CEO Inside Debt and Internal Capital Market Efficiency

CEO Inside Debt and Internal Capital Market Efficiency Abstract Agency theory argues that managerial equity-based incentives are more effective when firm solvency is likely while debt-based incentives are more effective when firms face a greater likelihood of bankruptcy. We examine the relation between chief executive officers inside debt holdings and the internal capital market efficiency of multi-segment firms. We find that CEO inside debt holdings are associated with conservative capital allocation to firm segments, with the result driven by financially distressed firms. Further analysis indicates that although CEO inside debt, on average, is negatively related to firm value, the relation is positive for financially distressed firms. Our evidence indicates that inside debt holdings align the interests of managers and external creditors, inducing managers to pursue conservative capital allocation strategies that appear to be optimal for firms facing insolvency. JEL classifications: G30, G31, G32 Keywords: CEO Inside Debt; Investments; Capital Allocation; Internal Capital Market Efficiency; Firm Value 2

1. Introduction The availability of an internal capital market, free of the constraints imposed by an imperfect external capital market, is a primary feature that distinguishes conglomerates from single-segment firms. While the ability to transfer funds across divisions could lead to excess value for the multi-segment firm, managers' self-interest could just as easily erode this advantage. Datta, D Mello, and Iskandar-Datta (2009) find that equity-based compensation, designed to mitigate manager-shareholder agency conflict, motivates conglomerate chief executive officers (CEOs) to allocate more resources to divisions with better investment opportunities resulting in greater excess value. Our paper considers the relation between inside debt, another form of executive compensation, and internal capital allocation efficiency (allocation efficiency). Edmans and Liu (2011) argue theoretically that equity-based incentives are more effective when firm solvency is likely while debt-based incentives are more effective when firms face a greater likelihood of bankruptcy. We investigate empirically the effects of inside debt on both resource allocation and excess value of multi-segment firms and how these effects vary with their financial conditions. Inside debt (pensions and deferred compensation) represents firms fixed payment obligations to the managers upon their retirement and are generally unsecured and unfunded. Inside debt holders are exposed to the same risk of firm insolvency as other unsecured creditors, thus inside debt holdings can align the interests of managers and creditors and reduce the agency cost of debt (Jensen and Meckling, 1976). From the firm perspective, the most important benefit of inside debt in the CEO compensation package is to reduce the agency cost of debt, which is ultimately borne by shareholders since debtholders would recognize the risk-shifting tendency of a CEO compensated strictly by equity (Edmans and Liu, 2011). Recent research on the relations between inside debt and corporate investment and financing decisions demonstrates that greater 3

inside debt leads to conservative corporate policies and firm risk reduction (e.g., Wei and Yermack, 2011; Cassell, Huang, Sanchez, and Stuart 2012; Phan, 2014). The efficiency of internal capital allocation is rooted in the concept that greater resources should be allocated to segments with greater opportunities (Rajan, Servaes, and Zingales 2000). However, segments with greater opportunities could be riskier. Inside debt can nudge CEOs toward a debtholder-like attitude with regard to firm risk. To the extent that inside debt makes CEOs risk averse, they would shift funds away from a higher expected return but potentially riskier segment toward a less risky segment with more stable cash flows. Capital transfers to segments with potentially lower risk could be beneficial to shareholders when firms face financial distress that accentuates the risk of bankruptcy and inefficient liquidation. Although models of firms engaging in risk-shifting (asset substitution) when facing financial distress abound in theory (e.g. Jensen and Meckling, 1976), there is less empirical support for its existence. 1 On the other hand, there is increasing evidence for risk management (Eckbo and Thorburn, 2003; Rauh, 2008; Almeida, Campello, and Weisbach, 2011) under similar circumstances. Consequently, while the presence of equity-based compensation may result in an increase in allocation efficiency (Datta et al., 2009), particularly if the measure of efficiency is constructed such that it is increasing in greater allocations toward the more productive but riskier investment, the presence of inside debt in the CEO's compensation package may exert an opposite effect on the same measure. We test this hypothesis in this paper. We begin our analysis by examining the effect of CEO inside debt holdings on internal capital allocation. Following Wei and Yermack (2011), Cassell et al. (2012), and Phan (2014), we 1 One exception is Eisdorfer (2008). 4

construct four measures of CEO inside debt holdings: relative CEO leverage, which is measured as the ratio of CEO s debt to equity scaled by the firm s debt to equity ratio; relative CEO incentive, which is the marginal change of the CEO s inside debt over the marginal change of his inside equity, given a unit change in the overall value of the firm, divided by the marginal change of firm debt over the marginal change of firm equity given the same unit change in the overall value of the firm; relative CEO leverage > 1 dummy, which is an indicator variable set to 1 if the relative CEO leverage is greater than 1, and 0 otherwise; and relative CEO incentive > 1 dummy, which is an indicator variable set to 1 if the relative CEO incentive is greater than 1, and 0 otherwise. Using a sample that includes 1,617 firm-year observations of an unbalanced panel of 694 multi-segment firms over the period 2006-2015, we find evidence that CEO inside debt is associated with conservative capital allocation, which is biased toward segments with lower investment opportunities but more stable cash flows. We further find that our results are driven by a subset of firms that are likely experiencing financial distress. Intuitively, financial distress increases the probability of default on debt payment and subsequent bankruptcy. Shifts in investment allocations to lower-risk segments, which reduce overall firm risk while improving cash flow stability, would help firms avoid insolvency and inefficient liquidation. Having established the relation between CEO inside debt and internal capital allocation, we examine how CEO inside debt affects the value of multi-segment firms. We note that the evidence regarding the effect of corporate diversification on firm value is mixed in the literature. Early research on conglomerates suggests that multi-segment firms typically suffer from a diversification discount. Berger and Ofek (1995) estimate the effect of diversification on firm value by comparing the value of the multi-segment firm with the value of a portfolio of pure-play 5

