Understanding the Recovery Rates on Defaulted Securities1

Viral V. Acharya London Business School Sreedhar T. Bharath University of Michigan

Anand Srinivasan University of Georgia Keywords: Recovery, Default, Credit risk, Distressed securities, Bankruptcy. J.E.L. Classication Code: G33, G34, G12. First Draft: February 2003, This Draft: April 20042
1 Viral

Acharya:

Department of Finance, London Business School and is a Research Aliate of the Centre for Economic Policy Ree-mail:

search (CEPR). Address: London Business School, Regents Park, London - NW1 4SA, UK. Phone: +44(0)20 7262 5050 x 3535.

vacharya@london.edu. Sreedhar T. Bharath: University of Michigan Business School, Department of Finance, D6209, Davidson Hall, Ann Arbor, MI-48103, U.S.A Phone: (734) 763-0485. e-mail: sbharath@umich.edu. Anand Srinivasan: University of Georgia, Department of Banking and Finance, Brooks Hall, Athens, GA 30602-6253, U.S.A. Phone: (706) 542-3638. e-mail: asriniva@uga.edu.

2 We thank Sergei Davydenko, Amy Dittmar, Denis Gromb, Rohit Guha, Julian Franks, Herbert Rijken, Ilya Strebulaev, and Per Stromberg for
useful discussions, the seminar participants at Bank of England, Dynamic Corporate Finance Workshop at Copenhagen Business School, European Finance Association Meetings (EFA) 2003, Federal Reserve Board at Atlanta, Lehman Brothers, London Business School, New York University, University of Michigan Business School, Anderson School of Business UCLA, Wharton School of Business, and the 14th Annual Financial Economics and Accounting Conference (2003), for comments, and Deepak Bhandari, Karthik Balakrishnan, Jugal Parekh, and Anupam Sharma for excellent research assistance. The authors acknowledge the help of Edward Altman, Brooks Brady, and Standard and Poors for the provision of data employed in the paper and its documentation. The authors are grateful to the Institute for Quantitative Investment Research (INQUIRE), UK for its nancial support for the project. Acharya is grateful to the Research and Materials Development (RMD) grant from London Business School. Srinivasan is grateful for nancial support from the University of Georgia Research Foundation research grant and the Terry-Sanford research grant. The views expressed and errors that remain in the paper are our own and should not be attributed to these supporting institutions.

Understanding the Recovery Rates on Defaulted Securities

Abstract We document the empirical determinants of the recovery rates on defaulted securities in the United States over the period 19821999. In addition to seniority and security of the defaulted securities, industry conditions at the time of default are found to be robust and important determinants of the recovery rates. Recovery in a distressed state of the industry (median annual stock return for the industry rms being less than 30%) is lower than the recovery in a healthy state of the industry by 10 to 20 cents on a dollar. A better liquidity position of the peers of the defaulted rm also implies higher recovery at emergence from bankruptcy. Furthermore, recovery is negatively correlated with asset-specicity of the industry when the industry is in distress, but not otherwise. Our ndings are thus consistent with the industry equilibrium hypothesis of Shleifer and Vishny (1992). In contrast to perceived wisdom and recent literature, we nd that recovery is not aected by the aggregate supply of defaulted securities, once the industry distress eect is accounted for.

Introduction

Since the seminal work of Altman (1968) and Merton (1974), the literature on credit risk has burgeoned especially in modeling the likelihood of default of a rm on its debt. Extant credit risk models, such as the industry standard KMV (www.kmv.com), Litterman and Iben (1991), Jarrow and Turnbull (1995), Jarrow, Lando, and Turnbull (1997), Madan and Unal (1998), Schnbucher (1998), Due and Singleton (1999), Das and Sundaram (2000), and o Acharya, Das, and Sundaram (2002), employ varying choices for modeling the likelihood of default. The credit spreads or prices of risky bonds and loans determined in these models depend also on the loss given default or inversely on the recovery rates on the bonds under consideration.1 However, many of the extant models and their calibrations assume that recovery is deterministic.2 Some preliminary evidence, see e.g., Carty and Hamilton (1999), Carty, Gates, and Gupton (2000), Brady (2001), and Altman (2002), suggests, however, that recovery rates on defaulted instruments exhibit substantial variability. This paper studies the empirical determinants of recovery risk - the variability in recovery rates over time and across rms - using the data on observed prices of defaulted securities in the United States over the period 19821999. We measure recovery rates using the prices of defaulted securities at the time of default or bankruptcy and at the time of emergence from default or bankruptcy. We focus our investigation on the contract-specic, rm-specic, industry-specic, and, nally, macro-economic determinants of recovery rates. We document systematically the impact of these factors on recovery rates and study whether they are dierent from the determinants of the likelihood of default for these securities. Among contract-specic characteristics, we nd that seniority and security are important determinants of recovery rates: Bank Loans recover on average 20 cents more on a dollar than Senior Secured and Unsecured instruments, which in turn recover 20 cents more than Subordinated instruments. In terms of security, the dierence in recovery at emergence between instruments backed by Current Assets and Unsecured instruments is about 25 cents on a dollar.3 Recoveries at emergence are found to be aected adversely by the length of time the rm spent in bankruptcy.
For instance, under the specic recovery assumption of Due and Singleton (1999), called the Recovery of Market Value (RMV) assumption, the credit spread to be added to the risk-free rate in order to obtain the discount rate applicable for defaultable securities is precisely equal to the risk-neutral hazard rate of default multiplied by the loss given default. 2 Notable exceptions which model recovery risk are Das and Tufano (1996), Frye (2000a, 2000b), Jokivuolle and Peura (2000), Bakshi, Madan, and Zhang (2001), Jarrow (2001), and Guntay, Madan, and Unal (2003). We discuss some of these models in Section 7. See also the Introduction in Altman et al. (2003) for a survey of the literature on recovery risk. 3 Our data-sets do not provide collateral information matched with instruments for which we have recoveries at default.
1

Among rm-specic factors, a 1% marginal increase in the protability of assets (measured as EBITDA/Sales, in the year preceding default) increases recoveries at default by 0.25 cents on a dollar. Recoveries at default are also negatively aected by greater number of defaulted issues and by greater debt dispersion, consistent with the theory that predicts greater coordination problems among creditors in these cases, and in turn, greater incidence of costly liquidations. Surprisingly, we nd the tangibility of assets (measured as Property, Plant, and Equipment/Assets in the year preceding default) does not aect either recoveries at default or at emergence. The eect of tangibility is captured entirely by the Utility industry dummy: Utility industry rms recover in default 30 to 40 cents on a dollar more than other industries.4 Finally, past stock return of the rm positively aects recovery at default whereas stock return volatility negatively aects recovery at default. However, none of the rm-specic characteristics appear to be signicant determinants of recoveries at emergence. Firm-specic characteristics thus determine some variation in recovery rates but not in a consistent manner across dierent measures of recovery. Our most striking results concern the eect of industry-specic and macroeconomic conditions in the default year. Shleifer and Vishny (1992) develop a theoretical model where nancial distress is more costly to borrowers if they default when their competitors in the same industry are experiencing cash ow problems. To the best of our knowledge, no study has examined directly the implications of this theory for recoveries on defaulted instruments.5 A lower asset value in liquidation should translate into a lower debt value: Either bankruptcy entails liquidation of assets to pay creditors, or the anticipation of low prices for asset sales gives greater bargaining power to equityholders in writing down creditor claims. Therefore, the empirical implications of Shleifer and Vishny (1992) model can be translated into implications for debt recoveries. In particular, we test the following four hypotheses: (1) Poor industry or poor macroeconomic conditions when a company defaults should depress recovery rates; (2) Companies that operate in more concentrated industries should
Also, consistent with the ndings of Altman and Kishore (1996), we do not nd any explanatory power for recoveries in coupon, issue size, and outstanding maturity of the instruments, and leverage ratio measured as Long Term Debt/Assets of the defaulting rms in default year minus one. 5 Using airline industry data, Pulvino (1998) examines data on asset sales in the airline industry. He nds evidence supportive of the Shleifer and Vishny (1992) model: Companies that sell aircrafts when they are nancially constrained, or companies that sell aircrafts when the industry is doing poorly, receive a lower price for these aircraft than companies that sell aircrafts at other times. Asquith, Gertner and Scharfstein (1994) nd in their study of junk bonds during the 1970s and 1980s that the use of asset sales in restructuring of distressed rms is limited by industry conditions such as poor performance and high leverage. Using a sample of 39 highly levered transactions, Andrade and Kaplan (1998) nd supporting evidence as well that poor industry and economic conditions adversely aect company performance or value. These studies however do not study prices of bonds. Their ndings however suggest though industry and economic factors may be important determinants of recovery rates.
4

have lower recoveries due to the lack of an active market of bidders; (3) Poor liquidity position of industry peers when a rm defaults should lower the recovery on its debt; (4) Industries that have more specialized assets (those that have few alternative uses) should have lower recoveries in the event of default. We identify the industry of the defaulted rm using the 3-digit SIC code of the company. First, in the spirit of Gilson, John, and Lang (1990) and Opler and Titman (1994), we dene an industry to be distressed if the median stock return for this industry in the year of default is less than or equal to 30%. We nd that when the defaulting rms industry is in distress, its instruments recover about 1012 cents less on a dollar compared to the case when the industry is healthy, both for recoveries at default as well as emergence. Defaulting companies whose industries also have suered adverse economic shock thus face signicantly lower recoveries. In fact, the magnitude of the eect is about half the relative eect of seniority of the instrument (Bank Loans vs. Senior Debt vs. Subordinated debt). We also consider an alternative denition of industry distress that employs stock returns and sales growth in the two years prior to default. The eect is robust to this denition and for recoveries at default it is in fact twice as large as that under the previous denition of distress. Importantly, this eect of industry distress is always non-linear: Industry return employed as a continuous return variable has no explanatory power for recoveries. Furthermore, this eect is found after controlling for contract-specic and rm-specic characteristics, and also controlling for Median Q of the industry. Median Q controls for the fact that industry distress may capture worsening of industrys future prospects, and in turn, be associated with low economic worth of assets. Second, we calculate a sales-based Herndahl index for the industry of the defaulted rm as a measure of its concentration. We do not nd any evidence supporting the hypothesis that industry concentration lowers the recovery rates. Third, we construct a measure of industry liquidity based on the median Quick ratio, the ratio of Current Assets minus Inventories to Current Liabilities, and another variant based on the median interest coverage ratio (EBITDA/Interest Expense). Controlling for the industry distress condition, we nd that poor industry liquidity negatively aects the recovery rates at emergence. The eect is both statistically and economically signicant. However, poor industry liquidity does not appear to depress the recovery rates at the time of default. Finally, we calculate the specicity of an industry using the mean (or the median) of Specic Assets of all rms in that industry. As in Berger, Ofek, and Swary (1996) and Stromberg (2001), we dene Specic Assets of a rm as the book value of its machinery and equipment divided by the book value of total assets. Then, we correlate industry-level recoveries with industry-level asset-specicities for industry-year pairs when the industry is in distress and when it is not in distress. We nd that the correlation between recoveries at emergence 3

and asset-specicity when industries are not in distress is statistically and economically insignicant. The correlation is however signicantly negative when industries are in distress: 0.63 (0.76) for mean (median) recovery and mean (median) asset-specicity. Industries with highly specic assets (e.g., Transportation, and Energy and Natural Resources) experience a signicant drop in debt recoveries (about 30 cents on a dollar) when they are in distress relative to their no-distress levels. Introducing an interaction term between industry distress dummy and asset-specicity also conrms this nding in the regression framework. Thus, except for the hypothesis concerning industry concentration, the variables motivated by Shleifer and Vishny (1992) explain well the time-series variation in recovery rates. We also test the hypothesis that poor macroeconomic conditions at time of default should depress recoveries. We employ macroeconomic data from Altman, Brady, Resti, and Sironi (2002) who study aggregate recovery rates and nd that aggregate recovery rates on defaulted debt are negatively related to aggregate default rates and to aggregate supply of defaulted bonds.6 We nd that the aggregate default rate in the default year and the aggregate supply of defaulted bonds measured mid-year in the default year (Altman et al. variables BDR and BDA, respectively) signicantly lower recovery rates when employed in the absence of industry conditions. However, once industry conditions are employed, their eect becomes insignicant. The eect on recovery of S&P 500 stock return, GDP growth rate in the default year, and of the Fama and French factors (SMB and HML), is generally insignicant. These ndings are in contrast to those of Altman et al. in that there is no eect of bond market conditions over and above the industry conditions. The ndings are complementary to theirs in that the eect of industry conditions is robust to inclusion of bond market conditions and macroeconomic conditions. In particular, the industry distress eect is present for recoveries at emergence even when 1990, the NBER recession year in our sample, is excluded from the data. The linkage between bond market aggregate variables and aggregate recoveries stressed by Altman et al. as arising due to supply-side eects in segmented bond markets appears to be a manifestation of omitted industry conditions. It is dicult to conclude that illiquidity in the nancial markets for trading defaulted instruments causes lower recoveries: The more likely candidate is in fact the illiquidity in the market for asset sales. More broadly, our results demonstrate that recovery risk exists and its magnitude is economically signicant. Does this imply that a credit-risk model must incorporate factors over and above the ones that determine default risk? We provide preliminary evidence that models incorporating recovery risk would likely also need to model industry conditions as a separate state variable. We include in the regression specication a Zscore based on the credit-scoring models of Altman (1968, 2000), or the credit-score of Zmijewski (1984), or the
Frye (2000a, 2000b) and Hu and Perraudin (2002) also show that aggregate quarterly default rates and recovery rates are negatively correlated.
6

Distance to Default as computed by the KMV adaptation of the Merton (1974) model. We nd that the default-risk proxies show up as signicant, but the eect of contract-specic and industry-specic characteristics remains unaected. The determinants of risk of default and the risk of recovery thus seem positively, but not perfectly, correlated. The remainder of the paper is structured as follows. Section 2 discusses additional related literature. Section 3 discusses the data we employ. Section 4 presents summary descriptive statistics. Section 5 presents the regression analysis. Section 6 compares the determinants of recovery risk with the determinants of default risk. Section 7 presents implications for future credit risk models. Section 8 concludes.

Other Related Literature

In addition to the related papers discussed above, a separate literature examines the recovery rates of dierent classes of creditors in the event of distress. Franks and Torous (1994) examine the recovery rates of dierent classes of creditors in the event of a distressed exchange of securities or a bankruptcy. The recovery rates in their sample are largely based on book values of securities received in a reorganization or bankruptcy. Hotchkiss (1995) documents that rms that emerge from Chapter 11 tend to default again subsequently. Therefore, recovery rates based on book values are likely to overstate true recoveries. Our proposed study uses the market prices of debt after the default event thereby circumventing this problem. Franks and Torous (1994) do report recovery rates for a sub-sample of 10 cases of distressed exchanges and 12 cases of bankruptcy based on market values for all securities received by the given creditor class.7 Our sample of market-value based recoveries is more comprehensive, covering 379 rm defaults and 645 instruments for recoveries at default, and 465 rm defaults and 1,511 instruments for recoveries at emergence. Thorburn (2000) looks at the recovery rates for debt in a set of bankruptcy auctions in Sweden. Institutional and legal dierences between the bankruptcy codes in the U.S. and Sweden may make some of her results and conclusions inapplicable to the U.S. Also, our data set has bankruptcies with a Chapter 11 ling and also cases of distress and cure (where there is a default and a rapid resolution), and pre-packaged bankruptcies or distressed exchanges which involve no Chapter 11 ling. Furthermore, Chapter 11 gives the incumbent management a substantial advantage in the reorganization process in the U.S. As explained in the paper, our data enables us to test hypotheses concerning fall in defaulted debt prices in anticipation of strategic write-downs by equityholders who exploit their bargaining power
The median recovery for this sub-sample was signicantly lower than their overall sample, providing a further justication for using market-based recovery measures.
7

when asset sales are expected to fetch low prices. Finally, Thorburn (2000) does not consider the eect of industry conditions on debt recoveries, a focal point in the analysis of this paper. Note that in the spirit of our ndings, Franks and Torous (1994) do nd in their sample a positive relationship between the past performance of the overall stock market and recoveries, and Thorburn (2000) documents a signicant eect on recoveries of a binary variable signifying if bankruptcy occurred during an economic downturn (year 1991) or not. Perhaps our paper is closest to the work of Izvorski (1997) who examines the recovery ratios for a sample of 153 bonds that defaulted in the United States over the period 1983 1993. Consistent with our results, Izvorski nds seniority and type of industry to be the major cross-sectional determinants of recovery.8 At an industry level, Izvorski nds a positive eect on recoveries from xed to total assets and from industry growth. Somewhat surprisingly, he nds a negative eect of industry concentration on recoveries. In contrast, we nd that once the industry is controlled for, tangibility of assets has no signicant eect on recoveries and neither does industry concentration. The industry growth eect is similar to the eect of median industry Q in our results. The key dierences between our work and Izvorskis arise from the facts that (i) we examine both recoveries at default and at emergence, (ii) our data enables us to study eect of collateral on recoveries, and (iii) crucially, our tests provide support for industry distress and industry liquidity eects. A study of recoveries on bank loans and corporate debt can be found in reports by Moodys Investors Service (Global Credit Research).9 These studies focus on description of average recoveries on loans and bonds across dierent seniorities, securities, and industries, and on the average length of time to default-resolution, but do not study the cross-sectional variation of recoveries across rms and through time. Moodys LossCalcT M model (Gupton and Stein, 2002) explains the variation in recoveries using contract-specic, rm-specic, industry-specic, and macroeconomic factors. Being a proprietary model, its description does not disclose the exact point estimates or their standard errors. More important, in contrast to our industry and macroeconomic variables, the LossCalc model employs the moving average of industry recoveries and an index of prices of bankrupt bonds. Since these are precisely the variables whose variation we attempt to explain, the choice of these variables in lagged forms may be desirable from a numerical-tting perspective but is not very useful from an economic perspective. Finally, we examine in detail the eects of industry distress and industry liquidity on recoveries.
Izvorski also nds the type of restructuring attempted after default to be a signicant determinant, a variable that is unavailable in our data set and somewhat cumbersome to obtain given the exhaustive nature of our data set (over 1500 instruments). 9 See, e.g., Carty and Lieberman (1996), Carty (1998), and Carty and Gupton (2000) for recoveries on defaulted bank loans, Carty and Hamilton (1999) for recoveries on corporate debt in general, and Moodys LossCalc model by Gupton and Stein (2002).
8

