Timothy J. Curry
Financial Economist
Federal Deposit Insurance Corporation
Washington, D.C. 20429
(202) 898-7372
tcurry@fdic.gov

The views stated here are those of the authors and do not necessarily reflect those of the Federal Deposit Insurance Corporation. The authors would like to thank Audrey Clement and Justin Combs for extensive research assistance. Sarah Belanger assisted with the tables. Peter J. Elmer was a senior financial economist at the Federal Deposit Insurance Corporation at the time that the authors conducted this research.

This paper assesses the extent to which stock market information may help bank regulators identify bank financial distress. The research specifies a variety of stock return and other market-related variables that might contain elements of longer-term trends and be capable of anticipating changes in regulatory ratings of commercial banks and thrift institutions. Univariate tests confirm a remarkable tendency for market-related variables to decline, or otherwise move, far in advance of formal regulatory rating downgrades, a finding suggesting that these variables may have useful predictive content. Furthermore, multivariate tests support the notion that market-related variables add predictive value to the value contained in publicly available Call Report financial data. The evidence supports the use of market-related variables in off-site monitoring applications.

JEL Classification: G1, G23

During the past two decades, research on the relation between market signals and bank financial health has repeatedly implied that bank supervision might benefit from using information embedded in stock market prices, returns, or other market-related data. This implication was first voiced by Pettway (1980) after he found that stock returns of 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. More recently, Flannery (1998) reiterates the theme, pointing out that regulators might use market data either to reduce the time required to recognize problems or to increase the accuracy of forecasts of future changes in bank condition.

In theory, it is not clear whether stock price, return, or other market-related data should lead, move contemporaneously with, or follow regulatory awareness of problems. The theory of efficient markets suggests that the market evaluates and prices public information at the fastest rate possible. Because the primary source of bank financial information, the quarterly Report of Condition and Income (Call Report), is available to regulators and the market at about the same time, market efficiency suggests that market awareness of Call Report-related problems could precede regulator awareness.¹ However, managers can often hide "bad" news from public scrutiny, whereas regulators have authority to access nonpublic information through confidential on-site examinations.² Regulator access to confidential information could easily offset any processing speed advantage held by the market and enable regulators to recognize problems either contemporaneously with or ahead of the market.

Empirical evidence does not consistently either support or reject the hypothesis that information embedded in market prices has useful predictive value to regulators. Pettway (1980) examines a small sample of large-bank failures and finds market signals that preceded regulatory awareness by long lead times, but Simons and Cross (1991) analyze a more recent sample to find that regulators appear to have been aware of problems before the market recognized them. Berger and Davies (1998) find that the market anticipates upgrades in regulatory ratings but follows downgrades. Flannery and Houston (1999) find that the market placed a high value on regulatory certification of bank accounting data in 1988, but a much lower value in 1990. Berger, Davies, and Flannery (2000) find that regulators acquire information sooner than bond rating agencies do, but the regulatory assessments are less accurate in predicting future changes in the performance of bank holding companies than are either stock or bond market indicators.

This study examines the extent to which market data can be used by regulators in evaluating bank financial condition. Market-related information offers the potential for assisting regulators at two points in their assessment, or rating, of bank capital, assets, management, earnings, and liquidity ("CAMEL" ratings).³ First, during the period preceding a CAMEL rating downgrade, market-related information might be used to anticipate the need for the downgrade. Second, during the period after a CAMEL rating downgrade, market information might help distinguish institutions that will subsequently recover (rating upgrades) from institutions that subsequently encounter more serious problems (rating downgrades or failure).

Since our interest is to identify variables that are useful for early-warning purposes, we focus on longer-term trends that are observable in practice and that precede rating changes with sufficient advance warning to provide regulators with a timely tool for policy change. Although market variables may meet these criteria on a univariate basis, their usefulness is enhanced by an ability to add marginal predictive value to other information used by regulators to monitor financial health—information such as Call Report financial data.

Some previous studies have incorporated market data into traditional default or bankruptcy models to assess the riskiness of institutions. For example, Altman (1968), KMV Corporation (1995), and Shumway (2001), have adopted this approach each using a different statistical methodology to analyze the issue. In contrast to these earlier studies, this paper utilizes logistic regression analysis and incorporates significantly more and a greater variety of market variables into the model to identify financially distressed institutions.

Section II of this paper describes the institutional setting in which regulators assess bank financial health, and Section III discusses conceptual issues relating to regulatory use of market data. Section IV describes a sample of publicly traded banks and thrifts that received CAMEL rating downgrades, and this sample is used in Sections V and VI to examine the performance of market-related variables around the time of the CAMEL rating changes. Section VII specifies a logistic regression model to test the marginal predictive content of market-related variables compared with accounting data from bank financial reports in anticipating changes in CAMEL ratings. Section VIII concludes.

