Family Firms, Debtholder-Shareholder Agency Costs and the Use of Covenants in Private Debt. Mark Bagnoli. Hsin-Tsai Liu. Susan G.

Transcription

1 Family Firms, Debtholder-Shareholder Agency Costs and the Use of Covenants in Private Debt by Mark Bagnoli Hsin-Tsai Liu Susan G. Watts Krannert School of Management Purdue University West Lafayette, IN March 2007 We thank Jason Abrevaya, Jennifer Altamuro, Anil Arya, John Fellingham, Chris Hogan, Marilyn Johnson, Rick Johnston, Yinghua Li, Scott Liao, Sam Pae, K. Ramesh, Doug Schroeder, Karen Sedatole, Eric Spires, Dave Williams, Rick Young, Helen Zhang, and workshop participants at Michigan State University, Ohio State University and the BKD Brown Bag Workshop at Purdue for many very helpful discussions, comments and suggestions. Bagnoli and Watts gratefully acknowledge the financial support of the Krannert School of Management and Purdue University. 1 Electronic copy of this paper is available at:

2 Family Firms, Debtholder-Shareholder Agency Costs and the Use of Covenants in Private Debt Abstract We ask whether the unusually close alignment of manager and shareholder interests in family firms is associated with increased use of restrictive covenants in private debt contracts. Our examination of Dealscan data indicates that S&P 500 family firms are more likely to include accounting-based covenants that limit the lender(s) risk that managers will divert cash or assets to shareholders than are S&P 500 non-family firms. The likelihood is further increased by presence of a dual class stock system that includes supervoting shares. This suggests that accounting numbers play an important role in mitigating debtholder-shareholder agency costs in family firms. JEL codes: G21, M41 Keywords: Debt covenants, Dealscan, Family Firms 2 Electronic copy of this paper is available at:

3 1. Introduction. Recent research indicates that family firms (companies in which the founders or their families exert significant influence through either their equity stake in the firm and/or their presence in senior management and/or on the board of directors) are surprisingly common. LaPorta et al. (1999), Claessens et al. (2000), and Faccio and Lang (2002) show that family firms are at least as prevalent as non-family firms in Europe and Asia, and Anderson and Reeb (2003a) report that in the U.S., one-third of Standard and Poor s (S&P) 500 firms can be classified as family firms. Further, Anderson and Reeb (2003a) provide evidence that family members tend to have a significant equity stake in their firms: On average, family members hold approximately 19% of their company s shares. To date, accounting researchers have focused on the impact of founding family ownership structure on shareholder-shareholder conflicts by studying how it affects the incentive to manipulate earnings (Ali et al. 2007, Wang 2006), and the way in which executive compensation contracts are tied to accounting numbers (Chen 2005). Finance researchers have focused on its impact on firm performance (e.g., Anderson and Reeb 2003a, Bennedsen et al. 2006, Villalonga and Amit 2006) and the cost of public debt (Anderson et al. 2003, Ellul et al. 2005). In this paper, we contribute to this literature by focusing on the debtholder-shareholder conflict in family firms and its effect on the use of financial covenants in private bank loans. Because we examine covenants that rely on accounting numbers, our paper contributes to the accounting literature in particular by assessing the usefulness of financial statement information in private debt contracting between family firms and their lenders. We focus on the use of covenants in private loan agreements for two reasons. First, private loans dominate the market for corporate debt. Dichev and Skinner (2002) report that private debt represents 80% of funded debt for their sample of large Compustat firms, a result that is consistent with Houston and James s (1996) estimate that only 17% of outstanding debt is public. Second, private debt, which includes syndicated loans, is both easier to renegotiate and easier to liquidate than public debt (Edwards 1986, Eichengreen and Mody 2000, Dichev and Skinner 2002, Allen and Gottesman 2005, 3

4 Altunbas et al. 2006). 1 As a result, private debt is more likely to include covenants designed to address a variety of agency problems (Smith and Warner 1979, Gilson and Warner 1998, DeAngelo et al. 2002). Family firms have an ownership structure that lends itself well to the examination of the agency costs arising from the debtholder-shareholder conflict and whether and how they affect the decision to include covenants in debt contracts. In particular, family firm ownership tends to be highly concentrated and family members tend to be relatively poorly diversified. According to Anderson and Reeb (2003a), family members have over 69% of their wealth, on average, invested in their firms, and 45.7% of family firms have founder or descendent CEOs. In addition, unaffiliated blockholders, who could serve as strong external monitors, tend to hold significantly smaller percentages of shares in family firms (Anderson and Reeb 2003b). Although the structure of family firms might mitigate certain agency problems (in particular, the owner-manager agency problem), Jensen and Meckling (1976) and Smith and Warner (1979) argue that when management s and shareholders interests are closely aligned, as they are in family firms, the conflict of interest between shareholders and debtholders gives rise to opportunistic behavior by managers in which debtholder wealth is expropriated. One method of mitigating this agency cost is through the use of covenants that protect lender(s) from the borrower using cash or assets in ways that increase the lender(s) risk. For example, liquidity and net worth covenants specify minimum limits on financial measures such as the current ratio, EBITDA, interest coverage, and the firm s (tangible) net worth and thus inhibit the transfer of cash and assets to shareholders. They also help to ensure that the current debt can be serviced, and if not, that there are assets available for the lender(s) to claim. Leverage covenants that restrict additional borrowing reduce the possibility of opportunistic borrowing by managers and thus help to maintain the value of existing debt by placing maximum limits on the debt already in place. 1 As Altunbas et al. (2006) observe, [c]ompared to a larger number of non-coordinated investors in the case of bonds, syndicated loans have a small number of relatively well organized lenders acting uniformly against any repayment problems. (p. 690) 4

