The Determinants of Recovery Rates in the USCorporate Bond Market

Rainer Jankowitsch∗ Florian Nagler†

Marti G. Subrahmanyam‡

First Version: April 26, 2012This Version: September 3, 2012

Abstract

We analyze the recovery rates of defaulted bonds in the US corporate bond market over the time pe-riod 2002 to 2010. Our data set is obtained from the Trade Reporting and Compliance Engine (TRACE)database maintained by the Financial Regulatory Authority (FINRA) and provides us with a completeset of traded prices and volumes around the default events. The analysis of the microstructure of tradingactivity allows us to estimate reliable market-based recovery rates. We investigate the relation betweenthese recovery rates and a comprehensive set of bond characteristics, firm fundamentals and macroeco-nomic variables. In addition, we explore the effect of the liquidity of the individual bonds on their tradedprices after default. Our panel regression analysis explains 64% of the total variance in the recovery ratesacross bonds. We find that the type of default event, the seniority of the bond, and the industry in whichthe firm operates, are important determinants of the recovery rate. However, balance sheet ratios moti-vated by structural credit risk models, macroeconomic variables, and transaction costs metrics measuringliquidity, are of equal importance.

JEL-Classification: G12, G33Keywords: recovery rate, corporate bonds, liquidity

∗WU (Vienna University of Economics and Business), Department of Finance, Accounting and Statistics, Heili-genstadter Straße 46-48, 1190 Vienna, Austria; email: rainer.jankowitsch@wu.ac.at†VGSF (Vienna Graduate School of Finance), Heiligenstadter Straße 46-48, 1190 Vienna, Austria; email: flo-

rian.nagler@vgsf.ac.at‡New York University, Stern School of Business, Department of Finance, 44 West Fourth, Room 9-68, New York,

NY 10012; email: msubrahm@stern.nyu.eduWe thank Edward Altman and the Salomon Center of New York University for providing us access to the

Master Default Database.

1 Introduction

The global financial crisis has highlighted the importance of credit risk in the pricing of financial

contracts, and emphasized the multifaceted nature of its key determinants: the probability of

default and the recovery rate in the event of default. Traditionally, credit risk modelling has

been focused on the probability of default, while setting the recovery rate to parametric values

that do not necessarily recognize its potential cross-sectional and time-series variation. However,

the magnitude and variability of defaults during the crisis have emphasized the importance of

obtaining more precise estimates of recovery rates, and explaining their variation across issues

and issuers. It is now intuitively understood that recovery rates are potentially driven by many

different factors: endogenous variables (such as specific characteristics of the assets involved and

the characteristics of the firm and industry), or exogenous factors (such as overall macroeconomic

conditions or market liquidity). It is important, therefore, to understand the determinants of this

risk factor and to analyze their interaction effects with other dimensions of default risk.

This paper aims at investigating these relationships at the issue and issuer level for the US

corporate bond market. The Trade Reporting and Compliance Engine (TRACE ) database, main-

tained by the Financial Regulatory Authority (FINRA), allows us to analyze, for the first time,

the prices of defaulted bonds, based on a complete set of transaction data, over the time period

2002 to 2010, covering most of the period before and after the onset of the global financial crisis.

Specifically, we investigate the relation between the recovery rates and a comprehensive set of

bond characteristics, firm fundamentals and macroeconomic variables based on panel data regres-

sions. In addition, we explore the influence of market liquidity on recovery rates, an innovation in

relation to the existing literature, which is largely silent on this issue.

Most credit risk instruments, such as bonds and credit default swaps (CDS), trade over-the-

counter (OTC). This makes research in this area challenging, as traded prices and volumes for

these instruments cannot be directly observed from a central database. Therefore, most studies

have to rely, of necessity, on quotation or trade data from a particular dealer, leaving open the

question of whether the data are representative of the market as a whole. This is even more

of a problem for defaulted financial instruments, as their trading can often be infrequent, with

stale prices, with some quotations or trades of individual dealers even “off market“. However, the

market for US corporate bonds is an ideal laboratory for this study as detailed data on prices and

1

volumes are available since 2002 from the TRACE database, covering all trades following default

events as well. The data not only permit a reliable estimate of a market-based recovery rate,

but also provide an opportunity to study trading activity, and thus liquidity, at different stages

following default.

We combine the TRACE data set with the Mergent Fixed Income Securities Database and the

NYU Salomon Center Master Default Database, which allows us to study a broad set of default

events capturing formal bankruptcy filings, distressed exchanges and downgrades to default status

by rating agencies. Furthermore, we add firm-specific information from the Compustat Database

and macroeconomic indicators from multiple sources. Overall, we cover more than 2, 000 default

events for about 800 bonds, and approximately 1, 700, 000 trades in the relevant time period around

default.

We make three contributions in this paper. First, we provide a detailed analysis of the mi-

crostructure of trading in defaulted bonds offering some interesting new insights. This allows us

to derive reliable market-based estimates of recovery rates. Second, we analyze these recovery

rates employing a broad set of explanatory variables in our regressions, in contrast to much of

the literature, where the analysis has typically been more narrowly focused. Third, we include

liquidity measures in our analysis of recovery rates, which turn out to be of particular importance

when dealing with defaulted bonds, which are potentially illiquid.

Our analysis of recovery rates yields several distinct sets of findings. We examine the trading

activity of the defaulted bonds, as defined by traded prices and volumes, 90 days before, at, and 90

days after the observed default event date. We find that although the price level is already rather

low before the default event, the traded price falls significantly to its lowest level on the default

day itself, around 35% of face value, on average. The price recovers, in the first 30 days following

default, to about 42% of face value and shows a more volatile evolution thereafter.1 Furthermore,

we find that the trading volume of a defaulted bond is relatively high on the default event day.

This high trading activity dies down quickly within the first 30 days after default to pre-default

levels. Based on these findings, we define the recovery rate of a defaulted bond as the average

traded price per unit of face value, between the default day and the next 30 days after default,

and thus, providing a market-based specification.

We analyze these recovery rates across bonds along various dimensions. Comparing different

1Note that a 40% recovery rate, which was the point estimate provided by Altman and Kishore (1996) in anearly paper in this area, has been widely used in calibrations in academia and industry.

2

default event types, we find that distressed exchanges have the highest recovery rates, whereas

Chapter 11 filings result in significantly lower recoveries. This finding provides further evidence

that out-of-court restructurings lead to higher recovery compared to formal bankrupcty proce-

dures. Furthermore, we find significant differences in the performance of the default grades of the

major rating agencies; in particular, the rating frameworks of Moody’s and Fitch seem to incor-

porate recovery rate information to a greater extent.2 Analyzing the differences in the recovery

rates of corporate bonds across industries, we find that among non-financial industries, utility

and energy related firms recover the most in default, while retailers recover the least. Interest-

ingly, among financial firms, the banking and credit & financing industries recover the most in

default, whereas the financial services industry recovers the least. Analyzing the seniority levels of

different defaulted bonds, we find, as expected, that secured bonds recover more than unsecured

and subordinated bonds. Furthermore, we document a substantial variation in recovery rates over

time, e.g., we find quarterly moving averages between 20% and 80% of face value, for the time

period 2002 to 2010.

In the main part of our analysis, we employ panel regressions to explain the variation in recovery

rates using a comprehensive set of bond characteristics, balance sheet ratios, macroeconomic

variables and liquidity measures (in addition to dummy variables, based on the default event type,

industry and seniority). Overall, the regression analysis explains 64% of the total variation in the

recovery rates, with all four groups of variables contributing to the explanatory power of recovery

rates. Among balance sheet ratios, we find significant effects for ratios motivated by structural

credit risk models, i.e., the higher the equity ratio, and the lower the default barrier, the higher the

recovery rate. Furthermore, we document the direct effect of the total assets of the firm, i.e., larger

firms have higher recoveries. Analyzing macroeconomic variables, we find a particularly strong

effect for the market-wide default rate. Thus, we find clear evidence that a high default rate in

the market as a whole, a systematic risk factor, is linked to significantly lower recovery rates of

individual bonds, following default. Along the same lines, we find a positive relation between short

term interest rates, an indicator of the business cycle, and recovery rates. In addition, we analyze

the effect of liquidity, which is an important innovation in this paper, and find a clear link between

the defined liquidity measures of bonds and their recovery rates. In particular, when measuring

2It is often claimed that the rating frameworks of Moody’s and Fitch focus on the expected loss, which involvesboth the probability of default and the recovery rate given default. In contrast, Standard and Poor’s ostensiblyconsiders only the probability of default in its ratings.

3

the transaction costs of trading using the price dispersion measure, we find that illiquid bonds with

high transaction costs recover less, following default. Additional evidence of the liquidity effects is

provided by the trading activity variables. Further, we find that bonds that can be delivered into

a CDS contract have a significantly higher recovery rate, possibly because of buy-side pressure

from protection buyers, who are required to physically deliver the underlying bond.

Overall, we provide a comprehensive analysis going beyond the results that have been presented

in the prior literature. We offer detailed insights into the stochastic nature and drivers of recovery

rates by analyzing a broad set of explanatory variables rather than providing evidence only on

the effects of any one factor. Our results on the effects of liquidity are particularly noteworthy,

since our paper is the first one, to our knowledge, to report findings on the effects of liquidity on

recovery rates.

The paper is organized as follows: Section 2 reviews the literature. Section 3 provides details

of the data used in our analysis. Section 4 states the main hypotheses that are being tested and

the research questions that are being addressed. Section 5 presents the methodology and explains

the setup of the subsequent analysis. Section 6 provides the descriptive analysis and the results

of the panel data regressions. Section 7 concludes.

2 Literature Review

The literature on recovery rates can be divided into two categories: theoretical papers dealing

with credit risk models, which make implicit or explicit assumptions about recoveries in default,

and empirical papers analyzing past default events. Traditionally, credit risk models have been

divided into structural and reduced-form models (see e.g., Altman et al. (2002) for a detailed

discussion). In the basic structural models, starting with Black and Scholes (1973) and Merton

(1974), the default risk of a firm is driven by the process generating the value of its assets; hence,

the risk of the firms’s default is explicitly linked to the volatility of its asset value. Default occurs

when the value of a firm’s assets is lower than that of its liabilities at maturity. In this case,

the debt holders receive the residual market value of the firm’s assets. Hence, in this setup, the

recovery rate, as the residual value of the defaulted company’s assets, is an endogenous variable

that is inversely related to the probability of default. This relation becomes even more evident

when structural models are used as the basis for credit portfolio analysis (see, e.g., Frye (2000) or

4

Gordy (2003)), where asset values are modelled by market-wide factors and idiosyncratic factors,

in which market factors lead to a negative relation between aggregate default and recovery rates.

Several authors provide extensions to the basic Merton (1974) model.3 They generally assume

that default may occur at any time between the issuance and maturity of the debt, that default is

triggered when the value of the firm’s assets reaches a lower threshold barrier, or that bankruptcy

costs arise exogenously. Interestingly, in most of these models the recovery rate is assumed to be

exogenous and independent of the firm’s asset value. It is generally defined as a fixed proportion of

the outstanding debt value, in terms of either face or market value, and is, therefore, independent

of the probability of default.

Reduced-form models of credit risk do not condition default on the structural features of the

firm (see, for example, Jarrow and Turnbull (1995), Duffie and Singleton (1997), Lando (1998), and

Madan and Unal (1998)). Rather, these models allow separate, explicit assumptions regarding the

dynamics of both the probability of default and the recovery rate. Although a complex dependence

structure can be used in such models, in principle, the recovery rate is usually assumed to be

exogenous, either deterministic or stochastic, and often independent of the probability of default.

It has been well documented that neither reduced-form models (see, e.g., Longstaff et al. (2005))

nor structural models (see, e.g., Huang and Huang (2002)) can fully explain observed yield spreads

satisfactorily. It is relevant, therefore, to understand the stochastic nature of recovery rates and

provide evidence from past defaults. During the last two decades, direct attempts have been

undertaken to empirically investigate the behavior of recovery rates. An important first analysis

is provided by Altman and Kishore (1996), who use a data set of over 700 defaulted bond issues

from 1978 to 1995 and analyze the effect of industry affiliation on recovery rates, and conclude

that the highest average recoveries come from public utilities (70%) and chemical, petroleum, and

related products (63%), and that the original rating of a bond has virtually no effect on recovery,

once seniority is accounted for. Hanson and Schuermann (2004) provide similar evidence for the

impact of seniority and industry affiliation, when analyzing a sample of around 2,000 defaults

of bonds and loans. Furthermore, they discuss the empirical distribution of recovery rates and

provide evidence that recoveries are lower in recessions. Along the same lines, Altman et al. (2005)

analyze the relationship at a macroeconomic level and conclude that the average annual recovery

3Leland (1994), Longstaff and Schwartz (1995), Leland and Toft (1996), Anderson and Sundaresan (1996),Collin-Dufresne et al. (2001), Goldstein et al. (2001), Mella-Barral and Perraudin (1997), Black and Cox (1976)and Acharya et al. (2006) are examples of such analyses.