Modern bank supervision uses information from on- and off-site supervisory tools as the starting point for its analysis. The largest banks and bank holding companies are monitored by on- and off-site analysts (examiners) who keep abreast of any information that can be found, including news reports, Wall Street analysis, and traditional Call Report financial data. Most smaller and mid-sized banks are monitored initially with automated analysis of Call Report information, and, if risk is identified, they are then reviewed by analysts.⁴

Periodic on-site safety-and-soundness examinations begin with off-site pre-exam reviews of Call Report and other pertinent data. On-site reviews then check the reported information and explore issues that might not be revealed in the quarterly reports.⁵ On-site examinations provide extensive financial information that is not generally available to the public, such as the payment histories of performing and nonperforming loans, loan classifications and the adequacy of loan-loss provisions, and bank capital.⁶

Bank examiners assign overall, or composite, CAMEL ratings of 1 or 2 to institutions in sound financial condition. Downgrading a bank’s rating to 3 sends an important signal of financial weakness and is normally accompanied by an agreement between the bank’s primary regulator and senior bank management specifying the nature of the bank’s weakness and procedures for changing bank policies to rectify the perceived problems. These agreements are classified by regulators as "informal" enforcement actions because they are not administratively or judicially enforceable in a court of law in the event of noncompliance.⁷ Nevertheless, the agreements represent a loud "shot across the bow" signaling significant regulatory concern and the need for change. Informal enforcement actions are kept confidential by regulators out of concern that public exposure of a bank’s problems may result in a decline in the deposit base or otherwise increase the difficulty or cost of recovery. As might be expected, the financial health of banks receiving a CAMEL rating of 3 varies, although regulators make no effort to distinguish composite quality ratings beyond the integers 1–5.⁸ Institutions downgraded to 3 may remain in that state for periods ranging from several months to several years before passing to a higher or lower grade.

Downgrading a bank’s CAMEL rating to 4 or lower indicates the existence of serious problems that could lead to insolvency. In practice, the term "problem" bank is often reserved for institutions with composite ratings of 4 or lower, and regulatory "problem-bank lists" tend to specify institutions with these ratings, although practices vary. Banks downgraded to 4 typically require immediate remedial actions and intensive monitoring by regulatory officials. In some cases, bank supervisory officials may not assign the more serious "formal" enforcement actions to 4-rated banks as long as bank management addresses regulatory concerns. If feasible, regulators prefer to work with informal enforcement actions because they are confidential and less confrontational than the more serious "formal" enforcement actions. However, most banks downgraded to ratings 4 and 5 receive formal enforcement actions, and these actions were made public beginning in 1989.⁹ Institutions with CAMEL ratings of 4 can continue in business for as long as several years before either returning to a higher grade, moving to a lower grade, or being declared insolvent by their primary regulator. A rating of 5 indicates an extremely high probability of failure, usually within the next 12 months.

Prudential bank supervision recognizes that bank examinations impose costs on banks and that higher costs should be justified by higher benefits, such as a comparable reduction in the expected cost of failure. Regulators have a responsibility to balance the need to limit the expected cost of failure against the need to minimize oversight costs, especially for well-run institutions.

With an eye toward balancing the costs against the benefits of bank supervision, we specify four criteria as a starting point for assessing whether information contained in market data is useful to regulators. First, the information should be "separable" at the bank level. This point simply recognizes that a change in a bank’s rating must be justified by the circumstances of that bank. Informational messages that are applicable to groups of banks, such as a change in the risk of all banks in a geographic region, are difficult (costly) to review or apply at the bank level because of their broad application. Second, the message must be "clear" in the sense that there exists a low likelihood of multiple interpretations. The hope is that the message specifies the nature of the market’s concern to the point that a meaningful regulatory response can be formulated. Unclear messages may motivate inappropriate regulatory response, raising the supervisory cost borne by the regulated institutions. Third, the informational message must precede problems with sufficient advance warning to provide regulators with a timely tool for policy change. Finally, the market signals should, on the margin, contain information not available in other sources of information commonly used by off-site monitoring, such as Call Report financial data. Market information that cannot add incremental benefit to the content of commonly used financial data may be viewed as redundant and could fail to justify an expenditure of regulatory resources.

For the purposes of this study, the first two hurdles can be overcome by a focus on elements of longer-term trend in stock prices, returns, and other market-related variables. Since stock prices and returns of firms in the same industry are correlated, short-term informational messages for specific institutions are often reflected in the prices and returns of many other firms in the same industry.¹⁰ This characteristic presents a relatively high cost for distinguishing and interpreting shorter-term signals of specific firms compared with those of other firms in the industry. Over longer periods, however, conflicting signals tend to offset, and it becomes easier to assign informational signals to specific institutions. Focusing on longer-term trends thereby allows regulators to limit the use of regulatory resources by targeting only the most appropriate firms for regulatory review.

The third hurdle requires that market signals have predictive content and sufficient advance notice. The vast resources of the market, which contains tens of thousands of traders, analysts, and similar participants, greatly exceed the resources of all banking industry regulators. Apart from vast resources, the market may also access certain types of nonpublic information that is not available to regulators, such as large stock sales by insiders and private contacts between individuals who may be old friends or former business associates. These points combine to suggest that market awareness of problems might precede regulatory awareness, possibly by a large margin, despite the ability of regulators to access nonpublic information through examinations.

The final hurdle—identifying the incremental value added by market signals—is the most difficult to deal with because it involves distinguishing the predictive content of market variables from the content of other common sources of information. But market variables should not be expected to be independent of other variables or effects. Rather, they need only provide sufficient additional value so that their contribution can be clearly distinguished from the contributions of the other variables. For example, if the market has a unique ability to interpret accounting data contained in quarterly reports, then in regressions that explain changes in financial position we expect market variables to provide significant additional explanatory power to financial ratios derived from the quarterly reports.

An intriguing 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 model anticipates a rise in return volatility as an institution approaches insolvency.¹¹ 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.¹² 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, may contain predictive content.

