Download Private Companies and Bond Pricing: A Comparative Analysis and more Study Guides, Projects, Research Finance in PDF only on Docsity!
This paper presents preliminary findings and is being distributed to economists and other interested readers solely to stimulate discussion and elicit comments. The views expressed in this paper are those of the authors and are not necessarily reflective of views at the Federal Reserve Bank of New York or the Federal Reserve System. Any errors or omissions are the responsibility of the authors.
Federal Reserve Bank of New York
Staff Reports
The Private Premium in Public Bonds
Anna Kovner
Chenyang Wei
Staff Report No. 553March 2012 Revised March 2014
The Private Premium in Public Bonds Anna Kovner and Chenyang Wei Federal Reserve Bank of New York Staff Reports , no. 553 March 2012; revised March 2014 JEL classification: G12, G
Abstract This paper is the first to document the presence of a private premium in public bonds. We find that spreads are 30 basis points higher for public bonds of private companies than for bonds of public companies, even after controlling for observable differences, including rating, financial performance, industry, bond characteristics, and issuance timing. Theestimated private premium increases to 40-56 basis points when a propensity matching methodology is used or when we control for fixed issuer effects. In contrast, in the same sample, there is no difference in pricing in private debt (syndicated loans). Despite the premium pricing, bonds of private companies are no more likely to decline in price, to default, or to be downgraded than are public bonds. We conclude that the costs ofinformation may be different across segments of the debt market.
Key words: bond pricing, private equity, debt costs, information
Kovner: Federal Reserve Bank of New York (e-mail: anna.kovner@ny.frb.org). Wei: AIG (e-mail:chenyang.wei@aig.com). The authors thank Eben Lazarus, Kevin Pan, and Phoebe White for outstanding research assistance. The views expressed in this paper are those of the authors and do notnecessarily reflect the position of the American International Group, the Federal Reserve Bank of New York, or the Federal Reserve System.
Unlike other studies of private firms, since companies with public bonds are required to file public financial statements in the U.S., 1 we are able to control for a full set of observable borrower characteristics. Our empirical tests include controls for an array of proxies for credit risk, including rating, industry, leverage and profitability. We also control for issuance quarter and differences in bond characteristics, such as maturity and putability or callability. While borrower and bond characteristics are associated with pricing, the average difference in bond spreads persists, suggesting that the difference is not due to observable characteristics. Although we cannot eliminate the possibility that there are differences in unobservable risk, estimates of the difference in bond spreads are actually higher when we use propensity matching techniques (45-56 bps). And of course, since we do not find evidence of a similar pricing difference in the private debt issued by these same companies, the differences in unobservable risk would have to be pertinent only to the pricing of bonds. Another way to measure risk is to see if companies with private equity are more likely to decline in price or default. We find no evidence that ex post outcomes for bonds of private companies are worse than those of public companies. Private issuers are no more likely to file for bankruptcy or to be downgraded than are their public peers. Among firms with traded credit default swap (CDS) contracts, we do not observe any significant difference between the CDS pricing of public and private firms. We also do not find evidence that private bonds perform worse post issuance, although the wedge between the pricing of public and private bonds persists in the secondary market. We also find no evidence that the private premium is related to aftermarket liquidity, nor do we find that bonds of private issuers are less liquid. Private
(^1) Although the financials of companies with private equity are not aggregated by Compustat, filing requirements for companies with public bonds are similar to those of companies with private equity with the exception of the proxystatement (14A), a form filed in advance of equity shareholder meetings and when soliciting shareholder votes; this form is filed only by companies with public equity.
companies pay a higher spread than do public companies with similarly highly concentrated equity ownership. It is not clear that private companies with public debt should be riskier than public ones. Pecking order theory suggests that in the presence of information asymmetry, higher quality firms should use less information-sensitive securities such as risky bonds (e.g., Myers and Majluf (1984), Myers (1984)). However, Fulghieri and Lukin (2001) show that this pecking order can reverse if investors produce additional information on the issuing firm and if the cost of becoming informed is low. Bolton and Freixas (2000) model the choice between equity, bank debt and bonds and conclude that riskier firms issue equity and bonds, while the safest companies issue only bonds.^2 Our paper extends the work of Saunders and Steffen (2011) who explore the costs of being private in the UK debt market. While they find that private firms pay higher prices for syndicated loans, they do not find a significant price differential in the cost of private debt for private companies with public bonds. We add to this the finding that when bond and loan borrowings co-exist, firms pay an extra premium in the public bond market relative to similar public companies. Therefore the availability of debt from alternative sources is not enough to prevent the emergence of price differences for public and private firms. Our findings also suggest another dimension to the trade-off between borrowing publicly and borrowing privately, as the price might be different depending on the issuer’s equity ownership structure.