firms that match the individual segments. 2 They find that diversified firms are valued 13% to 15% below their stand-alone entities. The diversification discount is often linked to the inefficient allocation of internal funds across the divisions of a multi-segment firm (e.g., Lamont, 1997; Shin and Stulz, 1998). However, Campa and Kedia (2002), Graham, Lemmon, and Wolf (2002), and Villalonga (2004a and 2004b) find that diversification is not necessarily associated with a value discount. We do not take a stand on the relation between diversification and firm value but rather employ the excess-value framework adopted by this line of research to examine the effect of CEO inside debt on firm value while controlling for variables that are documented to have power to explain excess value. Our analysis reveals an average negative effect of CEO inside debt on excess value, which is consistent with the evidence documented by previous research (e.g., Cassell et al., 2012; Phan, 2014); however, for the subset of firms experiencing financial distress, inside debt is positively related to excess value. This evidence implies that shareholders recognize the benefits of CEO inside debt, which motivates conservative capital allocation for firms that are facing increased probability of debt-payment default and insolvency, precisely when such conservative policies could reduce the likelihood of bankruptcy. Our research adds to the stream of literature on the relations between executive compensation and corporate policies their implications for firm value. First, Edmans and Liu (2011) demonstrate theoretically that equity-based compensation alleviates managerial agency problems, resulting in increased firm value in solvent states, whereas inside debt provides managerial incentives for increasing firm value in insolvent states. Bennett, Guntay, and Unal 2 Berger and Ofek (1995) refer to the difference as the excess value, but since the multi-segment firm is generally lower in value, the difference is also known as the diversification discount. 6

(2015) report that conservative policies, motivated by managerial inside debt holdings, are beneficial for banks during the financial crisis period. These authors see it as a trade-off of reduced returns during normal times that pay off with increased protection during a financial crisis period. Their finding suggests that managerial conservatism motivated by inside debt holdings can be good for shareholders depending on external market conditions. However, no prior research has examined the effects of CEO inside debt on the behaviors of firms facing insolvency. Our evidence that the conservative capital allocation strategy induced by CEO inside debt benefits the shareholders of financially distressed firms provides first empirical support to Edmans and Liu s (2011) theoretical arguments and complements the finding of Bennett et al. (2015). Second, to the best of our knowledge, our research is the first that provides evidence of the relation between inside debt and the capital allocation among the segments of a conglomerate. Due to their diversification, multi-segment firms tend to have lower risk than pure-play firms. Nevertheless, our finding that CEO inside debt is associated with a conservative internal capital allocation of multi-segment firm underscores the managerial conservative behavior motivated by CEO inside debt holdings. We also provide evidence of the relation between inside debt and conglomerate firm value. The rest of the paper is organized as follows. Section 2 presents the literature review. We provide a description of the data, variable construction, and descriptive statistics in Section 3. Section 4 presents the empirical predictions, models, and results. Section 5 discusses robustness checks and Section 6 concludes the paper. 2. Literature Review 2.1. Internal Capital Market Efficiency 7

There is a long history of finance literature that focuses on internal capital market efficiency of multi-segment firms. Weston (1970) relates, and summarily dismisses, two criticisms specific to internal capital markets: the cross-subsidization of unprofitable activities by profitable segments and the "deep pocket" advantage of conglomerates. During the 1960s and early 1970s, the U.S. experienced a boom in the number of conglomerates, and the prevailing economic view attributed the growth to the greater allocation efficiency of internal capital (Lang and Stulz, 1994). Empirical research on the value of diversification was stimulated following the adoption of Security and Exchange Commission (SEC) Regulation S-K and Financial Accounting Standards Board (FASB) Standards No. 14 that required firms to report segment information after December 15, 1977. Using the newly available data in Compustat, Lang and Stulz (1994) find a negative relation between the degree of diversification and Tobin's q. Comment and Jarrell (1995) observe lower stock returns for diversification, and Berger and Ofek (1995) document lower values for diversified firms compared to the imputed value of the individual segments if they were to exist as stand-alone entities. Berger and Ofek (1995) further examine the reasons for the reduced value and find that overinvestment and cross-subsidization are significant factors. Some dissenting views on the diversification discount include Villalonga (2004a) and Campa and Kedia (2002), who attribute the discount to self-selection bias in firms choice to diversify, or Graham et al. (2002), who find that additional target segments added by acquisitions are already trading at a discount prior to becoming part of the conglomerate. Villalonga (2004b) uses establishment-level data from the U.S. Bureau of Census, instead of the Compustat data, for analysis and finds a diversification premium. Scharfstein and Stein (2000) combine rent-seeking division managers with self-interested headquarters CEOs in a model with two layers of agency issue and obtain a result that weaker 8

