Debt Boundaries Matter: Evidence From The Subsidiary Debt

Transcription

1 Debt Boundaries Matter: Evidence From The Subsidiary Debt January 15, 2018 Abstract I exploit the introduction of an accounting reform in the US to investigate whether the presence of subsidiary debt affects the cost of borrowing of conglomerates. The accounting reform forces some firms to restate from standalone firms (declaring one segment unit) to conglomerates (declaring multiple segment units). I find that restating conglomerates suffer a 16% increase in the bond spread if they do not incorporate some debt on their segment units as separate legal entities, or subsidiaries. The effect is concentrated on firms with a high dispersion in the growth opportunities in the prereform period. This is consistent with the hypothesis that subsidiary prevents managers of low-growth divisions to capture part of the surplus of high-growth divisions when divisions differ in resource potential. JEL Classification: G12, G15, G30, L22. Keywords: Conglomerates, cost of debt, corporate socialism, subsidiary debt, internal capital markets. 1

2 1. Introduction Conglomerates that is, firms organized in units operating in several industries can incorporate their divisions as subsidiary firms that issue their own debt, instead of issuing debt at the parent-firm level. While the past literature devotes considerable attention to the relationship between corporate diversification and firm value [Lang and Stulz (1994); Berger and Ofek (1995); Campa and Kedia (2002)], it has not considered the role of subsidiary debt in the cost of borrowing of conglomerates. This is surprising, as subsidiary debt is economically relevant, representing 13% of the total corporate bond issues on the primary US bond market [Kolasinski (2009)]. The goal of this paper is to investigate the relevance of subsidiary debt for the borrowing cost of conglomerates. The unique characteristic of conglomerates, compared to standalone firms, is their ability to reallocate funds from one division to another. Several papers show that such reallocation generates agency problems when divisions differ in resource potential [among others: Scharfstein and Stein (1998); Rajan et al. (2000); Hoang and Ruckes (2015)]. However, conglomerates have an additional (mostly neglected) opportunity: to allocate some debt at the division level. This paper tests how the use of subsidiary debt affects the sensitivity of firm borrowing costs to the internal allocation of capital of conglomerates. I use a difference-in-difference methodology on a sample of corporate bonds issues from 1990 to 2003 to compare the cost of borrowing of conglomerates with and without outstanding subsidiary debt with the cost of borrowing of standalone firms. The treatment effect is generated by an accounting reform that changes the definition of a business unit (also called division or segment). This change forces some firms to restate the number of units according to the new definition introduced by the rule change, and to reveal detailed financial information of these new business units. Crucially, the reform also reveals the relative characteristics of business units that do not issue debt with respect to the business units that are separate legal entities and issue debt, namely the subsidiary firms. I exploit this information to compare the cost of borrowing of conglomerates, separately splitting conglomerates with and without subsidiary debt outstanding, with the cost of borrowing of standalone firms-not affected by the reform. 2

3 The first result of this paper is that bondholders charge a higher yield spread to the bonds of firms that restate from standalone (one single business unit) to conglomerates (several business units) after the reform. The latter firms experience an increase in the yield spread on new bond issues by 25 bps when compared to standalone firms. This is statistically and economically significant, as it represents 16% of the standard deviation of the yield spread in my sample and an increase in annual interest expenses of an average of $4m for each bond issue. The second and main result of this paper is that the effect of the reform is concentrated on conglomerates that never had subsidiary debt in their capital structure. I find that restating conglomerates without outstanding subsidiary debt experience an increase in the yield spread on the bond issues by 27 bps after the reform. Conversely, restating conglomerates with outstanding subsidiary debt do not experience a significant increase in the yield spread after the reform. These results show that the allocation of debt between the parent firm and its subsidiaries is not irrelevant to the cost of borrowing of conglomerates. To understand the economic channel making subsidiary debt beneficial for debt pricing, I investigate the pre-reform characteristics of firms that restate as conglomerates and do not have outstanding subsidiary debt. I find that conglomerates without subsidiary debt but with a low level and a high variability of market-to-book value in the pre-reform period suffer a significant increase in the yield spreads (35 bps) when compared to standalone firms. This is consistent with the investors anticipating transfers from high-growth divisions to low growth divisions [Rajan et al. (2000)] making them reluctant to provide low-price funding to conglomerates. The subsidiary debt disciplines divisional managers, either by forcing them to make fixed debt service payments or by reducing the available cash for transfers. The results are robust to alternative channels, model specifications, and dependent variables. Consistent with the notion that subsidiary debt affects the credit risk of conglomerates, I find that the pricing effect of the rule-change is not explained by a quantity effect, by a change in any other bond characteristics (e.g., securitization, maturity), or by any confounding factor contemporary to the reform. Consistent with the investors aware of an agency problem after the rule change, I also find 3