Data

The data source is the Credit Pro database (version 4.0) developed by Standard and Poors (S&P). The database consists of two component databases, the S&P database and the Portfolio Management Data (PMD) database. The S&P database provides detailed information on all companies that have defaulted between Jan. 1, 1981 and Dec. 31, 1999. At the issuer level, the database provides company names, industry codes, and wherever available, the CUSIP and SIC codes. At the issue level, the database provides bond names, coupons, seniority rankings, issue sizes in dollars, price at default, and default dates. The S&P database contains information only about bonds and does not include collateral information. The PMD database contains recovery data developed by Portfolio Management Data (a part of S&P). The PMD data includes recovery information in the form of price at default and also price at emergence on more than 1,200 bank loans, high-yield bonds, and other debt instruments, totaling over $100 billion. The information is derived from more than 300 non-nancial, public and private U.S. companies that have defaulted until the end of 1999. The coverage becomes extensive after 1987. In addition, the PMD database also provides information on collateral backing the instruments in default. Collateral for each secured instrument is specically identied and grouped into several categories including all assets, inventory and/or receivables, real estate, equipment, non-current assets, and other. For debtors that have emerged from bankruptcy, emergence dates also are furnished. The source data was obtained by S&P from bankruptcy documents: reorganization and disclosure statements, Securities and Exchange Commission lings, press articles, press releases, and their internal rating studies on the issuer. Although there is some overlap between the two databases, by and large the information in one is not replicated in the other. Out of the 1,511 observations in the PMD database 399 are found in the S&P database. The S&P database has a total of 646 observations. While the sample of prices at default from S&P database is smaller (perhaps due to greater illiquidity of corporate bonds at the time of default), the sample of prices at emergence from PMD database is close to being exhaustive and captures well the set of defaults in the United States over the period 1981 to 1999. Combining these two data sets, we obtained our overall sample of bank loans and corporate bonds. Note that our data does not contain any trade credit or project nance instruments, the recovery rates for which likely behave dierently from the instruments we examine. The PMD database measures recoveries at emergence (henceforth, denoted as P e) using three separate methods: (1) Trading prices of pre-petition instruments at the time of emergence; (2) Earliest available trading prices of the instruments received in a settlement; (3) Value for illiquid settlement instruments at the time of a liquidity event the rst 7

date at which a price can be determined, such as the subsequent acquisition of the company, signicant ratings upgrade, renancing, subsequent bankruptcy, or distressed exchange. The S&P database measures recoveries at default. In the case of price at default (henceforth, denoted as P d), the last trading price at the end of the month in which default took place is recorded in the database. For 399 instruments, we have both the price at default and the price at emergence. Both these measures of recovery are given in nominal terms and should be interpreted as Recovery of Face Value or Recovery of Par. This way of measuring recovery is often used in practice and is justied by the fact that when a rm defaults on any one of its obligations, cross-acceleration clauses typically cause all its other claims to also le for default. Furthermore, the amount payable on defaulted instruments once the rm is in bankruptcy is usually close to par. However, it is to be noted that collateralized instruments continue to accrue post-petition interest (after ling for bankruptcy) and thus the amount payable can exceed par.10 We obtain the rm and industry variables for our analysis by cross-matching the CUSIPs of these rms with the CRSPCOMPUSTAT merged database.11 We also use the macroeconomic variables identied by Altman, Brady, Resti, and Sironi (2002) in our analysis. These were obtained from Edward Altman.

Recovery Rates: Summary

Before discussing the general patterns observed in recovery rates data, we present the adjustments we make to the recovery prices at emergence. Since each defaulted rms bankruptcy procedure and reorganization take a dierent period of time, the time-period between default
Guha (2003) discusses the institutional underpinnings of Recovery of Face Value as the appropriate measure of recovery. In particular, Guha documents with examples (Enron Corp. and WorldCom Inc.) that prices of bonds of a corporation with dierent maturities and coupons but the same seniority dier substantially before bankruptcy; once the bankruptcy is announced however, the prices of these bonds converge to identical or close to identical values, since bond covenants contain a provision that makes the principal amount immediately payable upon default. We have conrmed this nding by also examining prices on defaulted instruments provided by LoanX, a new data service started jointly by a group of the largest investment banks. Our private communication with distress hedge-fund managers also revealed that they often engage in convergence trades between bonds of the same rm with diering maturities in order to bet that the rm will default, an event that would make prices of these bonds identical. 11 The S&P database does not always have the CUSIP of the issuing rm. Therefore, we hand-matched the list of defaulted companies in this data set with the CRSPCOMPUSTAT database. Our matching procedure is conservative in that we assign a match only when we can absolutely conrm the identity of the defaulted company. Several of the defaulted companies were private at the time of default as they had undergone leveraged buy-outs prior to the default event. Therefore, we are unable to obtain accounting or stock market data for these rms around the time of default or even one year prior to default.
10

date and emergence date is not identical for dierent default instances. In order to compare emergence recovery prices for dierent default instances, we adjust them for the time between emergence and default dates. In particular, in addition to raw emergence prices (P e), we construct two other measures, emergence prices discounted at high yield index (P ehyld) and emergence prices discounted at coupon rate (P ecoup). For P ehyld, we used the formula P ehyld = P e Id , Ie (1)

where Id and Ie are the level of a high-yield index at default date and at emergence date, respectively. We employed Lehman Brothers, Salomon Brothers, and Merrill Lynch highyield indices since none of these indices were available for use over the entire sample period.12 For P ecoup, we used the formula P ecoup = P e 1 , (1 + c)(T t) (2)

where c is the coupon rate on the defaulted instrument under consideration, T is the emergence date (in years), and t is the default date (in years). Our results are qualitatively robust across these measures, and in most cases, the economic magnitudes are similar too. Hence, in most of the tables we report the results only with P ehyld as the emergence recovery rate. Note that the default recovery prices P d are all measured within one month of default and such time adjustment is not as important. Time-series behavior: In Table 1, we describe the time-series behavior of recovery prices at default and at emergence. The number of defaults is quite small during the period from 1982 to 1986 (under ten in terms of rm defaults), picks up rapidly reaching its maximum during the recessionary phase of 1987 through 1992, and reduces somewhat in the mid-1990s. The number of defaults for which we have recovery prices at default rises steeply again in 1999 (64 rm defaults). However, most of these rms had not yet emerged from bankruptcy when our data was collected. Hence, they do not appear in the recovery prices at emergence. The mean (median) recovery rates at default are 41.96 (38.00) cents on a dollar with a sample standard deviation of 25.34. The mean (median) recovery rates at emergence for
These indices are total return indices. For example, Merrill Lynch High Yield Master II index is a market value-weighted index comprised of 1,800 domestic and yankee high-yield bonds, including deferred interest bonds and payment-in-kind securities. Issues included in the index have maturities of one year or more and have a credit rating lower than BBB/Baa3, but are not in default. The index is a fully invested index, which includes reinvestment of income.
12

P ehyld and P ecoup are 51.11 (49.09) and 52.27 (49.99), respectively, with sample standard deviations of 36.58 and 36.90. There is a clear and a substantial variation in these recovery rates through time. Figure 1 plots the time-series variation in the number of rm defaults (corresponding to P d series) and median recovery price at default (P d) in each year. There is a strong negative relationship between P d and aggregate default intensity (correlation of 0.53), the relationship being particularly strong for the period starting 1987. For example, the mean and the median P d are lowest in 1990, with respective values of 26.82 and 19.50, with a small standard deviation of 20.90 for the year. This coincides with a period of deep recession in the U.S. The number of rm defaults in 1990 and 1991 were respectively 41 and 52. In another instance of this correlation, the mean and the median P d values are 32.07 and 31.00, respectively, in 1999, a year that again coincides with the sharp increase in aggregate default intensity. There were 64 rm defaults in 1999 as compared to a similar number of defaults over the entire period from 1994 to 1998. This lends justication for examining the time-series variation in recovery rates and its correlation with aggregate default intensity and macroeconomic conditions as a potential dimension of recovery risk.13 Eect of Industry: In Table 2, we present the industry-based summary for recovery prices at default and at emergence. We include only P ehyld since the patterns are similar for P ecoup and P e. Our data divides the defaulting rms into 12 industries using S&Ps classication: Utility, Insurance and Real Estate, Telecommunications, Transportation, Financial Institutions, Healthcare and Chemicals, High Technology and Oce Equipment, Aerospace and Auto and Capital Goods, Forest and Building Production and Homebuilders, Consumer and Service Sector, Leisure Time and Media, and Energy and Natural Resources. Based on default date recovery data, the highest number of rm defaults have been for the Consumer and Service sector, Leisure Time and Media sector, and Aerospace, Auto and Capital Goods industries, the numbers being 97, 62, and 50, respectively. Consistent with the evidence of Altman and Kishore (1996) who examine 696 defaulted bond issues over the period 1978 to 1995, we nd the recovery rates are the highest for the Utility sector. The mean (median) recovery at default is 68.37 (77.00) and at emergence is 74.49 (76.94). These levels are statistically dierent from mean recoveries for other industries (at 5% level using the Schee, 1953, test). However, it should be noted that while the number
While we do not have complete data on recoveries after 1999, the recent evidence on recoveries is a point in case for the negative correlation between aggregate default intensity and recovery levels. In 2002, global defaults hit a record amount of $157.3 billion and simultaneously bank loans achieved their lowest recovery rate of 72% (in terms of value of instruments received at emergence). This recovery is 8% to 10% below the 15year mean for bank loan recoveries of 81.6%. Indeed, unsecured bondholders have recovered even less: 28% in 2002 and 22.1% in 2001 compared to the 15year mean of 46%. See Unsecured Bondholders Hit Hardest in 2002 Amidst Declining Recovery Rates, Standard and Poors (www.risksolutions.standardandpoors.com).
13

10

of instrument defaults is large for the Utility sector (55 and 82 based on P d and P e data, respectively), the number of rm defaults in this sector has been quite low (8 and 9 for P d and P e data, respectively). The mean recoveries are not statistically dierent across the other 11 industries, though the Energy and Natural Resources sector does stand out with mean (median) recoveries at emergence of 60.41 (58.80). This latter result also is consistent with the ndings of Altman and Kishore (1996) who nd that chemical, petroleum, and related products had average recoveries of 63% in their sample. Thus, the Utility sector appears as being dierent from other industries (perhaps partly due to regulatory issues), but the simple industry classication by itself does not have much power in explaining the cross-sectional variation of defaults. Eect of Seniority: In Table 3a, we classify defaulted instruments by seniority. The categories in decreasing seniority are: Bank Loans, Senior Secured, Senior Unsecured, Senior Subordinated, Subordinated, and Junior Subordinated. Though we do not have any bank loans in the recovery data at default (S&P database), we have 358 defaulted loans from 219 defaulting rms in the recovery data at emergence (PMD database). The median recoveries at default decline from 48 cents on a dollar for Senior Secured instruments to around 30 cents on a dollar for Senior Subordinated, Subordinated, and Junior Subordinated instruments. In terms of recoveries at emergence, Bank Loans earn on average 30 cents more on a dollar than the next class of seniority, i.e., Senior Secured instruments. This dierence is striking, especially given the lack of much prior evidence on recovery rates on bank loans and given the somewhat smaller relative variation in recoveries across other seniority classes. In level terms, median recoveries at emergence decline from 91.55 cents on a dollar for Bank Loans, to 26.78 cents for Senior Subordinated instruments, and further down to 6.25 cents for Junior Subordinated instruments. In addition to their highest seniority, it is possible that a part of the higher recovery at emergence on Bank Loans arises from a superior ability of banks and nancial institutions to coordinate a reorganization plan for the rm and from their greater bargaining power in the bankruptcy proceedings compared to the dispersed bondholders. Though banks often provide supra-priority debtor-in-possession (DIP) nancing to rms in bankruptcy, such DIP loans are not included in our data. Comparing the mean recoveries across these dierent seniority categories and for both types of recoveries, we nd that 11 out of 15 pair-wise means are statistically dierent at 5% condence level using Schees test. This underscores the importance of seniority of a defaulting instrument as a determinant of its recovery.14
Note that recoveries at emergence in our sample for public bonds of dierent seniorities are roughly similar to the numbers reported in Altman and Kishore (1996) and in Izvorski (1997). Their sample however does not contain data on recoveries at emergence for Bank Loans. A notable exception to the lack of research
14

11

Eect of Collateral: Finally, we document the behavior of recovery rates as a function of the collateral backing the defaulting instruments. Unfortunately, no collateral data is available for recoveries at default (S&P database). For recoveries at emergence (PMD database), instruments are classied into eight categories depending on type of collateral: Current Assets, Plants and Property and Equipment (PPE), Real Estate, All or Most Assets, Other Assets, Unsecured, Secured, and Unavailable Information. Table 3b documents the behavior of P ehyld, the recoveries at emergence, across these collateral categories. Note that about two-thirds of our sample (1,005 of 1,511 defaulting instruments) has no collateral information. We have veried that most of these instruments are in fact un-collateralized bonds. The Unsecured category corresponds to un-collateralized loans. Among the collateralized instruments, most common are those backed by All or Most Assets (228 of 1,511) and those backed by PPE (83 of 1,511). It is, however, the instruments backed by liquid, Current Assets, that have the highest mean (median) recovery of 94.19 (98.81) cents on a dollar. Instruments backed by All or Most Assets have the second-highest mean (median) recovery of 80.05 (89.16) cents. The other collateral categories (PPE, Real Estate, Other Assets, Unsecured, and Secured) have mean recoveries ranging from 63.0 to 72.0 cents. The category of instruments where no information is available on collateral has the lowest mean (median) recovery of 38.64 (30.91). Similar numbers are reported for the eect of collateral on recoveries of just bank loans by Carty and Lieberman (1998). Although there is a certain amount of cross-sectional variation in recoveries across these categories, only the mean recovery for instruments collateralized with Current Assets and the mean recovery for instruments where no collateral information is available are statistically dierent from the mean recoveries on other collateral categories at 5% statistical signicance level using a Schees test. We conclude that the relevant collateral categories for determining recovery rates are thus suciently captured by just liquid Current Assets and Unsecured (un-collateralized bonds) categories. This descriptive summary of our data suggests that contract-specic characteristics such as seniority and security (collateral), industry of defaulting rm (utility sector or other sector), and macroeconomic condition (aggregate default intensity), are likely to play an important role in explaining variation in recovery rates. Within all categories, there is substantial variation in recoveries around the means. In order to develop a more formal model of factors that determine recovery rates, we proceed to a multi-variate regression analysis where we exploit rm-specic characteristics as well as industry-specic conditions at time of default. Studies such as Altman and Kishore (1996) and Carty and Lieberman
on bank loans is the study by rating agencies on prices of defaulted loans, in particular from Moodys Investors Service (Global Credit Research) cited in Section 2. Our ndings are consistent with this research.

12

(1998) contain summary data on recoveries with reported magnitudes similar to those in our data. These studies, however, do not undertake a comprehensive, statistical analysis of the explanatory power of dierent variables. Crucially, they do not examine the industry conditions of the defaulting rm that we show below to be robust and economically important determinants of recovery rates. In Table 4, we report summary statistics of rm-specic, industry-specic, and macroeconomic variables we employ in our regressions. We visit these summaries in sub-sections corresponding to each set of variables.

Determinants of Recovery Rates

Our primary tests relate the price at default and at emergence to the contract, rm, industry, and macroeconomic characteristics for pooled data that combines the entire time-series and the cross-section of recoveries on defaulted instruments. We assume the price at default of each instrument is an unbiased estimate of its actual recovery. An alternative interpretation is that for investors who sell their instruments once default occurs, this is indeed the relevant measure of recovery. Furthermore, many credit risk models do not explicitly capture the bankruptcy proceeding, reorganization, emergence, etc., in their framework. They simply assume a loss given default, which is one minus the recovery rate. For these models, the price of defaulted instruments right after default is a more appropriate measure of recovery. In all our tests, we use ordinary least squares estimates, and standard errors of these estimates are adjusted for heteroscedasticity using the White (1980) estimator and also adjusted for the existence of rm-level clusters as described in Williams (2000) and Wooldridge (2002). That is, we consider each rms debt instruments as a single cluster. This helps us address the issue that a single bankrupt rm may have multiple defaulted instruments and all these instruments show up in our data as separate observations. The average number of defaulted instruments per rm in our sample is about 4.5 and only eight rms in the whole sample have multiple rm default observations.15 All our regressions include industry dummies using the classication employed by S&P. In the tables, we report only the coecient on the Utility dummy since other industry dummies do not show up as being signicant.
Defaults on instruments of the same rm that are separated by more than one year are counted as being parts of separate rm default observations. The eight rms experiencing such multiple defaults are: Ballys, Caldor, Cherokee, Greyhound, Heileman, New Valley Corp, TWA, and Zale. The rest of the rms have defaults of dierent securities within a 3-month period of the rst default and these are counted as being part of a single rm default observation. The dierence of about 3-month period arises due to the fact that bank loans typically default rst followed by bonds.
15

13

5.1

Contract-specic characteristics

We rst consider the eect of contract-specic characteristics. We estimate the specication Recovery = + 1 Coupon + 2 Log(Issue Size) + 3 Dummy(Bank Loans) + 4 Dummy(Senior Secured) + 5 Dummy(Senior Unsecured) + 6 Dummy(Senior Subordinated) + 7 Dummy(Subordinated) + 8 Time in Default + 9 Maturity Outstanding + [10 Dummy(Current Assets) + 11 Dummy(Unsecured) OR 10 Collateralized Debt] + 12 Private Debt + . (3)

