Are Banks Opaque?

K. Stephen Haggard* Department of Economics, Finance and International Business University of Southern Mississippi 118 College Drive #5072 Hattiesburg, MS 39406 E-mail: stephen.haggard@usm.edu John S. Howe Department of Finance University of Missouri - Columbia 447 Cornell Hall Columbia, MO 65211 E-mail: HoweJ@missouri.edu

Abstract We use the model of Jin and Myers (2006) to examine the relative opacity of banks. Our results show that banks have less firm-specific information in their equity returns than industrial matching firms, consistent with banks being more opaque than industrial firms. Additionally, we show that banks with less firm-specific information in their equity returns are more likely to experience significant declines in stock price. We also provide new evidence on the opacity of specific bank assets. We show that lower proportions of agricultural and consumer loans are related to higher levels of bank opacity. Our results provide a new tool for regulators to aid in the identification of publicly traded banks that deserve greater regulatory scrutiny. Our results are robust to the inclusion of various controls and consideration of the stock exchange on which shares trade. First version: April 19, 2005 This version: January 11, 2007
Contact author. This paper is based on a chapter of Haggard’s dissertation at the University of Missouri. We wish to thank Brian Balyeat, Rakesh Bharati, Paul Brockman, David Carter, Kiyoung Chang, Lane David, Riza Demirer, Steve Ferris, Mike Finke, Dan French, Rik Hafer, John Howe, Tom Lindley, Sandra Mortal, Raynolde Pereira, Andy Puckett, Sean Salter, Maria Schutte, Sterling Yan, and seminar participants at the University of Missouri – Columbia, Oklahoma State University, Southern Illinois University – Edwardsville, Indiana University – South Bend, and the University of Southern Mississippi for their helpful comments and suggestions. We thank I/B/E/S for providing analyst data. All remaining errors are our own.
*

Are Banks Opaque?
Abstract We use the model of Jin and Myers (2006) to examine the relative opacity of banks. Our results show that banks have less firm-specific information in their equity returns than industrial matching firms, consistent with banks being more opaque than industrial firms. Additionally, we show that banks with less firm-specific information in their equity returns are more likely to experience significant declines in stock price. We also provide new evidence on the opacity of specific bank assets. We show that lower proportions of agricultural and consumer loans are related to higher levels of bank opacity. Our results provide a new tool for regulators to aid in the identification of publicly traded banks that deserve greater regulatory scrutiny. Our results are robust to the inclusion of various controls and consideration of the stock exchange on which shares trade.

1.

Introduction The rationale for the regulation and protection of banks rests on the assumption

that banks are opaque, that outsiders cannot observe the risks involved in financial intermediation. Such opacity exposes banks and the entire financial system to runs and contagion, in which even healthy banks fall victim because opacity prevents outsiders from being able to distinguish between sound institutions and unsound ones. Thus, the logic goes, government regulation, the discount window as lender of last resort, and deposit insurance are necessary to protect healthy banks and the banking system. (Morgan (2002), Flannery et al. (2004)) But are banks really more opaque than industrial firms? To address this question, Morgan (2002) examines the ratings of new bonds issued by banks and industrial firms. If a firm is completely transparent, then the two major rating agencies should reach the same conclusion regarding the risk of any given bond issued by the firm. However, if a firm is opaque, rating agencies must use partial information and speculation to arrive at a rating, creating the possibility of disagreement between the two agencies. Therefore, disagreement between the agencies (a “split” bond rating) is an indication of firm opacity. Morgan shows that banks are more likely to receive such “split” ratings than industrial firms, consistent with banks being more opaque than industrial firms. Flannery et al. (2004) also empirically examine the differences in opacity between banks and industrial firms. The authors examine analyst and microstructure data to arrive at the conclusion that banks are no more opaque than industrial firms. In fact, they show that analysts forecast bank earnings more accurately than industrial firm earnings. Thus, these two empirical studies appear to reach opposite conclusions regarding bank opacity.

1

Given the mixed evidence, the issue of bank opacity remains an open question. In this paper, we revisit this issue using the theoretical model of Jin and Myers (2006), which links a firm’s opacity to its stock price movements. We also re-interpret the analyst forecast findings of Flannery et al. (2004) given the theoretical model of Jin and Myers (2006). Jin and Myers (2006) define firm opacity (opaqueness) as reduced firm information available to outside investors. Two implications flow from their theory. First, the authors argue that opacity reduces firm-specific information available to outside investors and affects the division of risk bearing between firm insiders and outside equity holders. Outside investors, in the presence of limited firm-specific information, replace unknown firm-specific information with its expected value, conditioned on the information available to outsiders. Thus, Jin and Myers (2006) contend that the stock returns of opaque firms are less likely to reflect firm-specific information and more likely to reflect market (and perhaps industry) information. Jin and Myers (2006) also theorize that more opaque firms are more likely to suffer “crashes,” defined as sudden and large negative equity returns. In their model, insiders abandon the firm if forced to absorb too much negative firm-specific information, leading to a sudden release of all such information. Such a release will result in a large negative equity return. Based on this prediction of the model, one might speculate that banks are more likely to experience crashes than industrial firms, and that relatively opaque banks are more likely to experience crashes than relatively transparent banks.

2

Previous studies examine the opacity of different types of assets. We use coefficients of determination from asset pricing model regressions (a measure of stock “synchronicity” or co-movement) to provide evidence consistent with banks being more opaque than matching industrial firms. We also re-interpret the analyst findings of Flannery et al. and premises and fixed assets. industry. We show that banks with less firm-specific information in their returns are more likely to experience crashes than banks with more firm-specific information in their returns. If managers of opaque firms only release information that varies with market and industry conditions. In contrast to the model’s predictions. Given the new evidence of our Morgan and Stiroh (2001) examine the relation between proportions of different loan types and yield spreads on new bond issues. assets held in trading accounts. bank returns contain less firm-specific information than matching firm returns. but not for industrial firms. Therefore. which are composed primarily of bank loans. 1 3 . and Diamond (1991) all lead to the conclusion that bank loans are opaque. we perform an analysis to determine which types of loans impact bank opacity. Studies by Campbell and Kracaw (1980).1 In this study. Ceteris paribus. Berlin and Loeys (1988). possibly due to the additional disciplining forces of regulators and depositors that are present for banks. we find that banks are less likely to experience crashes than industrial firms. which vary with market. then their earnings will be easier to predict than the earnings of transparent firms. are also opaque. but no previous study examines the opacity of individual loan types. This analysis captures differences in yield spreads due to differences in the riskiness of the underlying assets. such as loans. (2004) in light of the Jin and Myers (2006) model. and firm-specific information.Theory suggests that the opacity of bank assets is the source of bank opacity. not necessarily yield spread differences due to differences in opacity. consistent with the prediction of Jin and Myers (2006). bank assets.

The opacity of bank assets As Flannery et al. are related to bank opacity. Section 5 describes the data and methods used. therefore. Using stock returns and FR Y-9C financial statement information for our sample of publicly traded bank holding companies. Berlin and Loeys (1988) discuss the choice between public bonds and bank loans. Such banks are more likely to fail. conventional wisdom says that bank loans are informationally opaque. Section 7 discusses implications of our findings and concludes. Campbell and Kracaw (1980) posit that one reason borrowers use bank loans is that they have confidential information that they do not wish to disclose to the public. We show that lower proportions of agricultural and consumer loans are related to higher levels of bank opacity. 2. (2004). Section 3 provides a review of the stock synchronicity literature and summarizes the Jin and Myers (2006) model. and Section 6 examines our results. and non-insured depositors. Section 4 presents hypotheses and empirical predictions. opaque banks provide greater opportunities for rent extraction and excessive risk-taking by insiders. we conclude that banks are more opaque than industrial firms. deserve a greater amount of regulatory scrutiny. and specifically which loan types.study and the re-interpreted analyst findings of Flannery et al. shareholders. Both 4 . Ceteris paribus. we determine which assets. The rest of the paper is organized as follows: Section 2 summarizes the current knowledge on bank opacity. This conventional wisdom is consistent with several theoretical works. potentially causing losses to deposit insurance funds. Our results provide a new tool for identifying publicly traded banks that are more opaque and. (2004) point out.