4 that the amount of covenants required by investors on the debt issues of restating conglomerates increases by 18% after the reform. To the best of my knowledge, this is the first study to establish a link between the cost of borrowing of conglomerates and the issuance of subsidiary debt. The paper close to mine is the work of Hann et al. (2013), who find an inverse relationship between the cost of capital and corporate diversification because of coinsurance, as predicted by Lewellen (1971). Similarly, Franco et al. (2016) show that a more detailed segment reporting decreases the cost of borrowing of conglomerates by 35 bps, because it provides valuable information to the investors. My paper supports the importance of segment disclosure for the pricing of the debt of conglomerates, but it shows that the relationship between conglomeration and cost of borrowing is not necessarily negative. To the extent that most of the literature on conglomerates concentrates on the efficiency of internal capital markets [see Maksimovic and Phillips (2013) for a complete overview], my paper can reconcile past findings of a contemporary firm discount and low cost of borrowing, because it shows that a neglected variable- how the parent firm allocates the debt across business units - is a further determinant of the cost of capital of conglomerates. The paper also complements the results of Berger and Hann (2003), who finds that the discount for these hidden diversifiers increases in the post-reform period. Kolasinski (2009) is the first author to study the external debt issued by operating subsidiaries of conglomerates. He finds that firms with high variability in the growth options across segment units are more likely to issue subsidiary debt, and he explains this result with Jensen and Meckling (1976) asset substitution problem at the parent level. I do not find evidence that subsidiary debt mitigates asset substitution problems inside conglomerates when looking at the risk of the subsidiary debt. Instead, I find that the reform is detrimental only for firms that decide to not issue subsidiary debt and still have a high variability over time of the growth options. I, therefore, contribute to these findings with a novel empirical strategy that addresses the concern that financing costs and corporate diversification are jointly determined. The economic channel of managers incentives and agency costs in internal capital markets also relates to the work of Inderst and Müller (2003). They build a theoretical model to show that the use of decentralized (subsidiary) debt in conglomerates facilitates the access of conglomerates to financial 4

5 markets, by mitigating the severity of information asymmetries when the cash flow cannot be verified. My paper confirms their predictions, as it shows that debt decentralization can be beneficial from investors perspective in particular when conglomerates face financing constraints and contemporary reveal new information about their internal capital market. Finally, the paper relates to the empirical literature on the determinants of debt pricing [Collin- Dufresne et al. (2001); Fama and French (1992)], and it shows that subsidiary debt matters. The remainder of the paper is organized as follows. Section 2 describes the reform and the testable hypotheses. Section 3 presents the data, the variables, and the empirical strategy. Section 4 presents the empirical results, and Section 5 reports the robustness tests. Section 6 concludes. 2. Institutional Background and Hypotheses Development 2.1 The Segment Reporting Reform: Revealing Corporate Diversification Several studies assess the impact on a firm s information environment on their cost of capital by investigating changes surrounding accounting regulation [Easley and O Hara (2014)]. Because bondholders rely on publicly available information, a change in segment disclosure that reveals information about a firm s organizational structure is also crucial to assess firm risk [Franco et al., 2016]. Segment reporting obligations started in 1970 with the issuance of the Financial Reporting for Segments of Business Enterprise n. 14 ( Statement of Financial Accounting Standards - SFAS14). According to this regulation, a company should reveal information about revenues, assets, depreciation and capital expenditures for each industry segment representing 10% of firm revenues. The definition of a segment unit under the SFAS14 follows the industry approach. Specifically, it is the firm itself that identifies and indicates the industry of each segment unit. This approach has some limitations. The FASB Research Report conducted in 1992 shows that only 25% of 6,935 public companies indicated that their business involved more than one segment in the period from 1985 to Taking a closer look at the 1,035 companies with sales over $1bn, 43% of them declared themselves to be one-segment firm under the SFAS14 segment regulation. 5

6 On December 15, 1997, FASB issued a new regulation and replaced the SFAS14 with the new standard SFAS131. The new regulation changed the parameters for the identification of a segment unit. The reform replaced the industry approach with the management approach. The management approach implies that the identification of a segment unit follows the management organization of lines of business. The underlying assumption is that the management has superior information about the risk and the investment options of a segment unit. Therefore, according to the new definition, an operating segment is defined as a line of business of the company that generates at least 10% of the total group revenues, reviewed on regular basis. 1 As a direct consequence of this new approach, a significant number of firms (20%) restated from standalone firms (declaring one segment unit previously collapsed under an industry broad definition) to conglomerates (declaring multiple segment units). Berger and Hann (2003) show that firms reveal previously hidden information about their diversification strategies that affected their market valuations. Figure 1 reports the complete sample of all segment-restating firms in Compustat. In panel A, I report the number of conglomerates from 1995 to Figure 1.A shows that a remarkable number of firms restated as conglomerates in 1998 and in In Panel B of Figure 1, I report the number of segments after the reform for each firm that restates from standalone to conglomerate firm. The average restatement is from one segment to three segment-units, but some firms restate from one to nine segment units. Figure 2 reports an example of the new information provided by a restating firm in my sample: Sysco Corporation. Before 1998, the firm declared to operate in the business of Wholesale-Groceries and Related Products - SIC Panel A reports some financial information disclosed to the market before After 1999, Sysco declares to operate in three businesses, named Broadline, SYGMA, and Other (segments 1, 2 and 3). Panel B reports the same information of 1 Report available at 2 Figure 1.A shows a reverse tendency in the second half of 2000, when the number of conglomerates drops due to strategic merger started in 1993 and culminated in 2000 in the industries of electronics, communications, IT, and utilities [Ljungqvist and Wilhelm (2003)]. I analyze these issues in my robustness tests. 6