The specication considers seniority of the instrument (dummies for Bank Loans, Senior Secured, Senior Unsecured, Senior Subordinated, Subordinated), and collateral of the instrument (dummies for Current Assets and Unsecured) or the extent of Collateralized Debt of the rm as a fraction of its total debt. Some of these were found to be relevant for recoveries in the summary statistics of Section 4. We also include the fraction of the rms debt that is Private Debt. In addition, we include the coupon rate on the instrument, log of the issue size to which the instrument belongs, and its outstanding maturity. For recoveries at emergence, we also include the time in default (years). A common clause in bond indentures is that the accelerated amount payable to bondholders in bankruptcy equals its remaining promised cash ows discounted at the original issue yield, that is, the yield at which the bonds were issued. If a bond was issued at par, its coupon equals the original issue yield and the accelerated amount would be the par value. However, if a bond was issued at discount or premium, then the coupon on the bond will aect the accelerated amount payable to bondholders in bankruptcy. Though most bonds get issued at par, we include the Coupon on the instrument to allow for such an eect. Note that if the bond was indeed issued at discount or premium, then the discounted value of remaining promised cash ows would also depend on the Maturity Outstanding of the instrument, a variable that we also include. Larger issues may earn higher recoveries as a larger stakeholder may be able to exert greater bargaining power in the bankruptcy proceedings. Hence, we include Log (Issue Size) in our tests. Finally, the Time in Default is included since protracted bankruptcies may result in (or capture) lower debt recoveries. Table 5 reports the point estimates for some variants of this specication with Recovery being proxied by recovery at default, P d, and recovery at emergence, P ehyld. Recall that collateral information is not available for data with P d information and there are no Bank Loans in P d data. Furthermore, some data is lost upon requiring that Coupon, Issue Size, 14

and Maturity Outstanding information be available for defaulting instruments. As the point estimates reveal, seniority is statistically and economically an important contract-specic characteristic of recovery rates at default (Columns 12) and at emergence (Columns 37). The coecients on dummies for seniority, 3 through 7 , decline monotonically as seniority declines. The coecients are positive and statistically signicant, typically at 1% condence level (except for the coecients on Subordinated dummies). The dierences between the coecients for Senior and Subordinated seniorities have also been veried as being statistically signicant. On average, the bank loans recover at emergence 20 cents on a dollar more than senior bonds which in turn earn 20 cents more than senior subordinated bonds.16 The extent of private debt in the capital structure of the rm does not appear to have any eect for either the bank loans or the public bonds (Columns 6 and 7). The coecients on collateral dummies for Current Assets and Unsecured are both statistically signicant at 1% condence level (Column 5). These coecients are positive (about 14 cents on a dollar) and negative (about 11 cents on a dollar), respectively, mirroring the descriptive evidence that instruments that are collateralized with liquid, current assets recover more than other instruments. The extent of collateralized debt of the rm does not appear to aect the recoveries on defaulted instruments (Column 6), either for secured or unsecured instruments (Column 7). The sign on Coupon is always negative, but the eect is not robust in its statistical signicance. This is potentially consistent with higher coupon instruments being more likely to be issued at discounts and thus being discounted at higher yields to obtain the amounts payable in bankruptcy. Another possibility is that coupon is in fact an endogenous variable and higher coupons reect issuance by weaker credits, which, in turn, gives rise to lower recoveries. Log (Issue Size) and Maturity Outstanding are usually insignicant statistically. In fact, the sign on Log (Issue Size) is not robust across specications. The fact that Maturity Outstanding is insignicant is consistent with the provision of cross-default clauses leading to payment amounts that are close to par for all obligations of a rm. Since maturity information is available for a small subset of our data (about one-fourth for recoveries at emergence), we exclude this variable in tests that follow. The Time in Default is always signicant at 1% condence level. An additional year in bankruptcy reduces recovery at emergence by about 5 cents on a dollar. Finally, distressed utility rms recover on average about 2535 cents more at default and at emergence compared to other industries (whose
Another proxy for seniority commonly employed in industry is the extent of debt below or above (in terms of seniority) a specic issue in the capital structure of the rm. In regressions not reported here for sake of brevity, we nd this measure of seniority to be statistically and economically signicant when it is included as the only proxy for seniority: a greater debt above the issue at hand in rms capital structure reduces recoveries on that issue. For sake of parsimony, we do not include in our specications this additional variable that captures seniority of the instruments in rms capital structure in a continuous manner.
16

15

dummy coecients are statistically not dierent from zero). Two other observations are in order based on Table 5. First, the explanatory power of comparable specications (without collateral dummies) for recoveries at emergence is substantially greater than for recoveries at default. The adjusted-R2 s are of the order of 40% for P ehyld, whereas they are close to 30% for P d. Second, for all specications and for all recovery measures (except the fourth column), the intercept term is between 20 to 50 cents on a dollar and is usually signicant. This suggests that seniority and security, even though important in determining recoveries on defaulted instruments, are not sucient in explaining the time-series and the cross-sectional variation in recoveries.

5.2

Firm-specic characteristics

As a rst step toward explaining the residual variation in recovery rates, we examine the role played by characteristics of defaulting rms. Since accounting data is dicult to obtain in the year of default, we follow prior literature by using rm level accounting data one year prior to default. This potentially biases our tests against nding any predictive power from rm-specic characteristics. We consider the specication Recovery = + Contract Characteristics + 1 Prot Margin + 2 Tangibility + 3 Debt Ratio + 4 Log(Assets) + 5 Q Ratio + 6 No. of Issues + 7 Debt Concentration + 8 Firm Return + 9 Firm Volatility + . (4)

In this specication (and in specications that follow), Contract Characteristics employed are exactly as in the specication of equation (3), except that Maturity Outstanding, Collateralized Debt, and Private Debt are not employed. The corresponding vector of coecients is denoted as . The rst ve rm-specic characteristics considered in the specication are: Prot Margin, dened as EBITDA/Sales for the defaulting company; Tangibility, proxied by the ratio of Property, Plant and Equipment (PPE) to Assets; Leverage, measured as Long-Term Debt to Assets ratio - we report results only with book leverage as employing market leverage yielded similar results; Log (Assets), the natural log of total assets; and Q Ratio, the ratio of market value of the rm (estimated as book value of total assets book value of equity + market value of equity) to the book value of the rm (estimated as book value of total assets). From Table 4, the median values of these characteristics for rms in our sample are: 7% protability, 33% tangible assets, 48% leverage ratio, 6.05 for log of asset size which corresponds to $424 million of assets, and a Q ratio of 0.76 (all in the year preceding default). 16

The next four characteristics employed are: No. of Issues, the total number of issues defaulting for the defaulted company; Debt Concentration, the Herndahl index measured using the amount of the debt issues of the defaulted company; Firm Return, the stock return of the defaulted company; and Firm Volatility, measured as the standard deviation of daily stock returns of the rm. The median values of these characteristics are: 3.5 number of defaulted issues, 0.40 Herndahl index of debt concentration among defaulted rms, and a gross (net) stock return of 0.30 (70%), and a daily rm return volatility of 7% (again, all in the year preceding default). There is substantial cross-sectional variation in all rm-specic characteristics as revealed by the standard deviation and quantiles reported in Table 4. The recoveries on defaulted bonds should depend on the expected value at which the rm gets acquired or merged in a reorganization or on the expected value fetched by the assets of the rm in liquidation. The protability of a rms assets should thus positively aect recoveries: The greater the protability, the more a potential buyer would be willing to pay for it (all else being equal). Furthermore, many rms default due to liquidity problems and not economic problems per se, so the protability of assets of defaulted rms prior to default does exhibit substantial cross-sectional variation. We include the rms Q to proxy for the growth prospects of the assets which also should aect recovery rates positively. The tangibility of assets also is expected to enhance recovery rates: Intangible assets may be less easily transferrable to acquiring rms and may fetch little or no value in liquidation. We include the leverage of the rm to capture the possibility that bankruptcy proceedings of high-leverage rms may be more dicult to resolve: Higher leverage may be associated with greater dispersed ownership requiring greater coordination among bargaining parties. In a similar spirit, rms with greater number of issues and more dispersed creditor base, that is, lower debt concentration, may experience greater coordination problems and in turn lower recovery rates for creditors.17 We also consider the size of the rm, its total asset base, in order to allow for potential economies or diseconomies of scale in bankruptcy. On the one hand, a part of bankruptcy costs may be xed in nature giving rise to some scale economies; on the other hand, larger rms may be dicult in terms of merger and acquisition activity giving rise to greater ongoing concern value and lesser value in a reorganization or a liquidation. We also include the rms stock return to proxy for its nancial health: Firms with poorer stock returns prior to default, controlling for rms Q ratios, should be closer to
Shleifer and Vishny (1992) show that the expected recovery rates should aect the ex-ante debt capacity of rms and industries. Bolton and Scharfstein (1996) argue how the number of creditors can be optimally chosen by a rm to trade o strategic defaults by management (equityholders) against the costs of default resulting from liquidity shocks. We do not model this endogeneity aspect of rm leverage. Building a model that simultaneously explains design of leverage and recovery rates of rms is indeed a worthy goal to pursue but beyond the scope of the current paper.
17

17

binding nancial constraints, and thus have lower recovery rates. Finally, we include the rms stock return volatility. The greater the volatility of rms total asset value, the lower would be the anticipated value of creditor claims when bankruptcy is resolved. It should be noted however that stock return volatility is an imperfect proxy for rm volatility. In particular, there is a mechanical increase in equity volatility as the rm approaches default even if rm volatility remains unaected. Table 6a and 6b report the point estimates and standard errors for the specication in equation (4) for recoveries at default and emergence, respectively. The qualitative nature of eects for contract-specic characteristics is similar to the results in Table 5. A few rm-specic characteristics show up as being signicant in aecting recoveries at default (Table 6a). Consistent with our hypothesis, we nd that rms with greater protability of assets have greater P d. All else being the same, the marginal eect of a percentage point increase in EBITDA/Sales margin one year prior to default is between 0.25 to 0.35 cents on a dollar. After controlling for the Utility dummy, tangibility of assets has little eect on recoveries at default. If the Utility dummy is not included, then the eect of tangibility of assets is positive and statistically signicant (results are available from the authors upon request). Firm leverage, size, and Q, however, do not appear to be important determinants of recoveries at default. The magnitudes of coecients on these characteristics are mostly not dierent from zero at 5% condence level. In contrast, debt design measures show up as signicant and with signs that are consistent with starting hypotheses. Instruments of rms with greater number of issues in debt structure and with more dispersed ownership of debt experience lower recoveries suggesting greater bankruptcy and/or liquidation costs arising from coordination problems between creditors. The economic magnitude of these eects is however small: For example, an additional issue depresses recoveries by less than a cent on a dollar. Finally, a greater stock return for the rm in the year prior to default is associated with better recoveries, and a greater stock return volatility with lower recoveries. Noticeably, the explanatory power of the regression for recoveries at default, P d, is signicantly enhanced with the addition of rm-specic variables. The adjusted-R2 for P d specications in Table 5 is about 30% whereas in Table 6a it is close to 50%.18 Somewhat surprisingly, not a single rm-specic characteristic shows up as a signicant determinant of the recoveries at emergence (Table 6b). There is little improvement in the
We employed the cash position of the rm, measured as Cash and Cash Equivalents, in the specication. Overall, it does not show up as a signicant determinant of recovery, neither as Cash itself nor as Cash by Assets. We also employed the following variable: Cash minus the amount of debt in the capital structure that is senior to the defaulted instrument. This variable did not have any explanatory power for recoveries either. These results are excluded for sake of parsimony, but available upon request.
18

18

explanatory power for P ehyld, and the intercept term in most specications of Table 6b is positive and statistically signicant. One plausible explanation is that since these characteristics are measured one year prior to default, these are stale by the time of emergence. It should be noted though that for many rms in our sample the emergence and default year are the same (median Time in Default is about 1.5 years).

5.3

Industry characteristics

Shleifer and Vishnys(1992) model provides the theoretical insight that nancial distress is more costly to borrowers if they default when their competitors are experiencing cash ow problems. Asset sales through cash auctions when most of the industry is in distress would fetch prices that are lower than the economic worth of assets if assets cannot be deployed easily by rms outside the industry. To the best of our knowledge, no study has examined directly the implications of this model on debt recovery rates in the event of default. A lower asset value in liquidation should translate into lower debt recoveries. Thus, the Shleifer and Vishny model gives rise to the following implications for debt recoveries: (1) Poor industry or poor macroeconomic conditions when a company defaults should depress recovery rates; (2) Companies that operate in more concentrated industries should have lower recoveries due to the lack of an active market of bidders; (3) Poor liquidity position of industry peers when a rm defaults should lower the recovery on its debt; (4) Industries that have more specialized assets (those that have few alternative uses) should have lower recoveries in the event of default. The studies of Asquith, Gertner and Scharfstein (1994), Andrade and Kaplan (1998), and Pulvino (1998) test the implication of the Shleifer-Vishny model on asset sales and rm values. However, we test the implications of the model for debt recoveries, something not tested in the current literature. We believe that this approach has some advantages over studies that look only at actual asset sales. In particular, looking at debt recoveries enables us to understand the magnitude of losses expected from distressed asset sales, even when such asset sales do not actually occur. How would this happen? In Chapter 11 in the U.S., debtors have the rst-mover advantage in ling a reorganization plan. For sake of argument, suppose the debtors can make a take-it-or-leave-it oer to creditors (Anderson and Sundaresan, 1996, and Mella-Barral and Perraudin, 1997). If the take-it-or-leave-it oer is rejected by creditors, the rm would be liquidated. Then, the rst-mover advantage enables the debtors to strategically oer to creditors only the value that the creditors would receive if the rms assets were liquidated. The rm may eventually get reorganized, acquired, or merged in the bankruptcy, but the creditors receive simply their expected value from the scenario where the rm is liquidated. Such strategic behavior would manifest itself as 19

violations of absolute priority rule (APR), documented in Lopucki and Whitford (1990), Weiss (1990), and Franks and Torous (1994).19 Finally, our sample of defaults covers more than 600 instrument defaults (for emergence prices), more than 200 instrument defaults (for default prices), and spans a period of about two decades from 1982 to 1999. We believe the comprehensive nature of our data has the potential to shed additional light on the eect of industry conditions on defaulted rms. In order to test the rst three hypotheses, we estimate the following specication: Recovery = + Contract Characteristics + Firm Characteristics + 1 Dummy(Industry Distress) + 2 Median Industry Q + 3 Industry Concentration + 4 Industry Liquidity + 5 Median Industry Leverage + . (5)

Note that is the vector of coecients on rm-specic characteristics. Based on the results in Table 6, we include only Prot Margin and Debt Concentration as the rm-specic characteristics. Requiring stock market data reduces sample size which is detrimental to nding any eect from industry distress (which occurs in only 10% of our data). Hence, we include only a representative specication where Firm Volatility is also included. Tangibility is included to verify that the result of Table 6 that tangibility of assets does not aect recoveries is not due to omitted variable bias. According to the hypotheses, we expect that 1 < 0, 3 < 0, 4 > 0, and 5 < 0. Note that median industry leverage also could be employed as a proxy for diculty in raising new nancing. Firms in industries with high leverage may be closer to being nancially constrained or may nd taking on more leverage costly. Following the literature, the industry of a defaulted rm is identied as the set of rms with the same 3-digit SIC code as the defaulted rm. All industry variables are computed using data from CRSP and COMPUSTAT and are measured contemporaneous to default, that is, in the year of default.20 The defaulted rm is excluded from calculation of industry variables. If the 3-digit SIC code of a defaulted rm does not include at least ve other
The recent example of the bankruptcy of United Airlines illustrates this point well: Some of the USs leading companies, including Ford and Philip Morris, are facing billions of dollars of losses on United Airlines leases... The US airline believes it can slash its costs by renegotiating its $8bn of aircraft leases that are spread among 300 companies, ranging from Walt Disney to Pitney Bowes and DaimlerChrysler. It plans to send revised terms to leaseholders over the next three days... Uniteds advisers argue it is in a strong negotiating position because of the weak market for used aircrafts. (US groups face UAL lease losses, by Robert Clow in New York, Financial Times, December 13, 2002). Indeed, general creditors have even less bargaining power than lessors do, and such eects should magnify their losses. 20 We were not able to classify any industry as distressed if we measured industry conditions at the time of emergence of the defaulted rm.
19

20

rms, then we do not include the observation in the tests. We construct two versions of dummy variables for Industry Distress. First, in the spirit of Gilson, John, and Lang (1990) and Opler and Titman (1994), we dene an industry to be distressed if the median stock return for the industry of the defaulting rm in the year of default is less than or equal to 30%. This dummy is Distress1. We also consider Distress2, an alternative denition of industry distress: In addition to Distress1 being one, Distress2 requires that one year or two year median sales growth for the industry in the year of default or the preceding year (based on data availability) be negative. Distress2 is thus based on stock market performance of the industry as well as on the book measure of industry performance. In our data, these industry distress dummies take on the value of 1 for about 10% of the sample size in terms of defaulting instruments. As Table 4 reveals, 13% of industry-year pairs have distressed industries classied on the basis of Distress1, the number being 3% when classied on the basis of Distress2. We employ only one of Distress1 or Distress2 at a time in a specication. Median Industry Q is the median of the ratio of market value of the rm (estimated as book value of total assets book value of total equity + market value of equity) to the book value of the rm (estimated as book value of total assets). The median is taken over all other rms in the 3-digit SIC code of the defaulted rm. Industry Concentration is proxied by the sales-based Herndahl index of all other rms in the industry. Industry Liquidity is proxied using two measures. First, Industry Liq1 is the median Quick ratio, dened as the ratio of [Current Assets - Inventories] to Current Liabilities, the median being taken over all other rms in the industry. Second, Industry Liq2 is the median Interest Coverage ratio, measured as EBITDA/Interest. Both measures are frequently employed in empirical corporate nance to proxy for industry liquidity conditions.21 Finally, Median Industry Leverage is the median Long-Term Debt to Assets for all rms in the industry. We employ either Industry Liq1 or Industry Liq2 in a given specication. From Table 4, the median values of these variables for industry-year pairs in our sample are: 0.91 for Q ratio, 0.17 for Industry Concentration, 0.99 for Quick ratio (Industry Liq1), 3.08 for interest coverage ratio (Industry Liq2), and 18% industry leverage. The median rm Q (leverage) is substantially smaller (greater) compared to the respective medians at industry-level, as one would expect for defaulting rms. In Table 7a, we report the results from estimation of equation (5) for recoveries at default. Table 7b contains the results for recoveries at emergence. We nd that when the defaulting rms industry is in distress, as dened by Distress1 (based on median stock return for industry), its instruments recover about 1012 cents less on a dollar compared to when the industry is healthy. This statement holds for both recoveries at emergence as well as recoveries at default. Thus, defaulting companies whose industries also have suered adverse
See, for example, Stromberg (2001) for the use of Quick ratio, and Asquith, Gertner and Scharfstein (1994) and Andrade and Kaplan (1998) for the use of Interest Coverage ratio.
21