new firms) will turn to banks for their borrowing needs. In all of these models. even during financial crises.forms of debt are characterized by information asymmetry between borrowers and lenders. Bank loans are granted based on public information plus information gathered by banks through costly monitoring. 2002.” Calomiris and Mason (1997) examine the 1932 Chicago banking panic and show Consolidated Financial Statements for Bank Holding Companies – FR Y-9C. making bank loans informationally opaque by definition. However. firms without such a reputation (e.028 bank holding companies in the determination of the average percentage of assets represented by loans. Commercial paper is a contract with terms and loan-granting decisions based only on public information. Musumeci and Sinkey (1990) examine the 1987 Brazilian debt moratorium and show that “the market reacted rationally and penalized banks in direct proportion to their exposure to Brazilian debt. bank assets are opaque due to the opacity of loans. Borrowers with excellent credit ratings will choose to issue commercial paper because they can borrow less expensively in the open market due to their reputation for repayment. However. If banks assets are truly opaque.44 percent of its assets2. several studies show that investors and depositors can identify troubled banks. The empirical evidence on the opacity of bank assets is mixed. although banks partially overcome this asymmetry through costly monitoring. Loans form a sizeable portion of the assets of banks. Diamond (1991) posits a difference in information asymmetry between commercial paper and bank loans. December 31. the average bank holding company’s loans accounted for 64. United States Federal Reserve. 2 5 . investors and depositors will not be able to distinguish troubled financial institutions from healthy ones. Thus.. We use the financial reports of 2. including the borrower’s track record. bank insiders have more information about bank loans than do equity investors or depositors.g. At the end of our sample period.

The information content of stock returns Several studies seek to explain the relation between information asymmetry (opacity) and R2 from asset pricing regressions. but have lower trade volume and return volatility than non-financial firms of similar size and stock price. Morgan (2002) uses “split ratings” from bond rating agencies as a measure of opacity. ceteris paribus. “the panic did not produce significant social costs in terms of failures among solvent banks.that. leading the authors to conclude that the assets of large banks and the assets of large non-financial firms are similar in opacity. (2004) use microstructure data and analyst estimate data to show that large banks (defined as those traded on NYSE and AMEX) exhibit similar microstructure properties to large non-financial firms. The logic behind this interpretation is that. However. which he interprets as indicating that bank assets are more opaque than the assets of non-financials. a transparent firm’s bonds would receive identical ratings from the two major bond rating agency because information regarding the firm is readily available. 3. NASDAQ banks have similar bid-ask spreads. if the firm is opaque. Roll (1988) in his “R2” Presidential Address to the AFA. rating agencies might reach different conclusions about the prospects of the firm based on the limited available information. At least one study provides evidence that bank assets are opaque. discusses the large proportion of stock price movement that cannot 6 . and finds that banks are more likely to carry split ratings than non-financials.” All three of these studies indicate that bank assets are relatively transparent. (2004) interpret the latter findings as the assets of NASDAQ banks being “boring. although depositors were temporarily confused about bank asset quality. Flannery et al.” Flannery et al.

and posits that lower market model R2s reflect greater activity on the part of informed traders. insiders capture a portion of the firm’s operating cash flows. Jin and Myers (2006) explain how control rights and information affect the division of risk bearing between inside managers and outside investors. Durnev et al.R2 from an asset pricing model regression). Information asymmetry exists between insiders and outside investors. consistent with lower R2 firms exhibiting lower opacity. Morck et al. When firm-specific 7 . and posit that stronger public investor property rights in rich economies promote informed arbitrage. which are assumed to impound firm-specific information into equity prices. 1 . Outside investors are unable to observe firm-specific news perfectly.be attributed to systematic factors (in essence. we are able to use R2 for asset pricing model regressions as a proxy for the opacity of a firm. Jin and Myers (2006) develop an alternative model to explain the findings of Morck et al. which allows insiders to capture a portion of the unexpected positive earnings on good news. firms with more firm-specific information in their returns exhibit lower R2 for asset pricing model regressions. Piotroski and Roulstone (2004) show that stock return synchronicity is inversely related to insider trades. (2004) show that stocks exhibit lower R2 in countries with a freer press and a more developed financial analysis industry. which allows the incorporation of firm-specific information into asset prices. Thus. Bushman et al. As a result. (2000) provide international evidence of higher R2 in poor economies than in rich economies. (2003) show that firms with lower R2 exhibit a higher association between current returns and future earnings. The limits to capture are based on outside investors’ perception of the value of the firm. In their model. (2000).

Because insiders report neither good nor bad firm-specific news. and outsiders bear mostly systematic risks. or even to exaggerate. 4. insiders bear most of the firm-specific risk. Such an opaque firm will display a higher R2 in asset pricing model regressions. the bad news so they can maintain or increase the proportion of firm earnings that they capture. bad news only becomes credible when insiders pass it along at a cost to themselves. As a result. Crashes occur when insiders can absorb no more firm-specific bad news and decide to abandon the firm. we predict higher asset pricing regression R2 for banks than for a matching set of non-regulated industrial firms. finding strong positive relations between R2 and several measures of opacity. Outside investors recognize the incentive for insiders to announce bad news (whether it be true or not) and. 8 . insiders have the incentive to pass along. Hypotheses and empirical predictions Banks are more opaque than matched firms in other industries. consistent with insiders bearing the firm-specific risk. Jin and Myers test their predictions using stock returns from 40 stock markets from 1990 to 2001. as a result. The Jin and Myers (2006) model also predicts that more opaque stocks are more likely to “crash. the stock price for a firm with high opacity tends to vary with systematic factors. Conventional wisdom holds that bank assets are more informationally opaque (less transparent) than assets of industrial firms. H1.news is bad. These measures also explain the frequency and magnitude of large negative returns or crashes.” that is. Given the positive relation between R2 and opacity posited by the recent literature. Thus. insiders must reduce their capture of the firm’s cashflows in bad times to avoid costly reporting of bad news. to deliver large negative returns.

and the fact that IBES analyst earnings forecasts are more accurate for NASDAQ banks than for matching industrial firms. then multivariate regressions of opacity measures on independent variables including an indicator for NASDAQ firms and an indicator for banks should produce coefficients on these two indicators that cancel out or at least sum to a number not statistically significantly different from zero. If the conclusion of Flannery et al. However. then the coefficient on the bank indicator should be positive and significant and the coefficient on the NASDAQ indicator should be either be positive. the low levels of trading and accurate analyst forecasts are also consistent with a lack of firm-specific information in returns. then little firm-specific news exists upon which to trade. resulting in more accurate analyst earnings forecasts for more opaque firms. then the reported earnings of the firm will most likely move in step with other firms in the industry. We also perform regressions 9 . Such earnings are easier for analysts to forecast than earnings containing large amounts of firm-specific information. not significantly different from zero. (2004) conclude that NASDAQ banks are not more opaque than their industry counterparts. and not include firmspecific information (see Kirschenheiter and Melumad (2002)). they are merely “boring. Flannery et al. resulting in low trading frequency. However. Also.” Their conclusion is based on NASDAQ banks trading much less often than industrial matching firms despite having similar bid-ask spreads. if opacity allows managers to “manage” earnings. if such banks are more opaque than their industrial counterparts as we predict. Banks traded on NASDAQ are more opaque than matching firms.H2. or if negative and statistically significant. If banks are opaque. (2004) is valid. of a smaller magnitude than the positive coefficient on the bank indicator.

and finds that the composition of bank assets significantly affects the probability of a split rating. Morgan (2002) uses “split ratings” from bond rating agencies as a measure of opacity. H5. Banks with higher information opacity are more likely to “crash. We predict a positive relation between R2 values from asset pricing model regressions including industry returns and the proportion of a bank’s assets represented by these “opaque” assets.including the interaction of these two indicators to examine the difference in opacity between NASDAQ and non-NASDAQ banks. If banks are more opaque than matching industrial firms. then one might expect banks to experience more crashes than matching industrial firms. ceteris paribus. The model of Jin and Myers (2006) predicts that more opaque firms are more likely to experience crashes than less opaque firms. we expect banks with high R2 values from asset pricing model regressions including industry returns. Firm-level opacity of banks depends on the composition of their assets. Flannery et al. to have a higher probability of a crash. H4. more opaque banks are more likely to experience crashes than less opaque banks. (2004) define types of bank assets that are more opaque. thus. Banks are more likely to experience “crashes” than matching industrial 10 . H3. which are representative of greater opacity. more opaque.” Similar to the previous hypothesis. We also determine which types of loans are associated with higher R2 values and are. Therefore. firms.