In summary, it appears conceptually feasible for a number of market-related variables to provide regulators with the ability to speed their identification of risk without imposing burdensome costs on regulated institutions. Debate about regulatory use of market-related information in prudential bank supervision should therefore focus on empirical questions, not conceptual ones. In particular, do market-related variables add marginal predictive value to Call Report or other information easily available to regulators in off-site monitoring systems? In the absence of marginal predictive value, market signals may be viewed as redundant information with little supervisory value.

The empirical analysis begins with a sample of publicly traded banks and thrifts whose ratings were downgraded to problematic levels over the 1988–1996 period. Since a CAMEL rating of 3 signifies significant regulatory concern, but ratings of 4 and 5 signify more-severe financial distress that is often followed by failure, we separate institutions downgraded to 3 from those downgraded to 4 or 5. Recognizing that the financial industry had many problems in the early and middle 1980s, we limited the sample to institutions that were downgraded to 3 beginning in 1988 or were downgraded to either 4 or 5 beginning in 1989.¹³ Combining the 4s and 5s into a single group appeared reasonable, given that institutions may pass to failure from these two ratings but almost never from a rating of 3. As noted earlier, informal enforcement actions, which are not made public, are associated with a rating downgrade to 3, whereas formal enforcement actions, required by FIRREA to be made public, are associated with the assignment of 4 and 5 ratings.

To improve the integrity of the analysis, we also imposed several additional restrictions. Since our focus is on longer-term stock market behavior around the time of rating downgrades, the sample was limited to institutions that had a lengthy period of superior ratings before their downgrade. This condition is implemented by the requirement that institutions have CAMEL ratings in the 1–2 range at least three years before downgrade to a 3. Similarly, institutions downgraded to a 4 or 5 were required to have ratings in the 1–3 range for at least three years preceding downgrade to 4 or 5. The sample is also limited to banks and thrifts that either were not affiliated with bank holding companies or were members of holding companies that held only a single banking-related institution. Restricting the sample in this fashion ensures that the extensive financial data reported on bank Call Reports corresponds closely to the institution that issues the stock.¹⁴ This restriction also reduces contamination from activities of nonbank subsidiaries of bank holding companies.¹⁵ Since the empirical analysis combines Call Report financial data with stock market information reported by the Center for Research in Securities Prices (CRSP), both sources of data were required for a bank to be included in the sample.

Table 1 provides summary statistics for the two groups of downgraded institutions. The sample size is relatively large for both groups, with 122 institutions downgraded to 3 and 148 downgraded to 4 or 5. The sample varies slightly from quarter to quarter for several reasons, including the lack of available data on individual firms and the de-listing rules of the various exchanges (rules such as minimum capital requirements or minimum trading activity). The number of institutions in the 4 or 5 group is larger than the number in the 3 group primarily because the three-year constraint on previous ratings limits the group of 3s more than the group of 4s and 5s.

Considerable diversity is apparent in the sample. For example, both groups of downgraded institutions had a wide range of asset sizes, including institutions with total assets under $100 million as well as institutions with assets over $5 billion. More than 75 percent of the institutions had assets under $1 billion, while almost 20 percent had assets in the $1–5 billion range and about 5 percent had assets in the over $5 billion range. The relatively healthier condition of institutions downgraded to 3 is reflected in the higher book equity-to-assets and return-on-assets ratios compared with the ratios reported for institutions downgraded to 4 or 5. The stronger financial health appears to be recognized by the market, as the market price summary statistics are also higher for institutions downgraded to 3 compared with those downgraded to 4 or 5. The diversity in asset size is accompanied by diversity in market capitalization (average price times number of shares at the end of the quarter), which declines with market price. The sample breakout by charter type shows that banks (65 percent) have a higher representation than thrifts (35 percent).

The first point at which market information has the potential to help bank regulators is the period preceding a CAMEL rating downgrade. As explained above, downgrades to ratings of 3 are distinguished from downgrades to ratings of 4 or 5 by the distinct financial and regulatory problems of banks in the latter category. Our focus on longer-term trends is implemented by an analysis of quarterly data for several years preceding the downgrade.

Table 2 displays the univariate characteristics of stock prices, returns, and other market-related variables for banks and thrifts eight quarters (two years) preceding their downgrades to CAMEL rating 3, 4, or 5. The zero quarter contains the start date of the examination that results in the rating downgrade.¹⁶ As noted above, the sample of institutions in the 3- and 4- or 5-rated categories varies slightly from quarter to quarter because of the de-listing rules of the various exchanges (rules such as minimum capital requirements or minimum trading activity).

The stock price data show prices falling consistently throughout the two years before the downgrades, with the result that the change in stock price is negative in nearly every quarter during that period. For the 3-rated group, the average stock price starts at around $15 per share in the eighth quarter before the downgrade, then falls to $10 per share in the zero quarter, whereas for the 4- or 5-rated group, the average price declines from about $11 to $6 per share for the same period. The lower prices for the 4- or 5-rated institutions suggests that the market is able to distinguish the more serious financial problems of institutions approaching a 4 or 5 rating compared with those approaching a rating of 3.

In an effort to test the consistency of changes in stock prices across the sample, we used a t-test to test the hypothesis that the mean of each quarterly sample equals zero. For the 3-rated group, this test shows that the change in stock price becomes statistically significant in the sixth quarter preceding the downgrade. For the 4- or 5-rated group, the change is significant for all eight quarters before the downgrade, reflecting the more distressed nature of this group. The consistency of the t-test results, across many quarters before the rating downgrades, confirms that market prices have considerable univariate predictive content long before regulators formally alter an institution’s rating.