5

rates and default rates are indeed negatively correlated. They show that realized default rates in

a particular year are important drivers of recoveries, whereas other macroeconomic variables, i.e.,

the performance of the economy measured by GDP or the GDP growth rate, are less predictive

than most theoretical papers would suggest.

Acharya et al. (2007) provide a detailed analysis of industry-wide distress and its relation to

recovery rates. They argue that when an industry is in distress, defaulting firms in the industry

experience lower recoveries. One mechanism causing this effect is the lower ability of the distressed

firm to sell assets to competitors in the same industry, as discussed in Shleifer and Vishny (1992).

Using a data set from 1982 to 1999, with about 800 observations, they provide evidence that the

defaulted debt of industries in distress recover 10% to 15% less on average. They also document

a negative effect of aggregate default rates on the recovery rate of individual issues. Furthermore,

they provide evidence that balance sheet ratios, motivated by structural models, are of importance.

Altman and Kalotay (2010) provide further evidence for industry-driven effects, focusing on the

modelling of the distribution of recovery rates, based on ultimate recoveries for defaulted loans

and bonds.4

Additionally, Bris et al. (2006) and Davydenko and Franks (2008) provide evidence that differ-

ences in creditors’ rights and reorganization practices are reflected in the level of recovery rates.

Thus, the particular default event type is of importance. They provide evidence comparing de-

faults across different bankrupcty procedures, e.g., Chapter 7 versus Chapter 11 filings, as well

as across different countries or jurisdictions. Altman and Karlin (2009) provide further evidence

of the importance of the default event by discussing distressed exchanges. They find that in

distressed exchanges, recoveries are higher, compared with other default events.

Our paper extends the existing literature in new directions and provides detailed empirical

evidence analyzing the driving factors of recovery rates. In contrast to prior studies, we employ

market-based estimates of the recovery rates, based on a detailed analysis of the trading mi-

crostructure. Furthermore, we make use of a set of explanatory variables, which have not been

analyzed in the existing literature comprehensively, but rather on a stand-alone basis, and we

include liquidity measures in the analysis, which is an additional contribution.

4See Altman et al. (2010) for a further discussion of the differences between the recovery rates of loans andbonds.

6

3 Data

This paper relies on several data sources that we combine to analyze recovery rates in the

US corporate bond market. First, we identify the default events by type, using the Mergent

Fixed Income Securities Database and NYU Salomon Center Master Default Database. These

databases provide all Chapter 11 filings, distressed exchanges and downgrades to a default rating

grade.5 These events cover basically the whole spectrum of defaults, i.e., formal bankruptcy

filings to informal unlikely-to-pay events. Overall, we observe 1, 270 default events for 534 firms

for the time period July 2002 to October 2010. Table 1 presents the list of event types and their

definitions. Chapter 11 filings represent formal bankruptcy procedures handled by federal courts,

i.e., when a firm is unable to service its debt or repay its creditors, it or its creditors can file

with a federal bankruptcy court for protection under Chapter 11. A trustee can act as debtor

in possession, and thus, operate the business. Chapter 11 filings can be used to restructure the

debt or liquidate the assets. The second type of default event are distressed exchanges, in which

the debtor attempts to avoid formal bankruptcy by proposing a fundamental change in existing

contractual commitments to its creditors. Thus, the creditors can voluntarily agree to avoid

potential costs arising in a formal restructuring. Distressed exchanges became popular in recent

years, particularly since the financial crisis. Altman and Karlin (2009) report that, between 2002

and 2009, around 41 companies engaged in distressed exchange offers, which affected 299 bonds.

The third type of default event consists of ratings downgrades. We retrieve ratings from Moody’s,

Standard and Poor’s and Fitch. Ratings rank the obligor according to creditworthiness (AAA, AA,

. . . , C, D), where the rating agencies provide different default-classes. The worst rating grade (e.g.,

D) indicates an actual default (payment default on a financial commitment). The second worst

rating grade (e.g., C) is meant for highly speculative obligations that are considered as unlikely-

to-pay. This stage is often considered as being already a default event in many regulations (see,

for example, the definitions of default in the Basel II standards).

The second important data set we use is obtained from the TRACE database maintained by

the FINRA, which provides transaction information such as prices and volumes for the whole

universe of US corporate bonds.6 In the US corporate bond market, reporting of any transaction

5We exclude Chapter 7 and 15 filings, as we find almost no events where bonds are traded, after these events.Consequently, for these events, the recovery rate estimates can only be based on ultimate recovery, as in e.g. Briset al. (2006).

6The reported trade volume is capped at $1 million for high yield and unrated bonds, and at $5 million forinvestment grade bonds. However, the exact trade volume is released by FINRA after an 18 month delay.

7

to TRACE is obligatory for broker-dealers, and follows a set of rules approved by the Securities

and Exchange Commission (SEC ), whereby all transactions must be reported within a time-frame

of 15 minutes. This data source is rather unique for an OTC market since, in almost all other

cases, price information must usually be obtained either from an individual dealer’s trading book,

which provides a very limited view of the market, or by using bid-ask quotations. We implement

standard filters to exclude potential errors in TRACE.7

We match the default events with the individual bonds affected by the respective event, within

the time window starting 90 days before default, and ending 90 days after the default event.

However, some minimum requirements must be fulfilled for a bond to be included in our analysis.

We do not include bonds with an amount issued smaller than $10 million. We exclude from our

sample bonds with complex structures, mostly related to embedded derivative features, since the

prices of bonds with such payoff structures are quite different, and could potentially bias the

analysis. In particular, therefore, we drop bonds that are rating-sensitive, convertible, sinkable,

extendible, structured, or posses any other kind of complex optionality. Thus, the bonds included

for our analysis are either straight bonds, or simply puttable or callable. Matching the TRACE

data-set with the default events results in 2, 235 event/bond combinations, covering 818 bonds

issued by 259 firms, and account for approximately 1, 734, 000 trades, with an aggregate volume

of $500 billion.

We add bond and firm characteristics from Bloomberg, covering the amount issued, maturity,

coupon, industry and seniority level, where we group the bonds into four main categories of

seniority: guaranteed, secured, unsecured and subordinated. Furthermore, we match the data-set

with data from Markit.8 This allows us to identify bonds that can be delivered to settle credit

default swaps (CDS). In addition, we match the data-set with balance sheet and income statement

information obtained from Compustat. This permits us to analyze the effect of balance sheet ratios,

motivated by various models for recovery rates. We retrieve macroeconomic data covering interest

rates (US Federal Funds rate and Treasury yields) from Bloomberg, to assess the impact of overall

economic conditions on the level of recovery rates. The final combined data-set allows to assess

the impact of several groups of variables on the level of recovery rates in a comprehensive manner.

7Dick-Nielsen (2009) provides an extensive description of possible reporting errors, and its implications forliquidity analysis. Such errors include (i) trade corrections within the same day, (ii) trade cancelations within thesame day, and (iii) reversals across days, i.e., a mistake that had not been detected on the traded day. Furthermore,we implement price filters eliminating potentially erroneous reported prices.

8Markit provides consensus valuations of credit default swaps (CDS) across different maturities and restructuringclauses.

8

4 Research Questions and Hypotheses

In this section, we discuss the research questions we address and the hypotheses that we test in

this paper. In particular, we consider the underlying trading activity of defaulted bonds and focus

on the potential effects of bond characteristics, firm fundamentals, macroeconomic indicators, and

liquidity measures on the level of recovery rates.

The microstructure of the trading activity of defaulted bonds allows us to analyze interesting

research questions relating to how bond prices, trading volumes and number of trades evolve in

different stages of default, and how a reliable market-based recovery rate can be estimated. We

examine the trading activity of the defaulted bonds 90 days before, at, and 90 days after the

observed default event date. In particular, we search for a “grace period” after the default event,

during which prices are mainly driven by the effects of the default event itself, and test whether

the trading activity levels are significantly different before and after this window. Furthermore, we

analyze the trading microstructure of various sub-samples based on industry, rating and default

event type and compare the trading activity of the defaulted bonds with previous studies of non-

defaulted bonds. It turns out that a grace period of 30 days is optimal in the definition of recovery

rates, i.e., we use the average traded price of a bond between the default day and the next 30 days

after default to compute its recovery rate (see Section 6).

In the main part of our analysis, we explore the time-series and cross-sectional variations of

the recovery rates along various dimensions. First, we focus on three aspects that were found to

be of importance in the previous literature: default event type, industry and seniority. Starting

with the default event type, we cover the full range of default events from formal bankruptcy

to informal unlikely-to-pay events. We test the hypothesis that formal procedures are a sign

of more severe economic problems within a firm, and lead to higher costs to bondholders than

informal procedures. Therefore, we anticipate that Chapter 11 filings have lower recoveries than

distressed exchanges and rating defaults. Furthermore, we expect that default ratings have lower

recoveries than unlikely-to-pay ratings. As for industry affiliation, we would expect that within

non-financial industries, utility and energy firms should recover more than other industries as

reported by various studies (see Section 2), due to their higher proportions of tangible assets.

Similarly, among financial firms, commercial banks should recover more than investment banks,

possibly because of their larger holdings of liquid assets. As for the seniority of the bonds, we

9

hypothesize that the greater the seniority and collateral value of their assets, the higher the

recovery rate.

Going beyond these simple dimensions, we analyze the effects of bond characteristics, firm

characteristics, macroeconomic variables and liquidity variables on recovery rates. The potential

effects of bond characteristics, such as amount issued, maturity, coupon, rating grade one year

before default and CDS availability, on recovery rates address interesting research questions. In

particular, we conjecture that larger bond issues recover more, since they offer greater liquidity,

and bonds with longer maturities recover less, since bonds with a maturity of over ten years are

often held by buy-and-hold investors, such as insurance companies, and are either illiquid or are

sold in large blocks upon default. We expect the coupon rate to be positively related to the

recovery rate, since a higher coupon is of value to the bondholders for certain outcomes of the

default event. Regarding the rating grade one year before default, we hypothesize that the lower

the rating grade, the lower would be the recovery rate. Furthermore, we expect a higher recovery

rate if the bond is deliverable into a CDS contract, as it may generate greater demand upon

default, from protection buyers, compared to non-insurable bonds.

The characteristics of the firm are likely to determine the level of recovery rates. We hypothesize

that the value of equity and the default barrier affect recoveries as suggested by structural models

of credit risk: the lower the equity and the higher the default barrier, the lower the recovery of

debtholders, given a certain drop in the firm’s asset value, triggering default. Furthermore, we test

whether the earnings, tangible assets, receiveables, or firm size, before the default event, matter.

Macroeconomic variables, such as aggregate default rates and information based on interest

rate curves, are generally expected to have a significant impact on the level of recovery rates, since

both are indicators of the overall economic conditions. In particular, we expect that a high level of

the (aggregate) default rate in the overall economy signals that the economic conditions are poor

and, thus, could lead to lower recovery rates for individual firms. Similarly, when the (short-term)

interest rates are low, the economy is at the lower end of the business cycle with lower recovery

rates. We also investigate the impact of the slope of the interest rate term structure on recoveries.

Furthermore, our detailed data-set allows us, for the first time, to estimate liquidity measures,

such as trading activity variables (volume and number of trades) and transaction costs measures

(Amihud measure and price dispersion measure) for defaulted bonds. We test the hypothesis that

less liquid bonds have lower recovery rates. We expect that the liquidity effects on prices that have

10

been extensively documented in the literature on non-defaulted bonds (see, e.g., Bao et al. (2011),

Dick-Nielsen et al. (2012) and Friewald et al. (2012)) would be exacerbated following default.

5 Methodology

This section outlines the general approach to measuring the determinants of recovery rates

in the US corporate bond market. We present here our definitions of the recovery rate and the

various types of bond characteristics, firm fundamentals, macroeconomic variables and liquidity

measures, which are used to explain the level of bond recoveries (see Section 4). We also present

the panel data regression setup that we use in our analysis.