21

economic shock face signicantly lower recoveries. This is further conrmed when we use Distress2, based also on median sales growth for the industry, as the dummy variable for industry distress. In this case, if the defaulting rms industry is in distress, it recovers between 1620 cents on a dollar less at default compared to the case when the industry is healthy. The magnitude of the eect is as high as the relative eect of seniority of the instrument (Bank Loans vs. Senior Debt vs. Subordinated debt). The eects discussed above are all statistically signicant at 5%, and mostly at 1%. The eect of Distress2 on recoveries at emergence is weaker statistically, though still of the order of 710 cents on a dollar. We believe this provides support for the rst hypothesis that poor industry conditions when a company defaults depress recovery rates on the defaulting companys instruments. An alternative hypothesis is that a very high negative median stock return for the industry may arise also if assets of this industry are not expected to be protable in the future. That is, the median stock return for industry may in fact proxy for expected protability of assets of the defaulting rm, a very high negative return generating lower recoveries simply because defaulting rms assets are fundamentally not worth much. This counterargument would generate a negative coecient on Industry Distress dummies without any role for conditions of peer rms of the defaulting rm. We address this issue along three dimensions: First, we have employed Prot Margin of the defaulting rm one year prior to default. We have also included in the specication Median Industry Q. Our assumption is that median industry Q proxies for future growth prospects of the industry and in turn of the defaulted rms assets. Thus, examining the coecient of distress dummies in the presence of rm-level and industry-level protability measures helps isolate the eect on recoveries due to nancial and economic distress of the peers of the defaulted rm. This assumption is borne out in the estimates: In Table 7a, the coecient on Prot Margin is positive and signicant for P d, and in Table 7b, the coecient of Median Industry Q is positive and at least marginally signicant in most specications. The coecients on distress dummies are negative in both Tables 7a and 7b, and generally statistically signicant, controlling for Prot Margin and Median Industry Q. Second, Distress2 dummy is based not only on stock returns but also on the sales growth for the industry which is a book measure of the industrys condition. This measure is less likely to embed expectations of future protability. The eect of Distress2 on P d is even stronger than that of Distress1, specically about twice as large. Thus, even if one were to attribute the entire eect of Distress1 dummy to expectations about future growth prospects, there is a residual eect of industry distress in the Distress2 dummy. Third, if industry conditions merely capture the expectation of future growth prospects, then the positive linkage between industry conditions and recoveries should be symmetric: When industry is doing well, debt recoveries should be higher; when industry is not doing 22

well, debt recoveries should be lower. In Columns 7 and 8 of Tables 7a and 7b, we drop industry distress dummies and instead include median stock return for the industry as a continuous, un-truncated variable. We nd that controlling for Median Industry Q, the level of median industry return has no explanatory power at all for debt recoveries. That is, the eect of industry return on recoveries is always non-linear and restricted to situations where the industry is in distress. This is more consistent with the Shleifer and Vishny hypothesis than with the alternative hypothesis. To examine this eect of industry distress with a microscope, we identify in Table 8a the industries that experience distress based on Distress1 and the year in which they do so. The table shows the 23 industry-year distress pairs: Insurance and Real Estate sector experienced distress in 1990 and 1994; Transportation in 1984 and 1990; Financial Institutions in 1987, 1990 and 1991; Healthcare and Chemicals in 1987, 1990, 1994, 1995, and 1998; High Technology and Oce Equipment in 1990; Aerospace, Auto and Capital Goods in 1990; Forest, Building Products and Homebuilders in 1990; Consumer and Service Sector in 1990, 1993, 1995 and 1996; Leisure Time and Media in 1990, 1994 and 1995; and Energy and Natural Resources in 1986. In Panel A of Table 8b, we examine non-parametrically the dierence in recoveries (measured in dierent forms) between no industry distress years and industry distress years. The dierence is 19.5 cents on a dollar for P d, 14.6 for P e and P ehyld, and 12.6 for P ecoup, all dierences being statistically signicant with p-values close to zero. The alternative z-statistics for Wilcoxon rank sum test between no industry distress and distress samples also have a p-value close to zero.22 Interestingly, the magnitudes of the dierences based on non-parametric tests are close to the ones implied by the coecients on Distress1 and Distress2 in the parametric regressions of Table 7 where we employ contract-specic, rm-specic, and other industry-specic characteristics. Furthermore, the coecients on contract-specic and rm-specic characteristics in these regressions are of similar magnitudes as in the earlier tables (Tables 5, 6a, and 6b). This evidence implies that the eect of industry distress on recovery rates is orthogonal to that of contract-specic and rm-specic characteristics. In Panel B of Table 8b, we examine the dierence in recoveries between no industry distress years and industry distress years where we exclude 1990, the NBER recession year, in which nine out of the twenty-three industry distress events occur. This is to conrm that our results are not driven by just one year of economy-wide distress in which recoveries were skewed toward zero. This year was also special in that a large number of defaults occurred in the aftermath of the leveraged buy-out (LBO) phenomenon. Except for P d, we nd that the dierence in recoveries between no industry distress years and industry
While we do not report these numbers, we nd that the standard deviation of P d and P ehyld are of the same order between no industry distress and industry distress years.
22

23

distress years (excluding 1990) is of similar magnitude as in Panel A which includes 1990 in industry distress years: 17.1 for P e, 13.1 for P ehyld, and 12.4 for P ecoup, all dierences being statistically signicant with p-values close to zero. This illustrates that it is not per se the existence of an economy-wide recession year which depresses recoveries at emergence. What is crucial is whether the industry of the defaulting rm is itself in distress or not. If an industry is in distress, the recoveries for rms defaulting in this industry are signicantly depressed even when the overall economy is not in distress or recession.23 Continuing our examination of Tables 7a and 7b, we nd little evidence supporting the second hypothesis that industry concentration lowers the recovery rates. The coecient on revenue-based Herndahl index for the defaulting rms industry is always negative for recoveries at emergence as well as at default. The eect is, however, never statistically signicant. This result is in contrast to the ndings of Izvorski (1997) who documents a positive relationship between industry concentration and recoveries for a set of 153 bonds that defaulted in the United States between 19831993. Izvorski considers his nding a puzzle since it is opposite to the theoretical priors based on Shleifer and Vishny (1992). The third hypothesis, that poor liquidity position of industry peers when a rm defaults should lower the recovery on its debt, nds little support in Table 7a for recoveries at default, but nds strong support in Table 7b for recoveries at emergence. We nd that controlling for the industry distress condition and industry concentration, the coecient on Industry Liquidity is mostly insignicant for P d in terms of statistical condence. Furthermore, the sign of the coecient is not robust. The coecient is negative for Industry Liq1 and positive for Industry Liq2. In contrast, both industry liquidity proxies have positive and signicant eect on P ehyld. Using Table 4 and the coecient estimates, we nd the eect of lowering industry illiquidity by one standard deviation is to depress the recovery at emergence by about 5 cents on a dollar. Finally, the coecient on Median Industry Leverage is insignicant for both P d and P ehyld. The liquidity hypothesis is thus supported to some extent, but the eect is not as strong or robust as the eect of industry distress. In terms of explanatory power, note that the incremental explanatory power of industryspecic characteristics (over and above the explanatory power of contract-specic and rmspecic characteristics) is relatively small. Nevertheless, F-tests performed to check that industry-specic eects are jointly signicant have p-values less than 0.05 for all the specications. The smaller incremental power for industry-specic eects is an artifact of the pooling of data in the cross-section as well as in the time-series. The total variability in recoveries that is to be explained consists of cross-sectional variability (i.e., across rms,
Note that the dierence in P d, the price at default, between no industry distress and distress years is close to zero when 1990 is excluded from distress years. It must be observed though that excluding 1990 leaves only 10 instrument observations in the P d sample of industry distress years.
23

24

seniority, and collateral classes) as well as time-series variability. The eect of whether the industry is in distress or not explains well the time-series variability in recovery rates, which is small in magnitude compared to the cross-sectional variability in recoveries. So far we have not included a measure of asset-specicity to test the fourth hypothesis that rms in industries that have more specic assets (with few alternative users outside of the industry) should have lower debt recoveries. In a regression specication, this requires interacting the industry distress variable with rm-level or industry-level asset-specicity. Since rm-level asset-specicity is available for only a small fraction of our sample, we employ industry-level asset-specicity. Also, since industry distress dummy takes the value of one in only 10% of our sample, there is little power to identify the interaction with the sample of recoveries at default (about 250 data points). Hence, we focus on recoveries at emergence (about 750 data points). We calculate the specicity of an industry using the mean (or the median) of Specic Assets of all rms in that industry and over the entire sample period. As in Berger, Ofek, and Swary (1996) and Stromberg (2001), we dene Specic Assets of a rm as the book value of its machinery and equipment divided by the book value of total assets. Panels A and B of Table 8c provide the time-series mean and median of Specic Assets for the twelve industries. The four most asset-specic industries are Transportation (44% median Specic Assets), Telecommunications (30%), Energy and Natural Resources (25%), and Consumer and Service sector (24%). Insurance and Real Estate, and Financial Institutions, have close to zero asset-specicity, whereas all other industries have moderate asset-specicity lying between 14% and 17%. The estimates with the interaction term are reported in Columns 8 and 9 of Table 7b, where the industry dummies are replaced by industry-level asset-specicities. Consistent with the starting hypothesis, the interaction term is negative and statistically signicant. When in distress, creditors in an industry that is more asset-specic than a benchmark industry by 10% experience recoveries that are lower by 16 cents on a dollar compared to the creditors of the benchmark industry. The eect is of similar magnitude whether we employ industrys asset-specicity calculated as mean of rm-level asset-specicity ratios (Column 8) or as median (Column 9). We conrm this result non-parametrically as well. Panel A of Table 8c also reports mean and median recovery at emergence, P e, for industry-year pairs with no distress and with distress. Panel B reports P ehyld. Finally, Panel C correlates industry-level recovery with industry-level asset-specicity for industry-year pairs when industry is in distress in a year and when it is not in distress. Panel C illustrates that there is no correlation between recovery and asset-specicity in industry-year pairs when the industry is not in distress. The correlation coecients have unstable signs and are not statistically signicant. The correla25

tion is however signicantly negative in industry-year pairs when the industry is in distress: 0.63 (0.76) for mean (median) recovery and mean (median) asset-specicity. Industries with highly specic assets (e.g., Transportation, and Energy and Natural Resources) experience a signicant drop in debt recoveries (about 30 cents on a dollar) when they are in distress relative to their no-distress levels.24 It should be noted that for Financial Institutions, and for High Technology, Computers, and Oce Equipments, recoveries are actually higher when these industries are in distress compared to when they are not in distress.25 This is due to the small sample size of defaulted rms in distress years for these industries. The correlation patterns are qualitatively unaected by the exclusion of these two industries (Panel D, Table 8c). In particular, when Financial Institutions and High Technology, Computers, and Oce Equipments, are excluded, the correlation between mean (median) P ehyld and mean (median) asset-specicity is 0.90 (0.69) for industry distress years. The corresponding correlations for no industry distress years are insignicantly dierent from zero. We believe this nding is dicult to reconcile with the alternative hypothesis that debt recoveries are low during periods of industry distress simply because the fundamental worth of assets has gone down: it is not clear why this eect should be sensitive to the assetspecicity of industries. Overall, except for industry concentration, the industry variables motivated by Shleifer and Vishny (1992) explain well the time-series variation in recovery rates. We conclude our pursuit by examining macroeconomic conditions which if poor also may depress recoveries, consistent with the rst hypothesis.

5.4

Macroeconomic and bond market conditions

We examine the bond market variables shown by Altman, Brady, Resti, and Sironi (2000) to be signicant in explaining the time-series of average annual recoveries. These variables are: BDR, the aggregate weighted average default rate of bonds in the high yield market where weights are based on the face value of all high-yield bonds outstanding in the year; and BDA, the total face value amount of defaulted bonds in a year measured at mid-year and in trillions of dollars. From Table 4, we see that both BDA and BDR are highly skewed
Unfortunately, there is no data in our sample period for Utilities and Telecommunications when these industries are in distress. It is striking to note though that 19992002 constituted years of industry distress for Telecommunications sector and they were also characterized by extraordinarily low debt recoveries for defaulted telecom rms. In particular, many rms (e.g., Exodus and PSINet) were unable to sell most of their core telecom assets and their creditors recovered value only from the sale of oce-space and other such non-specic assets. 25 This is consistent with the regression results in Columns 8 and 9 of Table 7b where the eect of industry distress is seen to be positive at low levels of asset-specicity.
24

26

variables; median aggregate default rate is about 2% reaching a maximum value of 10%. Similary, median face value of defaulted bonds in a year is about 4 billion dollars with a maximum of 23.5 billion dollars (in 1999).26 If we interpret high values of BDR and BDA as capturing adverse macroeconomic conditions, then the negative eects of these variables on recovery rates would be consistent with the rst hypothesis discussed in Section 5.3: Poor macroeconomic conditions reduce the ability of potential buyers to pay high prices for these assets. Altman et al. in fact do nd such an eect. They nd that BDR and BDA (and their logarithms) aect average annual recoveries signicantly and negatively. In particular, a 1% increase in BDR, the aggregate default rate, is associated with a reduction of 4% in aggregate recoveries. Similarly, a 10 bln. USD increase in BDA, the supply of defaulted bonds, is associated with a lowering in aggregate recoveries of 4.5%. However, Altman et al. present another hypothesis which is that such a negative eect may capture supply conditions in the defaulted bond market: The set of investors participating in the defaulted bond market is segmented and limited to vulture funds, hedge funds, high-yield desks of banks and nancial institutions, and high net-worth individuals. A greater supply of defaulted bonds for a limited demand could imply that the prices on defaulted bonds must fall in order to clear the markets.27 We attempt to disentangle whether it is the illiquidity in the market for real assets (Shleifer and Vishny hypothesis) or the illiquidity in the market for defaulted securities (Altman et al. hypothesis) that causes recoveries to be low during periods of economywide distress. In addition to BDR and BDA, we capture comprehensively the extent of macroeconomic risk in the year of default by examining the eect of SR, the S&P 500 stock return for the year, GDP, the annual Gross Domestic Product growth rate, and the three Fama and French factors, Market, Size (Small Minus Big), and Book-to-Market (High Minus Low), obtained from the Web-site of Ken French and computed using the procedures described in Fama and French (1993). Using these variables, we estimate the specications
The timeseries variation in BDR and BDA is also quite large. For example, in the Altman et al. (2002) data, the aggregate default rate is 1.6% in 1998 and 9.6% in 2002. Similarly, the aggregate defaulted amount was $7.5bln in 1998 and $63.6bln in 2002, the latter being driven by large number of defaults in telecom, airlines, and steel sectors. 27 This is also the perceived wisdom in some industry literature concerning the depressed prices of defaulted securities in 20012002 period (a NBER recession period): As the huge volume of defaulted debt oods the market, trading prices for distressed debt have become depressed, a response to increased supply meeting a generally shallow, illiquid market. (Ultimate Recovery Remains High for Well-Structured Debt, Dropping for Poorly Structured Debt, Standard & Poors, Risk Solutions, January 2002)
26

27

 Recovery = + Contract Characteristics + Firm Characteristics + Industry Conditions + 1 SR + 2 GDP + 3 BDA + OR 1 BDA + OR 1 Market + 2 SMB + 3 HML + . (6)

In Table 9a, we report the point estimates and the standard errors from estimation of equation (6) for recoveries at default, P d (Columns 12), and recoveries at emergence, P ehyld (Columns 35). For ease of reporting, the coecients on seniority dummies are not reported in this table. As the table reveals, the macroeconomic and bond market conditions are not signicant determinants of recoveries, once industry dummies and industry conditions (Distress1, Median Industry Q, and Industry Liq1) are controlled for. SR, GDP, and the Fama and French factors do not appear to have any incremental explanatory power either. While SR and GDP do not show up as signicant also in the results of Altman et al., the insignicance of bond market conditions, BDR and BDA, is in striking contrast to their results. It is important to recognize that in our regressions, we control for the eect of industry conditions and also control for industry dummies. Since Altman et al. examine annual average recovery rates, such conditioning is not possible. To examine this issue further, in Table 9b we run the same specications as in Table 9a but without any industry variables (industry dummies are still included). Even in the absence of industry variables, stock market return, GDP growth rate, and economy-wide risk factors, have no explanatory power for either P d or P ehyld. In contrast, the eect of aggregate defaulted bond supply BDA is negative and statistically signicant for both P d and P ehyld, and the eect of aggregate default intensity BDR is negative and signicant for P ehyld. The magnitude of these eects in Table 9b is also comparable to those in Altman et al. For example, a 10 bln. USD increase in BDA in Columns 1 and 3 gives rise to a reduction in recoveries of about 7%. However, once industry distress eects are controlled for in Table 9a, these coecients fall by between 20% to 40%, and in particular, are rendered insignicant.28 In contrast, in Table 9a the bond market conditions do not drive out the eect of industry distress, Distress1, on P d and P ehyld. The eect of Distress1 in the presence of macroeconomic and bond market conditions is negative, signicant usually at 5% level, and of the
One possibility we must entertain is the following. Altman et al. (2000) results are for P d, recoveries at default, using the NYU Salomon Center data on the closing bid prices for about 1300 bonds (as close to the default as possible) over the period 19822002. Our sample of P d data is much smaller.
28

28

order of 1013 cents on a dollar as before. The eect of industry liquidity, Industry Liq1, on P ehyld is also unaected in magnitude and signicance. Finally, the intercept terms for P ehyld regressions in Table 9b are positive and statistically signicant, whereas in Table 9a, the intercept terms for P ehyld are not statistically dierent from zero. We conclude that industry conditions are an essential ingredient of a specication that explains well the recoveries at emergence. Also, our results suggest that the linkage between bond market conditions and recoveries stressed by Altman et al. as arising due to supply-side eects in segmented bond markets may be a manifestation of omitted industry conditions. It is dicult to conclude that illiquidity in the nancial markets for trading defaulted instruments causes lower recoveries: The more likely candidate is in fact the illiquidity in the market for asset sales.

5.5

Summary

We conclude from the results of Tables 59 that the following factors play an important role in explaining recovery rates. Seniority and security of defaulted instruments help explain the cross-sections of recoveries at default as well as at emergence. Protability of assets of defaulting rms and concentration of their debt structure also help explain the cross-section of recovery rates at default. Recoveries at default and at emergence are both aected significantly by the industry condition when a rm defaults (distressed or healthy), by the type of industry (utility or not), and by the asset-specicity of the industry when that industry is in distress. Finally, recoveries at emergence are aected adversely by the illiquidity of peer rms in industry of the defaulted rm and by the length of time the rm spends in bankruptcy. These sets of factors do not subsume each other, and each set has incremental power in explaining observed recovery rates in the United States over the period 19821999.