Two measures of trading activity are included in Table 2 to allow an examination of the hypothesis that higher trading activity should accompany market assessment of new information. But although the data clearly reflect a long-term trend of declining prices before downgrade, neither of the trading activity variables reflects the corresponding rise in trading activity hypothesized by financial theory. The most direct measure of trading activity—average daily trading volume—declines slightly for the 3-rated group and follows no consistent trend for the 4- or 5-rated group throughout the eight quarters before the downgrades. A second measure of trading activity, known as "turnover," divides the shares traded in any quarter by total shares outstanding. Like the trading volume variable, the turnover variable also shows no discernible trend for either of the groups. Therefore, the trading activity variables contain no easily observed univariate predictive content before CAMEL rating downgrades.

The remainder of Table 2 displays various measures of quarterly returns commonly found in previous empirical studies. The first return-related variable is the simple cumulative return, calculated by multiplying unity plus the daily return for each stock i on day t (1+r_it) across all trading days in each quarter, then subtracting unity. Consistent with the long-term decline in prices, cumulative returns are consistently negative preceding downgrade for both the 3-rated group and the 4- or 5-rated group. However, our t-tests also find that the cumulative returns are not significantly distinguishable from zero for institutions downgraded to 3, although they are significant for institutions downgraded to 4 or 5.

The cumulative return data are followed by a measure of return volatility, the standard deviation of daily returns. In contrast to the trading activity variables, the trend of volatility appears consistent with financial theory, which anticipates rising volatility as an institution encounters distress. For example, the volatility of the variable rises steadily for both groups as the downgrades approach, especially during the four quarters immediately preceding the downgrades. The level of volatility is noticeably higher for the most severely distressed institutions (downgraded to 4 or 5) compared with the moderately distressed institutions (downgraded to 3). The statistical content of the rising trend is confirmed by significant t-statistics beginning about a year before the rating downgrades.

The remaining columns examine market excess returns by calculating the differences between the cumulative quarterly return of each stock and the cumulative quarterly returns for three indexes of market performance. The first two indexes are the equal- and value-weighted indexes reported on the CRSP tapes. The third index is a value-weighted index constructed from CRSP data for the bank and thrift industries.¹⁷ Means tests are applied to all quarterly samples of excess returns to determine whether the mean excess return is statistically distinguishable from zero.

The market excess return variables consistently show patterns of negative returns that are statistically significant in most of the eight quarters before rating downgrades, although the significance varies by index. The results for the CRSP equal-weighted return index reflect the most consistent trend, as they are virtually always negative and significant at the 1 percent level. Excess returns calculated from the CRSP value-weighted index perform very similar to excess returns calculated from the CRSP equal-weighted index, with only slightly lower levels of significance for a few. The excess returns calculated from the industry value-weighted index did not precede or anticipate rating downgrades as consistently as the other excess return variables did.

From a broad perspective, the excess return data reported in Table 2 serves to reaffirm Pettway’s (1980) finding of negative excess returns for lengthy periods preceding financial distress, although our analysis is based on much more extensive data and statistical tests of the significance of each quarterly excess return.¹⁸ From the narrower interests of this paper, the excess return variables exhibit consistent negative patterns, effectively anticipating CAMEL rating downgrades long before they actually occur.

Table 3 performs a sensitivity analysis of the results in Table 2 by examining monthly trends of the most consistent quarterly variables. The trends are examined during the 17 months preceding the rating downgrade, in other words, for a period of almost five quarters. The monthly trends are not as consistent as the quarterly trends, although some data have reliable predictive content. The change in the stock price variable is always negative, but the change is not statistically significant for a number of months for the 3-rated group, although it is almost always significant for the more distressed firms in the 4- and 5-rated group. The performance of the volatility variable, the change in the standard deviation of daily returns does not have the anticipated negative sign in all quarters and is almost never significant at the 1 percent or 5 percent level. The CRSP equal-weighted excess return performs the best of all variables examined, showing negative returns that are significant at the 1 percent or 5 percent level in most months for both groups of firms. Consistent with our earlier findings, the remaining estimates of market excess returns, calculated with either the CRSP value-weighted index or the industry value-weighted index, are always negative, but not as consistently significant as excess returns based on the equal-weighted index. These results suggest that the equal-weighted market excess return retains, to a large degree, its predictive content over monthly as well as quarterly periods.

Table 4 extends the sensitivity analysis by examining the best-performing series from Table 3 over weekly periods. We limit the reported data to weeks 36 through 10 (months 9 through 2) preceding CAMEL rating downgrades. The weekly results are noticeably less conclusive than the monthly data. Somewhat surprisingly, the best-performing variable is the stock price, which is almost always negative and statistically significant at the 1 percent or 5 percent level in more than one-half of the reported observations, especially for the most distressed 4- and 5-rated group. The two measures of market excess returns are not consistently negative, and they are only occasionally significant at the 1 percent or 5 percent level. Thus, the predictive content of individual market variables is very limited in weekly applications. This result cautions against the use of short-term return trends for identifying longer-term market movements preceding rating downgrades.

