Results for Relational Distance

國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

26

financial covenants are less likely to fail. Borrowers’ financial characteristics seem to exert little influence on failure probability.

5.2. Results for Relational Distance

Results in the previous section suggest that communication amongst lenders does not seem to be affected by physical distance. In this section, I consider another proxy for communication segmentation, the relational distance, and test its effect on syndicated loan contracts.

5.2.1. Relational distance and loan spreads

First, I test the relationship between relational distance and loan spread. The regressions are performed using equation (6) to test hypothesis one. The results are presented in Table 2.3. In all specifications, my proxies for relational distance have statistically significant effects on loan spreads. Column (1) shows that the coefficient of average path length (Path) is positive and statistically significant at the 0.1% level.

Columns (2) and (3) show that the coefficients of clustering coefficient (Clustering) and density (Density) are negative and statistically significant at the 0.1% level. These results lend support to my hypothesis that the further the relational distance, the higher is the probability of a cascade occurring, hence the larger is the spread.

[Insert Table 2.3]

The effect of relational distance on loan spreads is also economically significant.

A one-standard deviation increase (decrease) in average path length (clustering coefficient/density) increases the all-in-drawn spread by 7.7 (11.5/15.0 ) basis points, which is approximately 3.4 (5.1/6.7) percent of the sample median spread of 225 basis points.

The coefficients of the control variables are by and large similar to those reported in column (1) of Table 2.2. In column (1), loan spreads significantly decrease with

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

27

loan amount. The reason may be due to the effect of economies of scale (Graham et al., 2008; and Godlewski et al., 2012) or because larger borrowers, which have greater transparency and lower default risk, typically issue larger loans (Focarelli et al., 2008;

and Carey and Nini, 2007). The negative coefficient on maturity seems to be consistent with ‘credit-quality hypothesis’ – banks limit their exposure by lending riskier borrowers shorter loans (see, for example, Gottesman and Roberts, 2004). The loan spreads on secured/guaranteed loans are significantly higher than those on unsecured/non-guaranteed loans. This result seems to support “observed-risk hypothesis” – banks charge riskier borrowers larger loan spreads and require more collateral (see, for example, Godlewski and Weill, 2011). The coefficient of the

“missing” variable is positive and significant. This means I may have underestimated the coefficient of Secured/Guaranteed because I treat the missing value as unsecured/non-guaranteed.

Loans with financial covenants seem to have lower spreads. The intuition may be that covenants is a contract design that can work as an ex ante monitoring device to moderate moral hazard, which in turn reduces the loan spread (Ivashina, 2009). Lee and Mullineaux (2004) and Bosch and Steffen (2011) showed that syndicate size is larger when the borrower has higher transparency and lower credit risk. This finding is consistent with my negative coefficient on lender number. I also find that borrowers’

characteristics such as larger size, higher growth opportunities, and greater profitability are negatively associated with loan spreads. Borrowers with higher leverage bear higher loan spreads. GDP per capita and credit spread have significant influence on loan spread, which is consistent with the findings of Giannetti and Yafeh (2012) and Graham et al. (2008).

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

28

5.2.2. Relational distance and non-price requirements

In this section, I run probit regressions to test my second hypothesis. The results are presented in Table 2.4. In Panel A of Table 2.4, I examine whether the probability of setting financial covenants increases with relational distance measures. The results show that the coefficients of my variables of interest, Path, Clustering, and Density, are statistically insignificant at conventional levels of significance. Relational distance does not seem to exert any influence on the probability of having financial covenants.

As for the coefficients of the control variables, they are qualitatively the same as those reported in column (2) of Table 2.2. Again I find that financially less constrained firms typically characterised by higher ROA and lower leverage are more likely to use financial covenants.

[Insert Table 2.4]

Panel B of Table 2.4 reports the regression results for collateral requirements. In column (1), the coefficient of average path length is positive and statistically significant with a value of 0.11. In columns (2) and (3) the coefficients of the clustering coefficient and density are negative and statistically significant at the 1%

level. These results suggest that the longer is the relational distance amongst lenders, the higher is the probability of requiring collateral, which is consistent with hypothesis two. In addition, the coefficients of the control variables show that loans with higher amount are less likely to require collateral. The probability of requiring collateral is lower for public firms than for private firms. Large sized borrowers tend to have lower probability of offering collateral. Unlike the results for financial covenants, there is evidence to suggest that borrowers that are more financially constrained, as characterised by lower profitability and higher leverage, are more

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

29

likely to offer collateral. Finally, macroeconomic factors such as GDP per capita and credit spread also have a significant impact on requiring collateral.

Panel C of Table 2.4 provides the regression results of equation (6). Column (1) shows that the average path length has a strong effect on the likelihood of guarantees requirements. The coefficient of Path is 0.03, which has the expected positive sign and is statistically significant at the 0.1% level. The same strong effect can also be found for the coefficient of Density which also has the expected sign and is significant at the 0.1% level. On the contrary, the coefficient of Clustering is insignificant at the 5% level. Given these results, I argue that the evidence in Panel C supports my hypothesis that the longer is the relational distance, the higher is the probability of guarantee requirements. Results in Panel C also show that loans with larger amount or with financial covenants have higher probability to require guarantees. Secured loans also exhibit higher probability of setting up guarantees. It is worth noting that given the negative and statistically significant coefficient of the variable Missing, I may overestimate the effect of secured loans on the probability of requiring guarantees.

Public firms, again, have lower probability to offer guarantees than private firms.

Value firms are more likely to offer guarantees than growth firms. Finally, the probability of requiring guarantees increases in borrowers’ leverage, GDP per capita, and term spread.

In summary, my evidence suggests that when a cascade is more likely to happen, the loan contracts have a greater tendency to include non-price agreements to attract potential lenders. This is especially true for non-price contract terms like collateral and guarantees but not for financial covenants. This observation may be in line with the intuition that even though financial covenants can prevent borrowers’ moral hazard, only collateral and guarantees can reduce lender’s damage once default occurs.

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

30

Therefore, contract terms on collateral and guarantees seem to be more attractive for lenders compared to financial covenants.

5.2.3. Relational distance and syndication failures

To test hypothesis three, I focus on sample with lender numbers between two and ten. The results are reported in Table 2.5. The coefficient of average path length is negative and significant at the 5% level, and the coefficient of density is positive and significant at the 5% level. The results from columns (1) and (3) suggest that the longer is the relational distance amongst lenders, the lower is the probability of syndication failure. However, I do not find support for the clustering coefficient. In column (2) of Table 2.5, the coefficient of Clustering is marginally significant but the sign is inconsistent with my model’s prediction.

[Insert Table 2.5]

The effect of control variables on the probability of syndication failures is similar to the results in column (5) of Table 2.2. Loans with larger amount or longer maturity have higher probability to fail. Contracts with financial covenants are less likely to fail. The probability of syndication failures is lower in expansions, as characterised by higher GDP per capita and term spread.

Due to the large difference between the number of failure cases and the number of success cases in my sample, I also test my third hypothesis using a matching sample method of failure and success syndication observations.⁸ I obtain 224 pairs of such observations. I find that there are significant differences for the three relational distance measures and the patterns are similar to those reported in Table 2.5. The

8 I find in Table 2.5 that the probability of syndication failure is significantly affected by loan amount,

maturity, financial covenants, and ROA, so I use these four variables as criteria to match the sample.

Results of matched sample are provided in Appendix B, Table B7.

‧ 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.