segments are subsidized by stronger segments. Billet and Mauer (2003) demonstrate that subsidies to financially constrained segments can increase the value of multi-segment firms. In Rajan et al. (2000), the CEO misallocates funds to force self-interested division managers to select efficient investments for a second-best solution to avoid the third-best outcome. Ozbas and Scharfstein (2009) find evidence to support inefficiencies in the allocation and conclude that internal capital markets are inefficient with headquarters management agency problems playing a factor. Datta et al. (2009) argue that if the misallocation is primarily due to the extraction of private benefits by headquarters CEOs, then equity-based executive compensation should improve allocation efficiency (Jensen and Meckling, 1976; Agrawal and Mandelker, 1987; Coles, Daniel, and Naveen 2006). Moreover, if CEOs private benefits increase in the misallocation, then CEO agency conflict will outweigh division managers' rent-seeking as the primary cause of value destruction. Datta et al. (2009) find support for both hypotheses. 2.2. Inside Debt In their seminal paper on agency costs of equity and debt, Jensen and Meckling (1976) introduce the term "inside debt," define it as debt held by the "owner-manager" of the firm, and suggest that a holding by the manager of the same fraction of the total debt of the firm as his fraction of the total equity would eliminate the shareholder-debtholder conflict leading to riskshifting. The most common forms of inside debt held by managers are pensions and deferred compensation (Sundaram and Yermack, 2007). Edmans and Liu (2011) model the optimal level of inside debt by considering not just the risk-shifting incentive of the manager induced by equity-based compensation, but also managerial effort, the probability of bankruptcy, and resulting liquidation value. They find that the optimal inside debt-equity ratio can vary from the firm debt-equity ratio as proposed in Jensen and 9

Meckling (1976) and that inside debt is a more effective solution to the agency cost of debt than either a solvency bonus (John and John, 1993) or dependence on manager's concern for his reputation (Hirshleifer and Thakor, 1992). 3 Since pensions and deferred compensation represent firms fixed payment obligations to the managers upon their retirement and are generally unsecured and unfunded, inside debt holders are susceptible to the same risk of firm insolvency as other unsecured creditors. Empirical and anecdotal evidence supports this proposition. Gerakos (2010) examines a sample of 172 firms and reports that only three of them provide protection to CEO pensions should these firms go bankrupt. The following piece of anecdotal evidence illustrates the loss of pension benefits incurred by executives when their firms fell into bankruptcy: More than 100 former General Motors executives who sued the automaker for cutting their retirement benefits during the company's 2009 bankruptcy had their appeal rejected today by a federal appeals court Most top executives pensions were cut by two-thirds, including former CEO Rick Wagoner, whose pension was reduced from $20 million to about $8.5 million. (Source: Automotive News, August 7, 2013). 4 Empirical research on inside debt lagged behind research on equity-based executive compensation primarily due to the lack of data. Bebchuk and Jackson (2005) rely on handcollected data, and actuarial assumptions and calculations, to estimate the pensions of 51 current and retired CEOs. They conclude that pensions, as a percentage of total CEO compensation, constitute a significant component (48.3%), but this measure also exhibits a large variation among 3 For a normative discourse on including inside debt in executive compensation, see Edmans (2012). 4 Available at the following link http://www.autonews.com/article/20130807/oem02/130809876/retired-gm-execslose-appeal-in-suit-over-pension-cuts. Retrieved on May 11, 2015. 10

the executives. Sundaram and Yermack (2007) also find a significant inside debt component in their larger hand-collected sample, and further conclude that CEO inside debt is positively related to CEO age and distance to default, an indication that greater inside debt results in more conservative corporate policies. A change in the SEC disclosure requirement in late 2006 and the accompanying availability of more easily accessible data stimulated empirical research related to inside debt. Wei and Yermack (2011) find that the 2007 filing of proxy statements post SEC disclosure reform on pension and deferred compensation was accompanied by an increase in bond prices, a decrease in equity prices, a decrease in volatility of both securities, and decreases in the implied volatility of exchange-traded options and in the spreads of default swaps associated with the firms. The transfer in value from equity to debt implies that the revealed inside debt was considered too high on average, while the lowered volatility and default swap spread indicates that investors expect lower price volatility induced by the revealed greater inside debt. Several recent studies argue that inside debt provides a greater alignment of managers and external creditors interests, leading to conservative corporate policies and a decrease in firm risk. Cassell et al. (2012) report that CEO inside debt holdings induce conservative investment and financing policy choices. Liu, Mauer, and Zhang (2014) find a positive relation between inside debt and cash holdings, whereas Phan (2014) finds a negative relation between CEO inside debt and corporate risk-taking in mergers and acquisitions. Bennett et al. (2015) use a sample of bank holding companies and show that greater inside debt measured in 2006 is associated with lower default risk and better performance during the subsequent crisis period. As a result, CEO inside debt leads to lower cost of debt and fewer restrictive debt covenants (Anantharaman, Fang, and Gong, 2014; Dang and Phan, 2016). 11