Are Determinants of Recovery Risk and Default Risk Identical?

We examine whether ex-ante measures of likelihood of default of a rm, found to be important by extant empirical literature, aect recovery rates or not. In particular, we examine three predictors of default risk of a rm employed in the literature and in practice. First, we employ the Zscore employed in credit-scoring models by rating agencies. The Zscore we employ is as dened in Altman (1968, 2000) and as modied by Mackie-Mason (1990): Z = (3.3 EBIT + Sales + 1.4 Retained Earnings + 1.2 Working Capital)/Assets.(7) 29

Second, we consider another credit-scoring model from the accounting literature, the Zmijewski Score, as dened in Zmijewski (1984): Zmijewski Score = 4.3 4.5 Net Income/Total Assets + 5.7 Total Debt/Total Assets 0.004 Current Assets/Current Liabilities. (8)

Finally, we also employ the Distance to Default as computed by KMV (www.kmv.com) using stock returns and stock return volatility of a rm, based on the Merton (1974) model. We have employed (but do not report the results for) the Expected Default Frequency (EDF), a variant of the Distance to Default measure. The exact computation of these measures is described in Appendix A. We estimate the specication Recovery = + Contract Characteristics + Firm Characteristics + Industry Conditions + ZScore or Zmijewski Score or Distance to Default + . (9)

Note that since the determinants of default risk are also based on rm-specic characteristics, we only include Debt Concentration among these variables. This lets us capture cleanly whether determinants of likelihood of default are also determinants of recoveries or not. The estimates are reported in Table 10 for recovery at default (Columns 13) and at emergence (Columns 46). The determinants of default risk are in general also signicant as determinants of recoveries: ZScore for both P d and P ehyld, Zmijewski Score for P ehyld, and Distance to Default for P d. The eects of seniority of instruments, time in default, debt concentration, and industry conditions remain overall unaected from the presence of the determinants of ex-ante default risk.

Implications for Credit Risk Models

Our results from Sections 5 and 6 put together show that while determinants of default risk and recovery risk are correlated, they are not perfectly correlated. Seniority and collateral, time in default, concentration of debt structure, industry distress, industry liquidity, and asset-specicity are factors that seem to aect recoveries over and above factors that aect default risk. How do these factors aect inputs of recovery rates in existing credit risk models? Seniority and collateral, and also concentration of debt structure, could be captured by allowing a constant recovery rate but one that varies depending on the rm and the 30

instrument (assuming debt structure does not change dramatically during the life of the instrument). It is not fully clear that uncertainty about time in default is a systematic risk and thus may also be reasonably captured in an average recovery rate. However, the state of the industry of the defaulted rm, distressed or healthy, is certainly a systematic risk factor. It constitutes a dimension of recovery risk that may in fact carry a risk-premium to it given its systematic nature. Our results thus underscore the need for modeling recovery risk as stemming from rm-specic factors as well as systematic, industry-specic factors. To the best of our knowledge, such an integrated credit risk model does not yet exist either in the structural class of credit risk models or in the reduced form variety. In a recent contribution to the limited literature that considers modeling recovery risk and associated risk premium, Guntay, Madan and Unal (2003) propose an approach to infer the risk neutral density of recovery rates implied by prices of debt securities of a rm of diering seniority. They demonstrate that interest rates, rm tangible assets, and the level of senior debt appear to be signicant determinants of the price of recovery. Their model however assumes that default risk and recovery risk are independent, in contrast to our nding that these risks are positively correlated but not perfectly so. In this regard, Das and Tufano (1996)s assumption where interest rate risk aects credit risk as well recovery risk is more consistent with our ndings. Their model however limits recovery risk to being determined completely by interest rates.29 The models of Frye (2000a, 2000b) capture in a reduced form the correlation of default risk and recovery risk as potentially arising from the risk of recessions. Jarrow (2001) also allows likelihood of default and recovery to be correlated and based on the state of the economy. He then uses equity prices to identify separately these two components of credit risk. As our tests reveal, it is the risk of an industry recession rather than an economy-wide recession which is the primary driver of recovery risk. Building a next generation of credit risk models that embed industry-specic factors thus appears to be a fruitful goal to pursue, and our empirical work should provide guidance concerning the additional factors to introduce in these models. Another possibility is to analyze in general equilibrium or asset-pricing frameworks the risk-premia arising from the industry eect, that is, from the risk of low recovery or the risk of asset re-sales when rms receive common shocks. Such an exercise would be valuable in understanding the implications of industry-driven recovery risk for prices of credit risky instruments.
Altman et al. (2000) nd that 1year and 10year treasury rates do not help explain aggregate annual recovery rates.
29

31

Conclusions

We have provided a comprehensive empirical analysis of recovery rates on defaulted loans and bonds in the United States over the period 19821999. Our main nding is that whether the industry of the defaulted rm is in distress or not is a robust and economically important determinant of recovery rates. We hope our research will serve as the empirical benchmark for recoveries on dierent kinds of debt and in dierent conditions. For example, studies that examine the determinants of credit spread changes, such as Collin-Dufresne, Goldstein, and Martin (2001), have not accounted for the rich cross-sectional and time-series variation in recoveries found in our study. A more complete analysis of credit spread changes is perhaps called for, especially one that employs some of the industry factors identied by us as important determinants of recovery rate changes. Finally, our results on industry-level asset-specicity and recoveries could also serve as useful benchmarks for future research attempting to link ex-post recovery outcomes to ex-ante capital structure of rms, and to understand the variation in leverage across industries and over the business cycle, along the lines suggested by Shleifer and Vishny (1992).

32

References
Acharya, V. V., Das, S. R. and R. Sundaram, 2002, Pricing Credit Derivatives with Rating Transitions, Financial Analysts Journal, v58(3), 2844. Altman, Edward I., 1968, Financial Ratios, Discriminant Analysis and the Prediction of Corporate Bankruptcy, Journal of Finance, Sep, 589609. Altman, Edward I., 2000, Predicting Financial Distress of Companies: Revisiting the ZScore and Zeta Models, Working Paper, Stern School of Business, New York University. Altman, Edward, 2002, Altman report on Defaulted Bonds and Bank Loans, Salomon Smith Barney United States Corporate Bond Research High Yield. Altman, Edward, Andrea Resti and Andrea Sironi, 2002, Analyzing and Explaining Default Recovery Rates, Working paper, New York University. Altman, Edward, Brooks Brady, Andrea Resti and Andrea Sironi, 2002, The Link between Default and Recovery rates: Implications for Credit Risk Models and Procyclicality, Working paper, New York University. Altman, Edward, Brooks Brady, Andrea Resti and Andrea Sironi, 2003, The link between Default and Recovery Rates: Theory, Empirical Evidence and Implications, Journal of Business, forthcoming. Altman, Edward I., and Vellore M. Kishore, 1996, Almost Everything You Wanted to Know about Recoveries on Defaulted Bonds, Financial Analysts Journal, Nov/Dec, 5764. Anderson, R. and S. Sundaresan, 1996, Design and Valuation of Debt Contracts, Review of Financial Studies 9(1), 3768. Andrade, Gregor and Steven N. Kaplan, 1998, How costly is Financial (Not Economic) Distress? Evidence from Highly Levered Transactions that Became Distressed, Journal of Finance, Vol. 53, 1443-1491. Asquith, P., Gertner, R. and D. Scharfstein, 1994, Anatomy of Financial Distress: An Examination of Junk-Bond Issuers, Quarterly Journal of Economics, August, 625658. Bakshi, G., Madan, D. and F. Zhang, 2001, Understanding the Role of Recovery in Default Risk Models: Empirical Comparisons and Implied Recovery Rates, Finance and Economics Discussion Series, 200137, Federal Reserve Board of Governors, Washington D.C.

33

Berger, P., E. Ofek, and I. Swary, 1996, Investor Valuation of the Abandonment Option, Journal of Financial Economics, Vol. 42, 257287. Bolton, P., and D. Scharfstein, 1996, Optimal Debt Structure and the Number of Creditors, Journal of Political Economy, 104(1), 125. Brady, B., 2001, Default Rate and Recovery Rate Correlation: Evidence and Implications for Modeling Corporate Bond Loss Rates and Loan Loss Rates, Working Paper, Stern School of Business, New York University. Carty, L. V., and D. Lieberman, 1996, Defaulted Bank Loan Recoveries, Moodys Risk Management Services. Carty, L. V., 1998, Bankrupt Bank Loan Recoveries, Moodys Risk Management Services. Carty, L. V., and D. Lieberman, 1999, Debt Recoveries for Corporate Bankruptcies, Moodys Risk Management Services. Carty, L. V., and D. Gates, 2000, Bank Loan Loss Given Default, Moodys Risk Management Services. Collin-Dufresne, Pierre, Golstein, Robert S. and J. Spencer Martin, 2001, The Determinants of Credit Spread Changes, Journal of Finance, 56(6), 21772207. Das, S. R. and R. Sundaram, 2000, A Discrete-Time Approach to Arbitrage-Free Pricing of Credit Derivatives, Management Science, v46(1), 46-62. Das, S. R., and P. Tufano, 1996, Pricing Credit Sensitive Debt when Interest Rates, Credit Ratings and Credit Spreads are Stochastic, Journal of Financial Engineering, v5, 161198. Due, Darrell J. and Kenneth Singleton, 1999, Modeling Term Structures of Defaultable Bonds, Review of Financial Studies, Vol. 12, pp. 687-720. Fama, E. and K. French, 1993, Common Risk Factors in the Returns on Stocks and Bonds, Journal of Financial Economics, 33, 356. Franks, Julian R. and Walter N. Torous, 1994, A comparison of nancial recontracting in distressed exchanges and Chapter 11 reorganizations, Journal of Financial Economics, Vol. 35, 349-370. Frye, J., 2000a, Collateral Damage, Risk, April issue, 9194. Frye, J., 2000b, Depressing Recoveries, Risk, November issue. 34

Gilson, Stuart, Kose John and Larry Lang, 1990, Troubled debt restructurings: An empirical investigation of private reorganization of rms in default, Journal of Financial Economics 27, 315-354. Guha, R., 2003, Recovery of Face Value at Default: Empirical Evidence and Implications for Credit Risk Pricing, Working Paper, London Business School. Guntay, Levent, Madan, Dilip and Haluk Unal, 2003, Pricing the Risk of Recovery in Default with APR Violation, Journal of Banking and Finance, Forthcoming. Gupton, G. and R. Stein, 2002, LossCalc: Moodys Model for Predicting Loss Given Default (LGD), Moodys Risk Management Services. Hotchkiss, Edith, 1995, Post-bankruptcy performance and management turnover, Journal of Finance, Vol. 50, 3-22. Hu, Yen-Ting and William Perraudin, 2002, The Dependence of Recovery Rates and Defaults, Working Paper, Birkbeck College. Izvorski, I., 1997, Recovery Ratios and Survival Times for Corporate Bonds, Working Paper, International Monetary Fund, Washington, D.C. Jarrow, R., 2001, Default Parameter Estimation Using Market Prices, Financial Analysts Journal, 57(5), 7592. Jarrow, R., D. Lando, and S. Turnbull, 1997, A Markov Model for the Term Structure of Credit Spreads, Review of Financial Studies, v10, 481-523. Jarrow, R., and S. Turnbull, 1995, Pricing Options on Financial Securities Subject to Default Risk, Journal of Finance, v50, 53-86. Jokivuolle, E. and S. Peura, 2000, A Model for Estimating Recovery Rates and Collateral Haircuts for Bank Loans, Working Paper, Bank of Finland. Litterman, Robert and Thomas Iben, 1991, Corporate bond valuation and the term structure of credit spreads, Journal of Portfolio Management, Volume 17, Number 3, 52-64. Lopucki, L. and W. Whitford, 1990, Bargaining over Equitys Share in the Bankruptcy Reorganization of Large, Publicly Held Companies, University of Pennsylvania Law Review, 139, 125152. MacKie-Mason, Jerey K., 1990, Do Taxes Aect Corporate Financing Decisions?, Journal of Finance, 45(5), 14711493. 35

Madan, D., and H. Unal, 1998, Pricing the Risks of Default, Review of Derivatives Research 2, 12160. Mella-Barral, P. and W. Perraudin, 1997, Strategic Debt Service, Journal of Finance 52, 531566. Merton, Robert C., 1974, On the Pricing of Corporate Debt: The Risk Structure of Interest Rates, Journal of Finance, Vol. 29, pp. 449-470. Opler, Tim and Sheridan Titman, 1994, Financial Distress and Corporate Performance, Journal of Finance Vol. 49, 1015-1040. Pulvino, Todd C., 1998, Do Asset Fire Sales exist: An Empirical Investigation of Commercial Aircraft Sale Transactions, Journal of Finance, Vol. 53, 939-978. Schee, H., 1953, A Method for Judging All Contrasts in the Analysis of Variance, Biometrika, 40, 87104. Schnbucher, P. J., 1998, Term-Structure Modeling of Defaultable Bonds, Review of Derivao tives Research 2, 16192. Shleifer, Andrei and Robert Vishny, 1992, Liquidation values and debt capacity: A market equilibrium approach, Journal of Finance, Vol. 47, 1343-1366. Stromberg, P., 2001, Conicts of Interest and Market Liquidity in Bankruptcy Auctions: Theory and Tests, The Journal of Finance, 55, 26412691. Thorburn, Karin S., 2000, Bankruptcy auctions: costs, debt recovery and rm survival, Journal of Financial Economics, Vol. 58, 337-358. Weiss, L. (1990) Bankruptcy Resolution: Direct Costs and Violation of Priority of Claims, Journal of Financial Economics 27, 419444. White, Halbert, 1980, A Heteroscedasticity-Consistent Covariance Matrix and a Direct Test for Heteroscedasticity, Econometrica, 48, 817838. Williams, R. L., 2000, A note on robust variance estimation for cluster-correlated data, Biometrics, 56, 645-646. Wooldridge, J. M., 2002, Econometric Analysis of Cross Section and Panel Data, The MIT Press, Cambridge, Massachusetts. Zmijewski, Mark E., 1984, Methodological Issues Related to the Estimation of Financial Distress Prediction Models, Journal of Accounting Research, Supplement, 5982.

36

Distance to Default in the KMV-Merton Model

Symbolically, the Merton (1974) model stipulates that the equity value of a rm satises E = V N (d1 ) erT F N (d2 ), (A.1)

where V is the value of rms assets, E is the market value of rms equity, F is the face value of the rms debt (assumed to be zero-coupon) maturing at date T , r is the instantaneous risk-free rate continuously compounded, N () is the cumulative standard normal distribution function, and, d1 and d2 are given by d1 =
2 ln(V /F ) + (r + 0.5V )T , d2 = d1 V T . V T

(A.2)

The KMV-Merton model makes use of two important equations. The rst is the BlackScholes-Merton equation (A.1), expressing the value of a rms equity as a function of the value of the rm. The second relates the volatility of the rm to the volatility of its equity. Under Mertons assumptions the value of equity is a function of the value of the rm and time, so it follows directly from Itos lemma that E = (V /E) E V . V (A.3)

E In the Black-Scholes-Merton model, it can be shown that V = N (d1 ), so that under the Merton models assumptions, the volatilities of the rm and its equity are related by

E = (V /E)N (d1 )V .

(A.4)

The KMV-Merton model uses the two nonlinear equations, (A.1) and (A.4), to translate the value and volatility of a rms equity into an implied probability of default. The value of a rms equity, E, is easy to observe in the marketplace by multiplying shares outstanding by the rms current stock price. The estimate of E is obtained from either the stock returns data or the implied volatility of the options written on the stock. We can then solve (A.1) and (A.4) for V and V . Using these, the distance to default and the expected default frequency (EDF) are computed as
2 ln(V /F ) + (r 0.5V )T , V T 2 ln(V /F ) + (r 0.5V )T EDF = N V T

DD =

(A.5) . (A.6)

37

Table 1: Time-series Pattern of Recovery Prices at Default (Pd) and at Emergence (Pehyld, Pecoup, Pe). This table documents the time series pattern of recovery prices in terms of cents per dollar. Pd is the price observed at default. Pe is the price observed at emergence. Pehyld is the price observed at emergence discounted by the high yield index for the period between default and emergence. Pecoup is the price observed at emergence discounted by the coupon rate of the instrument in default for the period between default and emergence. Note that one rm could have defaulted in multiple years. There was only one default in 1981. Table 2: Recovery Prices at Default and at Emergence based on Industry. This table documents the patterns of recovery prices in terms of cents per dollar across various industries. Pd is the price observed at default. Pe is the price observed at emergence. Pehyld is the price observed at emergence discounted by the high yield index for the period between default and emergence. Pecoup is the price observed at emergence discounted by the coupon rate of the instrument in default for the period between default and emergence. Note that one rm could be classied in multiple industries based on the division that defaulted. means signicantly dierent from other group means at 5% level using a Schees test. Table 3a: Recovery Prices at Default and at Emergence based on Seniority. Pd is the price observed at default. Pe is the price observed at emergence. Pehyld is the price observed at emergence discounted by the high yield index for the period between default and emergence. Pecoup is the price observed at emergence discounted by the coupon rate of the instrument in default for the period between default and emergence. All recovery prices are measured in cents per dollar. Note that one rm could be classied in multiple seniorities based on instruments that defaulted. 11 of the 15 pairwise means test for dierence is signicant at 5% level or lower using a Schees test. Table 3b: Recovery Prices at Default and Emergence based on Collateral. Pd is the price observed at default. Pe is the price observed at emergence. Pehyld is the price observed at emergence discounted by the high yield index for the period between default and emergence. Pecoup is the price observed at emergence discounted by the coupon rate of the instrument in default for the period between default and emergence. All recovery prices are measured in cents per dollar. Note that one rm could be classied in multiple collateral classes based on instruments that defaulted. means dierent from other group means is signicant at 5% level using a Schees test. Table 4: Summary Statistics of Firm, Industry and Macro Characteristics. Note that the number of data observations are dierent for each variable due to data availability. All rm-specic variables are measured as of the last scal year before the default and data is obtained from COMPUSTAT. Prot Margin is the ratio of EBITDA to Sales. Tangibility is the ratio of Property Plant and Equipment to Total Assets. Debt Ratio is the ratio of Long-Term Debt to Total Assets. Log (Assets) is the natural logarithm of the total assets. Q ratio is the ratio of Market Value of the rm (estimated as Book Value of total assets - book value of equity + market value of equity) to the book value of the rm (estimated as book value of total assets). No. of issues is the total number of debt issues of the rm that is currently under default. Debt concentration is the Herndahl index measure by amount of the debt issues of the rm that are in default. Firm return is the stock return of the rm that has defaulted in the year before default. Firm volatility is the standard deviation of daily stock returns of the rm in the year before default. Distress1 is a dummy variable that takes a value 1 if the median stock return of all the rms in the 3-digit SIC code of the defaulted rm is less than 30% and 0 otherwise. Distress2 is a dummy variable that takes on a value 1 if distress1 is 1 and if the median sales growth of all the rms in the 3-digit SIC code of the defaulted rm is negative in any of the 2 years before the default date. Med.Ind.Q is the median, of the ratio of Market value of the rm (estimated as Book Value of total assets book value of equity + market value of equity) to the book value of the rm (estimated as book value of total assets), of all the rms in the 3 digit SIC code of the defaulted rm. Ind. Liq1 is the median Quick ratio (ratio of (Current AssetsInventories) to Current Liabilities), Ind. Liq2 is the median Interest coverage ratio (EBITDA/Interest), Med. Ind. leverage is the median Long term debt to total assets, of all the rms in the 3 digit SIC code of the defaulted rm. Herndahl Index is the industry concentration measure based on sales. All Industry variables are measured in the year of default. SR, GDP, BDA, BDR are the macro variables used by Altman et.al (2002). SR is the annual return on the SandP 500 stock index. GDP is the annual GDP growth rate in percentage. BDA is the total amount of high yield bonds defaulted amount for a particular year (measured at mid-year in billions of dollars) and represents the potential supply of defaulted securities. BDR is the weighted average default rate on bonds in percentage in the high yield bond market. Weights are based on the face value of all high yield bonds outstanding each year.

