Bank financial reporting opacity and regulatory intervention

Similarly, Fischer and Verrecchia (2000) show that if investors are unable to fully unwind reporting bias, they will reduce their price response to reported fundamentals (e.g., earnings). Non-regulator bank stakeholders are likely to lack the training and private information necessary to unwind financial reporting opacity, especially relative to bank regulators (Stigler 1971; White 2012).

Prior work takes a more indirect approach, interpreting opaque reporting choices by banks (Huizinga and Laeven 2012) or changes in accounting standards (Blacconiere 1991; Blacconiere et al. 1991; Skinner 2008) as evidence that regulators may desire financial reporting opacity for forbearance purposes.

I discuss institutional details and prior research on U.S. bank supervision and intervention in Appendix A.

Forbearance and “too-big-to-fail” are distinct concepts. The latter refers to government bailouts and other support for systemically important institutions whose collapse could directly result in major problems for connected institutions. In contrast, regulators can practice forbearance on banks of any size, and the motives for such forbearance need not depend on the implications of the bank’s failure for the rest of the system. For example, it is widely accepted that regulators practiced forbearance on troubled thrifts during the U.S. savings and loan crisis in the 1980s, most of which would not have been considered “too big to fail” (Santomero and Hoffman 1998).

When the FDIC closes a bank, it sends a team to take over the bank’s operations. These teams can be large even for small banks. For example, the FDIC sent 80 agents to close a bank with 100 employees (Joffe-Walt 2009).

While forbearance can have benefits, it also has potential costs. If forbearance allows managers to increase risk-taking (“gambling for resurrection”), it could increase the ultimate cost of resolving the bank (Cole and White 2017). Furthermore, given the debtlike nature of their payoffs, uninsured creditors may prefer timely intervention in troubled banks, and they may run if they observe that the bank’s health is deteriorating (Diamond and Dybvig 1983). Ideally, regulators will balance the benefits of forbearance against these potential costs.

Regulators could affect provisioning behavior via less formal channels, such as conversations during on-site inspections. That said, enforcement actions can be used to address provisioning behavior.

While this discussion focuses on funding providers, other parties can inhibit forbearance. For example, politicians and the media can criticize regulators for a perceived lack of oversight. Furthermore, monitoring by one party can spark actions by another (e.g., political criticism of bank regulators could lead to depositor withdrawals).

Consistent with this idea, Rochet (2004) and Decamps et al. (2004) analytically show that if bondholders act as if a bank is in trouble (e.g., withdraw funds), regulators can be forced to intervene because forbearance becomes too costly for the regulator and/or government. While these models refer to bondholders, these bondholders are essentially providers of uninsured funding, similar to uninsured depositors.

Relatedly, Parlatore (2017) finds that if bank opacity impairs depositors’ ability to understand a bank’s deteriorating financial condition, they may be less likely to withdraw their funds, and Gao and Jiang (2018) show that reporting discretion can reduce the incidence of bank runs.

There are two reasons why non-regulator bank stakeholders may not assume the worst when observing financial reporting opacity. First, as Fischer and Verrecchia (2000) show, if stakeholders are not fully aware of the manager’s reporting strategy, then reporting bias adds noise to the report, leading to an attenuated response. Second, some studies suggest that bank stakeholders value bank opacity because it allows deposits to act as a money-like security whose value does not fluctuate with the underlying assets, which in turn facilitates liquidity risk sharing and creates a medium of exchange without fear of adverse selection (Chen et al. 2021).

The alternative—not including controls for key financial indicators such as earnings and capitalization—would likely introduce omitted variable bias.

I focus on the recent 2007–09 financial crisis rather than past crises (e.g., the Great Depression, the savings and loan crisis) for two reasons: (1) the set of call report and state-level macroeconomic variables have greatly expanded over time, enabling me to better mitigate correlated omitted variable concerns; and (2) during other crises, non-closure interventions were not publicly disclosed, precluding their investigation.

Alternatively, one could examine this association in an international setting. This is a less promising path for several reasons. First, many countries have highly concentrated banking sectors, with few banks and even fewer regulatory interventions. Second, it is challenging to compare interventions across countries because they differ substantially. Third, differences in institutions across countries are likely correlated with financial reporting, making it difficult to isolate the association between financial reporting opacity and intervention. Finally, data on non-U.S. banks is generally lacking in quality and number of variables relative to call report data.

This follows the SNL classification scheme. Prior research shows that severe enforcement actions are an important form of regulatory intervention: they are associated with reductions in lending (Danisewicz et al. 2018; Kleymenova and Tomy 2020; Peek and Rosengren 1995; Peek and Rosengren 1996) and asset risk (Delis et al. 2017), increases in capital (Kleymenova and Tomy 2020), and negative consequences for the local economy (Danisewicz et al. 2018).

I investigate these two intervention types together in my main analyses for parsimony and because both represent serious interventions. In Section 5.2, I separately examine different intervention types.

