This section summarizes the data and methodology used to test the hypotheses developed to help answer the four research questions and the empirical findings of this thesis.
1.5.1. Data
The final sample consists of 17,636 revolving and 6,625 term loans, for a total of 24,261 loans. It was collected from several databases: DealScan was the main source of information on loan contract terms and lender characteristics.
Compustat, the Institutional Brokers’ Estimate System, the Center for Research in Security Prices, and the National Information Center provided borrower information.
DataStream offered macroeconomic control variables.
Due to data availability, the final sample period is from 1 January 1994 to 31 December 2009. The data selection process, detailed later in Chapter 3, Section 3.2.4, includes several steps. First, information on all loans made by banks operating in the United States to U.S. non-financial firms from 1 January 1987 to 31 December
2009 is obtained from DealScan. Any loans without lead lender information or borrower Global Company Keys (GVKEYs) are then excluded. These GVKEYs are used to search for borrower information from Compustat and trace borrower–lender relationships. Loans without borrower information from Compustat are also omitted.
Those loans distributed under bilateral or club deals or without distribution information are dropped. Since many lead lenders make it difficult to identify their influences, only syndication loans with one lead lender or sole lender loans are included.10 Loans before 1994 are excluded since loan covenant information is very limited before then. Finally, given the focus on the two main loan types (i.e., revolving and term loans), other loans (e.g., letters of credit, demand loans, bridge loans) are excluded.
1.5.2. Methodology
This section first outlines the methodology associated with RQ1, RQ2, and RQ3 and then that for RQ4. Regarding RQ1, Durbin–Wu–Hausman exogeneity tests are first used to test the null hypothesis that the five loan terms should be determined separately (Lin, Phillips, and Smith, 2008). If the null hypothesis is rejected, then the loan terms should be determined simultaneously. This thesis then examines whether loan prices have unidirectional relations with non-price terms (i.e., collateral, maturity, covenants, and size) and whether non-price terms have bidirectional relations with other non-price terms. This simultaneous equation model provides the
10 To identify whether a loan is based on prior relationships, all its lead lenders are traced. So the sample of the second step, which includes multiple lead lender syndications, is used rather the final sample (see Section 4.2.3.2).
coefficients for independent loan terms and reveals evidence of their jointness. The overall model and its variables are presented in Chapters 3 and 4.
Since the answer for RQ1 is yes (i.e., loan terms are simultaneously determined), RQ2 and RQ3 are also addressed through the simultaneous equation model used for RQ1. This model provides estimated coefficients for the information asymmetry proxy to determine whether lenders and borrowers trade off the five key loan terms when addressing borrower information asymmetries, given loan term jointness (RQ2). In addition, the model offers estimated coefficients for the lending relationship measure. These indicate whether lending relationships have any impact on loan covenants and sizes, given the jointness of these loan terms (RQ3).
Next RQ4 is addressed by modifying the simultaneous equation model utilized to address RQ1, RQ2, and RQ3 through the addition of a new variable, the product of the information asymmetry proxy and the lending relationship measure.
This new variable helps investigate the interaction effect of information asymmetries and the lending relationships on loan terms and is expected to overcome the shortcomings of the single-equation approach, as discussed in Section 1.3.1.
1.5.3. Empirical Findings
This section summarizes the findings of the four research questions.
Regarding RQ1, Durbin–Wu–Hausman exogeneity tests first indicate that all of five loan terms are jointly determined for both revolving and term loans. Based on the simultaneous equation model, second-stage regressions then provide further information on the relations between these loan terms for revolving and term loans
separately. Although the findings indicate loan terms are jointly determined for both revolving and term loans, their relations differ across the two loan types. For revolving loans, loan prices have unidirectional relations with non-price terms and each non-price term has a bidirectional relation with the other non-price terms.
For term loans, loan prices also have unidirectional relations with the three loan terms collateral, maturity, and loan size but not loan covenants. This may be because term loans are based less on lending relationships and transaction-based lenders therefore find it costlier to apply covenants to their loans. In addition, for term loans, collateral, covenants, and loan size have bidirectional relations with other non-price terms, but maturity has a unidirectional relation with collateral (collateral impacts loan maturity). In contrast, term loan maturity has no impact on collateral, perhaps because these maturities are much longer and driven by their individual purposes. Therefore, when deciding on collateral, lenders may consider other factors (e.g., borrower characteristics, other loan characteristics, lender characteristics, and macroeconomic factors) instead of maturity. Generally, the answer to RQ1 is yes:
Loan terms (price, collateral, maturity, covenants, and size) are simultaneously determined.
The results of RQ2’s hypothesis test also differ across the two loan types but support the concept that, given their jointness, loan terms are traded off when addressing information asymmetries. The answer to RQ2 is therefore yes. For revolving loans, high information asymmetry borrowers accept higher prices, are more likely to be required collateral, and accept shorter maturities in return for larger loans with fewer covenants. High information asymmetry borrowers may also find it
difficult to borrow from other non-bank lenders since these lenders may have even less information. Such borrowers therefore prefer to borrow larger loans and accept other, less favourable terms (e.g., higher prices, likelier collateral requirements, shorter maturities). For such trade-offs, they are subject to fewer covenants. This may be because lenders may have difficulties designing and monitoring covenants for high information asymmetry borrowers and therefore use other loan terms (i.e., price, collateral, and maturity).
For term loans, borrowers with high information asymmetries accept higher prices, are more likely to be required to secure for their loans, but obtain larger loans.
Covenants are less important in term loans probably because lenders find that their costs exceed their benefits. Designing these covenants and monitoring high information asymmetry borrowers are costly. Given loan term jointness, lenders may instead use loan price and collateral to manage such borrowers.
For RQ3, the results show that lending relationship borrowers obtain loans with more covenants but larger sizes for both loan types. The answer for RQ3 is therefore yes.
Regarding RQ4, this thesis finds no evidence that the impact of information asymmetries on revolving loan terms depends on their prior lending relationships.
Borrowers with higher information asymmetries cannot obtain revolving loans with lower prices, are less likely to be required collateral, or receive longer loan maturities from their relationship lenders. For term loans, this thesis finds that borrowers with higher information asymmetries can obtain larger loans from their relationship lenders. Overall, the answer to RQ4 is no. Higher information
asymmetry borrowers do not benefit from their relationship lenders in terms of lower prices and better non-price terms, except for term loan size.