5 We hypothesize that family firms are more likely than non-family firms to include such financial covenants in their private debt contracts. We further hypothesize that their inclusion is even more likely if the family s control over operations is increased through the use of dual-class shares and/or if a family member is CEO. 2 We also note that because renegotiation of private debt contracts is relatively inexpensive (Dichev and Skinner 2002, Allen and Gottesman 2005), even family members who have no history or intention of expropriating debtholder wealth (consistent with their building or maintaining reputation, Anderson et al. 2003) are more likely, relative to managers in non-family firms, to find it beneficial, in net, to tie their hands through the use of covenants and alleviate any concern of the lenders. 3 We test our hypotheses by studying the financial (liquidity, net worth and leverage) covenants included in 2,687 private loan packages of S&P 500 firms identified as family firms by Business Week in its November 10, 2003, issue. Business Week defines a family firm as: any company [in the S&P 500] where founders or descendants continue to hold positions in top management, on the board, or are among the company s largest stockholders. To identify these firms, Business Week relies on the methodology developed by Anderson and Reeb (2003a) but fine-tunes the process by examining individual firms in more detail, as required, to ensure that family members do in fact exert significant influence on company operations. The private loan packages of the other (non-family) S&P 500 firms serve as a control sample. We gather covenant and loan package details, including average interest rate spread over LIBOR and maturity, from the 2005 Dealscan database. We gather dual-class share information from the Investor Responsibility Research Center database, CEO data from proxy statements and 2 As discussed in the next section, firms with a dual-class share structure (aka a two-tiered stock system) have one class of common stock with superior voting rights (more than one vote per share) and another class with inferior voting rights (one vote per share). In the case of family firms with dual-class shares, family members generally hold the super-voting shares and thus have significantly greater control over operations when compared to other shareholders. 3 Dichev and Skinner (2002) conclude that private debt covenants provide lenders with an option to step in and take action when conditions warrant, and that violations do not necessarily indicate severe financial problems. However, they also note that including covenants in debt agreements is not costless. Doing so requires management time, frequent reporting to and monitoring by lenders, and renegotiation and the possibility of additional covenants in the event of violation (Dichev and Skinner 2002, Beneish and Press 1993). 5

6 other necessary data from Compustat. Because our sample is restricted to S&P 500 firms, we note for the reader that the slice of the private debt market we focus on is accessed by some of the largest and most stable firms in the U.S. economy. Further, the private debt in our sample is relatively short-term in nature: the average (median) maturity is approximately three (two) years. Our multivariate tests control for factors other than family-firm status that prior research indicates affect the inclusion of covenants in debt contracts: leverage, growth, firm size, earnings variability, firm credit rating, macroeconomic conditions as proxied by LIBOR, deal amount, maturity and status as a financial firm. In addition, we supplement our basic probit analysis with two-stage conditional maximum likelihood estimation (2SML) to control for endogeneity in the interest rate and the inclusion of covenants that adjust for firm-specific risk. The results are consistent with our expectations: The private loan packages of family firms are more likely than those of non-family firms to include at least one financial covenant. We also show that covenant intensity, as proxied by the number of financial covenants in a loan package, is significantly greater for family firms. Also as expected, the presence of dual-class shares, while not pervasive in our sample, further increases covenant intensity and the probability of observing at least one financial covenant in a loan package. When specific categories of financial covenants (liquidity, leverage and net worth) are considered, we find that family firm status significantly increases the likelihood of observing liquidity and net worth covenants being included in the private debt contracts of our sample firms. Further, the presence of dual-class shares has a significant incremental impact on the probability of observing liquidity covenants. These results suggest that covenants that restrict managers ability to divert cash or assets to shareholders to the detriment of debtholders are viewed as more beneficial, in net, for family firms than non-family firms, especially in the presence of a two-tiered stock system that strengthens the family s operational control. Interestingly, we do not observe a similar effect for leverage covenants that restrict additional borrowing. However, it is possible that because of the short-term nature of the debt instruments in our sample, the incentive problem addressed by leverage covenants (i.e., the incentive to issue more debt 6