3. Sample, Variable Construction, and Descriptive Statistics We form our sample by combining CEO compensation data from Standard & Poor s (S&P) Executive Compensation (ExecuComp) database with firm-level accounting data from S&P s Compustat (Compustat) and segment-level data from the Compustat Industrial Segment (CIS) databases. Since a large number of firms do not report inside debt information, we follow Wei and Yermack (2011) and Cassell et al. (2012) in restricting our sample to multi-segment firms with positive CEO inside debt holdings to avoid a potential bias in our analysis. The sample spans the period 2006-2015. We follow previous research (e.g., Berger and Ofek 1995; Datta et al., 2009, among others) in applying the following filters to the sample: (i) we require firms to have nonmissing segment information on sales, assets, and capital expenditure; (ii) we exclude firms with $20 million or less in sales; (iii) we exclude firms whose sum of segment sales are not within 1% of the total firm sales or firms whose sum of segments assets are not within 25% of the firm s assets; (iv) we exclude firms with segments that have 1-digit SIC code equal to 0 (agriculture), 6 (finance), and 9 (non-operating divisions); (v) we exclude firms that have all segments in the same industry; (vi) we require at least five industry-matched pure-play firms for each division in the multi-segment firm based on a 3-digit SIC code match; and (vii) we require data on Compustat to calculate all other control variables for the multi-segment firms. Our merged inside debt, accounting, and segment data yield an unbalanced panel of 1,617 firm-year observations of 694 multi-segment firms, whose CEOs have positive inside debt holdings in the form of pension and deferred compensation over the sample period. We use this sample to examine the effects of CEO inside debt on internal capital allocation and firm excess value. 12

We follow Rajan et al. (2000) and Datta et al. (2009) in constructing a measure of allocation efficiency, the industry-adjusted relative value added by allocation (RVIA), as follows:, (1) where ωj is the proportion of segment j s book value of assets to firm assets, qj is segment j s Tobin s q proxied by the asset-weighted average Tobin s q of all stand-alone firms operating in the same 3-digit SIC code industry as that of segment j. is the asset-weighted average imputed qj s of the multi-divisional firm. Capexj is the capital expenditure of segment j and BAj is the book value of segment j s assets whereas / is the asset-weighted average capital expenditure to assets ratio for matched stand-alone firms operating in the same three-digit SIC industry as segment j. The variables qj and ωj are measured as of the beginning of the period. Segment j s Tobin s q is proxied by that of industry-matched pure-play firms since market values are not available for the individual segments. The objective is to measure the segment s investment opportunities compared to the other segments of the firm,, as well as the capital expenditure of the segment normalized by the book value of the segment s assets compared to the asset-weighted average equivalent measure for the industry-matched pure-play firms,. The weight is the proportion of segment j s book value of assets to the firm s book value of assets. If the investment opportunity is above average for the sector and the investment allocation is also above average, the weighted product,, will provide a positive contribution toward the measure; this will also occur if both the investment opportunity and the associated allocation are below average. If the investment opportunity and the allocation 13

move in opposite directions, the contribution will be negative. By construction, higher positive RVIA values indicate higher allocation efficiency while lower or negative values of RVIA indicate sub-optimal allocation efficiency. as: Similar to Berger and Ofek (1995) and Datta et al. (2009), we measure firm excess value, (2) where MV is the firm s market value defined as book value of assets plus the difference between market value and book value of equity, and I(MV) is the imputed value of the multi-segment firm calculated as the sum of the imputed values of the firm s n business segments: Salesi is segment i s sales value and (3) is the median industry multiple of total capital to sales of stand-alone firms matched to the business segment using the 3-digit SIC code. We use the following inside debt measures developed by Edmans and Liu (2011) and Wei and Yermack (2011) to estimate relative CEO leverage and relative CEO incentive: Relative CEO leverage = (DCEO/ECEO) (DFIRM/EFIRM), (4) where DCEO and ECEO is the manager s inside debt and inside equity, while DFIRM and EFIRM is the firm debt and equity, including the amount held by the inside manager, and: Relative CEO incentive = (ΔDCEO/ΔECEO) (ΔDFIRM/ΔEFIRM). (5) The relative CEO incentive captures the marginal change of the manager s inside debt over the marginal change of his inside equity, given a marginal change of the firm debt over the marginal change of the firm equity. Two other inside debt measures are the relative CEO leverage > 1 dummy, which is an indicator variable set to 1 if the relative CEO leverage is greater than 1 and 0 14

