* Reprinted with minor edits from Prompt Corrective Action in Banking: 10 Years Later, edited by George G. Kaufman, pp. 143 -202, © 2002, with permission of Elsevier.
** Lynn Shibut is the Section Chief for Consulting Services in the Division of Insurance and Research, FDIC. Tim Critchfield is a Senior Financial Analyst in the Division of Insurance and Research, FDIC. Sarah Bohn is in graduate school at the University of Maryland. The authors wish to thank Tyler Davis, Heather Gratton, Toni Holloman-Spinner, Terry Kissinger, Jennifer Merrill, and Katherine Samolyk for assistance, and Charles Collier, Robert Ferrer, Grovetta Gardineer, Alton Gilbert, Herb Held, James Marino, Rae-Ann Miller, Larry Mote, Dan Nuxoll, John O'Keefe, Bob Storch, Bob Walsh, and Jim Wigand for valuable comments.
1 If troubled banks overstate asset values, their capital is artificially increased: this may allow them to avoid or delay adverse consequences. The U. S. General Accounting Office (1992) documented the problem. There is also ample evidence that examinations of troubled banks often result in increased reserve levels (which cause capital to fall). See for example Dahl et al. (1998) and Gunther and Moore (2000).
2 Throughout the article, loss rates are defined as the FDIC loss divided by total assets of the failed bank.
3 Throughout this article, we use the term banks to mean all FDIC-insured depository institutions.
4 Since the 2 percent threshold allows the bank only 90 days to improve its condition before closure, the bank's viability at this point is almost entirely determined by events that occurred before it reached the threshold. If banks have not begun seeking capital well before reaching the 2 percent threshold, the 90-day time limit is tantamount to closure. However, regulators are allowed to delay closure for up to 270 days (inclusive of the first 90-day period) if the primary supervisor determines, and the FDIC concurs, that the delay would better achieve the purpose of FDICIA. The act also prescribes extremely narrow and explicit conditions for a delay beyond 270 days.
5 Peek and Rosengren (1996, p. 49). They tested only for the 4 percent leverage-ratio threshold. There are two other capital thresholds that define undercapitalized banks, and some of the banks with leverage ratios above 4 percent may have breached one of the other thresholds.
6 Jones and King found that regulators have already downgraded the institutions to a CAMEL 4/5 rating before the institutions meet the RBC threshold for an undercapitalized institution.
7 PCA may have had significant indirect effects on the behavior of both banks and regulators that were not addressed in these studies. For example, banks might protect themselves from the PCA restrictions by holding more capital than they would have held absent PCA.
8 Depending on the method used, from 214 to 314 banks would have been adequately capitalized under the RBC standards in place but would have been undercapitalized under the revised RBC standards. Very few banks that were undercapitalized under the RBC standard in place at that time would have become adequately capitalized under the revised method. The authors normalized the RBC levels across banks so that they captured only differences in the distribution of RBC across banks, instead of a combination of an increased standard and changes in the distribution across banks.
9 They used three definitions for troubled institutions: (1) all institutions with a CAMEL rating of 4 or 5, (2) all institutions with tangible capital below 2 percent, and (3) all institutions that met either of the first two standards. Their estimates of prediction error assumed that all such institutions should be classified as undercapitalized and that all other institutions should not.
10 The literature on public finance supports the assumption that government intervention should strive to maximize the net benefit to society. Therefore, when setting PCA thresholds, banking regulators should not focus solely on minimizing losses to the insurance funds.
11 More general discussions of the benefits of PCA can be found in Carnell (1997a) and Mishkin (1997).
12 Regulators have the authority to close banks when they are operating in an unsafe or unsound manner, regardless of capital levels. Presumably any bank with no market value would meet this criterion.
13 Several authors have noted the difficulties in measuring capital at banks. See, for example, GAO (1992).
14 Regulators are not allowed to set standards that are less stringent than GAAP. Under Section 37 of the FDI Act, regulators have the authority to set standards that are more stringent than GAAP. Note, however, that taking such an action introduces new costs into the equation (including a new regulatory burden for banks and reduced comparability of bank financial reports and information from other related industries) and thus should not be undertaken lightly-particularly if alternative means of meeting the objective are available.