7 that reduces the value of existing debt) may not be particularly severe for our sample firms. Overall, our analysis provides evidence that family firms are associated with greater use of financial covenants in private debt contracts, particularly when those covenants are designed to identify whether the family-firm managers are increasing the lender(s) risk by diverting cash or assets to shareholders. It also provides an indication of the value of accounting information in mitigating the agency costs that arise from the concentrated family ownership and control inherent in the structure of family firms. In particular, the availability of reliable measures of the firm s performance as presented in its financial statements appears to be very important in helping borrowers and lenders reduce these agency costs and contract more efficiently. Further, the importance of financial statements to family firms in contracting is consistent with the findings in Ali et al. (2007) and Wang (2006) that the financial disclosures of family firms are of higher quality and are less likely to be manipulated for opportunistic reasons. The rest of the paper is organized as follows. In Section 2, we review the literature and develop our hypotheses. In Section 3, we present the results of univariate and multivariate tests. In Section 4, we offer concluding remarks. 2. Motivation and Hypotheses Development. Bank loans are a significant source of financing for U.S. firms (Houston and James 1996, Denis and Mihov 2003). However, meeting the demands of their largest customers often requires banks to assume excessive risk or pushes them to regulatory limits (Allen and Gottesman 2005). 4 As a result, many borrowers and lenders turn to syndicated debt. Loan syndication allows banks to serve their largest customers while spreading risk and avoiding regulatory limits (Allen and Gottesman 2005). In syndication arrangements, the 4 According to Allen and Gottesman (2005), regulatory restrictions include the requirement for some U.S. banks that they not lend more than 25%, and sometimes 10%, of capital to an individual borrower and that they finance a significant portion of the loan with expensive sources of financing, reducing the profitability of the loan. 7

8 lead bank (the arranger) and the borrower set the terms of the loan (referred to as a credit facility or tranche), and then the lead bank finds other lenders (referred to as participants, usually other banks, insurance companies and other financial institutions) to join the syndicate. 5 Dichev and Skinner (2002) report that by the late 1990s, most sizable commercial loans in the Dealscan database were syndicated, a finding that they attribute to the increased prominence of this type of lending. Because of the way they are structured, private loans, whether they are syndicated or not, are much easier to renegotiate than public debt and therefore include more covenants (Edwards 1986, Eichengreen and Mody 2000, Dichev and Skinner 2002, Allen and Gottesman 2005, Altunbas et al. 2006). 6 Family firms have a unique ownership structure in that founders or their descendants hold senior management positions, sit on the board of directors, or are among the firm s largest shareholders (Anderson and Reeb 2003a). The family firm structure is common throughout the world: LaPorta et al. (1999) report that approximately 30% of large publicly traded firms worldwide are family firms. Anderson and Reeb (2003a) indicate that within the U.S., the structure is also common: Approximately one-third of the S&P 500 consists of family firms. Not only does ownership tend to be concentrated in such firms; family members also tend to be poorly diversified, with over 69% of their wealth, on average, invested in the firm (Anderson and Reeb 2003a). Further, unaffiliated blockholders (independent entities holding five percent or more of the firm s shares), that could serve as strong external monitors, have significantly smaller holdings in family firms when compared to non-family firms (Anderson and Reeb 2003a). Although some of the agency problems that arise in corporations might be mitigated by the structure of family firms in particular, the owner-manager (Type I) agency conflict, others could be exacerbated. In particular, when dealing with firms where ownership is concentrated in the family s hands and/or family members hold influential positions, lenders are likely to be particularly concerned about the debtholder-shareholder conflict of interest (Smith and 5 Private loan syndicate sizes range from 1 to 33, with an average of four lenders in a syndicate (Bradley and Roberts 2004). 6 Begley and Freedman (2004) provide evidence of a significant decline in the use of accounting-based covenants in public debt over the last fifteen years, a finding that is consistent with the difficulty of renegotiation when there are potentially thousands of bondholders. 8