Table 5: OLS Estimates of Regression of Recovery Prices at Default and at Emergence on Contract Characteristics. Pd is the price observed at default. Pehyld is the price observed at emergence discounted by the high yield index for the period between default and emergence. Both recoveries are measured in cents per dollar for each debt instrument. Coupon is the coupon rate of the instrument. Log (Issue size) is the natural logarithm of issue size (in millions of dollars). Bank Loans, Senior Secured, Senior Unsecured, Senior Subordinated, Subordinated are dummy variables that take on a value of 0 or 1 depending on the seniority of the instrument. Time in default is the time in years between the emergence and default dates. Maturity O/s is the remaining time to maturity of the instrument that has defaulted. Current Assets and Unsecured are dummy variables that take the value of 0 or 1 depending on the collateral for the instrument. Collateralized Debt is the ratio of dollar value of debt backed by collateral for the defaulting rm to the dollar value of all debt of the defaulting rm. Private Debt is the ratio of dollar value of bank debt for the defaulting rm to the dollar value of non-bank debt of the defaulting rm. Utility is a dummy variable if the rm belongs to the utility industry. All regressions have industry dummies (the coecients are not reported except for utility dummy). Cluster (based on each rms debt instruments as a single cluster) and heteroscedasticity corrected standard errors are reported in parentheses. , , represent signicance levels at 1%, 5%, and 10% respectively. Table 6a: OLS Estimates of Regression of Recovery Prices at Default on Contract and Firm Characteristics. Pd is the price observed at default measured in cents per dollar for each debt instrument. Coupon is the coupon rate of the instrument. Log (Issue size) is the natural logarithm of issue size (in millions of dollars). Senior Secured, Senior Unsecured, Senior Subordinated, Subordinated are dummy variables that take on a value of 0 or 1 depending on the seniority of the instrument. All rm specic variables are measured as of the last scal year before the default and data is obtained from COMPUSTAT. Prot Margin is the ratio of EBITDA to Sales. Tangibility is the ratio of Property Plant and Equipment to Total Assets. Debt Ratio is the ratio of Long-Term Debt to Total Assets. Log (Assets) is the natural logarithm of the total assets. Q ratio is the ratio of Market value of the rm (estimated as Book Value of total assets - book value of equity + market value of equity) to the book value of the rm (estimated as book value of total assets). No. of issues is the total number of debt issues of the rm that is currently under default. Debt concentration is the Herndahl index measure by amount of the debt issues of the rm that are under default. Firm return is the stock return of the rm that has defaulted in the year before default. Firm volatility is the standard deviation of daily stock returns of the rm in the year before default. Utility is a dummy variable if the rm belongs to the utility industry. All regressions have industry dummies (the coecients are not reported except for utility dummy). Cluster (based on each rms debt instruments as a single cluster) and heteroscedasticity corrected standard errors are reported in parentheses. , , represent signicance levels at 1%, 5%, and 10% respectively. Table 6b: OLS Estimates of Regression of Recovery Prices at Emergence on Contract and Firm Characteristics. Pehyld is the price observed at emergence measured in cents per dollar for each debt instrument and discounted by the high yield index for the period between default and emergence. Coupon is the coupon rate of the instrument. Log (Issue size) is the natural logarithm of issue size (in millions of dollars). Bank Loans,Senior Secured, Senior Unsecured, Senior Subordinated, Subordinated are dummy variables that take on a value of 0 or 1 depending on the seniority of the instrument. Time in default is the time in years between the emergence and default dates. Current Assets and unsecured are dummy variables that take the value of 0 or 1 depending on the collateral for the instrument. All rm-specic variables are measured as of the last scal year before the default and data is obtained from COMPUSTAT. Prot Margin is the ratio of EBITDA to Sales. Tangibility is the ratio of Property Plant and Equipment to Total Assets. Debt Ratio is the ratio of Long-Term Debt to Total Assets. Log (Assets) is the natural logarithm of the total assets. Q ratio is the ratio of Market value of the rm (estimated as Book Value of total assets - book value of equity + market value of equity) to the book value of the rm (estimated as book value of total assets). No. of issues is the total number of debt issues of the rm that is currently under default. Debt concentration is the Herndahl index measure by amount of the debt issues of the rm that are under default. Firm return is the stock return of the rm that has defaulted in the year before default. Firm volatility is the standard deviation of daily stock returns of the rm in the year before default. Utility is a dummy variable if the rm belongs to the utility industry. All regressions have industry dummies (the coecients are not reported except for utility dummy). Cluster (based on each rms debt instruments as a single cluster) and heteroscedasticity corrected standard errors are reported in parentheses. , , represent signicance levels at 1%, 5%, and 10% respectively.

Table 7a: OLS Estimates of Regression of Recovery Prices at Default on Contract, Firm and Industry Characteristics. Pd is the price observed at default measured in cents per dollar for each debt instrument. Coupon is the coupon rate of the instrument. Log(Issue size) is the natural logarithm of issue size (in millions of dollars). Senior Secured, Senior Unsecured, Senior Subordinated, Subordinated are dummy variables that take on a value of 0 or 1 depending on the seniority of the instrument. Time in default is the time in years between the emergence and default dates. Current Assets and unsecured are dummy variables that take the value of 0 or 1 depending on the collateral for the instrument. All rm-specic variables are measured as of the last scal year before the default and data is obtained from COMPUSTAT. Prot Margin is the ratio of EBITDA to Sales. Tangibility is the ratio of Property Plant and Equipment to Total Assets. Distress1 is a dummy variable that takes a value 1 if the median stock return of all the rms in the 3-digit SIC code of the defaulted rm is less than 30% and 0 otherwise. Distress2 is a dummy variable that takes on a value 1 if distress1 is 1 and if the median sales growth of all the rms in the 3-digit SIC code of the defaulted rm is negative in any of the 2 years before the default date. Med.Ind.Return is the median stock return of all the rms in the 3-digit SIC code of the defaulted rm. Med.Ind.Q is the median, of the ratio of Market value of the rm (estimated as Book Value of total assets book value of equity + market value of equity) to the book value of the rm (estimated as book value of total assets), of all the rms in the 3 digit SIC code of the defaulted rm. Ind. Liq1 is the median Quick ratio (ratio of (Current AssetsInventories) to Current Liabilities), Ind. Liq2 is the median Interest coverage ratio (EBITDA/Interest), Med. Ind. leverage is the median Long term debt to total assets, of all the rms in the 3 digit SIC code of the defaulted rm. Herndahl Index is the industry concentration measure based on sales. All Industry variables are measured in the year of default. Utility is a dummy variable if the rm belongs to the utility industry. All regressions have industry dummies (the coecients are not reported except for utility dummy). Cluster (based on each rms debt instruments as a single cluster) and heteroscedasticity corrected standard errors are reported in parentheses. , , represent signicance levels at 1%, 5%, and 10% respectively. Table 7b: OLS Estimates of Regression of Recovery Prices at emergence on Contract, Firm and Industry Characteristics. Pehyld is the price observed at emergence measured in cents per dollar for each debt instrument and discounted by the high yield index for the period between default and emergence. Coupon is the coupon rate of the instrument. Log(Issue size) is the natural logarithm of issue size (in millions of dollars). Bank Loans, Senior Secured, Senior Unsecured, Senior Subordinated, Subordinated are dummy variables that take on a value of 0 or 1 depending on the seniority of the instrument. All rm specic variables are measured as of the last scal year before the default and data is obtained from COMPUSTAT. Prot Margin is the ratio of EBITDA to Sales. Tangibility is the ratio of Property Plant and Equipment to Total Assets. Distress1 is a dummy variable that takes a value 1 if the median stock return of all the rms in the 3-digit SIC code of the defaulted rm is less than 30% and 0 otherwise. Distress2 is a dummy variable that takes on a value 1 if distress1 is 1 and if the median sales growth of all the rms in the 3-digit SIC code of the defaulted rm is negative in any of the 2 years before the default date. Med.Ind.Return is the median stock return of all the rms in the 3-digit SIC code of the defaulted rm. Med.Ind.Q is the median, of the ratio of Market value of the rm (estimated as Book Value of total assets book value of equity + market value of equity) to the book value of the rm (estimated as book value of total assets), of all the rms in the 3 digit SIC code of the defaulted rm. Ind. Liq1 is the median Quick ratio (ratio of (Current AssetsInventories) to Current Liabilities), Ind. Liq2 is the median Interest coverage ratio (EBITDA/Interest), Med. Ind. Leverage is the median Long term debt to total assets, of all the rms in the 3 digit SIC code of the defaulted rm. Herndahl Index is the industry concentration measure based on sales. Specicity of assets is dened as the ratio of machinery and equipment as percentage of total assets and follows Stromberg(2000) for all rms in Compustat over the sample period and using the S and P industry classication. All Industry variables are measured in the year of default. Utility is a dummy variable if the rm belongs to the utility industry. All regressions have industry dummies except specications eight and nine (the coecients are not reported except for utility dummy). Cluster (based on each rms debt instruments as a single cluster) and heteroscedasticity corrected standard errors are reported in parentheses. , , represent signicance levels at 1%, 5%, and 10% respectively. Coecients on Coupon, Log(Issue size), Prot Margin, Tangibility are not reported to conserve space. Table 8a: Industries in Distress. Industry Distress, Distress1, is a dummy variable that takes a value 1 if the median stock return of all the rms in the 3 digit SIC code of the defaulted rm in the default year is less than 30% and 0 otherwise. The following table lists the S and P Industry Code, the description of the industry, and the year in which it was classied as distressed using the above criterion.

Table 8b: Pattern of Recovery Prices at Default (Pd) and at Emergence (Pehyld, Pecoup, Pe) for Distressed and Non-distressed industries. Pd is the price observed at default. Pe is the price observed at emergence. Pehyld is the price observed at emergence discounted by the high yield index for the period between default and emergence. Pecoup is the price observed at emergence discounted by the coupon rate of the instrument in default for the period between default and emergence. All recoveries are measured in cents per dollar for each debt instrument. The table lists the recoveries as average over the entire sample, average over the sample whose industry is in distress in a given year, and average over the remaining sample. The medians are shown within brackets. Industry Distress is a dummy variable that takes a value 1 if the median stock return of all the rms in the 3-digit SIC code of the defaulted rm in the default year is less than 30% and 0 otherwise. The tstatistic tests for dierence of means (B)-(C). The zstatistic tests for dierences in medians (B)-(C) using the Wilcoxon rank sum test. , , represent signicance levels at 1%, 5%, and 10% respectively. Panel A is for the entire sample while Panel B excludes 1990 defaults. Table 8c: Pattern of Recovery Prices at Emergence (Pehyld, Pe) and Asset Specicity within each Industry. Pe is the price observed at emergence. Pehyld is the price observed at emergence discounted by the high yield index for the period between default and emergence. All recoveries are measured in cents per dollar for each debt instrument. Specic assets is dened as the ratio of machinery and equipment as percentage of total assets and follows Stromberg(2000) for all rms in Compustat over the sample period and using the S and P industry classication. The table lists the recoveries as average and median values over the entire sample when there is no industry distress, average over the sample whose industry is in distress in a given year. Industry Distress is a dummy variable that takes a value 1 if the median stock return of all the rms in the 3-digit SIC code of the defaulted rm in the default year is less than 30% and 0 otherwise. Panel A is for Pe and Panel B is for Pehyld. Panel C displays the correlation between specic assets and one of the recovery measures (mean Pe, median Pe, mean Pehyld or median Pehyld). Correlations are calculated separately for years when the industry in in distress and when it is not. , , represent signicance levels at 1% , 5% , and 10% respectively for the test that correlation is not equal to zero. Table 9a, 9b: OLS estimates of regression of Recovery Prices at default and emergence on Contract, Firm, Industry and Macro Characteristics. Pd is the price at default. Pehyld is the price observed at emergence discounted by the high yield index for the period between default and emergence. Both recoveries are measured in cents per dollar on each debt instrument. Coupon is the coupon rate of the instrument. Log(Issue size) is the natural logarithm of issue size (in millions of dollars). Bank Loans, Senior Secured, Senior Unsecured, Senior Subordinated, Subordinated are dummy variables that take on a value of 0 or 1 depending on the seniority of the instrument. To conserve space these variables are not reported in the table. Time in default is the time in years between the emergence and default dates. Current Assets and unsecured are dummy variables that take the value of 0 or 1 depending on the collateral for the instrument. Debt concentration is the Herndahl index measure by amount of the debt issues of the rm that are under default. All rm specic variables are measured as of the last scal year before the default and data is obtained from COMPUSTAT. Prot Margin is the ratio of EBITDA to Sales. Tangibility is the ratio of Property Plant and Equipment to Total Assets. Distress1 is a dummy variable that takes a value 1 if the median stock return of all the rms in the 3 digit SIC code of the defaulted rm is less than 30% and 0 otherwise. Med.Ind.Q is the median, of the ratio of Market value of the rm (estimated as Book Value of total assets book value of equity + market value of equity) to the book value of the rm (estimated as book value of total assets), of all the rms in the 3 digit SIC code of the defaulted rm. Ind. Liq1 is the median Quick ratio (ratio of (Current AssetsInventories) to Current Liabilities) of all the rms in the 3 digit SIC code of the defaulted rm. All Industry variables are measured in the year of default. SR, GDP, BDA, BDR are the macro variables used by Altman et.al (2002). SR is the annual return on the S and P 500 stock index. GDP is the annual GDP growth rate. BDA is the total amount of high yield bonds defaulted amount for a particular year (measured at mid-year in trillions of and represents the potential supply of defaulted securities. BDR is the weighted average default rate on bonds in the high yield bond market. Weights are based on the face value of all high yield bonds outstanding each year. Market, SMB, HML are the Fama-French factors in the 3 factor model. Utility is a dummy variable if the rm belongs to the utility industry. All regressions have industry dummies (the coecients are not reported except for utility dummy). Cluster (based on each rms debt instruments as a single cluster) and heteroscedasticity corrected standard errors are reported in parentheses. , , represent signicance levels at 1%,5%,and 10% respectively.

Table 10: OLS estimates of regression of Recovery Prices at default and emergence on risk factors that explain default. Pd is the price at default. Pehyld is the price observed at emergence discounted by the high yield index for the period between default and emergence. Both recoveries are measured in cents per dollar on each debt instrument. Coupon is the coupon rate of the instrument. Log(Issue size) is the natural logarithm of issue size (in millions of dollars). Bank Loans, Senior Secured, Senior Unsecured, Senior Subordinated, Subordinated are dummy variables that take on a value of 0 or 1 depending on the seniority of the instrument. Time in default is the time in years between the emergence and default dates. Current Assets and unsecured are dummy variables that take the value of 0 or 1 depending on the collateral for the instrument. Debt concentration is the Herndahl index measure by amount of the debt issues of the rm that are under default. All rm specic variables are measured as of the last scal year before the default and data is obtained from COMPUSTAT. Distress1 is a dummy variable that takes a value 1 if the median stock return of all the rms in the 3 digit SIC code of the defaulted rm is less than 30% and 0 otherwise. Med.Ind.Q is the median, of the ratio of Market value of the rm (estimated as Book Value of total assets book value of equity + market value of equity) to the book value of the rm (estimated as book value of total assets), of all the rms in the 3 digit SIC code of the defaulted rm. Ind. Liq1 is the median Quick ratio (ratio of (Current AssetsInventories) to Current Liabilities) of all the rms in the 3 digit SIC code of the defaulted rm. All Industry variables are measured in the year of default. ZScore is the Altman Zscore as modied by Mackie-Mason(1990). Zmij.Score is the Zmijeswki (1984) Score. Distance to default is the measure obtained by solving the Merton(1974) model for each rm. Utility is a dummy variable if the rm belongs to the utility industry. All regressions have industry dummies (the coecients are not reported except for utility dummy). Cluster (based on each rms debt instruments as a single cluster) and heteroscedasticity corrected standard errors are reported in parentheses. , , represent signicance levels at 1%,5%,and 10% respectively.

Table 1 : Time-series Pattern of Recovery Prices at Default (Pd) and at Emergence (Pehyld, Pecoup, Pe).

