Skip Header

Federal Deposit
Insurance Corporation

Each depositor insured to at least $250,000 per insured bank

Home > Industry Analysis > Research & Analysis > Working Paper Series > Market Information, Bank Holding Company Risk, and Market Discipline

Market Information, Bank Holding Company Risk, and Market Discipline

Timothy Curry

Gary Fissel

Gerald Hanweck*

Working Paper 2003-04

October 2003

* Timothy Curry and Gary Fissel are senior financial economists, Division of Insurance and Research, Federal Deposit Insurance Corporation (FDIC), Washington, DC 20429. Gerald Hanweck is a visiting scholar, Division of Insurance and Research, FDIC, Washington, DC, and Professor of Finance, School of Management, George Mason University, Fairfax, VA 22030.

The authors thank Audrey Clement, Justin Combs, and Brian Deitch for excellent research assistance; and Andrew Davenport, anonymous reviewers, and participants in the FDIC Research Seminar Series and in the George Mason University School of Management Finance and Accounting Seminar Series for comments on an earlier draft. The views expressed are those of the authors and do not necessarily reflect those of the FDIC or its staff.


This paper assesses the timing and magnitude of equity market valuations of bank holding companies (BHCs) in relation to changes in their risk assessments, as proxied by changes in supervisory ratings for these organizations. In particular, equity market indicators such as market-to-book value, abnormal returns, return volatility, market valuation, price covariance, and trading volume are used to determine if they can provide timely market signals as well as add incremental value to models predicting changes in supervisor-assigned BOPEC ratings of bank holding company risk. To analyze this issue, we took 3,974 bank holding company inspections from a sample of bank holding companies whose stock was publicly traded on major exchanges over the 1988–2000 period. We specify two statistical models: (1) an ordered logistic model, which is used to test the ability of lagged financial market variables, lagged financial accounting data, and past supervisory assessments to predict BOPEC rating changes; and (2) an OLS model, which is used to test the relationship of lagged BOPEC rating changes and lagged financial accounting ratios to predict financial market variables. These models taken jointly are used to test the hypotheses that (1) equity market information adds to the ability to forecast changes in banking company risk, as measured by changes in BOPEC ratings, and (2) supervisory risk ratings have the ability to lead stock market valuations of banking companies’ performance. The analysis is conducted for three distinct economic and banking periods: recession and banking crisis (1988–1992), economic recovery (1993–1995), and economic expansion (1996–2000).

The findings for the first model show that financial markets can add forecast value to financial accounting data and supervisory factors in predicting future BOPEC ratings. The results reveal a relationship between market indicators and supervisory rating changes of BHCs, reflecting risk conditions that flow from market valuations to supervisory rating changes. The findings for the first model were statistically significant for all three of the periods studied. Out-of-sample forecasts show that market information sometimes improves the ability of the models to forecast upgrades, downgrades and no-rating changes. This suggests that a multiple-model approach may be superior to one using a single model to forecast BHC risk assessments. The findings for the second model reveal that regressing financial market variables on lagged BOPEC changes and lagged financial accounting information is only moderately suggestive of a relationship between lagged supervisory rating changes and equity market variables. In summary, the findings provide empirical support for the presence of market discipline to the extent that the hypothesized market variables add incremental value to the model in predicting changes in bank holding company risk ratings. To this extent, the financial markets appear to be providing some degree of independent oversight to BHC management, besides the oversight provided by bank supervisors and holding company directors.


The prospects of using market information for the identification of risk in banking organizations have become a focus of bank supervision in recent years. This concern has arisen in part because of the rapid growth of large, global banking organizations whose portfolios and global operations have become increasingly complex in scope and increasingly opaque to supervisors and financial markets.1 In this regard, bank supervisors are attempting to glean any information that may assist them in identifying the changing risk structures of these firms. This interest in turning to markets for assistance has been accelerated by empirical research that suggests bank supervision might benefit from using information embedded in the capital market valuation of banking organizations’ debt and equity securities.2 This work has found that financial market information can be used to supplement and complement the traditional supervisory practices of on-site examinations and off-site monitoring. If markets are meaningfully efficient and investors are able to monitor and/or anticipate emerging risks of organizations, then investors' expectations become embedded in financial market information.3 Market assessment and the pricing of risky management policies may restrain risky behavior (direct market discipline) and/or assist supervisors in the monitoring process by providing timely signals on the changing risk patterns of these firms (indirect market discipline). The potential use of the signaling features of markets for modern bank supervision and discipline offers promise, as evidenced by the calls for including market indicators into the supervisory process that are being voiced at the highest levels of the U.S. bank regulatory agencies and the U.S. Congress and, on the international front, by the Basel Committee on Bank Supervision.4

The interest in using market information to assist in bank supervision is a fairly recent phenomenon in the United States. Historically, bank supervisory officials have questioned the ability of outsiders such as uninsured depositors, other unsecured creditors, Wall Street analysts, capital market investors, or other market participants to accurately evaluate the risks of bank portfolios without the confidential information acquired through on-site examinations. Bank supervisors have access to highly sensitive information, such as the payment histories of performing and nonperforming loans, the level of loan classifications, the adequacy of loan-loss reserves and bank capital, and a close-up view of managerial abilities and expertise. During intervals between examinations, regulators depend primarily on quarterly, unaudited financial statements submitted by commercial banking companies to monitor changes in financial condition. While supervisors expend considerable resources uncovering idiosyncratic information on bank financial condition, there are limitations to the process. On-site examinations take place only at widely spaced times, usually every 12 to 18 months, and the information may become outdated after only brief intervals.5 Also, bank examinations generally document past occurrences, such as changes in the credit quality of loan portfolios. In addition, accounting data used in off-site monitoring programs are reported with several months' delay, are released as of a single date, are usually unaudited, and often require considerable analysis and screening to yield a meaningful interpretation of financial condition.6

Financial markets, on the other hand, are fluid and forward looking in their appraisal of financial condition and, unlike the extended cycles between examinations, continuously incorporate and discount new information into security prices. While markets may not have access to extensive amounts of confidential information about financial condition as bank inspectors do, markets may be more timely and accurate in evaluating public information released in financial statements.7 Markets also have the ability to garner information about financial condition from a vast array of other sources, such as an institution's board of directors, market analysts, and industry observers. For these reasons, it has been suggested that markets may have information similar to the information of bank supervisors and may update it continuously, and that market participants have the profit incentive to expend resources to better forecast banking company profitability and risk taking.

The goal of this paper is to assess the timing and magnitude of equity market valuations of bank holding companies (BHCs) in relation to changes in supervisory ratings for these organizations. In particular, market indicator variables are used to determine if they can provide timely market signals as well as add incremental value in predicting changes in regulatory assigned BOPEC ratings for institutions whose stock was traded on major exchanges over the 1988–2000 period.8 To accomplish this, we conducted a univariate statistical analysis and specified an ordered logistic model to test for time lags in the relationship between measures of financial market valuation and the assignment of supervisory ratings, using methods similar to those presented by Berger, Davies, and Flannery (2000) and Gunther, Levonian, and Moore (2001). In addition, to investigate the possibility that supervisory changes are reflected in future market indicators—a reverse causality—we developed OLS regression models of market variables on changes in BOPEC ratings and financial statement variables used in the logistic models.

The main hypothesis tested is that, like rating assignments by Moody’s and Standard and Poor's, BOPEC rating evaluations lag by several quarters changes in the financial condition of bank holding companies. This occurs because of recognition, decision, and implementation lags associated with the examination process. For example, time is required to recognize the changing condition of banking companies, schedule special or extraordinary examinations or visitations, revise examination ratings, share the results with management, and rechart bank policies. Because of these lags, it has been suggested that market indicator variables such as price volatility, abnormal returns, return volatility, trading volume, and others may be observed to lead BOPEC rating changes.

Conversely, it is possible that early monitoring by supervisors leads to changes in the financial condition of the organization that are reflected in the next quarterly report so that a change in a BOPEC rating is nearly concurrent with changes in financial performance as shown in the financial reports. Also, banking companies may attempt to manage their financial reporting so that supervisors and markets are not alerted to deleterious changes that could trigger a review or even a downgrade. For these reasons, changes in market indicators may not lead changes in BOPEC ratings.