7 panel A after the restatement about the three segment units. The segment number 2 (SIC Meats and Meat Products), for example, invests less than its cash flow after the reform, while the segment number 3 (SIC Groceries and Related Products) invests more than its cash flow (therefore, it received funds from headquarters) after the reform. I now analyze how this information affects the cost of borrowing of conglomerates, conditional on the decision to allocate some debt at the subsidiary firm level. 2.2 Organizational Models and Debt Issues: Decentralized versus Centralized Debt In a Modigliani-Miller framework, the cost of debt only depends on a firm s risk class. A firm s organizational structure per se does not matter. It is, therefore, equal across conglomerates and standalone firms, provided they have the same overall risk. Also, the allocation of debt between a parent firm and its subsidiaries does not matter for the pricing of the debt. In a Pure Conglomerate Structure, the mother company is a multi-segments firm with segments engaged in different industries but raises debt only at the headquarter level. In this case, the mother company borrows funds and advances them to the operating companies. Since there are not creditors on the operating firm s level, the creditors of the mother company have full recourse to the operations in case of default. In a Parent-subsidiary Structure, the parent firm owns one (or several) operating company, and both the parent and its subsidiary firms issue some public debt. In most cases, the subsidiary debt funds a project that needs to get separate funding with respect to the other firm investments. Absent any guarantee, the parent firm enjoys limited liability for the subsidiary debt, as it is a separate legal entity. In a pure conglomerate structure, the new information provides a more clear picture of the transfers across segments units. The effect of the reform on the cost of debt of firms depends on the ability of conglomerates to allocate resources more efficiently compared to the standalone firms [Matsusaka and Nanda (1997); Gopalan et al. (2007)]. The redistribution of funds from segments with high investment opportunities to segments with low investment opportunities [ corporate socialism, Rajan, Servaes, and Zingales (2000)] may reduce firm value and profitability, and thus increase the risk of default. In addition, the divisions endowed with consistent amounts of cash may generate 7

8 incentives for the managers to overinvest [Jensen (1986)]. 3 The effect of the reform can have two directions. If the reform reveals new information about the degree of coinsurance across segment units with good investment opportunities [Khanna, N., and S. Tice (2001)], investors charge a lower cost of borrowing to restating firms after the reform. In contrast, if the reform reveals new information about transfers from high-growth segments to low-growth segments, or about poorly performing business units hidden before the reform, investors charge a higher cost of borrowing to restating firms after the reform. I sum up my reasoning in the following hypothesis: H1: Restating conglomerates have a higher (lower) cost of borrowing compared to standalone firms after the reform. In a parent-subsidiary structure, the new information provides a more clear picture of the transfers across segments unit that do not issue debt, because subsidiaries disclose financial information to investors already before the rule change. It follows that the effect of the reform can be biased because of different incentives in place between units issuing debt and units internally subsidized. As pointed in Kolasinski (2009), subsidiary debt commits subsidiary CEO to invest in projects that are more profitable, and to avoid cash-transfers from stronger to weaker segments. Subsidiary debt may prevent that managers of low-growth divisions capture part of the surplus of high-growth divisions, because it ensures that the cash flow is committed to subsidiary creditors instead of being transferred to other segment units. In addition, subsidiary debt may mitigate the empire-building temptation of the CEOs, because it avoids excessive cash pooling across divisions. This is consistent with Gertner et al. (2002), who find that capital allocation of conglomerates becomes more sensitive to their investment opportunities following a spin-off, and with Inderst and Müller (2003), who show that the benefits of the subsidiary debt depend on the probability of obtaining high cash flows from the different units of the conglomerate. 3 Consistent with this hypothesis, Glaser et al. (2013) find that there is a negative correlation between managerial power and the Tobin s q of the business unit. 8

9 Another stream of literature, however, argues that the subsidiary debt can increase the moral hazard problems inside the group. Mother companies can be tempted to raise the debt and move the riskier assets at a subsidiary level [asset substation as postulated by Jensen and Meckling (1976)] in order to keep the other affiliates safer in case of distress [Flannery, Houston, and Venkataraman s (1993); Khan and Winton (2005)]. These incentives prevailing, investors require higher yield spreads on the debt of the corporation. I sum up my arguments in the following hypothesis: H2: Subsidiary debt reduces (increases) the costs of debt of restating conglomerates, because it reduces (increases) the moral hazard costs. 3 Data and Methodology 3.1 Data and Sample Selection Bond data. I use FISD-Mergent to identify all public domestic bonds issued between 1990 and 2003 on the primary US bond market. Following previous literature, I exclude financial and utility firms. I also exclude asset-backed, convertible, warrants, and floating-rate bonds. I obtain the information on bond maturity, rating, seniority, and offering amount for each bond. 4 I exclude bonds that do not have information on the credit rating. For each bond observation, the dataset reports the identity of the ultimate parent of the bond issuer. I identify the subsidiary bonds when the identity of the ultimate parent is different from the identity of the issuer. I manually check the subsidiaries ownership in my sample, and I find that 98% of them are private firms (fully owned). My variable of interest is the yield spread of the bond at the time of issuance, computed as the difference between the bond yield and the corresponding yield on Treasury bonds with the same timeto-maturity. This variable captures the cost of borrowing of firms on the bond market. Bond characteristics include time-to-maturity (logarithm of months to maturity), the logarithm of the bond amount, and the debt seniority. I control for differences in default risk using credit ratings (Rating), where I use the average rating of Moody s, S&P, and Fitch. For each parent firm in my sample, I 4 As covenant protection is affected by the reform, it is not included in the difference-in-difference estimation. 9

