Empirical Estimates

This section presents the estimates derived from the empirical model presented in the previous section. We take several steps to understand empirically the extent of bank loan syndication in relation to its primary determinants based on the literature-rule of law, information asymmetry, and other bank-and loan-specific characteristics. First, we examine this relationship while controlling for broad bank- and loan-specific characteristics. Bank variables include the tier 1 ratio, liquidity, and nonperforming loans; loan variables are loan maturity, loan security, and loan-tranche size. The interaction variable of the cost-to-income ratio and the regulatory quality index was also included to capture bank structure and banking regulation.

Table 5.2 shows these specifications and the ordinary least squares regression results of the debt concentration variable on the explanatory variables. We find that regressions have a high degree of explanatory power: all regressions have an R2 of at least 0.70. Regressions are conducted with and without fixed effects to check and control for unobserved heterogeneity across sectors, quarters, and countries. Standard errors are clustered at the project level to account for correlation among projects in the sample.

Table 5.2 shows the coefficients of the rule of law variable in columns 1 to 5 are all positive, implying a strong direct association between the rule of law index and debt concentration (or an inverse relationship between legal risk and debt concentration). Greater rule of law (or a lower legal risk), therefore, increases the extent of loan syndication. This relationship holds even after controlling for bank-and loan-specific characteristics. The same positive relationship robustly holds with and without sector and time (quarter) fixed effects. The coefficient of the legal risk variable in all the specifications are also significant. These results show that, in the seven countries, bank lenders to PPP infrastructure projects are more likely to form a more concentrated syndicate when they can rely on the efficiency of courts and other legal institutions to protect and enforce contract arrangements, which confirms the empirical findings of Esty and Megginson (2003). Bank lenders, however, tend to create larger and more diffuse syndicates to discourage strategic default when they cannot rely on courts for legal enforcement mechanisms. When banks experience inefficiencies in the legal system, they are exposed to greater legal risk, which affects their lending behavior.

Table 5.2: Baseline Reģressions on Debt Concentration

EW = equally weighted, FE = fixed effects, S&P = Standard & Poor's.

Notes:

1. The table presents ordinary least squares regression results to examine the determinants of the extent of bank loan syndications on project finance deals in Asia.

2. The dependent variable is debt concentration.

3. Standard errors (in parentheses) are clustered at the project level to account for correlation among projects in the sample.

*** p < 0.01 ** p < 0.05 * p < 0.10

Source: Authors' estimates.

Table 5.2 shows that the coefficients of the tier 1 ratio in columns (1) to (4) are negative and significant. In turn, this shows that stronger banks with more solid capital bases are less likely to form concentrated loan syndicates on lending, which confirms the inverse relationship between capital requirements and debt concentration in the context of syndicated lending. Consistent with the findings of Simons (1993), a higher capital-to-asset ratio provides an incentive for bank lenders to form and participate in a loan syndicate because they will be unwilling to put a large loan on their balance sheet that will lower their ratios. In which case, banks may opt to participate in a loan syndicate to pursue lending. Godlewski (2008) supports this. The existence of capital requirements should positively influence the formation of loan syndicates through the motivation of "lending limit respect," where a stronger capital requirement increases the motivation relevance. A higher capital-to-asset ratio, therefore, increases the size of a loan syndicate and decreases its debt concentration.

The coefficients of the liquid-to-total-assets ratio are all negative and significant in columns (1) to (4). The empirical result implies that higher bank liquidity discourages the formation of more concentrated loan syndicates for lending to PPP infrastructure projects in the seven countries. While the relationship between bank liquidity and the level of debt concentration in a loan syndicate have not been established in the literature, some empirical evidence might support our finding. Using bank survey data, Pavel and Phillis (1987) find a positive and significant relationship between bank liquidity and the prospect for selling a loan. In the context of loan syndication, this may imply that higher bank liquidity results in the formation of large, diffuse syndicates. The authors note in this connection that if loan sales are primarily driven by liquidity and diversification prospects, then selling a loan should be encouraged. If this is the case, then higher bank liquidity is associated with a greater chance that banks will form less concentrated (or bigger, diffuse syndicates) in the event of syndication.

Among loan-specific characteristics, the coefficients of loan-tranche size appear to be significant in all the specifications. The robust positive relationship of loan-tranche size and debt concentration is also in line with studies that show the size of bank loans greatly influences the monitoring incentive of banks. A bank that lends a large amount to a borrower has a greater incentive to monitor the borrower's management of earnings and capacity to pay. Ahn and Choi (2009), as noted earlier, find that, as banks tend to lend bigger loan amounts, they are also more prone to risk and have a greater motivation for monitoring their borrowers. This implies that, as loan-tranche sizes increase, banks form more concentrated syndicates to observe the conduct of bank monitoring. Khalil and Parigi (1998) show that loan size signals a bank's greater stake in enhancing monitoring prospects because it also affects the income reporting of borrowers. Further, Kang et al. (2000) and Lee and Mullineaux (2001) find a positive relation between the size of a bank loan and a bank's incentive to do monitoring.