5.1 Recovery Rate

The recovery rate π of a bond i issued by firm j is defined in our analysis as the mean of the

transaction prices p, per 100 of face value, between the default day t and the subsequent T = 30

days after default. If Ki,j,s is the number of trades of bond i of firm j on day s indexed by ki,j,s,

then

πi,j,t =1

T + 1

t+T∑s=t

1

Ki,j,s

∑ki,j,s

ps,ki,j,s

(1)

Thus, this specification of the recovery rate suggests that the level of πi,j,t can be interpreted

as what a buyer (seller) would have to pay (receive), on average, and hence, in expectation,

given that a default event occurred, and given that the transaction takes place within the time

window between the default day and the next 30 days after default. It should be noted that the

accrued interest is set to zero, as most defaulted bonds are traded flat, i.e., without exchange of

accrued interest; thus, all prices under investigation are “clean“ rather than “dirty“ prices. The

specification presented above represents a market-based definition of the recovery rate, in which

a certain grace period is considered. We will further elaborate on our definition based on the

analysis of the transaction data in Section 6.

11

5.2 Bond Characteristics

We use a set of bond characteristics to explain differences in the recovery rates of corporate

bonds. The most basic information available about a bond consists of its amount issued, maturity

and coupon. In addition, we consider the seniority level, which is, of course, very important when

analyzing recovery rates. Specifically, in this paper, we use four different levels of seniority: (i)

guaranteed, (ii) secured, (iii) unsecured and (iv) subordinated.9

Bond ratings from Fitch, Standard and Poor’s and Moody’s, one year before the default events

are retrieved and are mapped to natural numbers, e.g., AAA = 1, AA+ = 2, . . . , D = 21. With

this data, we can analyze whether the rating grade before the default event is of importance in

the determination of the recovery rate, i.e., we can compare “expected“ vs. “unexpected“ credit

events. Additionally, we collect the information about whether the bond is deliverable into a CDS

contract and, hence, is insurable in the CDS market. We consider this a bond-specific event,

since only a selected list of bonds of a particular firm can be delivered into its CDS contract. For

example, if only CDS for unsecured debt are traded for a firm, subordinated bonds cannot be

insured.

5.3 Firm Fundamentals

We employ certain firm characteristics in our analysis. First, we use the industry in which the

firm operates as an important characteristic. Second, we use balance sheet and income statement

information as explanatory variables, which is available for the fiscal year prior to the default

event. We use the following six accounting ratios, which are directly motivated by structural

credit risk models (see Section 2):

Equity =Market Value of Equity

Total Assets(2)

Default Barrier =Short Term Debt + 1

2Long Term Debt

Total Assets(3)

LTD Issuance =Long Term Debt

Total Debt(4)

Profitability =EBITDA

Total Sales(5)

9Note that, for some bonds, the financial data venders provide finer classifications of seniority. However, thesefour classes are important and relevant for all bonds, with data being generally available for almost all bonds.

12

Intangibility =Intangible Assets

Total Assets(6)

Receiveables =Total Receivables

Total Assets(7)

We use the value of equity over total assets as a general indicator of the financial condition

of the firm.10 The value of equity is used in many structural credit risk models to infer the asset

value of the company and also to define the leverage. Furthermore, we use the default barrier as

defined by Moody’s KMV and widely used in structural credit risk modelling, i.e., in assessing

the distance to default measure of firms. In addition, we define LTD issuance as the ratio of long

term debt to total debt, since long-term debt is regarded as a more stable funding source and less

likely to cause default in the short run. We measure profitability using the EBITDA as motivated

by structural models based on cashflows. In addition, we use intangible assets and receivables

over total asset as ratios that are indicative of the potential irrecoverable proportions of assets.

Finally, we use total assets and number of employees as size proxies for firms.

5.4 Macroeconomic Variables

We consider three different macroeconomic indicators: the aggregate (market-wide) default

rate, the Federal Funds rate and slope of the term structure of interest rates. The aggregate default

rate, at time t, is defined as the fraction of defaulted bonds in relation to the total outstanding

bonds in the whole US corporate bond market, in the time interval between day t, and T = 90

days before t.

Aggregate Default Ratet,T =Defaulted Bondst,t−T

Outstanding Bondst,t−T

(8)

We consider the aggregate default rate on the default event day in our analysis. In addition, we

consider the Federal Funds rate on the default event day as the relevant short term interest rate,

to avoid issues of default risk and illiquidity, particularly after the financial crisis. We define the

slope of the yield curve on the default event day as the difference between the Federal Funds rate

and the ten year US Treasury yield.11

10In a few cases, we replace the market value of equity by the book value of equity, when reliable data were notavailable for the former.

11We tried alternative measures of the short term interest rate and the slope of the term structure as explanatoryvariables. However, the results are basically identical and we report only the results of the definitions above.

13

5.5 Liquidity Measures

As we have all the necessary transaction data available, we can define various liquidity proxies

that we use as additional explanatory variables. We employ simple trading activity variables, e.g.,

volume and the number of trades, and more sophisticated liquidity measures, e.g., the Amihud

and price dispersion measures, that have been used in literature (see e.g. Dick-Nielsen et al. (2012)

and Friewald et al. (2012)). We use the time window between the default day t and T = 30 days

after this day for the estimation of the measures.

Volume. The volume variable, vi,j,s, is the average transaction volume, per trade, of bond i, of

firm j, within the default day t, and T = 30 days after this day:

vi,j,t =1

T + 1

t+T∑s=t

1

Ki,j,s

∑ki,j,s

vs,ki,j,s

(9)

Number of trades. This variable, ni,j,t, is the average number of trades of bond i, of firm j,

between the default day t, and T = 30 days after this day:

ni,j,t =1

T + 1

t+T∑s=t

Ki,j,s (10)

Amihud measure. The Amihud measure (Amihud (2002)) of bond i, of firm j, on day s, given

Ni,j,s observed daily returns r indexed by ki,j,s is defined as:

Amihudi,j,s =1

Ni,j,s

∑ki,j,s

|rki,j,s|

vki,j,s

(11)

This measure, based on Kyle (1985), originally designed for limit order markets, assesses the

price impact of the traded volume, and hence the depth of the market. Intuitively, a market is

considered illiquid, if a low transaction volume causes relatively large price changes.

14

Price Dispersion. Similar to Jankowitsch et al. (2011) and Friewald et al. (2012), we define the

price dispersion, di,j,s, of bond i, of firm j, on day s, as:

di,j,s =

√√√√ 1∑ki,j,s

vki,j,s

·∑ki,j,s

(pki,j,s

mi,j,s− 1

)2

· vki,j,s(12)

where mi,j,s is the mean transaction price representing the fair value of the bond, and pi,j,s are

the individual trade prices. The (volume-weighted) volatility of the individual trades around the

fair value permits a direct estimation of transaction costs based on transaction data. The intuition

behind this measure is motivated by market microstructure models: a low dispersion of traded

prices around its market wide valuation indicates that the bond can be bought or sold close to its

fair value and, thus, at lower transaction costs, indicative of a more liquid instrument.

5.6 Panel Data Regression

We rely on a panel data regression approach to analyze the determinants of recovery rates

in the US corporate bond market. As motivated by the discussion in the previous section, the

recovery rate π of bond i, issued by firm j, for default at day t, is assumed to be given by:

πi,j,t = α+ β · (Bond Characteristics)i,j + γ · (Firm Fundamentals)j,t (13)

+ φ · (Macroeconomic Indicators)t + δ · (Liquidity)i,t

+ λ · (Default Event Type)i,j,t + µ · (Industry)j

+ ζ · (Seniority)i,j + εi,j,t

This specification is tested on a pooled data-set that combines the entire time-series and the

cross-section of recovery rates. We use ordinary least squares regressions adjusting the standard

errors for the existence of default event-firm clusters as described in Williams (2000) and Petersen

(2009). This approach addresses the issue that in a particular default event a firm may have

several bonds outstanding, and that all these defaulted bonds show up as separate observations in

our data. In addition, all our regressions include the default event, industry and seniority dummy

variables.

15

6 Results

6.1 Descriptive Analysis

This section analyzes the underlying trading activity of defaulted bonds in the US corporate

bond market and presents descriptive statistics for the resulting recovery rates. We first explore

the traded prices and volumes on the default day, and in the 90 days windows before and after

default. Focusing on the recovery rate itself, we analyze its empirical distribution and quantify the

effects of the default event type, industry and seniority on the recovery rates. We also document

the variation in recovery rates over time. In addition, we provide summary statistics for the

explanatory variables that are used in the panel data regressions.

6.1.1 Trading Microstructure of Defaulted Bonds

In this section, we analyze the underlying trading activity of defaulted bonds. Figure 1, Panel

A, provides the evolution of mean transaction prices per day as a percentage of face value, for a

time window starting 90 days before, and ending 90 days after the default event day, across all

default event/bond combinations. In addition, Panel B shows the mean number of trades and

trading volume in defaulted bonds per day. By investigating transaction prices, we find that the

lowest mean price is observed on the default day itself, and is around 35% of face value. The price

level 90 days before the default is already low and shows a declining trend from about 57% to

45%. However, the default event day witnesses a significant drop in prices and is, thus, not fully

anticipated by the market. Interestingly, we find that the transaction prices recover steadily to

42%, in the ensuing 30 days after default, whereas the prices after 30 days show a more volatile

evolution. Using the Kolmogorov-Smirnov test, we find that the transaction prices within the 90

day window before default are significantly different from those on the default day itself, and in

the subsequent 30 days, which are, in turn, significantly different from those in the time frame of

31 to 90 days after default.

The analysis of the mean number of trades and traded volume across all default event/bond

combinations exhibits interesting patterns. In particular, the average number of trades per bond

on the default day of around 35 is significantly higher than on all other days. This number of

trades is also remarkably high, compared to the market-wide average of the whole corporate bond

market of 3 to 4 trades per day per bond (see, for example, Friewald et al. (2012)). The number of

16

trades decreases rapidly in the subsequent 30 days after default to around 8 to 10 trades per day

per bond, which is still higher than the market-wide average. The average daily traded volume

per bond is around $10 million on the default day, and decreases to the same extent as the number

of trades to about $3 million. Again, the traded volume on the default day is higher than the

market-wide average of around $5 million, see e.g. Friewald et al. (2012). Overall, we find a lower

price and higher level of trading activity on the default day. This price on the default day recovers

quickly in the next 30 days, reaching a stable price level, with the trading activity returning to

pre-default levels thereafter. Based on these findings, we define the recovery rate in our analysis

as the mean transactions price in the window between the default date, and 30 days after default,

as we find evidence that the price discovery in this time window is driven mostly by the default

event itself. Furthermore, this time window represents a reasonable period to split up and sell

larger positions in defaulted bonds. We consider this estimation of a market-based recovery rate

more reliable than quotations or last-trade information from the default date alone, as has been

used by many prior studies due to data limitations. This significant methodological distinction

makes our subsequent analysis all the more robust.

Figure 2 further investigates the trading activity among four important sub-groups: for non-

financial versus financial firms, and investment versus speculative grade bonds, we again analyze

the mean transaction prices, number of trades and traded volume in the 90 days before and after

the default event. The general patterns obtained in the previous analysis of the full sample are

confirmed in each sub-group: first, the lowest transaction price is reported on the default day itself.

Second, trading activity is especially high on the default day and gradually declines afterwards.

By comparing non-financial and financial firms, we find that for non-financial firms, the price

decline leading towards the default day is smoother, and the price decline on the default day itself

is less severe (from 43% to 35%, compared to 47% to 33% for financials), indicating that the actual

default is more of a surprise to the market, in the case of financial firms. For both groups, the

number of trades and the traded volume are especially high on the default day. In addition, the

general level of trading activity around default seems to be higher for financial firms than for non-

financial firms, e.g., the mean number of trades is around 22 for non-financial firms, compared to

around 70 for financial firms, on the default day. The comparison of investment versus speculative

grade bonds yields interesting insights, as well. While the mean transaction prices for speculative

grade bonds decline gradually towards default, prices drop rather steeply on the default day in the

17

case of investment grade bonds (from 42% to 35% compared to 47% to 27%). This might indicate

a greater surprise element in the case of default for more credit-worthy, investment grade bonds.

The number of trades and traded volumes are higher for investment grade bonds, as expected.

Figure 3 presents mean transaction prices 90 days before, and after, default for the different

default event types. Transaction prices for Chapter 11 liquidation and restructuring filings exhibit

very similar patterns. In both cases, the default day induces a relatively sharp decline in prices

from about 50% down to 25%. Within the first 30 days after default, prices recover to around

40%. Especially interesting is the analysis of mean transaction prices in the case of distressed

exchanges. The pattern reveals that these cases are the only default events where transaction

prices before default are lower than after default, indicating that the default itself is seen as a sign

of relief by the market, after an uncertain negotiation process. In particular, distressed exchanges

exhibit the highest transaction prices in the post-event phase. For rating-based default events,

we find that unlikely-to-pay announcement events by all rating agencies lead to a sharp drop of

prices, indicating an element of surprise, whereas the event of downgrading to an actual default

rating class seems to be generally anticipated by the market. The analysis of the number of trades

and traded volumes for different default event types yields very similar results, compared to the

previous analyses, and is, therefore, not presented in the interest of conserving space.

