Federal Deposit
Insurance Corporation

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

Home > Industry Analysis > Research & Analysis > FDIC Banking Review

FDIC Banking Review

*Daniel A. Nuxoll is Senior Economist, John O'Keefe is Chief of the Financial Risk Measurement Section, and Katherine Samolyk is Senior Financial Economist in the Division of Insurance and Research, Federal Deposit Insurance Corporation. The views expressed here are those of the authors and not necessarily those of the Federal Deposit Insurance Corporation or its staff.

1For discussions of off-site monitoring models, see: Cole, Cornyn, and Gunther (1995); Gilbert, Meyer, and Vaughan (1999); and Reidhill and O'Keefe (1997).

2Samolyk (1994a) finds linkages between state banking conditions and state personal-income growth during the 1980s and early 1990s that are consistent with the existence of a regional credit channel. Neely and Wheelock (1997) conclude that the dispersions in state-level bank earnings can be attributed largely to disparities in state economic conditions; similarly, Samolyk (1994b) finds that state economic conditions explain significant amounts of observed differences in bank asset quality and bank profitability during the1980s and the early 1990s.

3But since bank failure is an extreme event, its correlation with standard measures of local economic conditions (such as income growth or unemployment rates) may be more complex than the correlation of continuous performance measures, such as bank asset-quality ratios. In addition, external capital injections or friendly mergers can prevent bank failures from occurring.

4CAMEL stands for Capital, Asset quality, Management, Earnings, and Liquidity. In 1997, the ratings became CAMELS with the addition of a market Sensitivity rating. However, because most of our data are from the period before 1997, we refer to CAMEL ratings.

5State-level economic variables can contribute to off-site monitoring models without being perfect measures of the relevant economic conditions because they bear on all banks. What is necessary is only that the economic variables provide reasonable approximations of the relevant "local" conditions for most banks in the sample.

6In contrast, although employment and (annual) income data are produced at the county level, the latter are not available until 18 months after the end of the year.

7For more detail, see Nuxoll (2003).

8Cole, Cornyn, and Gunther (1995) report on the development of the Federal Reserve System's failure-prediction and CAMEL-prediction models. Various prototypes included state-level data on unemployment rates, personal-income growth, and housing permits; however, the explanatory power of the state-level economic variables "is attenuated by the inclusion of bank-specific variables in the model" (p. 8). Other researchers have estimated bank failure-prediction models that include economic proxies, but they do not assess the contribution of the economic variables in their models.

9The nonperforming-asset ratio equals the sum of total loans and leases more than 90 days past due plus nonaccruing loans and leases plus other real estate owned as a share of total assets.

10Because the nonperforming-asset ratios of very large banks reflect the national and international scale of their activities, banks with more than $20 billion (1994) in assets were excluded from the calculations illustrated in figure 4.

11Thus, for each bank size class and each sample period, we estimate the following models:
             (bank model)
   (2) (banking & economic model)
   (3)              (naïve model)
where j = jth bank size class (1-6).
         i = ith observation in size class j.
         k = kth right-hand-side banking variable.
         l = lth right-hand-side economic variable.
In sample, the RMSE of the naïve model regressions will be very close to the standard deviation of the dependant variable for each sample of banks. Out of sample, the RMSE of the naïve model forecasts can differ from the standard deviation of realized asset-quality changes because the forecasts are based on the average changes in nonperforming-asset ratios evident historically, and these average changes can differ from the realized mean.

12Observations for all four years in a given sample period are pooled in what is called a cross-sectional time-series analysis. The four-quarter change in a bank's asset-quality ratio is measured as the percentage change in the ratio of nonperforming assets to total assets. Nonperforming assets include loans 90 days past due and still accruing, nonaccruing loans and leases, and other real estate owned.

13Because we are linking bank data over time, we adjust data where necessary to reflect bank mergers so as to get a consistent historical series for each bank.

14To control for variations in the national economy during a given sample period, the set of economic variables also includes one lag of U.S. personal-income growth and one lag of the percentage-point change in the GDP deflator (as a proxy for inflation).

15A thorough analysis of the causes of the U.S. banking crises of the 1980s and early 1990s found that a "boom/bust" cycle in banking markets was a common feature; the analysis also examined the implications of these cycles for bank growth. See Federal Deposit Insurance Corporation (1997).

16Bank supervisors also can place restrictions on bank growth. Regulatory capital requirements are perhaps the most general restriction and limit the degree to which a bank can engage in leveraged growth. Moreover, bank management may be required to obtain supervisory approval before engaging in some types of new activities.

17For an extensive description of the FDIC's GMS during the late 1980s and early 1990s, see Reidhill and O'Keefe (1997).

18The mean percentage change in state unemployment rates for examined banks is weighted by the number of banks examined within a state each quarter. This was done to ensure that the economic conditions shown in figure 6 reflect those faced by the banks whose CAMEL rating changes also are shown in figure 6.

19The Pearson's correlation coefficient (and p-values in parentheses) between the mean percentage change in unemployment rates and changes in CAMEL ratings for the period 1984 through 1997 is 0.24 (0.0734) for the Southwest and 0.73 (0.0001) for the Northeast.

20Volatile liabilities are defined here as the sum of time deposits over $100,000, foreign deposits, federal funds and securities sold under repurchase agreements, demand notes issued to the U.S. Treasury, and other borrowed money.

21We use the same approach to constructing the loan concentration index that Reidhill and O'Keefe (1997) used. Specifically, certain risky loan concentrations are weighted more heavily in the HHI.

22National rankings are used for all measures of portfolio change as well as for the summary measure of portfolio concentration. All remaining ratios are ranked with the use of peer groups. To form peer groups, we stratified banks into eight broad U.S. geographic regions and two asset-size classes ("large" or "small" depending on whether the asset size is greater or less than $1 billion).

23Specifically, we used the year-end percentile rankings of the 9 financial measures and the raw values of the 2 supervisory measures in the bank model as explanatory variables in a logistic regression model to explain the incidence of composite CAMEL downgrades during the subsequent three-year period. The weights obtained from a given three-year estimation period are applied out-of-sample as weights to the 11 variables, and the weighted sum is used as the growth index.

24Reidhill and O'Keefe (1997) indicate that there may be a three- to five-year lag between periods of excessive growth and subsequent declines in bank safety and soundness.

25A study by Avery and Gordy (1998) examines the extent to which recent loan growth (that is, growth during the previous two years) has been associated with a bank's current profitability and asset-quality ratios. The models in their study include a broad range of economic variables constructed from economic data at the county, state, and national levels. Although their study does not attempt to predict emerging banking problems, it does indicate that loan growth should be measured relative to economic fundamentals.

Last Updated 7/25/2003 Questions, Suggestions & Requests

Skip Footer back to content