9 Warner 1979). In such firms, there is a greater likelihood that management will act in ways that benefit shareholders at the lender s expense by, for example, expropriating lenders wealth either by altering the firm s investment decisions or by disgorging assets to shareholders (Tirole 2006, Grinblatt and Titman 2002). 7 Business Week s descriptions of family connections (i.e., ownership stake, senior positions, and influence) in S&P 500 family firms, as they define them, indicate that family members influence is indeed significant (see the Appendix for examples). As a result, we expect the debtholder-shareholder conflict described by Smith and Warner (1979) to be relatively severe in these firms. We also expect its severity to increase if the family s influence is increased through the presence of a family-member CEO and/or dual class stock, in which family members hold supervoting shares that give them significantly more control over important corporate decisions. 8 Including covenants in debt contracts is one method of protecting debtholders from the increased risk that managers will use cash in ways that lower the value of debt (i.e., increase risk), and thus we hypothesize that they are more likely to be included in the private debt contracts of family firms than non-family firms, particularly if the firm has dual class shares and/or a family member is CEO. Further, even if family members have no intention or history of taking such actions (consistent with, for example, their building or maintaining reputation, as argued in Anderson et al. 2003), they still might be relatively more likely to find it beneficial, in net, to tie their hands through the use of covenants because of the lender(s) s concern. As noted in the introduction, this trade-off is particularly salient in private debt contracts where renegotiation is relatively inexpensive. For example, Dichev and Skinner (2002) report violations of approximately one-third of the net worth and current ratio covenants in their sample of bank loans. This suggests to them that these 7 More specifically, lenders must be cautious about manager/family members investment strategies creating a debt overhang, asset substitution, short-sighted investment or a reluctance-to-liquidate problem (Grinblatt and Titman 2002, p. 563). They must also be cautious about stockholders diverting cash or assets toward share repurchases, dividend payments or extensive capital expenditures in illiquid assets or liquidating working capital to cover short-term losses. 8 A recent example of such a two-tiered stock structure is Google s initial public offering that included Class B shares that have 10 times the voting power of Class A shares. This dual-class share structure gave one-third of the voting power to the firm s two founders and its chief executive. See Fan and Wong (2002), Gompers et al. (2004), Hauser and Lauterbach (2004), Francis et al. (2005) and the references therein for more details regarding dual-class companies. 9

10 covenants act as monitoring devices or trip wires that provide the lenders with an option to step in and take action when conditions warrant. They also show that such covenants are generally set relatively tightly and that violations do not necessarily indicate severe financial difficulty, consistent with Chen and Wei s (1993) and Smith s (1993) observation that when covenants are violated, lenders often choose to not call the loan but instead opt for a less punitive alternative such as resetting the covenant. 9 As a result, covenants appear to be effective monitoring devices. The restrictive covenants that firms and their lenders might choose protect, albeit in different ways, against the borrowing firm s managers using cash or assets in a way that increases the lender s risk, consistent with their reducing the agency cost expected to be more severe in family firms. Liquidity covenants (that are based on, for example, the current or quick ratio and interest coverage) ensure that the firm s operations generate sufficient cash to service the debt. Similarly, net worth covenants (that are based on financial statement measures of net or tangible net worth) ensure that cash and/or assets are not diverted to shareholders but instead are used to service the debt or be available to (partially) repay the loan in the event of default. Because of the way these covenants are designed, we expect that both types will be more frequently included in the private debt of family firms relative to non-family firms. In addition, the arguments in Begley and Feltham (1999) suggest that leverage covenants (that are based on measures of the level of debt usually relative to some measure of earnings or assets/equity) might also be more frequently included in the debt contracts of family firms. Specifically, if lenders perceive family firms as being more likely to opportunistically increase the value of equity through future borrowing that also reduces the value of existing debt, then covenants that restrict future borrowing also lower agency costs and may be more attractive to family firms. However, we note that this problem might not be as severe for private debt, which tends to be short-term in nature (average maturity of approximately three years; see, for example, Dichev and Skinner 2002). 9 We want to emphasize that we are not implying that renegotiation is costless. It involves the time and cost of dealing with lender review, discussing forecasts and strategy with the lender(s) (Skinner and Dichev 2002), and possibly the addition of new covenants (Beneish and Press 1993). This means that borrowers and lenders trade off costs with benefits when deciding whether to include particular covenants in their debt contracts. 10

11 3. Sample and Data. Our sample consists of the loan packages, gathered from the 2005 Dealscan database, of the 177 S&P 500 firms identified as family firms by Business Week in their November 10, 2003, issue. Business Week defines a family firm as any company where founders or descendants continue to hold positions in top management, on the board or among the company s largest stockholders. Business Week bases its definition on Anderson and Reeb s (2003a) methodology and enlists the help of Spencer Stuart to identify the family firms. 10 The loan packages of the remaining S&P 500 firms as of June, 2003, serve as the non-family-firm (control) sample. Our sample period spans We use the 2003 Business Week family firm/non-family firm classification throughout our sample period based on the notion that family-firm status is sticky (Ali et al. 2007). 11 However, because CEO change is likely over our sample period, we examine proxy statements to determine whether a family member is CEO at the inception of each loan package in our sample. In addition, we use the Investor Responsibility Research Center database to determine whether the firm has a dual-class stock system in place at the inception of each loan package. We note for the reader that because of the way we identify family firms and the control sample, the slice of the private debt market that we examine is not representative of the entire private debt market. More specifically, our focus on S&P 500 firms means that we examine the private debt of the largest firms in the U.S. We identify loan facilities (tranches) and packages and gather interest rate spread over LIBOR (All-in-Spread Drawn), and other loan-specific and covenant information from the Dealscan database. All facilities in a loan package are covered by all of the covenants in the loan package in which they are contained, and so our unit of analysis is the 10 See Defining Family, Business Week, November 10, 2003, pp , for details. Business Week notes that it examined individual firms more closely, when necessary, in order to ensure that companies identified as family firms were in fact those whose operations were significantly influenced by family members. 11 Movement across classifications is generally from family-firm to non-family-firm status; thus, misclassifications from the use of the 2003 designation are likely to bias against our finding results that support our hypotheses. 11

