Does a Bank’s Loan Screening and Monitoring Matter?

When Fields et al. (2006) decompose the loans into new and renewal loans, they find that the decline in the abnormal loan announcement return is in renewal, rather than new, loans.

Coleman et al (2006) use the term “monitoring” to include both the pre-lending screening of loan applications and the post-lending monitoring of the borrower’s performance.

The fixed effects regression uses annual panel data drawn from the Call Reports of the 1,000 largest U.S. banks in each year between 1994 and 1998. Coleman et al. (2006) has a detailed discussion of the model, independent variables used for the fixed effects regression, and the results. The independent variables include asset and liability composition, fee income, number of branches, profitability, and size.

Coleman et al. (2006) note that the use of the fixed effects parameter to proxy loan screening and monitoring is similar to the use of fixed effects parameters from estimates of the stochastic cost function to measure cost efficiency (see Berger et al. 1993).

The choice of sample period reflects the time horizon used by Coleman et al. (2006) in deriving their MONITORING EFFORT proxy.

The estimation period used in similar studies include [−170, −21] in James (1987), [−200, −51] in Billett et al. (1995), and [−140, −21] in Preece and Mullineaux (1996).

This is the approach taken in previous studies of the borrowing firm’s abnormal loan announcement return (e.g., Billett et al. 1995). There is little evidence of industry clustering in our sample (i.e. no cross-sectional dependence) and Binder (1998) notes that time series dependence in the average abnormal return is not important where the event period is short relative to the estimation period. Although it is not reported in our results, we also calculated test-statistics using the time-series standard deviation method (Brown and Warner 1980). This method controls for cross-sectional correlation of stock returns. However, it does not control for heteroscedasticity in the security returns. The test-statistics for the time-series standard deviation method are similar in significance to the t-statistics calculated assuming both cross-sectional independence and constant variance and confirm that cross-sectional dependence is a “minor problem” in our sample.

More recently, however, Fields et al. (2006) have found that the difference in the borrower’s abnormal return for syndicated and single lender loans is insignificant in the 1980 to 2003 period.

Refinancing loans involve the replacement of an existing debt obligation with another debt bearing different terms. As refinancing loans may involve different lenders from those involved in the debt that is being refinanced, they are not equivalent to loan renewals as examined by Lummer and McConnell (1989).

Billett et al. (1995) find a significant negative relationship (at the 10% level) between the cumulative pre-announcement return and the announcement period return.

The z-statistic tests for the significance of the standardized abnormal return, assuming cross-sectional independence (Patell 1976). The t-statistic, which is also reported in Table 3, measures the significance of the CAR, rather than the standardized CAR.

The results in Panel A are robust to the benchmark used (e.g. market-adjusted abnormal returns).

Billett et al. (1995) find that the percentage of positive abnormal returns on the event day is 54.1%, significant at the 5% level.

The abnormal return reported by Fields et al. (2006) is not standardized, and is equivalent to our CAR.

However, the median CAR of 0.31% for the (0,+1) event window is not statistically significant.

Lummer and McConnell (1989) find a mean CAR over a two-day window of 0.61%, significant at the 1% level, for a sample of 728 loans from 1976 to 1986. Preece and Mullineaux (1996) find a mean CAR over a 2-day window of 1.00%, significant at the 1% level, for a sample of 446 loans from 1980 to 1987.

The sub-samples are not of equal size, as loans from the same bank are placed in the same sub-sample.

This feature is ignored in Cook et al.’s (2003) study of the effect of lender certification on loan spreads.

In earlier versions of this paper we used a lead bank group means regression to handle the clustering problem. In effect, all loans for a bank were combined into a single averaged observation, resulting in a sample of 31 bank observations. The group mean results were similar to the cluster corrected results reported below. However, with data averaging within banks and the limited degrees of freedom, these results are not as reliable as the cluster corrected OLS regression results.

The maximum adjusted R ² in Fields et al. (2006, p. 1206) is 0.04.

In contrast, Lummer and McConnell (1989) find an insignificant negative coefficient for term loans.

James (1987) finds no significant difference between the share price response for refinancing loans vis-à-vis loans used for capital expenditures.

We use the formulation interacting the loan screening and monitoring proxy with the revolver and mixed loan indicator variables (i.e. Reg 4 of Table 4). Similar results are obtained using Reg 1 of Table 4.

In contrast to our results, Billett et al. (1995) find that SD RESID has a positive and significant coefficient while BETA has a negative but insignificant coefficient.