Using Subordinated Debt as
an Instrument of Market Discipline
2. Evidence on the Potential Market-Discipline Effects of Subordinated Debt
The previous section argued that SND issues have the potential to impose direct and indirect market discipline on banks and bank holding companies. This section examines the available evidence on this potential through a review of the literature, a summary of the views of market participants, and an analysis of new econometric evidence. It ends with the study group's conclusions as to whether the available evidence suggests that SND provide statistically and economically significant market discipline and whether such discipline could be augmented.11
The most common test for market discipline in banking has been analysis of the cross-sectional relationship between interest rates paid on bank liabilities (typically large, uninsured CDs) and various measures of bank risk. Using inside information on the risk of the firm (for example, CAMELS ratings), accounting measures as proxies for risk, or market measures of risk, most studies have found rates to be positively and significantly associated with the risk measures. Additionally, the studies found that ''bad'' news was quickly incorporated into the cost of issuing CDs. A few of the earlier studies did not find evidence of a risk premium, but most did. 12 In fact, even the largest banks, which many would argue were ''too big to fail'' (TBTF) and therefore may have been viewed as having liabilities essentially guaranteed by an implicit safety net, were shown to have risk premiums embedded in their CD rates. One study even found that insured deposits at riskier thrifts included risk premiums.13 Similarly, studies that have examined the relationship between deposit growth and portfolio risk have generally found a relationship consistent with market discipline: Uninsured depositors' holdings at riskier institutions decline relative to those at safer institutions.
More relevant for this study, however, is an assessment of the evidence of market discipline in the market for subordinated debt issued by banking organizations. Again, the standard method to test for market discipline is to analyze the relationship between debt prices, or yield spreads over the rate on Treasury securities, and accounting measures of risk specific to banking organizations. 14 These studies can be divided into three groups. Early studies tested for market discipline in the subordinated debt market by investigating the relationship between the interest rate premium (defined as the rate on subordinated debt minus the rate on matched maturity U.S. Treasury securities) and various risk measures derived from balance sheets and income statements (for example, leverage ratios, measures of profit variability, and loss ratios). Most of these studies did not find much of a statistical relationship between measures of risk and the expected return demanded by investors. 15
The second group of studies improved upon the methodology employed in the earlier studies in a number of ways. The yields on SND were adjusted to account for the value of any embedded call options. 16 The option-adjusted spread over a Treasury bond with matched maturity was calculated and related to balance sheet measures of banking organization risk. After incorporating these methodological adjustments, Avery, Belton, and Goldberg (1988) analyzed banking organization SND data for 1983 and 1984 and still found no evidence of market discipline. 17
Gorton and Santomero (1990) improved upon the methodology of earlier studies by incorporating alternative measures of bank risk and, most important, by demonstrating that the relationship between spreads on large, uninsured bank liabilities and risk cannot be assessed by using a linear function. Rather, they imputed the implied volatility of the bank's assets from a highly stylized valuation model and related those volatilities, which can be shown to be linearly related to risk measures, to bank-specific measures of risk. Using the Avery, Belton, and Goldberg data sample, Gorton and Santomero continued to find virtually no relationship between bank risk measures and the bank's implied asset volatility. Thus, even with the methodological improvements in these more recent studies, the results offer little support for the argument that there was statistically significant market discipline in the banking organization SND market during 1983-84. 18
A more recent study (Flannery and Sorescu, 1996) analyzed SND secondary market data for a longer period and generated empirical results that were consistent with the earlier findings of both Avery, Belton, and Goldberg and Gorton and Santomero and also with the hypothesis that market discipline can be exerted in the SND market. Flannery and Sorescu argue that the apparent lack of market discipline in the earlier studies was most likely a result of real or implied government guarantees during the 1980s. These perceived guarantees were reinforced by the regulatory treatment of SND holders during Continental Illinois's rescue in 1984 and the formalization of the TBTF doctrine by the Comptroller of the Currency in congressional testimony. 19 Flannery and Sorescu argue that, as a result of these actions, the market believed that banking policy would at least partially protect the owners of banks during this period. Holders of SND, which were senior to bank equity, could have rationally believed that they were protected as well.
The implication of this perceived guarantee is that the evidence concerning market discipline in SND secondary markets should vary over 1983-91, with market discipline more apparent near the end of the period. Indeed, Flannery and Sorescu found that firm-specific risk measures were correlated with option-adjusted spreads in 1983-91 for a sample of 422 bonds issued by eighty-three banking organizations. Further, this correlation appears to have increased as conjectural government guarantees weakened in the late 1980s and early 1990s. The option-adjusted spreads on subordinated debt also appear partially to have reflected the market's banking-organization-specific estimate of a government bailout. Of the banking organizations included in Flannery and Sorescu's sample, those included in either the Comptroller's list or The Wall Street Journal's list of TBTF banks paid significantly lower option-adjusted spreads on their subordinated debt in 1985-87 and in 1991 after one takes into consideration accounting and market-based risk measures. 20 Thus, participants in the subordinated debt market appear to be willing to invest in evaluating bankingorganization-specific risks as long as they believe that they are at risk and that their investment is not protected by an implicit or explicit guarantee. 21
The results of the Flannery-Sorescu analysis were affirmed by DeYoung et al. (1998), who used data from 1989-95. Over this period, spreads were found to be closely related to balance sheet and market measures of bank risk. 22
Although SND policy proposals have typically focused on individual banks, the market discipline studies discussed earlier have used data primarily on SND that were issued by bank holding companies. In large part, they have done so because publicly traded SND were and continue to be issued mainly at the bank holding company level.