Most previous studies that have examined market signaling and discipline focused on the subordinated debt (subdebt) markets. It was generally thought that concerns of investors in these markets were more closely aligned with those of bank supervisors. It has also been suggested that equity market prices are unsuitable for monitoring purposes because the relationship between equity prices and bank default risk is not as readily apparent as that of the pricing of unsecured subdebt.9 However, the equity markets may have some advantages over the debt markets. These markets are generally deeper and more liquid than the debt markets. They are generally thought to be more efficient in pricing behavior than the debt markets, with liquidity premia playing a smaller role in the pricing of the securities. Also, data are more readily available from the equity markets with many more financial institutions being publicly traded than issuing traded subordinated debt securities.

Our univariate results show that changes in bank holding company BOPEC ratings clearly track changes in financial condition as reflected in quarterly bank accounting and market data. They also show that the dominant BOPEC rating assignment for the three different periods analyzed was “no change” in rating. This finding indicates that in specifying models that attempt to predict regulatory rating changes, the no-rating-change option must be accounted for simultaneously with the other supervisory options of rating downgrades and upgrades. This has generally not been done in previous studies that have examined this issue; they have principally focused on the market effects of downgrades.

The results from the ordered logistic regressions show that one-quarter lags of market data add to the identification of bank holding company BOPEC changes for all three of the periods examined when these data are used in conjunction with quarterly financial data and past rating information. On the basis of these results, we conclude that equity market data provide an independent source of timely information and/or a different perspective in predicting BOPEC rating changes beyond those provided by quarterly accounting and other supervisory data for the three periods examined. Considering the comprehensive sample used in this study, our findings support those reported in the most recent empirical literature and extend them to periods primarily before FDICIA and during the recent banking expansion.10 We go further to provide extensive out-of-sample tests and show that market information improves the ability of the models to forecast upgrades and downgrades. We also show that, with the use of financial data alone for common 2001 out-of-sample data, models based on 1993–1995 data improve forecasts for upgrades, and models using 1996–2000 data substantially improve forecasts for downgrades. This suggests that a multiple-model approach to identifying banking company risk conditions may be superior to an approach using a single combined model.

To further investigate the timeliness of supervisory versus market assessments of BHC condition, we pose the additional hypothesis that supervisory rating changes can forecast changes in market variables (the reverse of the above analysis). The direction of influence, from markets to supervisory action or vice versa, is important because it establishes the timing, which may be concurrent, of the influence of supervisory rating changes. As noted in Berger, Davies, and Flannery (2000), supervisory and market assessments may be concurrent, implying that each may be emphasizing or finding different information. This direction of influence from markets to regulatory actions or vice versa is not a matter of exogeneity or endogeneity but is important to identify in order to establish any significant role of supervisory rating changes influencing market variables. We accomplish this by regressing market variables against lagged BOPEC rating changes and financial ratio variables (in other words, information that the market would know from the published financial reports of the BHC). The results suggest that lagged BOPEC ratings are statistically significant, at least at the 10 percent level, in only 6 of 15 possible tests (five separate market variables for three periods). Although a Granger causality test could not be directly performed, these results indicate that the predominant relationship is one of market information as predicting future banking company risk rather than changes in supervisory ratings predicting market variables. This finding reveals a relationship between financial market data and bank risk assessment—a necessary and fundamental condition for market discipline to be effective.

The next section discusses the empirical literature. The subsequent section presents an overview of BOPEC ratings migration behavior over the 1988–2000 period. This is followed by an overview of the general conceptual approach of this study. The data and sample are then discussed, followed by a presentation of the univariate results. The final sections present the ordered logistic regression and OLS models, the findings, and the conclusions.

Empirical Literature

As mentioned above, most studies that have examined the usefulness of market signaling and discipline have focused on the subordinated debt markets.11 The empirical evidence shows that subordinated debt spreads generally increase as a bank’s risk increases. More recent work focuses on equity market valuations, which generally support the notion that equity market variables do add value in providing timely information. Berger and Davies (1998) use event study methodology and find that the equity market anticipates upgrades in regulatory ratings but follows downgrades. Berger, Davies, and Flannery (2000), using a sample of 184 bank holding companies over the period 1989:Q4 to 1992:Q2, find that regulators acquire information sooner than bond rating agencies and the equity markets, but the regulatory assessments are less accurate in predicting the future performance of bank holding companies than either stock or bond market indicators unless the regulatory ratings have been recently assigned.

Several studies have incorporated market data into traditional default or bankruptcy models to determine if the information adds value in identifying troubled institutions. For example, Elmer and Fissel (2000), with a sample of 93 failed banks for the years 1989–1995, found that equity market variables can be used to augment accounting-related information to predict bank failures several years before failure. Curry, Elmer, and Fissel (2001) found that equity market variables when combined with bank financial data help predict downgrades and upgrades in bank and thrift CAMEL ratings. Krainer and Lopez (2001, 2003) find that equity market variables, such as stock returns and equity-based expected default frequencies, can be useful to bank regulators for assessing the financial condition for bank holding companies over the 1990–1999 period. Using event study methodology, they found that equity market measures provide timely information by anticipating supervisory rating changes by up to four quarters. However, they found no evidence that combining supervisory and market data improves the out-of-sample forecasting of BOPEC ratings in a statistically significantly way. Gunther, Levonian, and Moore (2001) found that a measure of financial viability based on equity market data (expected default frequency, or EDF, from KMV) helps predict the financial condition of bank holding companies as reflected in their supervisory ratings.12

This paper adds to existing empirical work in a number of ways. First, we use equity market variables along with quarterly financial data and past supervisory information to predict bank holding company BOPEC rating changes over three very different economic and banking periods—recession and banking crisis (1989–1992), recovery (1993–1995), and expansion (1996–2000—and we do so for a substantial sample, which varies from over 3,500 to 3,900 new, full-scope bank holding company inspections. Previous empirical work has generally not focused on such a broad time horizon under such varying economic conditions or with a comparable sample size.13 Our approach has several benefits. First, breaking out the analysis for such disparate periods will allow for testing the stability of the models during these different periods, allowing us to compare the findings with those of the previous literature and to identify models able to forecast upgrades and downgrades. Second, an ordered logistic model is specified to account for all possible choices available to bank supervisors in assessing changes in the financial condition of BHCs; these choices consist of rating upgrades, downgrades, and no change in ratings. By simultaneously modeling the dominant rating event of “no change” along with downgrades and upgrades, we overcome some of the limitations of earlier work, where the primary focus was on predicting either the rating levels or downgrades. We expect our approach to yield improved predictive power over previous models. Third, unlike some of the previous work, we focus on predicting newly assigned BOPEC rating changes rather than past rating levels. The estimation of downgrade, upgrade, and no change models should allow for a more complete focus on the contribution of market data to the analysis.

BOPEC Ratings Migration Behavior

"Table 1 displays the frequency and migration behavior of BOPEC rating assignments over the 1988–2000 period, broken down into the three periods under observation. The column on the left shows the initial BOPEC rating (1–5), and the columns to the right are the newly assigned ratings from an inspection. Thus, each cell shows the percentage in that category for newly assigned BHC inspections.14 Only certain rating categories can be upgraded or downgraded or both. For example, institutions in the highest BOPEC rating category of 1 can only remain within the same category or be downgraded as the result of new inspections. These data show that for the 1988–1992 period, 68.5 percent of institutions inspected remained in the highest rating category of 1 from the previous inspections; 27.1 percent were downgraded to a 2 rating, 1.7 percent went to a 3, and 2.8 percent migrated to a 4 over this period. Conversely, those in the lowest rating category of 5 can only remain in that category or be upgraded. For example, 81.8 percent of the 5 rated remained in this category, while only 18.2 percent were upgraded to the 4 level. Those in the middle categories (BOPEC ratings 2, 3, and 4) could either be upgraded, downgraded, or remain the same. Reading down the diagonal shows the proportions of inspections that remained in the initial categories, and reading across to the left or right shows the number of upgrades or downgrades. Following the diagonal reveals that 71.1 percent of the companies in the 2-rated group remained in the same category they began in, as did 57.8 percent of the 3-level companies and 62.2 percent of the 4-rated group.