Year 41.96 35.80 49.25 53.31 44.70 36.65 53.99 44.55 43.67 26.82 47.02 53.88 50.20 56.09 49.88 45.34 56.40 41.50 32.07 38.00 34.00 54.75 50.50 35.00 37.25 41.00 40.00 35.54 19.50 38.50 52.75 38.00 55.00 43.50 45.00 52.50 30.00 31.00 25.34 15.01 20.50 15.67 20.43 16.96 27.70 19.91 31.31 20.90 29.96 23.27 28.89 19.82 23.30 25.32 27.31 25.77 19.63 1511 12 5 6 12 37 56 101 110 245 326 137 103 60 97 75 38 49 42 51.11 44.86 46.17 50.70 21.71 21.53 55.59 56.59 43.76 41.24 48.97 58.80 55.84 66.02 63.22 60.64 61.18 36.69 67.18 49.09 51.66 35.94 48.57 10.82 15.48 58.80 64.64 36.02 34.14 47.62 62.58 49.09 82.54 68.30 62.40 73.71 38.76 80.00 36.58 16.57 34.95 26.91 30.13 23.49 36.11 33.73 37.49 35.78 35.06 33.89 38.18 38.23 36.96 36.55 40.27 29.47 37.19

Defaults

Pd Average Median St.Dev. Defaults Median St.Dev. Firm defaults 465 5 4 3 8 16 11 24 29 69 81 53 36 25 35 27 11 16 12

Pehyld Average

Overall 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999

645 16 6 11 19 43 23 55 52 77 111 42 11 10 23 17 14 21 94

Firm defaults 379 9 5 6 12 20 14 25 28 41 53 25 9 9 18 13 9 19 64

Table 1 (continued) : Time-series Pattern of Recovery Prices at Default (Pd) and at Emergence (Pehyld, Pecoup, Pe).

Year 52.27 61.15 45.93 60.13 19.06 24.05 53.31 52.30 42.12 44.67 54.21 59.75 54.92 66.43 62.21 60.44 59.50 33.75 64.61 49.99 68.81 42.46 58.58 8.43 14.90 58.70 59.93 30.77 36.09 52.32 67.04 42.12 81.45 68.00 65.15 73.30 35.69 72.23 36.90 23.27 31.57 32.44 28.35 27.26 34.83 30.48 36.51 37.54 37.32 34.35 38.44 38.70 36.85 35.83 39.25 27.71 36.15 1541 12 5 6 12 37 56 101 110 244 344 138 111 61 100 75 38 49 42 62.02 78.63 53.81 78.31 29.18 38.03 63.03 72.07 51.73 54.27 66.99 68.03 58.61 71.96 70.45 65.16 65.88 36.78 67.28 63.07 88.06 57.66 75.46 15.13 20.56 72.04 89.25 36.71 51.88 68.50 80.39 50.35 93.57 80.17 65.13 77.50 38.76 80.00 42.98 33.53 32.28 40.48 41.83 44.57 42.05 42.31 44.96 43.05 45.57 37.81 41.01 42.02 40.63 38.73 44.28 29.46 37.20

Defaults

Pecoup Average Median St.Dev. Defaults Median St.Dev. Firm defaults 475 5 4 3 8 16 11 24 29 70 85 54 38 26 36 27 11 16 12

Pe Average

Overall 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999

1510 12 5 6 12 37 56 101 110 245 326 137 103 60 97 74 38 49 42

Firm defaults 464 5 4 3 8 16 11 24 29 69 81 53 36 25 35 26 11 16 12

Table 2 : Recovery Prices at Default and at Emergence by Industry.

S&P Code Overall 1 2 3 4 5 6 7 8 9 10 11 12 646 55 33 16 43 24 22 16 82 43 165 87 60 41.96 68.37** 39.79 27.33 36.98 29.70 36.51 46.78 41.99 39.69 37.33 42.88 45.07 38.00 77.00 33.50 22.31 36.00 28.75 36.25 49.38 40.00 29.25 38.00 35.00 37.38 25.34 20.82 26.70 9.99 26.10 24.63 26.33 22.97 21.59 29.46 21.25 26.90 25.68 1511 82 77 26 99 76 111 63 138 114 472 167 86

Industry Description Overall Utility Insurance and Real Estate Telecommunications Transportation Financial Institutions Healthcare / Chemicals High Technology/ Oce Equipment Aerospace / Auto / Capital Goods Forest,Building Prod / Homebuilders Consumer / Service Sector Leisure Time / Media Energy and Natural Resources

Def

Firm defaults 365 8 15 5 24 15 18 11 50 24 97 62 36

Pd Avg Mdn St.Dev. Def Firm defaults 424 9 23 6 20 24 35 22 46 30 126 54 29

Pehyld Avg 51.11 74.49** 37.13 53.01 38.92 58.79 55.67 47.05 52.08 53.50 47.22 51.82 60.41

Mdn 49.09 76.94 27.92 49.49 18.69 51.94 49.41 40.11 48.43 53.33 41.09 48.50 58.80

St.Dev. 36.58 18.79 30.96 44.29 40.76 42.13 38.13 38.07 37.18 32.35 35.57 36.05 35.41

Table 3a : Recovery Prices at Default and at Emergence by Seniority.

Def 598 No Data 183 107 153 148 7 110 57 133 113 5 48.33 51.06 34.01 36.07 43.70 48.00 42.50 29.00 32.00 28.90 23.10 26.89 23.13 24.11 33.00 42.07 37.75 25.17 1511 358 267 236 266 346 38 51.11 81.12 59.14 55.92 34.37 27.07 18.28 49.09 91.55 61.99 54.63 26.78 16.66 6.25

Seniority Code Overall 1 2 3 4 5 6

Seniority Description Overall Bank Loans Senior Secured Senior Unsecured Senior Subordinated Subordinated Junior Subordinated

Firm defaults 418

Pd Avg Mdn St.Dev. Def Firm defaults 829 219 119 98 172 186 35 Mdn St.Dev. 36.58 26.26 30.18 34.58 30.39 30.37 27.11

Pehyld Avg

Table 3b : Recovery Prices at Default and at Emergence by Collateral.

Def

Firm defaults

Pd Avg Mdn St.Dev. Def

Pehyld Avg

Mdn

St.Dev.

Collateral Code Overall 1 2 3 4 5 6 7 8 No data

Collateral Description Overall Current Assets PP and E Real Estate All or Most assets Other Unsecured Secured Information Not available

1511 52 83 38 228 33 32 40 1005

Firm defaults 644 46 44 23 126 20 25 17 343

51.11 94.19** 71.36 71.83 80.05 60.94 63.71 63.59 38.64**

49.09 98.81 77.74 77.77 89.16 53.67 63.79 67.42 30.91

36.58 15.96 27.51 31.07 26.35 31.21 33.48 36.43 33.48

Table 4 :Summary Statistics of Firm, Industry and Macro Characteristics. S.D. 0.20 0.25 0.35 1.33 0.43 3.97 0.28 0.36 0.03 0.34 0.18 0.43 0.13 0.34 2.08 0.10 0.12 1.99 6.84 2.82 -0.77 0.00 0.00 3.43 0.08 1.00 0.09 0.02 0.01 0.00 0.00 0.20 0.01 0.25 -0.83 0.01 -0.03 -2.03 0.30 0.84 0.07 0.33 0.48 6.05 0.76 3.50 0.40 0.30 0.07 0.00 0.00 0.91 0.17 0.99 3.08 0.18 0.17 3.54 4.07 1.81 0.94 0.93 1.93 10.46 2.45 33.00 1.00 1.70 0.18 1.00 1.00 3.52 0.69 2.05 10.32 0.50 0.38 7.27 23.53 10.27 Min Median Max

Variable

Mean

Prot Margin Tangibility Debt Ratio Log(assets) Q ratio No of issues Debt Concentration Firm Return Firm Volatility Distress1 Distress2 Med Ind Q Herndahl Index Ind Liq1 Ind Liq2 Med Ind Leverage SR GDP BDA BDR

163 163 165 168 113 196 196 86 149 94 94 94 91 86 87 90 18 18 18 18

0.08 0.38 0.49 6.23 0.86 4.27 0.49 0.43 0.07 0.13 0.03 1.02 0.20 1.01 3.43 0.18 0.18 3.22 6.47 3.06

25th Percentile 0.01 0.17 0.23 5.35 0.63 2.00 0.27 0.17 0.05 0.00 0.00 0.79 0.09 0.81 1.86 0.10 0.08 2.67 2.29 1.25

75th Percentile 0.14 0.57 0.69 7.04 0.98 5.00 0.61 0.56 0.09 0.00 0.00 1.16 0.27 1.23 4.65 0.24 0.29 4.17 7.49 3.50

Table 5 - Contract Characteristics - Pd and Pehyld Pd (1) 24.46

(11.10)

Const. Coupon Log(Issue Size) Bank Loans Senior Secured Senior Unsecured Senior Subordinated Subordinated Time in Default Maturity O/s Current Assets Unsecured Collateralized Debt Collateralized Debt * Unsecured Collateralized Debt * (1-Unsecured) Private Debt Private Debt * Bank Loans Private Debt * (1-Bank Loans) Utility Obs. R2

Pd (2) 25.30
(11.43)

Pehyld (3) 27.22

(9.06)

Pehyld (4) 9.45

(12.99)

Pehyld (5) 37.25

(10.22)

Pehyld (6) 25.49

(9.38)

Pehyld (7) 27.27

(9.46)

-0.96

-0.92

-0.14
(0.39)

-1.20 4.87

-0.17
(0.39)

-0.11
(0.38)

-0.11
(0.39)

(0.55)

(0.55)

(0.66)

2.03
(1.84)

2.17
(1.84)

-0.21
(0.91)

0.09
(0.93)

-0.14
(0.91)

-0.27
(0.92)

(1.66)

59.52 27.39 23.15

47.28 39.88 40.03

62.62 38.28 34.82 14.19

58.54
(5.74)

(5.36)

(6.63)

(6.05)

26.73 23.90

37.37 34.95 14.28

31.36 34.86 14.11

37.55
(5.87)

(7.65)

(7.91)

(5.79)

(7.90)

(5.93)

(5.76)

34.56
(6.07)

(8.16)

(8.54)

(6.23)

(8.48)

(6.22)

(6.09)

12.73
(7.94)

12.38
(8.21)

19.09

14.15
(5.07)

(5.06)

(7.82)

(5.09)

(5.07)

11.80
(8.19)

12.01
(8.46)

7.12
(5.23)

14.79

7.07
(5.27)

7.27
(5.22)

7.16
(5.22)

(7.46)

-5.41
(0.99)

-3.10
(1.22)

-5.20
(0.98)

-5.33
(1.02)

-5.44
(1.01)

-0.27
(0.28)

-0.22
(0.27)

13.27
(3.26)

-11.36
(4.52)

-2.98
(6.83)

-3.77
(8.03)

-2.03
(7.08)

4.43
(7.70)

-1.82
(8.98)

7.54
(8.56)

26.10
(8.34)

26.63
(7.95)

33.93
(6.23)

37.79
(6.61)

34.48
(6.10)

34.19
(6.21)

34.23
(6.10)

373 0.30

370 0.31

1375 0.43

370 0.48

1375 0.44

1375 0.43

1375 0.43

Table 6a - Contract, and Firm Characteristics - Pd Pd (1) 13.49

(18.19)

Const. Coupon Log(Issue Size) Senior Secured Senior Unsecured Senior Subordinated Subordinated Prot margin Tangibility Leverage Log(assets) Q Ratio No.of Issues Debt Concentration Firm return Firm Volatility Utility Obs. R2

Pd (2) 7.24
(23.15)

Pd (3) 32.28
(17.48)

Pd (4) 18.58
(17.04)

Pd (5) 8.81
(16.05)

Pd (6) 15.95
(19.77)

Pd (7) 43.45
(18.21)

-0.64
(0.63)

-0.53
(0.60)

-1.21

-0.54
(0.57)

-1.03 4.66

-0.94
(0.68)

-1.28
(0.57)

(0.60)

(0.60)

3.76

3.27
(2.22)

4.25

4.03

3.79
(2.39)

3.14
(2.22)

(2.17)

(2.49)

(2.06)

(2.18)

26.74
(8.52)

28.06
(9.33)

20.69
(8.38)

26.86
(7.51)

21.70
(6.93)

23.52
(10.38)

28.43
(9.05)

27.38

28.29

17.23

27.15

23.14

18.55

22.88
(8.41)

(8.72)

(9.18)

(8.35)

(7.62)

(6.62)

(9.09)

1.74
(8.43)

3.28
(9.26)

-5.84
(9.32)

-0.40
(7.26)

-3.32
(6.58)

-4.07
(9.40)

0.94
(9.27)

10.38
(8.38)

11.91
(9.44)

4.02
(8.97)

7.33
(7.32)

3.69
(6.70)

1.32
(8.94)

4.33
(8.84)

25.81
(9.14)

24.97
(9.84)

26.42
(14.78)

27.20
(9.37)

30.10
(8.92)

32.55
(19.47)

36.25
(12.06)

-13.75
(12.66)

-13.00
(12.53)

-37.32

-7.38
(11.13)

-11.80
(12.03)

-29.22 19.33

-29.21
(14.66)

(17.69)

(16.24)

11.81

12.31 0.81

7.89
(11.10)

8.20
(7.82)

12.86

15.52
(8.67)

(7.11)

(7.27)

(7.15)

(9.27)

(1.83)

-1.09
(8.56)

-0.91
(0.44)

23.32
(7.68)

17.81
(7.22)

-235.27
(69.52)

27.11

25.78

42.58

38.81

30.36

35.79

27.39
(10.43)

(10.32)

(10.65)

(9.14)

(12.5)

(10.08)

(11.98)

241 0.47

241 0.47

190 0.51

241 0.50

241 0.51

166 0.60

180 0.58

Table 6b - Contract, and Firm Characteristics - Pehyld Pehyld (1) 39.70

(14.22)

Const. Coupon Log(Issue Size) Bank Loans Senior Secured Senior Unsecured Senior Subordinated Subordinated Years in Default Current Assets Unsecured Prot margin Tangibility Leverage Log(Assets) Q ratio No.of Issues Debt Concentration Firm return Firm Volatility Utility Obs. R2

Pehyld (2) 30.49

(17.52)

Pehyld (3) 49.85

(18.91)

Pehyld (4) 41.19

(14.32)

Pehyld (5) 38.43

(14.32)

Pehyld (6) 49.11

(20.22)

Pehyld (7) 51.96

(21.97)

0.05
(0.47)

0.15
(0.45)

-0.03
(0.55)

0.04
(0.47)

0.01
(0.47)

0.21
(0.58)

0.34
(0.52)

1.54
(1.08)

0.86
(0.98)

1.85
(1.15)

1.55
(1.07)

1.69
(1.08)

2.17
(1.35)

2.18
(1.22)

37.53

38.02

33.73

37.18

36.79

25.38
(16.09)

25.55
(16.49)

(9.60)

(9.55)

(13.09)

(9.50)

(9.63)

20.11
(9.16)

20.19
(9.08)

19.79
(12.92)

20.45
(9.10)

19.86
(9.16)

4.43
(17.31)

4.29
(17.43)

24.90
(8.81)

24.86
(8.75)

24.31
(12.28)

25.40
(8.68)

24.97
(8.81)

14.09
(16.60)

11.85
(16.99)

-1.34
(8.31)

-0.60
(8.33)

-9.13
(12.83)

-1.29
(8.24)

-1.61
(8.35)

-22.40
(16.49)

-22.64
(16.89)

-4.35
(8.37)

-3.49
(8.40)

-6.88
(12.32)

-4.32
(8.30)

-4.74
(8.40)

-20.15
(16.54)

-20.72
(16.80)

-7.51

-7.89

-6.84

-7.43

-7.40

-8.71

-8.62
(1.40)

(1.16)

(1.25)

(1.49)

(1.15)

(1.17)

(1.53)

18.67

18.67

25.53

18.56

18.41

16.74

19.54
(5.92)

(5.23)

(5.25)

(6.44)

(5.22)

(5.19)

(6.14)

-4.29
(4.84)

-4.53
(4.78)

-6.07
(5.37)

-4.66
(4.78)

-4.83
(4.83)

-0.47
(5.76)

-2.82
(5.46)

7.14
(9.30)

5.09
(9.52)

10.40
(11.13)

7.75
(9.21)

7.82
(9.37)

5.94
(11.00)

5.17
(8.29)

-3.07
(9.85)

-1.78
(9.85)

-15.5
(11.37)

-2.64
(9.95)

-3.58
(9.81)

0.76
(12.08)

-4.64
(11.14)

-6.72
(5.31)

-5.78
(5.52)

-14.13
(8.51)

-7.01
(5.37)

-6.64
(5.32)

-12.20
(8.69)

-9.74
(8.45)

1.59
(1.55)

-0.14
(7.39)

-0.21
(0.32)

5.81
(6.70)

5.78
(5.58)

-55.05
(64.31)

43.00

41.45

45.17

46.25

43.86

41.78

48.86
(8.95)

(8.06)

(7.98)

(8.40)

(10.42)

(8.01)

(10.06)

670 0.57

670 0.57

468 0.60

670 0.57

670 0.57

348 0.67

396 0.66

Table 7a - Contract, Firm, and Industry Characteristics - Pd Pd (1) 11.30

(14.35)

Const. Coupon Log(Issue Size) Senior Secured Senior Unsecured Senior Subordinated Subordinated Prot margin Tangibility Debt Concentration Firm Volatility Distress1 Distress2 Med Ind Q Herndahl Index Ind Liq1 Ind Liq2 Med Ind Leverage Med Ind Return Utility Obs. R2

Pd (2) 6.14
(13.81)

Pd (3) 13.04
(13.68)

Pd (4) 10.16
(14.19)

Pd (5) 4.88
(13.35)

Pd (6) 33.25
(16.76)

Pd (7) 11.84
(16.57)

Pd (8) 5.06
(13.97)

-0.92
(0.59)

-0.57
(0.47)

-0.83
(0.57)

-1.00 5.57

-0.74 4.75

-1.15
(0.74)

-0.95 5.39

-0.70
(0.47)

(0.59)

(0.45)

(0.58)

5.47

4.33

4.71

5.30

4.59
(2.00)

(2.39)

(1.98)

(2.13)

(2.41)

(1.98)

(2.85)

(2.42)

26.02
(9.10)

23.55
(9.09)

25.83
(8.94)

25.53
(8.89)

23.38
(8.90)

25.15
(9.16)

24.51
(9.18)

22.37
(9.04)

24.85

24.27

25.17

25.11

24.71

18.25

24.28

24.03
(8.63)

(8.53)

(8.80)

(8.53)

(8.33)

(8.56)

(7.34)

(8.55)

2.07
(7.66)

1.20
(7.85)

2.75
(7.83)

1.00
(7.47)

0.33
(7.65)

-4.76
(7.48)

0.64
(7.71)

0.03
(7.79)

8.86
(8.25)

7.80
(8.34)

8.85
(8.17)

8.46
(8.14)

7.67
(8.31)

3.82
(8.89)

6.76
(8.29)

6.14
(8.26)

27.48
(7.86)

29.52
(7.83)