10 retrieve the consolidated financial ratios from Compustat North America. I require not missing information for leverage, size, ROA, market-to-book ratio, and firm cash-flow volatility for the firms in my sample. Details of the variables construction are in Appendix A. Matching with restating conglomerates. The Compustat-Historical Segments database provides detailed financial information about each segment unit of conglomerates. I identify firms that work in multiple industries (declaring multiple segment units) after the reform and in a single industry (declaring one segment unit) before the reform. I merge this information with the FISD-Mergent dataset, and I identify the bonds issued by restating firms to assemble my treated group. Firms that declare one segment unit before and after the reform compose my control sample. In a final step, I identify restating conglomerates with and without outstanding subsidiary debt along the overall sample. After merging these data, I have 494 remaining bond observations (240 firm-year observations) issued by firms that restate as conglomerates after 1998, and 912 bond observations (678 firm-year observations) of standalone firms (control group). The sample of treated (restating) firms represents 15% of the total number of conglomerates that issue debt on the primary bond market. This number is in line with the proportion of firms that restate as conglomerates in Compustat (20%). Of the 494 bond issues of restating conglomerates, 218 bonds belong to firms with outstanding subsidiary debt (56 firm-year obs.), while 276 belong to firms without outstanding subsidiary debt (184 firm-year obs.). In term of firms, 86 firms belong to the treated group and 142 firms compose the control group, for a total of 228 firms. The number of bonds of treated and control groups is spread across the years with the exception of a peak in 2001 for bonds of conglomerates with outstanding subsidiary debt, due to a federal aid to the firms in the construction industry. In 2001, a federal aid of $26bn is provided to the construction industry, thus generating an increase in the debt issued in this industry. 5 I exclude these bonds in my robustness tests to control whether this abnormal peak drives my results. Figure 3 - Panel A - reports the distribution of my treated sample across industries. In Panel B, I report the industry composition of the control group. Overall, both treated and control groups are widespread across all industries. 5 Report available at census.gov/library/publications/2002/compendia/statab/121ed.html. 10

11 Proxies for the agency problems. I construct some variables to assess the severity of agency problems within treated and control groups. I use the pre-reform level of the market-to-book value as a proxy for firm investment opportunities, and I construct the standard deviation of the market to book value as the asset-weighted average of the time series market-to-book values over the twenty quarters prior to the bond issuance. This variable tests the corporate socialism mitigation of subsidiary debt. I expect the reform to be detrimental for conglomerates with a high variability of the investment opportunities if corporate socialism mitigation explains the beneficial effect of subsidiary debt. As my control group is composed of standalone firms, I do not use the cross segments variability of cash flow and market-to-book value as in Kolasinki (2009). I assume that a high cross-sectional variability of the market-to-book value across segments also reflects a high variability of the market-to-book value over time. To test the free cash flow mitigation [Jensen (1986)] of the subsidiary debt, I construct the variable net cash flow. This is computed as the difference between the firm s cash flow and its capital expenditures before the reform. I also construct the volatility of the cash flow as the asset-weighted average of the time series standard deviations of quarterly cash flow (EBITDA/assets) over the twenty quarters prior to the bond issuance to serve as a proxy for firm operating risk. I expect the reform to be detrimental for conglomerates with a high level of the cash flow if free cash flow mitigation explains the beneficial effect of subsidiary debt. 3.2 Empirical Strategy I take advantage of the segment reporting reform to investigate the relationship between corporate diversification and the cost of borrowing in the presence of subsidiary debt. The reform reveals the internal capital market of firms that restate from standalone to conglomerate, but it does not affect the information provided by subsidiaries that issue debt. My treatment effect is the firm that restates from declaring one segment unit before the reform to declare multiple segment units after the reform and issues corporate bonds. The dependent variable is the yield spread of the bonds issued on the primary US bond market. I first investigate the net effect of corporate diversification on the cost of borrowing of restating conglomerates. My empirical methodology is a difference-in-difference estimation with covariates. I label all corporate bonds 11