otherwise, and the relative CEO incentive > 1 dummy, which is an indicator variable set to 1 if the relative CEO incentive is greater than 1 and 0 otherwise. To account for the effect of equity-based compensation, we calculate and control for CEO delta, which measures the change in CEO equity, both stock and option holdings, to a one dollar change in the firm s stock price, and CEO vega, the change in CEO equity to a 0.01 change in the firm s stock return volatility (Core and Guay, 2002). Previous research (e.g., Rajan et al., 2000) documents that greater diversity decreases allocation efficiency. Therefore, similar to Datta et al. (2009), we construct three alternative measures to capture the degree of diversification by multi-segment firms. The inverse Herfindahl index is the inverse of a firm s sales-based Herfindahl index, calculated at the beginning of the year as:, (6) where j indicates segment j and n is the total number of segments. Our second measure of diversity is number of segments, which is the number of segments reported by the firm and available in the CIS data. Our final measure is diversity, developed by Rajan et al. (2000) and defined as:, (7) where qj is market-to-book ratio of the segment proxied by the asset-weighted q of stand-alone firms in the same industry as the firm, ωj is the weight in terms of assets of segment j, n is a number of the segments of the diversified firm, and ωj and qj are beginning of year values. This measure 15

can be interpreted as an asset-weighted coefficient of variation of the individual segment's Tobin's q ratios. Table 1 reports the summary statistics for the allocation efficiency measure, RVIA, firms excess value, EV, all four CEO inside debt measures, CEO delta, CEO vega, the three diversity measures, new CEO dummy, CEO age, CEO tenure, favorable tax status dummy, R&D/sales, book value of assets, the capital expenditures to book value of assets ratio, capex/assets, and Tobin s q. Appendix A provides definitions of the variables. The four CEO inside debt measures are similar in magnitude to those reported by Phan (2014) for the period 2006-2009. A significant number of CEOs have inside leverage greater than their firm s leverage, as indicated by the mean of relative CEO leverage > 1 dummy, 0.47 (0.42 in Phan, 2014). The average book value of the sample firm is $8.56 billion and the average number of segments is 4.7 ($4.8 billion and 2.84, respectively, in Datta et al., 2009). 5 The allocation efficiency measure, RVIA, and diversification measures are also similar to those reported by Datta et al. (2009). 4. Empirical Predictions, Models, and Result Discussions 4.1. CEO Inside Debt and Internal Capital Allocation Inside debt aligns the interests of managers and external creditors, and is expected to motivate managers to act more conservatively with respect to risk (Jensen and Meckling, 1976; Edmans and Liu, 2011). Previous empirical research documents that managers debt-based compensation induces risk-decreasing strategies (Sundaram and Yermack, 2007), leading to 5 The difference arises from different sample periods and our screening out firm-year observations that do not have positive CEO inside debt holdings. 16

reduced R&D and financial leverage, increased working capital and firm diversification (Cassell et al., 2012), increased bond and decreased stock prices, and decreased volatility (Wei and Yermack, 2011). A shift in capital allocation from a higher expected return but riskier segment to a less risky one in a multi-segment firm can reduce risk but may also decrease overall firm performance, implying internal capital allocation inefficiency. Following the foregoing discussion, we predict a negative relation between CEO inside debt and allocation efficiency. We first examine the effect of CEO inside debt on allocation efficiency using the following regression model: RVIAi,t = α + βceo Inside Debti,t-1 + θxi,t + γfirm dummies + δyear dummies + εi,t, (8) where RVIAi,t is a measure of allocation efficiency as defined in Equation 1 for firm i in year t. CEO inside debt is proxied by either relative CEO leverage or relative CEO incentive. Since these two variables are heavily skewed to the right, we use their natural logarithm transformation in the analysis. Xi,t is a vector of control variables. 6 We control for firm diversity since Rajan et al. (2000) demonstrate that an increase in diversity will decrease allocation efficiency. Similar to Datta et al. (2009), we control for CEO change by using an indicator variable that takes the value of 1 if the firm has a new CEO, and 0 otherwise. We also control for firm characteristics that include R&D, firm size, capital expenditures, and Tobin s q. We use the natural logarithm of the book value of assets as a proxy for firm size. Datta et al. (2009) report that CEO equity-based compensation is positively related to allocation efficiency, therefore we additionally include CEO delta and CEO vega in some specifications. Since inside debt and capital allocation could be related to timevarying macroeconomic conditions, we control for year-fixed effects. CEO inside debt and capital 6 Similar to previous research, we use contemporaneous variables as controls. However, our results are qualitatively similar if we lag the control variables by one period. 17

allocation could also be related to unobserved time-invariant firm characteristics, such as the financial conditions of the firms, which raises a concern for potential endogeneity. We alleviate this endogeneity concern by controlling for firm-fixed effects in the regressions. Table 2 reports the regression results. Since we have three different proxies for diversity with qualitatively similar impact on the findings, for brevity we report the results with the inverse Herfindahl index (results for the other diversity measures are available from the authors upon request). Columns 1-4 of Table 2 report the results of the RVIA regressions on CEO inside debt measures, firm characteristics, and firm- and year-fixed effects, but do not include CEO delta and CEO vega. The coefficients on all proxies of CEO inside debt holdings are negative, ranging from -0.023 to -0.015, and statistically significant. In Columns 5-8, we further include managerial equity-based compensation proxied by CEO delta and CEO vega. The coefficients of all measures of CEO inside debt holdings remain negative, ranging from -0.024 to -0.017, and statistically significant. The economic effect of CEO inside debt on RVIA is also important. Since the test variable, relative CEO leverage (relative CEO incentive), is in the natural logarithm form, its coefficient estimate indicates that, holding other variables unchanged at their sample means, a 1% increase in relative CEO leverage (relative CEO incentive) is associated with 1.5-1.6 basis points (1.7 basis points) decrease in RVIA, which is equivalent to approximately 10% of its sample mean in absolute terms. These results indicate that CEO inside debt is negatively related to allocation efficiency. Boards of directors may anticipate the effects of CEO inside debt on internal capital allocation when they design executive compensation contracts, which implies a possible joint determination of CEO compensation structure and internal capital allocation. Thus, failing to control for endogeneity due to a possible joint determination of CEO compensation and internal 18

