Are the Borrowing Costs of Large Financial Firms Unusual?

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. Are the Borrowing Costs of Large Financial Firms Unusual? Javed I. Ahmed, Christopher Anderson, and Rebecca E. Zarutskie 2015-024 Please cite this paper as: Javed I. Ahmed, Christopher Anderson, and Rebecca E. Zarutskie (2015). “Are the Borrowing Costs of Large Financial Firms Unusual?,” Finance and Economics Discussion Series 2015-024. Washington: Board of Governors of the Federal Reserve System, http://dx.doi.org/10.17016/FEDS.2015.024. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Are the Borrowing Costs of Large Financial Firms Unusual? Javed Ahmed∗ Christopher Anderson† Rebecca Zarutskie∗ March 12, 2015 Abstract Estimates of investor expectations of government support of large financial firms are often based on large financial firms’ lower borrowing costs relative to smaller financial firms. Using pricing data on credit default swaps (CDS) and corporate bonds over the period2004to2013, however, wefindthattheCDSandbondspreadsoffinancialfirms are no more sensitive to borrower size than the spreads of non-financial firms. Outside of the financial crisis period, spreads are more sensitive to borrower size in several non-financial industries. We find that size-related differences in spreads are partially driven by higher liquidity and recovery rates of larger borrowers. Prior to the financial crisis, we also find that financial firms exhibited generally lower spreads that were less sensitive to size than spreads for several other industries. Our results suggest that estimates of implicit government guarantees to financial firms may overemphasize sizerelated borrowing cost differentials. However, our analysis also suggests that, prior to the financial crisis, investor expectations of government support, or generally reduced risk perceptions, may have reduced borrowing costs for the financial industry as a whole. JEL classification: G21, G22, G24, G28 Keywords: Financial industry, too-big-to-fail, implicit government guarantee, size effect, borrowing costs, credit default swaps ∗Board of Governors of the Federal Reserve System. 20th & C Streets, NW; Washington, DC 20551. Ahmed can be reached at javed.i.ahmed@frb.gov or (202) 912-4649. Zarutskie can be reached at rebecca.e.zarutskie@frb.gov or (202) 452-5292. †Harvard University. Anderson can be reached at chanderson@hbs.edu or (415) 300-6085. We thank Gary Gorton, Michael Gordy, Richard Herring, Matthew Plosser and seminar participants at the University of Maryland and the Federal Reserve System Conference on Financial Structure and Regulation hosted by the Federal Reserve Bank of Dallas Houston Branch for useful comments and suggestions. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by the Federal Reserve Board of Governors or its staff.

1. Introduction Whether market participants view certain financial firms as likely to receive government support in future times of stress is an open empirical question. Understanding and quantifying such expectations is relevant since such expectations could give financial firms perceived as too-important-to-fail an advantage in doing business over other firms due to their lower perceived risk. In addition, expectations of implicit government guarantees could further weaken financial stability by lowering the incentives of investors and other stakeholders to monitor firms with perceived implicit guarantees or by increasing risk-taking by financial firms (e.g., Flannery and Sorescu (1996), Iyer, Puri, and Ryan (2013), and Gropp, Hakenes, and Schnabel (2011)). A common approach to measuring expectations of implicit government guarantees is to test whether the borrowing costs of firms that investors may view as being more likely to receive government support, typically measured by firms size, are lower than those of other firms. Sincethebondsofsuch“too-big-to-fail”firmsshouldbelesslikelytodefault, investors should demand lower returns on these bonds, all else equal. However, evenintheabsenceofagovernmentguarantee, largefinancialfirmsmayborrow more cheaply than small financial firms. Relating size to otherwise unobserved risk factors is common in asset pricing literature (e.g. Chan and Chen (1988)), and size is often used as a control variable in studies of credit spreads. To evidence a government guarantee, size- 1

related funding cost differentials should seem unusual in banking relative to industries not subject to a guarantee (e.g., Kroszner (2013)). In this paper, we show that borrowing cost differentials between large and small financial firms are not unusually large relative to differentials we observe in other industries. We compare size-related borrowing cost differentials across industries to determine how size affects the cost of debt financing. Our methodology allows for industry-specific size effects in borrowing costs and controls for firm-specific default risk. While we find that large financial firms borrow more cheaply than small financial firms, the financial industry does not seem unusual in terms of this size effect. However, we do find evidence that financial firms as a group, particularly commercial and investment banking firms, have lower average costs of borrowing compared with similar firms in other industries, particularly prior to the financial crisis. This suggests that financial firms, taken as a whole, may be more likely to be viewed as less risky compared to firms in other industries, perhaps due to investor views that the government is more likely to intervene in this industry in times of stress.1 Specifically, we examine the quarterly credit default swap (CDS) and bond spreads for financial and non-financial firms over the period 2004 to 2013 and how these spreads relate to firm size, controlling for a number of factors including firm-specific risk. We find that larger 1This finding is consistent with a recent paper by Kelly, Lustig, and Van Nieuwerburgh (2011) who find using option data that the financial industry, as a whole, is perceived as less risky. 2

firms have lower spreads in most industries. The negative relation between firm size and spread is larger in many non-financial industries than it is in financial industries, especially prior to the financial crisis. During the financial crisis, the negative relation between firm size and spreads became weaker across all non-financial industries and slightly stronger for commercial and investment banking firms. However, following the financial crisis, this pattern has reversed. To understand the reasons behind the size effect in CDS and bond spreads, we explore explanations related to differences in liquidity and recovery rates. Greater liquidity could lower the cost of borrowing for larger firms. In addition, economies of scale in debt issuance could advantage larger or more familiar issuers. Similarly, recovery rate advantages related to issuer size could lead to lower borrowing costs for equivalent levels of default risk. When we control for these factors, we find that liquidity affects borrowing costs, but we still find size-related borrowing cost advantages in most industries outside the financial crisis. We also findpreliminaryevidenceofarelationshipbetweenrecoveryratesandsize, butinter-industry analysis suggests this relationship is driven by a small number of non-financial industries. Ourresultssuggestthatsize-relateddifferencesinthecostofborrowingamongstfinancial firms do not necessarily reflect investors’ expectations of a higher likelihood of government support for the larger firms. If anything, our results suggest that investors in the bonds and CDS of financial firms may view firms in these industries as overall less risky compared 3

to firms in non-financial industries, and that this was especially true prior to the financial crisis. Whether or not any lower overall perceived risk is due to investor expectations of government support or due to other factors is an open empirical question. Our study is unique on several dimensions, allowing us to generate new results and insightsintotheliteratureonmeasuringimplicitgovernmentguarantees. First,byusingCDS data, in addition to bond data, we are more able to make an “apples-to-apples” comparison between firms of different sizes and in different industries. Examining spreads on standard CDS instruments eliminates the need to control for a range of security characteristics, as mustbedonewhenexaminingbondspreadsordepositratesCDSspreadsshouldintheorybe equal to the spread of the underlying corporate bond since the CDS acts as insurance against the event of default of the underlying bond, and in practice CDS and corporate bond spreads are highly correlated (e.g., Houweling and Vorst (2005))2 Thus, examining CDS spreads, in addition to directly examining corporate bond spreads, reduces the likelihood that observed differences in borrowing costs are driven by differences in characteristics of the underlying bonds rather than differences in firm characteristics and allows us to analyze a large sample of firms. Second,thesetoffirmsforwhichtherearebothCDSandpublicbondstradinglietowards the top of the firm-size distribution, allowing a more meaningful comparison of borrowing 2Only a handful of other studies in the implicit government guarantee literature examine CDS spreads Li and Zhang (2011) examine CDS spreads of financial firms only. Demirguc-Kunt and Huizinga (2013) examine CDS spreads and equity prices in an international sample of banks and find that in some countries banks may have become “too-big-to-save” rather than “too-big-to-fail”. 4

costs by firm size since comparing borrowing costs of very small banks to very large banks may shed less light on any “too-big-to-fail” subsidy than comparing the largest banks to large, but still smaller, banks. Third, since we conduct our exercise across all industries, we are able to more directly control for other differences between small and large firms that may cause their costs of borrowing to differ besides implicit government guarantees (e.g., Kroszner (2013)). We examine a number of different firm size variables to account for differences in the underlying firm size distributions across industries. Our study is most similar to studies which explicitly examine the costs of debt between financial firms, such as Acharya, Anginer, and Warburton (2013) and Santos (2014), who examine differences in bond spreads between large and small financial firms, and Jacewitz and Pogach (2012) and Bassett (2014), who examine differences in the cost of deposits by large and small banks. We study differences in CDS and bond spreads between large and small banks, as well as large and small financial firms and non-financial firms more generally, but adopt a more comprehensive approach to econometrically modelling credit spreads as a function of firm industry and sizein a large sample of firms. Several studies have examined differences in the size sensitivity of bond spreads of financial firms relative to non-financial firms. However, these studies either group non-financial firms together in to one broad category over many years (e.g., Acharya et al. (2013)) or are limited to small samples (e.g., 5

Santos (2014)). The remainder of the paper proceeds as follows. Section 2 discusses related literature on why we may observe firm-size-related differential costs of borrowing. Section 3 describes the data. Section 4 presents our main results. Section 5 explores the roles of liquidity and recovery rates in explaining the observed size effects in credit spreads. Section 6 discusses the economic magnitudes of estimates. Section 7 concludes. 2. Related literature In this section, we briefly review the literature which discusses why larger financial firms may have borrowing costs, how these explanations have been examined in other studies, and how they are examined in our setting. 2.1. Size anomaly The strong empirical negative relation of firm size on equity and debt returns has long been recognized (e.g. Banz (1981), Fama and French (1992)). While the negative empirical relation between size and returns is typically cited as evidence of mis-specification of the underlying asset pricing model or mis-measurement of the model’s risk factors (e.g., Berk (1995), Gandhi and Lustig (forthcoming)), it is clear that firm size is correlated in most cases with factors or omitted variables indicating lower risk. We also find that firm size is strongly negatively correlated with the cost of borrowing, as 6

measured by CDS and bond spreads. This is true across most industries. Thus, our analysis is consistent with the findings in the “size anomaly” literature. We attempt to parse out the components of the size effect in our data, in particular, by measuring the liquidity of a firm’s CDS and bonds and the expected recovery rates on the bonds, as we discuss below. Still after controlling for these factors, we find that size is still negatively correlated with CDS and bond spreads in both financial and non-financial firms. 2.2. Too-big-to-fail (TBTF) subsidy One possible explanation of the size effect in borrowing costs and expected returns in financial firms is that larger financial firms may be perceived by investors as more likely to be bailed out by governments in the event of distress. This explanation has received a lot of study recently in light of the financial crisis of 2008 and 2009 and the widely publicized government support received by several large financial institutions. Approaches to estimating any TBTF subsidy vary. Many focus on estimating the difference in borrowing costs between large and small banks controlling for observable characteristics of the banks and borrowing instruments. For example Acharya et al. (2013), Santos (2014), and GAO (2014) focus on relating bond spreads to financial firm size. Tsesmelidakis and Merton (2013) use bond data and a structural model to form estimates of implicit government guarantees. Jacewitz and Pogach (2012), O’Hara and Shaw (1990), and Bassett (2014) focus on differences in bank deposit rates of large and small banks. Araten 7

and Turner (2013) examine the cost of funding for a variety of funding sources. Ueda and Weder di Mauro (2013) focuses on differences in credit ratings of large and small banks. Others such as Brewer III and Jagtiani (2013) estimate how much financial firms pay to acquire other firms to reach a certain size threshold. Gandhi and Lustig (forthcoming) use equity returns to show that investors may perceive the largest banks as being more likely to receive government support in times of stress. Kelly et al. (2011) present evidence using options data that investors place a higher probability on the financial sector receiving government support as a whole compared to individual financial firms. Several recent studies have adopted this empirical approach and have concluded that the value of an implicit guarantee to large financial firms can be sizable (e.g., Acharya et al. (2013), Tsesmelidakis and Merton (2013), Ueda and Weder di Mauro (2013), Araten and Turner (2013), Noss and Sowerbutts (2012), Jacewitz and Pogach (2012), Penas and Unal (2004)). Many of these studies find evidence of a TBTF subsidy during the financial crisis, but also in the periods before and after. Given the size anomaly literature discussed above, however, it is difficult to say with certainty that correlations between size or other measures of the likelihood of government support do not proxy for other factors, which we discuss below.3 3For a recent discussion of some of the challenges to estimating implicit government guarantees, see Kroszner (2013). 8