Other explanatory variables in the regressions that are found to affect the debt concentration of loan syndicates in the literature, such as Standard & Poor's ratings, nonperforming loans, and loan maturity and security, are not significant. And, while the coefficients of the loan maturity variable are not significant, the negative sign of the coefficients confirms our hypothesis that banks are less likely to lend more in a loan syndicate amid lengthening loan tenors, since longer tenors are associated with higher project risk. This empirical result supports research findings that short-term loans are more effective in resolving agency problems in the context of debt financing (Farinha and Santos 2002; Jones, Lang, and Nigro 2005, for example). The negative coefficients of the loan security variable also confirm the association between loan collateral and loan syndication. Because loan collateral signals a borrower's credit worthiness, the presence of loan collateral accordingly reduces the problems associated with information asymmetry that lead to the formation of large, diffuse loan syndicates.

Special mention of the information asymmetry variable needs to be made. The coefficient of Standard & Poor's variable is negative, but not significant. The relationship is, however, consistent with the findings in the literature that banks form more concentrated loan syndicates when problems of information asymmetry in loan transactions are potentially severe. The motivation for monitoring and due diligence encourage the lead arranger and syndicate members to form a more concentrated syndicate (Dennis and Mullineaux 2000; Lee and Mullineaux 2001; Sufi 2007). One possible explanation of why the information asymmetry variable is not significant across all the regression specifications is because the dataset contains all project finance transactions, which may have already captured and treated agency cost problems in bank lending.

To assess the robustness of the empirical results, we run the baseline regressions without controlling for loan-specific characteristics. Table 5.3 shows the ordinary least squares estimates of the baseline regression without the loan-specific variables. We find evidence that the regression results on the impact of the ratio of the rule of law index to debt concentration is indeed robust and significant. Other explanatory variables, such as the tier 1 ratio and the liquid-to-assets ratio, also show robust and significant results. Moreover, the bank regulation variable is now significant among these specifications. Columns (1) to (4) show the coefficient of the bank regulation variable is negative and significant at 1%, which confirms the positive relationship between cost efficiency and debt concentration.

Table 5.3: Baseline Reģressions without Loan-Specific Characteristics

FE = fixed effects, S&P = Standard & Poor's.

Notes:

1. The table presents ordinary least squares regression results to examine the determinants of the extent of bank loan syndications on project finance deals in Asia.

2. The dependent variable is debt concentration.

3. Standard errors (in parentheses) are clustered at the project level to account for correlation among projects in the sample.

*** p < 0.01 ** p < 0.05 * p < 0.10

Source: Authors' estimates.

We now investigate the determinants of the likelihood of loan syndication for project finance using the dummy for loan syndication as the dependent variable. For consistency, the same set of explanatory variables in the baseline regression is used to determine the factors that influence the likelihood that banks will form and participate in a loan syndicate. The probit specifications show that, except for bank liquidity, most of the variables used in the baseline regression are not robustly significant. The main explanatory variables, such as the rule of law index, tier 1 ratio, and bank regulation, do not significantly affect the likelihood that banks will form and participate in a loan syndicate. Appendix A5.1 gives the regression specification results.

To further examine the factors that influence the likelihood of loan syndication, we run specifications using a set of explanatory variables that may influence a bank's decision to form and participate in a loan syndicate. Table 5.4 shows the probit specification where the dummy variable is regressed to a set of macroeconomic variables that capture monetary policy (repo rate), macroeconomic stability (inflation rate), and a country's credit risk profile (credit default swap spread). Loan-specific characteristics, such as loan-tranche size, loan maturity, and loan security, are also included. Regressions are conducted with and without fixed effects.

Table 5.4: Probit Reģression on the Likelihood of Loan Syndication

CDS = credit default swap, EW = equally weighted, FE = fixed effects.

Notes:

1. The table presents probit regression results to examine the determinants of the likelihood of bank loan syndications on project finance deals in Asia.

2. The dependent variable is the loan syndication dummy variable

3. Standard errors (in parentheses) are clustered at the project level to account for correlation among projects in the sample.

*** p < 0.01 ** p < 0.05 * p < 0.10

Source: Authors' estimates.

Columns (1) and (2) in Table 5.4 show the repo rate coefficients are negative and significant at 1%. This implies that tighter monetary policy reduces the prospect for loan syndication. While the result is not robust across all specifications, the empirical result provides some evidence that the bank lending channel may also be affected by the impact of monetary policy in the willingness of banks to take risk (the risk-taking channel) (Bernanke and Gertler 1995; Borio and Zhu 2008). The coefficient sign in column (5), however, is positive and significant at 10% where sector, quarter, and country fixed effects are present. This implies the impact of monetary policy on bank lending in the context of syndicated lending is ambiguous and remains an empirical question.

Table 5.4 also shows the coefficients of the spread variable for credit default swaps are all positive and robustly significant in all the specifications (significant at 1% in four out of five specifications, significant at 10% in one). This robust finding indicates that, because credit default swaps reduce the credit risk exposure of banks to potential lenders, bank lenders are more likely to form and participate in a loan syndicate. Credit default swaps also reduce the incentives of banks to conduct bank monitoring, which in turn encourages the formation of a loan syndicate. This empirical finding is highly consistent with the view that the potential risk to banks caused by agency problems reduces the likelihood of syndication.

Loan-tranche size coefficients are all positive and significant in three out of five regressions. Consistent with the literature, banks are more likely to syndicate a loan when the size of the loan is significantly large. Also, the coefficient for loan security is significant in columns (1) and (2), which confirms the findings in the literature that loan collateral reduces the problems associated with information asymmetry, which in turn encourages the formation of a loan syndicate.