines the intensity of hedging (*HI*). In all panels the dependent variable is the annual excess equity return calculated from a firm *i*'s stock return over year *t* − 1 to year *t*, *ri,t*, net of *R<sup>B</sup> r*,*t* , the return of Fama and French (1993) size and book-to-market matched portfolio from year *t* − 1 to year *t*. Δ*Xi,t* is the notation for the one-year change of variable *X* for firm *i* over year *t* − 1 to year *t*, i.e., *Xt* − *Xt*−1; *HIi,t* is hedging principal (dollar amount of total notional value of derivatives) at time *t*, and Δ*HIi,t* = *HIi,t* − *HIi,t*−1; *Mi,t* is market value of equity at time *t*; *Ci,t* is cash plus marketable securities; *Ei,t* is earnings before extraordinary items plus interest, deferred tax credits, and investment tax credits; *NAi,t* is total assets minus cash holdings; *RDi,t* is research and development expense; *Ii,t* is interest expense; *Di,t* is common dividends; *NFi,t* is total equity issuances minus repurchases plus debt issuances minus debt redemption; *Li,t* is the ratio of long-term debt plus debt in current liabilities divided by the market value of assets at time *t*. The robust standard errors are to calculate T-Statistics reported in parentheses below estimates. \*, \*\* and \*\*\* denote significance at the 10%, 5% and 1% levels, respectively.

Across the groups, we observe that the coefficients on the variables used in Faulkender and Wang (2006) are generally consistent with the literature. For example, cash and earnings are positively related to excess shareholder return, while leverage is negatively related to excess equity return. Considering that the dependent variable is excess equity return, the coefficient of explanatory variables measures the additional value accrued to the equity holder for one unit change of the explanatory variables. The coefficients of hedging propensity (*HPi,t*) and hedging intensity (*HIi,t*) mean the additional value flowing to shareholder per one dollar investment in hedging position. Particularly, Panel A shows the incremental value contributed by hedging adoption to the shareholders. We find the value creation of hedging is more significant when pension incentive is higher. This indicates that pension incentive functions an effective mechanism in motivating executives to implement active risk management to create firm value. In Panel B, the existing hedging position (*NIi,t*−1/*Mi,t*−1), incremental hedging amount (Δ*NIi,t*/*Mi,t*−1), and the interaction term of them are included. We find that the interaction term has a significant and positive impact on firm value. For example, the result in Model 1 shows that one dollar investment in existing hedging position brings USD 0.374 to shareholders and one dollar increase in hedging position creates an additional value of USD 0.204. In contrast, when pension incentive is in the lower level (Model 2), the additional value created by the hedging investment is only USD 0.059 per dollar of investment. This distinct difference suggests that the pension incentive promotes a firm's hedging strategy and also strengthens the value creation of hedging for shareholders.

Given the above finding as to how hedging creates value for shareholders, we advance further to consider the interplay of pension incentives and governance mechanisms. In particular, we augment the model specified in Equation (2) by adding measures of pension incentive and governance. For pension incentives, we construct a dummy variable, *PI*, for each of the pension incentive measures. Specifically, *PI* takes the value of one when *CEO Pension*/*Equity* is greater than the median value of this variable, and takes the value of one when *CEO Pension Relative Leverage is* above the median value or greater than one, respectively. In addition, we construct the indictors of strong governance, a dummy variable that takes the value of one when *Board Independence* is greater than the median value, and zero otherwise.6 The regression model to examine the value of hedging with the consideration of pension incentive and governance is specified below in Equation (3):

$$\begin{split} r\_{l,l} - R\_{l,l}^B &= \beta\_1 \star \text{Hodge}\_{l,l} + \beta\_2 \star \text{PI}\_{l\mid l} + \beta\_3 \star \text{Gov}\_{\gamma\text{S}\otimes\bar{\jmath},l\mid l} \star \text{PI}\_{l\mid l} \star \text{Hodge}\_{l\mid l} + \beta\_4 \star \text{Gov}\_{\gamma\text{S}\otimes\bar{\jmath},l\mid l} \star \text{PI}\_{l\mid l} \star \text{Hodge}\_{l\mid l} \\ &+ \gamma\_1 \text{I}\_{l\mid l} + \gamma\_2 \frac{\Delta \mathcal{C}\_{l\mid l}}{\overline{\mathcal{M}}\_{l\mid l-1}} + \gamma\_3 \frac{\Delta \mathcal{E}\_{l\mid l}}{\overline{\mathcal{M}}\_{l\mid l-1}} + \gamma\_4 \frac{\Delta \mathcal{A}\_{l\mid l}}{\overline{\mathcal{M}}\_{l\mid l-1}} + \gamma\_5 \frac{\Delta \mathcal{I}\_{l\mid l}}{\overline{\mathcal{M}}\_{l\mid l-1}} + \gamma\_7 \frac{\Delta \mathcal{D}\_{l\mid l}}{\overline{\mathcal{M}}\_{l\mid l-1}} + \gamma\_8 \frac{\text{NF}\_{l\mid l}}{\overline{\mathcal{M}}\_{l\mid l-1}} \\ &+ \sum\_{l=1}^{10} \text{Year}\_l + \varepsilon\_{l\mid l} \end{split} \tag{3}$$