2.3. Economies of scale Larger firms may be more efficient and have lower costs or be less likely to fail relative to smaller firms. Indeed recent studies, such as Hughes, Mester, and Moon (2001), Wheelock and Wilson (2012), Hughes and Mester (2013), and Kovner, Vickery, and Zhou (2014), find evidence of economies of scale in the banking industry. To the extent that firm size and associated economies of scale in production lead to lower bankruptcy risk, then size should benegativelyrelatedtothecostofborrowing. Ifmeasuresofprobabilityofdefaultaccurately capture the impact of production economies of scale, then size may be less likely to proxy for economies of scale in pricing regressions. A second form of economies of scale may be in the frequency with which a firm issues bonds. If firms issue debt more frequently there may be lower transactions costs and greater transparency surrounding the debt issues, leading to lower spreads and costs of borrowing. We form a variables, which we discuss below, to measure potential economies of scale in debt securities issuance. 2.4. Liquidity Securities issued by larger firms may be more liquid, leading to lower spreads. Dick- Nielsen, Feldhu¨tter, and Lando (2012) find evidence that bonds experienced greater spreads due to illiquidity during the financial crisis, suggesting that to the extent size serves as a proxy for liquidity risk, that lower spreads on larger firms might be partially driven by 9

liquidity. We form two variables meant to capture the degree of liquidity in firms CDS and bond trading markets, which we discuss in Section 3. 2.5. Recovery rates Finally the cost of borrowing may be lower if expected recovery rates in the event of default are higher (e.g., Doshi (2011) and Acharya, Bharath, and Srinivasan (2007), Carey and Gordy (2009)). We relate firm size to expected recovery rates in the event of a bond default to parse out the extent to which the size sensitivity of bond and CDS spreads may be driven by differences in expected recovery rates. As we discuss below, we attemp to control for these factors in addition to firm size. In addition, and most importantly, we compare the size effect across industries over time to assess whether the size effect seems unique or unusual to financial firms. 3. Data We analyze quarterly CDS data (2004Q1-2013Q2) for a broad sample of U.S. firms. We collect data on credit default swap (CDS) spreads from Markit and quarterly balance sheet information from Standard & Poor’s Financial Services LLC Compustat via Wharton ResearchDataServices(Compustat). Werestrictattentiontofive-yearnon-governmentCDS spreads from Markit for senior unsecured debt denominated in U.S. dollars with a modified restructuring documentation clause. This restriction ensures that CDS spreads for different 10

firms are comparable in terms of debt type and maturity. For each quarterly observation, we choose the last available CDS spread. Our CDS sample includes all firms for which these data are available. The CDS sample contains 17,486 observations for 665 unique firms. We rely on estimates of firms’ default risk based on five-year EDF from Moody’s Analytics, CreditEdge. Moody’s calculates the five-year EDF using a structural model based on stock valuations, balance sheet information, and realized asset volatility. The CreditEdge model first calculates a theoretical default risk using a structural model similar to Merton (1974). Moody’s calculates the EDF by comparing the distribution of theoretical default risk to subsequent realized defaults. The five-year EDF calculations do not rely on bond yields or CDS spreads. WecomplementCDSdatawithdataonbondtradingfromFinancialIndustryRegulatory Authority: Trade Reporting and Compliance Engine (TRACE). We drop observations for whichreportedyieldsexceed25%astheylikelyincludedataerrors. Wekeeptradesexceeding $10 million. For each bond on each day, we record the median reported yield. We merge the bond data with Moody’s CreditEdge to obtain the EDF on the same day as the bond trade. WeobtaindataonbondcharacteristicsfromMergent’sFixedIncomeSecuritiesDatabase (FISD) and Securities Data Corporation (SDC) Platinum. We exclude bonds with floating interest rates, callable bonds, convertible bonds, and issues flagged as preferred stock. For each bond, we subtract the yield on zero-coupon U.S. treasury bonds with the same maturity 11

to obtain the yield spread.4 We retain the last quarterly observation for each matched bond. For issuers with multiple bonds, we retain the bond with maturity closest to 5 years. The bond trading sample contains 16,006 observations for 319 unique firms. The difference in CDS and bond sample sizes is due to the requirement that bonds not have floating interest rate, becallabeorconvertible, whereasCDSwrittenonsuchbondsareincluded. Inaddition, some bond sample observations are dropped because we cannot match them to the FSID or SDC data. We classify firms into industries using their primary SIC code. Non-financial firms are classified according to the Fama-French 12 Industry classifications. Financial firms (which fall under the “Money” Fama-French 12 industry classification) are further classified into banking, trading, and other financial based on the Fama-French 48 industry classifications.5 We end up with fourteen industries: Banking, business equipment, chemicals, consumer durables, consumer non-durables, energy, health care, manufacturing, other financial, other non-financial, shops, telecommunications, trading, and utilities. Typically banking consists ofcommercialbanks,tradingconsistsofinvestmentbanksandbrokerages,andotherfinancial consists of insurance companies. To measure the effect of liquidity or economies of scale in debt issuance, we analyze two variables designed to reflect liquidity benefits for particular firms. We measure CDS liquidity 4Weaddafixedquantity,onethousandbasispoints,beforetakingthelogtoincludebondswithnegative yield spreads in the regression analysis. 5A detailed mapping from SIC codes to Fama-French industry classifications is available at Ken French’s web page: http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/. 12

using Markit data. Depth measures the daily number of contributor prices used to calculate the 5 year CDS spread. It is essentially a measure of the amount of trading in comparable CDS contracts. We measure economies of scale in debt issuance using Raise, the amount of debt (in millions) issued by a firm over the previous five years according to the Mergent FSID database. Appendix Table A.1 in the provides details on the definition of each variable we form and the associated data source. Tables 1 presents means for the analysis variables in the CDS sample by firm industry. CDS spreads range in value from 401 (for Consumer Durables) to 119 (for Utilities). The financial industries have mean spreads that range from 178 for Banking to 269 for Other Financial. Banking firms have the largest mean size at 12.1 in log units, which translates to 18 billion dollars. Banking firms are also the most highly levered, given their unique capital structure consisting of deposits as well as bonds and equity. Trading firms have leverage ratios in line with Telecommunications and Utilities firms at around 47 percent. Half of banking firms in the sample lie in the top 5 percent of the size distribution compared to 20 percent of trading firms. These percentage are in line with other industries. Table 2 presents mean statistics for the bond sample and shows similar patterns. Variation of these statistics over time periods we analyze is summarized in the Appendix Tables A.2 and A.3. Both spreads and EDFs are quite low for banking firms prior to the crisis (2004Q1:2008Q2) compared to other industries and then spike during the crisis period 13

(2008Q3:2009Q2). BankingfirmspreadsandEDFsfallinthepost-crisisperiod(2009Q3:2013Q2). Table 3 presents the number of unique firms by industry and year for CDS (Panel A), and bond (Panel B) samples. There are a fairly constant number of firms in each year in both samples. 4. Results We begin our inter-industry analysis of the effect of size on borrowing costs by focusing on CDS spreads. To this end, we focus generally on the coefficient on size in credit spread regressions featuring various controls. We begin by analyzing all industries together and measuring the average impact of size on credit spreads for all borrowers across several specifications. After presenting aggregate results, we consider models that allow size-related coefficients to vary across industries, and interpret variation in these coefficients as industryspecific size effects. This approach is based on the premise that the existence of material bailout subsidies for financial borrowers would lead to rejection of the null hypothesis that the size effect for financial borrowers is no larger than that for borrowers in other industries. We also consider several size variables and specifications to account for the fact that industries have different underlying size distibutions. 14

4.1. The size effect for banking and trading financial firms We begin by examining the raw differences in borrowing costs and the size effect for the two largest categories of financial firms - banking and trading - compared to all other industries. We do this to set a baseline and to compare our results to other studies that compare funding costs of financial firms to those of non-financial firms. In Table 4, we present regression results based on estimating a baseline model with two financial industry indicator variables: Log(Spread) =β LogSize +β EDF +β EDF2 i,j,t 1 i,t 2 i,t 3 i,t +β ∗I ∗EDF +β ∗I ∗EDF2 +β ∗I ∗LogSize 4 Banking i,t 5 Banking i,t 6 Banking i,t +β ∗I ∗EDF +β ∗I ∗EDF2 +β ∗I ∗LogSize 7 Trading i,t 8 Trading i,t 9 Trading i,t +Industry FE +Quarter FE +ε j t i,t (1) where i indexes firms, j indexes industries, t indexes time (quarter), and I and I Banking Trading represent indicators for banking and trading industries. Log(Spread) is the log of the 5-year CDS spread. Our primary measure of size, LogSize, is the logarithm of book assets. We include both EDF and its square in specifications starting in Column 2 to capture nonlinear default risk-related effects. We also estimate the model in Column 3 (which corresponds directly to Equation 1) for pre-crisis, crisis, and post-crisis sub-samples (Columns 4 to 6). In Table 4, we highlight banking and trading industry fixed effects, as well as interactions 15

ofbankingandtradingindicatorvariableswithdefaultriskandsizemeasures. Thecoefficient on size (measured by log assets) is negative and significant across all specifications except that for the crisis subsample (Column 5). This can be interpreted as an average effect: larger firms generally borrow more cheaply than smaller firms. This result remains even after controlling for non-linear effects of credit risk. Results for banking and trading industries here are surprising: aside from the crisis period, the fixed effects for each industry are negative, but each industry’s interaction with LogSize is positive. These results suggest that generally, banking and trading exhibit lower overall borrowing costs, but do not enjoy size-related borrowing cost advantages relative to the set of other industries.6 During the crisis period, these patterns reverse, with positive coefficients on financial industry indicator variables and negative coefficients on their interaction with size. We don’t find a negative result on LogSize during the crisis, but the result for the interaction of the size and indicators for banking and trading are negative during the crisis. These crisis results are significant for Trading firms but not for Banking firms. This suggests some evidence of financial industry size-related borrowing advantages during the financial crisis, which we will examine in more detail later. To complement our CDS-based analysis, we examine corporate bond trade yield spreads 6These estimates of the size effect stand in contrast to estimates in other papers, such as Acharya et al. (2013))TheloweroverallCDSspreadsofbankingfirmspriortothefinancialcrisisisconsistentwitharecent study by Kelly et al. (2011) who find that options price lower overall risk in the financial industry. 16

using bond trading data in TRACE. We estimate the following equation: Log(Spread) =β LogSize +β EDF +β EDF2 i,j,t 1 i,t 2 i,t 3 i,t +β ∗I ∗EDF +β ∗I ∗EDF2 +β ∗I ∗LogSize 4 Banking i,t 5 Banking i,t 6 Banking i,t +β ∗I ∗EDF +β ∗I ∗EDF2 +β ∗I ∗LogSize 7 Trading i,t 8 Trading i,t 9 Trading i,t +β Maturity +Industy FE +Quarter FE +ε 10 i,t j t i,t (2) We present analogous results based on estimating Equation 2 in Table 5. The model we estimate is is similar to that in Equation 1 but controls for bond maturity. Results from bond data present some similar patterns compared with CDS results: larger firms generally enjoyborrowingcostadvantages,asdobankingandtradingfirms. Bankingandtradingfirms have lower overall average costs outside the crisis (as exhibited by the coefficients on industry indicator variables). However, size-industry interactions are actually positive and significant for these firms, suggesting little evidence of unusual size-related financial industry borrowing cost advantages. The coefficient on size is negative and significant across specifications, and positive coefficients on size-industry interactions for banking and trading go from positive to insignificant during the crisis. During the financial crisis period (Column 5 of Table 5), we also see different results here than for CDS data we analyze in Table 4: the size coefficient remains negative and significant, and industry fixed effects for banking and trading, as well as corresponding industry-size interactions, are statistically insignificant. 17

4.2. The size effect across industries WhileresultsinTables4and5presentsuggestiveevidence, theyalsoinvolvecomparisons of industry-size effects for two financial industries with a set of all other industries. To allow forcomparisonofsucheffectsinfinancialindustrieswithanalogouseffectsinotherindividual industries, we further consider estimating effects separately for each industry. We augment the regression in Equation 1 to include size-related and default-risk effects for each individual industry, and estimate the following baseline model: (cid:88) (cid:88) (cid:88) Log(Spread) = β LogSize + β EDF + β EDF2 i,j,t 1,j i,t 2,j i,t 3,j i,t j j j (3) +Industy FE +Quarter FE +ε j t i,t Table 6, Column 1, displays full-sample estimates of β , the coefficient on size-industry 1,j interactions for each industry. Columns 2, 3 and 4 estimate the same specification across time periods. This table illustrates the primary observation in this paper: the size effect in financial firm borrowing costs is not unusual. Here, size-industry interactions are generally negative for financial firms, as we might expect. However, the magnitude of these negative effects is not unusually high in financial industries relative to analogous estimates for effects in other industries. In fact, it is not even highest in financial industries, as we find larger (negative) effects for industries like business equipment and energy. For several industries not considered bailout candidates, the size effect is larger than it is for financial industries. For example, the banking industry has coefficients on size of -0.003 in the Pre-Crisis period 18

and -0.24 in the Post-Crisis period; while the Shops industry has size coefficients of -0.045 and -0.030 and the Business Equipment industry has size coefficients of -0.032 and -0.028. There is a notable exception: during the financial crisis, the size effect is higher for banking and trading firms than for firms in other industries. The crisis affects banking and trading differently than other industries. Most industries see the magnitude of the size effect decrease during the financial crisis, whereas banking and trading see the size effect increase to-0.057and-0.036, respectively. Howeverthesizeeffectforthosetwoindustriesdecreasesin the post-crisis period. The evidence is consistent with both expectations and actualizations of government support for banking and trading during the crisis, but disappearing before and after. The consumer durables industry seems anomalous in several specifications, with positive size effects. These results emerge because of the lower number of firms in this industry, combined with distress or near-distress in larger firms (auto manufacturers) during much of the sample period. Consumer durables exhibits a significant positive size effect, indicating that larger firms borrow at higher rates. 4.3. Nonlinear size effect Because implicit guarantees may only apply to the largest firms, and because the underlying firm size distribution varies across industries, we also consider a specification that includes both our size measure (logarithm of book assets) and an indicator variable for firms 19

above the 95th percentile for size in their industry (as measured by all firms in the industry within the Compustat universe of firms). This allows estimation of both discrete and continuous size effects. We estimate the following specification for pre-crisis, crisis, and post-crisis periods: (cid:88) (cid:88) (cid:88) (cid:88) Log(Spread) = β LogSize + β Size95 + β EDF + β EDF2 i,j,t 1,j i,t 2,j i,t 3,j i,t 4,j i,t j j j j +Industy FE +Quarter FE +ε j t i,t (4) Results, presented in Table 7, illustrate a consistent pattern: the importance of size for financial firms’ borrowing costs does not appear unusually large relative to that for other industries. Here, we estimate a discrete effect that enters for the largest firms, measured by β in Equation 4 for each industry. Our results suggest that while there may be borrowing 2,j cost advantages for the largest firms, they are not highest for financial firms, as other firms exhibit higher size-related borrowing cost advantages for the overall size effect (coefficient on LogSize) and for a nonlinear large firm effect (coefficient on Size95). While they are not unusually high or highest across industries, we do find some evidence ofhighsize-relatedborrowingcostadvantagesforthelargest5%ofbankingfirmsbothoverall and during the financial crisis, but the coefficients are statistically insignificant. 20