(^2) There is also a long literature about the choice between private and public debt built on Diamond (1984), who shows the value of banks as intermediaries that save on monitoring costs relative to direct financing from investors.Many subsequent models make predictions about firm characteristics such as age, assets and growth opportunities and the choice between bank debt and bonds (notably Hoshi, Kashyap and Scharfstein (1993), Chemmanur andFulghieri (1994), Boot and Thakor (1997), Holmstrom and Tirole (1997) and Repullo and Suarez (2000)).
are very small (equity value less than $1 billion) pay just as high spreads as do private companies, relative to the largest public companies.^4 If the amount of information produced is proportional to the amount of public equity, our results are consistent with the hypothesis that information produced by the equity market about public companies is valuable to bond investors. We examine next how much of the premium in bond prices that remains can be attributed to differences in costs of information. Proxies for opacity of the issuer’s assets such as first bond offering, earnings variability, underwriter quality, split rating and existence of CDS contracts reduce the private premium by approximately 7 bps (more than 25%). Results are inconclusive as to whether the penalty for opacity is different for private companies. Most similar to this study is the paper by Saunders and Steffen (2011), which examines the relative costs of private debt for private and public firms in the UK. Similar to our results, they document a 29-to-42-basis-point difference in loan spreads for private debt of private firms without public bonds. In Italy, Pagano, Panetta and Zingales (1998) document that the cost of bank credit falls after an initial public offering.^5 Santos and Winton (2008) find that companies with public debt pay lower bank loan spreads, but they argue that this effect is driven by differences in bargaining power. This paper complements this empirical literature by providing the first direct evidence in the U.S. market of price differences for public bonds with private equity. However, unlike previous papers that look at private lending, this paper focuses on the public bond market where monitoring differences should be less important. In the bond
(^4) Black and Scholes’ (1973) and Merton’s (1974) option pricing models imply a direct relationship between equity values and the risk of credit default. If investors rely on models that need stock prices, this would preclude themfrom investing in private companies (but not small public companies). (^5) In the empirical literature on why firms go private (or public), Boehmer and Ljungqvist (2004), Helwege and Packer (2009) and Chemmanur, He and Nandi (2010) also consider the costs of borrowing.
literature, it is the only paper that we are aware of that explicitly account for the impact of issuer financial characteristics in pricing bonds without public equity. The rest of the paper is organized as follows. Section 2 describes the data and variables used in the analysis. We begin in Section 3 by documenting that differences in bond pricing persist after controlling for observable differences in the earnings, leverage, ownership and likely payoffs of private and public issuers. We review similar analysis for loans, but find that observable differences explain pricing differences. We then use a propensity matching methodology to confirm the results and finally show that ex post outcomes are similar for public and private companies in terms of bankruptcy, downgrades and pricing and liquidity in the secondary market. In Section 4 we explore the role of public equity as a signal and as a security subordinate to debt. We then look directly at measures of information opacity that may affect pricing. Section 5 concludes.