For example, the St. Louis Federal Reserve’s financial crisis timeline starts in February 2007 (https://​www.​stlouisfed.​org/​financial-crisis/​full-timeline). Intervention was less common before the crisis. For example, there were no bank closures in either 2005 or 2006.

I employ a 20-quarter window in order to balance two objectives: (a) having a sufficiently long time-series to estimate the component regressions, and (b) allowing banks to change their delayed loan loss recognition over time. For these regressions, bank-level variables are winsorized at the 1st and 99th percentiles within each quarter.

I include DELR as a standalone variable, whereas some studies (e.g., Chen et al. 2021) interact DELR with bank fundamentals. The key difference is that other studies seek to examine how DELR affects stakeholder responses to bank fundamentals, and their findings suggest that DELR is a bank attribute associated with reduced stakeholder sensitivity to such fundamentals. This suggests that including DELR as a standalone variable will capture the regulator’s preference to forbear on banks for which the potential for stakeholder reactions to jeopardize forbearance is reduced.

The FDIC lists 325 banks that were closed during my sample period. Of these, 49 did not file call reports as of December 31, 2006, because they were savings banks or savings and loans (and therefore not commercial banks). Therefore, my closure sample begins with the remaining 276 commercial banks. Likewise, SNL has data on 1763 bank-level severe enforcement actions during my sample period, representing 1458 unique banks. But 209 did not file call reports as of December 31, 2006, because they were credit unions, savings banks, or savings and loans. An additional four banks filed call reports but received their enforcement action several years after filing their final call report; I exclude these enforcement actions from my analysis due to their unusual nature. Therefore, my severe enforcement action sample begins with the remaining 1245 unique commercial banks.

Continuous variables are winsorized at the 1st and 99th percentiles.

In untabulated analyses, I find that my inferences are unaffected if I instead cluster standard errors by bank, state, or regulatory combination (the OCC for nationally chartered banks, or the combination of the applicable state regulator with either the FDIC or Federal Reserve for state chartered banks).

It is important to note that this approach assumes that a given regulator’s leniency and/or policies equally affect all banks under its supervision, which may not be the case. Therefore, I cannot conclusively rule out this story. In an untabulated test, I find that including the Agarwal et al. (2014) leniency index does not affect my inferences.

The propensity scores are created by using the coefficients from a regression of High DELR on the full set of control variables. The matching is done without replacement and using a caliper of 0.001. My inferences are also robust to a PSM approach that matches on Intervention using all independent variables except DELR.

Additionally, in untabulated analyses I find that DELR exhibits low (< 0.1) correlations with key measures of bank health, such as the non-performing loans ratio, tier 1 capital ratio, or the level or change in the loan loss reserve, highlighting a benefit of the within-bank approach to measuring DELR.

I follow Oster (2019)‘s suggestion of setting R(max) to 1.3 times the R-squared from the fully specified regression. Furthermore, I conservatively assume that no control variables are fully observed. If I instead assume that certain key variables are fully observed (specifically Tier 1 Ratio, ROA, NPL, and Size), the Oster deltas increase to 2.56 and 252 in the panel and cross-sectional samples, respectively.

Including these additional variables removes banks that were closed in year t + 1 from the sample, as these banks would likely not file year-end call reports in year t+1. I only conduct this test using the panel sample; a similar test using the cross-sectional sample would eliminate all banks that did not file call reports in December 2010.

Beatty and Liao (2020) suggest an alternative approach to addressing this issue: using the addition to the nonaccrual asset in place of the change in non-performing loans. However, since additions to nonaccrual assets are not reported by commercial banks until Q3 2006, I cannot employ this approach. Beatty and Liao (2020) conclude that in the absence of data on additions to nonaccrual assets, one should use the change in non-performing loans (see p. 10 of their study), which I employ in my primary DELR measure.

In my primary analyses, I estimate DELR using the provision through the fourth quarter of year t, so that the timing of DELR matches the key metrics it impacts, such as capital ratios and earnings (which are also measured as of the fourth quarter of year t). However, this estimation requires ΔNPL_t + 1, meaning that ΔNPL from the first quarter of year t+1 is used. If the bank’s intervention occurs in the first quarter of year t+1, there is an overlap between the measurement period of DELR and Intervention. I conduct two untabulated tests using the panel sample to examine whether the overlap is responsible for my findings. First, I exclude observations where intervention occurred in the first quarter of year t+1. Second, I re-estimate Eq. 1 measuring all bank-level independent variables as of the third quarter of year t (which eliminates the overlap). In each test, I find a strong negative association between DELR and intervention. Another potential issue is that requiring ΔNPL in the first quarter of year t+1 might result in the observation not being included in the analysis if the intervention resulted in the bank not filing a call report in the first quarter of year t+1. I find no such examples in the cross-sectional sample. In the panel sample, I find 20 observations that filed a call report in the fourth quarter of year t but not in the first quarter of year t+1. However, these observations had missing DELR values in prior quarters; thus, the missing call report was not responsible for their exclusion.