15 Benston and Kaufman (1997, p. 146) argue that the closure of an institution that is solvent on a market-value basis "would not be a 'taking' by the government as any remaining capital would be returned to the shareholders." From a legal standpoint, they are probably correct. However, there are real costs, including the overhead expenses associated with administering a receivership, prepaid expenses that are rendered worthless at failure, transaction costs associated with transferring information from the seller to a buyer, and any lost asset value associated with the stigma of failure. Owners could potentially be subject to a Directors and Officers claim as well. In addition, the FDIC must pay the cost of administering insurance and preparing the resolution. From an economic standpoint, these costs should be considered when an increase in the PCA threshold is being contemplated.
16 Some of the bank's creditors could suffer losses if contracts are abrogated or if the receivership does not pay creditors as much as they would have received had the bank stayed open. In addition, if the bank was not located in a competitive market, the bank's customers (and others in the community) may not be able to obtain credit or make deposits at equivalent terms. These effects could, in turn, harm the local economy more broadly. This appears to be most likely for small, rural banks in situations where the FDIC closes the bank with no acquirer (a payout) or where the acquirer has less interest in the local community than the failed bank. See Gilbert and Kochin (1989) for an analysis of these effects.
17 If crossing the 2 percent threshold did not cause any material change in strategy, the cost would probably be negligible. If there was a change in strategy that improved the allocation of capital, there could be an economic benefit; if the opposite occurred, there could be an economic loss. But these results relate to the PCA threshold to the extent that the threshold (rather than the weak condition of the bank) caused the change. In addition, for the most part, the owners would retain the freedom to control their assets.
18 Like owners that suffer from closure, owners of a bank that was purchased would lose the freedom to control their assets, and they would have to pay the transaction costs associated with transferring information to a buyer. Because of the 90-day deadline, they might also be required to sell the franchise for a below-market price.
19 Stronger action by the bank could also yield improvements to the bank owner's earnings.
20 There have been some receiverships with a surplus. However, a surplus can occur for reasons other than an avoidable closure. For example, it can occur because of positive developments in the markets after failure or because the FDIC was able to reverse or mitigate problems experienced at the bank (such as excessive employee compensation or self-dealing on the part of management). Shibut and Critchfield (2000) found no evidence of wrongful closure in their review of low-cost failures.
21 The tradeoff may also be influenced by the nature of the problems experienced by the banking industry. For example, a PCA threshold that is optimum for banks undergoing a regional recession may not work as well for stresses associated with high interest rates or subprime lending.
22 Federal Deposit Insurance Corporation (1997, chapter 5). While it appears that the sudden application of PCA rules as the crisis mounted would have had terrible consequences, it is difficult to gauge what the results might have been if PCA triggers and the current accounting rules had been in place long enough to alter the banks' behavior well in advance of the crisis.
23 He states that the systemic-risk exception in FDICIA allows regulators enough discretion to avoid excessive costs associated with the PCA thresholds. Thus concerns about extreme situations need not be considered unless they do not involve systemic risk. If, however, an unexpected event were to make the PCA threshold inappropriate for a large number of banks at the same time, the systemic-risk exception would become untenable.
24 The study concentrated on the pace of recovery and the implications for selecting an appropriate time horizon for Value-at-Risk (VaR) models.
25 Recovery was defined as reaching a leverage ratio of 5 percent.
26 The Meriden Trust and Safe Deposit Bank, Meriden, Connecticut, was eliminated because it had no insured deposits and was part of a cross-guaranty case. The First National Bank of Keystone, Keystone, West Virginia, and BestBank in Boulder, Colorado, were eliminated because the fraud involved in these failures was large enough to seriously distort their financial reports.
27 We use the term Call Report as shorthand for both the Call Report and the Thrift Financial Report.
28 All but 7 of the 32 institutions that fell below the threshold failed within one year. Three institutions fell below the PCA threshold more than one year but fewer than three years before failure, and for them we decided to use the PCA date for our analysis. Four institutions fell below the PCA threshold more than three years before failure, but because they fell below the PCA threshold between 1990 and 1993 (before PCA was fully implemented), we decided to use their failure date.