29.04
(6.76)

26.29
(7.79)

28.19
(7.63)

30.33
(7.71)

28.70
(8.18)

29.45
(7.97)

0.20
(14.22)

-0.96
(12.26)

-2.66
(12.56)

3.30
(14.37)

2.14
(12.20)

-9.94
(18.84)

-0.49
(13.93)

0.51
(11.72)

24.39
(8.22)

20.66
(8.17)

20.98
(7.29)

25.44
(8.09)

22.33
(8.16)

21.87
(9.21)

24.73
(8.24)

21.30
(8.17)

-276.88 -12.68

(87.27)

-10.06

-10.64

(4.23)

(4.16)

(4.29)

-20.65
(6.72)

-15.98
(7.49)

-19.60
(7.14)

-5.12
(8.90)

-3.88
(8.50)

-4.11
(8.40)

-5.87
(8.80)

-4.63
(8.66)

-0.72
(9.97)

-1.87
(9.19)

-1.31
(8.60)

-13.71
(15.00)

-18.00
(14.42)

-19.78
(14.70)

-5.75
(14.64)

-11.13
(13.95)

28.58

-22.28
(18.25)

-21.23
(15.80)

(13.70)

-1.09
(5.49)

-1.71
(5.45)

-5.66
(9.57)

-1.19
(5.96)

2.44

2.16
(1.32)

2.53
(1.21)

(1.21)

6.48
(20.79)

-0.45
(6.15)

-0.06
(5.58)

19.14 243 0.53

18.60 254 0.53

18.03
(11.17)

19.01 243 0.53

18.58 254 0.53

24.61

18.61 243 0.51

17.84
(10.47)

(10.94)

(10.43)

(10.87)

(10.36)

(11.47)

(11.11)

262 0.52

186 0.62

254 0.52

Table 7b - Contract, Firm, and Industry Characteristics - Pehyld

Pehyld (1) 8.66
(14.74)

Const. Bank Loans Senior Secured Senior Unsecured Senior Subordinated Subordinated Years in Default Current Assets Unsecured Distress1 Distress2 Med Ind Q Herndahl Index Ind Liq1 Ind Liq2 Med.Ind Leverage Med.Ind Return Specicity (mean) Specicity (median) Distress1 * Specicity (mean) Distress1 * Specicity (median) Utility Obs. R2

Pehyld (2) 12.90

(14.32)

Pehyld (3) 20.20

(15.27)

Pehyld (4) 9.20

(14.74)

Pehyld (5) 12.86

(14.37)

Pehyld (6) 8.60

(15.69)

Pehyld (7) 11.57

(15.28)

Pehyld (8) 3.07

(16.91)

Pehyld (9) -0.29

(16.74)

44.67

42.26

44.49

42.41

40.07

42.18

39.88

51.13 32.25 35.49

51.32
(9.99)

(9.58)

(9.73)

(9.63)

(9.36)

(9.59)

(9.39)

(9.64)

(9.93)

22.38 26.60

21.30 26.99

22.87 27.86

21.00 25.24

19.98 25.74

20.68 25.03

19.81 25.62

32.83
(8.76)

(8.95)

(9.19)

(8.88)

(8.84)

(9.13)

(8.89)

(9.17)

(8.71)

35.67
(8.54)

(8.38)

(8.67)

(8.46)

(8.35)

(8.69)

(8.39)

(8.74)

(8.52)

4.41
(8.15)

3.16
(8.36)

4.54
(8.34)

2.77
(8.06)

1.32
(8.24)

2.87
(8.09)

1.35
(8.29)

11.03
(7.70)

11.12
(7.70)

-0.37
(7.93)

-2.24
(8.26)

-1.01
(8.09)

-1.83
(7.80)

-3.88
(8.13)

-2.25
(7.85)

-4.25
(8.16)

1.64
(7.83)

1.49
(7.82)

-5.40
(1.13)

-6.04
(1.16)

-6.03
(1.14)

-5.40
(1.15)

-6.05
(1.18)

-5.48
(1.16)

-6.10
(1.18)

-3.82
(1.36)

-3.65
(1.39)

18.45
(7.39)

16.69
(7.16)

17.99
(7.13)

18.91
(7.39)

16.48
(7.09)

19.13
(7.40)

16.59
(7.14)

15.01
(6.83)

14.96
(6.87)

-3.35
(4.82)

-5.85
(4.89)

-4.53
(4.86)

-3.81
(4.85)

-6.20
(4.95)

-3.90
(4.86)

-6.33
(4.97)

-3.80
(6.19)

-3.60
(6.25)

-10.78
(4.72)

-9.89
(4.38)

-10.78
(4.28)

31.23
(11.17)

28.70
(10.11)

-9.65
(5.85)

-7.08
(6.31)

8.80
(5.20)

8.22
(5.41)

8.92
(5.33)

10.19
(5.32)

9.60
(5.51)

10.94
(5.21)

10.14
(5.49)

11.76
(6.31)

11.81
(6.31)

-8.12
(13.99)

-4.65
(14.59)

-6.56
(14.65)

-6.34
(15.01)

-2.16
(15.65)

-11.42
(13.90)

-5.03
(14.54)

-24.06
(17.66)

-23.94
(17.46)

12.62
(4.66)

12.31
(4.65)

12.93
(4.86)

7.70
(5.22)

7.94
(5.28)

2.13
(0.86)

2.18
(0.88)

2.25
(0.85)

-6.45
(16.56)

0.94
(3.76)

1.18
(4.00)

-0.24
(0.25)

-0.12
(0.25)

-1.61
(0.38)

-1.69
(0.37)

41.89

42.06

43.8

42.16

42.09

42.27

42.02

(6.98)

(7.36)

(7.25)

(7.02)

(7.39)

(7.05)

(7.38)

729 0.58

754 0.58

765 0.58

729 0.58

754 0.58

729 0.58

754 0.58

729 0.50

729 0.50

Table 8a: Industries in Distress. S and P Code 4 12 5 6 2 4 5 6 7 8 9 10 11 5 10 2 6 11 6 10 11 10 6 Description Transportation Energy and Natural Resources Financial Institutions Healthcare/Chemicals Insurance and Real Estate Transportation Financial Institutions Healthcare/Chemicals High Technology/Oce Equipment Aerospace/Auto /Capital goods Forest, Building Products/Home Builders Consumer/Service Sector Leisure Time/Media Financial Institutions Consumer/Service Sector Insurance and Real Estate Healthcare/Chemicals Leisure Time/Media Healthcare/Chemicals Consumer/Service Sector Leisure Time/Media Consumer/Service Sector Healthcare/Chemicals Year 1984 1986 1987 1987 1990 1990 1990 1990 1990 1990 1990 1990 1990 1991 1993 1994 1994 1994 1995 1995 1995 1996 1998

Table 8b: Industry Distress Behavior of Recovery Prices at Default (Pd) & Emergence (Pehyld, Pecoup, Pe). Recovery rates (Panel A) Pd Pe Pehyld Pecoup Recovery rates (Panel B) Pd Pe Pehyld Pecoup Full sample Mean (Median) 44.2 61.8 50.8 52.0 (39.5) (61.7) (48.4) (49.0) Obs 387 1473 1443 1442 Obs 330 1237 1209 1209 No Industry Distress Mean(Median) 46.0 63.4 52.4 53.4 (41.0) (65.0) (50.3) (51.3) Obs 350 1312 1285 1285 Obs 320 1194 1167 1167 Distress Mean(Median) 26.5 48.8 37.8 40.8 (19.0) (35.0) (24.9) (27.9) Obs 37 161 158 157 Obs 10 43 42 42 t-statistic (z- statistic) 4.52*** 4.07*** 4.77*** 4.03*** (4.85)*** (4.21)*** (4.92)*** (4.24)***

Full sample Mean (Median) 47.4 63.5 52.8 53.4 (41.7) (65.0) (50.5) (51.3)

No Industry Distress Mean(Median) 47.5 64.1 53.2 53.9 (42.0) (66.8) (51.2) (52.4)

Distress Mean(Median) 42.5 47.0 40.2 41.5 (38.0) (32.5) (27.5) (34.3)

t-statistic (z- statistic) 0.62 (0.62) 2.57*** (2.62)*** 2.30** (2.33)** 2.15** (2.15)**

Table 8c: Industry Distress Behavior of Recovery Prices at Emergence (Pehyld, Pe) and Asset Specicity. Panel A (Pe) Description Utility Insurance and Real Estate Telecommunications Transportation Financial Institutions Healthcare / Chemicals High Technology / Computers / Oce Equipment Aerospace / Automotive / Capital Goods / Metals Forest and Building Products / Homebuilders Consumer / Service Sector Leisure Time / Media Energy / Natural Resources Panel B (Pehyld) Utility Insurance and Real Estate Telecommunications Transportation Financial Institutions Healthcare / Chemicals High Technology / Computers / Oce Equipment Aerospace / Automotive / Capital Goods / Metals Forest and Building Products / Homebuilders Consumer / Service Sector Leisure Time / Media Energy / Natural Resources Panel C (Correlations) No Industry Distress Specic Specic Specic Specic Assets Assets Assets Assets (mean), Pe (mean) (median), Pe (median) (mean), Pehyld (mean) (median), Pehyld (median) -0.066 0.224 -0.015 0.200 No Industry Distress -0.167 0.188 -0.086 0.072 Industry Distress -0.800 -0.579 -0.763 -0.626 Industry Distress -0.826 -0.566 -0.896 -0.688 14.9% 8.4% 36.5% 46.1% 2.7% 16.8% 18.6% 18.5% 19.7% 26.3% 19.8% 29.6% 14.1% 1.6% 30.0% 44.2% 0.0% 14.6% 15.8% 16.5% 16.9% 23.6% 16.3% 25.0% 73.99 36.06 53.01 47.35 50.42 60.85 43.31 53.37 56.04 48.54 52.99 60.80 76.80 26.17 49.49 47.93 40.35 71.63 28.59 50.09 56.64 42.91 47.79 58.80 no data 42.42 no data 9.46 81.35 33.29 72.77 36.81 36.73 34.37 28.74 14.88 no data 30.73 no data 3.02 100.00 14.85 75.40 32.91 23.96 34.10 25.86 15.54 Specic Assets Mean 14.9% 8.4% 36.5% 46.1% 2.7% 16.8% 18.6% 18.5% 19.7% 26.3% 19.8% 29.6% Median 14.1% 1.6% 30.0% 44.2% 0.0% 14.6% 15.8% 16.5% 16.9% 23.6% 16.3% 25.0% No Industry Distress Mean 109.60 44.98 68.52 55.71 61.04 69.08 52.59 63.67 74.04 57.25 58.89 67.36 Median 103.75 36.46 100.00 57.51 61.55 75.80 39.45 55.30 80.00 53.67 55.30 72.04 Industry Distress Mean no data 66.51 no data 14.96 89.10 35.26 82.53 47.75 52.74 50.57 41.28 20.44 Median no data 39.00 no data 5.14 100.00 16.63 100.00 51.75 38.00 47.65 28.25 20.56

Panel D (Correlations)- Excluding High Tech and Financial Institutions Specic Specic Specic Specic Assets Assets Assets Assets (mean), Pe (mean) (median), Pe (median) (mean), Pehyld (mean) (median), Pehyld (median)

Table 9a - Contract,Firm,Industry and Macro Characteristics Pd (1) 5.47

(26.14)

Const. Coupon Log(Issue Size) Time in Default Current Assets Unsecured Prot margin Tangibility Debt Concentration Distress1 Med Ind Q Ind Liq1 SR GDP BDA BDR Market SMB HML Utility Obs. R2

Pd (2) 14.83
(11.79)

Pehyld (3) 22.74

(15.26)

Pehyld (4) 16.06

(14.20)

Pehyld (5) 10.29

(14.34)

-0.91
(0.62)

-0.98
(0.60)

0.10
(0.46)

0.14
(0.46)

0.07
(0.47)

5.98
(2.44)

5.54
(2.31)

1.58
(1.12)

1.31
(1.11)

1.28
(1.13)

-5.98
(1.13)

-5.43
(1.12)

-4.08
(1.96)

18.13

17.42

19.27
(7.28)

(8.22)

(7.77)

-4.39
(4.79)

-4.56
(4.72)

-3.27
(4.68)

25.61
(7.72)

26.02
(7.57)

12.52
(9.14)

14.03
(9.51)

14.17
(9.11)

2.00
(12.73)

0.31
(12.64)

-0.69
(9.04)

-0.78
(9.08)

2.48
(9.17)

22.08
(8.63)

23.77
(7.97)

-9.80

-13.19
(4.77)

-11.60
(4.80)

-10.04
(4.51)

-11.10
(4.57)

(5.05)

-8.18
(9.07)

-6.18
(8.79)

5.69
(5.14)

5.70
(5.18)

8.31
(5.25)

3.89
(7.73)

-2.03
(5.19)

12.41
(5.17)

14.30
(5.09)

12.00
(5.26)

14.11
(18.48)

-17.05
(12.87)

-167.04
(150.50)

-44.41
(139.81)

-558.03
(294.00)

-381.92
(300.44)

-17.85
(80.13)

-62.56
(43.31)

-3.61
(12.69)

8.25
(18.14)

-16.97
(11.22)

20.57 242 0.54

21.25

41.78

41.48

39.48
(7.18)

(10.57)

(10.77)

(7.35)

(7.15)

242 0.52

709 0.59

709 0.58

702 0.58

Table 9b - Contract, Firm and Macro Characteristics Pd (1) 18.91

(12.97)

Const. Coupon Log(Issue Size) Time in Default Current Assets Unsecured Prot margin Tangibility Debt Concentration SR GDP BDA BDR Market SMB HML Utility Obs. R2

Pd (2) 18.39
(12.04)

Pehyld (3) 49.73

(14.42)

Pehyld (4) 42.89

(13.61)

Pehyld (5) 36.91

(13.49)

-0.71
(0.57)

-0.87 3.92

-0.19
(0.49)

-0.11
(0.49)

-0.26
(0.50)

(0.53)

4.32

1.84

1.34
(1.04)

1.32
(1.07)

(2.43)

(2.32)

(1.05)

-7.60
(1.13)

-6.95
(1.16)

-6.15
(2.04)

17.82
(7.65)

16.66
(7.13)

18.88
(6.66)

-7.00
(4.94)

-7.54
(4.94)

-5.85
(4.75)

8.30
(4.92)

9.23
(4.92)

12.31
(3.49)

13.15
(3.44)

13.34
(3.45)

7.67
(10.88)

6.69
(11.12)

-0.47
(8.49)

-1.91
(8.55)

1.79
(8.80)

11.57
(7.63)

15.79
(7.42)

21.70
(16.67)

-15.98
(12.90)

-181.56
(145.18)

-83.42
(140.07)

-678.52
(260.37)

-717.77
(317.06)

-39.61
(70.88)

-101.75
(44.82)

-4.73
(13.13)

-3.47
(18.59)

-14.38
(11.27)

18.95
(9.23)

20.37
(10.36)

42.04
(7.51)

42.61
(7.44)

41.10
(7.30)

266 0.46

266 0.44

760 0.57

760 0.56

753 0.56

Table 10: OLS estimates of regression of Recovery Prices at default and emergence on risk factors that explain default. Pd (1) Const. Coupon Log(Issue Size) Bank Loans Senior Secured Senior Unsecured Senior Subordinated Subordinated Time in Default Current Assets Unsecured Debt Concentration Distress1 Med Ind Q Ind Liq1 Z-Score Zmij. Score Distance to Default Utility Obs. R2 32.01

Pd (2) 12.12
(15.36)

Pd (3) 8.84
(19.16)

Pehyld (4) 13.24

(15.17)

Pehyld (5) 15.65

(14.37)

Pehyld (6) 10.87

(18.33)

1.43
(17.89)

-0.99
(0.63)

-1.15
(0.67)

-0.65
(1.03)

-0.20
(0.50)

0.03
(0.48)

0.65
(0.47)

6.14
(2.38)

6.58
(2.48)

3.54
(3.60)

1.85
(1.15)

2.26
(1.28)

0.96
(1.10)

44.62
(10.50)

44.98
(10.18)

36.90
(14.09)

24.80

21.68

18.33

25.31

22.79

9.63
(14.47)

(9.19)

(8.59)

(7.79)

(9.12)

(8.88)

22.28
(8.01)

20.15
(7.62)

5.79
(7.90)

27.40
(8.42)

24.84
(8.44)

16.59
(14.44)

2.63
(8.18)

-1.72
(7.72)

-8.10
(7.84)

6.67
(8.78)

3.52
(8.57)

-17.07
(14.16)

11.76
(9.26)

6.48
(8.16)

-1.62
(8.94)

4.96
(8.18)

4.13
(8.01)

-7.95
(13.92)

-6.14
(1.23)

-6.32
(1.14)

-6.22
(1.39)

15.60

15.22
(9.49)

18.86
(6.78)

(8.89)

-5.81
(5.71)

-4.79
(5.54)

-0.51
(5.09)

22.74
(9.94)

28.46
(9.87)

31.71
(13.15)

-14.45

-13.30

-20.70
(8.69)

-14.25
(5.16)

-14.50
(4.88)

-14.14
(8.19)

(4.36)

(4.77)

5.56
(13.11)

0.67
(11.73)

-1.77
(11.76)

6.96
(4.94)

5.26
(5.09)

5.55
(5.57)

-4.85
(5.19)

-5.27
(5.11)

-1.71
(8.94)

13.30
(5.43)

13.61
(5.18)

16.22
(9.76)

2.94

3.32 -0.19
(0.36)

(1.39)

(1.20)

-0.92
(0.36)

0.48 26.91

0.16
(0.10)

(0.20)

28.03

47.57

43.14

48.55
(6.09)

(8.50)

(7.87)

(8.64)

(6.24)

(5.18)

211 0.51

212 0.50

165 0.65

598 0.59

609 0.57

395 0.68

Time-series behavior of Recoveries at Default (Pd) and at Emergence (Pehyld)

90.00 80.00 70.00 Cents per dollar 60.00 50.00 40.00 30.00 20.00 10.00 0.00 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Years - 1982 to 1999 10 0 # Defaulted firms 50 40 30 20 70 60

Median Pd Median Pehyld #Firm Defaults

Figure 1: Time-series behavior of Recoveries at Default (Pd) and at Emergence (Pehyld) This figure plots the time-series variation in the number of firm defaults (corresponding to Pd series), median recovery price at default (Pd) in each year, and median recovery price at emergence (Pehyld) in each year.