Matching firms must survive the entire calendar year.5. Using a sample period of 1993-2002 and their filters leaves me with 243 bank holding companies. If not. We define industries using 2-digit SIC codes. (2004) in the elimination of “observations that seem likely to produce unrepresentative values. We match firms first by exchange. Our final sample consists of 1. market value of equity.186 bank-year observations and an equal number of matching industrial firm-year observations. which we will refer to interchangeably as BHCs or banks. and share price. If the share price of the industrial firm is within 25 percent of the bank’s share price.” Specifically. calculate value-weighted weekly industry 11 . we omit any firm-year (bank or match) for which the average share price is less than $2 or the stock has fewer than 400 trades. Data and methods We follow Flannery et al. We also follow Flannery et al. then by market equity. We re-select each bank’s matching firm at the start of each calendar year. We also perform the same regression including an additional industry excess return factor. We select matching firms from CRSP (except for financial firms (SIC code 6000-6999) or regulated utilities (SIC code 4800-4900). Each calendar year. we then choose the industrial firm as a match. we select the next-closest market equity firm and subject it to the same share price test until we find an appropriate matching firm listed on the same stock exchange as the bank. we calculate R2 measures for all banks and matching firms through regression of excess weekly returns on the excess market return. (2004) in the selection of banks and matching firms. which are excluded from the sample) on the basis of stock exchange. We obtain share price and number of trades from CRSP.

These variables are: DIVERS = revenue-based Herfindahl index of firm diversification using business segments reported in Compustat segment data. available from 1996. 12 . into a continuous variable with a more normal distribution (Piotroski and Roulstone (2004)).t (2) (1) BETAINDRSQ: + β4INDRETi.t = α + β1MARETi. We use control variables suggested by Piotroski and Roulstone (2004) and Flannery et al.t + β2MARETi. Specifically.t + β2MARETi. bound by zero and one. we create two measures of the incremental power of industry-level returns for explaining firm-specific returns as DIFFRSQBETA = BETAINDRSQ – BETARSQ and log transform all three of these measures as follows:  BETARSQ  SYNCHB = log  1  BETARSQ      BETAINDRSQ  SYNCHBI = log  1  BETAINDRSQ      DIFFRSQBETA  DIFFB = log  1  DIFFRSQBETA     (4) (3) (5) (6) Log transformation changes the R2 variable.t = α + β1MARETi.excess returns. (2004) to demonstrate the robustness of our findings to inclusion of items previously shown to be related to opacity or R2.t-1 + β3INDRETi. and include lagged values of all. we calculate the R2 measures for the following regressions: BETARSQ: RETi. Piotroski and Roulstone (2004) include the lagged values of regressors “since the presence of informed parties can impact the timing of the market and industry information’s incorporation into prices” (page 1123).t-1 RETi.t-1 Also following Piotroski and Roulstone (2004).

MB. ΔINST. REV = ΔINST = MVE = TRDSZE = NEST = CSD = FE = SHRTURN = log(. calculated by multiplying share price by the number of shares outstanding from CRSP. FE.HERF = STDROA = RETCORR = FUNDCORR = revenue-based Herfindahl index of 2-digit SIC industry-level concentration using Compustat annual data. and TRDSZE or the log of one plus REV. annual standard deviation of quarterly return on assets (ROA) from Compustat. number of analysts revising their forecast during the month divided by the number of analysts following the firm.) = We also control for leverage (the proportion of assets funded by debt) due to the large difference in this attribute between banking and industrial firms. CSD. the absolute change in the number of shares held by institutions. Spearman correlation between weekly market returns and valueweighted industry returns. SHRTURN. NEST. the log of MVE. FUNDCORR is calculated each year using the previous twelve quarterly observations. NIND. average number of shares per transaction during the calendar year (in thousands). the logarithmic transformation of the R2 from a regression of the firm’s quarterly return on assets (ROA) on a value-weighted index of ROA. number of analysts posting an earnings forecast for the firm’s current fiscal year.000 (to measure forecast error in basis points). median absolute EPS forecast error. divided by share’s price at the start of the fiscal year. 13 . computed only for firms with more than one analyst. and multiplied by 10. and HERF. INST. as a fraction of all shares outstanding. annual share volume divided by shares outstanding (from CRSP) multiplied by 1000. the market value of equity in thousands of dollars at the beginning of the year. cross-sectional standard deviation of analysts’ forecasts.

We treat DIVERS differently than other control variables because of the way in which we define DIVERS. the coefficient on the indicator variable accounts for variation between observations with and without data for the control variable. (2004) we create the following variables for the composition of each bank’s portfolio of assets: TRADE = the fair value of assets held in trading accounts (includes government debt. commercial paper. normalized by the market value of equity. Firms with no data in the Compustat segment dataset are likely to be undiversified. A value of one for DIVERS represents an undiversified firm. OPAQUE = HIGH_CB = 14 . We include the indicator for a variable with missing data in all regressions in which the control variable is used. we set missing observations equal to zero and create an indicator value equal to one if the variable has been set to zero and equal to zero otherwise. an indicator equal to one if the BHC’s subsidiary commercial bank assets (summed from individual bank call reports) exceed the sample median proportion of total BHC assets. For most variables with a sizable proportion of observations missing (ΔINST. Following Flannery et al.” normalized by the market value of equity. we set missing observations equal to one and create an indicator value equal to one if DIVERS has been set to one and equal to zero otherwise.We follow Palia (2001) and Fama and French (2002) in taking steps to preserve sample size in the presence of missing data. and bankers acceptances). For DIVERS. and “other assets. REV and FE). NEST. whereas a value of zero represents an infinitely diversified firm. CDs. CSD. intangible assets. plus investments in unconsolidated subsidiaries. Thus. the book value of bank premises and fixed assets.

Our final sample of 243 banks contains 217 NASDAQ firms. and 2 AMEX firms. 24 NYSE firms. No sample bank switches exchanges during the sample period. percentage of loan portfolio accounted for by real estate loans. BETACT. and zero if the firm is listed on the American Exchange (AMEX) or the New York Stock Exchange (NYSE). We also create a variable. to represent the portion of a bank’s assets represented by loans. We calculate the number of residual returns from the market model regression (see equation (1)) exceeding 2. We subtract the upside frequencies from the downside frequencies to arrive at BETACT. Panel A describes the construction of the opacity proxies.Following Beatty et al. Table 1 presents variable descriptions and descriptive statistics for sample banks and matching firms. percentage of loan portfolio accounted for by individual (consumer) loans. and indicators for each Federal Reserve District. Panel B 15 . percentage of loan portfolio accounted for by loans to depository institutions.576 standard deviations above and below the mean to generate a frequency of 1 percent in the lognormal distribution. percentage of loan portfolio accounted for by commercial and industrial loans. we construct two indicator variables. NASDAQ equals one if the firm is listed on the NASDAQ exchange. and equals zero if the firm is a matching firm. (2002). BANK equals one if the firm is a bank. we create variables to represent the composition of each bank’s loan portfolio as follows: LOANAG = LOANCI = LOANDEP = LOANIND = LOANRE = percent of loan portfolio accounted for by agricultural loans. Finally. PCTLOANS. We follow Jin and Myers (2006) in construction of our crash prediction variable.

but analysts produce better estimates of bank earnings (mean REV = 0. mean ΔINST = 9. on the concentration of banking.910. (2004).130) than banks (mean NEST = 3. mean 16 .880).409) makes banks appear to be more diversified than their matching firms (mean DIVERS = 0. If we assume that firms with no segment information in Compustat are not diversified at all (DIVERS = 1). Panel E provides descriptive statistics for sample banks. one must be careful to note that the Compustat segment data reports only five firm-year observations for banks.695) than of their industrial counterparts (mean INST = 38.724. then banking firms have a mean DIVERS of 0. The banking industry is also less concentrated (mean HERF = 0. Large differences exist between banks and matches in firm characteristics. and Panel D defines our control variables. Large differences also exist between banks and matches in analyst statistics.details the construction of our crash prediction variable. consistent with historical regulatory restrictions. mean ΔINST = 3. Panel C details construction of the bank asset characteristics variables. Institutions own and trade smaller portions of banking firms (mean INST = 17. as opposed to 565 firm-year observations for matching firms. Consistent with the findings of Flannery et al.476). banks are less liquid (mean SHRTURN = 4.286. Although the lower value of DIVERS for banks (mean DIVERS = 0. indicating that banks are less diversified than matching firms.427) than matching industrial firms (mean SHRTURN = 21. Panel F provides the same statistics (for variables common to both) for matching firms.071).176).811). More analysts follow matching firms (mean NEST = 5. such as branching limitations.554.025) versus matching firm industries (mean HERF = 0.998 versus their industrial matches mean DIVERS of 0.