capital allocation may render our coefficient estimates biased and inconsistent. It is ideal if there were exogenous shocks to CEO inside debt that we could use to identify its effects on internal capital allocation. However, CEO inside debt compensation is a corporate decision and previous literature does not provide a clear guidance on any shocks to CEO inside debt that we can use for our analysis. Therefore, we address this endogeneity concern by running two-stage instrumental variable (IV) regressions. Following previous research (e.g., Anantharaman et al., 2014; Cassell et al., 2012; Phan, 2014), we use the natural logarithm of CEO tenure, natural logarithm of CEO age, and a favorable tax status dummy, all known to be important determinants of inside debt compensation, as instruments for relative CEO leverage (relative CEO incentive). 7 Intuitively, the older a CEO is and the longer he works for a given firm, the higher are his pension benefits, which imply positive relations between CEO inside debt holdings and CEO age and tenure. Sundaram and Yermack (2007) suggest that taxation affects stock option and pension compensation since these compensation forms enable managers to defer their incomes to future years, which could lead to net tax savings for both the firm and the executives depending on their marginal tax rates. We use the presence of a tax-loss carry-forward in a firm s balance sheet as a proxy for its favorable tax status. Our selected instruments should be valid because they are directly related to CEO inside debt measures but there are no obvious reasons to argue that they are directly related to allocation efficiency other than through CEO inside debt. 7 Since CEO age and CEO tenure are highly correlated, in an unreported test, we alternately drop each from the set of instruments but our results are qualitatively unchanged. We also consider other instruments, such as liquidity constraint, state tax rate on individual income, and industry median relative CEO leverage (relative CEO incentive) suggested by previous studies but these variables do not pass the instrument validity tests in our analysis. 19

Table 3 reports the results of the two-stage IV regressions with relative CEO leverage (Columns 1 and 2) and relative CEO incentive (Columns 3 and 4). The coefficients of the instruments have the expected signs and are highly significant, indicating that each instrument is relevant by itself. The F-statistics of the Cragg-Donald weak instrument tests are greater than 10, suggesting that the selected instruments are not weak. The Hausman endogeneity test validates the need to correct for endogeneity while the overidentification test indicates that our selected instruments are valid. Consistent with the results in Table 2, the coefficients on the instrumented relative CEO leverage and instrumented relative CEO incentive are negative (-0.035 and -0.038 in Columns 2 and 4, respectively) and statistically significant. This evidence indicates that our results are robust to the correction for endogeneity. Next, we investigate the relation between CEO inside debt and internal capital allocation for subgroups of firms characterized by their degrees of financial distress. The agency problem of debt suggests that firms that face financial distress or a threat of insolvency may engage in riskshifting to benefit shareholders (Jensen and Meckling, 1976). The intuition is that as shareholders of these firms hold residual claims on firm assets, investing in riskier projects may increase the payoff to the shareholders on the upside but does not affect shareholder value on the downside. In the worst-case scenario, the payoff to the shareholders is zero when the debt value exceeds the asset value. In contrast, if these firms follow a risk-management strategy that favors less risky investment projects, they can safeguard the firm value, which better serves the debtholders interest. It is worth noting that there are theoretical arguments (e.g. Almeida et al., 2011) as well as growing empirical evidence (e.g., Eckbo and Thorburn, 2003; Rauh, 2008) that support risk management. Following these discussions, we expect the effect of CEO inside debt on conservative capital allocation to be more pronounced for firms that face financial distress. 20

Similar to previous research (e.g., Denis and Mihov, 2003), we use firms Altman s (1977) Z-scores to sort firms into financially distressed and undistressed subgroups. The Altman Z-score is calculated as: Z = 1.2(Working Capital/Total Assets) + 1.4(Retained Earnings/Total Assets) + 3.3(Earnings Before Interest and Taxes/Total Assets) + 0.6(Market Value of Equity/Book Value of Liabilities) + 0.999(Net Sales/Total Assets). Firms with calculated Altman s Z scores below 1.81 are considered financially distressed with a high probability of debt payment default and bankruptcy. Following this classification, our sample includes 531 firm-year observations with Z- scores < 1.81 and 1,086 firm-year observations with Z-scores 1.81. Table 4 reports the results of the RVIA regressions for the two subgroups. In Columns 1-4, the coefficients of CEO inside debt measures are all negative, ranging from -0.022 to -0.017, and statistically significant for the subgroup of financially distressed firms. In contrast, the coefficients of CEO inside debt measures are statistically insignificant in Columns 5-8 for financially undistressed firms. This evidence is consistent with our expectation that CEOs of financially distressed firms with larger inside debt holdings are more likely to follow a conservative capital allocation strategy. To address endogeneity concern due to a potential joint determination of CEO inside debt and internal capital allocation, we again turn to the IV regressions using the set of instruments similar to the one in Table 3. Table 5 reports the second-stage results of the RVIA IV regressions separately for the financially distressed and undistressed firms. Consistent with the results in Table 4, the coefficients of instrumented relative CEO leverage and instrumented relative CEO incentive are negative (-0.034 and -0.039, respectively) and statistically significant for the financially distressed firms but insignificant for the undistressed firms. This result indicates that our finding is robust to the correction for endogeneity. 21