One can observe trends in migration patterns over time by comparing the matrices for the three different time periods (table 1). Financial conditions are seen improving as reflected in newly assigned BOPEC ratings, as we move out of the recession toward expansion. For example, 68.5 percent remained in the highest rating category over the 1988–1992 period, increasing to 88.5 percent in the 1993–1995 period before sliding slightly for 1996–2000 period to 87.0 percent. Conversely, 81.8 percent remained in the worst category—5—in the 1988–1992 period, declining to no 5-rated institutions by the 1996–2000 period. As another example, while 62.2 percent of the BOPEC 4-rated BHCs remained 4s in the1988–1992 period, by the 1996–2000 inspections, only 33.3 percent remained in this category, with a larger share (50.0 percent) being upgraded than downgraded (16.7 percent).15

The key observations from the rating migration matrices are, first, that while rating patterns were changing throughout the economic cycle, the likelihood or probability of remaining in the same rating category, or “no rating change,” was significantly higher than that of being downgraded or upgraded; and second, that downgrades and upgrades performed very differently over the cycles. Downgrades dominated rating changes in the 1988–1992 period, while upgrades dominated in the other two periods. These data suggest that analysis by period is important. (See Figure 1).

Data and Sample

The data consist of quarterly Y-9C bank holding company financial reports, confidential bank holding company (BOPEC) ratings, confidential CAMELS ratings, and quarterly financial data on lead banks of the holding company. Equity market information was obtained from the Center Research Security Prices (CRSP). CRSP data contain historical information such as daily stock prices, daily returns, equal and value-weighted indexes of market returns, dividend information, daily trading volume, and other variables for all organizations publicly traded on national exchanges. To obtain information for only commercial banking companies, we matched CRSP against historical Y-9C holding company files and quarterly financial data (Call Reports) back to 1984. We then matched these files against bank holding company BOPEC ratings. All quarterly data for the market indicator variables used in the analysis were calculated from the daily stock price and trading information provided by CRSP.

From this universe of bank holding company and CRSP matches, we obtained a sample for analysis. For the univariate analysis, which focuses on individual financial accounting and financial market variables, we used a sample of 3,974 full-scope bank holding company inspections. For each inspection, there is a BOPEC rating assignment. For the logistic and OLS regression models, the samples differ slightly from those in the univariate analysis because it was necessary for this analysis that each holding company have quarterly financial data, CRSP market data, and inspection information for the time it was in operation over the 1988–2000 period. To meet these requirements, we used a slightly reduced sample of 3,534 full scope inspections for the logistic analysis, which resulted in 458 BOPEC downgrades, 461 BOPEC upgrades, and 2,615 inspections with no rating changes. For the OLS regression analysis, the number of BOPEC inspections analyzed declined further to 3,068. The reduction in the OLS sample was due to the lagging of certain variables in the regression, which led to a loss of some observations. Only full-scope inspections with assigned composite ratings of 1, 2, 3, 4, or 5 are considered because other on-site reviews, such as limited-scope inspections and visitations, generally do not produce as much complete information on the financial condition of the entire holding company organization as do full-scope examinations. For each inspection, data were collected on the date when each began, the date it ended, and the date when the institution was notified of its assigned BOPEC rating. To analyze the timeliness issue, we selected a benchmark or “event” quarter. The event quarter is the one in which the board of directors of the holding company was informed by bank supervisors of the final rating from the inspection.

No action was taken to deal with merged institutions over the 1988–2000 period. As long as a bank holding company was publicly traded, remained on CRSP, and filed quarterly financial reports, the institution remained in the sample. Acquired institutions disappeared from the sample after acquisition, when they no longer filed financial reports. It is not anticipated that this approach for dealing with mergers would have any important effects on the analysis. One possible problem may arise if a merger was randomly announced in the quarter preceding a bank holding company inspection, such that changes in market information would reflect the anticipated effects of the proposed merger and not necessarily the current financial condition of the company. Like most other studies in this area, this one made no attempt to control for this possibility.


BOPEC rating changes are the result of inspections by trained field examiners. The process by which these changes occur forms the foundation for estimation of the models. We consider the changes in ratings in terms of time series where supervisors can choose to either change a rating or leave it the same. Thus, in contrast to failure models, once a bank is assigned a BOPEC rating, it need not remain there. With the exception of an initial BOPEC rating of 1 as indicated above, a bank holding company can be upgraded or remain the same, and with the exception of an initial BOPEC rating of 5, a holding company can be downgraded or remain the same. On the basis of these supervisory options and the empirical evidence presented, the underlying conceptual model of the formation of decisions by supervisors to change a banking company’s BOPEC rating is that the no-change option must be explicitly considered.

The approach of this paper is to use an ordered logitistic regression model to better account for the broader decision-set open to bank supervisors in assessing the condition of a BHC. The logit model predicts BOPEC ratings as a function of several factors, including institution size, past supervisory information, current financial data, and market information. Data for these variables are taken from one quarter preceding the inspection quarter, when possible.16 In this regard, the information set is drawn to closely approximate the state of knowledge that bank supervisors should possess about the institution immediately before an inspection.17 The model is then run at a one-quarter interval before the quarter in which the new rating is assigned. The model estimates the probability for the three possible outcomes of the inspection: a downgrade, an upgrade, or no rating change. A dependent variable of 1 is specified for institutions that were downgraded, a 2 for institutions whose rating remained the same, and a 3 if the institution was upgraded, thus ordering the regression. For classification purposes, the critical level of probability is determined by the size of each group as a proportion of the total sample.18 If the market variables significantly add to the prediction of BOPEC ratings, they then provide an independent source of information about BHC risk apart from the other explanatory variables.

An OLS model is specified to determine if confidential BOPEC ratings can predict selected market variables. In this instance, the five market variables are specified as the dependent variables and the BOPEC ratings as the independent variables. The purpose of a second set of regressions is to validate the time sequence of market indicators of financial condition to measured assessments of banking company financial condition as reflected in the BOPEC rating assignments. If BOPEC ratings, controlling for information contained in book-value performance ratios, can predict market indicators, this will be taken as evidence that on-site BHC inspections reveal new information concerning the institutions’ performance. If there is little or no relationship, the market may be interpreted as having incorporated the underlying information of BOPEC rating changes and valued the company accordingly.

Univariate Analysis

Sample means for supervisory, financial and market variables for bank holding companies broken out by period and by BOPEC rating category are presented in tables 2 to 4. The data show a high correlation between financial condition and examination rating. Reading across the first row of each of the three tables reveals that the number of days from the last BOPEC inspection generally declines with poorer ratings, as expected, as they are inspected and monitored more frequently than healthier banks. Also, weak BOPEC ratings for bank holding companies tend to be associated with unsatisfactory performance at the bank level as well as reflected in CAMELS ratings of the lead banks. Also, BOPEC ratings from the prior inspection are highly correlated with the current assignment of the rating. In addition, those with the weakest BOPEC ratings tend to have larger asset-quality problems, a greater reliance on large time deposits, and higher loan-loss provisions. Those with poorer ratings also have lower levels of capital and earnings, as expected.

The market valuation of the firm follows a similar path tracking the financial data and thus shows a high correlation with financial condition. The weakest-rated firms exhibit relatively lower returns, increased volatility of returns, lower market valuations, and greater trading volume. The degree of association between all BOPEC ratings and variables is measured statistically by the Van der Waerden test and reflected in the p values shown on the three tables.19 As shown, there is a high degree of association between the BOPEC rating category and the financial condition of banking companies, with most variables being statistically significant at better than the 1 percent level. With few exceptions (abnormal or excess returns for the 1993–1995 period and loans past due 90 days to total assets in the 1996–2000 period), these results are consistent across all three periods analyzed.

Multivariate Analysis (One-Quarter Lag)

The subsections below describe the variables used in the regression models as well as the empirical results. The estimated coefficients for the ordered logit regression are interpreted in terms of the effect on the likelihood of a downgrade.

Size of Institution

The first independent variable is the log of total assets (LN_ASSET), which is a control variable for differences in institution size and is measured as the natural logarithm of total assets. The log form of total assets is taken to account for the potential nonlinearity in the variables between firm size and rating assignments. To the extent that an institution’s size provides for large-scale diversification, economies of scale and scope, access to the capital markets, and an overall source of support for the entire organization, it should reduce the likelihood of experiencing a BOPEC rating downgrade. Thus, a negative relationship is anticipated between the probability of a downgrade and institution size.