2. Data 2.1 Sample composition Since we are interested in a set of private companies for which we can have full financial information, we focus on U.S. companies that raise publicly traded debt in the domestic corporate bond market. Using Mergent’s Fixed Income Securities Database (FISD), we begin with all U.S. corporate bonds issued by industrial (non-financial, non-utility) firms between 1993 and 2009. It has been used in other studies (e.g., Billet, King and Mauer (2007)) as a comprehensive data source for the U.S. corporate bond market. For each bond issue, FISD provides the offering yield, offering date, amount, coupon, security level, callability, putability
only the company’s current ownership status as a data field. When Capital IQ does not provide enough information, we search SEC filings, media coverage, company websites and other online company descriptions (e.g., Google Finance, Wikipedia, etc.) for further verification. Through this procedure, we are able to unambiguously classify 1,276 bond issues as offered by a subsidiary of a publicly listed parent company (e.g., Bell Atlantic).^7 In doing so, we also find 34 cases where issuers are subsidiaries of foreign public companies. Overall, we are able to confirm the public/private issuer status for 9,034 bonds, with 7,287 issues offered by 1,384 public firms and 1,747 issues by 1,023 private companies. We further research the private companies and identify 28% of the private issues as being associated with leveraged buyouts. Panels A and B of Table 1 summarizes the number of bond issues with information on equity ownership. We supplement the Mergent data with quarterly company financial information on firm size, leverage and profitability from Compustat and Capital IQ. While both Compustat and Capital IQ collect data from SEC filings, Compustat collects data only for firms with public equity above a certain size. Thus, Capital IQ has better coverage of the private firms in our sample. For each bond, we collect three accounting numbers as of the end of the quarter prior to bond issuance: total assets, total debt and earnings before interest, taxes, depreciation and amortization (EBITDA). We define firm size as the log of total assets and profitability as the ratio of the latest 12 months (LTM) EBITDA to total assets. We define leverage as the ratio of total book debt divided by total book assets.^8 In addition to issuing public bonds, the vast majority (96%) of bond issuers in our sample for which we also have financial information from Compustat or Capital IQ have also borrowed
(^78) Our results are unaffected by dropping all subsidiaries of public companies. structure of the company has changed with the bond issuance.We also measure leverage in the quarter ended immediately following the bond issuance, in case the capital Using this measure of leverage does not change the results significantly.
money in the syndicated loan market at some time. Since we hand research company ownership at the date of bond issuance, and the risk of a company may not be fixed over time, we search for syndicated loan packages underwritten within 15 months of the bond issuance. We match approximately 70% of our bonds with financial data to loan packages arranged within 15 months of the bond issuance date. The result is a sample of 4,986 bonds (456 private) and 4,697 loan facilities (913 private) for 1,857 issuers with financial information. Without restricting the sample to companies that are also issuing private debt, we have a broader sample of 7,155 bonds (619 private) with financial information. Bond pricing results are similar when estimated in the sample matched to loans or in the full sample.
2.2 Financial Characteristics Table 1 presents the distribution of issuances through time and across industries for public bonds and syndicated loans for the broadest possible sample of companies. For this table, we use our hand research to identify private companies issuing bonds and use the Dealscan public/private indicator to identify private companies issuing syndicated loans. Between 1993 and 2009, U.S. public companies raised more than $2 ($26) trillion in fixed, rated, non- convertible public debt (syndicated loans), as compared with less than $400 billion ($17 trillion) borrowed by private companies. The average bond (loan facility) issuance size of public companies is $287 ($278) million as compared to $224 ($156) million for the private firms. Over the 17-year period, issuance numbers and volumes of the two borrower types followed generally similar patterns, with a 57% (64%) correlation in number of bonds (loan facilities) issued and a 40% (91%) correlation in issuance volume.
As well as observable differences in financial ratios, public and private companies are also different in informational opacity. The “Opacity Measures” sections of Table 2 present several measures of opacity calculated at issuance. Some offers are specific to the debt type and we summarize those separately: First offer(bond) is a dummy variable equal to 1 if this issuance is the company’s first public bond offering (measured since 1988) and First offer(loan) is a dummy variable equal to 1 if this loan facility is the company’s first syndicated loan offering (measured since 1996). SD ROA is the standard deviation of the 4 quarters of return on assets (ROA) following bond issuance. 144A is a dummy variable equal to 1 if bonds were first issued only to qualified institutional buyers under Rule 144A. 9 Top underwriter is a dummy variable equal to 1 if the company’s bond underwriter had a market share in the previous year of greater than 1% (equivalent to a top 15 ranking).^10 Finally, S plit rating is a dummy variable equal to 1 if the bond rating from S&P is different from Moody’s.^11 Rating agencies provide arguably the most important independent assessments of the credit quality of a bond issue/issuer. Therefore a disagreement among them is likely associated with heightened uncertainty with respect to the issue/issuer’s default risk. As shown in Table 2, bond issuances by private firms are generally more opaque, are more likely to have split ratings and to be issued under Rule 144A, are less likely to have a top underwriter and have more volatile accounting performance.