4.4. Results with bond data In Tables 8 and 9, we present results of estimating Equations 3 (size effect across industries) and 4 (nonlinear size effect) for the bond sample. These results illustrate a similar pattern: size-based borrowing cost advantages for financial firms are not unusually large compared with similar effects in other industries. Results from Table 8 largely mirror those from Table 6: analyzing both samples suggest the size effect within financial firms is largest in the other financial category, although they are still not largest among all industries. Similarly, results using bond data presented in Table 9 are similar to those using CDS data in Table 7. The continuous and discrete size effects are both higher in a number of non-financial industries than they are in the financial industries. The bond sample presents stronger evidence of a full-sample nonlinear size effect in banking, although this effect does not appear in the post-crisis sample. During this period, by contrast, we find that the largest trading firms experience significant cost disadvantages in the presence of size-based borrowing advantages for the industry overall. 4.5. Robustness to size variable While we focus primarily on the logarithm of book assets, a commonly used size measure, we also examine results for alternative measures: the Size95 variable described in the previous section, and a standardized version of log total assets, which measures size in units of standard deviation from the mean log of total assets for all other Compustat firms in the 21

same industry. The standardized size measure allows another way of adjusting for the fact that the underlying firm size distributions across industries varies. Analysis with standardized size variables is presented in Appendix Tables A.4 (for the CDS sample) and A.5 (for the bond sample). We also restrict our analysis to the set of firms with both non-missing CDS and bond data, reported in Appendix Table A.6, and find that our main results are robust. To aid in the comparison of the esimated size effects by size variable, we graph he coefficients for pre-crisis, crisis and post-crisis size-industry interaction terms are illustrated for CDS data in Figures 1 through 3 and for bond data in Figures 4 through 6. Since the coefficients on different size measures are in different units, we instead plot normalized size coefficients which are comparable across different size measures. We normalize the size coefficients by subtracting the mean coefficient for a given specification and dividing by the standard error of the coefficients. The figures illustrate that both the pre- and post-crisis size effects for financial firms are similar to those for other industries. These results generally support our conclusions. For the banking industry during the crisis and for other financial firms post-crisis, the estimated coefficient on the industry interaction with Size95 is higher than it is for other industries. 22

4.6. Robustness to default risk specification In the previous analysis, we relied on the EDF measure to proxy for firm-speciffic default risk. It is possible that EDF may not accurately capture non-government support default risk if the EDF reflects lower default probability due to higher equity values that may result from implicit guarantees of financial firms’ bonds. If this is the case, then EDF could be biased downward, reflecting the implicit government guarantee. Thus, we also estimate our baseline regressions using alternative measures of default risk, namely firm leverage and credit rating. This analysis with alternative default risk specifications based on leverage and credit ratings are presented in Appendix Tables A.7 (for the CDS sample) and A.8 (for the bond sample). Our results our robust to these alternative default risk specifications. In particular, we only observe larger than average size effects for the financial industries during the crisis period. 5. Explaining the size effect Having documented the size effect under using a number of size variables and specifications, we proceed to examine possible explanations. We consider two candidates: liquidity and recovery rates. If larger firms enjoy economies of scale in debt issuance, their borrowing costs could be systematically lower. Similarly, higher expected recovery rates could lead to lower borrowing costs holding the probability of default constant. 23

5.1. Liquidity In order to test the hypothesis that larger firms have lower funding costs due to more liquid markets for their securities or economies of scale from seasoned issuance, we examine the impact of adding additional control variables. Raise is debt issuance over the past five years and Depth measures the amount of trading in a particular CDS contract. See Section 3 for a full explanation of these variables. We interact these additional variables by industry to capture industry-specific effects. To measure the effect of our liquidity variables, we estimate the following equation: (cid:88) (cid:88) (cid:88) Log(Spread) = β LogSize + β Log(Raise )+ β Depth + i,j,t 1,j i,t 2,j i,t 3,j i,t j j j (5) (cid:88) (cid:88) β EDF + β EDF2 +Industy FE +Quarter FE +ε 4,j i,t 5,j i,t j t i,t j j Table 10 presents results of estimating Equation 5 for the liquidity-augmented specification. The main result remains: the size effect in financial industries is similar to the size effect in non-financial industries outside of the financial crisis, and in fact size effects are larger in several industries than in any financial industry. During the financial crisis, trading firms appear to enjoy unusual size-related borrowing cost advantages, while there appears to be no size-related borrowing cost advantage in non-trading financial industries. The dramatic changes in the size effect documented in Table 10 suggest that liquidityrelated factors explain an unusually large portion of the size effect during the crisis period. After controlling for liquidity, the size effect in banking drops in magnitude during the 24

financial crisis to nearly zero. In contrast, the size effect in trading approximately doubles in size as compared to both the pre-crisis period and the sub-sample regressions from Table 6. The size effect for other financial industries does not respond significantly to liquidity-related controls. 5.2. Recovery rates Ourearlieranalysiscontrolsfordefaultrisk,butrecoveryrates-ortheirreverse,lossgiven default (LGD) - are another important component of credit risk. If larger firms have lower expected LGD, they should have lower CDS spreads holding default probabilities constant. LGD by itself is unlikely to fully explain the size effect; the important question is what causes changes in risk-neutral estimates of LGD. A firm could have a lower LGD due to the type of assets it has, due to public bailouts, or due to correlation between expected LGD and marginal utility. Determining whether LGD is related to borrower size cannot fully address these questions, but it can shed light on the avenue through which the size effect operates. To consider these issues, we compare firm size to realized LGD. Our sample includes all default events from April 1987 to April 2013 from Moody’s Analytics, Default and Recovery Database with LGD > 0. We classify firms into industries using the methodology outlined in Section 3. We calculate firm size as the face value of debt outstanding at the time of the default, which assumes that the value of equity is negligible during default. Table 11 documents the number of defaults by industry along with the average size and LGD. 25

As illustrated in Table 11, the three financial industries have few realized default observations. Additionally, these observations consist of relatively small firms (by contrast, Table 1 suggests financial firms in our CDS sample are relatively large). Because government intervention could occur to prevent failure, we are cautious about interpreting realized LGD results for financial institutions. However, the existence of a relationship between size and LGD7 in other industries suggests the possibility that larger firms have higher recovery rates for non-bailout related reasons. If large financial firms had a higher recovery rate than small financial firms even excluding the possibility of bailouts, then we would expect to see a size effect in financial industries. While we cannot empirically test this possibility due to the limited number of failures of large financial institutions, we can see that larger firms have higher recovery rates in other industries in situations that do not involve government intervention. If some of the same factors that apply to large non-financial firms also apply to large financial firms, then fundamental-based differences in recovery rates may drive part of the size effect. Future research is needed to identify the factors which drive the relationship between recovery rates and size and determine whether they would apply to the financial industry as well. 7In Appendix Table A.9, we present results of regressions of realized LGD on log size (Columns 1-3) and realizedLGDonlogsize-industryinteractions(Column4). WefindthatlargerfirmshaveloweraverageLGD. Interacting size with industry variables suggests that this relationship holds for most industries, although it is statistically significant only for three: business equipment, consumer durables, and other non-financial. 26

6. Interpreting our results Most of our analysis involves regressing log spread on log size interacted across industries. To help interpret the magnitude of our results, we consider the impact of a 10% increase in size on borrowing costs implied by our results. For coefficient β on log of assets and a borrower with spread s, the change in spread implied by a 10% increase in assets is given by: ∆s = 1.1β −1. (6) s In Table 12, we summarize results across these Tables 6, 8, and 10 in terms of the expected percentage change in borrowing costs from a 10% increase in assets. For the full sample, a 10% increase in size reduces borrowing costs for banking, trading, and other financial firms by approximately 1%, while this quantity is around 4% for firms in the shops, business equipment, and chemicals industries. The effect for financial firms is highest (2.8%) for trading firms during the crisis when we control for liquidity (Table 10). This is the only case where our estimated size effect is highest in a financial industry. In other periods and other specifications, there is always a non-financial industry with a larger estimated size effect than that for the banking, trading, and other financial industries. 27

7. Conclusion Our analysis examines both industry-specific and time-varying effects of borrower size on the cost of borrowing, using both CDS and corporate bond spreads over the period 2004 to 2013. Across industries, the largest borrowers enjoy borrowing cost advantages. Consideration of variation in the size effect across industries suggests that borrowing advantages for the largest financial firms do not seem unusual. With the exception of investment banking and trading firms during the recent financial crisis, the size effect is not generally largest for financial borrowers. We find qualitatively similar results using a variety of size measures including both continuous measures that adjust for firm size distribution with an industry and indicator variables identifying the largest borrowers. We present evidence that suggests the size effect is partially driven by liquidity and recovery rates. Our results suggest that researchers may overestimate the size of too-big-to-fail subsidies if they do not take into account the lower borrowing costs of larger firms across a variety of industries. However, prior to the financial crisis, we also find that financial firms exhibited generally lower spreads that were less sensitive to size than spreads for several other industries. Our results suggest that estimates of implicit government guarantees to financial firms may overemphasize size-related borrowing cost differentials. This suggests that investor expectations of government support, or generally reduced risk perceptions, may have reduced borrowing costs for the financial industry, as a whole. 28

References Acharya, V., D. Anginer, and J. Warburton. 2013. The end of market discipline? Investor expectations of implicit state guarantees. Working paper . Acharya, V. V., S. T. Bharath, and A. Srinivasan. 2007. Does industry-wide distress affect defaulted firms? Evidence from creditor recoveries. Journal of Financial Economics 85:787–821. Araten, M., and C. Turner. 2013. Understanding the funding cost differences between global systemically important banks (GSIBs) and non-GSIBs in the USA. Journal of Risk Management in Financial Institutions pp. 387–410. Banz, R. W. 1981. The relationship between return and market value of common stocks. Journal of Financial Economics 9:3–18. Bassett, W. F. 2014. Using Insured Deposits to Refine Estimates of the Large Bank Funding Advantage. Working paper . Berk, J.B.1995. Acritiqueofsize-relatedanomalies. Review of Financial Studies 8:275–286. Brewer III, E., and J. Jagtiani. 2013. How much did banks pay to become too-big-to-fail and to become systemically important? Journal of Financial Services Research 43:1–35. 29

Carey, M., and M. Gordy. 2009. The bank as grim reaper: Debt composition and bankruptcy thresholds. Working paper . Chan, K., and N.-F. Chen. 1988. An Unconditional Asset-Pricing Test and the Role of Firm Size as an Instrumental Variable for Risk. The Journal of Finance 43:309–325. Demirguc-Kunt, A., and H. Huizinga. 2013. Are banks too big to fail or too big to save? International evidence from equity prices and CDS spreads. Journal of Banking and Finance 37:875–894. Dick-Nielsen, J., P. Feldhu¨tter, and D. Lando. 2012. Corporate bond liquidity before and after the onset of the subprime crisis. Journal of Financial Economics 103:471–492. Doshi, H. 2011. The term structure of recovery rates. Working paper . Fama, E. F., and K. R. French. 1992. The cross-section of expected stock returns. Journal of Finance 47:427–462. Flannery, M.J., andS.M.Sorescu.1996. Evidenceofbankmarketdisciplineinsubordinated debenture yields: 1983–1991. The Journal of Finance 51:1347–1377. Gandhi, P., and H. Lustig. forthcoming. Size anomalies in U.S. bank stock returns. Journal of Finance . 30

GAO, U. S. 2014. Evidence from the bond market on banks’ “too-big-to-fail” subsidy. GAO 14-621, United States Government Accountability Office Report to Congress . Gropp, R., H.Hakenes, andI.Schnabel.2011. Competition, risk-shifting, andpublicbail-out policies. Review of Financial Studies 24:2084–2120. Houweling, P., and T. Vorst. 2005. Pricing default swaps: Empirical evidence. Journal of International Money and Finance 24:1200–1225. Hughes, J.P., andL.J.Mester.2013. Whosaidlargebanksdontexperiencescaleeconomies? Evidence from a risk-return-driven cost function. Journal of Financial Intermediation 22:559–585. Hughes, J. P., L. J. Mester, and C.-G. Moon. 2001. Are scale economies in banking elusive or illusive?: Evidence obtained by incorporating capital structure and risk-taking into models of bank production. Journal of Banking & Finance 25:2169–2208. Iyer, R., M. Puri, and N. Ryan. 2013. Do depositors monitor banks? National Bureau of Economic Research Working Paper No. 19050 . Jacewitz, S., and J. Pogach. 2012. Deposit rate advantages at the largest banks. Working paper . Kelly, B. T., H. Lustig, and S. Van Nieuwerburgh. 2011. Too-systemic-to-fail: What option 31

markets imply about sector-wide government guarantees. National Bureau of Economic Research Working Paper No. 17149 . Kovner, A., J. Vickery, and L. Zhou. 2014. Do big banks have lower operating costs? Economic Policy Review December. Kroszner, R. S. 2013. A review of bank funding cost differentials. Working paper . Li, Q. S., Zan, and J. Zhang. 2011. Quantifying the value of implicit government guarantees for large financial institutions. Moody’s Analytics Report . Merton, R. C. 1974. On the pricing of corporate debt: The risk structure of interest rates. The Journal of Finance 29:449–470. Noss, J., and R. Sowerbutts. 2012. The implicit subsidy of banks. Bank of England Financial Stability Paper 15. O’Hara, M., and W. Shaw. 1990. Deposit insurance and wealth effects: the value of being too big to fail. The Journal of Finance 45:1587–1600. Penas, M., and H. Unal. 2004. Gains in bank mergers: Evidence from the bond markets. Journal of Financial Economics 74:149–179. Santos, J. 2014. Evidence from the bond market on banks’ “too-big-to-fail” subsidy. Economic Policy Review 20. 32