6.1.2 Recovery Rates and Effects of Event Type, Industry and Seniority

Analyzing the resulting recovery rates, i.e., the mean transaction prices between the default

day, and 30 days after default, we first present the empirical distribution of the recovery rates

of defaulted US corporate bonds between 2002 and 2010, in Figure 4. The mean recovery rate

is equal to 38.6% with a standard deviation of 27.4%. While the mean recovery rate is close to

the 40% estimate provided by Altman and Kishore (1996), which is widely used in academia and

industry, the standard deviation around this number suggests substantial variation in recovery

rates across different dimensions; therefore, a comprehensive analysis of the driving factors is

important. Specifically, three peaks can be identified in the empirical distribution - one up to

20%, one between 40% to 50% and one between 60% to 70%. The lowest peak is likely to have

been driven by the recovery rates of bonds issued by Lehman Brothers, which traded at about

15%, after filing for protection under Chapter 11 on September 15, 2008. Overall, the distribution

documents the stochastic nature of the recovery rate.

18

Figure 5 displays the time series of mean recovery rates in the US corporate bond market as a

quarterly moving average. We find that recovery rates are highly volatile over time: around 60%

in 2007, compared to 20% at the end of 2008. The lowest mean recovery rates can be found during

the financial crisis. Thus, cross-sectional average recovery rates are clearly not constant over time

and should be used with caution.

We present summary statistics in Table 2 displaying the recovery rates across different default

event types, industries and seniority levels, to analyze the important determinants of recovery

rates. We report, in total, 2, 235 default event/bond combinations for which transaction data are

available. Panel A displays the statistics for the overall sample and confirms the results discussed

when presenting the empirical distribution. Panel B displays recovery rates across different default

event types. Chapter 11 restructuring filings form the largest group, consisting of 492 observations,

whereas we observe only 13 Chapter 11 liquidations. Interestingly, we find only an insignificant

difference between Chapter 11 restructuring and liquidation filings (i.e., 37.1% vs. 40.7%). Dis-

tressed exchanges exhibit, on average, the highest recovery rate of 51.3%, confirming that avoiding

formal bankruptcy procedures leads to significantly higher recovery rates, by preserving more of

the “going concern” value for bondholders. Furthermore, we find significant differences in recovery

rates, when analyzing the default grades of the rating agencies. We find, within the Fitch and

Moody’s credit classifications, that the actual default grade has a significantly lower recovery than

the unlikely-to-pay grade (Fitch: 31.4% vs. 41.3% and Moody’s: 16.0% vs. 44.9%). This differ-

ence is pronounced, especially in the case of Moody’s, indicating that their rating framework is

indeed more sensitive to the expected loss, as is generally assumed by many market participants.

For Standard and Poor’s such a difference is not observable, which could indicate that Standard

and Poor’s does not incorporate recovery aspects, as suggested by their rating framework.

Panel C displays recovery rates across non-financial industries, while panel D reports recovery

rates for financial firms. One should note that the sample is fairly balanced between non-financial

and financial firms (1, 160 observations belong to non-financial firms, while 1, 075 belong to fi-

nancial firms). We find that among non-financial industries, utility and energy firms recover, on

average, the highest (e.g., electricity 48% and oil & gas 44.4%), while retail firms recover the

least at 33.4%. Among financial industries, we find that the overall highest recovery rate of 56.6%

is reported for the credit & financing industry, while the financial services industry exhibits the

lowest recovery rates - this result is mainly driven by the low recoveries of Lehman Brothers debt.

19

The averages across financial firms (38.8%) and non-financial firms (38.5%) are almost identical,

while the standard deviations are high in every industry group.

Panel E displays the average recovery rates across seniority levels. As expected, secured bonds

recover, on average, the most (around 49.3%), while bonds that are subordinated recover, on av-

erage, the least (around 15.1%). Interestingly, we find only a small difference in recovery rates

between guaranteed (40.3%) and unsecured bonds (39.1%). This can arise possibly due to guar-

antees provided to bonds of subsidiaries by the holding company, making the guarantee worthless

in case of default of the holding company.

Figure 6 displays the average recovery rates by rating class, one year before default. Interest-

ingly, no consistent pattern emerges from an inspection of the figure. Surprisingly, bonds with

A+, A or A− ratings, one year before default, have rather low recoveries, following default. Thus,

rather unexpected events, which cause erstwhile highly rated firms to default, seem to result in

lower recoveries. However, when only rating grades from BBB+ to C are considered, a slight

tendency for lower recoveries for lower rated bonds can be discerned.

6.1.3 Summary Statistics of Explanatory Variables

The previous analysis shows a pattern of significant cross-sectional and time-series variation

of recovery rates, which might be explained by a more detailed analysis of a comprehensive set of

explanatory variables.12 Table 3 reports the summary statistics for the main explanatory variables

in our empirical analysis, covering bond characteristics, firm fundamentals and liquidity proxies.

Table 4 provides the same statistics, for the sub-samples of non-financial and financial firms. We

first discuss the results of the full sample and then highlight the differences between non-financial

and financial firms.

Panel A of Table 3 summarizes the results for the bond characteristics: the average issue size

of defaulted bonds is $400 million. The average maturity and coupon rate are 6.82 years and

7.48% respectively. The average bond rating one year before default is BB, i.e., most of defaulting

bonds are from the speculative grades. All these variables show considerable variation, e.g., the

standard deviation of the credit rating is five notches. Interestingly, we find that for financial

firms, the average bond rating is A− (investment grade), while for non-financial firms, the average

bond rating is B− (speculative grade), indicating that the default of a financial firm is often not

12See Section 5 for a detailed description of the explanatory variables.

20

considered as very likely by rating agencies one year before the actual event, whereas for non-

financial firms, the economic situation of the company is already perceived as being weak. A

similar difference with a parallel reasoning can be found for the coupon rate (financial firms: 5.8%

vs. non-financial firms: 8.6%). Furthermore, defaulted financial bonds have, on average, a longer

maturity by 2.5 years.

Panel B in Table 3 presents statistics for firm fundamentals. On average, firms have an equity

ratio of 6.62% of total assets. The average default barrier equals 47.80% of total assets. Comparing

financial and non-financial firms in Table 4, we find that there is only a small difference in the

equity ratio between these two groups. Interestingly, the default barrier is higher for financial

firms (around 53%) than for non-financial firms (around 23%), indicating that the former use more

short-term financing. On average, receivables are 50% of total assets for financial firms, but only

10% are for non-financial firms. There is also a huge gap in intangibility: for non-financial firms,

intangibility is ten times as high as for financial firms. Despite being close to default, financial

firms seem to be highly profitable, on average, with a profitability of around 60%. Analyzing the

size proxies, the average firm size is equal to $139 billion with 2,970 employees, respectively. While

the average firm size of defaulted financial firms is ten times as large as that of non-financial firms,

the latter have 2.2 times as many employees as financial firms.

Panel C in Table 3 summarizes the trading activity variables and liquidity measures. The

trading activity variables confirm that the number of trades and trading volume are above the

market-wide average and, thus, the market for defaulted bonds is not an obviously illiquid segment

of the corporate bond market for all measures. Interestingly, bonds of financial firms have five

trades per day on average more compared to non-financial firms. However, the volume per trade

is much lower (around $260,000 vs. $470,000). Analyzing the liquidity measures, we find that

trading in defaulted bonds results in relatively high transaction costs. Thus, defaulted bonds are

illiquid in this sense. The average price impact given by the Amihud measure of $1 million is

1.49%, while the market-wide average is at 0.36% (see Friewald et al. (2012)). Transaction costs

amount to 2.80%, estimated by the price dispersion measure, which is six times as high as the

overall market average of 0.43% (see Friewald et al. (2012)).

Figure 7 provides the time series evidence for the key macroeconomic variables, i.e., the aggre-

gate default rate, the Federal Funds rate and the 10 year Treasury rate (used in the slope variable).

Analyzing the default rate, three regimes can be identified: the “dotcom“ bubble in 2002 and 2003,

21

the Ford and General Motors crisis in 2005 and the financial crisis in 2008 and 2009.13 The period

during the financial crisis experienced the highest aggregate default rate, around 3.60%. We find

a significant variation in default rates during our observation period, allowing us to analyze the

relation between recovery and default rates. The Federal Funds rate shows similar patterns, but

in the opposite direction, i.e., low rates in crisis periods, and high rates in the boom phases. The

difference between the 10 year Treasury rate and the Federal Funds rate shows large differences

at the end of crisis, and low or negative differences at the beginning of crisis, whereas the 10 year

rate itself is rather stable over time, albeit with a decrease since the financial crisis, due to central

bank intervention, through quantitative easing.

6.2 Panel Data Analysis

In this section, we present the results of the panel regressions. The recovery rates are explained

by bond characteristics, firm fundamentals, macroeconomic variables and liquidity proxies, as well

as dummy variables for the default event types, industries and seniority classes. The results of

this model specification are presented in the next section. Additionally, we provide an alternative

empirical analysis. In particular, we test the model on two subsamples, i.e., for non-financial and

speculative grade rated bonds, to analyze whether our results are driven by either financial firms

or by firms that were hitherto investment grade, but defaulted subsequently.14

6.2.1 Regression Model Explaining Recovery Rates

In this section, we present the results of our regression analysis. Table 5 presents six different

regression specifications. Model 1 represents a regression including only the dummy variables

for the default event types, industries and seniority levels. This specification allows to explore

the increase in explanatory power compared to the other specifications. We find an adjusted R2

of 37% for Model 1, i.e., reasonable explanatory power can be found even for this specification,

as important dimensions are already included and the results of the descriptive analysis can be

confirmed. The next four specifications, i.e., Models 2 to 5, control for each of the four defined

groups of variables, i.e., bond characteristics, firm fundamentals, macroeconomic variables or

13See Friewald et al. (2012) for a related analysis on the different regimes in the US corporate bond market.14In addition, we perform a factor analysis and identify five factors representing a balance sheet factor, a size

factor, a macroeconomic factor, a trading activity factor and a transaction costs factor, and run the regressions(joint model and subsamples) afresh based on these factors to avoid potential multi-collinearity. The results basedon the factors confirm the overall findings (sign and significance) and are, thus, not presented in detail.

22

liquidity proxies. We find that all four groups add to the explanation of recovery rates. Bond

characteristics (Model 2) increase the adjusted R2 by five percentage points to 42%. However,

these characteristics are not as important as the other groups. Firm fundamentals (Model 3) and

macroeconomic variables (Model 4) seem to be of similar importance, as they exhibit adjusted R2

of 47% and 46% respectively. We obtain the highest adjusted R2 of 53% by including liquidity

measures (Model 5). Thus, the trading activity and liquidity measures are important additional

variables necessary to explain recovery rates.

Model 6 in Table 5 includes all four sets of variables. We focus on this complete model to

discuss the effect of the individual variables. We find that this model is able to capture 64%

of the variation of recovery rates. Among the bond characteristics, four variables turn out to

be significant. As expected, we find a positive association between the amount issued and the

recovery rate, i.e., bonds with higher amounts outstanding trade at favorable prices post-default.

A $10 million increase in amount issued is associated with a 2.85% increase in recoveries per

100% of face value, which represents an economically significant effect. We find that bonds with

a longer maturity exhibit lower recoveries, i.e., an increase in the time to maturity by one year

decreases the recovery rate by around 0.6% of face value. Thus, this effect has rather low economic

impact, and might be caused by selling-pressure imposed by large institutional investors, such as

insurance companies, which typically hold bonds with longer time to maturity, but may be forced

to sell following default, due to mandate restrictions. Furthermore, we find a small positive effect

for the coupon rate, potentially indicating that bonds with a higher coupon are of higher value

for certain outcomes of the default event. Interestingly, the rating variable is insignificant in

the complete model, indicating that the rating one year before the default event conveys little

information concerning the recovery rate. This result is consistent with the findings of Altman

and Kishore (1996). The most interesting result among the bond characteristics is provided by

the CDS dummy, indicating whether a bond can be delivered into a CDS contract. In particular,

we find that bonds that are deliverable into a CDS contract exhibit around 6.70% higher recovery

rates of face value than bonds that are non-deliverable. This effect is quite significant in economic

terms, and may arise due to the potential buy-side pressure of protection buyers, who are obliged

to deliver certain bonds to protection sellers in the case of default.