mean FE = 0. Next. with the null hypothesis that the medians of these measures are equal for banks and matching firms. We calculate t-statistics based on standard errors derived from the empirical distribution of the annual coefficient estimates following Fama and MacBeth (1973). managers smooth earnings until firm-specific news is sufficiently bad. perhaps their earnings are “smoother” and therefore easier to predict. Assuming a normal distribution of these measures (which holds more strictly for the logarithmic transformations of the R2 measures). leading to the analyst statistics we observe and the observations of Flannery et al.006) than of matching firm earnings (mean REV = 0. (2004) regarding NASDAQ banks. with the null hypothesis that the means of these measures are equal for banks and matching firms.073. We begin with univariate tests of the statistical significance of the difference in the opacity measures between banks and matching firms.032. we perform the Mann-WhitneyWilcoxon z-test for differences in medians. Following Piotroski and Roulstone (2004). The resulting coefficients are the average coefficients from ten annual estimations. we perform Fama and MacBeth (1973) regressions to determine the relation between our opacity proxies and the independent variables. we perform Student’s t-test for the comparison of two population means.CSD = 0. we perform multivariate analysis to examine differences in opacity measures between banks and matching firms. In the model of Kirschenheiter and Melumad (2002). earnings management is easier to achieve in opaque firms.431. 17 . mean CSD = 0. If banks are more opaque.” Presumably. Additionally. when they under-report earnings by the maximum amount in what has come to be known as a “big bath.016). mean FE = 0.

SYNCHBI.We use control variables suggested by Flannery et al. We also perform Fama and MacBeth (1973) regressions to determine the relation between our firm-specific opacity proxy. on the BANK indicator and various control variables. and bank asset composition. We use proxies for proportions of different types of assets (TRADE and OPAQUE) developed by Flannery et al. and also control for differences between Federal Reserve Districts by creating indicator variables for districts one through ten (to avoid over-specifying the model). We include the control variables of Piotroski and Roulstone (2004) and Flannery et al. We also use the percentage of assets 18 . (2004) to control for variables previously shown to be related to opacity. SYNCHBI. tend to be geographically undiversified. BETACT. Controlling for geographic differences is important because smaller banks. we perform Fama and MacBeth (1973) regressions to determine the relations between our firm-specific opacity proxy. following Jin and Myers (2006). The resulting coefficients are the average coefficients from ten annual estimations. we perform Fama and MacBeth (1973) regressions of our crash predictor variable. BETACT. we calculate t-statistics based on standard errors derived from the empirical distribution of the annual coefficient estimates following Fama and MacBeth (1973). although perhaps diversified over several industries and types of loans. (2004). Finally. Our information regarding bank assets comes from the FR Y-9C bank holding company reports. which might affect our results if the opacity of loans varies by locale. (2004) and Piotroski and Roulstone (2004). and our crash predictor variable. To determine whether the greater opacity of banks leads to more frequent crashes among banks than matching firms. for a banks-only sample.

10.3 percent commercial and industrial loans. in direct opposition to any hypothesis of bank asset opacity.represented by loans (PCTLOANS). The average bank holding company in our sample holds 64. The average sample bank holding company’s loan portfolio is composed of 1. and the percentage of the loan portfolio represented by agricultural (LOANAG). 6. However. positive and significant differences should exist. commercial and industrial (LOANCI). the log transformation of R2 from the market model 19 . and 66.2 percent agricultural loans.7 percent of its assets in the form of loans.9 percent consumer loans. if banks are more opaque than matching firms.3 percent is composed of loans to foreign governments and “other” loans. 0. Panel E of Table 1 provides descriptive statistics for these variables. All statistically significant differences recorded in Table 2 are negative. The remaining 3. consumer (LOANIND). Market model: multivariate analysis Table 3 presents the results of multivariate tests of the difference between banks and matching firms in SYNCHB. Ceteris paribus.2 percent real estate loans. 18. banks are fundamentally different from matching firms along many dimensions previously shown to be related to R2. We test whether the various opacity proxies are significantly different between banks and matches by examining whether the differences in means and medians (banks – matches) are significantly different from zero. Results Univariate tests Table 2 presents the results of the univariate examination. so we withhold judgment on the opacity of bank assets pending the results of multivariate tests. and real estate (LOANRE) loans. depository institution (LOANDEP).1 percent loans to depository institutions.

Larger firms experience higher R2 in asset pricing model regressions. consistent with the findings of Roll (1988). and two sets of control variables suggested by Piotroski and Roulstone (2004) and Flannery et al. However. while models 6 through 10 use both sets. consistent with greater R2 among banks. but is never significant in any regression containing the BANK indicator and the full set of control variables. (2004). This result is supportive of H1. that banks are more opaque than matched firms in other industries. given the large difference in leverage between banks and matching firms. Neither the coefficient on the BANK indicator nor the coefficient on the NASDAQ indicator achieves significance in the any of the first five models. the BANK indicator coefficient becomes positive and significant. Ceteris paribus. The coefficient on the size control is positive and significant in all models. Flannery et al.regression. the NASDAQ indicator. The coefficient on leverage is significant in about half the regressions. after inclusion of the second set of controls.07 percent of a standard deviation higher among banks than matching firms using the coefficient on the BANK indicator from model 10. (2004) assert that this is the appropriate control for size (as opposed to assets) given that “equity traders experience valuation uncertainty in proportion to their equity claim. We perform Fama and MacBeth (1973) regressions of SYNCHB on the BANK indicator. 20 . Models 1 through 5 use only the former set of controls. The coefficients on the NASDAQ indicator and the NASDAQBANK interaction are not significantly different from zero in any of the models. which supports H2.” All models also control for leverage. All models control for firm size through inclusion of the natural log of the market-value of equity. log(MVE). SYNCHB is 47.

model 5 produces a different result. Table 4 presents the results of these regressions. but more opaque than matching firms. The coefficient on BANK is larger in magnitude than the coefficient on the interaction term. The coefficient on BANK is larger in magnitude than the coefficient on the interaction term. To address this problem. resulting in less firm-specific information in matching firm returns than in bank returns. Matching firms might load more heavily on industry returns. Similar to the set of models with only the Piotroski and Roulstone (2004) controls. Once again.Market model plus industry returns: multivariate analysis The transformed R2 used in the Table 3 regressions (SYNCHB) represents the proportion of return volatility that can be accounted for using market returns. the coefficient on the size proxy is positive and significant in every model. with a positive and significant coefficient on BANK and a negative and significant coefficient on the interaction term. so one cannot conclusively state from these results that bank returns contain less firm-specific information than matching firm returns. consistent with NASDAQ banks being less opaque than non-NASDAQ banks. consistent with NASDAQ banks being less opaque than non-NASDAQ banks. which would indicate greater opacity among banks. with no statistically significant coefficients on either of the indicator variables. The unexplained variation contains both industry-specific information and firm-specific information. and the coefficient on leverage only attains significance in the models with all control variables when the BANK indicator is absent. but 21 . we run identical regressions using SYNCHBI as the dependent variable. A similar pattern to Table 3 emerges for models 1 through 4. the final model has a positive and significant coefficient on BANK and a negative and significant coefficient on the interaction term. However.