29 (1) New Meridian Trust was excluded because it was a bridge bank created by the FDIC to operate a failed institution until a buyer could be found. (2) Valley First Community Bank, Scottsdale, Arizona, reported a capital level below 2 percent on its very first regulatory report after establishing its operations. (3) Chicago-Tokyo Bank, Chicago, Illinois, was held by a foreign bank holding company and it reported virtually zero assets on its last report. (4) Suntrust Bankcard, National Association, Orlando, Florida, had one quarter during a downsizing that produced a low capital ratio because the denominator was based on average assets that did not reflect the end-of-period level of assets. (5) Continental Savings Bank, Seattle, Washington, filed a report with a missing value for capital in one quarter.
30 Of the 21, 14 were absorbed by mergers that eliminated their charters and 7 were purchased by a holding company.
31 There were 6 institutions out of the 84 institutions studied that failed to file a financial report in the quarter after their last full examination before falling below the PCA capital limit.
32 To create a problem bank index we summarized the branch-level deposits of all FDIC-insured institutions that were rated CAMELS 4 or 5 for every market which was located in a metropolitan statistical area (MSA) or county for any county that was not in an MSA. We divided this sum by the total deposits for the MSA or county and then multiplied these percentages by the percent of each institution's deposits located in each market. Then we summed the totals for each market that an institution had a presence in for the mid-year before the institution fell below the PCA capital limit of 2 percent or before failure.
33 This was calculated as of the middle of each year before the institution falls below the PCA capital limit using data from the annual Summary of Deposits. We calculated the percentage of an institution's deposit in each market that it operates and then squared these percentages. Then we summed the squared percentages for all markets in which an institution operates so that the results ranged from 0 to 1. This measure was based on the work of Katherine Samolyk.
34 All parameter estimates were statistically significant at the 1 percent level. The R2 was .33. The sample included commercial banks during 1984-1989.
35 Under GAAP, the allowance for loan-and-lease losses (ALLL) should cover probable estimated credit losses in the loan portfolio, not a specified number of years of expected charge-offs.
36 Of the 45 receiverships, 32 (71 percent) had terminated by year-end 2000. As of year-end 2000, the remaining receiverships held $79 million in assets for sale, which was 6.8 percent of the assets held by these receiverships at failure.
37 In fact, comparisons of simple averages for analysis were unsatisfactory because of outliers, and comparisons of weighted averages were unsatisfactory because they tended to represent only the largest banks in the sample. Thus, we relied most heavily on medians, distributions, and the Wilcoxon test for analysis.
38 Throughout the section on results, we use the terms significant and insignificant to mean statistically significant and statistically insignificant.
39 Examiners use four classification levels to indicate the level of impairment. The distribution of classifications across these levels has a strong influence on the Jones and King estimate.
40 Our results are consistent with those of Barakova and Carey (2001). They briefly compared the components of changes in equity of failed banks with the components of banks that recovered. The authors found no marked differences in performance between these two groups, although failed banks raised somewhat less equity and experienced higher cumulative losses than the banks that recovered.
41 All correlations and tests of significance are based on the Kendall's tau test and compare the FDIC loss rate with the item being discussed.
42 Eight low-cost failures and five high-cost failures did not breach the PCA threshold before failure. For these institutions, we treated the final Call Report date as the PCA failure date.
43 The correlation was 0.16 and insignificant. This could relate to idiosyncratic differences between banks (particularly the amount of loss attributable to fraud or mismanagement) and the small sample size.
44 McDill (2002) confirmed that a relationship exists. Using data from 1980 through 2000, she regressed the FDIC loss rates of failed banks against various business-cycle items and bank-specific characteristics. She found that loss rates were higher for institutions located in states with falling personal-income levels-particularly if the reduction occurred after a "boom." She also found that loss rates increased for institutions located in states that had concentrations of problem banks. As an example of the effects, she used the results to estimate the loss rate for a typical Texas failure in 1988 and 1999. The loss rate in 1988 was 6 percentage points higher than the loss rate in 1999.