In addition to anticipating a rating downgrade to the 3 or 4/5 level, regulators also have a special interest in the period immediately after the downgrade. In this regard, the interests of this paper prompt us to ask, "Can the market distinguish future ‘winners’ (subsequent upgrades) from ‘losers’ (subsequent downgrades) in the immediate aftermath of the rating downgrades?" Distressed institutions require close regulatory supervision. Most either recover (are subsequently upgraded) or encounter more serious problems (either are subsequently downgraded or fail) in reasonable periods of time after the rating changes, although some institutions languish in these rating categories for several years after their initial rating change. If market variables can anticipate the path taken, this information might be used to allocate supervisory resources, or otherwise improve regulatory oversight of problematic institutions. Although the ability of the market to anticipate the subsequent performance of a downgraded institution is interesting by itself, the supervisory value of the anticipation is enhanced if it is observed relatively soon after the downgrade—such as during the first quarter after the downgrade.

Tables 5 and 6 present univariate results similar to the results in Table 2. The previous groups of 3- and 4/5-rated firms are broken into two subsamples, depending on whether the downgraded institutions subsequently recover (Table 5) or whether they slide into deeper financial distress (Table 6). Since a subsequent rating change can occur at any time after the initial downgrade, Tables 5 and 6 show declining samples during the eight quarter period after the downgrades.¹⁹ For example, the sample of 3-rated banks that recover declines from 54 observations during the zero quarter to 48 in the eighth quarter after the downgrades, while the 4/5-rated group goes from 94 to 87. A similar pattern is observed for firms that did not recover after their downgrade (Table 6).

Table 5 suggests that immediately after the initial rating change, the market does not anticipate the recovery of institutions that have experienced CAMEL rating downgrades. The exact nature of the reaction is different, however, depending on the financial condition of the group. For example, for the 3-rated group after the downgrade, average stock prices follow a downward trend for only the first quarter, going from about $11 per share in quarter zero to $10 per share in quarter one before rebounding slightly in quarter two. In other words, for this group, the market soon determines that recovery is under way. For the 4/5-rated group, stock prices follow a downward trend for about two quarters on average after the downgrade, with prices declining from almost $7 per share in the zero quarter to about $6 per share in quarter two. Thus, the upward trend in prices associated with market anticipation of recovery for the 4/5-rated group does not occur for about six months. The downward trend in prices after the downgrades causes all estimates of market excess returns to be negative after the initial downgrade, so these variables retain no more predictive content than the stock prices. In addition, contrary to the results in Table 2, the volatility variables fail to develop the downward trend anticipated by financial theory. Consistent with the results in Table 2, no clear trend appears in the trading activity variables (trading volume and turnover). Thus, market-related variables provide little univariate predictive content for institutions that recover from a rating downgrade.

In contrast to the results in Table 5, Table 6 finds the market capable of anticipating the performance of institutions whose distress deepens after a CAMEL rating downgrade to the 3 or 4/5 level. In Table 6, stock prices for the 3-rated group follow a steady downward trend after the downgrade. A similar pattern is observed for the 4/5-rated group, with prices declining continuously for all eight quarters after the downgrade. For both groups, the downward average price trend during the quarters after downgrade causes negative cumulative and market excess returns, and the t-statistics associated with the negative returns tend to be statistically significant. The volatility variables also recover much of their earlier explanatory power, as volatility rises steadily in the period after downgrade. Consistent with our earlier results, no clear trend appears in the trading activity variables.

Comparative interpretation of results in Tables 5 and 6 is difficult to do. Table 6 in isolation suggests that market-related variables have univariate predictive content. However, similarities between the downward trends of prices and returns in Tables 5 and 6 make it more difficult to distinguish winners from losers after rating downgrades. For example, the market appears to identify future "losers" in the 4/5-rated institutions in Table 6 with declining average stock prices and returns in the first quarter after the downgrade. However, the identification of "winners" in Table 5 takes up to four quarters for the 3-rated group and up to six quarters for the 4/5-rated group. Thus, univariate market signals regarding an institution’s prospects for recovery, subsequent to a rating downgrade, are difficult to interpret.

Testing the marginal importance of stock price and return variables against Call Report financial data allows us to formally gauge the statistical strength of the two types of explanatory variables. Our approach proceeds by initially specifying a traditional CAMEL rating prediction model and then extending the model to include stock prices, returns, and other market related variables. Although the stock price and return variables need not dominate the traditional ratio-based model, a minimum level of competency is required to justify a conclusion that market-related variables are a meaningful addition to traditional analysis.

In this section, logistic regression equations are estimated to explain changes in financial institution supervisory (CAMEL) ratings with publicly available financial information.²⁰ Table 7 defines the variables used in the regressions, along with related means and standard definitions. Two sets of regressions are estimated, one for predicting rating downgrades and the other for predicting either recovery or ultimate failure subsequent to a rating downgrade. A dichotomous variable is specified for both equations. The dependent binary variable in the first equation (CAMELCAT) tests our ability to explain CAMEL rating downgrades, whether to ratings of 3, 4, or 5. To this end, each downgraded institution is "matched" with a randomly chosen, publicly-traded bank or thrift that is healthy and is in the same asset class as the downgraded institution. If Call Report or CRSP return data are not available for an institution in any quarter, that institution is dropped from the sample during the quarter. The dependent variable in the first set of regressions takes a value of 1 if the institution is rated 1 or 2 and is downgraded to 3, and 0 if the institution is rated 1 or 2 and does not experience a downgrade. In the second set of regressions, the dichotomous dependent variable tests our ability to explain whether an institution either recovers or suffers further distress after a rating downgrade. In this case, the dependent variable takes a value of 0 if the institution recovers, and a value of 1 if its distress deepens after a downgrade.