Tsesmelidakis, Z., and R. C. Merton. 2013. The value of implicit guarantees. Working Paper . Ueda, K., and B. Weder di Mauro. 2013. Quantifying structural subsidy values for systemically important financial institutions. Journal of Banking & Finance 37:3830–3842. Wheelock, D. C., and P. W. Wilson. 2012. Do large banks have lower costs? New estimates of returns to scale for US banks. Journal of Money, Credit and Banking 44:171–199. 33

snaem elpmas SDC :1 elbaT gnitaR 59eziS esiaR htpeD egareveL eziSdtS eziSgoL FDE daerpS yrtsudnI 3.22 5.0 0.03 9.5 6.36 1.2 1.21 35.1 871 gniknaB 5.12 2.0 6.2 2.7 9.82 2.1 6.01 85.1 962 laicnaniF rehtO 5.91 2.0 3.3 1.5 5.74 1.1 2.9 81.1 602 gnidarT 4.91 7.0 1.4 5.5 5.81 8.1 4.9 07.0 761 tnempiuqE ssenisuB 1.12 4.0 1.5 4.6 2.02 7.1 0.9 07.0 821 slacimehC 4.91 1.0 8.0 7.5 0.73 0.1 7.8 83.2 104 selbaruD remusnoC 7.91 2.0 8.1 5.6 8.72 2.1 8.8 40.1 202 selbaruD-noN remusnoC 7.81 3.0 5.2 6.6 3.42 4.1 5.9 00.1 251 ygrenE 6.71 6.0 4.2 8.5 9.52 7.1 8.8 05.0 461 eraC htlaeH 8.02 2.0 4.1 1.6 2.62 2.1 9.8 19.0 561 gnirutcafunaM 1.81 5.0 9.1 2.6 9.63 5.1 0.9 83.2 743 laicnaniF-noN rehtO 0.81 4.0 0.2 2.7 1.72 3.1 0.9 41.1 222 spohS 1.71 2.0 9.3 0.6 5.64 9.0 5.9 40.2 203 snoitacinummoceleT 1.02 1.0 6.1 8.5 5.64 8.0 7.9 55.0 911 seitilitU tluafed detcepxe ,daerps SDC raey-5 edulcni selbairaV .smrfi 566 dna snoitavresbo mrfi-retrauq 684,71 fo stsisnoc elpmas SDC gniylrednu ehT tatsupmoC rof stessa fo gol fo naem eht morf noitaived dradnats fo stinu ni ezis serusaem( stessa gol dezidradnats ,stessa gol ,)FDE( ytilibaborp revo desiar tbed fo mus ,)stcartnoc SDC elbarapmoc ni gnidart fo tnuoma eht fo erusaem a( htped SDC ,egarevel tekram ,)yrtsudni emas eht ni smrfi elacs 82-1 a no desserpxe( gnitar tiderc dna ,smrfi tatsupmoC yrtsudni-nihtiw ssorca ezis fo elitnecrep ht59 rof rotacidni na ,)M$ ,esiaR( sraey 5 tsap 21 hcnerF-amaF eht ot gnidrocca defiissalc era smrfi laicnanfi-noN .snoitinfied elbairav deliated erom rof .1.A elbaT xidneppA eeS .)AAA=82 htiw ,gniknab otni defiissalc rehtruf era )noitacfiissalc yrtsudni 21 hcnerF-amaF ”yenoM“ eht rednu llaf hcihw( smrfi laicnaniF .snoitacfiissalc yrtsudnI .snoitacfiissalc yrtsudni 84 hcnerF-amaF eht no desab laicnanfi rehto dna ,gnidart 34

snaem elpmas dnoB :2 elbaT gnitaR 59eziS ytirutaM egareveL eziSdtS eziSgoL FDE daerpS yrtsudnI 3.22 07.0 6.6 1.07 6.2 1.31 2.1 812 gniknaB 6.91 11.0 8.9 7.92 1.1 5.01 1.1 192 laicnaniF rehtO 6.22 97.0 2.4 5.58 4.2 0.31 8.0 661 gnidarT 0.12 58.0 8.8 0.61 1.2 2.01 8.0 761 tnempiuqE ssenisuB 0.32 28.0 2.9 5.91 3.2 5.01 5.0 241 slacimehC 5.31 98.0 2.12 5.17 0.2 0.21 4.1 826 selbaruD remusnoC 2.02 65.0 8.9 0.42 5.1 7.9 2.0 091 selbaruD-noN remusnoC 5.91 37.0 0.02 7.02 6.1 9.9 6.0 722 ygrenE 5.02 18.0 3.01 0.41 7.1 8.8 2.0 302 eraC htlaeH 8.91 56.0 5.01 4.82 6.1 9.9 8.0 202 gnirutcafunaM 7.71 47.0 3.7 9.73 8.1 8.9 3.2 433 laicnaniF-noN rehtO 0.12 27.0 4.9 6.42 0.2 6.01 6.0 091 spohS 8.51 40.0 6.3 7.44 0.1 1.9 4.01 524 snoitacinummoceleT 2.71 50.0 9.7 1.05 7.0 4.9 6.0 352 seitilitU ytilibaborp tluafed detcepxe ,daerps dnob edulcni selbairaV .smrfi 913 dna snoitavresbo mrfi-retrauq 600,61 fo stsisnoc elpmas dnob gniylrednu ehT eht ni smrfi tatsupmoC rof stessa fo gol fo naem eht morf noitaived dradnats fo stinu ni ezis serusaem( stessa gol dezidradnats ,stessa gol ,)FDE( tiderc dna ,smrfi tatsupmoC yrtsudni-nihtiw ssorca ezis fo elitnecrep ht59 rof rotacidni na ,sraey ni ytirutam dnob ,egarevel tekram ,)yrtsudni emas defiissalc era smrfi laicnanfi-noN .snoitinfied elbairav deliated erom rof .1.A elbaT xidneppA eeS .)AAA=82 htiw elacs 82-1 a no desserpxe( gnitar era )noitacfiissalc yrtsudni 21 hcnerF-amaF ”yenoM“ eht rednu llaf hcihw( smrfi laicnaniF .snoitacfiissalc yrtsudnI 21 hcnerF-amaF eht ot gnidrocca .snoitacfiissalc yrtsudni 84 hcnerF-amaF eht no desab laicnanfi rehto dna ,gnidart ,gniknab otni defiissalc rehtruf 35

Table 3: Number of unique firms by industry and year Industry 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Panel A: CDS sample Banking 12 15 18 20 20 18 18 18 19 19 Other Financial 28 31 37 39 39 35 34 34 32 31 Trading 23 27 32 39 39 35 35 36 37 32 Business Equipment 30 36 46 49 43 38 35 34 33 34 Chemicals 36 45 49 52 50 44 39 35 36 35 Consumer Durables 10 11 14 14 15 14 13 13 12 9 Consumer Non-Durables 29 32 36 37 36 33 32 33 33 33 Energy 23 25 27 33 33 31 30 28 28 23 Health Care 11 11 14 16 16 14 13 13 14 12 Manufacturing 58 61 69 75 72 67 66 63 60 56 Other Non-Financial 46 53 64 69 71 67 67 62 65 59 Shops 42 44 50 54 52 45 45 41 41 40 Telecommunications 15 18 24 29 30 25 22 17 17 17 Utilities 33 36 41 45 45 46 48 43 40 35 Panel B: Bond sample Banking 22 25 34 32 30 27 25 31 29 25 Other Financial 16 21 28 28 24 21 19 17 15 13 Trading 7 10 12 10 12 6 6 6 5 3 Business Equipment 10 12 15 13 8 8 6 4 4 4 Chemicals 12 17 22 22 17 12 10 9 9 6 Consumer Durables 2 2 4 3 3 2 3 3 2 2 Consumer Non-Durables 15 17 21 15 15 15 13 12 12 12 Energy 5 3 8 7 6 6 4 5 4 4 Health Care 3 4 5 4 4 3 3 3 3 3 Manufacturing 16 23 26 23 22 23 22 21 17 11 Other Non-Financial 16 21 26 22 20 14 14 13 13 10 Shops 17 21 23 20 13 12 11 12 8 8 Telecommunications 4 7 7 6 3 2 2 1 2 1 Utilities 4 8 20 16 8 6 7 4 3 1 36

Table 4: Regressions of log CDS spread on firm size for banking and trading financial firms (1) Full (2) Full (3) Full (4) Pre-Crisis (5) Crisis (6) Post-Crisis EDF 0.042 (***) 0.072 (***) 0.073 (***) 0.088 (***) 0.081 (***) 0.063 (***) (0.003) (0.005) (0.005) (0.009) (0.008) (0.005) Banking 0.045 0.057 0.116 -0.158 (***) 0.985 (**) 0.147 (0.126) (0.088) (0.110) (0.055) (0.460) (0.175) Banking*EDF -0.022 (*) 0.031 -0.032 -0.040 (**) (0.012) (0.035) (0.022) (0.016) Banking*EDF2 0.000 -0.006 0.001 0.001 (0.001) (0.004) (0.001) (0.001) EDF2 -0.001 (***) -0.001 (***) -0.002 (***) -0.001 (***) -0.001 (**) (0.000) (0.000) (0.001) (0.000) (0.000) LogSize -0.016 (***) -0.012 (***) -0.011 (***) -0.014 (***) 0.026 (***) -0.018 (***) (0.004) (0.004) (0.004) (0.004) (0.009) (0.004) LogSize*Banking -0.005 -0.006 -0.009 0.011 (**) -0.083 (**) -0.006 (0.010) (0.007) (0.009) (0.005) (0.036) (0.014) LogSize*Trading 0.002 -0.005 -0.005 0.002 -0.062 (***) 0.001 (0.006) (0.006) (0.006) (0.005) (0.017) (0.007) Trading -0.033 0.029 0.039 -0.059 0.658 (***) -0.004 (0.051) (0.052) (0.052) (0.047) (0.157) (0.069) Trading*EDF -0.005 0.081 (*) -0.004 -0.015 (*) (0.009) (0.044) (0.019) (0.007) Trading*EDF2 0.000 -0.014 (**) -0.000 0.000 (0.000) (0.006) (0.001) (0.000) Adj.R2 0.585 0.618 0.621 0.571 0.625 0.628 N 17,486 17,486 17,486 8,227 1,984 7,275 ThistablepresentsresultsfromestimatingEquation1forCDSsampleoverthefulltimesample(04Q1-13Q2, Columns 1-3) and for the pre-crisis sample (04Q1-08Q2, Column 4), the financial crisis sample (08Q3-09Q2, Column5),andthepost-crisissample(09Q3-13Q2,Column6). EDFistheexpecteddefaultfrequency,Log- Sizeislogofbookassets, andBankingandTradingrepresentindustryindicatorvariables. Allspecifications includeindustryandquarterfixedeffectsandreportstandarderrorsclusteredatthefirmlevel. ***,**,and * indicate statistical significance at the 1%, 5% and 10% levels, respectively. 37

Table 5: Regressions of log bond spread on firm size for banking and trading financial firms (1) Full (2) Full (3) Full (4) Pre-Crisis (5) Crisis (6) Post-Crisis EDF 0.022 (***) 0.055 (***) 0.069 (***) 0.062 (***) 0.090 (***) 0.078 (***) (0.002) (0.008) (0.005) (0.008) (0.009) (0.010) Banking -0.212 (**) -0.148 (*) -0.091 -0.086 0.165 -0.287 (**) (0.096) (0.083) (0.081) (0.061) (0.247) (0.130) Banking*EDF -0.026 (**) 0.052 (**) -0.060 (***) -0.012 (0.012) (0.021) (0.015) (0.013) Banking*EDF2 0.001 (*) -0.005 (***) 0.002 (***) 0.001 (**) (0.000) (0.002) (0.001) (0.001) EDF2 -0.001 (***) -0.002 (***) -0.001 (***) -0.002 (***) -0.003 (***) (0.000) (0.000) (0.000) (0.000) (0.001) LogSize -0.044 (***) -0.036 (***) -0.033 (***) -0.018 (***) -0.052 (***) -0.048 (***) (0.005) (0.005) (0.005) (0.005) (0.008) (0.006) LogSize*Banking 0.023 (***) 0.015 (**) 0.011 (*) 0.006 -0.001 0.029 (***) (0.008) (0.007) (0.007) (0.006) (0.019) (0.010) LogSize*Trading 0.036 (***) 0.029 (***) 0.026 (***) 0.014 (**) 0.025 0.040 (***) (0.009) (0.009) (0.009) (0.006) (0.026) (0.012) Maturity 0.001 (***) 0.001 (***) 0.001 (***) 0.001 (***) -0.001 (**) 0.001 (***) (0.000) (0.000) (0.000) (0.000) (0.001) (0.000) Trading -0.398 (***) -0.332 (***) -0.268 (**) -0.172 (***) -0.137 -0.450 (***) (0.113) (0.116) (0.113) (0.064) (0.340) (0.162) Trading*EDF -0.044 (***) -0.011 -0.071 (***) 0.024 (0.014) (0.013) (0.015) (0.041) Trading*EDF2 0.001 (**) -0.003 (***) 0.002 (***) -0.014 (0.000) (0.001) (0.001) (0.010) Adj.R2 0.610 0.648 0.659 0.610 0.573 0.657 N 15,972 15,972 15,972 8,303 1,831 5,838 ThistablepresentsresultsfromestimatingEquation2forbondtrading(TRACE)sampleoverthefullsample (04Q1-13Q2, Columns 1-3) and for pre-crisis (04Q1-08Q2, Column 4), financial crisis (08Q3-09Q2, Column 5), and post-crisis (09Q3-13Q2, Column 6) subsamples. EDF is the expected default frequency, LogSize is log of book assets, and Banking and Trading represent industry indicator variables. All specifications includeindustryandquarterfixedeffectsandreportstandarderrorsclusteredatthefirmlevel. ***,**,and * indicate statistical significance at the 1%, 5% and 10% levels, respectively. 38