12 issued by firms that restate from standalone firms to conglomerates as the Treated observations. This is an indicator variable equal to one when a standalone firm restates as a conglomerate in 1998 and issues debt on the primary bond market. I choose the standalone firms as my Control group because they are not affected by the reform. For a change in segment disclosure at time t, the control firms are those that did not change their classification from standalone to conglomerates. I identify the years after 1998 as my after period. The idea behind this approach is that two firms (one a fake standalone firm and the other a real standalone firm) share a parallel trend in the cost of debt before the reform. Figure 4 tests this assumption graphically. The figure reports the trends of the yield spread in the years before the reform. I plot the average of the annual yield spreads of the bond issues for the treated and control firms at t = 0 (1998), in years t = 3 to t = +1. Figure 4.A reports the annual yield spread for each firm-year across the overall sample. In Panel B, I control for firms heterogeneity, and I subtract, for each firm, the average of the yield spread across the overall sample of bond issues. The figure confirms that, in the years before the reform ( ), treated and control firms share parallel trends of yield spreads, but only after controlling for firm heterogeneity. In year t+1 (1998), the figure shows a noticeable increase in the yield spreads of bonds issued by the treated group with respect to the control group. I estimate a difference-in-difference regression with covariates of the following form: y Treated After υ μ τ X ϵ (1) where the dependent variable is the yield spread of bond b of firm i on the primary bond market issued at time t. Equation (1) tests H1 that is, it estimates the causal effect of the reform on the cost of borrowing of conglomerates compared to standalone firms. In my estimations, I always exclude subsidiary bonds from the sample, as they are not directly comparable with the bonds issued by their parents. I investigate the pricing of subsidiary bonds in the robustness section. The coefficient (Treated After) in Equation (1) gives the causal treatment effect of segment reporting change on the yield spreads of the bond issues. I expect a positive sign of the coefficient if the benefits of the debt pricing of firms that restate as conglomerates are smaller than their costs. The 12

13 vector X bit controls for standard bond and (previous year) firm characteristics. For bond characteristics, I include the credit rating, the seniority of the bond, the amount at the issuance (logarithm), bond maturity, and whether bonds have a callable option. I identify firm age, size, and leverage as good controls. While firm size and leverage restore the randomness of the treatment, other variables like the cash flow, the ROA, and the market-to-book value are potential bad controls, because affected by the reform. The vector i is a vector of firm fixed effects, which allows that firms in the treatment and control groups have different average yield spread over the sample. I include industry (τ and year (υ fixed effects to control for time and industry-varying factors. I cluster the standard errors at the firm level. The hypothesis that corporate diversification affects the cost of borrowing implies that this effect is exacerbated in the presence of agency problems. To test this channel, I perform crosssectional tests with the variables that proxy for the agency costs of the restating firms. I split my sample according to the pre-reform level of: i) the net cash flow ratio (cash flow minus capital expenditure scaled by total assets), ii) the investment opportunities, measured by the market-to-book value, iii) the standard deviation of the cash flow over the past five years, and iv) the standard deviation of the market-to-book value over the past five years. Any alternate explanation based on omitted factors explaining the baseline results should also explain why they are concentrated on specific subsamples. Investigating the causal relationship between corporate diversification and the cost of borrowing is a necessary step to explore how the use of subsidiary debt affects this relationship. Because the reform does not provide new information about subsidiaries, I expect that the effect of the reform is homogenous across firms with and without outstanding subsidiary debt. To test this hypothesis, I split my sample of treated bonds into treated firms with outstanding subsidiary debt (s i = 1) and treated firms without outstanding subsidiary debt (s i = 0) along the overall sample. I follow the model: y Treated After υ μ τ X ϵ if s i = {0,1} (2) 13

14 where the breakpoint s i is determined by splitting the sample into treated firms with outstanding subsidiary debt (s i = 1) and treated firms that never had outstanding subsidiary debt in the sample period ( ). Equation (2) tests H2: the variable Treated After gives the causal treatment effect of the reform on the cost of debt of parent firms that restate as conglomerates and do, or do not, have outstanding subsidiary debt. If the use of subsidiary debt has no relationship with the cost of borrowing of restating conglomerates, the coefficient is homogenous across the two estimations. I expect a positive sign of the coefficient β for conglomerates without outstanding subsidiary debt if the latter mitigates the agency problems of conglomerates (H2). Finally, I expect a positive sign of the coefficient β for conglomerates with outstanding subsidiary debt if the use of subsidiary debt increases the credit risk of restating firms. Issuance decisions depend on both the informational environment and the equilibrium pricing of debt. Both of these things changed after SFAS 131, so there is selection bias in terms of which firms issue bonds before and after the rule change, as well as the terms of issuance (amount, secured vs. unsecured, holding company vs. subsidiary). These firms, however, are big conglomerates issuing corporate bonds in significant amounts. Therefore, it is unluckily that these firms stop (or start) to issue public debt only according to the change in the SFAS14. In order to deal with these issues, I look into the amount and the characteristics of bond issuances in my robustness tests. 3.3 Univariate Analysis Table 1 presents descriptive statistics on bonds issued by restating conglomerates with and without outstanding subsidiary debt (panels A and B), as well as standalone firms (panel C). Conglomerates tend to have, on average, lower yield spreads and a higher credit rating than standalone firms. This is consistent with the fact that investors consider safer bonds issued by conglomerates because corporate diversification reduces firm idiosyncratic risk [Hann et al. (2013)]. Not surprisingly, conglomerates are also larger than standalone firms [Villalonga (2006)]. Restating conglomerates without outstanding subsidiary debt do not have significant differences in ROA and market-to-book value compared to standalone firms. Restating conglomerates with outstanding subsidiary debt have instead a higher ROA, market-to-book value and level of the cash flow compared to conglomerates without 14