Past Supervisory Information

To help predict individual BOPEC rating changes, we included three variables that reflect past supervisory inspections. The first is a dummy variable (BOPEC-1 to BOPEC-4) that reflects the BOPEC rating from the most recent holding company inspection before the opening of the current inspection. The probability or likelihood of being downgraded from the BOPEC-1 to BOPEC-4 level is interpreted relative to the omitted BOPEC rating, which in this instance is the BOPEC-5 rating. As mentioned the composite or overall BOPEC ratings are assigned from 1 to 5, with 1 being the highest and 5 being the lowest rating. It is anticipated that firms with the most favorable ratings are the ones most likely to experience a BOPEC rating downgrade. Thus, a positive relationship is anticipated.

Another dummy variable is included to capture the relationship between the last CAMELS rating of the lead bank of the holding company and the assignment of the current BOPEC rating of the parent.20 Changes in examination ratings on the bank level can lead to changes in ratings on the organizational or holding-company level. Like the BOPEC ratings, CAMELS ratings for banks carry a value of 1 to 5, with 1 being the highest and 5 the lowest. The rating is taken from the most recent examination record of the largest or lead bank in the holding company system before to the quarter in which the current inspection of the parent holding company occurs. A dummy variable (PROB_BK) is assigned a value of 1 if the previous rating of the lead bank was a problem-bank rating of 3, 4, or 5, and a 0 if the rating was a favorable 1 or 2. A positive relationship is anticipated between an existing problem-bank rating for the lead bank of the holding company system and the likelihood of a BOPEC rating downgrade for the parent holding company.

Since earlier work has shown that bank examination ratings can become outdated in relatively short periods following an inspection, a variable is included that captures the age of the last BOPEC inspection before the most recent inspection under review (INSP_AGE). This variable is calculated as the number of days from the last holding company inspection to the start date of the current inspection. Hence, the older the prior rating, the greater the age of the inspection and the more likely that there will be a change in rating. Experience suggests that this change is more likely to be a downgrade because more frequently inspected BHCs are most likely to have poorer past BOPEC ratings, and those inspected less frequently will have better BOPEC ratings and are more likely to be downgraded.

Financial Accounting Variables

Capital adequacy is measured by two ratios: equity to assets (EQ_Asset) and loan-loss reserves to assets (LLR_ASSET). High levels of capital and reserves provide a cushion to absorb loan losses and thus reduce the likelihood of insolvency, so a negative relationship is expected between this variable and the probability of a downgrade. The credit quality of the loan portfolio is captured by loans past due 90 days or more (PD90_ASSET), those in nonaccrual status (NA_ASSET), and the ratio of the provision for loan losses to assets (PROV_ASSET); provision for loan losses is income allocated to loan-loss reserves. All three variables are expected to be positively related to the likelihood of a downgrade because higher levels of delinquent loans and loan provisions to rebuild loan-loss reserves generally indicate credit quality deterioration and increasing levels of financial distress.

The overall profitability of the institution is measured by the return-on-assets variable (ROA), which is expected to be negatively related to future downgrades. This means that a higher ROA implies a greater likelihood of an upgrade and a lesser likelihood of a downgrade. Two measures of liquidity are posited: the ratio of liquid assets to total assets (LIQ_ASSET) and total deposits held in accounts greater than $100,000 (TD100_ASSET). The level of liquid assets ratio is expected to be negatively related to future financial distress, reflecting that higher levels of securities to assets provide sources of additional liquidity in difficult times. However, the amount of deposits held in accounts greater than $100,000 (TD100_ASSET) is expected to have a positive sign, reflecting a more aggressive, expensive, and volatile funding strategy.

Financial Market Variables

The market variables are specified into five categories reflecting different aspects of equity market valuations; these categories are: price volatility, abnormal returns, volatility of returns, capitalization, and share turnover. These market data are based on daily information for each quarter from CRSP. The first variable captures stock price volatility (COVAR_PRICE) and is measured by the coefficient of variation in equity prices. COVAR is defined as the standard deviation of the quarterly stock price divided by the average quarterly price. This variable is expected to be positively related to the likelihood of a rating downgrade. Market abnormal or excess returns are captured by the EX_RETURN variable, which measures the differences between the cumulative quarterly return of each stock and the cumulative quarterly return of an index of market performance. The simple cumulative quarterly return is calculated by multiplying 1 plus the daily return for each stock i on day t (1 + rit) across all trading days in each quarter, then subtracting 1. The index of market performance chosen is the CRSP value-weighted index, which covers all publicly traded institutions.21 The EX_RETURN measure is expected to be negatively related to the future downgrades.

An interesting aspect of market data is that various aspects of financial theory effectively extend the list of market-related variables beyond prices and returns. For example, Merton’s (1973) option pricing model implies a direct relationship between stock return volatility and asset return volatility conditioned on asset values. This relationship indicates that greater stock return volatility will increase the likelihood of insolvency and is directly associated with the likelihood of a downgrade.22 As investors become concerned over financial distress and potential insolvency, more variation may appear in return patterns. Wang (1994) ties trading volume to the flow of information about a firm’s financial health, suggesting that trading volume should rise as information about financial distress is released.23 Although a comprehensive analysis of market-related variables goes beyond the scope of this paper, return volatility and trading volume nevertheless are two variables that are easily observed and, according to financial theory, should contain predictive content.

Return volatility is captured by the standard deviation of the quarterly returns (STD_RETURN) and is expected to be positively related to future downgrades. Trading volume within each quarter is identified by the TURNOVER variable. This is calculated as the percentage of shares traded during the quarter divided by the number of shares outstanding at the end of the quarter. This measure is expected to exhibit a positive relationship with the likelihood of being downgraded. The market valuation of a firm is captured by market equity divided by the book equity (MKT_BKEQ), with both values measured at the end of the quarter. This variable is expected to be negatively related to the probability of a downgrade. That is, the higher the value of the ratio, the lower the likelihood of financial distress and downgrade.

Definitions of the variables used in the regression equations along with their means and standard deviations over the 1988–2000 period are presented in tables 5 and 6. The sample means show that the financial variables clearly track the different economic cycles over these periods, including the recession and banking crisis of the late 1980s and early 1990s, the recovery in the middle 1990s, and the prosperity of the late 1990s. From the start to the end of these periods, financial statements showed significant improvement. Earnings and capital increased significantly, loan-loss risk declined, and firm liquidity improved. However, strong loan growth was being fueled in part from increases in volatile uninsured liabilities on the balance sheet. Financial markets also responded favorably to the changes in firm condition over this period, as witnessed by significant increases in firm valuations as recognized by the market-to-book value ratios and a slightly reduced volatility of share prices and earnings.

The ordered logistic regression equation follows the following format:

[D]formula 1

where α1 and α2 are constants estimated for the ordered logit with three groups (downgrades, upgrades, and no change),24 β1 is the coefficient for firm size, β2, …β7 are coefficients for a vector of variables representing past supervisory opinions unique to each BHCi , β8…β16 are coefficients for a vector of variables for financial condition based on quarterly accounting information for each BHCi, and β17 …β21 are coefficients for a vector of variables for selected market measures for each BHCi. The eit is assumed to be distributed as a logit function. The regression results are estimated for the three different economic cycles: recession and banking crisis (1988–1992), recovery (1993–1995), and expansion (1996–2000). As a forecasting model, all independent variables are lagged relative to the BOPEC rating of the event quarter that is being predicted. As mentioned, we ordered the logistic regression model by relating the specified variables to the probability or likelihood of receiving a BOPEC downgrade.25

The empirical results for the logistic models are shown in tables 7,8,9. Specification 1 for each table provides a traditional default or rating downgrade prediction model by specifying past supervisory information and quarterly accounting data. These variables are accompanied in each specification by information that accounts for firm size. Specification 2 in each table launches the analysis of equity market variables by isolating the predictive power of these variables in the absence of the financial data but including past supervisory information. Specification 3 combines both financial and market data into one equation to determine if market information adds incremental predictive value to the financial and supervisory data in identifying firms likely to be downgraded. Although the equity market variables need not dominate the traditional ratio-based model, a minimum level of competence is required to justify a conclusion that market-related variables are a meaningful addition to traditional analysis. In the absence of incremental predictive value, market signals may be viewed as redundant information with little supervisory value. Also as mentioned, the variables in each regression are lagged for one quarter before the event quarter, which is the quarter when a new inspection rating is assigned—the rating that the model is attempting to predict.