Table 6: Regression of log CDS spread on firm size across industries and time periods (1) Full (2) Pre-Crisis (3) Crisis (4) Post-Crisis LogSize* Banking -0.020 (**) -0.003 -0.057 (*) -0.024 (*) (0.008) (0.003) (0.035) (0.013) LogSize* Other Financial -0.014 (*) -0.009 (***) 0.037 (**) -0.036 (***) (0.007) (0.003) (0.015) (0.013) LogSize* Trading -0.016 (***) -0.013 (***) -0.036 (***) -0.017 (***) (0.004) (0.003) (0.014) (0.006) LogSize* Business Equipment -0.030 (***) -0.032 (***) -0.000 -0.028 (***) (0.006) (0.007) (0.012) (0.008) LogSize* Chemicals -0.027 (***) -0.030 (***) -0.012 -0.021 (***) (0.007) (0.006) (0.011) (0.006) LogSize* Consumer Durables 0.066 (***) 0.059 (***) 0.199 (***) 0.024 (*) (0.017) (0.015) (0.063) (0.014) LogSize* Consumer Non-Durables -0.034 (**) -0.009 0.003 -0.042 (***) (0.014) (0.010) (0.032) (0.013) LogSize* Energy -0.023 (***) -0.022 (***) -0.013 -0.029 (**) (0.007) (0.006) (0.014) (0.012) LogSize* Health Care -0.024 -0.004 -0.028 -0.052 (**) (0.018) (0.009) (0.029) (0.022) LogSize* Manufacturing -0.012 (**) -0.013 (***) 0.004 -0.016 (***) (0.005) (0.005) (0.015) (0.006) LogSize* Other Non-Financial 0.003 0.003 0.036 (***) -0.008 (0.007) (0.010) (0.013) (0.007) LogSize* Shops -0.044 (***) -0.045 (***) -0.001 -0.030 (**) (0.014) (0.017) (0.017) (0.014) LogSize* Telecommunications -0.011 -0.004 -0.003 -0.029 (*) (0.011) (0.013) (0.014) (0.015) LogSize* Utilities 0.004 0.000 0.016 0.004 (0.006) (0.006) (0.016) (0.009) Adj.R2 0.665 0.646 0.675 0.719 N 17,486 8,227 1,984 7,275 This table presents results of estimating Equation 3 across the financial crisis (2008Q3-2009Q2), pre-crisis (2004Q1-2008Q2),andpost-crisis(2009Q3-2013Q2)subsamples. Specificationsincludeindustryandquarter fixed effects and interactions between industry and EDF and industry and EDF2. Standard errors are clustered at the firm level. ***, **, and * indicate statistical significance at the 1%, 5% and 10% levels, respectively. 39

Table 7: Regression of log CDS spread across industries and time period, nonlinear size effect (1) Full (2) Pre-Crisis (3) Crisis (4) Post-Crisis LogSize* Banking -0.012 0.001 -0.032 -0.016 (0.011) (0.004) (0.044) (0.018) Size95* Banking -0.035 -0.017 -0.115 -0.033 (0.023) (0.012) (0.099) (0.036) LogSize* Other Financial -0.012 -0.006 0.013 -0.031 (**) (0.011) (0.005) (0.033) (0.015) Size95* Other Financial -0.010 -0.015 0.119 -0.028 (0.029) (0.013) (0.112) (0.046) LogSize* Trading -0.014 (**) -0.010 -0.031 -0.013 (0.007) (0.007) (0.022) (0.009) Size95* Trading -0.010 -0.012 -0.026 -0.019 (0.030) (0.033) (0.114) (0.030) LogSize* Business Equipment -0.029 (***) -0.031 (***) -0.020 -0.025 (***) (0.006) (0.008) (0.021) (0.007) Size95* Business Equipment -0.002 -0.004 0.064 -0.013 (0.027) (0.027) (0.089) (0.026) LogSize* Chemicals -0.026 (**) -0.029 (***) -0.014 -0.018 (*) (0.011) (0.010) (0.021) (0.011) Size95* Chemicals -0.003 -0.003 0.005 -0.009 (0.018) (0.019) (0.042) (0.024) LogSize* Consumer Durables 0.015 0.002 0.044 -0.020 (0.026) (0.024) (0.070) (0.016) Size95* Consumer Durables 0.288 (***) 0.308 (***) 0.988 (***) 0.240 (***) (0.109) (0.097) (0.277) (0.070) LogSize* Consumer Non-Durables -0.033 -0.009 0.047 -0.040 (*) (0.021) (0.015) (0.052) (0.021) Size95* Consumer Non-Durables -0.005 -0.000 -0.167 (*) -0.007 (0.038) (0.026) (0.093) (0.040) LogSize* Energy -0.028 (***) -0.028 (***) -0.018 -0.043 (***) (0.007) (0.008) (0.017) (0.012) Size95* Energy 0.018 0.019 0.016 0.041 (0.019) (0.014) (0.050) (0.028) LogSize* Health Care -0.022 0.013 -0.054 -0.067 (***) (0.016) (0.010) (0.036) (0.026) Size95* Health Care -0.004 -0.037 (**) 0.059 0.030 (0.019) (0.015) (0.051) (0.023) LogSize* Manufacturing -0.002 -0.003 0.033 -0.010 (0.010) (0.010) (0.028) (0.009) Size95* Manufacturing -0.033 (*) -0.028 -0.093 -0.023 (0.019) (0.020) (0.061) (0.020) 40

Table 7: Regression of log CDS spread across industries and time period, nonlinear size effect (1) Full (2) Pre-Crisis (3) Crisis (4) Post-Crisis LogSize* Other Non-Financial 0.008 0.011 0.038 (**) 0.001 (0.010) (0.016) (0.019) (0.008) Size95* Other Non-Financial -0.019 -0.024 -0.008 -0.028 (0.024) (0.031) (0.061) (0.023) LogSize* Shops -0.056 (***) -0.056 (**) 0.008 -0.039 (**) (0.021) (0.027) (0.032) (0.017) Size95* Shops 0.034 0.028 -0.022 0.025 (0.028) (0.032) (0.050) (0.036) LogSize* Telecommunications 0.018 0.041 0.035 -0.016 (0.020) (0.027) (0.026) (0.023) Size95* Telecommunications -0.140 (**) -0.214 (**) -0.182 (**) -0.056 (0.062) (0.091) (0.081) (0.054) LogSize* Utilities 0.009 0.003 0.024 0.010 (0.008) (0.007) (0.018) (0.012) Size95* Utilities -0.023 -0.016 -0.054 -0.022 (0.018) (0.013) (0.048) (0.022) Adj.R2 0.668 0.653 0.684 0.721 N 17,486 8,227 1,984 7,275 ThistablepresentsresultsofestimatingEquation4overcrisis(2008Q3-2009Q2),pre-crisis(2004Q1-2008Q2), and post-crisis (2009Q3-2013Q2) subsamples. Regression specifications augment Equation 3 with an additional size variable (Size95) to examine nonlinear size effects for the largest firms in each industry. Specifications include industry and quarter fixed effects and interactions between industry and EDF and industry and EDF2. Standard errors are clustered at the firm level. ***, **, and * indicate statistical significance at the 1%, 5% and 10% levels, respectively. 41

Table 8: Regression of log bond spread on firm size across industries and time periods (1) Full (2) Pre-Crisis (3) Crisis (4) Post-Crisis LogSize* Banking -0.021 (***) -0.011 (***) -0.052 (***) -0.019 (**) (0.005) (0.003) (0.016) (0.008) LogSize* Other Financial -0.034 (***) -0.018 (***) -0.030 -0.068 (***) (0.011) (0.007) (0.021) (0.021) LogSize* Trading -0.007 -0.004 -0.027 -0.007 (0.008) (0.003) (0.025) (0.011) LogSize* Business Equipment -0.013 0.017 0.013 -0.073 (***) (0.021) (0.024) (0.041) (0.006) LogSize* Chemicals -0.020 (**) -0.013 -0.036 (**) -0.031 (0.010) (0.009) (0.015) (0.020) LogSize* Consumer Durables 0.005 -0.026 0.036 0.007 (0.033) (0.035) (0.046) (0.028) LogSize* Consumer Non-Durables -0.025 -0.015 -0.016 -0.032 (0.016) (0.012) (0.034) (0.020) LogSize* Energy -0.002 -0.005 0.071 (**) 0.018 (0.013) (0.008) (0.032) (0.023) LogSize* Health Care -0.015 -0.006 -0.045 -0.018 (***) (0.023) (0.004) (0.029) (0.006) LogSize* Manufacturing -0.050 (***) -0.030 (***) -0.051 (***) -0.052 (***) (0.011) (0.009) (0.019) (0.013) LogSize* Other Non-Financial -0.039 (***) -0.029 (***) -0.071 (***) -0.044 (***) (0.006) (0.007) (0.013) (0.007) LogSize* Shops -0.048 (***) -0.017 (*) -0.064 (***) -0.059 (***) (0.008) (0.009) (0.014) (0.011) LogSize* Telecommunications -0.016 (**) -0.020 (**) -0.046 (***) -0.028 (***) (0.007) (0.010) (0.014) (0.002) LogSize* Utilities 0.009 0.004 -0.009 0.029 (0.009) (0.007) (0.022) (0.033) Maturity 0.001 (***) 0.001 (***) -0.001 (**) 0.001 (***) (0.000) (0.000) (0.001) (0.000) Adj.R2 0.677 0.643 0.608 0.679 N 15,972 8,303 1,831 5,838 This table presents results of regressions of (log) yield spread using bond trading data from TRACE on firm size (Equation 3). Pre-crisis period is 2004Q1-2008Q2, financial crisis period is 2008Q3-2009Q2, post-crisis period is 2009Q3-2013Q2. All specifications control for maturity and include industry interactions for EDF and EDF2 as well as industry and year-quarter fixed effects. Standard errors are clustered at the firm level. ***, **, and * indicate statistical significance at the 1%, 5% and 10% levels, respectively. 42

Table 9: Regression of log bond spread across industries and time period, nonlinear size effect (1) Full (2) Pre-Crisis (3) Crisis (4) Post-Crisis LogSize* Banking -0.010 (*) -0.005 -0.020 -0.020 (*) (0.005) (0.003) (0.019) (0.011) Size95* Banking -0.042 (**) -0.025 (**) -0.130 (*) 0.006 (0.020) (0.011) (0.072) (0.032) LogSize* Other Financial -0.029 (**) -0.007 -0.018 -0.059 (***) (0.015) (0.009) (0.028) (0.020) Size95* Other Financial -0.023 -0.040 -0.066 -0.132 (**) (0.033) (0.028) (0.058) (0.057) LogSize* Trading -0.008 -0.004 -0.029 -0.114 (***) (0.008) (0.003) (0.026) (0.030) Size95* Trading 0.009 0.003 0.013 0.528 (***) (0.006) (0.003) (0.039) (0.146) LogSize* Business Equipment -0.023 0.025 -0.011 -0.077 (***) (0.020) (0.031) (0.041) (0.005) Size95* Business Equipment 0.071 -0.037 0.155 0.579 (**) (0.096) (0.083) (0.165) (0.231) LogSize* Chemicals -0.020 -0.009 -0.058 (**) -0.051 (*) (0.013) (0.009) (0.026) (0.030) Size95* Chemicals 0.002 -0.018 0.090 0.085 (0.045) (0.034) (0.080) (0.084) LogSize* Consumer Durables -0.126 (***) -0.172 (***) -0.257 (***) -0.082 (***) (0.034) (0.017) (0.038) (0.004) Size95* Consumer Durables 0.598 (***) 0.733 (***) 1.121 (***) 0.427 (***) (0.098) (0.080) (0.133) (0.020) LogSize* Consumer Non-Durables 0.001 0.006 -0.001 0.002 (0.014) (0.012) (0.046) (0.029) Size95* Consumer Non-Durables -0.060 (*) -0.057 (*) -0.033 -0.072 (0.033) (0.032) (0.113) (0.057) LogSize* Energy -0.016 -0.012 0.115 (**) -0.006 (0.012) (0.008) (0.055) (0.031) Size95* Energy 0.068 (***) 0.025 (*) -0.059 0.055 (*) (0.015) (0.013) (0.043) (0.033) LogSize* Health Care -0.047 -0.001 -2.372 (***) 0.016 (0.031) (0.009) (0.786) (0.014) Size95* Health Care 0.056 -0.010 2.545 (***) -0.044 (**) (0.035) (0.016) (0.838) (0.019) LogSize* Manufacturing -0.033 (***) 0.005 -0.032 -0.045 (***) (0.011) (0.013) (0.027) (0.016) Size95* Manufacturing -0.044 (*) -0.078 (***) -0.045 -0.018 (0.027) (0.025) (0.063) (0.036) 43