We use their method to examine the difference in incremental impact of industry returns between banks and matching industrial firms. SYNCHBI is 57. Therefore. Differential impact of industry returns: multivariate analysis Piotroski and Roulstone (2004) subtract the market model R2 from the market model plus industry returns R2 to examine the incremental impact of industry returns on the returns of individual firms. the interaction term NASDAQBANK also loses significance. consistent with H2. Neither the BANK indicator coefficient nor the NASDAQ indicator coefficient is significant in any specification of Table 5. indicating no difference between banks and matches and no difference between NASDAQ and nonNASDAQ firms in the proportion of return variation that can be attributed to industry information. 22 . Table 5 presents the results from regressions of DIFFB on the same independent variables used in Tables 3 and 4. We use the logarithmic transformations of the difference in R2 between the two models (DIFFB) to perform an examination similar to that of Piotroski and Roulstone (2004). and that no difference exists in this respect between NASDAQ and non-NASDAQ firms. Ceteris paribus. leading to a conclusion of greater opacity among banks.61 percent of a standard deviation higher among banks than matching firms using the coefficient on the BANK indicator from model 10.more opaque than matching firms. we conclude that bank returns vary more with market. Once all control variables are included. not industry. This result is consistent with bank returns containing less firm-specific information than matching firm returns. consistent with bank returns varying more with market and industry returns than matching firm returns. returns in response to the lack of firm-specific information which characterizes bank opacity.

bank insiders have at least two other disciplining bodies.” or significant negative equity return. then banks should be more likely to crash than industrial firms. One might argue that deposit insurance disarms this disciplining force.000) and providers of Fed funds are not insured and are. large depositors (> $100. However. Diamond and Rajan (2001) show how the sequential service constraint (first come. The extra 23 . However.Bank opacity and “crash” likelihood We test the prediction of Jin and Myers (2006) that more opaque firms have a higher likelihood of experiencing a “crash. log(MVE). on the BANK indicator. the NASDAQ indicator. Does this unexpected result mean that banks are less opaque than industrial firms? Not necessarily. Our regression results show the exact opposite. shareholders are the only source of discipline for insiders. and more opaque firms are more likely to crash. Such supervision might reduce the incentives for banks to take on risky projects. and such discipline can only be achieved at a cost. H3 is not supported. bank insiders are less willing to take on risky projects. Thus. If banks are more opaque than matching industrial firms. with a negative and significant coefficient on the BANK indicator in every regression model in which the indicator is included. incentivized to monitor banks and pull their funds at the first sign of problems. Additionally. our crash predictor. depositors and regulators. Table 6 presents the results from regressions of BETACT. and various controls. In the model of Jin and Myers (2006). thus. banks are subject to supervision by government regulators. first served) on deposit withdrawals leads depositors to withdraw their funds when they perceive (or suspect) that managers are undertaking projects of greater risk than depositors desire. leverage. As a result.

a crash likelihood proxy. Table 7 presents the results from regressions using the reduced sample of banking firms (no matches) of SYNCHBI. the coefficient on BETACT is positive and significant. with a negative coefficient.discipline provided by depositors and regulators might be responsible for banks experiencing fewer crashes than matching industrial firms. and controls suggested by Piotroski and Roulstone (2004) and Flannery et al. 24 . After all controls (including the NASDAQ indicator) are included. healthy competitor banks might acquire struggling banks. In any acquisition. (2002). on BETACT. Additional control variables include the Federal Reserve District indicators. OPAQUE. as in all previous specifications. the acquiring firm must offer the target firm’s shareholders a premium to induce them to sell. (2004). including bank size. that banks with higher information opacity are more likely to “crash” than more transparent banks. The opacity of specific bank assets In this section. Table 8 presents results from regressions of SYNCHBI on the percentage of the bank’s assets held as loans. the TRADE. and HIGH_CB variables suggested by Flannery et al. an opacity proxy. consistent with H4. The nonzero probability of such a takeover combined with this premium may limit downward movements in a struggling bank’s equity prices. our proxy for bank size. but the NASDAQ indicator attains significance for the first time. (2004). and various controls. The coefficient on the size proxy is positive and significant. and the loan portfolio characteristic variables suggested by Beatty et al. Finally. we decompose bank opacity to determine which types of assets are responsible for bank opacity.

Therefore. a one standard deviation decrease in LOANAG relates to an increase in SYNCHBI of 16. In all specifications containing the loan portfolio characteristics variables. Active spot and futures markets provide price information for these loans’ underlying crops. LOANIND has a negative and significant coefficient. albeit with a ten percent alpha. but loses significance once the Federal Reserve District indicators are included.The coefficients on TRADE and OPAQUE are positive and significant in models 1 and 5. LOANAG has a negative and significant coefficient. one can surmise for which crops the bank has loaned money. Local weather reports and agricultural reports are available free to the public. it is reasonable to consider that 25 . a one standard deviation decrease in LOANIND relates to an increase in SYNCHBI of 33.93 percent of a standard deviation.44 percent of a standard deviation. The only variable to maintain significance throughout all models is the proxy for size. Similarly. Government-backed crop insurance removes some risks associated with agricultural loans. The coefficient on HIGH_CB is negative and significant in models 2 and 3. consistent with Roll (1988). Using coefficients from model 8 of Panel A. ceteris paribus. Why might agricultural loans be more transparent than commercial and industrial loans? Given the time of year and geographic location of a bank. Our evidence is consistent with greater opacity among banks with lower proportions of agricultural and consumer loans in their portfolios. In all specifications containing the loan portfolio characteristics variables and control variables. which states that the opacity of banks depends on the composition of their assets. ceteris paribus. This evidence supports H5. but lose significance when control variables are introduced.

consistent with the smaller magnitude of both the coefficient and t-statistic for LOANIND than for LOANAG. consistent with regulators and depositors providing additional disciplining forces upon banks which are not experienced by industrial matching firms. consumer loans might be more transparent due to the high proportion of automobile loans contained in this category. One could make the reasonable assumption that the cross-section of autos serving as collateral for the bank is similar to the cross-section of vehicles registered in the region within the last five years. This conclusion is robust to inclusion of industry returns in the asset pricing model. (2004). Implications and conclusions In this study. knowing how many of each make of automobile serves as collateral for loans granted by a given bank is difficult. 7. We find that banks are less likely to experience crashes than matching industrial firms. We show that. while NASDAQ banks might be less opaque than banks traded on NYSE or AMEX. in contrast to the conclusions of Flannery et al. they are not less opaque than matching industrial firms. Why might consumer loans be more transparent than commercial and industrial loans? Although not as straightforward to explain as agricultural loans. We also present evidence that more opaque banks are more likely to experience sudden drops in their 26 . the composition and risk of which cannot be so easily ascertained. but the composition of consumer loans would still be estimated with less certainty than the composition of agricultural loans. However. we provide evidence consistent with banks being more opaque (less transparent) than matching industrial firms. An active secondary market in automobiles provides market prices for the assets underlying the loans.these loans might be more transparent than commercial and industrial loans.

equity valuations than less opaque banks. but they also represent 53. Assets previously identified as affecting bank opacity tend to lose statistical significance after the inclusion of appropriate control variables and loan portfolio characteristics. Finally. insiders take advantage of opacity to capture a portion of firm cash flows. Morgan and Stiroh (2001) 27 . the BHCs of our sample represent only 15. In the model of Jin and Myers (2006). therefore. deserve greater regulatory oversight. we present evidence consistent with agricultural loans and consumer loans being more transparent than other types of loans made by banks.83 percent of all assets held by BHCs. consistent with a theoretical prediction of Jin and Myers (2006). Banks with high R2 values from asset pricing model regressions are more opaque. Therefore. Banks with greater R2 provide insiders with a greater opportunity to extract rents from shareholders and depositors and. Admittedly. Our results provide a new tool for regulators to aid in the identification of publicly traded banks that deserve greater regulatory scrutiny. our measure could be a useful tool for monitoring the majority of bank holding company assets in the United States.97 percent of all BHCs at the end of 2003.