The logit model is estimated by the use of data for market variables from quarter four preceding the downgrade, while Call Report data are taken from quarter five. The reason the market variables from a given quarter are matched with Call Report financial data from the preceding quarter is that the Call Reports are routinely released in the quarter following the quarter to which they apply.

The first independent variable is a control variable for charter type. This variable is important because in 1989 FIRREA provided funds to deal with a backlog of troubled thrifts whose financial condition had slipped below the condition of troubled banks during that era.

A dummy variable, INSBIF, distinguishes banks and thrifts by assigning a value of unity to institutions that come under the umbrella of the Bank Insurance Fund (BIF), and a value of zero to institutions that come under the Savings Association Insurance Fund (SAIF). As shown in Table 7, the mean of the INSBIF variable equals 0.69, signifying that commercial banks represent 69 percent of the 3-rated group, and therefore thrifts 31 percent of the same group. For the 4/5-rated group, banks represent 60 percent and thrifts 40 percent. For the overall sample, banks represent 64 percent and thrifts 36 percent. The coefficient for the "charter" dummy is expected to be positive in CAMEL rating models using financial ratios because, in the post-FIRREA period, banks tended to have a higher likelihood of downgrades than thrifts with similar asset sizes: many of the most-troubled thrifts had already been sold or liquidated by the Federal Savings and Loan Insurance Corporation or handed off to the Resolution Trust Corporation for resolution.

The traditional CAMEL prediction model is specified with variables from the Call Reports. The first variable is the equity-to-assets ratio (EQ_AS), which measures the ability of a firm to absorb loan losses before bankruptcy and is expected to be negatively related to future distress. The credit quality of the loan portfolio is captured in the NC_RES variable, which measures the amount of delinquent assets less loan-loss reserves relative to total assets.

A positive sign is expected for this variable, signifying that higher levels of delinquent loans are associated with a higher likelihood of rating downgrades and failure. The overall profitability of the institutions is measured by the return-on-assets variable (ROA), which is expected to be inversely related to future downgrades. Two measures of liquidity are posited: the securities-to-assets ratio (SC_AS) and the volatile liabilities-to-assets ratio (VL_AS). The SC_AS is expected to be negatively related to future distress, reflecting that higher levels of securities to assets provide sources of additional liquidity in troubled times. However, the volatile liabilities ratio VL_AS is expected to have a positive sign, reflecting that higher levels of volatile liabilities are normally associated with potential for problems of liquidity and funding during times of crisis.

The market variables are segmented into two groups, one specifying "core" financial variables that reflect market activity, the other accounting for market risk. In the first group, the first variable is market price, which is measured as the natural logarithm of the average quarterly price. This variable is expected to be inversely related to rating downgrades. Market excess returns is captured by EXRET, which measures the CRSP equal-weighted excess quarterly returns for each observation and is discussed extensively above in the univariate analysis connected with Table 2. Given the high degree of negative excess return persistence observed in Table 2, we expect EXRET to possess at least some failure predictive content and to be negatively related to the future downgrades. Firm dividend policy is captured by the dummy variable DIV, which is equal to unity if the bank paid a dividend in the last four quarters, or zero otherwise. The coefficient for DIV is expected to have a negative sign, reflecting financial weakness if the firm fails to pay a dividend.

In the second group of variables, those that account for market risk as suggested either by the market model of Fama and French (1993) or the option model of Merton (1974), the first variable, SDRET, is the standard deviation of annualized quarterly returns and is expected to be positively related to future downgrades. As investors become concerned over potential distress and potential bankruptcy, more variation may appear in return patterns. Another stock market model variable, TURN, which measures stock turnover in particular quarters, is also expected to be positive during periods of financial distress. The third market model variable, the book equity-to-market value ratio (BE/ME), is expected to have a positive coefficient as the ratio moves directly with changes in stock prices, holding book equity constant. This ratio, therefore, also serves as proxy for financial distress.

Equation (1) shows the basic logit estimation equation, which sequentially adds stock market data to Call Report financial data on the right-hand side of the regression:

(1)

The results for the pre-downgrade period are presented in Table 8. As noted, the regressions were run at four quarters before the event date (or, in this case, before the date the institutions experienced their downgrades to the 3, 4, or 5 levels). Panel A shows the results for firms that were downgraded to 3, and Panel B shows the results for firms that were downgraded to 4 or 5. Specification 1 shows the CAMEL prediction model primarily using Call Report data. The control variable for charter type, INSBIF, has a positive coefficient and is significant only for those banks that are in the worst financial condition—the 4/5-rated group. This finding reveals that commercial banks generally tended to be more distressed than thrifts during the early 1990s and therefore exhibit a higher likelihood of being downgraded.

The first Call Report variable, the equity-to-asset ratio (EQ_AS), has a negative sign as expected for both groups, thereby confirming the importance of equity levels in models predicting distressed CAMEL ratings. The portfolio quality variable (NC_RES) has its anticipated positive coefficient and is significant for only the 3-rated group, confirming the relationship between downgraded supervisory ratings and credit quality.²¹ The return-on-asset variable (ROA) also exhibits a negative sign, as expected, and is highly significant. The two liquidity measures (SC_AS and VL_AS) also perform as expected: the former is negative and significant, confirming the fact that there is an inverse relationship between the level of security holdings and financial distress, and the latter shows a positive and significant relationship between volatile liabilities and future downgrades. For specification 1 for each group, all signs are correct and most coefficients are significant at the 1 percent level. These results suggest that this equation serves as a good starting point or benchmark for assessing the incremental value of information embedded in stock returns compared with the information contained in commonly used financial ratios.