15 outstanding subsidiary debt and standalone firms. This is not consistent with the findings of Duchin (2010), who shows that conglomerates hold significantly less cash than do standalone firms, thanks to the presence of the internal capital market that reduces the need for precautionary savings. Table 2 presents the characteristics of my sample in the pre-reform period ( ). Table 2 shows a lower yield spread for conglomerates with subsidiary debt outstanding before the reform compared to standalone firms. This is not justified by a noticeable difference in the investment opportunities (market to book value) and it is consistent with coinsurance inside conglomerates reducing idiosyncratic risk (Hann et al., 2014). The net cash flow (cash flow minus capital expenditures) tends to be lower for restating conglomerates without outstanding subsidiary debt compared to restating conglomerates with outstanding subsidiary debt. As the values of ROA and of the cash flow did change after the reform, the table confirms that the rule-change did affect the estimation of the profitability of the restating firms, as shown in Berger and Hann (2003). 4 Empirical Results 4.1 Corporate Diversification and Cost of Borrowing In this section, I test H1 regarding the relationship between corporate diversification and firms cost of borrowing. I use Equation (1) to estimate the causal effect of the reform on the bond yield spreads of restating firms (treated group) compared to standalone firms (control group). Table 3 reports the results. In column (1), I estimate Equation (1) excluding any control. In column (2), I add bond characteristics, in column (3) I add additional firm characteristics to my estimation. The coefficient Treated After in column (3) is economically and statistically significant and implies that treated firms have a higher cost of borrowing by 25 bps after the reform compared to standalone firms. This effect is economically relevant, as it represents 16% of the standard deviation of the yield spreads in my sample (154 bps). This result is also in line with Berger and Hann (2003), who find that, after the reform, conglomerates have a lower firm value and a lower ROA compared to the pre-reform period. This result confirms H1: firms that restate as conglomerates have a higher cost of borrowing after the reform. This result rejects the irrelevance of corporate diversification for firms 15

16 cost of borrowing. Restating conglomerates suffer a relevant increase in their cost of borrowing when compared to standalone firms. To test the economic channel making corporate diversification detrimental for the cost of borrowing, I split the sample of restating conglomerates according to high/low pre-reform level of the variables that proxy for the agency problems: i) net cash flow (cash flow minus the capital expenditures scaled by total assets), ii) market-to-book value, iii) cash flow volatility, and iv) marketto-book volatility. The results are in Table 4. The table shows that the coefficient Treated After is statistically and economically significant for conglomerates with a low level (57 bps) but a high variability (31 bps) of the market-to-book value in the pre-reform period. These results support the corporate socialism channel of the detrimental effect of the reform on the cost of borrowing of conglomerates. A low level of the market-to-book value that is not stable over time may signal that managers have an incentive to decrease their effort and taking safe investment decisions, because they are able to get a partial surplus of the segments that perform the best. From the bondholders perspective, the benefits arising from coinsurance across segments are lower than the costs of too much capital allocated to low-growth segment units. The coefficient Treated After is also statistically significant when conglomerates have a low amount of cash but a high risk before the reform. This result excludes that the free cash flow channel is the main channel that explains the detrimental effect of the reform on the cost of borrowing of conglomerates. 4.2 Subsidiary Debt and Cost of Borrowing of Conglomerates The hypothesis H2 regarding the relevance of subsidiary debt for the borrowing costs of the restating conglomerates implies that the effect of the reform is not homogenous in the cross-section of conglomerates with and without outstanding subsidiary debt. To test this hypothesis, I estimate Equation (2) and I compare the cost of borrowing of restating firms with and without outstanding subsidiary debt with the cost of borrowing of standalone firms. Results are in Table 5. Columns (1) - (3) of Table 5 report the estimation of Equation (2) when restating conglomerates with outstanding subsidiary debt compose the treated sample. Columns (4) through (6) of Table 5 report the estimation 16

17 of Equation (2) when restating conglomerates that do not issue subsidiary debt compose the treated sample. The coefficient Treated After is neither economically nor statistically significant when comparing (restating) conglomerates with outstanding subsidiary debt and standalone firms. Treated firms that issue subsidiary bonds have a similar cost of borrowing compared to standalone firms after the reform. In contrast, conglomerates without outstanding subsidiary debt have a 27 bps higher cost of borrowing when compared to standalone firms (column (6)). In economic terms, conglomerates without outstanding subsidiary debt experience an increase in the cost of the debt by 17% of the standard deviation of the yield spreads in the sample. To test the economic channel making the use of subsidiary debt beneficial for the cost of borrowing (corollaries of H2) of restating conglomerates without outstanding subsidiary debt, I split the sample according to the variables that proxy for the agency problems. The estimations are in Table 6. Columns (1) through (8) report the coefficients for the sample splits of the restating firms without outstanding subsidiary debt compared to standalone firms. Again, the coefficient Treated After is economically and statistically significant for conglomerates with a high variability (27 bps) but a low level (47 bps) of the market-to-book value in the pre-reform period. These results support the corporate socialism channel of the detrimental effect of the reform on the cost of borrowing of conglomerates. Bondholders require a higher yield spread with respect to the pre-reform period to firms without outstanding subsidiary debt that disclose a low level but a high variability in the growth options compared to similar standalone firms. 4.3 Borrowing at The Subsidiary Firm Level: A Financial Contracting Approach. Inderst and Müller (2003) argue that, if the productivity of a new project is sufficiently high, the centralized borrowing does not provide enough incentives to the CEO to reveal the true cash flow to the external investors. This implies an ex-ante increase of the financing constraints of conglomerates unless they do not allocate some debt on their subsidiaries. Consistent with this theory, Billet and Mauer (2003) find that transfers to financially constrained segments increase conglomerates value, irrespective of the efficiency of these transfers. The model does not make empirical predictions for the 17