45 Capital is basically a measure of current condition and largely ignores the likelihood of future deterioration. Mailath and Mester (1993) developed a theoretical model on optimum bank closure policy and found that regulators should consider the future prospects and probable actions of a failing bank in order to minimize losses to the insurance funds. A more forward-looking measure of capital might mitigate losses during periods of stress, but it would introduce more subjectivity into the equation as well.
46 If regulators imposed either extremely tight restrictions on risk (that is, narrow banking) or very high capital requirements, it might be feasible to largely eliminate insurance fund losses during periods of stress. However, there would probably be substantial costs in the form of tighter credit availability and higher credit costs for a large number of borrowers.
47 The market for deposits was strong throughout the period. There could be a different result during periods of lower demand.
48 We investigated the amount of foreclosure activity during receivership but found that the median change in OREO (as a percentage of total assets) between closing weekend and the final receivership balance increased about the same amount (2 percentage points) for both groups.
49 Appendix 3 provides more detail about the asset composition at failure and in receivership.
50 These loss rates were calculated as the difference between the gross book value recorded by the receivership and the sales proceeds. If the costs associated with selling the assets had been included, the difference between the two groups of institutions would have been larger.
51 There are several reasons why reserves may be sufficient under GAAP for an operating institution but still be smaller than the asset losses experienced at liquidation. Some of the reasons are changes in market conditions, changes in interest rates, and differences in the value anticipated for an asset held in a long-term portfolio versus the value if the asset is sold relatively quickly.
52 Receivership income and expenses were calculated on a cash basis, and they excluded gains and losses from asset sales, asset write-offs, holding costs, income from the bid premium and other equity adjustments. Holding costs were defined as the FDIC's lost income from funding the deposits up front and waiting for recovery from the receivership. Appendix 2 provides more information on the composition of these items.
53 Because the level of activity and the duration (in time) of receiverships vary substantially, all income and expense items are reported as the total amount over the life of the receivership divided by total assets at failure.
54 One cannot merely subtract receivership income from receivership expenses and assume that the difference (as a percentage of total assets at failure) is the amount of loss in percentage points attributable to receivership operations. Many receiverships start with substantial amounts of positive equity (partly because reserves are reversed at failure). In addition, some losses and gains do not flow through the receivership income statement, and some receivership losses are borne by creditors other than the FDIC.
55 An acquirer is defined as the institution (or institutions) that purchased the deposits.
56 The simple average for the percentage of assets sold to the acquirer was 30.01 percent for the high-cost institutions and 65.73 percent for the low-cost institutions.
57 National vacancy rates for office space were 16.8 percent at the beginning of our sample period. They dropped to 13.8 percent at year-end 1995, and 9.9 percent at year-end 1997. The changes in sales volume were larger. According to Torto-Wheaton Research, there were only 18 sales made of Class A office buildings nationwide in 1996, and 35 in 1997-but 238 were made in 1998. The results were similar for other types of office buildings.
58 The results concerning changes in the FDIC loss estimates over time were also consistent with the concept that losses may increase when markets are weak. For high-cost failures, the FDIC loss estimates increased substantially between the original estimate and the latest estimate; for low-cost failures, they increased only a little bit. This supports the theory that markets deteriorated during the life of the receivership, or it might indicate that asset valuation at closure was more difficult for the high-cost institutions that carried riskier assets.
59 See Shibut and Critchfield (2000) for details.
60 Regulators often use cease-and-desist orders to force troubled banks to use appropriate methods for setting reserves. Because we did not collect data on formal and informal actions, we do not know whether there was a relationship between such orders and coverage ratios. In addition, the heterogeneity of banks, coupled with the amount of subjective judgment inherent in reserve levels, may make the task of limiting bank discretion through industry-wide rules a daunting one.
61 This idea is not original with us. See also U. S. Department of the Treasury (1991), Berger et al. (1991), and Jones and King (1995).
62 Based on our research, the formula might use a combination of classified assets, asset types, changes in and levels of noncurrent loans and OREO, and data about local-market conditions. Given the results of the Jones and King (1995) estimate, the appropriate formula for banks approaching insolvency may differ from the optimum formula for less-troubled banks. Thus it would be important to match the regulatory use of the formula to the methods and data used in its development.
63 A formulaic approach may not meet GAAP, in which case the regulatory burden associated with a change may increase. The regulatory burden could be mitigated if the formula were used solely for setting regulatory capital standards.