12 package, not the individual facility. 12 We require that an observation have maturity and interest rate spread information in Dealscan in order to be included in our sample. We gather other required financial and firm-specific information not contained in the Dealscan database from Compustat and Investor Responsibility Research Center (IRRC) databases. Table 1 describes the way our final sample consisting of 2,687 loan packages, spanning the period , was formed. One-hundred-and-forty (79%) of the family firms and 275 (85%) of the non-family firms in the S&P 500 as of June, 2003, are associated with loan packages in our final sample. The types of covenants included in the loan packages for our sample as well as brief descriptions and break-down by specific covenants are presented in Table 2. (We code a covenant as existing if it is indicated as existing in the Dealscan database and not existing otherwise.) Table 2 shows that in our sample, financial covenants (liquidity, leverage level and net worth covenants) are frequent: 1,154 packages contain at least one financial covenant. Liquidity covenants (minimum limits placed on the firm s ability to service its debt obligations, N=1,118) are the most common financial covenant in our sample, followed by leverage level covenants (maximum limits placed on leverage, N=734), and then net worth covenants (minimum limits placed on the level of the firm s net worth, N=327). Of the liquidity covenants, minimum interest coverage occurs most frequently (N=1,125), followed by minimum debt service coverage (N=418). 13 (These subtotals indicate that loan packages in our sample often contain more than one liquidity covenant.) Of the leverage level covenants, maximum leverage ratio occurs most frequently (N=371), followed by maximum debt to EBITDA (N=302). Like Dichev and Skinner (2002), we observe variations in the ratios used in leverage and liquidity covenants, consistent with Leftwich s (1983) argument that private loan agreements are often tailored to borrower characteristics. These variations are not critical for our purpose (examining the existence of covenants in family-firm versus non-family-firm 12 Consistent with prior research, we find that a large number of deals in Dealscan are packages or bundles of facilities (e.g., a one-year line of credit and a longer term note). On average, loan packages of S&P 500 firms contain 1.4 facilities. (The averages for S&P 500 family and non-family firms are also 1.4.) Dealscan data is organized so that each facility has a unique identifier of its own as well as a package identifier. 13 Some packages have multiple facilities with different interest coverage ratios. 12

13 private loans), and so we focus on the broader categories of financial covenants (liquidity, leverage and net worth) as opposed to specific covenants in the empirical analysis that follows. 4. Empirical Results Univariate Tests. Table 3 presents univariate comparisons between family firms and non-family firms and their loan packages. Table 4 contains Pearson and Spearman correlations among most of the variables examined in Table 3. Panel A of Table 3 shows that family firms associated with the loan packages in our sample are not significantly different from the non-family control firms in terms of leverage (long-term debt scaled by total assets), consistent with Anderson and Reeb (2003b), and growth (book-to-market). However, these same family firms are significantly smaller in terms of total assets, on average, than the non-family firms (p=0.000), and there is some evidence that their earnings are more volatile than those of non-family firms (p=0.062). The percent of family-firm loan packages where the borrower is a financial institution (SIC ) is smaller than it is when the nonfamily-firm loan packages are considered: 10.8% versus 13.7%. Further, family firms associated with the loan packages in our sample have, on average, a slightly worse credit rating than the non-family firms (3.743 versus on a 1 to 7 scale where 1=AAA credit rating and 7=below B credit rating see the footnote to Table 3 for details), although the median credit ratings are the same for both groups and the statistical significance of the difference in means is marginal (p=0.083). 14 They are also more likely to have dual class stock systems (13.4%) than the non-family firms (4%), although we note that such systems are not pervasive in our sample. The purpose of the majority of facilities in our sample loan packages (approximately 60% of the family firm facilities and 76% of the non-family firm facilities) is for everyday use (e.g., general operations, short-term purchases, commercial paper backup). Less 14 We use firm credit rating, instead of issue-specific bond ratings, because it is a measure of the rating agency s assessment of the firm s inherent default risk rather than the default risk of a particular debt instrument, which can depend in part on the type and number of covenants attached. 13

