A. Alternative bondholder concentration proxies



The table reports the estimates of a model:

where Y denotes the yield spread, H denotes bondholder concentration, and x is the set of control variables used throughout. The model is estimated with instrumental variables, where bondholder concentration H is instrumented by the interaction between the previous-period component of concentration of the Katrina-affected investors and the hurricane Katrina indicator, _, as described in the text and in the previous tables. For convenient tabulation, only the coefficient on bondholder concentration H is reported for each regression. The model is estimated in a number of sub-samples, and in both panels A and B, it is run at the issuer-seniority class level, i.e. one observation corresponds to a given issuer-seniority class in a given quarter. In panel A, the sample is split based on credit rating (investment and speculative grade, columns (1)-(2)) and investor horizon (proxied by Investor portfolio duration, columns (3)-(4)). In panel B, the sample is split based on bond liquidity, proxied by the Illiquidity index (columns (1)-(2)) and Trading volume (columns (3)-(4)). Following Petersen (2009) the t-statistics are based on standard errors clustered around issuer-seniority class in all regressions. The sample consists of all the issuer-seniority class groups of bonds in the Bank of America Merrill Lynch Corporate and High Yield Master Bond Index Database over the period between 1998Q1 and 2007Q4, excluding 24 bonds issued by (re)insurance companies affected by hurricane Katrina as described in Appendix A. To include a given bond in the sample, we require the availability of bond characteristics in the Mergent Fixed Income Securities Database (FISD) and firm characteristics in the Compustat Fundamentals quarterly database. The symbols *, **, and *** denote statistical significance at the 10%, 5%, and 1% levels.

The table reports the estimates of a model:

where Y denotes the yield spread, H denotes bondholder concentration, and x is the set of control variables used throughout. The model is estimated with instrumental variables, where bondholder concentration H is instrumented by the interaction between the previous-period component of concentration of the Katrina-affected investors and the hurricane Katrina indicator, _, as described in the text and in the previous tables. For convenient tabulation, only the coefficient on bondholder concentration H is reported for each regression. The model is estimated in a number of sub-samples, and in both panels A and B, it is run at the issuer-seniority class level, i.e. one observation corresponds to a given issuer-seniority class in a given quarter. In panel A, the sample is split based on credit rating (investment and speculative grade, columns (1)-(2)) and investor horizon (proxied by Investor portfolio duration, columns (3)-(4)). In panel B, the sample is split based on bond liquidity, proxied by the Illiquidity index (columns (1)-(2)) and Trading volume (columns (3)-(4)). Following Petersen (2009) the t-statistics are based on standard errors clustered around issuer-seniority class in all regressions. The sample consists of all the issuer-seniority class groups of bonds in the Bank of America Merrill Lynch Corporate and High Yield Master Bond Index Database over the period between 1998Q1 and 2007Q4, excluding 24 bonds issued by (re)insurance companies affected by hurricane Katrina as described in Appendix A. To include a given bond in the sample, we require the availability of bond characteristics in the Mergent Fixed Income Securities Database (FISD) and firm characteristics in the Compustat Fundamentals quarterly database. The symbols *, **, and *** denote statistical significance at the 10%, 5%, and 1% levels.

The table reports the estimates of our baseline model, splitting the sample depending on the degree of bank lender concentration faced by a given bond issuer. Columns (1) and (3), labelled “low bank loan concentration” restrict the attention to issuers with a bank lender Herfindahl index below 0.05 (approximately equal to the sample median), columns (2) and (4) above 0.05. The specifications in columns (1)-(2) are estimated using OLS, those in columns (3)-(4) using instrumental variables estimation, where bondholder concentration H is instrumented by the interaction between the previous-period component of concentration of the Katrina-affected investors and the hurricane Katrina indicator, _. The hurricane Katrina indicator Kat is equal to 1 in 2005Q3, and 0 otherwise. All specification include the full set of control variables used throughout, including coarse rating, industry, and date indicators, as well as issuer-seniority class fixed effects. In all specifications, the t-statistics are based on standard errors clustered around issue-seniority class. The sample consists of all the issuers of bonds in the Bank of America Merrill Lynch Corporate and High Yield Master Bond Index Database with available bond characteristics in the Mergent Fixed Income Securities Database (FISD), firm characteristics in the Compustat Fundamentals quarterly database, over the period between 1998Q1 and 2007Q4. The symbols *, **, and *** denote statistical significance at the 10%, 5%, and 1% levels.