Table 9: Regression of log bond spread across industries and time period, nonlinear size effect (1) Full (2) Pre-Crisis (3) Crisis (4) Post-Crisis LogSize* Other Non-Financial -0.035 (***) -0.029 (***) -0.058 (***) -0.043 (***) (0.005) (0.007) (0.013) (0.006) Size95* Other Non-Financial -0.023 0.003 -0.064 -0.013 (0.020) (0.013) (0.061) (0.030) LogSize* Shops -0.043 (***) -0.018 (*) -0.071 (***) -0.035 (***) (0.009) (0.010) (0.010) (0.009) Size95* Shops -0.019 0.003 0.038 -0.097 (***) (0.024) (0.021) (0.059) (0.027) LogSize* Telecommunications -0.014 (**) -0.019 (**) -0.045 (***) -0.028 (***) (0.007) (0.009) (0.013) (0.002) Size95* Telecommunications -0.057 (**) -0.028 (*) 0.000 0.000 (0.024) (0.016) (0.000) (0.000) LogSize* Utilities 0.013 0.008 -0.009 0.029 (0.009) (0.008) (0.022) (0.033) Size95* Utilities -0.051 (***) -0.041 (**) 0.000 0.000 (0.016) (0.020) (0.000) (0.000) Maturity 0.001 (***) 0.001 (***) -0.001 (**) 0.001 (***) (0.000) (0.000) (0.001) (0.000) Adj.R2 0.684 0.652 0.620 0.690 N 15,972 8,303 1,831 5,838 ThistablepresentsresultsofestimatingEquation4usingbondtradingdatafromTRACEoverfinancialcrisis (2008Q3-2009Q2), pre-crisis (2004Q1-2008Q2) and post-crisis (2009Q3-2013Q2) subsamples. Regression specifications augment Equation 3 with an additional size variable (Size95) to examine nonlinear size effects forthelargestfirmsineachindustry. Allspecificationscontrolformaturityandincludeindustryinteractions for EDF and EDF2 as well as industry and year-quarter fixed effects. Standard errors are clustered at the firm level. ***, **, and * indicate statistical significance at the 1%, 5% and 10% levels, respectively. 44

stneicffieoc yrtsudni-ezis sisirc-erp SDC .1 .giF 3 noitauqE morf ezis gol selbairav ezis eerht rof elpmasbus sisirc-erp SDC eht rof snoitcaretni yrtsudni-ezis eht no stneicffieoc stneserp erugfi sihT rebmun eht serusaem hcihw )neerg ni nwohs( elbairav ezis dezidradnats a dna ,)der ni nwohs( 4 noitauqE morf rotacidni 59eziS eht ,)eulb ni nwohs( gnitcartbus yb dezilamron era stneicffieoc eht ,nosirapmoc fo esae roF .yrtsudni sti fo stessa gol naem eht morf si ezis s’mrfi a snoitaived dradnats fo stneicffieoc stneserp erugfi sihT .stneicffieoc eht fo noitaived dradnats eht yb gnidivid dna elbairav ezis nevig a rof stneicffieoc eht fo eulav naem eht 45

stneicffieoc yrtsudni-ezis sisirc SDC .2 .giF rotacidni 59eziS eht ,)eulb ni nwohs( 3 noitauqE morf ezis gol selbairav ezis eerht rof elpmasbus sisirc SDC eht rof snoitcaretni yrtsudni-ezis eht no si ezis s’mrfi a snoitaived dradnats fo rebmun eht serusaem hcihw )neerg ni nwohs( elbairav ezis dezidradnats a dna ,)der ni nwohs( 4 noitauqE morf rof stneicffieoc eht fo eulav naem eht gnitcartbus yb dezilamron era stneicffieoc eht ,nosirapmoc fo esae roF .yrtsudni sti fo stessa gol naem eht morf rof snoitcaretni yrtsudni-ezis eht no stneicffieoc stneserp erugfi sihT .stneicffieoc eht fo noitaived dradnats eht yb gnidivid dna elbairav ezis nevig a 46

stneicffieoc yrtsudni-ezis sisirc-tsop SDC .3 .giF ,)derninwohs(4noitauqEmorfrotacidni59eziSeht,)eulbninwohs(3noitauqEmorfezisgol-selbairaveziseerhtrofelpmasbussisirc-tsopSDCeht sti fo stessa gol naem eht morf si ezis s’mrfi a snoitaived dradnats fo rebmun eht serusaem hcihw )neerg ni nwohs( elbairav ezis dezidradnats a dna gnidividdnaelbairavezisnevigarofstneicffieocehtfoeulavnaemehtgnitcartbusybdezilamronerastneicffieoceht,nosirapmocfoesaeroF .yrtsudni rof elpmasbus sisirc-erp dnob eht rof snoitcaretni yrtsudni-ezis eht no stneicffieoc stneserp erugfi sihT .stneicffieoc eht fo noitaived dradnats eht yb 47

stneicffieoc yrtsudni-ezis sisirc-erp dnoB .4 .giF elbairav ezis dezidradnats a dna ,)der ni nwohs( 4 noitauqE morf rotacidni 59eziS eht ,)eulb ni nwohs( 3 noitauqE morf ezis gol selbairav ezis eerht ,nosirapmoc fo esae roF .yrtsudni sti fo stessa gol naem eht morf si ezis s’mrfi a snoitaived dradnats fo rebmun eht serusaem hcihw )neerg ni nwohs( fo noitaived dradnats eht yb gnidivid dna elbairav ezis nevig a rof stneicffieoc eht fo eulav naem eht gnitcartbus yb dezilamron era stneicffieoc eht morf ezis gol selbairav ezis eerht rof elpmasbus sisirc dnob eht rof snoitcaretni yrtsudni-ezis eht no stneicffieoc stneserp erugfi sihT .stneicffieoc eht 48

stneicffieoc yrtsudni-ezis sisirc dnoB .5 .giF serusaem hcihw )neerg ni nwohs( elbairav ezis dezidradnats a dna ,)der ni nwohs( 4 noitauqE morf rotacidni 59eziS eht ,)eulb ni nwohs( 3 noitauqE yb dezilamron era stneicffieoc eht ,nosirapmoc fo esae roF .yrtsudni sti fo stessa gol naem eht morf si ezis s’mrfi a snoitaived dradnats fo rebmun eht stneserp erugfi sihT .stneicffieoc eht fo noitaived dradnats eht yb gnidivid dna elbairav ezis nevig a rof stneicffieoc eht fo eulav naem eht gnitcartbus 49

stneicffieoc yrtsudni-ezis sisirc-tsop dnoB .6 .giF eht ,)eulb ni nwohs( 3 noitauqE morf ezis gol selbairav ezis eerht rof elpmasbus sisirc-tsop dnob eht rof snoitcaretni yrtsudni-ezis eht no stneicffieoc snoitaiveddradnatsforebmunehtserusaemhcihw)neergninwohs(elbairavezisdezidradnatsadna,)derninwohs(4noitauqEmorfrotacidni59eziS eht fo eulav naem eht gnitcartbus yb dezilamron era stneicffieoc eht ,nosirapmoc fo esae roF .yrtsudni sti fo stessa gol naem eht morf si ezis s’mrfi a .stneicffieoc eht fo noitaived dradnats eht yb gnidivid dna elbairav ezis nevig a rof stneicffieoc 50

Table 10: Regression of CDS spread, liquidity controls (1) Full (2) Pre-Crisis (3) Crisis (4) Post-Crisis LogSize* Banking -0.024 (**) -0.004 -0.028 -0.025 (0.012) (0.006) (0.050) (0.021) LogSize* Other Financial -0.007 -0.011 (***) 0.022 -0.024 (*) (0.007) (0.003) (0.018) (0.014) LogSize* Trading -0.018 (***) -0.012 (***) -0.059 (***) -0.021 (***) (0.005) (0.004) (0.015) (0.006) LogSize* Business Equipment -0.027 (***) -0.026 (***) -0.003 -0.030 (***) (0.007) (0.007) (0.014) (0.008) LogSize* Chemicals -0.026 (***) -0.025 (***) -0.022 -0.019 (**) (0.008) (0.007) (0.015) (0.008) LogSize* Consumer Durables 0.077 (***) 0.057 (***) 0.175 (*) 0.064 (***) (0.018) (0.018) (0.090) (0.017) LogSize* Consumer Non-Durables -0.027 (*) -0.007 0.003 -0.039 (*) (0.014) (0.009) (0.037) (0.022) LogSize* Energy -0.018 (**) -0.014 (**) -0.011 -0.024 (*) (0.007) (0.006) (0.018) (0.012) LogSize* Health Care -0.022 -0.002 -0.007 -0.038 (*) (0.017) (0.011) (0.030) (0.020) LogSize* Manufacturing -0.011 (*) -0.015 (**) 0.011 -0.019 (***) (0.006) (0.007) (0.019) (0.006) LogSize* Other Non-Financial 0.007 0.009 0.040 (**) -0.003 (0.009) (0.014) (0.016) (0.008) LogSize* Shops -0.035 (***) -0.020 (*) 0.002 -0.023 (0.013) (0.010) (0.016) (0.016) LogSize* Telecommunications -0.007 0.011 -0.013 -0.048 (***) (0.012) (0.017) (0.019) (0.015) LogSize* Utilities 0.007 0.002 0.007 0.001 (0.006) (0.006) (0.016) (0.007) Adj.R2 0.682 0.662 0.695 0.766 N 14,714 7,977 1,900 4,837 This table presents results for estimating Equation 5, which includes controls for liquidity (CDS depth and log of debt issued in past 5 years and CDS depth, interacted across industries) over financial crisis (08Q3- 09Q2), pre-crisis (04Q1-08Q2) and post-crisis (09Q3-13Q2) subsamples. All specifications include industry interactions for EDF and EDF2 as well as industry and year-quarter fixed effects. Standard errors are clustered at the firm level. ***, **, and * indicate statistical significance at the 1%, 5% and 10% levels, respectively. 51

Table 11: Summary statistics on realized loss given default Industry # Obs Mean LGD Mean Size ($M) Banking 4 38 924 Other Financial 10 58 917 Trading 12 63 335 Business Equipment 61 58 793 Chemicals 25 49 1,529 Consumer Durables 49 43 1,384 Consumer Non-Durables 86 47 657 Energy 54 44 541 Health Care 34 54 661 Manufacturing 144 48 592 Other Non-Financial 183 51 820 Shops 157 50 448 Telecommunications 103 56 2,321 Utilities 22 33 2,430 Sample consists of all default events from Moody’s Default and Recovery Database from April 1987 to April 2013withanLGD>0. Lossgivendefault(LGD)isexpressedinpercentagepoints. Firmsizebasedonface value of debt outstanding at the time of default. 52

Table 12: Coefficient magnitude summary Full Sample Pre-Crisis Table 6 Table 8 Table 10 Table 6 Table 8 Table 10 Panel A: Full Sample and Pre-crisis LogSize* Banking -0.2 -0.2 -0.2 -0.0 -0.1 -0.0 LogSize* Other Financial -0.1 -0.3 -0.1 -0.1 -0.2 -0.1 LogSize* Trading -0.2 -0.1 -0.2 -0.1 -0.0 -0.1 LogSize* Business Equipment -0.3 -0.1 -0.3 -0.3 0.2 -0.2 LogSize* Chemicals -0.3 -0.2 -0.2 -0.3 -0.1 -0.2 LogSize* Consumer Durables 0.6 0.0 0.7 0.6 -0.2 0.5 LogSize* Consumer Non-Durables -0.3 -0.2 -0.3 -0.1 -0.1 -0.1 LogSize* Energy -0.2 -0.0 -0.2 -0.2 -0.0 -0.1 LogSize* Health Care -0.2 -0.1 -0.2 -0.0 -0.1 -0.0 LogSize* Manufacturing -0.1 -0.5 -0.1 -0.1 -0.3 -0.1 LogSize* Other Non-Financial 0.0 -0.4 0.1 0.0 -0.3 0.1 LogSize* Shops -0.4 -0.5 -0.3 -0.4 -0.2 -0.2 LogSize* Telecommunications -0.1 -0.2 -0.1 -0.0 -0.2 0.1 LogSize* Utilities 0.0 0.1 0.1 0.0 0.0 0.0 Panel B: Crisis and Post-crisis LogSize* Banking -0.5 -0.5 -0.3 -0.2 -0.2 -0.2 LogSize* Other Financial 0.4 -0.3 0.2 -0.3 -0.6 -0.2 LogSize* Trading -0.3 -0.3 -0.6 -0.2 -0.1 -0.2 LogSize* Business Equipment -0.0 0.1 -0.0 -0.3 -0.7 -0.3 LogSize* Chemicals -0.1 -0.3 -0.2 -0.2 -0.3 -0.2 LogSize* Consumer Durables 1.9 0.3 1.7 0.2 0.1 0.6 LogSize* Consumer Non-Durables 0.0 -0.2 0.0 -0.4 -0.3 -0.4 LogSize* Energy -0.1 0.7 -0.1 -0.3 0.2 -0.2 LogSize* Health Care -0.3 -0.4 -0.1 -0.5 -0.2 -0.4 LogSize* Manufacturing 0.0 -0.5 0.1 -0.2 -0.5 -0.2 LogSize* Other Non-Financial 0.3 -0.7 0.4 -0.1 -0.4 -0.0 LogSize* Shops -0.0 -0.6 0.0 -0.3 -0.6 -0.2 LogSize* Telecommunications -0.0 -0.4 -0.1 -0.3 -0.3 -0.5 LogSize* Utilities 0.2 -0.1 0.1 0.0 0.3 0.0 This table summarizes coefficient magnitudes across Tables 6 (regression of log CDS spread), 8 (regression of log bond spread), 10 (CDS spread regression wth liquidity controls). Results represent estimated percent (not percentage point) change in borrowing costs for a 10% increase in size. Pre-crisis period is 2004Q1- 2008Q2, financial crisis period is 2008Q3-2009Q2, post-crisis period is 2009Q3-2013Q2. 53

Appendix A. Sample details and additional results Table A.1: Variable definitions Spread (CDS) Yield spread (in basis points) between 5-year senior CDS and 5-year treasury bond Spread (Bond) Yield spread (in basis points) between bond yield to maturity and treasury bond with closest maturity LogSize Log of book assets (millions) (Compustat: atq) StdSize Number of standard deviations from the mean of log assets of Compustat firms in the same industry Leverage Market leverage, 100 * (Debt Principal)/(Debt Principal + Market Cap) (Compustat: (100(dlttq + dlcq) / (cshoq * prccq + dlttq + dlcq)) Depth Daily number of contributor prices used to calculate the 5-year CDS spread Raise Sum of new public debt issuance (in Mergent FISD data) over previous 5 years (millions) Size95 Indicator for size above 95th percentile for Compustat firms in industry Rating S&P issue-level credit rating ordered from 28 (=AAA) to 1 (=Default) Maturity Bond issue maturity taken from FISD or SDC EDF 5-year (CDS) or maturity year (Bond) expected default frequency estimated by Moody’s (percentage) LGD 100 * (1-Realized recovery rate), from Moody’s Ultimate Recovery database (percentage) 54

