Alternative explanations and model specifications

國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

31

average path length and density have the expected signs, which support my third hypothesis.

5.3. Robustness checks

The results presented above suggest that using relational distance as a proxy for segmented communication lend supports to my model’s predictions. In order to identify the presence of information cascades and distinguish them from other possible explanations for the results, I conduct a number of additional analyses to gauge the robustness of my empirical findings. I first conduct robustness checks by repeating my analysis separately with interaction terms between relational distance measures and previous lead bank and borrower relationship, investment grades, covenant violation and foreign bank participation variables. I examine the robustness of the relationship between relational distance and loan spread in each of these subsamples. Finally, I examine the potential endogeneity between spread and relational distance and the impact of omitted variables.

5.3.1. Alternative explanations and model specifications

The first concern of my results is that it is information asymmetry between lead bank and participants instead of cascade effect which drives my results. The information asymmetry effect gives rise to adverse selection and moral hazard problems (e.g., Focarelli et al., 2008; Ivashina, 2009; Sufi, 2007, 2009; Lee and Mullineaux, 2004), and enable participants to require a higher spread to compensate for their informational disadvantage. To disentangle this alternative explanation, I first check the relationships between my relational distance measures and the adverse selection effect. Adverse selection happens when lead bank has private information on the borrower that is unknown to participants. Following Sufi (2007), I use previous lending relationships between the lead bank and the borrower to capture the lead

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

32

bank’s information advantage. If the relational distance measures mainly capture the adverse selection effect, we should expect that the impact of these measures on loan spreads is stronger when there are previous relationships between the lead bank and borrower.

The results are shown in Table 2.6.⁹ If my relational distance measures mainly capture adverse selection effect, I shall expect to see significant interaction terms, P_Previous_Relation, C_Previous_Relation, and D_Previous_Relation. As shown in Panel A of Table 2.6, none of the three interactions is statistically significant. It suggests that my results are not driven by adverse selection effect. The negative coefficients on the dummy variable, Previous_Relation, are consistent with Ivashina and Kovner (2011), although they are statistically insignificant.

[Insert Table 2.6]

An additional concern is that my results may be driven by moral hazard effect rather than cascade effect. I conduct a similar exercise that classifies borrowers into two categories, investment grade and non-investment grade, according to their S&P senior debt rating. The borrowers’ credit rating is used as a proxy for their informational opacity. Borrowers with non-investment grade require more intense monitoring and due diligence, hence moral hazard problem for the lead bank is more severe. If my relational distance measures are driven by the moral hazard effect, we should expect that the impact of the three measures on loan spreads will be stronger when borrowers are informationally opaque. The evidence in Panel B of Table 2.6 show that the interaction terms, P_Investment_Grade, C_Investment_Grade, and

9 I do not report the results of control variables for brevity. However, the full tables be found in Appendix B, Tables B8 to B11.

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

33

D_Investment_Grade, are statistically insignificant, which suggests that my results are not driven by moral hazard effect.

Another worry is that the relational distance amongst lenders may imply difficulties in ex post renegotiation, which should be priced ex ante. I use ex post violations of financial covenants as a proxy for expected probability of renegotiation.¹⁰ I assume that loans with covenant violations have higher ex ante probability of renegotiation. When it is difficult to execute renegotiation and the expected probability of renegotiation is high, the loan spread may be higher to reflect the renegotiation costs. I incorporate the covenant violation dummy and the interaction term of the dummy variable with my relational distance measures in the regression and the results are reported in Panel C of Table 2.6. If my measures of relational distance mainly capture renegotiation costs, we should expect to see significant coefficients on the interaction terms, P_Covenant_Violation, C_Covenant_Violation, and D_Covenant_Violation. Results in Panel C of Table 2.6 show that covenant violation dummy, Covenant_Violation, is significantly positive which means that the loans which the borrowers ex post violate financial covenants have ex ante higher spreads. However, most of the interaction terms are insignificant.

The evidence does not support the relational distance amongst lenders may imply difficulties in ex post renegotiation.

One may also question that behaviour may differ amongst different types of lenders, e.g., foreign banks and domestic banks, and my results may not be

10 To match my sample with covenant violation provided by Nini et al. (2011), I drop observations with facility end date before July 1, 1997 and the facility start date after March 31, 2008. However, my results are not sensitive to this sample adjustment. For example, the results are qualitatively the same if I drop the observations with facility end date before January 1, 1997 and facility start date after June 30, 2008.

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

34

generalized to different lender classes. To answer this question, I test whether my relational distance measures show different patterns between foreign lenders and domestic lenders. The results are presented in Panel D of Table 2.6. The coefficients on the foreign lender dummy, Foreign_Lender, is positive which indicates that loans with foreign participants have lower spreads, but most of them are statistically insignificant. The interaction terms between the dummy variable and the three distance measures, P_Foreign_Lender, C_Foreign_Lender, and C_Foreign_Lender, are mostly insignificant. The evidence show that my results can be generalized to different lender classes.