The optimal contracting hypothesis suggests the more pronounced effect of pension incentive on the equity value of hedging to well-governed firms. Intuitively, since pension incentive is intended to dampen CEO incentives to pursue active risk management that largely benefits equity holders, the optimal contracting hypothesis predicts that pension benefit has a positive influence on the marginal relation between equity value and hedging in firms with stronger shareholder power. We report the result of above specification in Table 9. Remarkably, among the coefficients on the triple interaction terms for each of three pension incentive measures, we find that those that interacted with the above-median governance dummies are statistically significant, but those that interacted with the below-median governance dummies are not significant. For example (in Panel A), a significant coefficient of the triple interaction term (0.068) is found in the column of *CEO Pension*/*Equity Compensation* for the firms with governance measure above the median. From the column using the dummy variable of (CEO pension relative leverage greater than one) we also observe that the triple interaction term with the governance (>Median) has a significant coefficient of 0.068. These results indicate that the marginal impact of pension incentive on value of hedging is higher for the firms with good governance structure or strong shareholder power. This finding is consistent with the optimal contracting hypothesis that the contract of liability-based incentive is an outcome of optimal governance structure, and the additional hedging amount promoted by such effective incentive creates more value for shareholders.

<sup>6</sup> We also examine the results by using institutional blockholder ownership and observe the consistent results. These results are not reported for brevity.


**Table 9.** The Value Creation of Pension Incentive through Hedging Interacted with Governance.

This table presents the OLS regressions of excess stock returns on CEO pension incentive and the changes in firm characteristics, including the explanatory variables from Faulkender and Wang (2006) specification augmented with the measures of pension incentive, hedging variables, and governance variable. Panel A examines the propensity of hedging (*HP*) and Panel B examines the intensity of hedging (*HI*). In all panels the dependent variable is the annual excess equity return calculated from a firm *i*'s stock return over year *t* − 1 to year *t*, *ri,t*, net of *R<sup>B</sup> r*,*t* , the return of Fama and French (1993) size and book-to-market matched portfolio from year *t* − 1 to year *t*. Δ*Xi,t* is the notation for the one-year change of variable *X* for firm *i* over year *t* − 1 to year *t*, i.e., *Xt* − *Xt*−1; *HIi,t* is hedging principal (dollar amount of total notional value of derivatives) at time *t*, and Δ*HIi,t* = *HIi,t* − *HIi,t*−1; *Mi,t* is market value of equity at time *t*; *Ci,t* is cash plus marketable securities; *Ei,t* is earnings before extraordinary items plus interest, deferred tax credits, and investment tax credits; *NAi,t* is total assets minus cash holdings; *RDi,t* is research and development expense; *Ii,t* is interest expense; *Di,t* is common dividends; *NFi,t* is total equity issuances minus repurchases plus debt issuances minus debt redemption; *Li,t* is the ratio of long-term debt plus debt in current liabilities divided by the market value of assets at time *t*; *PI* denotes one of three dummy proxies of pension incentive in each regression, namely, *PI* takes the value of one when *CEO Pension*/*Equity* is greater than the median value of this variable, and takes the value of one when *CEO Pension Relative Leverage* is above the median value or greater than one, respectively. *Governance* is a dummy variable that takes the value of one when *Board Independence* is greater than the median value, and zero otherwise. T-Statistics in the column next to the estimates are computed using robust standard errors. \*, \*\* and \*\*\* denote significance at the 10%, 5% and 1% levels, respectively.

Apparently, the tests based on hedging intensity in Panel B also report the consistent results of significantly strengthening influence of pension incentive on the value of hedging. For instance, when we use the dummy measure of CEO pension relative leverage, pension incentive significantly increases the value of hedging by USD 1.297 for the firms with strong governance but no significant value creation is found for the peers with weak governance. This result justifies our optimal contract hypothesis and, also, confirms our findings in Tables 6 and 7. With more influence of pension incentive, the marginal value creation of hedging attributed to such liability-based incentives is higher for firms with strong shareholder power.