14 than 15% are for the purpose of debt consolidation or refinancing. Consistent with prior research that examines private debt from Dealscan (e.g., Dichev and Skinner 2002), average loan package maturity in our sample (based on the maturity of the longestmaturity facility in the package) is approximately three years. 15 In addition, we find no statistical difference in mean maturities for family and non-family firm loan packages, N=851 and 1,836, respectively. When we weight each facility s maturity by its relative proportion of the deal amount to arrive at a weighted average maturity for each package, the means for both types of firms increase slightly (35.2 months versus 34.6 months), but are still statistically indistinguishable from one another. 16 Interestingly, the average deal amount for family firms (7.45) is significantly smaller than the average deal amount for non-family firms (11.4). In addition, the mean simple average spread (across facilities in the package) in terms of basis points over LIBOR is higher for family firms ( bps) than for non-family firms ( bps), which is consistent with the somewhat lower average credit rating for family firms in our sample. Even though this difference in spreads is statistically significant, we note that it is small in economic terms: the mean (median) difference in average spreads across the two groups is only 0.2% (0.1%). Further, when we weight the spread in each facility by its relative proportion of the deal amount to arrive at a weighted average spread for each package, the means for family and non-family firm packages change by very little ( bps versus bps). Consistent with our expectations, the covenant frequency information in Panel B of Table 3 indicates that significantly larger percentages of family firm loan packages include covenants than non-family firm loan packages. Specifically, 47.8% of family firm loan packages contain at least one financial covenant, compared to 40.8% of non-family firm loan packages. In addition, covenant intensity, as measured by the number of covenants in a loan package, is also significantly greater for family firm loan packages: 1.257, on average, for family firm packages versus 0.941, on average, for non-family firm packages. When specific types of covenants are considered, the same pattern 15 Approximately three-fourths of the family and non-family firm facilities are 364-day facilities or revolvers. 16 This weighting will provide some control for differences in maturity of the dominant facilities in each package. Although the means are not statistically different from one another with this weighting scheme, the medians significantly change, especially for the non-family firm packages. 14

15 continues to hold: 45.9% (39.6%) of family firm (non-family firm) loan packages contain liquidity covenants; 30.1% (26.0%) contain leverage covenants; and 15.9% (10.4%) contain net worth covenants. These univariate comparisons provide an initial indication that family firms are more likely to include covenants of all kinds and more of them in their private loan packages than are non-family firms, consistent with their being useful in reducing the agency costs arising from the potentially more severe debtholder-shareholder conflict in family firms. However, because factors other than ownership structure affect the decision to include covenants, we next turn to multivariate tests that control for other influencing factors. 4.2 Basic Models. Panels A and B of Table 5 contain in the results of basic probit regressions designed to predict the probability of at least one financial covenant being included in a single loan package and the Poisson and OLS regressions designed to explain covenant intensity (the number of covenants included in a single loan package), respectively. More specifically, the full probit model in Panel A is: Prob (at least one financial covenant) = F (β 0 + β 1 Family Firm Dummy i + β 2 Leverage i (1) + β 3 Growth i + β 4 Firm Size i + β 5 Earnings Variability i + β 6 Firm Credit Rating i + β 7 LIBOR i + β 8 Financial Firm Dummy i + β 9 Dual Class Dummy i + β 10 Deal Amount i + β 11 W.A. Maturity i + β 12 W.A. Spread i + ζ i ) The full Poisson and OLS models in Panel B are: Covenant Intensity Index i = α 0 + α 1 Family Firm Dummy i + α 2 Leverage i + α 3 Growth i (2) + α 4 Firm Size i + α 5 Earnings Variability i + α 6 Firm Credit Rating i + α 7 LIBOR i + α 8 Financial Firm Dummy i + α 9 Dual Class Dummy i + α 10 Deal Amount i + α 11 W.A. Maturity i + α 12 W.A. Spread i + ε i We also estimate the models without the Dual Class Dummy to see whether our results are sensitive to its inclusion. Analogous models that include an Outsider CEO Dummy are presented in Panels A and B of Table 6. 15

16 The dependent covenant variable in the probit models is dichotomous; and the covenant intensity index, as mentioned above, is a simple count of the number of financial covenants included in the package. The explanatory variables of interest to us are the Family Firm Dummy variable, which takes on the value of 1 if the borrower is a family firm and 0 otherwise (included in both Tables 5 and 6); the Outsider CEO Dummy variable, which takes on the value of 1 if the firm s CEO is not a family member and 0 otherwise (included in Table 6 only, with the Family Firm Dummy variable); and the Dual Class Dummy which takes on the value of 1 if the firm has a dual class stock system (included in one version of each model in both Tables 5 and 6). Control variables are those identified in prior research (e.g., Begley and Feltham 1999) as being associated with the presence of covenants: (1) Leverage as measured by ratio of long-term debt to total assets as of the beginning of the quarter in which the loan is initiated; (2) Growth as measured by the book-to-market ratio; (3) Firm Size as measured by the log of total assets; (4) Earnings Variability as measured by the standard deviation of the ratio of annual operating income to total assets using data from the prior three years; (5) Firm Credit Rating as measured by the S&P long-term domestic issuer credit rating and coded as 1, 2,, 7 where 1 = AAA rating which is the highest rating and 7= below B which is the lowest rating; (6) LIBOR as measured by the 12-month U.S. dollar denominated LIBOR rate as of June 30 of the current year; (7) Financial Firm Dummy which takes on the value of 1 if the firm is a financial institution (SIC ) and 0 otherwise; (8) Deal Amount as measured by the size of the loan deal; (9) W.A. (Weighted Average) Maturity as measured by the sum of the maturities of each facility in the package weighted by that facility s proportion of the total deal amount; and (10) W.A. (Weighted Average) Spread as measured the sum of the amounts the borrower pays in basis points over LIBOR for each faculty in the package weighted by that facility s proportion of the total deal amount Recall that a firm s credit rating is a measure of the rating agency s assessment of the firm s inherent default risk. The advantage of using this rating versus a bond rating is that the latter is affected by the covenants attached to the debt contract (Weber 2006). 16