Table A.2: CDS data variable means by period Industry Spread EDF LogSize StdSize Leverage Depth Raise Size95 Rating Panel A: Full Banking 178 1.53 12.1 2.1 63.6 5.9 30.0 0.5 22.3 Other Financial 269 1.58 10.6 1.2 28.9 7.2 2.6 0.2 21.5 Trading 206 1.18 9.2 1.1 47.5 5.1 3.3 0.2 19.5 Business Equipment 167 0.70 9.4 1.8 18.5 5.5 4.1 0.7 19.4 Chemicals 128 0.70 9.0 1.7 20.2 6.4 5.1 0.4 21.1 Consumer Durables 401 2.38 8.7 1.0 37.0 5.7 0.8 0.1 19.4 Consumer Non-Durables 202 1.04 8.8 1.2 27.8 6.5 1.8 0.2 19.7 Energy 152 1.00 9.5 1.4 24.3 6.6 2.5 0.3 18.7 Health Care 164 0.50 8.8 1.7 25.9 5.8 2.4 0.6 17.6 Manufacturing 165 0.91 8.9 1.2 26.2 6.1 1.4 0.2 20.8 Other Non-Financial 347 2.38 9.0 1.5 36.9 6.2 1.9 0.5 18.1 Shops 222 1.14 9.0 1.3 27.1 7.2 2.0 0.4 18.0 Telecommunications 302 2.04 9.5 0.9 46.5 6.0 3.9 0.2 17.1 Utilities 119 0.55 9.7 0.8 46.5 5.8 1.6 0.1 20.1 Panel B: Pre-crisis Banking 71 0.36 11.9 2.1 63.0 8.1 18.2 0.6 22.5 Other Financial 75 0.37 10.6 1.2 23.7 9.6 2.0 0.2 21.7 Trading 94 0.28 9.4 1.2 48.8 6.9 3.8 0.3 19.7 Business Equipment 127 0.47 9.2 1.9 15.3 6.9 5.0 0.8 19.1 Chemicals 113 0.66 8.9 1.7 19.3 8.4 4.9 0.4 20.9 Consumer Durables 242 1.75 8.7 1.0 35.5 8.1 0.6 0.1 19.1 Consumer Non-Durables 120 0.32 8.8 1.2 26.0 8.4 1.4 0.2 19.8 Energy 91 0.28 9.3 1.4 20.1 8.5 2.0 0.3 18.8 Health Care 124 0.29 8.6 1.8 22.0 7.7 2.3 0.6 17.6 Manufacturing 91 0.35 8.8 1.2 23.4 8.1 1.3 0.3 20.7 Other Non-Financial 261 1.43 9.0 1.5 32.3 8.2 2.1 0.5 18.2 Shops 177 0.46 8.9 1.4 23.4 9.3 1.8 0.4 18.0 Telecommunications 283 1.75 9.4 0.9 45.1 7.4 2.5 0.2 17.1 Utilities 74 0.39 9.5 0.9 45.4 7.5 1.8 0.1 19.9 55

Table A.2: CDS data variable means by period Industry Spread EDF LogSize StdSize Leverage Depth Raise Size95 Rating Panel C: Crisis Banking 516 3.78 12.1 2.1 72.4 5.1 49.6 0.6 22.3 Other Financial 549 3.82 10.6 1.1 40.7 6.0 3.2 0.2 21.3 Trading 619 3.18 9.0 1.1 51.4 4.3 3.1 0.2 19.3 Business Equipment 314 1.57 9.3 1.8 25.4 4.7 2.0 0.7 19.2 Chemicals 226 1.36 9.0 1.7 25.9 5.2 5.2 0.4 20.9 Consumer Durables 1418 6.89 8.6 1.0 52.8 4.4 0.7 0.1 19.6 Consumer Non-Durables 371 2.35 8.8 1.1 36.1 5.7 1.7 0.2 19.5 Energy 283 2.36 9.4 1.3 31.9 5.9 2.3 0.3 18.4 Health Care 264 1.39 8.7 1.6 32.1 4.8 2.2 0.5 17.4 Manufacturing 409 2.44 8.9 1.2 35.6 5.5 1.2 0.2 20.7 Other Non-Financial 691 4.57 8.9 1.5 46.7 5.5 1.7 0.4 18.0 Shops 359 2.32 9.1 1.3 35.7 6.4 2.1 0.4 18.0 Telecommunications 449 3.59 9.3 0.9 54.5 4.9 4.0 0.2 16.5 Utilities 215 0.88 9.6 0.8 52.2 5.1 1.2 0.1 20.2 Panel D: Post-crisis Banking 199 2.11 12.2 2.1 62.1 4.1 41.0 0.5 22.1 Other Financial 421 2.38 10.7 1.1 32.0 4.7 3.3 0.1 21.2 Trading 213 1.57 9.2 1.0 45.0 3.6 2.7 0.2 19.3 Business Equipment 173 0.75 9.6 1.7 20.5 3.9 3.6 0.7 19.7 Chemicals 118 0.55 9.3 1.7 19.7 4.3 5.5 0.4 21.5 Consumer Durables 288 1.80 8.7 0.9 34.3 3.5 1.0 0.1 19.8 Consumer Non-Durables 251 1.53 8.9 1.1 27.8 4.5 2.7 0.2 19.6 Energy 181 1.39 9.8 1.3 26.8 4.6 3.3 0.3 18.8 Health Care 180 0.49 9.0 1.6 28.6 4.0 2.6 0.5 17.6 Manufacturing 185 1.14 9.1 1.2 26.8 4.0 1.7 0.2 20.8 Other Non-Financial 347 2.80 9.0 1.5 38.9 4.2 1.8 0.5 18.1 Shops 242 1.67 9.2 1.3 29.5 4.8 2.3 0.4 18.0 Telecommunications 273 1.86 9.8 0.9 45.3 4.6 6.4 0.2 17.3 Utilities 141 0.63 9.8 0.8 46.1 4.1 1.3 0.1 20.3 This table presents CDS sample variable means across full sample (04Q1-13Q2), pre-crisis (04Q1-08Q2), crisis(08Q3-09Q2),andpost-crisis(09Q3-13Q2)subsamples. Variablesinclude5-yearCDSspread,expected default probability (EDF), log assets, standardized log assets (measures size in units of standard deviation from the mean of log of assets for Compustat firms in the same industry), market leverage, CDS Depth (a measure of the amount of trading in comparable CDS contracts), sum of debt raised over past 5 years (Raise, $M), an indicator for 95th percentile of size across within-industry Compustat firms, and credit rating (expressed on a 1-28 scale with 28=AAA). 56

Table A.3: Bond sample variable means by time period Industry Spread EDF LogSize StdSize Leverage Maturity Size95 Rating Panel A: Full Banking 218 1.2 13.1 2.6 70.1 6.6 0.70 22.3 Other Financial 291 1.1 10.5 1.1 29.7 9.8 0.11 19.6 Trading 166 0.8 13.0 2.4 85.5 4.2 0.79 22.6 Business Equipment 167 0.8 10.2 2.1 16.0 8.8 0.85 21.0 Chemicals 142 0.5 10.5 2.3 19.5 9.2 0.82 23.0 Consumer Durables 628 1.4 12.0 2.0 71.5 21.2 0.89 13.5 Consumer Non-Durables 190 0.2 9.7 1.5 24.0 9.8 0.56 20.2 Energy 227 0.6 9.9 1.6 20.7 20.0 0.73 19.5 Health Care 203 0.2 8.8 1.7 14.0 10.3 0.81 20.5 Manufacturing 202 0.8 9.9 1.6 28.4 10.5 0.65 19.8 Other Non-Financial 334 2.3 9.8 1.8 37.9 7.3 0.74 17.7 Shops 190 0.6 10.6 2.0 24.6 9.4 0.72 21.0 Telecommunications 425 10.4 9.1 1.0 44.7 3.6 0.04 15.8 Utilities 253 0.6 9.4 0.7 50.1 7.9 0.05 17.2 Panel B: Pre-crisis Banking 125 0.2 13.0 2.7 66.1 6.2 0.72 23.3 Other Financial 182 0.4 10.5 1.2 25.8 8.9 0.17 20.0 Trading 105 0.3 12.9 2.4 86.2 4.3 0.76 23.0 Business Equipment 175 1.2 9.6 2.0 20.9 7.3 0.75 18.2 Chemicals 102 0.6 10.2 2.2 16.5 8.4 0.78 23.3 Consumer Durables 543 1.2 12.2 2.1 71.1 23.0 0.91 14.3 Consumer Non-Durables 135 0.1 9.4 1.4 24.5 8.4 0.42 20.3 Energy 144 0.3 9.8 1.7 18.0 15.6 0.81 19.4 Health Care 131 0.0 8.8 1.8 8.1 10.0 0.84 20.6 Manufacturing 153 0.4 9.8 1.6 27.4 10.1 0.66 19.6 Other Non-Financial 300 3.1 9.7 1.8 38.8 7.4 0.78 17.2 Shops 149 0.2 10.2 1.9 22.3 8.4 0.72 20.4 Telecommunications 420 12.1 9.2 1.0 46.9 3.5 0.05 15.5 Utilities 159 0.2 9.5 0.8 47.0 5.8 0.07 17.3 Panel C: Crisis Banking 600 4.6 13.2 2.7 83.4 7.0 0.69 21.9 Other Financial 687 2.4 10.5 1.1 38.4 9.9 0.09 19.5 Trading 562 3.5 13.1 2.5 85.4 4.1 0.76 22.1 Business Equipment 444 1.2 10.2 2.2 20.5 8.6 0.88 20.0 Chemicals 329 0.6 10.5 2.3 27.5 9.7 0.83 22.6 Consumer Durables 1347 2.7 12.1 2.0 84.3 22.0 0.91 10.6 Consumer Non-Durables 436 0.6 9.7 1.5 29.3 10.7 0.60 20.1 Energy 280 0.5 10.2 1.6 17.8 14.8 0.63 20.3 Health Care 557 0.4 8.7 1.6 19.6 13.1 0.78 20.6 57

Table A.3: Bond sample variable means by time period Industry Spread EDF LogSize StdSize Leverage Maturity Size95 Rating Manufacturing 415 1.3 10.0 1.6 32.6 11.5 0.70 20.2 Other Non-Financial 666 1.5 9.8 1.7 43.3 7.8 0.55 18.0 Shops 422 1.4 10.7 2.0 28.8 10.5 0.78 21.4 Telecommunications 698 2.1 7.8 0.9 39.4 4.4 0.00 16.1 Utilities 732 1.8 9.4 0.7 65.6 11.4 0.00 15.9 Panel D: Post-crisis Banking 215 1.4 13.3 2.6 70.5 6.8 0.68 21.3 Other Financial 317 1.8 10.4 1.0 33.0 11.1 0.03 19.0 Trading 196 1.1 13.1 2.4 83.7 3.9 0.87 21.7 Business Equipment 97 0.3 11.1 2.3 9.1 10.7 0.99 24.3 Chemicals 142 0.3 10.9 2.3 21.9 10.5 0.90 22.7 Consumer Durables 475 1.2 11.4 1.8 67.0 15.1 0.78 12.6 Consumer Non-Durables 165 0.3 10.0 1.5 22.0 10.7 0.67 20.2 Energy 299 0.9 10.0 1.5 24.1 25.7 0.67 19.3 Health Care 184 0.4 8.9 1.6 20.4 9.8 0.77 20.4 Manufacturing 200 1.0 10.0 1.5 28.2 10.8 0.63 19.9 Other Non-Financial 262 0.8 10.1 1.8 33.5 6.8 0.72 18.6 Shops 177 0.9 11.0 2.1 26.2 10.3 0.71 21.7 Telecommunications 315 1.1 8.1 0.9 30.1 3.9 0.00 17.2 Utilities 295 1.2 9.0 0.4 51.3 13.3 0.00 17.6 This table presents bond sample variable means across full sample (04Q1-13Q2), pre-crisis (04Q1-08Q2), crisis(08Q3-09Q2),andpost-crisis(09Q3-13Q2)subsamples. Variablesinclude5-yearCDSspread,expected default probability (EDF), log assets, standardized log assets (measures size in units of standard deviation fromthemeanoflogofassetsforCompustatfirmsinthesameindustry),marketleverage,maturityinyears, an indicator for 95th percentile of size across within-industry Compustat firms, and credit rating (expressed on a 1-28 scale with 28=AAA). 58