64 If such a policy were adopted, it might be beneficial for a separate team of FDIC staff (including examiners) to be involved in such decisions. This might reduce the potential for inefficiencies stemming from the principal-agent problem, which is eloquently described by Mishkin (1997).
65 The steps of the calculations are (1) estimate the net present value of the assets; (2) estimate the volume of creditors in each creditor class; (3) estimate payments to each creditor class, given the discounted asset value estimates; and (4) estimate the FDIC loss as the difference between the FDIC's claim and its anticipated recoveries.
66 For terminated receiverships, all cash flows are undiscounted. For ongoing receiverships, actual recoveries to date are undiscounted, and expected future recoveries are discounted to the date of the estimate. For example, assume that a failure occurred in 1995. The FDIC loss as of year-end 2000 would be calculated as the 1995 outlay minus all dividends paid through year-end 2000 minus estimated future dividends (which would be based on the anticipated future recoveries from the receivership, discounted to year-end 2000). In addition, the interest due to the FDIC is calculated and placed on the receivership books in cases in which the FDIC expects some or all of the interest to be paid in the near future. This interest is excluded from the cost reported in the Failed Bank Cost Analysis (FBCA) even in cases in which it is reported and/or paid by the receivership.
67 Differences in the cost of insurance determination are considered when the FDIC determines the least-cost bid. The cost of preparing for resolution is a sunk cost that would not influence the selection of a winning bidder and thus is excluded from the least-cost test.
68 Our estimates do not follow FDIC regulations and practices for the payment of interest to receivership creditors.
69 This adjustment still does not result in a fully consistent comparison of economic costs. The initial cost estimate uses risk-adjusted discount rates based on the asset composition of the bank, whereas the adjustment to the 2000 FBCA cost estimate effectively uses the FDIC's cost of funds for discounting.
70 For example, assume that a bank sells a portfolio of mortgages but retains the servicing rights. The servicing rights may have real value, but they are typically recorded on the books of the receivership at one dollar.
71 Gross book value is typically defined as the historical cost or unpaid loan balance minus any charge-offs that were recorded by the bank. If there are partial charge-offs recorded by the failed bank that are identifiable at failure, they will normally be reversed; however, this is frequently not the case. Thus, the treatment of charge-offs is not always consistent across receiverships.
72 By asset categories, we mean types of assets grouped by loan purpose (such as single-family mortgages or C&I loans).
73 Banks frequently categorize assets on their general ledgers on the basis of the subsystems used in servicing the assets. For example, all fixed-rate amortizing instruments may be boarded on one subsystem (regardless of loan purpose), and the general ledger may carry one set of accounts that is used for all such instruments. To prepare the Call Reports, a bank would typically use information from both the general ledger and the subsystems. At failure, the FDIC attempts to align the general ledger categories with those used internally by the FDIC receiverships (stratified by loan purpose, as in the Call Report), but time constraints and the differences in asset categories limit the FDIC's ability to succeed in this effort.
74 For many creditors, receiverships pay post-insolvency interest only after the principal is paid in full. The FDIC's current policy is to calculate and record the interest cost due to creditors when 95 percent of the principal balance has been repaid. Such costs are treated as an adjustment to equity rather than an interest expense, because the bulk of such claims are typically accrued but not paid and receiverships use cash-accounting principles.
75 Charge-offs are not reversed.
76 Although accounting policy excludes all asset-sales expenses from these figures, occasionally certain asset-sales expenses incurred at the time of sale are included.
77 This could also occur because of losses incurred by other creditors. However, most at-risk creditors flee banks before failure, so the FDIC and the bank's stockholders typically bear almost all the losses at failure.
78 Some of the adjustments made during the receivership relate to post-closing activities rather than the institution's asset balance at closing. We dropped all adjustments that we could identify as occurring after failure.
79 Unfortunately, we were not always able to determine which report provides the most accurate picture of the institutions. For some of these institutions, there was a change in receivership that appears to bring the institution's balance sheet into closer conformity with the Call Report. This type of change probably indicates that the asset types recorded on the institution's general ledger did not align well with the Call Report.