Among the firm characteristics, we find significant effects for the ratios motivated by structural

credit risk models. We find that the higher the equity value and the lower the default barrier, the

23

higher the recovery. In particular, we find roughly similar partial net effects of these two ratios,

i.e., an increase in the equity ratio and a decrease in the default barrier by ten percentage points,

increases the recovery by around 1.50% and 1.80% respectively. In addition, we find that firm size

matters, i.e., a $100 billion increase in total assets leads to an increase in recoveries by approx-

imately 3.60%. The other firm characteristics employed (long term debt issuance, intangibility,

receivables, profitability, and employees) are statistically insignificant in the joint model. Thus,

the information from these characteristics might be already contained in the industry dummies.

The third group of explanatory variables are macroeconomic characteristics, of which the most

important one is the aggregate (market-wide) default rate. Several studies (see e.g., Altman et al.

(2002)) have concluded that aggregate default rates and aggregate recovery rates are negatively

associated. As already mentioned in Section 5, we employ a more precise estimate of the aggregate

default rate; based on the default event date, we derive for each recovery rate a market-wide default

rate in a trailing 90 day window, to be able to measure the contemporaneous interaction effect

between these two variables. In addition, we consider the Federal Funds rate on the default event

date as the relevant short term rate and we explore the slope between the 10 year Treasury rate and

this short term interest rate. All three variables are highly statistically significant. In particular,

we find that high aggregate default rates and low short term interest rates imply lower recoveries.

Thus, as expected, poor overall economic conditions result in lower recovery rates, i.e., an increase

in the market-wide default rate, and a decrease in the short term interest rates by one percentage

point leads to a decrease in recoveries by around 4.60%, and 6.60% respectively. In addition,

we find a positive effect of the slope factor, i.e., in regimes that could be associated with higher

optimism, we observe higher recoveries. Overall, as expected, systematic risk factors such as the

market-wide default risk influences the level of recoveries. The high explanatory power of the

model points to the fact that a significant part of the variation in recovery rates can be attributed

to this market-wide risk factor.

The fourth group of explanatory variables consists of the liquidity measures (volume, number

of trades, Amihud measure and price dispersion measure). We find that liquidity effects are

of particular importance in explaining the variation of recovery rates across different bonds in

default. In particular, the price dispersion measure introduced by Jankowitsch et al. (2011) is

highly significant and exhibits a negative coefficient - indicating that illiquid bonds suffer more of

a decline in the event of default than liquid bonds. In particular, we find that an increase in the

24

price dispersion measure by 100 basis points leads to a decrease in recovery rates by around 5.05%.

The trading activity variable and the volume variable turn out to be insignificant, indicating that

pure trading activity itself has virtually no impact on recovery rates; rather, transaction cost

metrics seem to be of greater importance. Interestingly, the Amihud measure is insignificant and,

thus, fails to provide explanatory power in addition to the price disperion measure.15

Overall, we find important factors driving the recovery rates of corporate bonds following

default. As expected, bond characteristics are of minor importance. However, we document the

strong effect of deliverability (into the CDS contract) on the recovery rate. On the other hand,

firm characteristics motivated by structural credit risk models and macroeconomic variables are

clearly linked to recovery rates. Interestingly, liquidity variables, especially those proxies that

measure transaction costs, are significant factors in explaining recovery rates.

6.2.2 Non-Financial and Speculative Grade Bonds

In this section, we present the results for two important subsamples of our overall data-set. This

analysis allows to validate the results of the previous section and to analyze whether certain results

are potentially driven by financial firms (especially, Lehman Brothers) or by large investment grade

firms. Table 6 provides the results for non-financial (Model 1) and speculative grade bonds (Model

2). We find an adjusted R2 of around 54% for these two subsamples. Analyzing the effects of

the individual variables, we again find similar results, i.e., all groups of characteristics add to the

explanatory power and most variables that are significant in the overall model are again so. Thus,

the main results stay much the same for these sub-groups as the overall sample.

However, some interesting differences compared to the overall sample need to be highlighted.

Among the bond characteristics, the partial effect of maturity (around −0.96%) is more negative

for both non-financial and speculative grade bonds, i.e., longer maturity bonds recover less in

the two subsamples as compared to the full sample (around −0.61%). The amount issued and

coupon are of minor importance, potentially indicating that bonds of these samples are more

similar compared to the full sample, in this respect. Interestingly, the rating variable is significant

for both samples, i.e., a one notch downgrade is associated with approximately a 1.45% decrease

in recoveries for non-financial and speculative grade bonds. Thus, once the low recoveries that

15This result could be triggered by retail trades, which have to bear very high transaction costs due to their smallsize. The price dispersion measure uses a low weight for this observations, whereas the Amihud measure gives aparticular high weight to low volume transactions.

25

we observe in the A+, A and A− rating grade are disregarded, we find the expected result that

a lower rating grade one year before the default indicates lower recovery. The option to deliver

the bond into a CDS contract reveals interesting insights in these two subsamples. While for

non-financial bonds, the possibility of delivering the bond into the CDS contract still results

in significantly higher recovery rates, the coefficient is insignificant for speculative grade bonds,

perhaps indicating that CDS contracts for these firms are often illiquid before an expected default

event.

As for firm characteristics, we find similar results in the subset-regressions, compared to the

overall sample. We find that the partial effects of equity and the default barrier become stronger for

both of these groups, compared to those for the full sample. In addition, receiveables and number

of employees seem to matter for non-financial firms and speculative grade bonds. Specifically, a one

percentage point increase in receivables is associated with an increase in recoveries of around 0.21%

to 0.27% for these two sub-groups. This effect could not be detected in the overall sample, possibly

because of the higher percentage of receivables for financial firms. Additionally, an increase in the

number of employees by 1000 leads to an increase of recoveries for these two groups by around

0.61%.

The significance and directional effects of macroeconomic variables remain basically unchanged

in the sub-samples compared to the full sample. Further interesting insights can be obtained from

the liquidity measures. For the transaction cost measures (both Amihud and price dispersion

metrics), we find results similar to those for the full sample. The effect of the price dispersion

measure is more pronounced in these two subsets, i.e., an increase in transaction costs by 100

basis points is associated with a decrease in recoveries by 7.7% to 8.8%, indicating that illiquidity

effects are of particular importance for non-financial and speculative grade bonds in explaining the

variation across bond recoveries. Interestingly, the trading activity variables (volume and number

of trades) are significant in the subsets as well. The volume variable has a negative impact, whereas

the number of trades has a positive effect. This could indicate that a large trading volume in a

bond with only a few observed trades has a negative impact on its recovery rate following default.

Such an effect cannot be found in the results for the overall sample, as (large) financial firms and

investment grade firms have potentially more active trading after default, even when recoveries

are lower for other reasons (e.g., Lehman Brothers).

26

7 Conclusion

The recovery rate in the event of default is an important risk factor for pricing financial

contracts exposed to credit risk. Many defaults during the recent past have highlighted the

stochastic nature of recovery rates. Therefore, it is important to understand the determinants

of this risk factor in greater detail. In this paper, we analyze recovery rates of defaulted US

corporate bonds, based on a complete set of transactions data over the time period 2002 to 2010.

In particular, we investigate the underlying trading activity for a broad set of default event types

and provide reliable market-based estimates of the recovery rates. The focus of our analysis is

on the relation between these recovery rates and a comprehensive set of bond characteristics,

firm fundamentals and macroeconomic variables. Furthermore, we use measures quantifying the

liquidity effects for individual bonds, an additional innovation relative to the prior literature.

Analyzing the microstructure of the trading activity reveals that the lowest bond prices indeed

occur on default event day itself (around 35% of face value, on average). Interestingly, the trading

activity on this day, measured by volume and number of trades, is quite high in comparison with

non-defaulted bonds. The prices recover to around 42% in the following 30 days after default with

the trading activity still remaining high. Thereafter, prices show a more volatile evolution and

the trading activity dies down quickly. Based on these findings, we define the recovery rate of a

defaulted bond as the average traded price between the default day and the next 30 days after

default, as we conjecture that price discovery in this time window is mostly driven by the default

event itself.

The subsequent panel regression analysis explains 64% of the total variation in recovery rates

employing bond characteristics, firm fundamentals, macroeconomic variables and liquidity mea-

sures as explanatory variables. As expected, we find that the type of default event, the industry

in which the firm operates, and the seniority, are important determinants of the recovery rate.

However, of equal importance are balance sheet ratios motivated by structural credit risk models

and overall macroeconomic conditions. Furthermore, we find that transaction costs metrics and

trading activity variables measuring liquidity are important determinants of recovery rates. In

summary, we provide a comprehensive analysis offering detailed insights into the stochastic nature

of recovery rates and quantify the effects of various endogenous and exogenous factors on recovery

rates.

27

Figures and Tables

Figure 1: Trading Activity in Defaulted US Corporate BondsThis figure shows the mean transaction prices and volumes as well as the average number of trades per bond on thedefault day, and in the time window 90 days before, and after, default across all default event/bond combinations.In Panel A, the evolution of mean transaction prices around default is presented. In Panel B, the average numberof trades per bond and the traded volume per bond on a daily basis is shown. The data-set consists of transactiondata reported by TRACE for the period July 2002 to October 2010 and amounts to approximately 1, 734, 000trades with an aggregate volume of $500 billion covering 2, 235 default event/bond combinations. The defaultevents are obtained from the Mergent Fixed Income Securities Database and the NYU Salomon Center MasterDefault Database and cover bankruptcy filings, out-of-court restructurings and rating downgrades.

3540

4550

55

Days to Default

Pric

e in

% o

f Fac

e V

alue

Panel A

−90 −60 −30 0 30 60 90

05

1015

2025

3035

Days to Default

Num

ber

of T

rade

s / V

olum

e in

$1,

000,

000

Panel B

−90 −60 −30 0 30 60 90

TradesVolume

28

Figure 2: Trading Activity in Non-Financial/Financial and Investment/Speculative Grade BondsThis figure shows the mean transaction prices and volumes as well as the average number of trades per bond on thedefault day, and in the time window 90 days before, and after, default for bonds of non-financial firms and financialfirms, as well as for investment and speculative grade bonds. The top half refers to bonds of financial versus nonfinancial firms, while the bottom half refers to investment versus speculative grade bonds. The data-set consistsof transaction data reported by TRACE for the period July 2002 to October 2010 and amounts to approximately1, 734, 000 trades with an aggregate volume of $500 billion covering 2, 235 default event/bond combinations. Thedefault events are obtained from the Mergent Fixed Income Securities Database and the NYU Salomon CenterMaster Default Database and cover bankruptcy filings, out-of-court restructurings and rating downgrades.

3540

4550

Days to Default

Pric

e in

% o

f Fac

e V

alue

Non−Financial

−90 −60 −30 0 30 60 900

510

1520

25

Days to Default

Num

ber

of T

rade

s / V

olum

e in

$1,

000,

000

Non−Financial

−90 −60 −30 0 30 60 90

TradesVolume

3540

4550

5560

Days to Default

Pric

e in

% o

f Fac

e V

alue

Financial

−90 −60 −30 0 30 60 90

020

4060

Days to Default

Num

ber

of T

rade

s / V

olum

e in

$1,

000,

000

Financial

−90 −60 −30 0 30 60 90

TradesVolume

3040

5060

Days to Default

Pric

e in

% o

f Fac

e V

alue

Investment Grade

−90 −60 −30 0 30 60 90

010

2030

4050

6070

Days to Default

Num

ber

of T

rade

s / V

olum

e in

$1,

000,

000

Investment Grade

−90 −60 −30 0 30 60 90

TradesVolume

3540

4550

Days to Default

Pric

e in

% o

f Fac

e V

alue

Speculative Grade

−90 −60 −30 0 30 60 90

05

1015

2025

Days to Default

Num

ber

of T

rade

s / V

olum

e in

$1,

000,

000

Speculative Grade

−90 −60 −30 0 30 60 90

TradesVolume

29

Fig

ure

3:T

ran

sact

ion

Act

ivit

yac

ross

Def

ault

Eve

nts

Th

isfi

gu

resh

ow

sth

em

ean

tran

sact

ion

pri

ces

acr

oss

the

diff

eren

td

efau

ltev

ent

typ

eson

the

def

au

ltd

ay

an

din

the

tim

ew

ind

ow

90

days

bef

ore

,an

daft

er,

def

au

lt.