1988–1992 Results

Specification 1 for the 1988–1992 period shows that institution size (LN_ASSET) is important and that, as expected, larger institutions are less likely to be downgraded (table 7). Also, past supervisory information is highly associated with the likelihood of future BOPEC downgrades when firms with the best ratings are most likely to be downgraded (BOPEC-1 to -4). The dummy variable representing the previous CAMELS rating of the lead bank (PROB_BK) of the holding company shows that when lead banks have problem CAMELS ratings, the result is a significantly higher likelihood of the holding companies' being downgraded. Contrary to what most previous literature predicts, the time lapsed from the previous inspection (INSP_AGE) is not a significant predictor of future downgrades. The financial accounting variables in the model perform largely as expected, with eight of the nine variable coefficients significant at mostly the 1 percent level. The Max-Rescaled R2, a measure of goodness of fit, for the model is 0.58.26 Since specification 1 provides a working downgrade prediction model, specified with the traditional capital, asset-quality, earnings, and liquidity ratios along with past supervisory information, it is used as a benchmark equation for comparing the predictive content of the market-based variables.

Specification 2 contains only the size, supervisory factors, and stock market variables, excluding the financial accounting information discussed above. In this equation, institution size is negative and significant, indicating that larger-sized firms are less likely to be downgraded. As before, almost all the variables associated with previous supervisory evaluations are highly significant at the 1 percent level. Of the five market variables, four are significant: the volatility of price (COVAR_PRICE), the value-weighted excess returns (EX_RETURN), the standard deviation of the returns (STD_RETURN), and the share turnover variable (TURNOVER), all of which are significant at the 1 percent level. The explanatory power of the regression is less than that of specification 1, with an Max-Rescaled R2 = 0.39.

Specification 3 forces the stock market-based variables to compete with the financial accounting variables used in specification 1 to test for any added predictive power for the equation. As before, institution size continues to be statistically significant and negatively related to the prospects of future downgrades. The prior BOPEC rating and past supervisory assessments, with the exception of inspection age (INSP_AGE), continue to be highly significant, with almost all coefficients being significant at the 1 percent level. The financial variables continue to be strong, with seven of the eight also being highly significant.

For the market variables, four of the five have the correct signs and are statistically significant. The benchmark comparisons between specifications 1 and 3 are performed through a likelihood ratio test, which is a chi-square statistic. If the likelihood ratio statistic is positive and statistically significant, we can conclude that the financial and market variables when combined into a single equation cause it to possess a better fit than the equation it is tested against. The likelihood ratio test is positive and significant at the 1 percent level, which supports the contention that the market variables do add some explanatory power to the model in comparison with specification 1. In addition, other measures of the overall fit of the regression indicate improved results. The Max-Rescaled R2 statistic of 0.60 is higher than that of specification 1. Also, the Akaike Information Criterion (AIC), which measures the overall fit of the equation, is lower for specification 3 relative to specifications 1 and 2, which indicates a greater association of the combined variable set with the BOPEC rating changes.27 For the 1988–1992 period, we conclude that market variables provide an independent source of information and add significant explanatory power to known financial and supervisory data.

1993–1995 Results

The regression results for the 1993–1995 economic recovery period are displayed in table 8. The finding for the asset-size variable generally carries the expected negative sign but is significant only for specification 1. The coefficients for the past supervisory inspection variables for specifications 1–3 are similar to the previous period’s results, with the BOPEC dummies and the problem-bank dummy of the lead bank all exhibiting the correct signs and all exhibiting statistical significance at the 1 percent level. Previous BOPEC ratings continue to indicate that firms with the most favorable past ratings are the most likely to be downgraded in the future relative to a BOPEC rating of 5. The inspection age variable (INSP_AGE) continues to be insignificant. The results for the financial variables in specification 1 for the 1993–1995 period are not as consistent as for the previous period. Of the nine variables measuring financial condition, five are significantly related to downgrades: the level of holding company equity (EQ_ASSET), past due 90 days (PD90_ASSET), nonaccrual loans (NA_ASSET), the return on assets (ROA), and large deposits (TD100_ASSET).

The market variables in specification 2 show that all variables have the correct signs and that three of the five variables are statistically significant from the 1–10 percent level: the coefficient of variation of price (COVAR_PRICE), the standard deviation of the returns (STD_RET), and the market equity to book equity variable. The explanatory power of specification 2, with only the market variables, has a Max-Rescaled R2 of 0.44, which is lower than the R2 of 0.52 for specification 1. In addition, the AIC measure for the overall fit is higher for specification 2 than for specification 1, which indicates a lower overall fit for specification 2. These findings show that the market variables in isolation do not possess as much predictive power as do the financial variables by themselves.

Specification 3, which combines the financial with the market variables into one model for the 1993–1995 period, does improve upon specification 1. The Max-Rescaled R2 is slightly higher for specification 3 relative to specification 1, the AIC variable is lower, and the log likelihood ratio test is again significant, indicating that this specification continues to improve the predictive power over the first specification. In summary, the findings for the 1993–1995 recovery period are similar to those of the 1988–1992 recession in that market data continue to add some value to the overall regression, but overall the results are not as strong as those for the earlier period.

1996–2000 Results

The results for 1996–2000, a period of substantial economic growth and banking expansion, are presented in table 9. The log of asset size has a consistently negative sign as expected, but only the first specification is it significantly related to the likelihood of being downgraded. The level of past BOPEC ratings remains the most consistent predictor of future downgrades in specifications 1–3, showing that institutions with the highest ratings are again the most likely to be downgraded. The prior CAMELS rating of the lead bank (PROB_BK) is also a reliable predictor of future downgrades and is consistent with the previous two periods analyzed. Contrary to expectations, the inspection age variable (INSP_AGE) is negative and significant for all three specifications. All of the financial variables in specification 1 have the correct signs, and most are significant at the 1 percent level. The Max-Rescaled R2 of specification 1 is 0.45. In specification 2, four of the five market variables are significant from the 1–5 percent level. The Max-Rescaled R2 for the equation is 0.35 percent, or slightly below the first specification, and the value for the AIC variable is higher, which indicates a less favorable overall fit for specification 2 relative to specification 1.

Specification 3 shows the combined model for all of the variables for the 1996–2000 period. In this equation, six of the nine financial variables and three of the five market variables are significant. The overall fit of the model in specification 3 is better than for either of the first two specifications. The Max-Rescaled R2 is higher and the AIC variable is much lower than for either of the first two specifications, indicting an improved fit for the model. In addition, the likelihood ratio test, which compares the explanatory power of specification 3 with benchmark specification 1, is positive and significant at the 1 percent level, indicating improved predictive power. In summary, the empirical results for the 1996–2000 expansion period are similar to the other two periods but overall are relatively more robust than for the 1993–1995 transition period. They show that the market indicator variables contribute in the hypothesized direction to the identification of downgraded institutions for this period.

In-Sample Classification Results

Table 10 contains in-sample classification results for the ordered logit regressions summarized in tables 7–9. In the formation of these in-sample classifications, the a priori likelihood of being a downgrade, no change, or upgrade is the a priori frequency ratio of these outcomes in each classification category. We determined the critical level of probability for classifying each BOPEC change of a downgrade, no change, or an upgrade as correct or incorrect by taking the number of firms that experienced each of these BOPEC changes in the sample for each period analyzed and dividing by the number in the total sample. We then compared this critical level of probability or a priori probability with the predicted conditional likelihood of the model to determine the appropriate classifications. These classifications appear in each of the cells in table 10.

The in-sample predictive accuracy of the model for the 1988–1992 period shows that the correct classifications for the BOPEC downgrades increase slightly, going from 81 percent to 83 percent from specification 1 to specification 3, or an increase of six institutions with the addition of the market variables. The correct classification of those firms that had no change in ratings remained about the same at about 78 percent, while the identification of firms that were upgraded actually declined. For the 1993–1995 economic recovery period, the correct classifications for the downgrades improved, going from 56 percent to about 63 percent, while the no-change ratings and upgrade predictions were about the same. The in-sample classifications for the 1996–2000 period shows that the correct downgrade and upgrade predictions were about the same, but the no-change category improved from 64 percent to about 67 percent (34 additional institutions). While the overall results of the model show some improvement in classification accuracy for all periods and for most groups, the differences are not generally large in percentage terms. The main contribution of the inclusion of market variables is the identification of the BOPEC rating downgrades for both the 1988–1992 and 1993–1995 periods, when the correct classifications increased from about 2 percentage points to 7.