The analysis proceeds by selectively adding market variables that offer several measures of returns or stock-return trends to the benchmark regressions containing only financial ratios. This approach facilitates the ability to examine the predictive content of each of the market variables, through their individual coefficients and t-statistics. It also tests the predictive content of market-based models compared with the content of traditional models, through the likelihood ratio test statistic and the Akaike Information Criterion (AIC).²² If the likelihood ratio test statistic is positive and significant, we may conclude that a market-based regression has significantly higher predictive content than the Call Report model reported in the first specification of the model. A similar conclusion can be reached if the AIC measure exhibits a lower value for the market-based specifications relative to the Call Report model.

The logistic regressions incorporating market variables are presented in specifications 2 and 3. Of the three core market variables added into specification 2—the natural logarithm of the stock price (LN_PR), the equal-weighted market excess return variable (EXRET), and the dividend variable (DIV)—all have the expected negative signs for the coefficients. The log of price variable is significant primarily for the 3-rated group. For the 4/5-rated group, both the dividend and excess return variables are significant, portraying a mixed picture for the core market variables.²³

Specification 3 adds two additional market-risk variables: the standard deviation of the return variable (SDRET), which measures the variance of the returns, and the turnover ratio (TURN), which measures average trading volume. In this regression, these other market-related variables failed to show any significant increase in marginal predictive value for either of the groups. The SDRET variable was significant at the 5 percent level for the 3-rated group, and the TURN variable was significant at the 10 percent level for the 4/5-rated group. In the last specification (4), only the publicly available stock-market variables are included. Based upon the AIC, this equation does not improve upon specification 1, which incorporated only Call Report variables.

A likelihood ratio test is performed, measuring the equation specifications that contain stock-market variables (specifications 2 and 3) compared with the equation specification that does not contain stock-market information (specification 1). As shown in Table 8, equations 2 and 3 show greater explanatory power compared with specification 1, suggesting value added by the introduction of the market variables. Specification 4, with only the stock-market data, has significantly less explanatory power than specifications 1, 2, and 3. These results are shown by the AIC. The AIC value is lower for both regressions 2 and 3 relative to equation 1, suggesting that the stock-market variables are adding to the overall predictive content of the model. The AIC for the last equation is much higher than it is for the previous three regressions, suggesting that the model using only market-related variables performs at a lower level than the other three specifications.²⁴

Table 9 contains in-sample and out-of-sample tests of the model for both the 3-rated and 4/5-rated groups.²⁵ The critical probability that is used is 50 percent. This critical probability is used to determine how the model performs in identifying which banks or thrifts in the two groupings are properly classified as likely to experience future CAMEL rating downgrades. Within the in-sample classification group for the 3-rated institutions, the correct prediction of distressed or healthy banks and thrifts generally increases, moving from specifications 1 to 3; then the correct prediction level declines in specification 4. The group prediction columns generally reveal a similar pattern for the 4/5-rated group. Thus, for the in-sample forecasts, these findings show that adding stock variables increases the predictive accuracy of the model in identifying CAMEL rating downgrades in the first three specifications. This is reflected in the Type I and Type II errors, which are also displayed in Table 9.

A similar pattern is observed for the out-of-sample forecasts, although the classifications are not as accurate as for the in-sample forecasts. For example, for the specifications for the 3-rated group, the model correctly classified an average of 66 percent of the downgrades for the out-of-sample, compared with 76 percent for the in-sample. For the 4/5-rated group the classifications were relatively more accurate, identifying about 72 percent of the downgrades for the out-of-sample data relative to about 74 percent for the in-sample data. The overall accuracy of the model appears to improve as the conditions of the institutions deteriorate.

Table 10 shows the logistic regression results for the first quarter after the CAMEL downgrades for the 3- and 4/5-rated groups. As mentioned previously, the second model is run to determine if it is sensitive enough to predict which institutions will likely recover or experience future downgrades or failures. Banking regulators would benefit if the model were able to distinguish these differences within a relatively short period after the initial downgrades. The results show that all the coefficients for the Call Report variables have the anticipated signs and perform as expected, but the significance of the tests varies between the 3- and 4/5-rated groups. The log of market price and the excess return variables generally add to the CAMEL predictions in most specifications, although there is some variation in levels of significance, which range from 1 percent to 10 percent depending on the specification. As before, when the market-risk variables are combined with the Call Report data and core market variables, they generally do not add significantly to the equation. Therefore, the logistic regression tests tend to suggest that relatively simple measures of market price and market excess returns appear to offer the best hope of improving the predictive content of Call Report data. Although the market excess return variable is not a dominating variable for the 3-rated group, for the most distressed organizations—the 4/5-rated group—it adds significantly to standard models based on the accounting data in quarterly reports. Other market variables, such as return volatility, dividend payment performance, turnover in shares, and the ratio of book to market value equity, appear to have little marginal predictive value.