Th

ed

ata

-set

con

sist

sof

tran

sact

ion

data

rep

ort

edby

TR

AC

Efo

rth

ep

erio

dJu

ly2002

toO

ctob

er2010

an

dam

ou

nts

toap

pro

xim

ate

ly1,7

34,0

00

trad

esw

ith

an

aggre

gate

volu

me

of

$500

billion

cover

ing

2,2

35

def

au

ltev

ent/

bon

dco

mb

inati

on

s.T

he

def

au

ltev

ent

data

are

ob

tain

edfr

om

the

Mer

gen

tF

ixed

Inco

me

Sec

uri

ties

Da

taba

sean

dth

eN

YU

Sa

lom

on

Cen

ter

Ma

ster

Def

au

ltD

ata

base

an

dco

ver

ban

kru

ptc

yfi

lin

gs,

ou

t-of-

cou

rtre

stru

ctu

rin

gs

an

dra

tin

gd

ow

ngra

des

.2030405060

Day

s to

Def

ault

Price in % of Face Value

Cha

pter

11

Liqu

idat

ion

−90

−60

−30

030

6090

304050

Day

s to

Def

ault

Price in % of Face Value

Cha

pter

11

Res

truc

turin

g

−90

−60

−30

030

6090

45505560

Day

s to

Def

ault

Price in % of Face Value

Dis

tres

sed

Exc

hang

e

−90

−60

−30

030

6090

305070

Day

s to

Def

ault

Price in % of Face Value

Fitc

h D

−90

−60

−30

030

6090

20304050

Day

s to

Def

ault

Price in % of Face Value

Moo

dy's

C

−90

−60

−30

030

6090

40455055

Day

s to

Def

ault

Price in % of Face Value

Sta

ndar

d an

d P

oor's

D

−90

−60

−30

030

6090

40455055

Day

s to

Def

ault

Price in % of Face Value

Fitc

h C

−90

−60

−30

030

6090

42465054

Day

s to

Def

ault

Price in % of Face Value

Moo

dy's

Ca

−90

−60

−30

030

6090

455055

Day

s to

Def

ault

Price in % of Face Value

Sta

ndar

d an

d P

oor's

C

−90

−60

−30

030

6090

30

Figure 4: Recovery Rates in the US Corporate Bond MarketThis figure shows the empirical distribution of recovery rates of defaulted US corporate bonds. Recovery rates aredefined as the average transaction price per bond within 30 days after default. The data-set consists of transactiondata reported by TRACE for the period July 2002 to October 2010 and amounts to approximately 1, 734, 000 tradeswith an aggregate volume of $500 billion covering 2, 235 default event/bond combinations. The default events dataare obtained from the Mergent Fixed Income Securities Database and the NYU Salomon Center Master DefaultDatabase and cover bankruptcy filings, out-of-court restructurings and rating downgrades.

Recovery Rate

Den

sity

0 20 40 60 80 100 120

0.00

00.

005

0.01

00.

015

Figure 5: Recovery Rates across TimeThis figure shows the time series of mean recovery rates (quarterly moving average) in the US corporate bondmarket. Recovery rates are defined as the average transaction price per bond within 30 days after default. Thedata-set consists of transaction data reported by TRACE for the period July 2002 to October 2010 and amountsto approximately 1, 734, 000 trades with an aggregate volume of $500 billion covering 2, 235 default event/bondcombinations. The default events data are obtained from the Mergent Fixed Income Securities Database and theNYU Salomon Center Master Default Database and cover bankruptcy filings, out-of-court restructurings and ratingdowngrades.

2002 2004 2006 2008 2010

2030

4050

6070

8090

Time

Rec

over

y R

ate

31

Figure 6: Recovery Rates across Rating ClassesThis figure shows the mean recovery rates across rating classes (one year before default). Recovery rates are definedas the average transaction price per bond within 30 days after default. The data-set consists of transaction datareported by TRACE for the period July 2002 to October 2010 and amounts to approximately 1, 734, 000 tradeswith an aggregate volume of $500 billion covering 2, 235 default event/bond combinations. The default events dataare obtained from the Mergent Fixed Income Securities Database and the NYU Salomon Center Master DefaultDatabase and cover bankruptcy filings, out-of-court restructurings and rating downgrades.

A+ A A− BBB+ BBB BBB− BB+ BB BB− B+ B B− CCC+ CCC CCC− C

Rating Class

Mea

n R

ecov

ery

Rat

e

010

2030

4050

Figure 7: Aggregate Default Rate, Federal Funds Rate and Treasury YieldThis figure shows the time series of the aggregate default rate (quarterly moving average), which is given as thefraction of defaulted bonds and outstanding bonds in the US corporate bond market, the Federal Funds rate andthe ten year US Treasury yield for the time period July 2002 to October 2010. The data-set consists of transactiondata reported by TRACE for the period July 2002 to October 2010 and amounts to approximately 1, 734, 000 tradeswith an aggregate volume of $500 billion covering 2, 235 default event/bond combinations. The default events dataare obtained from the Mergent Fixed Income Securities Database and the NYU Salomon Center Master DefaultDatabase and cover bankruptcy filings, out-of-court restructurings and rating downgrades. The Federal Funds rateand the Treasury yield are retrieved from Bloomberg.

2002 2004 2006 2008 2010

01

23

45

6

Time

Agg

rega

te D

efau

lt R

ate

/ Fed

eral

Fun

ds R

ate

/ 10

Year

US

Tre

asur

y Y

ield

Aggregate Default RateFederal Funds Rate10 Year US Treasury Yield

32

Tab

le1:

Def

ault

Eve

nts

Th

ista

ble

list

sth

ed

iffer

ent

def

au

ltev

ent

typ

esu

sed

inth

ean

aly

sis

an

dth

eir

defi

nit

ion

s.W

eco

nsi

der

thre

ecl

ass

esof

def

au

ltev

ents

:fi

lin

gs

for

pro

tect

ion

un

der

Ch

ap

ter

11

(res

tru

ctu

rin

gan

dliqu

idati

on

),d

istr

esse

dex

chan

ges

an

dra

tin

gd

ow

ngra

des

from

the

thre

em

ajo

rra

tin

gagen

cies

,i.e.

,F

itch

Moo

dy’s

,an

dS

tan

da

rda

nd

Poo

r’s.

Typ

eD

esc

ripti

on

Chapte

r11

liquid

ati

on

Ifa

busi

nes

sis

unable

tose

rvic

eit

sdeb

tor

pay

its

cred

itors

,ei

ther

the

busi

nes

sit

self

or

any

of

its

cred

itors

can

file

wit

ha

feder

al

bankru

ptc

yco

urt

for

pro

tect

ion

under

Chapte

r11.

As

thedebtor

inpossession

the

trust

eem

ayliquid

iate

the

ass

ets

of

the

firm

.

Chapte

r11

rest

ruct

uri

ng

Ifa

busi

nes

sis

unable

tose

rvic

eit

sdeb

tor

pay

its

cred

itors

,ei

ther

the

busi

nes

sit

self

or

any

of

its

cred

itors

can

file

wit

ha

feder

al

bankru

ptc

yco

urt

for

pro

tect

ion

under

Chapte

r11.

As

thedebtor

inpossession

the

trust

eem

ayre

stru

cture

the

firm

.

Dis

tres

sed

exch

ange

Deb

tor

pro

pose

sa

fundam

enta

lch

ange

inth

eco

ntr

act

ual

com

mit

men

tsto

cred

itors

who

may

volu

nta

rily

agre

e.

Fit

chD

Rati

ng

gra

de

indic

ate

sth

at

firm

has

ente

red

def

ault

.

Moody’s

CO

bligati

ons

rate

dC

are

the

low

est

rate

dcl

ass

and

are

typic

ally

indef

ault

,w

ith

litt

lepro

spec

tfo

rre

cover

yof

pri

nci

pal

or

inte

rest

.

Sta

ndard

&P

oor’

sD

Pay

men

tdef

ault

on

financi

al

com

mit

men

ts.

Fit

chC

Subst

anti

al

cred

itri

sk.

Def

ault

isa

real

poss

ibilit

y.

Moody’s

Ca

Obligati

ons

rate

dC

aare

hig

hly

spec

ula

tive

and

are

likel

yin

,or

ver

ynea

r,def

ault

,w

ith

som

epro

spec

tof

reco

ver

yof

pri

nci

pal

and

inte

rest

.

Sta

ndard

&P

oor’

sC

Curr

entl

yhig

hly

vuln

erable

obligati

ons.

33

Tab

le2:

Rec

over

yR

ates

acro

ssD

efau

ltE

ven

ts,

Ind

ust

ries

an

dS

enio

rity

Th

ista

ble

rep

ort

sth

esu

mm

ary

stati

stic

sfo

rth

ere

cover

yra

tes

(mea

ntr

an

sact

ion

pri

cew

ith

in30

days

aft

erd

efau

lt)

acr

oss

diff

eren

td

efau

ltev

ent

typ

es,

ind

ust

ries

an

dse

nio

rity

level

s.W

ere

port

the

low

est

reco

ver

yra

te(M

in),

25%

qu

anti

le(Q

0.25),

med

ian

,m

ean

,75%

qu

anti

le(Q

0.75),

hig

hes

tre

cover

yra

te(M

ax),

stan

dard

dev

iati

on

(SD

)an

dnu

mb

erof

ob

serv

ati

on

s(N

).P

an

elA

pro

vid

esth

est

ati

stic

sfo

rth

eto

tal

sam

ple

,P

an

elB

list

sth

esu

mm

ary

stati

stic

sfo

rre

cover

yra

tes

acr

oss

the

diff

eren

td

efau

ltev

ent

typ

es,

Pan

elC

pre

sents

the

stati

stic

sfo

rn

on

-fin

an

cia

lfi

rms,

wh

ile

Pan

elD

list

sfi

na

nci

al

firm

s.P

an

elE

giv

esth

est

ati

stic

sfo

rre

cover

ies

acr

oss

diff

eren

tle

vel

sof

sen

iori

ty.

Th

ed

ata

-set

con

sist

sof

tran

sact

ion

data

rep

ort

edby

TR

AC

Efo

rth

ep

erio

dJu

ly2002

toO

ctob

er2010

an

dam

ou

nts

toap

pro

xim

ate

ly1,7

34,0

00

trad

esw

ith

an

aggre

gate

volu

me

of

$500

billion

cover

ing

2,2

35

def

au

ltev

ent/

bon

dco

mb

inati

on

s.T

he

def

au

ltev

ents

data

are

ob

tain

edfr

om

the

Mer

gen

tF

ixed

Inco

me

Sec

uri

ties

Da

taba

sean

dth

eN

YU

Sa

lom

on

Cen

ter

Ma

ster

Def

au

ltD

ata

base

an

dco

ver

ban

kru

ptc

yfi

lin

gs,

ou

t-of-

cou

rtre

stru

ctu

rin

gs

an

dra

ting

dow

ngra

des

.T

he

ind

ust

rycl

ass

ifica

tion

san

dse

nio

rity

level

s(s

ecu

red

,gu

ara

nte

ed,

un

secu

red

,su

bord

inate

d)

are

retr

ieved

from

Blo

om

berg

.