Out-of-Sample Classification Analysis

The out-of-sample classification analysis provides a test of the accuracy and forecast potential for the model when equity market variables are added to the financial accounting variables. In performing the out-of-sample classifications, the a priori cutoff probabilities were calculated in a similar fashion as for the in-sample tests. Furthermore, when classifying the out-of-sample 1993–1995 period using the 1988–1992 model (panel A of table 11), we use the observed frequency ratio for the 1988–1992 sample to determine the cutoff probabilities. We chose this method because we did not want to bias the test, since in practice actual frequencies of future periods would not be known and an unbiased estimate of future frequencies is the frequency of the sample data of the prior period. The out-of-sample tests show that of the 18 classifications in table 11 (panels A–F)—six out-of-sample classification tests, each with three classifications groups—the addition of market variables shows an improvement in 11 classifications.

The out-of-sample classifications show that the FA, SM and the combined models mostly show improvement in the different categories of rating change, including the no-change category. For example, using the 1988–1992 model on the 1993–1995 and 1996–2000 out-of-sample data shows that the FA model (panels A and B in Table 11) is more accurate than the SM or combined models in classifying upgrades, and the SM model is much more accurate than the FA or combined models in classifying downgrades. In addition, the combined model is more accurate than either in classifying no changes.

To further explore the differences in out-of-sample classification precision for each of the estimated models for the three periods examined (1988–1992, 1993–1995, 1996–2000), we apply each of the models to a common out-of-sample test using 2001 data as a common period (panels D–F of table 11). The findings indicate that the model that has the highest overall correct classifications varies according to the period in which the estimation occurred. For example, when the 1988–1992 model estimated from data for a recessionary and banking crisis period is used, the combined model has the most accurate overall classification (panel D), while for the 1993–1995 and 1996–2000 periods , the FA model overall does better than either the SM model or the combined model (panels E and F). In another example for specific categories rather than overall performance, the 1996–2000 model applied to the 2001 data (panel F) indicates that the FA model does the best in classifying downgrades (88 percent) and the SM model does the best in classifying upgrades (96 percent). These results provide further evidence that using financial accounting and/or market variables can improve forecasting changes in bank holding company risk assessments.

The implication of these results is that the models estimated from different periods in the economic and banking cycles can give different and sometimes improved forecasts of BOPEC rating changes and, by inference, risk assessments. One drawback to using different models to provide forecasts is that a BHC’s inspection may be classified in each of the BOPEC change categories. When this arises, a simple rule would be to choose to classify an inspection in the group that, if a misclassification, is least costly from a supervisory perspective or in the group in which its estimated likelihood of membership is greatest. This multiple-models approach to classification can provide improved forecasting and best integrates market data into the supervisory process.

BOPEC Risk Rating Changes and Financial Statement Variables as Predictors of Financial Market Factors

Berger et al. (2000) compared the timeliness and accuracy of when BHC supervisors assessed changes in BHC financial condition with the timeliness and accuracy of when the financial markets assessed the changes. In their analysis, they estimate models to forecast equity market assessments using lagged market variables and supervisory assessments.28 They find that supervisory assessments add little to the prediction of market assessments.

We conduct a similar analysis by considering the relationship between BOPEC rating changes and lagged market variables when, unlike what happens in the analysis above, the lagged BOPEC rating changes become predictors of market variables. For the testing of this latter relationship, the market variables used in the logit models are specified as dependent variables and are regressed against the lagged BOPEC rating changes and lagged financial statement ratios, including the natural logarithm of BHC assets, as independent variables. Although some of the market dependent variables are bounded from below by zero, we use OLS estimation. However, OLS should provide reasonable estimates and should not cause serious problems because these variables are never observed to reach the lower bound. For the sake of brevity, we discuss only those results when there was a statistically significant relationship between the lagged BOPEC rating changes (BPchgL1) and the financial market variables. This occurs only five times for the three periods and for four variables: the standard deviation of quarterly returns (STD_RETURN), the quarterly value-weighted excess return (EX_RETURN), the market value of equity to book value equity ratio (MKT_BKEQ), and the covariance of price (COVAR_PRICE). Tables 12–16 show the regressions for each of the five market variables for each of the three periods. The model is identified as follows:

[D]formula 2

where the financial accounting variables are those used in the logistic regressions and are defined above. The lagged BOPEC change, BPchgL1, takes on a value of 1 for a downgrade, 2 for no change, and 3 for a ratings upgrade. A negative sign for the BOPEC rating change coefficient, β1, means that a movement toward a downgrade will increase the dependent variable compared with no change or an upgrade. All variables are defined in table 5.

The relationship between the standard deviation of return (STD_RETURN) and the change in the BOPEC rating (BPchgL1) has the hypothesized negative sign and is statistically significant for the 1988–1992 and 1996–2000 periods (table 12). The implication is that a prior BOPEC downgrade will be associated with a higher return volatility for a BHC, indicating a worsening financial condition going forward. Additionally, most of the signs for the financial statement variables are according to expected relationships, and many are statistically significant. For example, a greater value of book equity to assets, EQ_ASSET, and ROA suggests that the STD_RETURN would be lower. Less leverage and greater earnings would be expected to reduce stock return volatility, ceteris paribus. The overall fit for this regression is statistically significant, with an Adjusted R2 of 45 percent, 58 percent, and 17 percent in the 1988–1992, 1993–1995, and 1996–2000 periods, respectively.

The relationship between the quarterly value-weighted excess return (EX_RETURN) and the change in the BOPEC rating (BPchgL1) has the hypothesized positive sign and is statistically significant at the 5 percent level or better only for the 1988–1992 period (table 13). The implication of this relationship is that a prior BOPEC downgrade will be associated with a lower excess return, indicating a worsening financial condition. Additionally, some of the signs for the financial statement variables are according to expectations, and many are significant. However an interesting set of relationships is that of loans past due 90 days or more to assets (PD90_ASSET) and loan charge-offs (CHARG_ASSET). The estimates show different signs and are all statistically significant and imply that a greater value of PD90_ASSET indicates a decrease in EX_RETURN, whereas an increase in CHARG_ASSET will increase EX_RETURN. It would be expected that greater values for all of these variables would tend to be associated with lower excess returns because of the greater riskiness of the company that an increase in their values would indicate. One interpretation of the opposite signs is that greater reported past-due loans suggest greater future charge-offs, whereas greater actual charge-offs reflect recognized losses and are viewed by the market as getting bad assets off the books. It is also interesting to note that ROA carries the expected positive sign. The overall fit for this regression is statistically significant but small, with an Adjusted R2 of 9.1 percent for the 1988–1992 period.

The relationship between the lagged change in the BOPEC rating and the market-to-book value variable (MKT_BK) is statistically significant at the 5 percent level or better and has the hypothesized positive sign for only the 1988-1992 period. The interpretation is that a bank holding company that had a previous rating downgrade will show a deteriorating market-to-book value in the following quarter. Furthermore, several financial statement variables are highly significant for each of the periods and usually have the expected signs. The variables loans on nonaccrual status (NA_ASSET) and ROA stand out in this regard. The overall fit for this regression is statistically significant, with an Adjusted R2 of 20.9 percent in the 1988–1992 period and 16 percent and 33 percent for the other two periods.

The relationship between the lagged change in the BOPEC rating and the coefficient of variation of price (COVAR_PRICE) is statistically significant at the 5 percent level or better and has the hypothesized negative sign for only the 1996–2000 period. The interpretation of this relationship is that a bank holding company that had a previous rating downgrade will show a rising coefficient of variation of price in the subsequent quarter. Furthermore, several financial statement variables are highly significant for the period and usually have the hypothesized signs. The variables EQ_ASSET, LPROV_ASSET, and CHARG_ASSET stand out in this regard. The coefficient of variation is hypothesized to decline as EQ_ASSET increases, because of the decreased use of leverage. The interplay between loan-loss provisions (a prospective indicator of future charge-offs) and revealed charge-offs in the previous period is consistent with the hypothesis that taking charge-offs is current recognition of bad loans and will decrease price volatility, whereas higher provisions are an indicator of greater future charge-offs (a positive sign). The overall fit for this regression is small but highly statistically significant, with an Adjusted R2 of 5.1 percent in the 1996–2000 period.