Table 11 contains information on the accuracy of the model in the post-downgrade period. The critical probabilities for these in-sample and out-of-sample classifications are 0.56 for the 3-rated group and 0.35 for the 4/5-rated group.²⁶

In general, these findings show that the addition of stock-market information to the Call Report data not only increases the correct predictions of banks that will either have more financial distress or recover but also minimize the incorrect predictions—although the results are not as strong as in the pre-downgrade period. As we move from specification 1 with only Call Report data to specifications 2 and 3, the model generally improves the correct prediction of financial distress especially for the most-distressed—banks in the 4/5-rated group. For specification 4 in relation to specification 1, the stock-market variables have only a mixed correct prediction rate when compared with Call Report variables alone.

This paper explores the notion that stock price, return, and other market-related variables can be used to improve the predictive content of Call Report financial ratios for the purpose of anticipating CAMEL rating changes. A sample of 122 banks and thrifts that were downgraded to the CAMEL 3 level and 148 banks and thrifts downgraded to the 4 or 5 level was analyzed over the period 1988–1996. Extensive univariate analysis confirms that relatively simple measures of stock prices and returns exhibit downward trends as much as two years before banks and thrifts experience CAMEL rating downgrades to 3, 4, or 5. The longer-term nature of these trends suggests that the univariate trends are not commonly found in stock returns of healthy institutions. Previous research has not confirmed the pre-downgrade returns of patterns found in this paper—certainly not for the length of time or for the variety of market variables used in this research. However, no simple relation appears in univariate comparisons of several other market variables, including average trading volume and average quarterly turnover of shares.

The second section of the paper adds stock return variables to regression equations that include financial ratios commonly used to predict CAMEL rating changes (both upgrades and downgrades) in off-site monitoring models. The results provide several additional points of interest. Most important, adding relatively simple measures of excess returns, stock prices, and an institution’s dividend record improve the CAMEL ratings predictive content of Call Report data and otherwise appears to have a limited independent role in anticipating financial distress. The predictive content of the models is most robust for institutions experiencing the greatest financial distress—those being downgraded to the 4 or 5 level. Other market-related variables, such as return volatility, trading volume, and the book-to-market equity ratio, appear to have limited predictive value.

References

Altman, E. I. 1968. Financial Ratios, Discriminant Analysis and the Prediction of Corporate Bankruptcy. Journal of Finance, no. 23:589–609.

Barberis, Nicholas, Andrei Shleifer, and Robert Vishny. 1998. A Model of Investor Sentiment. Journal of Financial Economics 49, no. 3:307-343.

Berger, Allen N., and Sally M. Davies. 1998. The Information of Bank Examinations. Journal of Financial Services Research 14, no. 2:117–144.

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. (Forthcoming.)

Chen, Joseph, Harrison Hong, and Jeremy C. Stein. 2000. Forecasting Crashes: Trading Volume, Past Returns and Conditional Skewness in Stock Prices. Journal of Finance. (Forthcoming.)

Cole, Rebel A., and Jeffrey W. Gunther. 1995. Separating the Likelihood and Timing of Bank Failure. Journal of Banking and Finance 19, no. 6:1073–89.

———. 1998. Predicting Bank Failures: A Comparison of On- and Off-site Monitoring Systems. Journal of Financial Services 13, no. 2:103–17.

Curry, Timothy, George Hanc, John O’Keefe, Lee Davison, and Jack Reidhill. 1997. Bank Examination and Enforcement. In History of the Eighties, Lessons for the Future: An Examination of the Banking Crises of the 1980s and Early 1990s, 421–75. Federal Deposit Insurance Corporation.

Curry, Timothy, John O’Keefe, Jane Coburn, and Lynne Montgomery. 1999. Financially Distressed Banks: How Effective Are Enforcement Actions in the Supervision Process? FDIC Banking Review 12, no. 2:1–18.

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 David M. Borowski. 1988. An Expert System Approach to Financial Analysis. Financial Management 17, no. 3:66–76.

Fama, Eugene F. 1998. Market Efficiency, Long-Term Returns, and Behavioral Finance. Journal of Financial Economics 49, no. 3:283–306.

Federal Deposit Insurance Corporation. 1997. History of the Eighties, Lessons for the Future: An Examination of the Banking Crises of the 1980s and Early 1990s, Vol. 1. (Available on the Internet at www.fdic.gov.)

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

Fissel, Gary S. 1994. Risk Measurement, Actuarially-Fair Deposit Insurance Premiums and the FDIC’s Risk-Related Premium System. FDIC Banking Review 7, no. 1:16–27.

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.

Flannery, Mark J., and Joel F. Houston. 1999. The Value of Government Monitor for U.S. Banking Firms. Journal of Money, Credit, and Banking 31, no. 1:14–34.

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.

Greene, William H. 2000. Econometric Analysis. Prentice Hall.

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

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.

KMV Corporation. 1995. Introducing Credit Monitor, Version 4. San Francisco: KMV Corporation.

Maddala, G.S. 1983. Limited Dependent and Qualitative Variables in Econometrics. University Press.

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

Nelson, Daniel B. 1991. Conditional Heteroskedasticity in Returns: A New Approach. Econometrica 59, no. 2:347–70.

Pettway, Richard H. 1976. Market Tests of Capital Adequacy of Large Commercial Banks. Journal of Finance (June):865–75.

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

Pindyck, Robert S. and Daniel L. Rubinfeld. 1991. Econometric Models and Economic Forecasts. McGraw–Hill.

Shumway, Tyler. 2001. Forecasting Bankruptcy More Accurately: A Simple Hazard Model. Journal of Business vol. 74, no. 1:101–124.

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

Wang, Jiang. 1994. A Model of Competitive Stock Trading Volume. Journal of Political Economy 102, no. 1:127–168.