(^9) Livingston and Zhou (2002) find evidence of lower liquidity, information uncertainty and weaker protection of investors for securities issued under 144A. Of the companies in this sample, the 88% that were issued under Rule144A also had registration rights that require a public registration within six months or an increase in the interest rate. (^10) See Livingston and Miller (2000) for evidence that investment banker reputation acts to certify the value of a debt issue to investors and an estimation of the impact of underwriter prestige on offering yields. (^11) Livingston and Zhou (2010) find that split-rated bonds average a 7-basis-point yield premium over non-split-rated bonds of similar credit risk and conclude that investors demand higher yields to compensate for the informationopacity of such bonds.
Finally, to understand if borrower performance is different post-issuance, we collect secondary-market bond and CDS pricing data for our sample firms. Bond prices and yields are gathered from two data sources. Transaction-based data (volume and yield) between July 2002 and December 2010 come from the Transaction Reporting and Compliance Engine (TRACE).^12 Since trading of corporate bonds is fairly infrequent, we also use Reuters’ DataScope to collect end-of-day price and yield quotes.
CDS pricing data come from Markit CDS Pricing. 13 In the period between 2001 and 2007, CDS pricing data are available for 412 firms in our sample from Markit CDS Pricing. Thirty percent of these firms are private as of the pricing and issuance date. We use year-end spread data for five-year, senior unsecured credit default swaps, the most common CDS contracts traded in that period. We focus on spreads classified under the “modified restructuring” document clause, a contract term that enumerates the contingencies under which settlement of a CDS contract would be triggered.
3. Establishing the Private Premium 3.1 OLS Specifications In order to understand if there are differences in debt pricing for public and private companies, we estimate a pooled OLS regression, of the following form:
[1] SPREADi , j , t ^ ( PRIVATEi , t )( ISSUEj )( COMPANYi. t )( QUARTERt )^ i , t. j
(^12) TRACE was introduced in July 2002 with the aim of enhancing the transparency of the corporate bond market. For a detailed description of the TRACE initiation and a general background on corporate bond trading in the U.S.,see Bessembinder and Maxwell (2008). Goldstein and Hotchkiss (2007) discuss a few exemptions in TRACE eligibility. (^13) While CDS contracts may be traded on the other bonds in the sample, to our knowledge Markit maintains the most comprehensive available data source for CDS data.
include additional controls for the financial condition of the borrower. We control for: (i) size (log assets), (ii) profitability (EBITDA to assets), and (iii) leverage (total debt to assets). We add industry controls, dummy variables equal to one for each of the manufacturing, media, retail, railroad, service and telecommunications sectors. In addition to the financial variables shown, we tried other financial ratios, such as interest coverage (EBITDA to interest) and other definitions of profitability (EBITDA less capital expenditures), but do not include the results in the final specifications, since the estimated coefficients were not statistically significant. The results are summarized in Table 3, Panel A for bonds and Panel B for loans. We begin by controlling only for bond characteristics (excluding rating). As is suggested by the univariate results, bonds (syndicated loans) of companies with private equity are issued at spreads that are 187 (30) bps higher than bonds of public companies. [TABLE 3] Of course, much of this is driven by differences in risk. After controlling for company financials and ratings, the difference shrinks to 71 bps for bonds (Column (5) of Table 3). This indicates that there are meaningful differences between private and public companies that finance themselves in the bond markets, differences that account for a 106 basis point pricing difference. These differences are not captured fully by ratings, since both the financial metrics and ratings dummies are statistically significant. For loans, the estimated coefficient on private companies falls to only 9 basis points, and the difference is no longer statistically significant after controlling for observable differences between public and private borrowers. While almost a third of the private companies are leveraged buyouts, the private spread premium is not an LBO effect. In fact, Huang, Ritter and Zhang (2013) find that yield spreads on private equity backed issues are actually lower, all else equal. After controlling for a fixed