The stock volume turnover variable, TURNOVER, showed no relationship for any period with BPchgL1. However, a number of the financial variables were statistically significant with anticipated signs. The overall fits were statistically significant for all periods and showed an R2 of 27.1 percent, 25.1 percent, and 91 percent for the 1988–1992, 1993–1995, and 1996–2000 periods, respectively.

The conclusion that we draw from these results is that BOPEC rating changes have minimal predictive power for stock market variables. With statistical significance in only 6 of a possible 15 regressions and only 2 of these for the 1996–2000 period, this conclusion seems justified. However, there is ample evidence from these regressions that lagged financial statement variables, being readily known to the market before an inspection, have some consistent predictive power in every period. This in itself suggests that previous financial performance has some predictive power of equity market variables.


This paper explores the notion that financial market-based variables may be used to augment financial statements and past supervisory risk assessments for the purpose of predicting future supervisory risk assessments as embodied in BOPEC rating changes. This analysis is conducted with the use of a sample of publicly traded bank holding companies with 3,974 full-scope inspections over the 1988–2000 period. This period is characterized as representing three distinct economic and banking cycles: recession and banking crisis (1988–1992), economic recovery (1993–1995), and economic and banking expansion (1996–2000). Focusing on different economic and banking climates allows for testing the stability of the model over these varying periods and permits a comparison of the findings with the previous literature. It should be noted that even though we are analyzing changes in BOPEC ratings arising from full-scope inspections, which are supervisory events, we are also implicitly analyzing changes in the stringency of regulatory regimes. It should be recalled that the 1988–1992 period was considered a “near-death” experience for commercial banking, while the 1996–2000 period witnessed an exuberant recovery, with market-to-book ratios often exceeding four for some of the larger banking companies.

To investigate the direction of causality from market variables to changes in BOPEC ratings or vice versa, two empirical models are estimated. The results for the first model (ordered logistic) show that market indicator variables were able to add value in predicting BOPEC rating changes when used in conjunction with lagged supervisory and financial accounting data for each of the three periods examined. These findings are consistent with earlier studies investigating the role of equity market indicators in identifying the changing patterns of bank financial risk, but the results show stronger relationships when analyzed by sub-periods. For example, through six different out-of-sample classification tests, we show that the use of stock market information generally improves out-of-sample forecasts.

A second model (OLS) was specified to determine if lagged BOPEC rating changes, when used with financial accounting information, were capable of predicting the equity market variables used in the ordered logistic model estimation. The results were only moderately successful in predicting the financial market variables and were statistically significant for only 6 of the possible 15 cases. We interpret these results as being supportive of an information flow from changes in equity market factors to changes in BHC supervisory risk ratings.

The findings in this paper provide some empirical support for the presence of market discipline to the extent that the hypothesized market variables add statistical and incremental value to the model in predicting changes in bank holding company BOPEC ratings. Thus, the results indicate that the market has been able to glean some independent information about BHC financial risk exposure in the period leading up to the inspection beyond the information that was known from publicly available financial reporting sources. To this extent, the financial markets appear to be providing some degree of independent oversight for management besides that of bank supervisors and holding company directors, thereby providing the fundamental conditions for market discipline to be effective.


Basel Committee on Banking Supervision (2001). Working Paper on Pillar 3-Market Discipline. Bank for International Settlements.

Berger, Allen N., and Sally M. Davies. 1998. The Information Content of Bank Examinations. Journal of Financial Services Research 14, no. 2:117–44.

Berger, Allen N., Sally M. Davies, and Mark J. Flannery. 2000. Comparing Market and Supervisory Assessments of Bank Performance: Who Knows What When? Journal of Money, Credit, and Banking 32, no. 3:641–67.

Cole, Rebel A., and Jeffery W. Gunther. 1998. Predicting Bank Failures: A Comparison of On- and Off-site Monitoring Systems. Journal of Financial Services Research 13, no. 2:103–17.

Curry, Timothy, Peter Elmer, and Gary Fissel. 2001. Regulator Use of Market Data to Improve the Identification of Bank Financial Distress. Working Paper 2001-01. FDIC.

———. 2003. Using Market Information to Help Identify Distressed Institutions: A Regulatory Perspective. FDIC Banking Review 15, no. 3:1-16.

Dahl, Drew, John O’Keefe, and Gerald Hanweck. 1998. The Influence of Examiners and Auditors on Loan-Loss Recognition. FDIC Banking Review 11, no. 4:10–23.

Elmer, Peter J., and Gary S. Fissel. 2000. Forecasting Bank Failure from Momentum Patterns in Stock Returns. Unpublished manuscript. FDIC.

Evanoff, D. D., and Larry Wall. 2001. SND Yield Spreads as Bank Risk Measures. Journal of Financial Services Research 19:121–46.

Fama, Eugene F., and Kenneth R. French. 1993. Common Risk Factors in the Returns on Stocks and Bonds. Journal of Financial Economics 33:3–56.

Federal Reserve Board of Governors and U.S. Treasury Department. 2001. The Feasibility and Desirability of Mandatory Subordinated Debt.

Fissel, G., G. Hanweck, and J. O’Keefe. 2002. A Comparative Analysis of Modeling Methodologies of Financially Distressed Banking Firms. FDIC.

Flannery, Mark J. 1998. Using Market Information in Prudential Bank Supervision: A Review of the U.S. Empirical Experience. Journal of Money, Credit, and Banking 30, no. 3:273–305.

———. 2000. The Faces of Market Discipline. Unpublished manuscript. University of Florida.

Flannery, Mark J., and Sorin M. Sorescu. 1996. Evidence of Bank Market Discipline in Subordinated Debenture Yields: 1983–1991. Journal of Finance (September): 1347–77.

French, Kenneth R., Schwert, G. William, and Robert F. Stambaugh. 1987. Expected Stock Returns and Volatility. Journal of Financial Economics 19, no. 1:3–29.

Gajewski, Gregory R. 1989. Assessing the Risk of Bank Failure. In Proceedings of a Conference on Bank Structure and Competition, 432–56. Federal Reserve Bank of Chicago.

Gallant, A. Ronald, Peter E. Rossi, and George Tauchen. 1992. Stock Prices and Volume. Review of Financial Studies 5, no. 2:199–242.

Goodhardt, C. A. E. 1988. The Evolution of Central Banks. MIT Press.

Gunther, Jeffery W. and Robert R. Moore. 2003. Loss Underreporting and Auditing Role of Bank Exams. Journal of Financial Intermediation. 12; 153-177.

Gunther, Jeffery W., Levonian, Mark E., and Robert R. Moore. 2001. Can the Stock Market Tell Bank Supervisors Anything They Don’t Already Know. Federal Reserve Bank of Dallas Economic and Financial Review. 2nd quarter.

Gupta, Atul, and Lalatendu Misra. 1999. Failure and Failure Resolution in the U.S. Thrift and Banking Industries. Financial Management 28, no. 4:87–105.

Hanweck, Gerald A., and Lewis J. Spellman. 2002. Forbearance Expectations and the Subordinated Debt of Bank Insolvency. Paper presented at the joint FDIC and Journal of Financial Services Research conference on pricing deposit insurance. Washington, DC.

Jordan, John, Joe Peek, and Eric Rosengren. 2000. The Market Reaction to the Disclosure of Supervisory Actions: Implications for Bank Transparency. Journal of Financial Intermediation 9:298–319.

Krainer, John, and Jose A. Lopez. 2001. Incorporating Equity Market Information into Supervisory Monitoring Models. Working Paper 2001-14. Federal Reserve Bank of San Francisco.

———. 2003a. How Might Financial Market Information Be Used for Supervisory Purposes? Federal Reserve Bank of San Francisco Economic Review, 29–45.

———. 2003b. Using Securities Market Information for Supervisory Monitoring. Unpublished manuscript. Federal Reserve Bank of San Francisco.

Levonian, Mark. 2001. Subordinated Debt and the Quality of Market Discipline in Banking. Unpublished manuscript. Federal Reserve Bank of San Francisco.

Merton, Robert C. 1973. Theory of Rational Option Pricing. Bell Journal of Economics 4 (Spring): 141–83.