R. Melumad. and J. 257-292. A. 689-721. B." Review of Financial Studies 15. Zarowin.G. "Can "Big Bath" And Earnings Smoothing Co-Exist as Equilibrium Financial Reporting Strategies?.Emprical Tests. 1988. 397-412.J.." American Economic Review 87. Piotroski and A. Rajan." Journal of Finance 35. and the Theory of Financial Intermediation. Market Signalling. 2002. "What Determines Corporate Transparency?. "Does Greater Firm-Specific Return Variation Mean More or Less Informed Stock Pricing?." Journal of Financial Economics 71. and N. T.R. Jin. and S. B.D. Kirschenheiter. and J. "Contagion and Bank Failures During the Great Depression: The June 1932 Chicago Banking Panic. "Market Evidence on the Opaqueness of Banking Firms' Assets. Myers. "Information Production.F.R. Ke and K." The Accounting Review 77. Campbell. Diamond.References Beatty.F. Berlin. 419460. S. and K. 2006. M." Journal of Accounting Research 41.A.. Fama. Calomiris. 1-33. 1980. E. 607-636. and R.. R. MacBeth. Return and Equilibrium .." Journal of Accounting Research 40. 207-252. Diamond.R." Journal of Financial Economics 79. "Bond Covenants and Delegated Monitoring." Journal of Finance 43. 2002.L. D. 2004.M.W. A. Nimalendran.W. 797-836. J." Journal of Political Economy 99. 1991. Mason. 28 . "Liquidity Risk. and J. D. and W. 1997. 863-882. 863-883. "Risk. 1973.S. Flannery.H.. Kwan and M. 2002. 2001." Journal of Political Economy 109. Durnev. Loeys. 2003. "R-Squared around the World: New Theory and New Tests. Bushman. M. "Monitoring and Reputation: The Choice between Bank Loans and Directly Placed Debt. Liquidity Creation. French. Kracaw. E. Smith.J. 761-796. Fama. Petroni. L. 287-327. "Earnings Management to Avoid Earnings Declines across Pubicily and Privately Held Banks. and Financial Fragility: A Theory of Banking.D.C. 2004. Yeung and P." The Journal of Political Economy 81. Morck. C." Journal of Accounting Research 42.W.D. M. "Testing Trade-Off and Pecking Order Predictions About Dividends and Debt. 547-570.

Palia.F.J. "Rating Banks: Risk and Uncertainty in an Opaque Industry. D. 874-888. "R2." Journal of Finance 43. Institutional Investors. Yeung and W.. 2000. and Bank Security Returns in 1987: The Brazilian Experience. Roulstone. 1988. "The Influence of Analysts. and Banking 22. Piotroski. R. 195-208.P. Morgan. and J. 1990.. Morgan.P." Journal of Financial Services Research 20. Credit. 2004. D. Stiroh. "The International Debt Crisis. J. Roll. Industry." American Economic Review 92. 215-260." Journal of Financial Economics 58. 735-764. 2001. 1119-1151. 2001.J. Yu. and Insiders on the Incorporation of Market. "Market Discipline of Banks: The Asset Test. Investor Contagion.Morck. Musumeci. 29 . B. and K. D.D." Accounting Review 79. "The Endogeneity of Managerial Compensation in Firm Valuation: A Solution. R." Review of Financial Studies 14. 541-566." Journal of Money. and FirmSpecific Information into Stock Prices. J..T.. 209-220. Sinkey. 2002. "The Information Content of Stock Markets: Why Do Emerging Markets Have Synchronous Stock Price Movements?. and D.

We subtract the upside frequencies from the downside frequencies. 30 .Table 1 Variable definitions and descriptive statistics Panel A. Returns informativeness proxies BETARSQ = SYNCHB = coefficient of determination from yearly regression of weekly returns on the value-weighted CRSP excess market return.576 standard deviations above and below the mean to generate a frequency of 1 percent in the lognormal distribution. Crash predictor BETACT = number of residual returns from the market model regression exceeding 2. BETAINDRSQ = SYNCHBI =  BETAINDRSQ  log  1  BETAINDRSQ     BETAINDRSQ – BETARSQ DIFFRSQBETA = DIFFB =  DIFFRSQBETA  log  1  DIFFRSQBETA     Panel B.  BETARSQ  log  1  BETARSQ     coefficient of determination from yearly regression of weekly returns on the value-weighted CRSP excess market returns and value-weighted excess 2-digit SIC code industry returns.

Bank asset portfolio characteristics TRADE = OPAQUE = PCTLOANS = HIGH_CB = LOANAG = LOANCI = LOANDEP = LOANIND = LOANRE = the fair value of assets held in trading accounts (includes government debt. and “other assets. Control variables DIVERS = HERF = STDROA = RETCORR = FUNDCORR = revenue-based Herfindahl index of firm diversification using business segments reported in Compustat segment data. percentage of loan portfolio accounted for by loans to depository institutions. as a fraction of all shares outstanding. percentage of loan portfolio accounted for by real estate loans. the book value of bank premises and fixed assets. an indicator equal to one if the BHC’s subsidiary commercial bank assets (summed from individual bank call reports) exceed the sample median proportion of total BHC assets. Panel D.Panel C. percentage of loan portfolio accounted for by commercial and industrial loans. average number of shares per transaction during the calendar year (in thousands). available from 1996. number of analysts posting an earnings forecast for the firm’s current fiscal year. FUNDCORR is calculated each year using the previous twelve quarterly observations. annual standard-deviation of quarterly return on assets (ROA) from Compustat. Spearman correlation between weekly market returns and value-weighted industry returns. the absolute change in the number of shares held by institutions. plus investments in unconsolidated subsidiaries. commercial paper. and bankers acceptances). number of analysts revising their forecast during the month divided by the number of analysts following the firm. the logarithmic transformation of the R2 from a regression of the firm’s quarterly return on assets (ROA) on a value-weighted index of ROA. percentage of loan portfolio accounted for by agricultural loans.” the percentage of bank assets accounted for by loans. revenue-based Herfindahl index of 2-digit SIC industry-level concentration using Compustat annual data. intangible assets. the market value of equity in thousands of dollars at the beginning of the year. calculated by multiplying share price by the number of shares outstanding from CRSP. CDs. REV = ΔINST = MVE = TRDSZE = NEST = 31 . percentage of loan portfolio accounted for by individual (consumer) loans.

32 . and equal to zero otherwise. the log of MVE. and multiplied by 10. FE. and equal to zero otherwise. zero otherwise.CSD = FE = SHRTURN = log(. an indicator equal to one if a missing value of log(CSD) is set to zero. an indicator equal to one if a missing value of DIVERS is set to zero. annual share volume divided by shares outstanding (from CRSP) multiplied by 1000.000 (to measure forecast error in basis points). and equal to zero otherwise. an indicator equal to one if a missing value of log(FE) is set to zero. ΔINST. and equal to zero otherwise. SHRTURN. and equal to zero otherwise. MB. CSD. NIND. an indicator equal to one if the firm is traded on NASDAQ. an indicator equal to one if a missing value of log(ΔINST) is set to zero. an indicator equal to one if a missing value of log(NEST) is set to zero. NEST.) = DIVERSmissing = log(ΔINST)missing = log(NEST)missing = log(CSD)missing = log(FE)missing = NASDAQ = cross-sectional standard deviation of analysts’ forecasts. median absolute EPS forecast error. computed only for firms with more than one analyst. and HERF. INST. and TRDSZE or the log of one plus REV. divided by share’s price at the start of the fiscal year.

435 PCTLOANS 0.601 0.236 0.005 0.000 0.420 396264.415 0.387 0.289 0.000 0.000 0.676 1.001 INST 17.023 0.230 -1.164 1.000 0.936 0.061 12.458 0.417 0.360 0.577 0.130 2291599.066 1.468 -0.589 0.012 0. Dev.299 0.238 0.417 0.920 120298.000 n 1186 1186 1186 1186 1186 1157 5 1186 1127 1009 929 1186 1109 1186 1186 1186 1186 1141 800 800 571 797 1055 1055 1055 1055 1055 1055 1055 1055 1055 1186 Firm characteristics DIVERS 0.286 CSD 0.090 0.308 0.880 REV 0.018 OPAQUE 0.357 0.000 75th Pctl.000 0.587 SHRTURN 4.001 25.143 -1.023 -3.000 0.0 -0.029 0.006 0.001 0.648 5.214 Std.902 4.017 -3.089 0.000 0.038 0.006 Asset portfolio characteristics TRADE 0.001 16.000 0.090 20797.892 0.744 TRDSZE 9.133 7.000 0.000 0.135 0.000 0.172 -1.109 0.427 LEVERAGE 0.790 0.563 0.4 FUNDCORR -2.172 0.901 1.256 551.730 1.9 -2.004 0.000 0.725 0.409 HERF 0.116 0.571 0.183 LOANDEP 0.000 Mean Opacity proxies BETARSQ SYNCHB BETAINDRSQ SYNCHBI DIFFRSQBETA DIFFB 0.620 0.025 4.000 0.820 13.100 0.544 0.550 4.250 0.0 -3.000 25th Pctl.002 0.162 0.911 IBES analyst coverage NEST 3.026 0.000 0.329 86.498 3.165 0.698 LOANAG 0. 0.001 LOANIND 0.913 0.025 STDROA 0.063 3.042 0.026 -3.006 -5.909 33 .778 2. Descriptive statistics: Banks 5th Ptcl.012 0.487 0.300 0.089 -2.000 0.196 0.000 0.882 20.722 1.026 0.8 2.554 ΔINST 3.145 1.721 0.751 0.458 0.096 0.662 1.002 -5.000 95th Pctl.877 RETCORR 0.130 0.750 0.323 0.000 0.365 -0.041 0.414 700.275 0.950 0.000 0.109 0.024 0.871 1.610 0.468 0.012 LOANCI 0.045 0.004 0.647 HIGH_CB 0.046 5.5 -7.790 4.008 0.064 -3.225 746.388 -1.662 Crash predictor BETACT 0.069 -2. 0.302 0.914 2.375 0.014 0.718 Median 0.820 0.553 0.122 1.1 -1.000 0.695 MVE 562609.883 5.851 671.032 FE 0.044 -3.000 0.078 0.785 8.026 0.377 0.036 0.013 0.927 4.089 0.804 NIND 699.210 0.313 0.829 0. 0.000 13.000 0.024 0. 0.109 0.969 0.000 0.733 0.111 0.520 0.113 0.131 -1.071 0.810 1.108 -2.769 0.941 13.856 11.002 50.230 0.594 1846717.109 LOANRE 0.710 52500.001 0.750 0. 0.000 4.112 0.564 1.068 -2.Panel E.111 789.023 0.200 -1.460 0.

