Information > Manual 172 > This page
First, we consider whether
the banking organization's risk, size, and frequency of SND issuance affect
its probability of issuance. In table
2, parameter estimates for a ''bare bones'' probit model with only
accounting measures of risk, bank holding company size, and an indicator
variable for whether the bank holding company issued in the previous six-month
interval are presented for separate two-year sample periods. In the first column, the
parameter estimates for the 1986:Q2 to 1987:Q4 period suggest that accounting
measures of risk individually did not significantly affect the SND issuance
decision. Indeed, the five accounting measures of risk taken together
did not provide significant explanatory power during this period. In the 1988:Q1 to 1989:Q4 period and in the 1990:Q1 to 1991:Q4 period, the parameter estimate for the ratio of accruing loans past due ninety days or more to total assets is significant and has the expected negative sign. In the latter of these two periods, the parameter estimate for the ratio of total book liabilities to the sum of market value of common stock and book value of preferred stock is also significant and is of the expected negative sign. Also, the five accounting measures of risk, taken together, have a significantly negative effect on the issuance decision both in the 1988:Q1 to 1989:Q4 period and in the 1990:Q1 to 1991:Q4 period. These results are also consistent with Flannery and Sorescu, who found that secondary SND prices were significantly affected by such accounting measures of risk in the 1989-91 period. These results also suggest that riskier banking organizations chose not to issue SND during 1988-91. In each of the next three
periods (1992:Q1 to 1993:Q4, 1994:Q1 to 1995:Q1, and 1996:Q1 to 1997:Q4),
the accounting measures of risk were as likely to have positive parameter
estimates as to have negative ones. Indeed, taken together the five accounting
measures of risk had a significant We first consider figure 2, which shows box plots for such spreads
in each quarter for the 1991:Q1 to 1997:Q4 period. These box plots are
graphical representations of the center and width of a distribution, along
with outliers. The height of each black box is equal to the interquartile
width, which is the difference between the third quartile and first quartile
of data. This width has narrowed considerably since 1992:Q1. It is also
notable that in the middle to late 1990s the top quartile of the SND spread
over comparable-maturity Treasuries is below the medians of such spreads
(which are represented by horizontal lines in the interior of each box)
that were observed during 1991. The brackets ([ ]) for each box plot are
located at extreme values of the data for the quarter or at a distance
equal to 1.5 times the interquartile distance from the center, whichever
is less. To interpret the latter result,
we consider the possibility that SND spreads during 1993-97 did not reflect
the risk of banking organizations to the same degree that spreads did
in the early 1990s. In all two-year periods considered in table 2, the coefficients on the frequency of issuance and the banking organization's asset size are of the expected (positive) sign, and almost all are significantly different from zero at the 5 percent level. These positive coefficients are consistent with studies of SND secondary market spreads over comparablematurity Treasuries, which find that larger banking organizations have narrower secondary market spreads than smaller banking organizations. The only discernible pattern in the significance of the coefficients is that issuance in the previous six-month interval is becoming more important over time. This finding suggests a trend toward more-frequent issuance by some top fifty bank holding companies in recent years, a view that was supported by study group interviews with market participants. We next consider how private
benefits, bond market volatility, and macroeconomic conditions are estimated
to affect the decision to issue SND by top fifty bank holding companies.
In table 3, parameter estimates
for a probit model that includes the ''bare bones'' model variables, additional
bank-specific variables, and variables for bond market and macroeconomic
conditions are presented. Based on the findings presented in
table 2, we provide parameter estimates for the period in which market
discipline is strongest (1989-92) and for the entire sample period (1986:Q2
through 1997:Q4). With regard to private benefits, neither the variable
for foreign and domestic income taxes as a percentage of net income ( -
The exception is that the first sample period is only eighteen months. -
The Wald test of the restriction that all accounting measures of risk coefficients are zero was not significant at the 5 percent confidence level. -
The Wald tests reject the restrictions that all five coefficients of the accounting measure of risk are zero at the 5 percent confi- dence level. Moreover, the sums of all the marginal effects were positive for each of the three periods. -
For data having a Gausian distribution, approximately 99.3 percent of the data fall inside the brackets. Horizontal dashes represent ''unusually deviant data points'' that are further than 1.5 times the interquartile distance from the center of the box. -
This interpretation is not inconsistent with market discipline studies that have found evidence of spread sensitivity to risk in the mid-1990s (for example, Jagtiani, Kaufman, and Lemieux, 1999). The interpretation requires only that spreads be relatively insensitive to a banking organization's risk, which does not rule out the possibility that the spreads are statistically sensitive to an organization's risk. -
An alternative explanation for the positive coefficients on risk variables is that relatively risky banks during 1988-91 were unable to issue SND and, therefore, issued in 1992-97 to ease the banks' pent-up desire to issue debt.
NOTE. All specifications include a constant term that was significant at the 5 percent level. Year indicator variables, which were equal to one in the first year of each panel and zero otherwise, were also included, though those coefficient estimates are not reported here. Numbers in parentheses are t-statistics.
NOTE. All specifications
include a constant term that was significant at the 5 percent level. Year
indicator variables, which were equal to one in each specific year of
each panel and zero otherwise, were also included for all years except
the first year, though those coefficient estimates are not reported here.
Numbers in parentheses are |