Min

Q0.2

5M

edia

nM

ean

Q0.7

5M

ax

SD

NP

anel

A:

All

reco

ver

yra

tes

All

0.0

112.8

938.5

338.6

165.4

1116.5

027.3

62235

Panel

B:

Rec

over

yra

tes

acr

oss

def

ault

even

tty

pes

Chapte

r11

liquid

ati

on

0.1

211.6

123.1

340.6

869.4

4103.6

039.3

213

Chapte

r11

rest

ruct

uri

ng

0.0

111.2

425.9

637.1

165.7

8110.8

028.7

6492

Dis

tres

sed

exch

ange

10.6

528.9

851.0

451.2

672.3

598.7

125.4

664

Fit

chD

0.4

18.2

226.0

631.3

660.4

463.7

025.5

424

Moody’s

C0.0

13.1

08.4

716.0

223.8

7100.0

019.0

3289

Sta

ndard

&P

oor’

sD

0.0

113.9

947.8

543.8

466.0

1109.1

027.4

3465

Fit

chC

0.2

918.1

143.4

341.2

855.2

3116.5

025.2

1361

Moody’s

Ca

0.2

624.9

743.6

144.8

759.2

8109.1

023.8

7456

Sta

ndard

&P

oor’

sC

0.5

521.4

538.5

343.5

965.9

3110.0

028.1

171

Panel

C:

Rec

over

yra

tes

by

indust

ry:

non-fi

nanci

al

firm

sM

anif

act

uri

ng

0.0

113.5

231.0

338.9

364.3

6110.8

028.5

5573

Med

ia&

Com

unic

ati

ons

0.0

14.3

221.0

634.7

066.8

5101.0

034.5

6163

Oil

&G

as

9.8

533.7

441.6

744.3

753.6

392.7

923.6

821

Ele

ctri

city

23.8

135.7

140.0

748.0

348.0

4102.8

022.6

739

Ret

ail

1.4

16.6

616.3

533.4

057.8

8100.5

034.1

933

Ser

vic

e&

Lei

sure

0.0

313.4

926.9

938.6

563.9

8116.5

030.3

7190

Tra

nsp

ort

ati

on

16.7

825.7

426.9

438.1

757.4

678.3

818.8

570

Rea

lE

state

13.5

634.2

940.6

341.9

749.0

195.9

716.0

571

Panel

D:

Rec

over

yra

tes

by

indust

ry:

financi

al

firm

sB

ankin

g14.3

221.9

859.5

449.2

661.2

769.5

020.1

962

Cre

dit

&F

inanci

ng

2.2

643.7

965.3

856.5

866.0

198.0

114.2

8588

Fin

anci

al

serv

ices

0.0

14.9

19.8

710.6

413.8

198.6

39.7

5363

Insu

rance

7.9

612.2

234.6

643.3

773.9

996.1

432.7

217

Sav

ings

&L

oan

0.0

10.3

70.6

910.7

428.4

930.2

513.4

144

Panel

E:

Rec

over

yra

tes

by

senio

rity

Sec

ure

d0.0

39.8

943.1

749.2

787.9

8110.0

037.4

784

Guara

nte

ed0.0

116.9

034.7

340.2

763.4

3116.5

028.1

3585

Unse

cure

d0.1

412.9

042.3

939.0

965.5

6101.0

025.6

91457

Sub

ord

inate

d0.0

10.5

85.3

115.1

317.8

698.7

123.7

9109

34

Tab

le3:

Bon

dC

har

acte

rist

ics,

Fir

mF

un

dam

enta

lsan

dL

iqu

idit

yP

roxie

sT

his

tab

lere

port

sth

esu

mm

ary

stati

stic

sof

bon

dch

ara

cter

isti

cs(P

an

elA

),fi

rmfu

nd

am

enta

ls(P

an

elB

)an

dliqu

idit

yp

roxie

s(P

an

elC

).W

ere

port

the

low

est

valu

e(M

in),

25%

qu

anti

le(Q

0.25),

med

ian

,m

ean

,75%

qu

anti

le(Q

0.75),

hig

hes

tvalu

e(M

ax),

stan

dard

dev

iati

on

(SD

)an

dnu

mb

erof

ob

serv

ati

on

s(N

).A

mou

nt

issu

edis

giv

enin

milli

on

s,m

atu

rity

inyea

rs,

cou

pon

inp

erce

nt

of

noti

on

al

an

dra

tin

gs

are

map

ped

onto

natu

ral

nu

mb

ers,

e.g.AAA

=1,AA

+=

2,

...,

D=

21.

Equ

ity,

def

au

ltb

arr

ier,

inta

ngib

ilit

y,re

ceiv

eab

les

are

giv

enin

per

cent

of

tota

lass

ets,

LT

Dis

suan

cein

per

cent

of

tota

ld

ebt

an

dp

rofi

tab

ilit

yin

per

cent

of

tota

lsa

les.

Tota

lass

ets

are

giv

enin

$100

billion

s,em

plo

yee

sin

1,0

00.

Volu

me

isgiv

enin

$100,0

00.

Am

ihu

dm

easu

rere

pre

sents

ap

rice

chan

ge

inp

erce

nt

base

don

$1

million

of

volu

me

an

dp

rice

dis

per

sion

isgiv

enin

per

cent.

Th

ed

ata

-set

con

sist

sof

tran

sact

ion

data

rep

ort

edby

TR

AC

Efo

rth

ep

erio

dJu

ly2002

toO

ctob

er2010

an

dam

ou

nts

toap

pro

xim

ate

ly1,7

34,0

00

trad

esw

ith

an

aggre

gate

volu

me

of

$500

bil

lion

cover

ing

2,2

35

def

au

ltev

ent/

bon

dco

mb

inati

on

s.T

he

def

au

ltev

ents

data

are

ob

tain

edfr

om

the

Mer

gen

tF

ixed

Inco

me

Sec

uri

ties

Da

taba

sean

dth

eN

YU

Sa

lom

on

Cen

ter

Ma

ster

Def

au

ltD

ata

base

an

dco

ver

ban

kru

ptc

yfi

lin

gs,

ou

t-of-

cou

rtre

stru

ctu

rin

gs

an

dra

tin

gd

ow

ngra

des

.B

on

dch

ara

cter

isti

csare

retr

ieved

from

Blo

om

berg

an

dth

efi

rmfu

nd

am

enta

lsfr

om

Co

mp

ust

at. M

inQ

0.2

5M

edia

nM

ean

Q0.7

5M

ax

SD

NP

anel

A:

Chara

cter

isti

csof

def

ault

edb

onds

Am

ount

issu

ed10.0

044.4

0250.0

0399.6

0500.0

04200.0

0517.8

02235

Matu

rity

0.0

12.5

24.6

96.8

27.6

834.5

06.6

22235

Coup

on

3.6

06.0

07.0

07.4

88.8

514.2

52.3

52235

Rati

ng

5.0

08.0

013.0

012.2

117.0

021.0

05.1

52235

Panel

B:

Fundam

enta

lsof

def

ault

edfirm

sE

quit

y−

97.2

13.5

97.1

66.6

215.3

897.3

321.2

62164

Def

ault

barr

ier

0.0

122.8

349.8

847.8

063.0

8100.0

025.1

42181

LT

Dis

suance

0.0

19.5

520.9

324.1

427.1

7201.9

030.5

82117

Inta

ngib

ilit

y0.0

10.7

81.2

811.6

012.3

788.8

519.0

12180

Rec

eivea

ble

s0.0

16.4

925.7

331.2

864.4

985.6

926.5

22140

Pro

fita

bilit

y−

149.2

09.6

632.4

237.4

759.1

4190.6

029.1

52133

Tota

lass

ets

0.0

10.0

45

0.9

01.3

91.8

612.6

41.8

42184

Em

plo

yee

s0.0

10.6

71.1

62.9

72.5

928.3

05.2

92158

Panel

C:

Liq

uid

ity

pro

xie

sof

def

ault

edb

onds

Volu

me

0.0

30.4

82.4

13.6

06.2

450.0

03.6

12235

Tra

des

1.0

02.7

55.1

010.7

111.0

8235.4

016.5

72235

Am

ihud

0.0

10.2

90.9

41.4

92.0

016.5

61.8

42024

Dis

per

sion

0.0

11.2

12.5

12.8

03.9

315.3

71.9

82024

35

Tab

le4:

Bon

dC

har

acte

rist

ics,

Fir

mF

un

dam

enta

lsan

dL

iqu

idit

yP

roxie

sfo

rN

on

-Fin

an

cial

an

dF

inan

cial

Fir

ms

Th

ista

ble

rep

ort

sth

esu

mm

ary

stati

stic

sof

bon

dch

ara

cter

isti

cs(P

an

elA

),fi

rmfu

nd

am

enta

ls(P

an

elB

)an

dliqu

idit

yp

roxie

s(P

an

elC

)se

para

ted

inn

on

-fin

an

cia

lfi

rms

an

dfi

na

nci

al

firm

s.W

ere

port

the

mea

n,

stan

dard

dev

iati

on

(SD

)an

dnu

mb

erof

ob

serv

ati

on

s(N

).A

mou

nt

issu

edis

giv

enin

million

s,m

atu

rity

inyea

rs,

cou

pon

inp

erce

nt

of

noti

on

al

an

dra

tin

gs

are

map

ped

onto

natu

ral

nu

mb

ers,

e.g.AAA

=1,AA

+=

2,

...,

D=

21.

Equ

ity,

def

au

ltb

arr

ier,

inta

ngib

ilit

y,re

ceiv

eab

les

are

giv

enin

per

cent

of

tota

lass

ets,

LT

Dis

suan

cein

per

cent

of

tota

ld

ebt

an

dp

rofi

tab

ilit

yin

per

cent

of

tota

lsa

les.

Tota

lass

ets

are

giv

enin

$100

billion

s,em

plo

yee

sin

1,0

00.

Volu

me

isgiv

enin

$100,0

00.

Am

ihud

mea

sure

rep

rese

nts

ap

rice

chan

ge

inp

erce

nt

base

don

$1

million

of

volu

me

an

dp

rice

dis

per

sion

isgiv

enin

per

cent.

Th

edata

-set

con

sist

sof

tran

sact

ion

data

rep

ort

edby

TR

AC

Efo

rth

ep

erio

dJu

ly2002

toO

ctob

er2010

an

dam

ou

nts

toapp

roxim

ate

ly1,7

34,0

00

trad

esw

ith

an

aggre

gate

volu

me

of

$500

billion

cover

ing

2,2

35

def

au

ltev

ent/

bon

dco

mb

inati

on

s.T

he

def

au

ltev

ents

data

are

ob

tain

edfr

om

the

Mer

gen

tF

ixed

Inco

me

Sec

uri

ties

Da

taba

sean

dth

eN

YU

Sa

lom

on

Cen

ter

Ma

ster

Def

au

ltD

ata

base

an

dco

ver

ban

kru

ptc

yfi

lin

gs,

ou

t-of-

cou

rtre

stru

ctu

rin

gs

an

dra

tin

gd

ow

ngra

des

.B

on

dch

ara

cter

isti

csare

retr

ieved

from

Blo

om

berg

an

dth

efi

rmfu

nd

am

enta

lsfr

om

Co

mp

ust

at.

Non-F

inancia

lfirm

sF

inancia

lfirm

sM

ean

SD

NM

ean

SD

NP

anel

A:

Bond

chara

cter

isti

csA

mount

issu

ed434.8

0459.4

01090

366.0

0565.9

01145

Matu

rity

5.5

75.0

71090

7.9

77.6

41145

Coup

on

8.6

31.9

91090

5.7

71.7

31145

Rati

ng

16.7

82.0

41090

7.8

63.0

01145

Panel

B:

Fir

mfu

ndam

enta

lsE

quit

y6.5

930.5

61023

6.6

54.4

61141

Def

ault

barr

ier

23.3

717.7

01040

53.0

215.1

11141

LT

DIs

suance

21.5

828.7

6979

26.3

531.9

11138

Inta

ngib

ilit

y22.0

722.5

71042

2.0

15.7

61138

Rec

eivea

ble

s10.2

69.8

3999

49.6

822.5

71141

Pro

fita

bilit

y12.9

315.4

41040

60.8

317.7

01093

Tota

lass

ets

0.2

30.6

51043

2.4

51.9

41141

Em

loyee

s4.1

67.0

81024

1.8

92.3

71134

Panel

C:

Liq

uid

ity

pro

xie

sV

olu

me

4.6

72.9

71090

2.5

83.8

61145

Tra

des

8.2

011.6

71090

13.0

919.8

71145

Am

ihud

1.3

61.9

1953

1.6

11.7

71071

Dis

per

sion

2.4

81.7

3953

3.0

92.1

51071

36

Table 5: Regression AnalysisThis table reports the results of the panel data regression analysis. The dependent variable is the recovery rate ofthe default event/bond combinations. The explanatory variables are given by bond characteristics (amount issued,maturity, coupon, rating and CDS dummy), firm characteristics (equity, default barrier, LTD issuance, intangibility,receiveables, profitability, total assets and employees), macroeconomic variables (aggregate default rate, federalfunds rate and slope of the interest rate curve), liquidity measures (volume, number of trades, Amihud measure andprice dispersion measure) and dummy variables representing default event types, industries and seniority levels.Model 1 represents a regression including only these dummy variables. Model 2 additionally controls for the bondcharacteristics. Model 3 controls for the firm characteristics. Model 4 controls for the macroeconomic indicators andModel 5 controls for the liquidity measures. Model 6 represents the complete model containing all variables. Thedata-set consists of transaction data reported by TRACE for the period July 2002 to October 2010 and amountsto approximately 1, 734, 000 trades with an aggregate volume of $500 billion covering 2, 235 default event/bondcombinations. The default events data are obtained from the Mergent Fixed Income Securities Database and theNYU Salomon Center Master Default Database and cover bankruptcy filings, out-of-court restructurings and ratingdowngrades. Bond characteristics and macroconomic data are retrieved from Bloomberg and firm fundamentalsfrom Compustat. Clustered standard errors at the default event-firm level (see e.g. Petersen (2009)) are given inparentheses. Significance is given by: *** < 0.01, ** < 0.05, * < 0.1.