18 borrowing costs of conglomerates, but it shows that firms that centralize all of their debt at the parent firm level also have a high level of the cash flow. In my setup, testing whether subsidiary debt also reduces the financing constraints of restating conglomerates has some empirical concerns. Intuitively, the use of subsidiary debt may affect financing constraints of conglomerates even before the reform. For example, subsidiary debt allows to enter new markets when there is high competition for external funds [Cestone and Fumagalli (2006)]. As the reform itself affects firms financing constraints, distinguishing between these effects is challenging. To solve this problem, I use the ratio of institutional investors [as computed in Dlugosz et al. (2006)] in the stock ownership of the restating firms as a proxy for the ability of the firm to mitigate financing constraints. The idea is that, under the hypothesis that subsidiary debt also decreases the financing constraints of the restating firms, conglomerates without subsidiary debt and with dispersed shareholders have high costs of borrowing after the reform. The presence of concentrated shareholders is a good instrument as it is stable over time, and less affected by the reform. 6 I then estimate Equation (1) by splitting my sample into firms with high/low institutional investors shares. The results are in Table 7. The coefficient Treated After is positive and statistically significant (70 bps) for conglomerates without outstanding subsidiary debt and with dispersed shareholders, compared to standalone firms. This result supports Inderst and Müller s (2003) hypothesis of an additional benefit of subsidiary debt on firms financing constraints. The channel is that the subsidiary debt provides enough incentives to the CEOs to reveal the true value of the realized cash flows. Anticipating these incentives, bondholders are more willing to lend to firms with outstanding subsidiary debt. 7 6 In an early version of their paper, Berger and Hann (2003) argue that the reform also improves the market for corporate control, thus affecting the ownership structure. 7 A sample split across the pre-reform level of the financing constraints [Hadlock and Pierce (2010); Whited and Wu (2006)] provides similar results (not reported in the table). 18

19 5 Robustness In this section, I conduct various robustness tests. I first conduct a placebo test by assuming a policy change two years before the reform, that is, in Table 8 reports the results. It shows no effect on the yield spreads of the treated firms if we assume a (fake) reform in Changing the dependent variable in yields, ratings, or logarithm of yield spreads does not change my baseline findings. I now investigate alternative channels explaining my results. 5.1 Other Candidate Model Subsidiary debt and asset substitution. Previous results show that firms with outstanding subsidiary debt do not have a different cost of borrowing with respect to standalone firms. The use of subsidiary debt can also be detrimental to the cost of borrowing of conglomerates. Jensen and Meckling (1976) argue that firms close to distress are tempted to increase the risk of their assets in a gamble for resurrection. Kahn and Winton (2005) argue that the subsidiary debt allows parent firms to move some risky asset at the subsidiary firm level to keep safe the other divisions. Consistently, Kolasinki (2009) find that firms with a high cross-sectional variability in the growth options of segments are more likely to issue subsidiary debt. Suppose that parent firms allocate some risky assets to the subsidiary firms to mitigate the effects of the reform on their cost of borrowing: parent firms can reduce the risk on their own debt at the expense of subsidiary s bondholders. Consequently, the average risk of the parent debt is lower after the reform, but subsidiary bondholders require a higher bond spread after the reform. In this case, the beneficial effect of the reform on the cost of borrowing of the parent firms should be compensated by a detrimental effect of the reform on the cost of borrowing of its subsidiaries. In order to test the asset substitution theory, I identify all the subsidiary bonds that belong to a restating conglomerate as my new Treated sample. I exclude guaranteed subsidiary debt as it might generate different incentives for the parent firm [Altieri, Manconi and Massa (2016)]. I estimate Equation (1) and I expect a positive and economically significant coefficient after the reform (higher cost of borrowing of subsidiary debt with respect to standalone debt) if asset substitution (from parent 19