4.2. Inside Debt and Multi-segment Firm Excess Value In this section, we investigate the effect of CEO inside debt holdings on the value of multisegment firms proxied by their excess values. The results from the previous section indicate that CEO inside debt has a negative effect on allocation efficiency of multi-segment firms. To the extent that higher capital allocation to greater investment opportunity but riskier segments creates more value, we predict a negative relation between CEO inside debt and excess value of the average multi-segment firm. However, since conservative investment policy may help financially distressed firms avoid bankruptcy and inefficient liquidation, a capital allocation strategy that is biased toward the less risky segments would enhance the value of financially distressed firms. Following this argument, we expect a positive relation between CEO inside debt and the excess value of financially distressed multi-segment firms. We examine the effect of CEO inside debt on excess value using the following regression: EVi,t = α + βceo Inside Debti,t-1 + θxi,t + Firm fixed effects + Year fixed effects + εi,t, (9) where EVi,t is a measure of excess value as defined in Equation 2 for firm i in year t and Xi,t is a vector of control variables including inverse Herfindahl index and firm size. Columns 1-4 of Table 6 reports the results of the firm excess value regressions on CEO inside debt measures, firm characteristics, and firm- and year-fixed effects, but without controlling for CEO delta and CEO vega. The coefficients on all proxies of CEO inside debt holdings are negative, ranging from - 0.217 to -0.159, and statistically significant. In Columns 5-8, we further control for managerial equity-based compensation proxied by CEO delta and CEO vega. The regression results indicate that all proxies of CEO inside debt holdings are negative, ranging in values from -0.202 to -0.135, and statistically significant. The economic effect of CEO inside debt on multi-segment firm excess value is also substantial. The coefficient estimates indicate that, holding other variables unchanged 22

at their sample means, a 1% increase in relative CEO leverage (relative CEO incentive) is associated with 16-19 basis points (14-17 basis points) decrease in firm excess value. This evidence suggests that, on average, CEO inside debt is negatively related to multi-segment firm value. The reduced value could be a direct consequence of the suboptimal capital allocation induced by CEO inside debt. Moreover, the finding also suggests that as segments with higher investment opportunities are deprived of funds, they may find it harder to compete with their pureplay counterparts. It is possible that a firm s excess value and CEO inside debt are both correlated with unobserved firm characteristics, raising the issue of potential endogeneity. To address this concern, we run IV regressions using the set of instruments similar to the one used in Tables 3 and 5. Table 7 reports results of the two-stage excess value IV regressions with relative CEO leverage (Columns 1 and 2) and relative CEO incentive (Columns 3 and 4). Consistent with the results in Table 6, the coefficients of instrumented relative CEO leverage and instrumented relative CEO incentive are both negative (-0.573 and -0.631, respectively) and statistically significant. This evidence indicates that our results are robust to the correction for potential endogeneity bias. In Table 8, we run the excess value regressions separately for financially distressed and undistressed subgroups. Interestingly, we find that the coefficients of CEO inside debt variables are positive, ranging from 0.217 to 0.318, and statistically significant for the financially distressed subgroup. In contrast, the coefficients of CEO inside debt variables are negative, ranging from - 0.245 to -0.147, and statistically significant for the subgroup of financially undistressed firms. We also run the IV regressions separately for each subgroup of firms sorted on their financial conditions and report the second-stage results in Table 9. Consistent with the results reported in Table 8, the coefficients of instrumented relative CEO leverage and instrumented 23

relative CEO incentive are positive (0.284 and 0.342, respectively) and statistically significant for financially distressed firms. In contrast, the coefficients of the instrumented variables are negative (-0.322 and -0.368, respectively) and statistically significant for financially undistressed firms. This evidence indicates that our finding of a positive (negative) relation between CEO inside debt and excess value of financially distressed (undistressed) multi-segment firms is not sensitive to the correction for endogeneity. To the extent that CEO inside debt motivates corporate risk-decreasing behavior, it may reduce the bankruptcy risk of financially distressed firms. In this context, investors may view the conservative capital allocation policy pursued by financially distressed firms as optimal and, thus be willing to assign a higher value to these firms. Conversely, we do not expect investors of financially undistressed firms to support a conservative capital allocation approach that hampers corporate growth and adversely affects firm performance. Our finding that CEO inside debt has a positive (negative) effect on excess value of financially distressed (undistressed) firms is consistent with this line of argument. Importantly, while the conservative internal capital allocation induced by CEO inside debt could be value-decreasing for an average firm or firms in solvent states, it appears to be an optimal investment strategy for financially distressed firms or firms in insolvent states. 5. Robustness Checks We run several additional tests to verify the robustness our results and summarize our findings in this section. Our measure of allocation efficiency, which follows Rajan et al. (2000) and Datta et al. (2009), is increasing in value when a firm's segments with relatively greater investment opportunity obtain a relatively greater share of firm resources. In the first robustness 24