808 0.130 REV 0.009 0.803 0.010 34.092 1.7 FUNDCORR -2.031 -3.0 2.239 2.037 0.213 9.592 6.799 0.000 n 1186 1186 1186 1183 1186 1143 565 1185 1137 934 852 1186 1120 1186 1186 1186 1186 1084 806 806 677 801 1186 Firm characteristics DIVERS 0.917 0.131 0.724 ΔINST 9.478 0.483 0.146 1.553 0.253 0.313 66.417 0.8 -6.061 5.991 0.748 0.003 0.044 -3.073 FE 0.005 16.069 0.000 0.811 HERF 0.000 25th Pctl.227 0. 0.314 0.908 0. Descriptive statistics: Matches Mean Opacity proxies BETARSQ SYNCHB BETAINDRSQ SYNCHBI DIFFRSQBETA DIFFB 0.001 -5.Panel F.000 75th Pctl.247 -1.422 IBES analyst coverage NEST 5.700 NIND 291.001 1.333 26.018 0.234 253.784 5th Ptcl.071 STDROA 0.077 0.015 -3.710 4.872 25.567 5.379 196.216 3.653 1.5 -2.157 -1.0 0. 0.123 0.092 -2.000 0.002 0.000 0.860 51744.214 -1.7 -1.007 -4.760 2042705.000 0.0 -3.021 0.620 0.808 0.627 0.923 11.038 0.808 14.018 3.114 0.594 0.000 0.527 0.034 1.399 3.016 Crash predictor BETACT 1.855 0.320 0.182 -1. 0.500 0.212 5.122 1.077 -2.815 122111.140 -1.176 LEVERAGE 0. Dev.477 0.685 4.768 20.210 20814.356 0.212 SHRTURN 21.035 -3.447 0.815 0.182 34.229 0.091 0.794 15.108 1.029 1.006 1.349 -0.021 60.625 0.476 MVE 521295.067 12.113 -2.000 Median 0.250 0.195 806.330 0.108 444.180 12.044 0.554 34 .332 0.840 1.000 -1.333 0.000 0.077 -2. 0.028 INST 38.073 0.925 RETCORR 0.625 0.679 0.746 81.069 -3.610 0.360 33.211 Std.722 8.039 1521851.261 1.101 26.553 0.431 CSD 0.471 -0.398 0.000 95th Pctl.080 0.630 395714.277 1. 0.273 0.302 0.099 84.656 TRDSZE 11.

799 0.594 0.59 0.009 -0.069 -3.157 -0.315 0.892 0.064 -3.500 -0.12 -2.7 21.108 -2.4 21.045 median match 122111.8 0.115 -0.000 -0.016 zstatistic 0.Table 2 Univariate tests: Comparing means and medians of matching parameters and opacity proxies for banks and matches We calculate t-statistics for the null hypothesis of no difference in means between banks and matches.077 0.56 -1. two-tails) using bold.250 0.46 0.14 -2.56 -2.207 -0.5 18.815 0.12 -3.211 difference 41313.143 -0.214 match 521295. mean n Market Equity Share Price BETARSQ SYNCHB BETAINDRSQ SYNCHBI DIFFRSQBETA DIFFB 1186 1186 1186 1186 1186 1186 1186 1157 bank 562609.6 0.9 19.029 difference -1812.140 -1.010 -0.005 -0.182 -1.71 -1. We calculate z-statistics using the Mann-Whitney-Wilcoxon test for the null hypothesis of no difference in medians between banks and matches.131 -1.55 -0.64 -3.44 -1. We indicate significant results (five percent alpha.07 0.107 0.003 tstatistic 0.068 -2.478 0.620 0.05 35 .913 0.044 -3.113 -2.044 -3.751 0.750 0.75 -2.005 -0.43 -0.077 -2.05 bank 120298.172 -1.010 -0.

362 (-1.52) 0.11) -0.175 216 0. (1) BANK (2) 0.40) 0.445 (1.252 (-1.13) -0.232 (2.605 (-2.07) (10) 0.214 216 0.210 (1.389 (2.244 (9.222 (8.Table 3 Regressions of SYNCHB on BANK indicator.325 (9.86) 0.34) Yes Yes 0.173 216 0.321 (8.008 (-0.215 (2.220 216 0.25) (3) (4) 0.016 (0.26) Yes Yes 0.56) Yes No 0.49) 0.170 216 0.324 (-1.033 (0. two-tails) using bold.10) -0.234 (7.95) Yes No 0. We indicate significant results (five percent alpha.79) 0. We calculate t-statistics (in parentheses) using the standard errors derived from the empirical distribution of the ten annual coefficient estimates.330 (-1.41) Yes Yes 0.14) 0.395 (2.16) 0.33) 0.174 (1.244 (7. and log(SHRTURN).186 (1. and log(ΔINST).96) -0. STDROA.18) Yes Yes 0.199 (1.329 (8. FK&N Controls consist of log(TRDSZE).020 (0.167 216 0.275 (-1. log(CSD). This table presents average coefficients from ten annual estimations each model following Fama and MacBeth (1973).608 (-2.89) -0. NASDAQ indicator.20) -0.691 (2.18) -0.17) 0.214 216 0.633 (-2.86) Yes No 0.283 (-1.323 (9.18) (8) (9) 0. P&R Controls consist of DIVERS.290 (-1.48) Yes No 0.48) -0. and controls.11) -0.12) (5) 0. log(REV).31) -0.328 (9.57) Yes No 0. log(HERF).97) -0. R2 Average n 36 . log(FE).232 (9.03) Yes Yes 0.165 216 0.224 216 NASDAQ NASDAQBANK log(MVE) LEVERAGE P&R Controls FK&N Controls Average Adj.221 216 (6) (7) 0.

and log(SHRTURN).066 (-0.007 (-0.302 (11.232 216 0.278 (-1. log(ΔINST).640 (3.009 (0.55) Yes No 0.56) -0.18) -0.29) 0.08) -0.40) Yes No 0. We calculate t-statistics (in parentheses) using the standard errors derived from the empirical distribution of the ten annual coefficient estimates.362 (-2. We indicate significant results (five percent alpha.39) 0. FK&N Controls consist of log(TRDSZE). STDROA.005 (0.171 (1.329 (3.Table 4 Regressions of SYNCHBI on BANK indicator.299 (10.52) Yes No 0.050 (-0.030 (0.54) (8) (9) 0.259 (-1. log(HERF). (1) BANK (2) 0.069 (-0. NASDAQ indicator.17) Yes Yes 0.229 216 0.463 (2.299 (10.016 (-0.433 (-2.224 216 0. log(REV).139 (0.218 (8.03) (6) (7) 0.28) -0.308 (11.16) -0. R2 Average n 37 .066 (-0.28) -0.69) -0. This table presents average coefficients from ten annual estimations each model following Fama and MacBeth (1973).217 (6.260 (1.46) -0. log(CSD).293 216 0.252 (-1.52) Yes Yes 0.248 (1.90) (5) 0. log(FE).54) 0.71) NASDAQBANK log(MVE) 0.64) NASDAQ -0.03) Yes Yes 0.154 (1.76) -0.48) Yes No 0.91) 0.211 (8.220 216 0.05) Yes Yes 0.30) Yes No 0. FUNDCORR.224 216 0.335 (3. two-tails) using bold.289 216 0.97) 0.282 216 0.55) Yes Yes 0.203 (8. and log(NIND).306 (11.295 216 LEVERAGE P&R Controls FK&N Controls Average Adj.286 216 0.223 (7.93) 0.62) (10) 0. P&R Controls consist of DIVERS.129 (0. and controls.13) -0.13) 0.85) (3) (4) 0.