Relatedly, several studies reports suggest that U.S. bank regulators likely practiced forbearance during the crisis (Brown and Dinç 2011; Government Accountability Office 2011; Rapoport 2013).

When the indicator variable is based on or is closely related to a continuous control variable, I remove the associated control variable from the regression in order to facilitate the interpretation of the indicator coefficient.

Chen et al. (2021) find that while opacity impairs depositor monitoring, the effect weakens during the early part of the financial crisis, which they attribute to depositors not attempting to distinguish between banks (i.e., not monitoring) during this period. In an untabulated analysis, I replicate their design in my sample, using High DELR as the measure of financial reporting opacity. I find that High DELR weakens the sensitivity of uninsured deposit flows to bank performance in my sample period, consistent with Chen et al. (2021)‘s main finding of opacity impairing depositor monitoring. Furthermore, this effect is weaker (stronger) in the earlier (later) part of my sample period, again consistent with their findings. My findings suggest that depositor monitoring, and thus the facilitating effect of DELR for forbearance, returned after the early crisis period. As noted above, this was also the period in which regulators were likely facing significant resource constraints and the DIF balance was at its lowest, suggesting that forbearance incentives had sharply increased. These findings are consistent with DELR facilitating forbearance by reducing stakeholder responses to bank fundamentals, and help explain why the DELR-intervention association is concentrated within the latter part of my sample period.

I find similar results when measuring state banking sector weakness using the deposit-weighted percentage of banks with a tier 1 capital ratio in close proximity to the 6% threshold for being considered well-capitalized.

The interaction term coefficients are not statistically different across models; this appears to be driven by the relatively large standard error in the federally chartered subsample. Furthermore, the main effect of DELR is statistically significant (insignificant) within the federally (state-) chartered banks. I interpret this as suggesting that both regulators exploit DELR for forbearance, but that the motivation for such forbearance differs across regulators: the state banking sector is a primary factor for state regulators, whereas non-regional factors (such as resolution costs) motivate forbearance for national regulators.

Put differently, regulators may still prefer to forbear on banks with less non-regulator monitoring, but they should be less dependent on financial reporting opacity to facilitate such forbearance. For example, while regulators may desire to forbear on banks with significant insured deposits, opacity should be less important for facilitating such forbearance since insured depositors, whose claims are not at risk, are less likely to monitor the bank.

In my sample, on average approximately 11% of assets are funded by brokered deposits and non-deposit liabilities. Given that the average equity-to-asset and cash-to-asset ratios are 11 and 5.5%, respectively, runs by these funding providers could cause solvency and liquidity problems for banks.

Consistent with this idea, prior research finds that non-banking firms exhibit less earnings management when there is greater analyst coverage (Yu 2008; Burgstahler and Eames 2003) and institutional ownership (Bushee 1998; Jiambalvo et al. 2002; Roychowdhury 2006).

I classify a bank as Private if I cannot link it or its holding company to a CRSP or CUSIP identifier (which signals the presence of traded equity or debt), either via the call reports or manual inspection. Since there are multiple potential sophisticated monitors (equity analysts, bond raters, hedge funds, the business press, etc.), this analysis instead examines banks that likely lack collective exposure to such monitoring. In untabulated analyses, I find that the DELR-intervention association is no longer significant in the presence of two specific sophisticated monitors (equity analysts and institutional owners), consistent with the tabulated findings.

I only examine bank-level enforcement actions in this analysis; personnel-level actions are excluded.

For example, a severe enforcement action may require that the regulator subsequently conduct an on-site inspection or additional off-site monitoring, and closing a bank requires the FDIC to send a team of employees.

A reading of a random selection of non-severe enforcement actions suggests that the issues addressed arose either during onsite inspections or via offsite monitoring, consistent with regulatory monitoring leading to their issuance.

I find similar results when using lagged time-varying bank covariates in the panel sample (results untabulated).

It is important to note that, despite this apparent benefit to DELR, banks and/or regulators may not uniformly seek to increase it for several reasons. First, prior research suggests that banking sectors with greater DELR have increased risk-taking (Bushman and Williams 2012); thus, there may be a cost for allowing greater DELR on average. Second, regulators may apply greater scrutiny to high DELR banks during non-crisis periods, while exploiting DELR during a crisis to avoid intervention. This is consistent with the intuition, from Morrison and White (2013), that regulators may prefer transparency ex ante, but opt for opacity during a crisis as it allows them to forbear secretly without creditors knowing. Finally, more sophisticated stakeholders, such as institutional investors, may be reluctant to provide capital or liquidity to high DELR banks. This is difficult to examine empirically in my sample but is consistent with prior research (Bushman and Williams 2015).

I follow prior research (Curry et al. 1999; Lucca et al. 2014) in using the term “enforcement actions” to collectively refer to the various formal actions that bank regulators can take against banks, including written agreements, cease-and-desist orders, prompt corrective action directives, and less severe interventions. Regulators can also issue enforcement actions against individuals, but these actions are not examined in this study.