Replication of Chava, Kumar, and Warga (2009)

Replication of Chava, Kumar, and Warga (2009) Table 2 Report

Mingxun “Richard” Wang

Georgia Institute of Technology

May 31st, 2018

(June 9th, 2018 revised)

Table of Contents

1. Introduction
2. Screening the Sample
3. Classification of Covenants
4. Replication and Comparison
5. Issues

1. Introduction

This report will illustrate the procedure of data collection and the replication of Table 2 in the Chava, Kumar, and Warga (2009) paper. By carefully following most of the sample screening steps provided in the paper, I was able to establish replications for three different periods. One (1993-2007 replication) had 4188 bonds by 1086 firms (with up-to-today FISD, CRSP, Compustat and ExecComp datasets), which was very close to 4478 bonds by 1064 firms from the Table 2 in the original paper. The other one included bonds from 2008 to 2017. The last one was composed by all the data (1993 to 2017). Then based on these samples, I replicated the Table 2 by categorizing all the covenants into exactly same groups that were described underneath the Table. The 1993-2007 percentage results were mostly like the original results. However, comparing the tables of original 1993-2007 with replication 2008-2017, there were several covenants, such as “Merger restrictions”, and “Stock sale restrictions”, that had obvious declining occurrences in contracts of noninvestment grade high-yield bonds.

1. Screening the Sample

1. Get all the U.S. Domiciled bonds (Firms: 10341)

        Dataset “fisd_issuer” → filter the column “country_domicile” to be “USA”.

1. Get all the nonfinancial names (Firms: 7144)

        Filtering the dataset “fisd_issuer” → make column “industry_code” exclude all the

financial codes (20, 21, 22, 23, 24, 25, 26).

1. Filtering by years

        Filtering the datasets by three different time periods.

• 1993-2007 (replication to the original period in the paper)

(Bonds: 195338; Firms: 9755)

• 2008-2017

(Bonds: 193092; Firms: 5380)

• 1993-2017

(Bonds: 388430; Firms: 13046)

1. Intersection of Compustat (Bonds: 16976; Firms: 3181)

WRDS compustat dataset “funda”, and matched by “cusip” (substring first 6 digits)

1. Intersection of CRSP (Bonds: 12867; Firms: 2600)

WRDS CRSP dataset “dsfhdr”, and matched by “cusip” (substring first 6 digits)

1. Intersection of Executive Compustat (Bonds: 9224; Firms: 1436)

WRDS ExecComp dataset “execcomp” and matched by “cusip” (substring first 6 digits)

1. Exclude Yankee, Canadian, and Foreign Currencies (Bonds: 9060; Firms: 1422)
2. Corporate Debentures (Bonds: 6289; Firms: 1127)

Let the column “asset_backed” to be “N”

1. Get all the ratings at the time of initial issuance (Bonds: 4183; Firms: 1085)

Combine datasets “fisd_rating”, “fisd_ratings”, and “fisd_rating_hist”, and keep only the very beginning ratings for each bond.

Convert all the ratings from Moody’s, Fitch, and S&P into categories of AAA, AA, A, BBB, BB, B, CCC and below groups.

• Key Numbers by steps (Updated on June 9th, 2018)

1. Classification of Covenants

All the covenants were from three FISD tables, “fisd_bondholder_protective” (18), “fisd_issuer_restrictive” (18), and “fisd_subs_restrictive” (14). In total of 50 covenants, along with “secured or not” covenant information from the table “fisd_issue”.

Table of Classification of Covenants

        ##: means having possibility that the covenant belongs to the category, however could not be determined to be included into the replication

        **: means highly possibility that the covenant belongs to the category, and had been considered and calculated into the replication