R2 Average n 38 . and log(SHRTURN).93) NASDAQ -0. log(CSD). and controls.209 (6.07) -0.506 (4.17) (8) (9) 0.119 (0.053 (0.230 (-1. (1) BANK (2) -0. FK&N Controls consist of log(TRDSZE). This table presents average coefficients from ten annual estimations each model following Fama and MacBeth (1973).208 (7.125 (3.133 209 LEVERAGE P&R Controls FK&N Controls Average Adj.002 (-0.078 (-0. log(FE).30) Yes Yes 0.79) 0.63) Yes No 0.98) 0.98) 0.478 (1.83) 0.353 (1.10) (10) 0.498 (4.18) 0. log(ΔINST).44) -0.132 209 0. and log(NIND).49) Yes No 0.76) 0.209 (6.261 (-1.083 (-0.75) 0.208 (7. t-statistics (in parentheses) are based on the standard errors derived from the empirical distribution of the ten annual coefficient estimates.44) (3) (4) -0. FUNDCORR.184 (0.92) 0.48) 0.109 209 0.60) Yes Yes 0.68) Yes Yes 0.105 209 0.67) -0.333 (3.490 (1.114 (1.49) Yes No 0. NASDAQ indicator. STDROA.02) -0.208 (7.124 209 0.106 209 0.120 (3.97) (6) (7) -0.039 (-0.14) -0.45) (5) 0. We indicate significant results (five percent alpha.375 (1. log(HERF).136 209 0.69) Yes Yes 0.354 (1.130 209 0.66) 0.124 (3. log(REV). two-tails) using bold.45) Yes No 0.080 (-0.471 (1.100 209 0.127 (2.105 209 0.008 (-0.021 (-0.31) Yes Yes 0. P&R Controls consist of DIVERS.66) NASDAQBANK log(MVE) 0.328 (3.25) 0.Table 5 Regressions of DIFFB on BANK indicator.62) 0.128 (2.33) Yes No 0.072 (-0.

46) 0.09) 0.33) -0.067 (0.215 (0. FK&N Controls consist of log(TRDSZE).550 (-3.099 (0.91) -0. NASDAQ indicator.Table 6 Regressions of BETACT on BANK indicator. log(ΔINST).024 (0. two-tails) using bold.522 (-2.053 (2.57) 0. log(CSD).062 (3.74) 0. We indicate significant results (five percent alpha. log(REV).40) Yes No 0.65) Yes Yes 0.116 216 0.48) -0.143 216 (6) (7) -0.191 (0. P&R Controls consist of DIVERS.120 (-0.042 (2.29) 0.045 (2.002 (-0.50) 0.076 (3.056 (3.082 (0.04) 0.37) 0.43) Yes No 0.134 216 0.90) 0.113 (0.020 (-0. and log(SHRTURN).68) 0.066 (3.144 (0.620 (-3.047 (2.02) (8) (9) -0. log(FE).57) 0.36) Yes No 0.574 (-3.82) Yes Yes 0.08) -0.53) Yes Yes 0. and log(NIND).63) -0.059 (-3.581 (-2.71) 0.29) (3) (4) -0.071 (-0.78) -0.25) Yes No 0.48) (10) -0. R2 Average n 39 .645 (-3.584 (-2.046 (0.054 (3.137 216 0.145 216 NASDAQ NASDAQBANK log(MVE) LEVERAGE P&R Controls FK&N Controls Average Adj. and controls.115 216 0.088 (0. STDROA.77) Yes Yes 0.608 (-2. FUNDCORR.135 216 0. (1) BANK (2) -0. This table presents average coefficients from ten annual estimations each model following Fama and MacBeth (1973).072 (3.118 216 0. log(HERF).42) Yes No 0.117 216 0.25) -0.142 216 0.650 (-3.30) (5) -0. t-statistics (in parentheses) are based on the standard errors derived from the empirical distribution of the ten annual coefficient estimates.46) -0.76) Yes Yes 0.

04 (1.99) NASDAQ log(MVE) 0. log(HERF). R2 Average n 40 .13) No No 0.250 119 0.83) Yes No 0. We calculate t-statistics (in parentheses) using the standard errors derived from the empirical distribution of the ten annual coefficient estimates. log(REV). and log(SHRTURN).23 (4.236 110 0.22 (5. This table presents average coefficients from ten annual estimations each model following Fama and MacBeth (1973).44 (-1. P&R Controls consist of DIVERS.67) Yes Yes 0.51) Yes Yes 0. log(CSD). log(FE).08 (2.335 110 P&R Controls FK&N Controls Average Adj.34 (11.54) (4) 0.79) (3) 0. FUNDCORR. FK&N Controls consist of log(TRDSZE). and log(NIND).09) -0. two-tails) using bold.34 (9.03 (0. log(ΔINST).02 (0.289 110 0. We indicate significant results (five percent alpha.45) (2) 0. (1) BETACT 0.Table 7 Regressions of SYNCHBI and on market model crash predictor and controls for banks only. STDROA.

018 (0.068 (0.14) -0.819 (-1.61) 0.318 98 Yes Yes 0.20) -0.095 (-0.064 (-0.151 (-0.043 (-1.67) 0.234 (-2.285 (5.17) -10.725 (-1.065 (-0.058 (-0.480 (-1.140 (-1.11) -0.129 (-2.490 (-2.38) -0.06) -0.371 (-1.847 (0.46) -4.542 (-1.237 (0. and log(SHRTURN). FUNDCORR.38) No Yes 0.024 (-0.35) -3.02) 0.189 (-2.544 (0.256 (-0.355 (-0.120 (-0.64) -4.01) -3.327 95 No No 0.021 (-0.905 (-1.271 (5.80) -2.136 (-0.655 (0. This table presents average coefficients from ten annual estimations each model following Fama and MacBeth (1973). We calculate t-statistics (in parentheses) using the standard errors derived from the empirical distribution of the ten annual coefficient estimates.062 (-0.96) -6. log(ΔINST).27) No No 0. two-tails) using bold.386 (-1.58) -0.16) 0.198 (-1.369 98 (8) 0.397 (2.247 (2. log(HERF).93) 0.454 (0.30) (2) -0.095 (-0.68) 1.71) -0.82) (4) -0. R2 Average n 41 .365 (9.085 (-1.046 (0.64) (7) 0.88) 0.242 106 No Yes 0. log(REV). We indicate significant results (ten percent alpha.259 (5.01) -0.51) -1.71) 0.71) -0.63) -3.01) -2.68) 0.28) -0.87) (3) -0. log(CSD). (1) PCTLOANS 0.24) -8.374 95 TRADE OPAQUE HIGH_CB NASDAQ log(MVE) LOANAG LOANCI LOANDEP LOANIND LOANRE FRD Controls Other Controls Average Adj.134 (-2.82) -0.145 (-0.374 98 (5) -0.017 (0.59) -0.119 (-1.66) 0.99) 0.281 106 0. log(FE).256 (2.985 (-1.26) -0.05) -3.607 (-2.584 (1.89) -7.58) -0.410 (16.34) -7. STDROA.87) -2.Table 8 Regressions of SYNCHBI on asset characteristics and controls.664 (-1.98) 0.412 (1.049 (0.325 98 No Yes 0.89) -2. P&R Controls consist of DIVERS.72) 0.51) 0.10) 0.26) (6) 0.56) Yes Yes 0.88) -0.86) -6.013 (-0. and log(NIND).486 (0.76) 0.44) 0.256 (2.020 (-0.174 (-1.468 (1.617 (-1.159 (-2.17) -0.86) 0.22) -10.40) -0.485 (0. FK&N Controls consist of log(TRDSZE).30) No Yes 0.070 (1.25) 0.87) 0.