Table A.4: Regressions of log CDS spread, standardized size variable (1) Full (2) Pre-Crisis (3) Crisis (4) Post-Crisis StdSize* Banking -0.047 (**) -0.010 -0.119 -0.052 (*) (0.019) (0.007) (0.076) (0.029) StdSize* Other Financial -0.048 (**) -0.029 (***) 0.119 (**) -0.110 (***) (0.022) (0.008) (0.048) (0.040) StdSize* Trading -0.045 (***) -0.037 (***) -0.096 (**) -0.048 (***) (0.011) (0.008) (0.039) (0.018) StdSize* Business Equipment -0.076 (***) -0.083 (***) -0.003 -0.076 (***) (0.017) (0.017) (0.033) (0.024) StdSize* Chemicals -0.076 (***) -0.079 (***) -0.036 -0.061 (***) (0.019) (0.017) (0.030) (0.018) StdSize* Consumer Durables 0.210 (***) 0.182 (***) 0.636 (***) 0.078 (0.058) (0.048) (0.201) (0.050) StdSize* Consumer Non-Durables -0.085 (**) -0.024 0.008 -0.099 (***) (0.033) (0.026) (0.081) (0.034) StdSize* Energy -0.072 (***) -0.060 (***) -0.038 -0.082 (**) (0.021) (0.016) (0.041) (0.033) StdSize* Health Care -0.057 -0.014 -0.076 -0.135 (**) (0.047) (0.024) (0.078) (0.062) StdSize* Manufacturing -0.033 (**) -0.034 (***) 0.008 -0.047 (***) (0.014) (0.013) (0.040) (0.016) StdSize* Other Non-Financial 0.012 0.010 0.108 (***) -0.019 (0.022) (0.031) (0.038) (0.024) StdSize* Shops -0.102 (***) -0.107 (***) -0.003 -0.074 (**) (0.032) (0.040) (0.040) (0.033) StdSize* Telecommunications -0.027 -0.015 -0.008 -0.088 (**) (0.035) (0.039) (0.040) (0.044) StdSize* Utilities 0.003 0.002 0.025 0.006 (0.010) (0.009) (0.026) (0.013) Adj.R2 0.664 0.646 0.674 0.718 N 17,486 8,227 1,984 7,275 This table presents results of estimating Equation 3 using a standardized size measure (which expresses size in terms of standard deviations from the mean of log assets by industry) across financial crisis (2008Q3- 2009Q2),pre-crisis(2004Q1-2008Q2)andpost-crisis(2009Q3-2013Q2)subsamples. Allspecificationsinclude industry interactions for EDF and EDF2 as well as industry and year-quarter fixed effects. Standard errors are clustered at the firm level. ***, **, and * indicate statistical significance at the 1%, 5% and 10% levels, respectively. 59

Table A.5: Regressions of log bond spread, standardized size variable (1) Full (2) Pre-Crisis (3) Crisis (4) Post-Crisis StdSize* Banking -0.045 (***) -0.021 (***) -0.104 (***) -0.043 (**) (0.011) (0.005) (0.037) (0.019) StdSize* Other Financial -0.108 (***) -0.056 (***) -0.131 (*) -0.204 (***) (0.032) (0.020) (0.067) (0.066) StdSize* Trading -0.023 -0.014 (*) -0.072 -0.020 (0.022) (0.008) (0.070) (0.030) StdSize* Business Equipment -0.048 -0.006 -0.166 (***) -0.183 (***) (0.055) (0.067) (0.052) (0.024) StdSize* Chemicals -0.050 (*) -0.030 -0.096 (**) -0.093 (*) (0.028) (0.026) (0.044) (0.054) StdSize* Consumer Durables 0.008 -0.111 0.112 0.027 (0.107) (0.104) (0.151) (0.097) StdSize* Consumer Non-Durables -0.049 -0.031 -0.014 -0.074 (0.047) (0.034) (0.090) (0.048) StdSize* Energy 0.026 -0.001 0.178 (*) 0.065 (0.049) (0.024) (0.106) (0.071) StdSize* Health Care 0.053 -0.029 (*) -0.022 -0.094 (***) (0.033) (0.016) (0.144) (0.032) StdSize* Manufacturing -0.134 (***) -0.069 (***) -0.151 (***) -0.149 (***) (0.031) (0.022) (0.053) (0.033) StdSize* Other Non-Financial -0.121 (***) -0.086 (***) -0.221 (***) -0.140 (***) (0.019) (0.021) (0.043) (0.024) StdSize* Shops -0.115 (***) -0.043 (***) -0.160 (***) -0.143 (***) (0.019) (0.015) (0.036) (0.026) StdSize* Telecommunications -0.098 (***) -0.126 (***) 0.195 (**) -0.133 (***) (0.027) (0.035) (0.078) (0.047) StdSize* Utilities 0.010 0.004 -0.016 0.034 (0.014) (0.013) (0.036) (0.065) Maturity 0.001 (***) 0.001 (***) -0.001 (**) 0.001 (***) (0.000) (0.000) (0.001) (0.000) Adj.R2 0.688 0.661 0.625 0.682 N 15,466 7,892 1,766 5,808 This table presents results of estimating Equation 3 using a standardized size measure (which expresses size in terms of standard deviations from the mean of log assets by industry) across financial crisis (2008Q3- 2009Q2),pre-crisis(2004Q1-2008Q2)andpost-crisis(2009Q3-2013Q2)subsamples. Allspecificationscontrol formaturityandincludeindustryinteractionsforEDFandEDF2 aswellasindustryandyear-quarterfixed effects. Standard errors are clustered at the firm level. ***, **, and * indicate statistical significance at the 1%, 5% and 10% levels, respectively. 60

Table A.6: Regressions of log CDS spread, matched CDS-TRACE sample (1) Full (2) Pre-Crisis (3) Crisis (4) Post-Crisis LogSize* Banking -0.014 (***) -0.004 -0.027 -0.017 (**) (0.005) (0.003) (0.021) (0.008) LogSize* Other Financial -0.012 -0.009 (***) 0.039 (***) -0.034 (**) (0.008) (0.002) (0.015) (0.017) LogSize* Trading -0.019 (***) -0.013 (***) -0.046 (***) -0.016 (**) (0.005) (0.004) (0.015) (0.007) LogSize* Business Equipment -0.035 (***) -0.041 (***) -0.009 -0.024 (***) (0.006) (0.008) (0.015) (0.006) LogSize* Chemicals -0.029 (***) -0.034 (***) -0.013 -0.019 (***) (0.008) (0.007) (0.012) (0.007) LogSize* Consumer Durables 0.079 (***) 0.068 (***) 0.224 (***) 0.038 (***) (0.010) (0.010) (0.033) (0.012) LogSize* Consumer Non-Durables -0.027 (*) -0.008 0.016 -0.035 (*) (0.014) (0.010) (0.032) (0.018) LogSize* Energy -0.023 (***) -0.027 (***) -0.015 -0.019 (0.009) (0.008) (0.016) (0.014) LogSize* Health Care -0.026 -0.006 -0.027 -0.052 (**) (0.019) (0.010) (0.029) (0.022) LogSize* Manufacturing -0.015 (**) -0.019 (***) -0.002 -0.018 (***) (0.006) (0.006) (0.017) (0.006) LogSize* Other Non-Financial -0.003 -0.005 0.020 -0.006 (0.006) (0.009) (0.013) (0.007) LogSize* Shops -0.033 (***) -0.022 (**) -0.002 -0.027 (0.010) (0.009) (0.019) (0.017) LogSize* Telecommunications -0.010 -0.000 -0.004 -0.021 (0.014) (0.019) (0.019) (0.015) LogSize* Utilities 0.002 0.003 -0.005 0.003 (0.007) (0.008) (0.019) (0.009) Adj.R2 0.671 0.670 0.696 0.722 N 12,812 5,658 1,547 5,607 This table illustrates results analogous to those in Table 6 but using a sample of CDS issues matched with bond data in TRACE. All specifications include industry interactions for EDF and EDF2 as well as industryandyear-quarterfixedeffects. Standarderrorsareclusteredatthefirmlevel. ***,**,and*indicate statistical significance at the 1%, 5% and 10% levels, respectively. 61

Table A.7: Regressions of log CDS spread, leverage and rating controls (1) Full (2) Pre-Crisis (3) Crisis (4) Post-Crisis LogSize* Banking -0.040 (***) -0.017 (***) -0.126 (***) -0.044 (***) (0.004) (0.003) (0.013) (0.009) LogSize* Other Financial -0.042 (***) -0.020 (***) -0.027 -0.074 (***) (0.012) (0.007) (0.020) (0.019) LogSize* Trading -0.026 (***) -0.005 -0.050 (***) -0.024 (**) (0.008) (0.006) (0.018) (0.010) LogSize* Business Equipment -0.025 -0.021 -0.005 -0.032 (**) (0.017) (0.015) (0.038) (0.016) LogSize* Chemicals -0.019 (**) 0.002 0.005 -0.037 (***) (0.009) (0.011) (0.024) (0.013) LogSize* Consumer Durables -0.064 (***) -0.032 (*) -0.033 -0.071 (***) (0.025) (0.018) (0.064) (0.015) LogSize* Consumer Non-Durables -0.007 -0.008 0.048 -0.015 (0.016) (0.013) (0.040) (0.021) LogSize* Energy -0.020 (**) -0.024 (***) -0.013 -0.023 (**) (0.008) (0.008) (0.018) (0.011) LogSize* Health Care -0.013 0.001 0.015 -0.024 (0.019) (0.019) (0.034) (0.019) LogSize* Manufacturing -0.024 (***) -0.016 (***) -0.007 -0.034 (***) (0.009) (0.006) (0.020) (0.012) LogSize* Other Non-Financial -0.022 -0.020 -0.012 -0.026 (0.014) (0.014) (0.029) (0.017) LogSize* Shops -0.030 (***) -0.030 (***) -0.002 -0.036 (**) (0.011) (0.009) (0.030) (0.015) LogSize* Telecommunications 0.009 0.026 0.018 -0.019 (0.017) (0.023) (0.023) (0.015) LogSize* Utilities -0.004 -0.000 -0.002 -0.018 (0.008) (0.008) (0.017) (0.013) Adj.R2 0.560 0.570 0.577 0.573 N 16,006 7,508 1,804 6,694 ThistablepresentsresultsforestimatinganequationbasedonEquation3butincludingcontrolsforleverage and credit rating, interacted across industries, over financial crisis (08Q3-09Q2), pre-crisis (04Q1-08Q2) and post-crisis (09Q3-13Q2) subsamples. All specifications include industry interactions for EDF and EDF2 as well as industry and year-quarter fixed effects. Standard errors are clustered at the firm level. ***, **, and * indicate statistical significance at the 1%, 5% and 10% levels, respectively. 62

Table A.8: Regressions of log bond spread, leverage and rating controls (1) Full (2) Pre-Crisis (3) Crisis (4) Post-Crisis LogSize* Banking -0.021 (***) -0.007 -0.076 (***) -0.029 (**) (0.005) (0.007) (0.014) (0.013) LogSize* Other Financial -0.013 -0.006 -0.022 -0.026 (*) (0.008) (0.007) (0.019) (0.014) LogSize* Trading 0.002 0.004 -0.350 (***) -0.049 (0.013) (0.013) (0.047) (0.032) LogSize* Business Equipment -0.049 (***) -0.033 (**) -0.079 -0.067 (***) (0.017) (0.016) (0.054) (0.013) LogSize* Chemicals -0.014 -0.008 -0.051 (**) -0.024 (0.012) (0.010) (0.020) (0.018) LogSize* Consumer Durables -0.089 (***) -0.169 -0.432 (***) -0.067 (***) (0.029) (0.157) (0.155) (0.017) LogSize* Consumer Non-Durables -0.000 0.006 0.036 -0.017 (***) (0.007) (0.004) (0.024) (0.003) LogSize* Energy 0.011 -0.011 0.035 0.003 (0.008) (0.009) (0.033) (0.032) LogSize* Health Care -0.022 -0.006 -0.158 (***) -0.052 (0.035) (0.009) (0.035) (0.048) LogSize* Manufacturing -0.029 (**) -0.035 (***) -0.027 -0.008 (0.014) (0.013) (0.027) (0.018) LogSize* Other Non-Financial -0.024 -0.011 -0.071 (***) -0.069 (***) (0.016) (0.017) (0.018) (0.022) LogSize* Shops -0.019 (*) -0.001 -0.009 -0.012 (0.010) (0.009) (0.010) (0.029) LogSize* Telecommunications -0.015 -0.007 -0.112 (***) -0.007 (***) (0.009) (0.011) (0.010) (0.002) LogSize* Utilities -0.022 (*) -0.010 (*) 0.026 -0.045 (0.012) (0.005) (0.058) (0.071) Maturity 0.001 (***) 0.002 (***) -0.001 0.001 (***) (0.000) (0.000) (0.001) (0.000) Adj.R2 0.687 0.613 0.700 0.704 N 14,376 7,057 1,679 5,640 This table presents results for estimating an equation based on Equation 3 using bond yield spreads from TRACE but including controls for leverage and credit rating, interacted across industries, over the financial crisis (08Q3-09Q2), pre-crisis (04Q1-08Q2) and post-crisis (09Q3-13Q2) subsamples. All specifications control for maturity and include industry interactions for EDF and EDF2 as well as industry and year-quarter fixed effects. Standard errors are clustered at the firm level. ***, **, and * indicate statistical significance at the 1%, 5% and 10% levels, respectively. 63

Table A.9: Regressions of LGD on log size Variable (1) No FE (2) Year FE (3) Year, Ind FE (4) FE+Size*Indus LogSize -2.747 (***) -2.562 (***) -2.730 (***) (0.713) (0.715) (0.733) LogSize* Banking 5.252 (8.881) LogSize* Other Financial -7.593 (5.518) LogSize* Trading -2.608 (7.668) LogSize* Business Equipment -6.058 (***) (2.286) LogSize* Chemicals -5.416 (3.388) LogSize* Consumer Durables -7.292 (***) (1.870) LogSize* Consumer Non-Durables -0.179 (2.576) LogSize* Energy -4.373 (4.018) LogSize* Health Care -3.448 (2.723) LogSize* Manufacturing -0.984 (2.058) LogSize* Other Non-Financial -5.166 (***) (1.469) LogSize* Shops 0.649 (2.173) LogSize* Telecommunications 0.784 (2.003) LogSize* Utilities -3.554 (3.105) N 944 944 944 944 R2 0.017 0.122 0.149 0.166 This table presents regressions of realized LGD on log size (Columns 1-3) and realized LGD on log sizeindustryinteractions(Column4)usingdefaultandrecoverydatafromMoody’sUltimateRecoveryDatabase withLGD>0. LGDismeasuredinpercentagepoints;LogSizeisthelogofthefacevalueofdebtoutstanding at the time of default. *** indicates statistical significance at the 1% level. 64