17 We expect Leverage, Earnings Variability, Firm Credit Rating, Deal Amount and W.A. Maturity to be positively associated with the inclusion of covenants since they indicate higher risk for lenders. We expect Firm Size, Growth and LIBOR to be negatively related to the inclusion of covenants because larger firms (in terms of total assets) and non-growth firms (as reflected in high book-to-market ratios) are likely to be less risky from the lender s perspective, and because LIBOR tends to rise (fall) with better (worse) economic conditions and thus provides a control for the effect of general economic conditions on the probability of including covenants. (We expect a negative sign on the coefficient for LIBOR because worse economic conditions can be reasonably expected to increase the probability of a covenant, Bradley and Roberts 2004.) We include a financial firm dummy as a control variable because financial firms have a different capital structure than non-financial firms and are often subject to additional regulation. As a result, their propensity to include covenants may be different from that of nonfinancial firms. We also expect the weighted average interest rate spread over LIBOR (W.A. Spread) to be related to the inclusion of covenants, but we are unable to predict the sign of the relation. It could be the case that there is a trade-off between spread and the presence of covenants as alternative means of protecting against risk, creating a negative relation; but it could also be the case that even in the presence of such a trade-off, loans with higher spreads are also more likely to include covenants because of higher perceived risk, creating a positive relation between the two. 18 The results of the basic probit regressions in Panel A of Table 5 indicate that family firm status significantly increases the probability of at least one financial covenant being included in private debt, consistent with our expectations, and that the presence of dual class shares further increases the probability. (Coefficients for both variables are positive and significant at p<0.01.) The conclusion drawn from these initial results that family firms are more likely than non-family firms to include financial covenants in their private debt contracts is further supported by the evidence in Panel B of Table 5. In both the 18 That is, since higher spreads and covenants are used together to reduce the lender s risk, using them in combination should result in riskier borrowers loans having higher spreads and more covenants. The results for the family firm variables are not sensitive to our using a simple average versus weighted average spread variable. 17

18 Poisson and OLS regressions, family firm status is positively and significantly associated with the covenant intensity index, a result that indicates that family firms tend to have greater numbers of financial covenants in their private debt packages than non-family firms. And, once again, the presence of dual class shares has significant incremental explanatory power beyond that of the family firm variable. This suggests that when family control is strengthened through the unusual voting rights afforded by dual class shares, additional covenants are generally placed on the firm s private debt. In Panel C of Table 5, we consider the effect of family firm and dual class share status on the inclusion of specific types of financial covenants (liquidity, leverage and net worth). The results of the basic probit estimations show that family firm status increases the probability of liquidity and net worth covenants being included in private debt, consistent with our expectations. However, it does not appear to be significant in determining the probability of leverage covenants. Further, the presence of dual class shares increases the likelihood of liquidity covenants but is not incrementally significant in explaining the probability of leverage or net worth covenants. These findings, when considered together, indicate that after controlling for other influencing factors, family firms are more likely to include covenants that monitor inflows or outflows of cash not associated with additional borrowing in their private debt contracts than are non-family firms, and that dual class stock structures are important determinants of the inclusion of liquidity covenants in particular. One possible explanation for the lack of family-firm influence on the inclusion of leverage covenants is that the private debt contracts we examine are relatively short-term in nature, making the ability to service the debt more important than the restriction of additional borrowing. (Not only is the average maturity of the facilities in our sample short; as noted earlier, a significant percent of the facilities in the packages in our sample are short-term.) As explained in the previous section, the agency problem that financial covenants address might well be mitigated if the chief executive officer is not a family member. Table 6 reports the results of the same models presented in Table 5 except that both a Family Firm Dummy and an Outsider CEO Dummy are included as explanatory variables. (We 18