check, we validate our conjecture that segments with greater opportunity are also more risky. Since investment opportunity is proxied by Tobin's q, which is not available for individual segments for a multi-segment firm, we use a sample of single-segment firms for analysis. Table A1 in the Internet Appendix presents the results of the regression of firm risk, measured by cash flow volatility of single-segment firms, on Tobin's q and other control variables. 8 Our risk model is motivated by previous corporate risk-taking studies (e.g., Coles et al., 2006; John, Litov, and Yeung, 2008). Cash flow volatility is a time series variable calculated as the standard deviation of seasonally adjusted quarterly EBITDA-to-assets ratios of a single-segment firm over a five-year period. The control variables include firm size, proxied by the natural logarithm of sales, book leverage, capital expenditure, and R&D investment, the last two scaled by the book value of assets. The regression results indicate a positive and significant relation between Tobin's q and singlesegment firms cash flow volatility, lending credence to our belief that higher allocations to segments with lower growth opportunities, which typically yield lower expected returns, may offer greater cash flow stability in the context of our analysis. Our argument about a negative relation between CEO inside debt and capital allocation efficiency of multi-segment firms is grounded on the premise that the capital allocation behavior induced by CEO inside debt reduces firm risk, which adversely affects the value of financially undistressed firms but increases value of the financially distressed firms. To validate our argument, we examine the direct relation between CEO inside debt and the risk of the segment of a multisegment firm. To the extent that CEO inside debt motivates risk-decreasing capital allocation 8 Cash flow volatility is the appropriate measure for firm risk in this test since it is a firm-based measure, which is likely under the control of the CEOs. 25

strategy, we expect a negative relation between CEO inside debt and the segment s cash flow volatility, a proxy for segment risk. Table 10 reports results of the segment-level cash flow volatility on CEO inside debt and other control variables. Cash flow volatility is calculated as the standard deviation of seasonally adjusted quarterly EBITDA-to-assets ratios of a segment of a multi-segment firm over a five-year period. For control variables that are unavailable at the segment level (e.g., market-to-book and financial leverage), we use firm-level data instead. Due to the missing data of segment-level cash flow and other variables, the regression sample is small. The estimated results indicate a negative and significant relation between CEO inside debt measures and segment-level cash flow volatility, which is consistent with our expectation. To alleviate the problem of missing segment-level cash flow data of multi-segment firms, in an alternative analysis, we use the average cash flow volatility of single-segment firms in the same 3-digit SIC industry as a proxy for the risk of a segment of a multi-segment firm and report the regression results in Table A2 in the Internet Appendix. We find that our results are qualitatively unchanged. We further examine the relation between CEO inside debt and segment-level risk for multi-segment firms sorted on the level of financial distress. The results reported in Table A3 in the Internet Appendix indicate that the negative relation between CEO inside debt and segment-level risk is only statistically significant for financially distressed firms. Intuitively, the risk choice embedded in the internal capital allocation of the multi-segment firms will ultimately be reflected in the firm risk. Therefore, we examine the relation between CEO inside debt and firm risk using both the firm-based cash flow volatility and market-based stock return volatility as alternate surrogates for firm risk. Stock return volatility is calculated as the standard deviation of daily stock returns of a firm in a given year. The regression results reported 26

in Table 11 indicate that the coefficients of CEO inside debt measures are all negative and highly significant. This finding suggests that CEO inside debt has a negative effect on multi-segment firm risks, which is consistent with our expectation. Next, to ensure that our results are not sensitive to the way we construct the allocation efficiency variable, we follow Datta et al. (2009) in constructing two additional measures of allocation efficiency for testing. The first alternative measure is the relative value added by allocation (RVA) that accounts for both firm and industry adjustments calculated as follows: (10) Whereas the RVIA measure adjusts for industry changes by subtracting out the investment ratio of the industry-matched stand-alone firms from the investment ratio of the segment, RVA subtracts out a third component,, which is the abnormal investment ratio averaged across all the segments of the firm. The RVA measure avoids treating the total differential funds available to the conglomerate as a transfer between segments. The second alternative measure of allocation efficiency is the absolute value added by allocation (AVA), which is calculated as: 1 (11) The only difference between the AVA and RVIA measures lies in the way that the relative investment opportunities of a segment is captured: the RVIA measure is based on the mean assetweighted imputed qj s of the diversified firm as the benchmark, whereas the corresponding benchmark for AVA measure is unity. We substitute RVA and AVA as alternative measures of allocation efficiency for RVIA and rerun our analysis. Tables A4 and A5 in the Internet Appendix report the results of the allocation 27