However, one study that evaluates publicly traded SND issued directly by banks is currently under way. Analyzing SND issues for nineteen banks and forty-one bank holding companies over 1992-97, Jagtiani, Kaufman, and Lemieux (1999) attempt to compare the extent of market discipline imposed on the banks with the extent of market discipline imposed on the bank holding companies. Although the analysis is ongoing and alternative specifications are being tested, the consistent finding, and the one most relevant for our purposes, is that the market does indeed impose a risk premium on SND issued at the bank level. These authors also find that the market tends to price risk more severely at poorly capitalized banks-that is, the spread-risk relationship is nonlinear based on the capitalization of the bank. This information is useful because most SND proposals would require that the debt be issued by the bank.
Finally, although the recent literature on market discipline in SND markets seems to indicate that risk premiums are imposed, there is reason to believe that these studies may underestimate the full extent of such discipline by ignoring banks that did not issue debt. If the decision not to issue debt is associated with the riskiness of the non-issuing banking organization, market discipline will not be captured fully in the studies that analyze only secondary prices of the SND that reached the market. This issue is addressed in the study group's own econometric research, which is discussed in the section ''New Evidence.''
To summarize, most of the literature suggests that the market can account for risk when pricing SND issued by banking organizations. During the periods that SND premiums were not found to be related to risk measures, there is significant evidence that SND holders viewed themselves as not at risk regardless of the riskiness of the debt-issuing bank. During these periods, debt holders were most likely relying on a presumed implicit government guarantee. As this guarantee was decreased through policy and legislative changes in the late 1980s and early 1990s, debt holders came to realize that they were no longer protected from losses, and they responded rationally by more effectively taking banks' risk into account. In short, SND holders appear to be willing and able to invest in evaluating the riskiness of bank assets but do so only when they need to.
A related topic concerns whether the information used by private market participants to discipline banks differs significantly from that available to bank supervisors, either in content or in the timing of its availability. It has been argued that through the on-site examination process, bank supervisors have access to inside information that the market generally does not have. Alternatively, the private market has the strongest incentive to obtain the necessary information to make informed investment decisions. Additionally, the various ''bank-watchers'' may seek different information because they have different roles. Equity holders, for example, may be concerned with the potential for a bank to generate efficiency gains instead of concentrating on the bank's probability of failure. In contrast, the objectives of SND holders and bank supervisors probably align quite well in that both are most interested in protecting against failure. Thus, it would be informative in considering an SND proposal to contrast the availability of information to the different bank watchers.
Most of the recent research in this area suggests that supervisors may temporarily have inside information not immediately available to the market. Dahl, Hanweck, and O'Keefe (1995) found that significant contributions to loan-loss reserves typically occurred immediately after a bank examination, a finding that suggests that new information may be uncovered during exams. Cole and Gunther (1998) attempted to predict bank failure with models using publicly available financial data and then augmented their model with the supervisor's CAMELS ratings to test whether the additional information improved the predictive power of the model. They found that the augmented model more accurately predicted bank failure but only if the CAMELS rating was less than six months old. After six months, the data appeared to be ''stale'' and to have already been incorporated into the market's information set. The finding that information becomes stale, rather typical in the literature, suggests that over time the additional information diffuses into the broader market. Berger and Davies (1994), for example, found that CAMELS upgrades were quickly integrated into market prices (a finding that suggests that banks may have been releasing the new examination information to the market) but that downgrades were incorporated only with a lag (suggesting that, at least for a time, banks were able to keep this bad information from the market but not from examiners).
More directly related to SND proposals, a study by DeYoung et al. (1998) considered the information content of bank examinations as it relates to secondary market spreads of the SND of the bank's holding company. The study compared CAMELS ratings with various market assessments of bank condition and found that bank examination ratings contained additional private information about a bank's safety and soundness. The authors tested whether the market incorporated new private supervisory information into the risk premium paid on holding company debentures with a lag. They concluded that bank exams provided significant new information that was not internalized by financial markets for several months. Unfortunately, the study did not consider the opposite effect: Whether there was information in the private market beyond that to which the examiners already had access.
This last issue was examined by Berger, Davies, and Flannery (1998) when they analyzed the information sets of examiners and private market participants to see ''who knows what when?'' 23 They tested whether private market assessments of the condition of bank holding companies changed before or after supervisors changed their assessments. Similarly, they tested whether information in private markets was preceded by changes in the assessments of supervisors. 24 This study's general conclusion was that supervisory assessments and private market participant assessments complement one another, in that pertinent information obtained by each group is only subsequently incorporated into the other group's assessments. Thus, each group appears to bring new and valuable information to the table, and that information is incorporated with a lag into the other group's information set.
Different market participants (supervisors, bond market participants, rating agencies, and others) appear to generate complementary information that could be useful in the discipline of bank risk-taking. This conclusion, however, is based on relatively few research studies, and more research is clearly warranted.