19 also continue to include a Dual Class Dummy variable in one version of each model.) The coefficient for the Family Firm Dummy variable continues to be significantly positive and of roughly the same magnitude in the basic probit regressions (for the inclusion of at least one financial covenant) in Panel A and in the Poisson and OLS regressions (to explain covenant intensity) in Panel B. The Dual Class Dummy also continues to be positive and significant in all of the regressions in which it is included. However, the Outsider CEO Dummy is not statistically significant in any of the regressions in Panels A and B. When we consider individual types of covenants (Panel C), the Family Firm and Dual Class Dummy variables continue to be positive and significant for liquidity covenants, and they continue to have no significant impact on the inclusion of leverage covenants. Consistent with what we find in Panels A and B, the Outsider CEO Dummy variable is not significant in either the liquidity or leverage covenant model. However, we observe some changes in the net worth covenant regressions: The Outsider CEO Dummy has a significant, negative impact on the inclusion of net worth covenants (consistent with non-family control in the CEO position mitigating the agency costs that might also be reduced by including a net worth covenant), and the Dual Class Dummy variable is marginally significant and negative. Even so, the results in Table 6, taken together, do not provide strong support for an outsider CEO having a significant impact on the inclusion of financial covenants in private debt contracts. One possible explanation for this is that Business Week s family firm classification scheme is designed to ensure significant family influence, regardless of its source, making it difficult to pick up an incremental effect for an outsider CEO in the regressions. Alternatively, significant family firm influence may simply mitigate the potential benefits of an outsider CEO. As expected, several of the control variables (Firm Size, Firm Credit Rating, LIBOR, Deal Amount) are significant and of the predicted sign regardless of the covenant measure considered and the regression that is estimated. In addition, the coefficient for W.A. Spread is generally positive and significant, suggesting that loan packages with higher spreads are more likely to include all types of financial covenants. The Growth and Earnings Variability control variables, however, are statistically insignificant in each 19

20 of the regressions, consistent with the weakness of the results of the univariate tests in Table 3. Leverage is significant and of the expected sign in all regressions except the net worth probit regressions. W.A. Maturity is either insignificant, or in the case of liquidity covenants and the probit regressions for the inclusion of at least one financial covenant, significant and negative. Again, these results could be due to the short-term nature of the loan packages that we examine. To summarize: Although there is some variability in the significance of the control variables, the basic models in Tables 5 and 6 provide strong evidence that family firms are more likely than non-family firms to include financial covenants, especially liquidity and net worth covenants, in their private debt agreements, and that the presence of a two-tiered stock system increases the likelihood their inclusion. The models also indicate that family firms have more financial covenants, on average, per loan package than do non-family firms SCML Estimation. As is true for other studies that examine the likelihood of covenants being included in debt contracts, we face an endogeneity problem. In particular, it is likely that the inclusion of covenants and the interest rate spread are jointly determined in the negotiations between borrower and lender. Before proceding, we note for the reader, as did Begley and Feltham (1999), that many of the other explanatory variables that are treated as exogenous in studies such as ours are also the outcomes of firm decisions. We focus on interest rate spread as the endogenous explanatory variable in our probit estimations because it is the one for which the endogeneity problem is likely to be most severe in the context of a loan package. As a result, we check the robustness of our basic results by using two-stage conditional maximum likelihood estimation (2SCML). In a system of two equations where one equation has a dichotomous variable and the other has a continuous variable on the left-hand side (as will be the case for us), Rivers and Vuong (1988) show that 2SCML produces consistent estimates. In our 2SCML estimation, the first stage for each type of covenant involves estimating a reduced form regression with weighted average spread as the dependent variable and all 20

21 explanatory variables from the basic probit regressions in Tables 5 and 6 except the (endogenous) W.A. Spread variable, plus a dummy variable for whether the debt is secured and another for the existence of other covenants as independent variables. This structure assumes that the interest rate charged for the debt is likely to be affected by whether the debt is secured (Asquith et al. 2005, Beatty et al and the references therein) and by the use of other covenants. In the second stage, we include the residuals from the first-stage regression as well as the (endogenous) W.A. Spread variable in a probit model for the covenant measure we are currently examining. An advantage of this procedure is that it allows for the testing of endogeneity through comparison of the log of the likelihood functions calculated with and without the residuals in the second stage (Alvarez and Glasgow 2000, Timpone 2003). We present the second-stage probit estimation results in Tables 7 and 8. Table 7 contains the estimation results with the Family Firm Dummy variable only, and Table 8 contains the estimation results with both the Family Firm and Outsider CEO Dummy variables. The results are largely consistent with the results from the basic models. In particular, family firm status continues to be associated with an increase in the probability of at least one financial covenant being included in a private loan package, as does the presence of dual class stock. The presence of an outsider CEO does not change the probability. The results for specific types of covenants are also generally consistent with the prior results with the following exceptions. The Family Firm Dummy variable loses significance (p= 0.139) in the liquidity covenant probit in Table 8 (however, the Dual Class Dummy retains its significance and positive sign); and the Outsider CEO Dummy variable becomes significant and has a negative sign in the net worth covenant regressions in Table 8. Likelihood ratio tests indicate significant endogeneity in all of the probits, and consistent with this, the spread residual is significant in each regression in Tables 7 and 8. This also indicates bias in the coefficients estimated in the basic probit regressions in Tables 5 and 6 (Rivers and Vuong 1988, Greene 2000). Consistent with this, a comparison of the coefficients in Table 7 (8) with those in Table 5 (6) reveals some differences. Firm size and deal amount continue to be consistently significant and of the predicted sign for all the types of covenants in Table 7. However, the leverage variable 21