price effect for bonds or loans issued as part of a leveraged buyout, the estimated coefficient on the private dummy remains statistically significant and of the same magnitude. In summary, after controlling for differences in observable bond and company characteristics, we find that bonds are much more expensive for companies with private equity, but there is no statistically significant difference in loan pricing between public and private companies. On average, spreads are 30 bps higher for private companies (see specification (5) of Table 3, Panel A). This is more than 6% of of mean bond spreads for private issuers (a present value of $4 million in interest for a bond of mean size and maturity).^14 In the final specification of Panels A and B of Table 3, we take advantage of the 443 companies in the sample that changed their ownership and estimate the same model controlling for company fixed effects. Assuming that unobservable risk is constant over time for companies, this specification should provide the best estimate of the private premium. There is still a positive, statistically significant coefficient on the private dummy -- bonds of the same companies are 58 bps more expensive when those companies have privately held equity. 3.2 Propensity Score Matching In the previous analysis (Section 3.1), we control for differences between private and public companies using observable characteristics and fixed company effects. To reduce the potential selection bias in estimating a causal effect on spreads of being private, we apply a propensity score matching methodology (Rosenbaum and Rubin (1983)). This methodology is useful when observable differences in covariates (such as size and rating) are related to the probability of being private. Furthermore, it uses only the matched subsample for estimation
(^14) Present value based on a spread premium of 30 basis points for a bond of mean size ($230M), an additional $69,000 annually for an average maturity of 9 years, discounted at 9.77% (mean spread of 463 bp plus mean yieldon 10 year treasury of 5.14% from 1993-2009).
are actually higher. Among issuers, private companies are smaller, less profitable and more highly levered. [TABLE 4] In order for the propensity matching method to work, we need to have an adequate control group of bonds of companies with public equity and issue/issuer characteristics similar to those of bonds with private equity. Because there are so many more issuances by companies with public equity, there is a sufficient overlap. Figure 1 shows the scale of the overlap in terms of ratings. Since industry is not significantly associated with the probability of being private (except for telecommunications), but is likely to be associated with pricing, we run two sets of matching variables, one that includes industry and one that does not. Many different methodologies for propensity score matching are proposed in the literature. We use two different matching methodologies, different variants of the matching procedure as well as different weightings of the matching characteristics. Propensity score matching is a trade-off between the quality of the match and the number of matches. Therefore, we estimate matches for 2 and 5 nearest neighbors (the 2 and 5 closest matches). We also use local linear matching, which can be a superior methodology when a large number of propensity scores approach the boundary, and use the local linear estimator proposed by Heckman, Ichimura and Todd (1997) with a Gaussian kernel. We also compare the standard errors to standard errors bootstrapped with 50, 100 and 300 replications. The results of these specifications are shown in Table 5. [TABLE 5] Matching bonds of private companies to similar public companies suggests that private companies pay 30 to 45 bps more for public debt than their public peers. These estimates are
again higher than those of the OLS specifications, suggesting that if anything private companies are paying much higher prices to access public debt markets. In contrast, we do not always find a statistically significant difference between pricing of loans of public and private companies, and never estimate a price difference greater than 12 basis points. This result is consistent with empirical studies of going-private transactions, in that these studies do not suggest that private companies are riskier than public companies. Mehran and Peristiani (2009) find that a primary reason for companies to abandon their public listing was a failure to attract significant visibility and interest from investors. They also find that firms with low stock price volatility are twice as likely to be taken private. Opler and Titman (1993) argue that firms with lower costs of financial distress (and thus possibly lower losses given default) are more likely to conduct leveraged buyouts and Kaplan (1989) finds incentive improvements in newly private LBOs.
3.3 Bonds vs. Loans Of course, the estimated bond premium could arise from time-varying differences in the unobservable riskiness of public and private companies. However, we find a large statistically significant difference in pricing only for public bonds, which suggests that to the extent there are differences in unobservable risk of private companies, they exist only for private companies’ bonds and not their loans. Since one difference between bonds and loans may be seniority, we replicate the analysis dropping subordinated bonds. This means we examine a sample of only senior bonds, which are typically pari passu with loans. The estimated coefficient in specification (5) of Table 3 falls to