20 to subsidiary) is in place. The variable Subsidiary After is an indicator variable equal to one when the subsidiary bond belongs to a firm that switches from standalone to conglomerate firm after the segment reporting reform. Results are in Table 9. In column (1), I estimate Equation (1) excluding any control. In column (2), I add bond characteristics, and in column (3) I add firm characteristics to my estimation. In all estimations, I include parent fixed effects for the subsidiary firms, and issuer fixed effects for standalone firms. The table shows that subsidiary firms tend to have a lower yield spread when compared to similar standalone firms. The coefficient Subsidiary After in column (3) is negative and economically (but not statistically) significant when comparing the yield spreads of subsidiaries and standalone bonds. These findings exclude that subsidiary bondholders require a higher risk premium after the reform. Adverse selection costs and subsidiary debt. Adverse selection costs are another candidate to explain a difference in the cost of borrowing across firms with and without subsidiary debt. Ex-ante information asymmetry usually increases the risk premium required by investors [Bessembinder, Maxwell, and Venkataraman (2006)], because they do not have the same information as the management. Franco et al. (2014) show that an accurate segment reporting decreases the yield spreads of conglomerates. Because the reform does not affect the subsidiary firms, the decrease in information asymmetry about each business unit of the firm should affect all the conglomerates that restate the number of their business units [Berger and Hann (2003)]. As my results go in the opposite direction, this effect may represent a factor that mitigates the magnitude of my results. 5.2 Confounding factors In this subsection, I investigate confounding factors that may explain my results. One potential confounding effect relates to the business cycle. For example, an increased demand for bonds of standalone firms may decrease their price, explaining a contemporary drop in their yield spreads. The business cycle in 1998 in the US is characterized by federal government surpluses, a low unemployment rate, near zero inflation, and robust growth in the national income. 8 On the one hand, 8 Report available at 20

21 the U.S. government dramatically decreased the supply of risk-free debt because of the elimination of the budget deficit. On the other hand, because of the Asian financial crisis of 1998, investors reallocated their portfolios into safe U.S. government bonds. An abnormal demand (supply) for bonds of the control (treated) group may drive bond yields down (up) and explain my results. I first report the amount of bond issues of standalone and treated firms before the reform (from 1996 to 1998) in Figure 5. The percentage of bond issues is computed as the ratio, for each year, of the total amount ($mil) of new bond issues and firms total assets. The figure shows that standalone and conglomerates decreased their public debt issuances in the two years before the reform. This is consistent with the presence of a demand effect that might drive the results. To empirically test whether the bond demand is affected by the reform, I conduct a falsification test, where my dependent variable is the ratio between the amount of yearly bond issues and firm total assets. I also test whether the reform affected any of the bond characteristics, namely covenants, time-to-maturity, and seniority. I estimate Equation (1) and I report the results in Table 10. The table shows that the reform did not affect any of these characteristics but the covenant protection. This is consistent with the investors requiring more covenants on the new debt issues after the reform to mitigate the agency problem inside the conglomerate. Another potential concern is the presence of contemporary industry shocks. Harford (2005) shows that an economic shock may lead to merger waves across different, sometimes unrelated, industries. If the subsidiary debt is concentrated in specific industries, a contemporary technological or regulatory shock can explain my results. To empirically test these arguments, I identify industry shocks occurring in 1998, retrieved from Harford (2005). The first of these is the commodity shock driven by the unexpected Russian default and Asian financial crisis. However, as I exclude utilities from my sample, I expect that US manufacturing and services firms are marginally affected by these events. The second shock is the introduction of the Internet in 1998, which shaped the computer industry. To control for these effects, I estimate Equation (1) after excluding bonds issued by firms belonging to this industry. The results are in Table 11. The coefficient Treated After holds statistically and economically significant for restating conglomerates without outstanding subsidiary debt after controlling for industry shocks. 21

22 Finally, there was a wave of corporate defaults in 2001 and 2002, with many large bankruptcies. This increased risk premia in the bond market, especially for speculative-grade firms. No firm in my treated sample, however, defaults on its debt. As mentioned before, I observe a big spike in treated issuance in for conglomerates with subsidiary debt in the construction industry because of federal aid provisions. My results are not affected when I exclude these bonds from my sample. One may argue that, since firms without subsidiary debt are rated lower and have high yield spreads before the reform, the results can be driven by lower rated firms experiencing the larger increases in yields during the periods. While this effect would bias a direct comparison between firms with and without subsidiary debt, it is unlikely that drives the comparison between the treated and the control groups, because the latter also has high yield spreads and low ratings before the reform. 6 Conclusions This study empirically examines how the allocation of debt between the parent and its subsidiaries affects the cost of borrowing of conglomerates (conglomerates). I take advantage of a segment reporting reform introduced in 1997 in the US to test the sensitivity of the borrowing costs of conglomerates to their use of subsidiary debt. The reform is a good laboratory, as it forces some firms to restate from standalone to conglomerate and to reveal complete information about their divisions that do not issue public debt. I find that conglomerates that fund their divisions only with parent debt experience a 16% increase in the cost of borrowing after the reform when compared to standalone firms. The effect is economically substantial, as it represents an increase of $4 million in interests expenses for each new bond issue. In contrast, firms with outstanding subsidiary debt do not suffer an increase in their borrowing costs after the reform. I investigate the economic channel making subsidiary debt beneficial for the cost of borrowing of conglomerates, and I find that the effect is concentrated on conglomerates without outstanding subsidiary debt but with a low level and a high variability of the growth options in the pre-reform period. This is consistent with the hypothesis that investors anticipate transfers from good to poorly 22