———. 1974. On the Pricing of Corporate Debt: The Risk Structure of Interest Rates. Journal of Finance 29:449–70.

Meyer, Lawrence H. 1998a. Remarks before the Washington Conference of the Institute of International Bankers. Federal Reserve Board, Washington, DC. March 2.

———. 1998b. Remarks before the Spring 1988 Banking and Finance Lecture. Widener University, Chester, PA. June 9.

———. 1998c. Remarks before the Bank Administration Institute's Finance and Accounting Management Conference. Federal Reserve Board, Washington, DC. June 9.

———. 1998d. Remarks at the 16th Annual Monetary Conference, Money in the New Millennium: The Global Financial Architecture. Cato Institute, Washington, DC. October 22.

Pettway, Richard H. 1980. Potential Insolvency, Market Efficiency, and Bank Regulation of Large Commercial Banks. Journal of Financial and Quantitative Analysis, 15, no. 1:219–36.

SAS Institute, Inc. 1996. SAS/STAT Software: Changes and Enhancements. Cary, NC.

Simons, Katerina, and Stephen Cross. 1991. Do Capital Markets Predict Problems in Large Commercial Banks? New England Economic Review (May/June): 51–56.

Sinkey, Joseph E., Jr. 1975. A Multivariate Analysis of the Characteristics of Problem Banks. Journal of Finance 20:21–36.

Tanoue, Donna. 2001. Remarks to the 100th Annual Meeting of the Conference of State Bank Supervisors, Traverse City, MI. May 18.

Wang, Jiang. 1994. A Model of Competitive Stock Trading Volume. Journal of Political Economy 102, no. 1:127–68.

1 Another reason is that the increasing size and scope of banking organizations have contributed to the organizations' ability to cause a systemic event that would have catastrophic market effects.

2 For some recent studies on this issue, see Flannery and Sorescu (1996); Flannery (1998, 2000); Berger and Davies (2000); Berger, Davies, and Flannery (2000); Curry, Elmer, and Fissel (2001); Krainer and Lopez (2001, 2003); Gunther, Levonian, and Moore (2001); and Evanoff and Wall (2001).

3 The issue of the opacity of bank assets has been studied extensively in the academic literature. The weight of the evidence suggests that investors in debt and equity securities do price changes in bank risk, although some studies disagree. For a review of some of this literature, see Krainer and Lopez (2003).

4 See Tanoue (2001). Also, the Gramm-Leach-Bliley Act of 1999 directed the Federal Reserve Board and the Treasury Department to analyze the value of requiring large banking organizations to issue subordinated debt. See also the comments from Federal Reserve Board Governor Lawrence Meyer (1998a, 1998b, 1998c, 1998d).

5 See Cole and Gunther (1998).

6 For example, it has been shown that there is a correlation between bank examinations and the level of loss recognition for commercial loans. Also, enhanced loan-loss charge-offs have been associated with the timing of examinations of commercial banks. This suggests that accounting information contained in the publicly available quarterly financial statements may not always capture the risk profile of the reporting bank. See Dahl, O’Keefe, and Hanweck (2000); Gunther and Moore (2003).

7 There is some evidence for this statement. See Berger, Davies, and Flannery (2000).

8 The bank holding-company-system rating is the acronym BOPEC, assigned by the Federal Reserve Board, and it stands for Banking subsidiaries, Other (nonbanking) subsidiaries, Parent company, consolidated Earnings, and Consolidated capital. A rating from 1 to 5 is assigned for each component, with 1 being the best and 5 the worst. A composite rating also from 1 to 5 is assigned, reflecting the overall condition of the organization.

9 However, Levonian (2001) has shown that equity and debt market information should produce similar results. See also Hanweck and Spellman (2002).

10 See Gunther, Levonian, and Moore (2001) and Krainer and Lopez (2001, 2003a, 2003b).

11 See Flannery (1998), Flannery and Sorescu (1996), Evanoff and Wall (2001), and the Federal Reserve Board of Governors and U.S. Treasury Department (2001).

12 Some early work on the use of the equity markets as an early-warning system shows mixed results. Pettway (1980) found that stock returns of 6 large banks destined for failure signaled problems almost a year (38 weeks) before regulators began the examination process that led to the banks’ being classified on the problem-bank list. Conversely, Simons and Cross (1991) analyzed a more recent sample of 22 bank holding companies and found that regulators appear to have been aware of problems before the market recognized them.

13 Krainer and Lopez (2001) have a roughly comparable period (1990–1999) and sample size but their model focuses primarily on predicting rating levels rather than new inspections.

14 These migration rates consider all full-scope inspections that occurred over the period in question. As a result, a BHC will appear and be counted as many times as it is inspected during the period.

15 The number of BHCs was increasing over the 1988–2000 period so that a comparison by number of companies or inspections is meaningless. To get an idea of the relative distribution of companies by initial BOPEC rating, see tables 2–4.

16 Financial and market data are taken from one quarter before the inspection quarter in the model. The previous examination ratings are taken from the examinations on record that were given before the current ratings were assigned.

17 While preliminary Call Report and bank holding company data are first released about 45 days after the end of the quarter, bank examiners will have access to the most recent information available before starting an examination, regardless of when the data are released to the public by the supervisors. Furthermore, banking firms typically release financial statement information within two weeks after the end of the quarter, so this information is available to the market well before the released Call Report data. It is for these reasons that we use only a one-quarter lag for the financial data in the analysis. Regardless, we ran the regressions with two-period lags for the financial data and found no significant differences in the results.

18 Thus, for example, if the critical probability from this calculation is 50 percent, then if the predicted probability from the model is equal to or greater than .5, the institution would be classified as belonging to that particular group.

19 The Van de Waerden procedure tests whether the distribution of a variable has the same location parameter across different groups.

20 The CAMELS acronym stands for each component of the examination: Capital, Assets, Management, Earnings, Liquidity and Sensitivity to market risk.

21 The CRSP value-weighted index of returns for all publicly traded companies was chosen over one for the banking industry alone so as to compare banking company returns with the entire market in which all banking company equities compete and to avoid spurious correlation of an industry index dominated by a few large companies.

22 French, Schwert, and Stambaugh (1987) lend empirical support to this view by documenting a positive relationship between the volatility of market returns and market excess returns (market return minus T-bill yield).

23 Gallant, Rossi, and Tauchen (1992) find supportive empirical evidence that large daily price movements are followed by high trading volume.

24 Two constant terms are estimated under the assumption that the proportional odds among the groups are independent of the explanatory variables such that the slope parameters are the same for each group. In general, to separate three groups (downgrade, no change, and upgrade) requires two planes (or hyperplanes for more than three dimensions). In the case of the ordered logit that we estimate, the planes are parallel. This means that the probability of observation it's belonging to group 1, conditional on the explanatory variables, is CL(α1+ß’Xi), the conditional probability of its belonging to group 2 is CL(α2+ß’Xi) - CL(α1+ß’Xi), and the conditional probability of its belonging to group 3 is [1 – CL(α2+ß’Xi)], where CL() is the cumulative logit function. Throughout our analysis the constant terms are found to be significantly different.

25 The model can be “ordered” to estimate the probability of a rating downgrade, no change, or upgrade. However, the results presented in the tables refer to the likelihood of a downgrade’s occurring.

26 The Max-Rescaled R2 statistic, like the pseudo-R2, is a measure of goodness of fit. However, the pseudo-R2 for discrete models has a maximum of less than 1.0. The Max-Rescaled R2 statistic adjusts the pseudo-R2 such that the statistic reaches a maximum of 1.0, in line with the standard notion of R2 (SAS Institute, Inc. [1996], 414).

27 The AIC statistic is a goodness-of-fit statistic similar to the maximum likelihood ratio test, with a lower value indicating a better fit.

28 Berger et al. use three equity market variables: abnormal quarterly return on a BHC’s common stock (AR), quarterly change in the proportion of outstanding shares held by institutional investors (INST), and the quarterly change in the proportion of outstanding shares held by officers and directors (INSIDER). Our excess return variable, EX_RETURN, attempts to measure similar market conditions as AR, but our measure does not impose an estimated time series model to compute excess returns. However, our findings, which include book-value financial ratios in the regressions, show little forecasting power of equity market variables by supervisory variables and are similar to those of Berger et al.

Last Updated 11/12/2003 Questions, Suggestions & Requests

Skip Footer back to content