Model(1) (2) (3) (4) (5) (6)

Intercept 36.5099∗∗∗ 57.8231∗∗∗ 32.3022∗∗∗ 34.7875∗∗∗ 50.5939∗∗∗ 20.3193∗∗∗

(1.7535) (6.0960) (4.9439) (1.9165) (2.0104) (10.0191)Amount issued −0.1420 0.2850∗∗∗

(0.0890) (0.0950)Maturity −0.7015∗∗∗ −0.6142∗∗∗

(0.0673) (0.0682)Coupon 0.1412 0.9323∗∗∗

(0.3901) (0.3471)Rating −1.0529∗∗∗ −0.6395

(0.3460) (0.4306)CDS availability 0.0849 6.7313∗∗∗

(1.9267) (2.1436)Equity 0.1603∗∗∗ 0.1487∗∗∗

(0.0573) (0.0543)Default barrier −0.2523∗∗∗ −0.1821∗∗∗

(0.0758) (0.0663)LTD issuance −0.1124∗∗∗ −0.0062

(0.0269) (0.0278)Intangibility −0.2401∗∗∗ −0.0659

(0.0571) (0.0471)Receiveables 0.2465∗∗∗ 0.0989

(0.0824) (0.0928)Profitability 0.0044 −0.0671

(0.0627) (0.0584)Total assets 1.5252 3.5960∗∗

(1.1822) (1.6576)Employees −0.5454∗∗∗ 0.0272

(0.1829) (0.1859)Aggregate default rate −5.6664∗∗∗ −4.5636∗∗∗

(0.9634) (1.0698)Federal Funds rate 11.4106∗∗∗ 6.5625∗∗∗

(1.2021) (1.4215)Slope 11.9764∗∗∗ 6.1983∗∗∗

(1.4512) (1.6013)Volume 0.0010 −0.0005

(0.0018) (0.0026)Trades 0.1299∗∗∗ 0.0038

(0.0263) (0.0219)Amihud −0.0790 0.1580

(0.2730) (0.3090)Dispersion −5.6500∗∗∗ −5.0700∗∗∗

(0.2900) (0.3400)Adjusted R2 0.37 0.42 0.47 0.46 0.53 0.64Observations 2235 2235 1972 2235 2024 1809Event dummies Y Y Y Y Y YIndustry dummies Y Y Y Y Y YSeniority dummies Y Y Y Y Y Y

37

Table 6: Regression Analysis for Non-Financial and Speculative Grade BondsThis table reports the results for two subsample based on the panel data regressions. The dependent variable is therecovery rate of the default event/bond combinations. The explanatory variables are given by bond characteristics(amount issued, maturity, coupon, rating and CDS dummy), firm characteristics (equity, default barrier, LTDissuance, intangibility, receiveables, profitability, total assets and employees), macroeconomic variables (aggregatedefault rate, federal funds rate and slope of the interest rate curve), liquidity measures (volume, number of trades,Amihud measure and price dispersion measure) and dummy variables representing default event types, industriesand seniority levels. Model 1 contains the results for non-financial firms, while Model 2 for speculative gradefirms. The data-set consists of transaction data reported by TRACE for the period July 2002 to October 2010and amounts to approximately 1, 734, 000 trades with an aggregate volume of $500 billion covering 2, 235 defaultevent/bond combinations. The default events data are obtained from the Mergent Fixed Income Securities Databaseand the NYU Salomon Center Master Default Database and cover bankruptcy filings, out-of-court restructuringsand rating downgrades. Bond characteristics and macroeconomic data are retrieved from Bloomberg and firmfundamentals from Compustat. Clustered standard errors at the default event-firm level (see e.g. Petersen (2009))are given in parentheses. Significance is given by: *** < 0.01, ** < 0.05, * < 0.1.

ModelNon-Financial Speculative Grade

(1) (2)Intercept 20.8767∗∗∗ 28.8214∗∗∗

(10.1022) (13.0564)Amount issued 0.0370 0.1340

(0.1780) (0.1450)Maturity −0.9675∗∗∗ −0.9569∗∗∗

(0.1484) (0.1585)Coupon 0.7992∗ 0.8480∗

(0.4571) (0.4350)Rating −1.4762∗∗∗ −1.4511∗∗∗

(0.4727) (0.4901)CDS availability 5.5402∗∗∗ 3.6716

(2.0981) (2.2061)Equity 0.2380∗∗∗ 0.2055∗∗∗

(0.0531) (0.0505)Default barrier −0.3336∗∗∗ −0.2833∗∗∗

(0.0656) (0.0629)LTD Issuance 0.0281 0.0169

(0.0271) (0.0266)Intangibility −0.0395 −0.0246

(0.0491) (0.0448)Receiveables 0.2777∗∗∗ 0.2087∗∗

(0.1013) (0.0956)Profitability −0.0661 −0.0661

(0.0618) (0.0541)Total assets 1.6273 0.3885

(1.4592) (1.0853)Employees 0.6108∗∗∗ 0.6099∗∗∗

(0.2075) (0.1581)Aggregate default rate −3.2360∗∗∗ −3.7297∗∗∗

(1.4925) (1.4209)Federal Funds rate 9.9700∗∗∗ 8.2605∗∗∗

(1.7594) (1.5538)Slope 10.8005∗∗∗ 8.6959∗∗∗

(1.9955) (1.6650)Volume −0.0079∗∗ −0.0059∗∗

(0.0036) (0.0034)Trades 0.1316∗∗ 0.1661∗∗∗

(0.0670) (0.0612)Amihud −0.4040 −0.6320

(0.4750) (0.4330)Dispersion −8.8260∗∗∗ −7.7480∗∗∗

(0.5700) (0.7500)Adjusted R2 0.54 0.54Observations 795 868Event dummies Y YIndustry dummies Y YSeniority dummies Y Y

38

References

Viral Acharya, Jing-zhi Huang, Marti G. Subrahmanyam, and Rangarajan Sundaram. When doesStrategic Debt-service Matter? Economic Theory, 29:363–378, 2006.

Viral V. Acharya, Sreedhar T. Bharath, and Anand Srinivasan. Does Industry-wide Distress AffectDefaulted Firms? - Evidence from Creditor Recoveries. Journal of Financial Economics, 85(3):787–821, 2007.

Edward Altman and Egon Kalotay. A Flexible Approach to Modelling Ultimate Recoveries onDefaulted Loans and Bonds. Working Paper, 2010.

Edward I. Altman and Brenda Karlin. The Re-Emergence of Distressed Exchanges in CorporateRestructurings. Journal of Credit Risk, 5(2):43–56, 2009.

Edward I. Altman and Vellore M. Kishore. Almost Everything You Wanted to Know aboutRecoveries on Defaulted Bonds. Financial Analysts Journal, 52(6):57–64, 1996.

Edward I. Altman, Andrea Resti, and Andrea Sironi. The Link between Default and RecoveryRates: Effects on the Procyclicality of Regulatory Capital Ratios. BIS Working Paper No. 113,2002.

Edward I. Altman, Brooks Brady, Andrea Resti, and Andrea Sironi. The Link between Defaultand Recovery Rates: Theory, Empirical Evidence, and Implications. The Journal of Business,78(6):2203–2228, 2005.

Edward I. Altman, Amar Gande, and Anthony Saunders. Bank Debt versus Bond Debt: Evidencefrom Secondary Market Prices. Journal of Money, Credit and Banking, 42(4):755–767, 2010.

Yakov Amihud. Illiquidity and Stock Returns: Cross-Section and Time-Series Effects. Journal ofFinancial Markets, 5(1):31–56, 2002.

Ronald W. Anderson and Suresh Sundaresan. Design and Valuation of Debt Contracts. TheReview of Financial Studies, 9(1):37–68, 1996.

Jack Bao, Jun Pan, and Jiang Wang. The illiquidity of corporate bonds. The Journal of Finance,66(3):911–946, 2011.

Fischer Black and John C. Cox. Valuing Corporate Securities: Some Effects of Bond IndentureProvisions. The Journal of Finance, 31(2):351–367, 1976.

Fischer Black and Myron Scholes. The Pricing of Options and Corporate Liabilities. The Journalof Political Economy, 81(3):637–654, 1973.

Arturo Bris, Ivo Welch, and Ning Zhu. The Costs of Bankruptcy: Chapter 7 Liquidation versusChapter 11 Reorganization. The Journal of Finance, 61(3):1253–1303, 2006.

Pierre Collin-Dufresne, Robert S. Goldstein, and J. Spencer Martin. The Determinants of CreditSpread Changes. The Journal of Finance, 56(6):2177–2207, 2001.

Sergei A. Davydenko and Julian R. Franks. Do Bankruptcy Codes Matter? A Study of Defaultsin France, Germany, and the U.K. The Journal of Finance, 63(2):565–608, 2008.

Jens Dick-Nielsen. Liquidity Biases in TRACE. Journal of Fixed Income, 19:43–55, 2009.

Jens Dick-Nielsen, Peter Feldhutter, and David Lando. Corporate bond liquidity before and afterthe onset of the subprime crisis. Journal of Financial Economics, 103(3):471–492, 2012.

Darrell Duffie and Kenneth J. Singleton. An Econometric Model of the Term Structure of Interest-Rate Swap Yields. The Journal of Finance, 52(4):1287–1321, 1997.

Nils Friewald, Rainer Jankowitsch, and Marti G. Subrahmanyam. Illiquidity or credit deteriora-tion: A study of liquidity in the US corporate bond market during financial crises. Journal ofFinancial Economics, 105(1):18–36, 2012.

Jon Frye. Collateral Damage: A Source of Systematic Credit Risk. Risk Magazine, 2000.

Robert Goldstein, Nengjiu Ju, and Hayne Leland. An EBIT Based Model of Dynamic CapitalStructure. The Journal of Business, 74(4):483–512, 2001.

Michael B. Gordy. A risk-factor model foundation for ratings-based bank capital rules. Journalof Financial Intermediation, 12(3):199–232, 2003.

Samuel G. Hanson and Til Schuermann. Estimating Probabilities of Default. Working Paper,2004.

Jing-zhi Huang and Ming Huang. How Much of the Corporate-Treasury Yield Spread is Due toCredit Risk? Working Paper, 2002.

Rainer Jankowitsch, Amrut Nashikkar, and Marti G. Subrahmanyam. Price Dispersion in OTCMarkets: A new Measure of Liquidity. Journal of Banking and Finance, 35(2):343–357, 2011.

Robert A. Jarrow and Stuart M. Turnbull. Pricing Derivatives on Financial Securities Subject toCredit Risk. The Journal of Finance, 50(1):53–85, 1995.

Albert S. Kyle. Continuous Auctions and Insider Trading. Econometrica, 53(6):1315–1335, 1985.

David Lando. On Cox Processes and Credit Risky Securities. Review of Derivatives Research, 2:99–120, 1998.

Hayne E. Leland. Corporate Debt Value, Bond Covenants, and Optimal Capital Structure. TheJournal of Finance, 49(4):1213–1252, 1994.

Hayne E. Leland and Klaus Bjerre Toft. Optimal Capital Structure, Endogenous Bankruptcy, andthe Term Structure of Credit Spreads. The Journal of Finance, 51(3):987–1019, 1996.

Francis A. Longstaff and Eduardo S. Schwartz. A Simple Approach to Valuing Risky Fixed andFloating Rate Debt. The Journal of Finance, 50(3):789–819, 1995.

Francis A. Longstaff, Sanjay Mithal, and Eric Neis. Corporate Yield Spreads: Default Risk orLiquidity? New Evidence from the Credit Default Swap Market. The Journal of Finance, 60(5):2213–2253, 2005.

Dilip Madan and Haluk Unal. Pricing the Risks of Default. Review of Derivatives Research, 2:121–160, 1998.

Pierre Mella-Barral and William Perraudin. Strategic Debt Service. The Journal of Finance, 52(2):531–556, 1997.

Robert C. Merton. On the Pricing of Corporate Debt: The Risk Structure of Interest Rates. TheJournal of Finance, 29(2):449–470, 1974.

Mitchell A. Petersen. Estimating Standard Errors in Finance Panel Data Sets: Comparing Ap-proaches. Review of Financial Studies, 22(1):435–480, 2009.

Andrei Shleifer and Robert W. Vishny. Liquidation Values and Debt Capacity: A Market Equi-librium Approach. The Journal of Finance, 47(4):1343–1366, 1992.

Rick L. Williams. A Note on Robust Variance Estimation for Cluster-Correlated Data. Biometrics,56(2):645–646, 2000.

40